blob: 6d2f561d517cf3e3df4b76941d09f75bba0bc9d0 [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
386 if (!mapping_cap_writeback_dirty(mapping))
387 return 0;
388
Tejun Heob16b1de2015-06-02 08:39:48 -0600389 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600391 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 return ret;
393}
394
395static inline int __filemap_fdatawrite(struct address_space *mapping,
396 int sync_mode)
397{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700398 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399}
400
401int filemap_fdatawrite(struct address_space *mapping)
402{
403 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
404}
405EXPORT_SYMBOL(filemap_fdatawrite);
406
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400407int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800408 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
410 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
411}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400412EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Randy Dunlap485bb992006-06-23 02:03:49 -0700414/**
415 * filemap_flush - mostly a non-blocking flush
416 * @mapping: target address_space
417 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 * This is a mostly non-blocking flush. Not suitable for data-integrity
419 * purposes - I/O may not be started against all dirty pages.
420 */
421int filemap_flush(struct address_space *mapping)
422{
423 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
424}
425EXPORT_SYMBOL(filemap_flush);
426
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800427static int __filemap_fdatawait_range(struct address_space *mapping,
428 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300430 pgoff_t index = start_byte >> PAGE_SHIFT;
431 pgoff_t end = end_byte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 struct pagevec pvec;
433 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800434 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200436 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700437 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 while ((index <= end) &&
441 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
442 PAGECACHE_TAG_WRITEBACK,
443 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
444 unsigned i;
445
446 for (i = 0; i < nr_pages; i++) {
447 struct page *page = pvec.pages[i];
448
449 /* until radix tree lookup accepts end_index */
450 if (page->index > end)
451 continue;
452
453 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800454 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 ret = -EIO;
456 }
457 pagevec_release(&pvec);
458 cond_resched();
459 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700460out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800461 return ret;
462}
463
464/**
465 * filemap_fdatawait_range - wait for writeback to complete
466 * @mapping: address space structure to wait for
467 * @start_byte: offset in bytes where the range starts
468 * @end_byte: offset in bytes where the range ends (inclusive)
469 *
470 * Walk the list of under-writeback pages of the given address space
471 * in the given range and wait for all of them. Check error status of
472 * the address space and return it.
473 *
474 * Since the error status of the address space is cleared by this function,
475 * callers are responsible for checking the return value and handling and/or
476 * reporting the error.
477 */
478int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
479 loff_t end_byte)
480{
481 int ret, ret2;
482
483 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700484 ret2 = filemap_check_errors(mapping);
485 if (!ret)
486 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488 return ret;
489}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200490EXPORT_SYMBOL(filemap_fdatawait_range);
491
492/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800493 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
494 * @mapping: address space structure to wait for
495 *
496 * Walk the list of under-writeback pages of the given address space
497 * and wait for all of them. Unlike filemap_fdatawait(), this function
498 * does not clear error status of the address space.
499 *
500 * Use this function if callers don't handle errors themselves. Expected
501 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
502 * fsfreeze(8)
503 */
504void filemap_fdatawait_keep_errors(struct address_space *mapping)
505{
506 loff_t i_size = i_size_read(mapping->host);
507
508 if (i_size == 0)
509 return;
510
511 __filemap_fdatawait_range(mapping, 0, i_size - 1);
512}
513
514/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700515 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700517 *
518 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800519 * and wait for all of them. Check error status of the address space
520 * and return it.
521 *
522 * Since the error status of the address space is cleared by this function,
523 * callers are responsible for checking the return value and handling and/or
524 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 */
526int filemap_fdatawait(struct address_space *mapping)
527{
528 loff_t i_size = i_size_read(mapping->host);
529
530 if (i_size == 0)
531 return 0;
532
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200533 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534}
535EXPORT_SYMBOL(filemap_fdatawait);
536
537int filemap_write_and_wait(struct address_space *mapping)
538{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800539 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800541 if ((!dax_mapping(mapping) && mapping->nrpages) ||
542 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800543 err = filemap_fdatawrite(mapping);
544 /*
545 * Even if the above returned error, the pages may be
546 * written partially (e.g. -ENOSPC), so we wait for it.
547 * But the -EIO is special case, it may indicate the worst
548 * thing (e.g. bug) happened, so we avoid waiting for it.
549 */
550 if (err != -EIO) {
551 int err2 = filemap_fdatawait(mapping);
552 if (!err)
553 err = err2;
554 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700555 } else {
556 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800558 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800560EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Randy Dunlap485bb992006-06-23 02:03:49 -0700562/**
563 * filemap_write_and_wait_range - write out & wait on a file range
564 * @mapping: the address_space for the pages
565 * @lstart: offset in bytes where the range starts
566 * @lend: offset in bytes where the range ends (inclusive)
567 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800568 * Write out and wait upon file offsets lstart->lend, inclusive.
569 *
570 * Note that `lend' is inclusive (describes the last byte to be written) so
571 * that this function can be used to write to the very end-of-file (end = -1).
572 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573int filemap_write_and_wait_range(struct address_space *mapping,
574 loff_t lstart, loff_t lend)
575{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800576 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800578 if ((!dax_mapping(mapping) && mapping->nrpages) ||
579 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800580 err = __filemap_fdatawrite_range(mapping, lstart, lend,
581 WB_SYNC_ALL);
582 /* See comment of filemap_write_and_wait() */
583 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200584 int err2 = filemap_fdatawait_range(mapping,
585 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800586 if (!err)
587 err = err2;
588 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700589 } else {
590 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800592 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593}
Chris Masonf6995582009-04-15 13:22:37 -0400594EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Randy Dunlap485bb992006-06-23 02:03:49 -0700596/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700597 * replace_page_cache_page - replace a pagecache page with a new one
598 * @old: page to be replaced
599 * @new: page to replace with
600 * @gfp_mask: allocation mode
601 *
602 * This function replaces a page in the pagecache with a new one. On
603 * success it acquires the pagecache reference for the new page and
604 * drops it for the old page. Both the old and new pages must be
605 * locked. This function does not add the new page to the LRU, the
606 * caller must do that.
607 *
608 * The remove + add is atomic. The only way this function can fail is
609 * memory allocation failure.
610 */
611int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
612{
613 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700614
Sasha Levin309381fea2014-01-23 15:52:54 -0800615 VM_BUG_ON_PAGE(!PageLocked(old), old);
616 VM_BUG_ON_PAGE(!PageLocked(new), new);
617 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700618
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -0700619 error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700620 if (!error) {
621 struct address_space *mapping = old->mapping;
622 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400623 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700624
625 pgoff_t offset = old->index;
626 freepage = mapping->a_ops->freepage;
627
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300628 get_page(new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700629 new->mapping = mapping;
630 new->index = offset;
631
Greg Thelenc4843a72015-05-22 17:13:16 -0400632 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700633 __delete_from_page_cache(old, NULL);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700634 error = page_cache_tree_insert(mapping, new, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700635 BUG_ON(error);
Michal Hocko4165b9b2015-06-24 16:57:24 -0700636
637 /*
638 * hugetlb pages do not participate in page cache accounting.
639 */
640 if (!PageHuge(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700641 __inc_node_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700642 if (PageSwapBacked(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700643 __inc_node_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400644 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700645 mem_cgroup_migrate(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700646 radix_tree_preload_end();
647 if (freepage)
648 freepage(old);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300649 put_page(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700650 }
651
652 return error;
653}
654EXPORT_SYMBOL_GPL(replace_page_cache_page);
655
Johannes Weinera5289102014-04-03 14:47:51 -0700656static int __add_to_page_cache_locked(struct page *page,
657 struct address_space *mapping,
658 pgoff_t offset, gfp_t gfp_mask,
659 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
Johannes Weiner00501b52014-08-08 14:19:20 -0700661 int huge = PageHuge(page);
662 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700663 int error;
664
Sasha Levin309381fea2014-01-23 15:52:54 -0800665 VM_BUG_ON_PAGE(!PageLocked(page), page);
666 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700667
Johannes Weiner00501b52014-08-08 14:19:20 -0700668 if (!huge) {
669 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800670 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700671 if (error)
672 return error;
673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -0700675 error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700676 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700677 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800678 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700679 return error;
680 }
681
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300682 get_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700683 page->mapping = mapping;
684 page->index = offset;
685
686 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700687 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700688 radix_tree_preload_end();
689 if (unlikely(error))
690 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700691
692 /* hugetlb pages do not participate in page cache accounting. */
693 if (!huge)
Mel Gorman11fb9982016-07-28 15:46:20 -0700694 __inc_node_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700695 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700696 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800697 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700698 trace_mm_filemap_add_to_page_cache(page);
699 return 0;
700err_insert:
701 page->mapping = NULL;
702 /* Leave page->index set: truncation relies upon it */
703 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700704 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800705 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300706 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 return error;
708}
Johannes Weinera5289102014-04-03 14:47:51 -0700709
710/**
711 * add_to_page_cache_locked - add a locked page to the pagecache
712 * @page: page to add
713 * @mapping: the page's address_space
714 * @offset: page index
715 * @gfp_mask: page allocation mode
716 *
717 * This function is used to add a page to the pagecache. It must be locked.
718 * This function does not add the page to the LRU. The caller must do that.
719 */
720int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
721 pgoff_t offset, gfp_t gfp_mask)
722{
723 return __add_to_page_cache_locked(page, mapping, offset,
724 gfp_mask, NULL);
725}
Nick Piggine2867812008-07-25 19:45:30 -0700726EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400729 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
Johannes Weinera5289102014-04-03 14:47:51 -0700731 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700732 int ret;
733
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800734 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700735 ret = __add_to_page_cache_locked(page, mapping, offset,
736 gfp_mask, &shadow);
737 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800738 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700739 else {
740 /*
741 * The page might have been evicted from cache only
742 * recently, in which case it should be activated like
743 * any other repeatedly accessed page.
Rik van Rielf0281a02016-05-20 16:56:25 -0700744 * The exception is pages getting rewritten; evicting other
745 * data from the working set, only to cache data that will
746 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700747 */
Rik van Rielf0281a02016-05-20 16:56:25 -0700748 if (!(gfp_mask & __GFP_WRITE) &&
749 shadow && workingset_refault(shadow)) {
Johannes Weinera5289102014-04-03 14:47:51 -0700750 SetPageActive(page);
751 workingset_activation(page);
752 } else
753 ClearPageActive(page);
754 lru_cache_add(page);
755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 return ret;
757}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300758EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Paul Jackson44110fe2006-03-24 03:16:04 -0800760#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700761struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800762{
Miao Xiec0ff7452010-05-24 14:32:08 -0700763 int n;
764 struct page *page;
765
Paul Jackson44110fe2006-03-24 03:16:04 -0800766 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700767 unsigned int cpuset_mems_cookie;
768 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700769 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700770 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700771 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700772 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700773
Miao Xiec0ff7452010-05-24 14:32:08 -0700774 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800775 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700776 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800777}
Nick Piggin2ae88142006-10-28 10:38:23 -0700778EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800779#endif
780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781/*
782 * In order to wait for pages to become available there must be
783 * waitqueues associated with pages. By using a hash table of
784 * waitqueues where the bucket discipline is to maintain all
785 * waiters on the same queue and wake all when any of the pages
786 * become available, and for the woken contexts to check to be
787 * sure the appropriate page became available, this saves space
788 * at a cost of "thundering herd" phenomena during rare hash
789 * collisions.
790 */
NeilBrowna4796e32014-09-24 11:28:32 +1000791wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
Linus Torvalds9dcb8b62016-10-26 10:15:30 -0700793 return bit_waitqueue(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794}
NeilBrowna4796e32014-09-24 11:28:32 +1000795EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Harvey Harrison920c7a52008-02-04 22:29:26 -0800797void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
799 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
800
801 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000802 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 TASK_UNINTERRUPTIBLE);
804}
805EXPORT_SYMBOL(wait_on_page_bit);
806
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700807int wait_on_page_bit_killable(struct page *page, int bit_nr)
808{
809 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
810
811 if (!test_bit(bit_nr, &page->flags))
812 return 0;
813
814 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000815 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700816}
817
NeilBrowncbbce822014-09-25 13:55:19 +1000818int wait_on_page_bit_killable_timeout(struct page *page,
819 int bit_nr, unsigned long timeout)
820{
821 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
822
823 wait.key.timeout = jiffies + timeout;
824 if (!test_bit(bit_nr, &page->flags))
825 return 0;
826 return __wait_on_bit(page_waitqueue(page), &wait,
827 bit_wait_io_timeout, TASK_KILLABLE);
828}
829EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100832 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700833 * @page: Page defining the wait queue of interest
834 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100835 *
836 * Add an arbitrary @waiter to the wait queue for the nominated @page.
837 */
838void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
839{
840 wait_queue_head_t *q = page_waitqueue(page);
841 unsigned long flags;
842
843 spin_lock_irqsave(&q->lock, flags);
844 __add_wait_queue(q, waiter);
845 spin_unlock_irqrestore(&q->lock, flags);
846}
847EXPORT_SYMBOL_GPL(add_page_wait_queue);
848
849/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700850 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 * @page: the page
852 *
853 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
854 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900855 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
857 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700858 * The mb is necessary to enforce ordering between the clear_bit and the read
859 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800861void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800863 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800864 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700865 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100866 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 wake_up_page(page, PG_locked);
868}
869EXPORT_SYMBOL(unlock_page);
870
Randy Dunlap485bb992006-06-23 02:03:49 -0700871/**
872 * end_page_writeback - end writeback against a page
873 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 */
875void end_page_writeback(struct page *page)
876{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700877 /*
878 * TestClearPageReclaim could be used here but it is an atomic
879 * operation and overkill in this particular case. Failing to
880 * shuffle a page marked for immediate reclaim is too mild to
881 * justify taking an atomic operation penalty at the end of
882 * ever page writeback.
883 */
884 if (PageReclaim(page)) {
885 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700886 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700887 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700888
889 if (!test_clear_page_writeback(page))
890 BUG();
891
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100892 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 wake_up_page(page, PG_writeback);
894}
895EXPORT_SYMBOL(end_page_writeback);
896
Matthew Wilcox57d99842014-06-04 16:07:45 -0700897/*
898 * After completing I/O on a page, call this routine to update the page
899 * flags appropriately
900 */
Jens Axboec11f0c02016-08-05 08:11:04 -0600901void page_endio(struct page *page, bool is_write, int err)
Matthew Wilcox57d99842014-06-04 16:07:45 -0700902{
Jens Axboec11f0c02016-08-05 08:11:04 -0600903 if (!is_write) {
Matthew Wilcox57d99842014-06-04 16:07:45 -0700904 if (!err) {
905 SetPageUptodate(page);
906 } else {
907 ClearPageUptodate(page);
908 SetPageError(page);
909 }
910 unlock_page(page);
Mike Christieabf54542016-08-04 14:23:34 -0600911 } else {
Matthew Wilcox57d99842014-06-04 16:07:45 -0700912 if (err) {
Minchan Kim2c290ee2017-02-24 14:59:59 -0800913 struct address_space *mapping;
914
Matthew Wilcox57d99842014-06-04 16:07:45 -0700915 SetPageError(page);
Minchan Kim2c290ee2017-02-24 14:59:59 -0800916 mapping = page_mapping(page);
917 if (mapping)
918 mapping_set_error(mapping, err);
Matthew Wilcox57d99842014-06-04 16:07:45 -0700919 }
920 end_page_writeback(page);
921 }
922}
923EXPORT_SYMBOL_GPL(page_endio);
924
Randy Dunlap485bb992006-06-23 02:03:49 -0700925/**
926 * __lock_page - get a lock on the page, assuming we need to sleep to get it
927 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800929void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800931 struct page *page_head = compound_head(page);
932 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800934 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 TASK_UNINTERRUPTIBLE);
936}
937EXPORT_SYMBOL(__lock_page);
938
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800939int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500940{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800941 struct page *page_head = compound_head(page);
942 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500943
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800944 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000945 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500946}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300947EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500948
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700949/*
950 * Return values:
951 * 1 - page is locked; mmap_sem is still held.
952 * 0 - page is not locked.
953 * mmap_sem has been released (up_read()), unless flags had both
954 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
955 * which case mmap_sem is still held.
956 *
957 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
958 * with the page locked and the mmap_sem unperturbed.
959 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700960int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
961 unsigned int flags)
962{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700963 if (flags & FAULT_FLAG_ALLOW_RETRY) {
964 /*
965 * CAUTION! In this case, mmap_sem is not released
966 * even though return 0.
967 */
968 if (flags & FAULT_FLAG_RETRY_NOWAIT)
969 return 0;
970
971 up_read(&mm->mmap_sem);
972 if (flags & FAULT_FLAG_KILLABLE)
973 wait_on_page_locked_killable(page);
974 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700975 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700976 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700977 } else {
978 if (flags & FAULT_FLAG_KILLABLE) {
979 int ret;
980
981 ret = __lock_page_killable(page);
982 if (ret) {
983 up_read(&mm->mmap_sem);
984 return 0;
985 }
986 } else
987 __lock_page(page);
988 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700989 }
990}
991
Randy Dunlap485bb992006-06-23 02:03:49 -0700992/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700993 * page_cache_next_hole - find the next hole (not-present entry)
994 * @mapping: mapping
995 * @index: index
996 * @max_scan: maximum range to search
997 *
998 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
999 * lowest indexed hole.
1000 *
1001 * Returns: the index of the hole if found, otherwise returns an index
1002 * outside of the set specified (in which case 'return - index >=
1003 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1004 * be returned.
1005 *
1006 * page_cache_next_hole may be called under rcu_read_lock. However,
1007 * like radix_tree_gang_lookup, this will not atomically search a
1008 * snapshot of the tree at a single point in time. For example, if a
1009 * hole is created at index 5, then subsequently a hole is created at
1010 * index 10, page_cache_next_hole covering both indexes may return 10
1011 * if called under rcu_read_lock.
1012 */
1013pgoff_t page_cache_next_hole(struct address_space *mapping,
1014 pgoff_t index, unsigned long max_scan)
1015{
1016 unsigned long i;
1017
1018 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001019 struct page *page;
1020
1021 page = radix_tree_lookup(&mapping->page_tree, index);
1022 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001023 break;
1024 index++;
1025 if (index == 0)
1026 break;
1027 }
1028
1029 return index;
1030}
1031EXPORT_SYMBOL(page_cache_next_hole);
1032
1033/**
1034 * page_cache_prev_hole - find the prev hole (not-present entry)
1035 * @mapping: mapping
1036 * @index: index
1037 * @max_scan: maximum range to search
1038 *
1039 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1040 * the first hole.
1041 *
1042 * Returns: the index of the hole if found, otherwise returns an index
1043 * outside of the set specified (in which case 'index - return >=
1044 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1045 * will be returned.
1046 *
1047 * page_cache_prev_hole may be called under rcu_read_lock. However,
1048 * like radix_tree_gang_lookup, this will not atomically search a
1049 * snapshot of the tree at a single point in time. For example, if a
1050 * hole is created at index 10, then subsequently a hole is created at
1051 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1052 * called under rcu_read_lock.
1053 */
1054pgoff_t page_cache_prev_hole(struct address_space *mapping,
1055 pgoff_t index, unsigned long max_scan)
1056{
1057 unsigned long i;
1058
1059 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001060 struct page *page;
1061
1062 page = radix_tree_lookup(&mapping->page_tree, index);
1063 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001064 break;
1065 index--;
1066 if (index == ULONG_MAX)
1067 break;
1068 }
1069
1070 return index;
1071}
1072EXPORT_SYMBOL(page_cache_prev_hole);
1073
1074/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001075 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001076 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001077 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001078 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001079 * Looks up the page cache slot at @mapping & @offset. If there is a
1080 * page cache page, it is returned with an increased refcount.
1081 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001082 * If the slot holds a shadow entry of a previously evicted page, or a
1083 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001084 *
1085 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001087struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088{
Nick Piggina60637c2008-07-25 19:45:31 -07001089 void **pagep;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001090 struct page *head, *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
Nick Piggina60637c2008-07-25 19:45:31 -07001092 rcu_read_lock();
1093repeat:
1094 page = NULL;
1095 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1096 if (pagep) {
1097 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001098 if (unlikely(!page))
1099 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001100 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001101 if (radix_tree_deref_retry(page))
1102 goto repeat;
1103 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001104 * A shadow entry of a recently evicted page,
1105 * or a swap entry from shmem/tmpfs. Return
1106 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001107 */
1108 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001109 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001110
1111 head = compound_head(page);
1112 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001113 goto repeat;
1114
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001115 /* The page was split under us? */
1116 if (compound_head(page) != head) {
1117 put_page(head);
1118 goto repeat;
1119 }
1120
Nick Piggina60637c2008-07-25 19:45:31 -07001121 /*
1122 * Has the page moved?
1123 * This is part of the lockless pagecache protocol. See
1124 * include/linux/pagemap.h for details.
1125 */
1126 if (unlikely(page != *pagep)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001127 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001128 goto repeat;
1129 }
1130 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001131out:
Nick Piggina60637c2008-07-25 19:45:31 -07001132 rcu_read_unlock();
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 return page;
1135}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001136EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Randy Dunlap485bb992006-06-23 02:03:49 -07001138/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001139 * find_lock_entry - locate, pin and lock a page cache entry
1140 * @mapping: the address_space to search
1141 * @offset: the page cache index
1142 *
1143 * Looks up the page cache slot at @mapping & @offset. If there is a
1144 * page cache page, it is returned locked and with an increased
1145 * refcount.
1146 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001147 * If the slot holds a shadow entry of a previously evicted page, or a
1148 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001149 *
1150 * Otherwise, %NULL is returned.
1151 *
1152 * find_lock_entry() may sleep.
1153 */
1154struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155{
1156 struct page *page;
1157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001159 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001160 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001161 lock_page(page);
1162 /* Has the page been truncated? */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001163 if (unlikely(page_mapping(page) != mapping)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001164 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001165 put_page(page);
Nick Piggina60637c2008-07-25 19:45:31 -07001166 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001168 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 return page;
1171}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001172EXPORT_SYMBOL(find_lock_entry);
1173
1174/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001175 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001176 * @mapping: the address_space to search
1177 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001178 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001179 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001180 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001181 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001182 *
Randy Dunlap75325182014-07-30 16:08:37 -07001183 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001184 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001185 * FGP_ACCESSED: the page will be marked accessed
1186 * FGP_LOCK: Page is return locked
1187 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001188 * @gfp_mask and added to the page cache and the VM's LRU
1189 * list. The page is returned locked and with an increased
1190 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001191 *
1192 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1193 * if the GFP flags specified for FGP_CREAT are atomic.
1194 *
1195 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001196 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001197struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001198 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199{
Nick Piggineb2be182007-10-16 01:24:57 -07001200 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001203 page = find_get_entry(mapping, offset);
1204 if (radix_tree_exceptional_entry(page))
1205 page = NULL;
1206 if (!page)
1207 goto no_page;
1208
1209 if (fgp_flags & FGP_LOCK) {
1210 if (fgp_flags & FGP_NOWAIT) {
1211 if (!trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001212 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001213 return NULL;
1214 }
1215 } else {
1216 lock_page(page);
1217 }
1218
1219 /* Has the page been truncated? */
1220 if (unlikely(page->mapping != mapping)) {
1221 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001222 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001223 goto repeat;
1224 }
1225 VM_BUG_ON_PAGE(page->index != offset, page);
1226 }
1227
1228 if (page && (fgp_flags & FGP_ACCESSED))
1229 mark_page_accessed(page);
1230
1231no_page:
1232 if (!page && (fgp_flags & FGP_CREAT)) {
1233 int err;
1234 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001235 gfp_mask |= __GFP_WRITE;
1236 if (fgp_flags & FGP_NOFS)
1237 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001238
Michal Hocko45f87de2014-12-29 20:30:35 +01001239 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001240 if (!page)
1241 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001242
1243 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1244 fgp_flags |= FGP_LOCK;
1245
Hugh Dickinseb39d612014-08-06 16:06:43 -07001246 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001247 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001248 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001249
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -07001250 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001251 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001252 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07001253 page = NULL;
1254 if (err == -EEXIST)
1255 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 return page;
1260}
Mel Gorman2457aec2014-06-04 16:10:31 -07001261EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
1263/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001264 * find_get_entries - gang pagecache lookup
1265 * @mapping: The address_space to search
1266 * @start: The starting page cache index
1267 * @nr_entries: The maximum number of entries
1268 * @entries: Where the resulting entries are placed
1269 * @indices: The cache indices corresponding to the entries in @entries
1270 *
1271 * find_get_entries() will search for and return a group of up to
1272 * @nr_entries entries in the mapping. The entries are placed at
1273 * @entries. find_get_entries() takes a reference against any actual
1274 * pages it returns.
1275 *
1276 * The search returns a group of mapping-contiguous page cache entries
1277 * with ascending indexes. There may be holes in the indices due to
1278 * not-present pages.
1279 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001280 * Any shadow entries of evicted pages, or swap entries from
1281 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001282 *
1283 * find_get_entries() returns the number of pages and shadow entries
1284 * which were found.
1285 */
1286unsigned find_get_entries(struct address_space *mapping,
1287 pgoff_t start, unsigned int nr_entries,
1288 struct page **entries, pgoff_t *indices)
1289{
1290 void **slot;
1291 unsigned int ret = 0;
1292 struct radix_tree_iter iter;
1293
1294 if (!nr_entries)
1295 return 0;
1296
1297 rcu_read_lock();
Johannes Weiner0cd61442014-04-03 14:47:46 -07001298 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001299 struct page *head, *page;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001300repeat:
1301 page = radix_tree_deref_slot(slot);
1302 if (unlikely(!page))
1303 continue;
1304 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001305 if (radix_tree_deref_retry(page)) {
1306 slot = radix_tree_iter_retry(&iter);
1307 continue;
1308 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001309 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -08001310 * A shadow entry of a recently evicted page, a swap
1311 * entry from shmem/tmpfs or a DAX entry. Return it
1312 * without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001313 */
1314 goto export;
1315 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001316
1317 head = compound_head(page);
1318 if (!page_cache_get_speculative(head))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001319 goto repeat;
1320
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001321 /* The page was split under us? */
1322 if (compound_head(page) != head) {
1323 put_page(head);
1324 goto repeat;
1325 }
1326
Johannes Weiner0cd61442014-04-03 14:47:46 -07001327 /* Has the page moved? */
1328 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001329 put_page(head);
Johannes Weiner0cd61442014-04-03 14:47:46 -07001330 goto repeat;
1331 }
1332export:
1333 indices[ret] = iter.index;
1334 entries[ret] = page;
1335 if (++ret == nr_entries)
1336 break;
1337 }
1338 rcu_read_unlock();
1339 return ret;
1340}
1341
1342/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 * find_get_pages - gang pagecache lookup
1344 * @mapping: The address_space to search
1345 * @start: The starting page index
1346 * @nr_pages: The maximum number of pages
1347 * @pages: Where the resulting pages are placed
1348 *
1349 * find_get_pages() will search for and return a group of up to
1350 * @nr_pages pages in the mapping. The pages are placed at @pages.
1351 * find_get_pages() takes a reference against the returned pages.
1352 *
1353 * The search returns a group of mapping-contiguous pages with ascending
1354 * indexes. There may be holes in the indices due to not-present pages.
1355 *
1356 * find_get_pages() returns the number of pages which were found.
1357 */
1358unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1359 unsigned int nr_pages, struct page **pages)
1360{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001361 struct radix_tree_iter iter;
1362 void **slot;
1363 unsigned ret = 0;
1364
1365 if (unlikely(!nr_pages))
1366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Nick Piggina60637c2008-07-25 19:45:31 -07001368 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001369 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001370 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001371repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001372 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001373 if (unlikely(!page))
1374 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001375
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001376 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001377 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001378 slot = radix_tree_iter_retry(&iter);
1379 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001380 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001381 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001382 * A shadow entry of a recently evicted page,
1383 * or a swap entry from shmem/tmpfs. Skip
1384 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001385 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001386 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001387 }
Nick Piggina60637c2008-07-25 19:45:31 -07001388
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001389 head = compound_head(page);
1390 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001391 goto repeat;
1392
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001393 /* The page was split under us? */
1394 if (compound_head(page) != head) {
1395 put_page(head);
1396 goto repeat;
1397 }
1398
Nick Piggina60637c2008-07-25 19:45:31 -07001399 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001400 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001401 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001402 goto repeat;
1403 }
1404
1405 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001406 if (++ret == nr_pages)
1407 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001408 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001409
Nick Piggina60637c2008-07-25 19:45:31 -07001410 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 return ret;
1412}
1413
Jens Axboeebf43502006-04-27 08:46:01 +02001414/**
1415 * find_get_pages_contig - gang contiguous pagecache lookup
1416 * @mapping: The address_space to search
1417 * @index: The starting page index
1418 * @nr_pages: The maximum number of pages
1419 * @pages: Where the resulting pages are placed
1420 *
1421 * find_get_pages_contig() works exactly like find_get_pages(), except
1422 * that the returned number of pages are guaranteed to be contiguous.
1423 *
1424 * find_get_pages_contig() returns the number of pages which were found.
1425 */
1426unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1427 unsigned int nr_pages, struct page **pages)
1428{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001429 struct radix_tree_iter iter;
1430 void **slot;
1431 unsigned int ret = 0;
1432
1433 if (unlikely(!nr_pages))
1434 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001435
Nick Piggina60637c2008-07-25 19:45:31 -07001436 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001437 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001438 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001439repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001440 page = radix_tree_deref_slot(slot);
1441 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001442 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001443 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001444
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001445 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001446 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001447 slot = radix_tree_iter_retry(&iter);
1448 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001449 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001450 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001451 * A shadow entry of a recently evicted page,
1452 * or a swap entry from shmem/tmpfs. Stop
1453 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001454 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001455 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001456 }
Nick Piggina60637c2008-07-25 19:45:31 -07001457
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001458 head = compound_head(page);
1459 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001460 goto repeat;
1461
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001462 /* The page was split under us? */
1463 if (compound_head(page) != head) {
1464 put_page(head);
1465 goto repeat;
1466 }
1467
Nick Piggina60637c2008-07-25 19:45:31 -07001468 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001469 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001470 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001471 goto repeat;
1472 }
1473
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001474 /*
1475 * must check mapping and index after taking the ref.
1476 * otherwise we can get both false positives and false
1477 * negatives, which is just confusing to the caller.
1478 */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001479 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001480 put_page(page);
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001481 break;
1482 }
1483
Nick Piggina60637c2008-07-25 19:45:31 -07001484 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001485 if (++ret == nr_pages)
1486 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001487 }
Nick Piggina60637c2008-07-25 19:45:31 -07001488 rcu_read_unlock();
1489 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001490}
David Howellsef71c152007-05-09 02:33:44 -07001491EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001492
Randy Dunlap485bb992006-06-23 02:03:49 -07001493/**
1494 * find_get_pages_tag - find and return pages that match @tag
1495 * @mapping: the address_space to search
1496 * @index: the starting page index
1497 * @tag: the tag index
1498 * @nr_pages: the maximum number of pages
1499 * @pages: where the resulting pages are placed
1500 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001502 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 */
1504unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1505 int tag, unsigned int nr_pages, struct page **pages)
1506{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001507 struct radix_tree_iter iter;
1508 void **slot;
1509 unsigned ret = 0;
1510
1511 if (unlikely(!nr_pages))
1512 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Nick Piggina60637c2008-07-25 19:45:31 -07001514 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001515 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1516 &iter, *index, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001517 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001518repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001519 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001520 if (unlikely(!page))
1521 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001522
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001523 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001524 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001525 slot = radix_tree_iter_retry(&iter);
1526 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001527 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001528 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001529 * A shadow entry of a recently evicted page.
1530 *
1531 * Those entries should never be tagged, but
1532 * this tree walk is lockless and the tags are
1533 * looked up in bulk, one radix tree node at a
1534 * time, so there is a sizable window for page
1535 * reclaim to evict a page we saw tagged.
1536 *
1537 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001538 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001539 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001540 }
Nick Piggina60637c2008-07-25 19:45:31 -07001541
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001542 head = compound_head(page);
1543 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001544 goto repeat;
1545
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001546 /* The page was split under us? */
1547 if (compound_head(page) != head) {
1548 put_page(head);
1549 goto repeat;
1550 }
1551
Nick Piggina60637c2008-07-25 19:45:31 -07001552 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001553 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001554 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001555 goto repeat;
1556 }
1557
1558 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001559 if (++ret == nr_pages)
1560 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001561 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001562
Nick Piggina60637c2008-07-25 19:45:31 -07001563 rcu_read_unlock();
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 if (ret)
1566 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 return ret;
1569}
David Howellsef71c152007-05-09 02:33:44 -07001570EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001572/**
1573 * find_get_entries_tag - find and return entries that match @tag
1574 * @mapping: the address_space to search
1575 * @start: the starting page cache index
1576 * @tag: the tag index
1577 * @nr_entries: the maximum number of entries
1578 * @entries: where the resulting entries are placed
1579 * @indices: the cache indices corresponding to the entries in @entries
1580 *
1581 * Like find_get_entries, except we only return entries which are tagged with
1582 * @tag.
1583 */
1584unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1585 int tag, unsigned int nr_entries,
1586 struct page **entries, pgoff_t *indices)
1587{
1588 void **slot;
1589 unsigned int ret = 0;
1590 struct radix_tree_iter iter;
1591
1592 if (!nr_entries)
1593 return 0;
1594
1595 rcu_read_lock();
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001596 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1597 &iter, start, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001598 struct page *head, *page;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001599repeat:
1600 page = radix_tree_deref_slot(slot);
1601 if (unlikely(!page))
1602 continue;
1603 if (radix_tree_exception(page)) {
1604 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001605 slot = radix_tree_iter_retry(&iter);
1606 continue;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001607 }
1608
1609 /*
1610 * A shadow entry of a recently evicted page, a swap
1611 * entry from shmem/tmpfs or a DAX entry. Return it
1612 * without attempting to raise page count.
1613 */
1614 goto export;
1615 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001616
1617 head = compound_head(page);
1618 if (!page_cache_get_speculative(head))
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001619 goto repeat;
1620
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001621 /* The page was split under us? */
1622 if (compound_head(page) != head) {
1623 put_page(head);
1624 goto repeat;
1625 }
1626
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001627 /* Has the page moved? */
1628 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001629 put_page(head);
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001630 goto repeat;
1631 }
1632export:
1633 indices[ret] = iter.index;
1634 entries[ret] = page;
1635 if (++ret == nr_entries)
1636 break;
1637 }
1638 rcu_read_unlock();
1639 return ret;
1640}
1641EXPORT_SYMBOL(find_get_entries_tag);
1642
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001643/*
1644 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1645 * a _large_ part of the i/o request. Imagine the worst scenario:
1646 *
1647 * ---R__________________________________________B__________
1648 * ^ reading here ^ bad block(assume 4k)
1649 *
1650 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1651 * => failing the whole request => read(R) => read(R+1) =>
1652 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1653 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1654 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1655 *
1656 * It is going insane. Fix it by quickly scaling down the readahead size.
1657 */
1658static void shrink_readahead_size_eio(struct file *filp,
1659 struct file_ra_state *ra)
1660{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001661 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001662}
1663
Randy Dunlap485bb992006-06-23 02:03:49 -07001664/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001665 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001666 * @filp: the file to read
1667 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001668 * @iter: data destination
1669 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001670 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001672 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 *
1674 * This is really ugly. But the goto's actually try to clarify some
1675 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 */
Al Viro6e58e792014-02-03 17:07:03 -05001677static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1678 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001680 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001682 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001683 pgoff_t index;
1684 pgoff_t last_index;
1685 pgoff_t prev_index;
1686 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001687 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001688 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
Wei Fangc2a97372016-10-07 17:01:52 -07001690 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
Linus Torvalds48466c42016-12-14 12:45:25 -08001691 return 0;
Wei Fangc2a97372016-10-07 17:01:52 -07001692 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1693
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001694 index = *ppos >> PAGE_SHIFT;
1695 prev_index = ra->prev_pos >> PAGE_SHIFT;
1696 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1697 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1698 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 for (;;) {
1701 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001702 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001703 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 unsigned long nr, ret;
1705
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707find_page:
Michal Hockob67c7d32017-02-03 13:13:29 -08001708 if (fatal_signal_pending(current)) {
1709 error = -EINTR;
1710 goto out;
1711 }
1712
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001714 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001715 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001716 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001717 index, last_index - index);
1718 page = find_get_page(mapping, index);
1719 if (unlikely(page == NULL))
1720 goto no_cached_page;
1721 }
1722 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001723 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001724 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001725 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001727 if (!PageUptodate(page)) {
Mel Gormanebded022016-03-15 14:55:39 -07001728 /*
1729 * See comment in do_read_cache_page on why
1730 * wait_on_page_locked is used to avoid unnecessarily
1731 * serialisations and why it's safe.
1732 */
Bart Van Asschec4b209a2016-10-07 16:58:33 -07001733 error = wait_on_page_locked_killable(page);
1734 if (unlikely(error))
1735 goto readpage_error;
Mel Gormanebded022016-03-15 14:55:39 -07001736 if (PageUptodate(page))
1737 goto page_ok;
1738
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001739 if (inode->i_blkbits == PAGE_SHIFT ||
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001740 !mapping->a_ops->is_partially_uptodate)
1741 goto page_not_up_to_date;
Eryu Guan6d6d36b2016-11-01 15:43:07 +08001742 /* pipes can't handle partially uptodate pages */
1743 if (unlikely(iter->type & ITER_PIPE))
1744 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001745 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001746 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001747 /* Did it get truncated before we got the lock? */
1748 if (!page->mapping)
1749 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001750 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001751 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001752 goto page_not_up_to_date_locked;
1753 unlock_page(page);
1754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001756 /*
1757 * i_size must be checked after we know the page is Uptodate.
1758 *
1759 * Checking i_size after the check allows us to calculate
1760 * the correct value for "nr", which means the zero-filled
1761 * part of the page is not copied back to userspace (unless
1762 * another truncate extends the file - this is desired though).
1763 */
1764
1765 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001766 end_index = (isize - 1) >> PAGE_SHIFT;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001767 if (unlikely(!isize || index > end_index)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001768 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001769 goto out;
1770 }
1771
1772 /* nr is the maximum number of bytes to copy from this page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001773 nr = PAGE_SIZE;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001774 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001775 nr = ((isize - 1) & ~PAGE_MASK) + 1;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001776 if (nr <= offset) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001777 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001778 goto out;
1779 }
1780 }
1781 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
1783 /* If users can be writing to this page using arbitrary
1784 * virtual addresses, take care about potential aliasing
1785 * before reading the page on the kernel side.
1786 */
1787 if (mapping_writably_mapped(mapping))
1788 flush_dcache_page(page);
1789
1790 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001791 * When a sequential read accesses a page several times,
1792 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 */
Jan Karaec0f1632007-05-06 14:49:25 -07001794 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 mark_page_accessed(page);
1796 prev_index = index;
1797
1798 /*
1799 * Ok, we have the page, and it's up-to-date, so
1800 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 */
Al Viro6e58e792014-02-03 17:07:03 -05001802
1803 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001805 index += offset >> PAGE_SHIFT;
1806 offset &= ~PAGE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001807 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001809 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001810 written += ret;
1811 if (!iov_iter_count(iter))
1812 goto out;
1813 if (ret < nr) {
1814 error = -EFAULT;
1815 goto out;
1816 }
1817 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
1819page_not_up_to_date:
1820 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001821 error = lock_page_killable(page);
1822 if (unlikely(error))
1823 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001825page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001826 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 if (!page->mapping) {
1828 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001829 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 continue;
1831 }
1832
1833 /* Did somebody else fill it already? */
1834 if (PageUptodate(page)) {
1835 unlock_page(page);
1836 goto page_ok;
1837 }
1838
1839readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001840 /*
1841 * A previous I/O error may have been due to temporary
1842 * failures, eg. multipath errors.
1843 * PG_error will be set again if readpage fails.
1844 */
1845 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 /* Start the actual read. The read will unlock the page. */
1847 error = mapping->a_ops->readpage(filp, page);
1848
Zach Brown994fc28c2005-12-15 14:28:17 -08001849 if (unlikely(error)) {
1850 if (error == AOP_TRUNCATED_PAGE) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001851 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001852 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001853 goto find_page;
1854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
1858 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001859 error = lock_page_killable(page);
1860 if (unlikely(error))
1861 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 if (!PageUptodate(page)) {
1863 if (page->mapping == NULL) {
1864 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001865 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 */
1867 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001868 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 goto find_page;
1870 }
1871 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001872 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001873 error = -EIO;
1874 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 }
1876 unlock_page(page);
1877 }
1878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 goto page_ok;
1880
1881readpage_error:
1882 /* UHHUH! A synchronous read error occurred. Report it */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001883 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 goto out;
1885
1886no_cached_page:
1887 /*
1888 * Ok, it wasn't cached, so we need to create a new
1889 * page..
1890 */
Nick Piggineb2be182007-10-16 01:24:57 -07001891 page = page_cache_alloc_cold(mapping);
1892 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001893 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001894 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001896 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08001897 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 if (error) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001899 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001900 if (error == -EEXIST) {
1901 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 goto out;
1905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 goto readpage;
1907 }
1908
1909out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001910 ra->prev_pos = prev_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001911 ra->prev_pos <<= PAGE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001912 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001914 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001915 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001916 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917}
1918
Randy Dunlap485bb992006-06-23 02:03:49 -07001919/**
Al Viro6abd2322014-04-04 14:20:57 -04001920 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001921 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001922 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001923 *
Al Viro6abd2322014-04-04 14:20:57 -04001924 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 * that can use the page cache directly.
1926 */
1927ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001928generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929{
Al Viroed978a82014-03-05 22:53:04 -05001930 struct file *file = iocb->ki_filp;
1931 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07001932 size_t count = iov_iter_count(iter);
1933
1934 if (!count)
1935 goto out; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
Al Viro2ba48ce2015-04-09 13:52:01 -04001937 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001938 struct address_space *mapping = file->f_mapping;
1939 struct inode *inode = mapping->host;
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001940 struct iov_iter data = *iter;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001941 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 size = i_size_read(inode);
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001944 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
1945 iocb->ki_pos + count - 1);
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001946 if (retval < 0)
1947 goto out;
Al Viroed978a82014-03-05 22:53:04 -05001948
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001949 file_accessed(file);
1950
1951 retval = mapping->a_ops->direct_IO(iocb, &data);
Al Viroc3a69022016-10-10 13:26:27 -04001952 if (retval >= 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001953 iocb->ki_pos += retval;
Al Viroed978a82014-03-05 22:53:04 -05001954 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001955 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001956
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001957 /*
1958 * Btrfs can have a short DIO read if we encounter
1959 * compressed extents, so if there was an error, or if
1960 * we've already read everything we wanted to, or if
1961 * there was a short read because we hit EOF, go ahead
1962 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001963 * the rest of the read. Buffered reads will not work for
1964 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001965 */
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001966 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001967 IS_DAX(inode))
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001968 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 }
1970
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001971 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972out:
1973 return retval;
1974}
Al Viroed978a82014-03-05 22:53:04 -05001975EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001978/**
1979 * page_cache_read - adds requested page to the page cache if not already there
1980 * @file: file to read
1981 * @offset: page index
Randy Dunlap62eb3202016-02-11 16:12:58 -08001982 * @gfp_mask: memory allocation flags
Randy Dunlap485bb992006-06-23 02:03:49 -07001983 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 * This adds the requested page to the page cache if it isn't already there,
1985 * and schedules an I/O to read in its contents from disk.
1986 */
Michal Hockoc20cd452016-01-14 15:20:12 -08001987static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988{
1989 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001990 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001991 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
Zach Brown994fc28c2005-12-15 14:28:17 -08001993 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08001994 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08001995 if (!page)
1996 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -07001998 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
Zach Brown994fc28c2005-12-15 14:28:17 -08001999 if (ret == 0)
2000 ret = mapping->a_ops->readpage(file, page);
2001 else if (ret == -EEXIST)
2002 ret = 0; /* losing race to add is OK */
2003
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002004 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
Zach Brown994fc28c2005-12-15 14:28:17 -08002006 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07002007
Zach Brown994fc28c2005-12-15 14:28:17 -08002008 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009}
2010
2011#define MMAP_LOTSAMISS (100)
2012
Linus Torvaldsef00e082009-06-16 15:31:25 -07002013/*
2014 * Synchronous readahead happens when we don't even find
2015 * a page in the page cache at all.
2016 */
2017static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2018 struct file_ra_state *ra,
2019 struct file *file,
2020 pgoff_t offset)
2021{
Linus Torvaldsef00e082009-06-16 15:31:25 -07002022 struct address_space *mapping = file->f_mapping;
2023
2024 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002025 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002026 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002027 if (!ra->ra_pages)
2028 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002029
Joe Perches64363aa2013-07-08 16:00:18 -07002030 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07002031 page_cache_sync_readahead(mapping, ra, file, offset,
2032 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002033 return;
2034 }
2035
Andi Kleen207d04b2011-05-24 17:12:29 -07002036 /* Avoid banging the cache line if not needed */
2037 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002038 ra->mmap_miss++;
2039
2040 /*
2041 * Do we miss much more than hit in this file? If so,
2042 * stop bothering with read-ahead. It will only hurt.
2043 */
2044 if (ra->mmap_miss > MMAP_LOTSAMISS)
2045 return;
2046
Wu Fengguangd30a1102009-06-16 15:31:30 -07002047 /*
2048 * mmap read-around
2049 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08002050 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2051 ra->size = ra->ra_pages;
2052 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002053 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002054}
2055
2056/*
2057 * Asynchronous readahead happens when we find the page and PG_readahead,
2058 * so we want to possibly extend the readahead further..
2059 */
2060static void do_async_mmap_readahead(struct vm_area_struct *vma,
2061 struct file_ra_state *ra,
2062 struct file *file,
2063 struct page *page,
2064 pgoff_t offset)
2065{
2066 struct address_space *mapping = file->f_mapping;
2067
2068 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002069 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002070 return;
2071 if (ra->mmap_miss > 0)
2072 ra->mmap_miss--;
2073 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07002074 page_cache_async_readahead(mapping, ra, file,
2075 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002076}
2077
Randy Dunlap485bb992006-06-23 02:03:49 -07002078/**
Nick Piggin54cb8822007-07-19 01:46:59 -07002079 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07002080 * @vma: vma in which the fault was taken
2081 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07002082 *
Nick Piggin54cb8822007-07-19 01:46:59 -07002083 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 * mapped memory region to read in file data during a page fault.
2085 *
2086 * The goto's are kind of ugly, but this streamlines the normal case of having
2087 * it in the page cache, and handles the special cases reasonably without
2088 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002089 *
2090 * vma->vm_mm->mmap_sem must be held on entry.
2091 *
2092 * If our return value has VM_FAULT_RETRY set, it's because
2093 * lock_page_or_retry() returned 0.
2094 * The mmap_sem has usually been released in this case.
2095 * See __lock_page_or_retry() for the exception.
2096 *
2097 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2098 * has not been released.
2099 *
2100 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002102int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
2104 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07002105 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 struct address_space *mapping = file->f_mapping;
2107 struct file_ra_state *ra = &file->f_ra;
2108 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002109 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002111 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07002112 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002114 size = round_up(i_size_read(inode), PAGE_SIZE);
2115 if (offset >= size >> PAGE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07002116 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002119 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002121 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002122 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002124 * We found the page, so try async readahead before
2125 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002127 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002128 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002129 /* No page in the page cache at all */
2130 do_sync_mmap_readahead(vma, ra, file, offset);
2131 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002132 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002133 ret = VM_FAULT_MAJOR;
2134retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07002135 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 if (!page)
2137 goto no_cached_page;
2138 }
2139
Michel Lespinassed88c0922010-11-02 13:05:18 -07002140 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002141 put_page(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002142 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07002143 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07002144
2145 /* Did it get truncated? */
2146 if (unlikely(page->mapping != mapping)) {
2147 unlock_page(page);
2148 put_page(page);
2149 goto retry_find;
2150 }
Sasha Levin309381fea2014-01-23 15:52:54 -08002151 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002152
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002154 * We have a locked page in the page cache, now we need to check
2155 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 */
Nick Piggind00806b2007-07-19 01:46:57 -07002157 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 goto page_not_uptodate;
2159
Linus Torvaldsef00e082009-06-16 15:31:25 -07002160 /*
2161 * Found the page and have a reference on it.
2162 * We must recheck i_size under page lock.
2163 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002164 size = round_up(i_size_read(inode), PAGE_SIZE);
2165 if (unlikely(offset >= size >> PAGE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002166 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002167 put_page(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002168 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002169 }
2170
Nick Piggind0217ac2007-07-19 01:47:03 -07002171 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002172 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174no_cached_page:
2175 /*
2176 * We're only likely to ever get here if MADV_RANDOM is in
2177 * effect.
2178 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002179 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180
2181 /*
2182 * The page we want has now been added to the page cache.
2183 * In the unlikely event that someone removed it in the
2184 * meantime, we'll just come back here and read it again.
2185 */
2186 if (error >= 0)
2187 goto retry_find;
2188
2189 /*
2190 * An error return from page_cache_read can result if the
2191 * system is low on memory, or a problem occurs while trying
2192 * to schedule I/O.
2193 */
2194 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002195 return VM_FAULT_OOM;
2196 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
2198page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 /*
2200 * Umm, take care of errors if the page isn't up-to-date.
2201 * Try to re-read it _once_. We do this synchronously,
2202 * because there really aren't any performance issues here
2203 * and we need to check for errors.
2204 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002206 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002207 if (!error) {
2208 wait_on_page_locked(page);
2209 if (!PageUptodate(page))
2210 error = -EIO;
2211 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002212 put_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002213
2214 if (!error || error == AOP_TRUNCATED_PAGE)
2215 goto retry_find;
2216
2217 /* Things didn't work out. Return zero to tell the mm layer so. */
2218 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002219 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002220}
2221EXPORT_SYMBOL(filemap_fault);
2222
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002223void filemap_map_pages(struct fault_env *fe,
2224 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002225{
2226 struct radix_tree_iter iter;
2227 void **slot;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002228 struct file *file = fe->vma->vm_file;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002229 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002230 pgoff_t last_pgoff = start_pgoff;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002231 loff_t size;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002232 struct page *head, *page;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002233
2234 rcu_read_lock();
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002235 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2236 start_pgoff) {
2237 if (iter.index > end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002238 break;
2239repeat:
2240 page = radix_tree_deref_slot(slot);
2241 if (unlikely(!page))
2242 goto next;
2243 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002244 if (radix_tree_deref_retry(page)) {
2245 slot = radix_tree_iter_retry(&iter);
2246 continue;
2247 }
2248 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002249 }
2250
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002251 head = compound_head(page);
2252 if (!page_cache_get_speculative(head))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002253 goto repeat;
2254
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002255 /* The page was split under us? */
2256 if (compound_head(page) != head) {
2257 put_page(head);
2258 goto repeat;
2259 }
2260
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002261 /* Has the page moved? */
2262 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002263 put_page(head);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002264 goto repeat;
2265 }
2266
2267 if (!PageUptodate(page) ||
2268 PageReadahead(page) ||
2269 PageHWPoison(page))
2270 goto skip;
2271 if (!trylock_page(page))
2272 goto skip;
2273
2274 if (page->mapping != mapping || !PageUptodate(page))
2275 goto unlock;
2276
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002277 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2278 if (page->index >= size >> PAGE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002279 goto unlock;
2280
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002281 if (file->f_ra.mmap_miss > 0)
2282 file->f_ra.mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002283
2284 fe->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2285 if (fe->pte)
2286 fe->pte += iter.index - last_pgoff;
2287 last_pgoff = iter.index;
2288 if (alloc_set_pte(fe, NULL, page))
2289 goto unlock;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002290 unlock_page(page);
2291 goto next;
2292unlock:
2293 unlock_page(page);
2294skip:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002295 put_page(page);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002296next:
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002297 /* Huge page is mapped? No need to proceed. */
2298 if (pmd_trans_huge(*fe->pmd))
2299 break;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002300 if (iter.index == end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002301 break;
2302 }
2303 rcu_read_unlock();
2304}
2305EXPORT_SYMBOL(filemap_map_pages);
2306
Jan Kara4fcf1c62012-06-12 16:20:29 +02002307int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2308{
2309 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002310 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002311 int ret = VM_FAULT_LOCKED;
2312
Jan Kara14da9202012-06-12 16:20:37 +02002313 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002314 file_update_time(vma->vm_file);
2315 lock_page(page);
2316 if (page->mapping != inode->i_mapping) {
2317 unlock_page(page);
2318 ret = VM_FAULT_NOPAGE;
2319 goto out;
2320 }
Jan Kara14da9202012-06-12 16:20:37 +02002321 /*
2322 * We mark the page dirty already here so that when freeze is in
2323 * progress, we are guaranteed that writeback during freezing will
2324 * see the dirty page and writeprotect it again.
2325 */
2326 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002327 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002328out:
Jan Kara14da9202012-06-12 16:20:37 +02002329 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002330 return ret;
2331}
2332EXPORT_SYMBOL(filemap_page_mkwrite);
2333
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002334const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002335 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002336 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002337 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338};
2339
2340/* This is used for a general mmap of a disk file */
2341
2342int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2343{
2344 struct address_space *mapping = file->f_mapping;
2345
2346 if (!mapping->a_ops->readpage)
2347 return -ENOEXEC;
2348 file_accessed(file);
2349 vma->vm_ops = &generic_file_vm_ops;
2350 return 0;
2351}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
2353/*
2354 * This is for filesystems which do not implement ->writepage.
2355 */
2356int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2357{
2358 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2359 return -EINVAL;
2360 return generic_file_mmap(file, vma);
2361}
2362#else
2363int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2364{
2365 return -ENOSYS;
2366}
2367int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2368{
2369 return -ENOSYS;
2370}
2371#endif /* CONFIG_MMU */
2372
2373EXPORT_SYMBOL(generic_file_mmap);
2374EXPORT_SYMBOL(generic_file_readonly_mmap);
2375
Sasha Levin67f9fd92014-04-03 14:48:18 -07002376static struct page *wait_on_page_read(struct page *page)
2377{
2378 if (!IS_ERR(page)) {
2379 wait_on_page_locked(page);
2380 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002381 put_page(page);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002382 page = ERR_PTR(-EIO);
2383 }
2384 }
2385 return page;
2386}
2387
Mel Gorman32b63522016-03-15 14:55:36 -07002388static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002389 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002390 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002391 void *data,
2392 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393{
Nick Piggineb2be182007-10-16 01:24:57 -07002394 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 int err;
2396repeat:
2397 page = find_get_page(mapping, index);
2398 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002399 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002400 if (!page)
2401 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002402 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002403 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002404 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002405 if (err == -EEXIST)
2406 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 return ERR_PTR(err);
2409 }
Mel Gorman32b63522016-03-15 14:55:36 -07002410
2411filler:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 err = filler(data, page);
2413 if (err < 0) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002414 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002415 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 }
Mel Gorman32b63522016-03-15 14:55:36 -07002417
2418 page = wait_on_page_read(page);
2419 if (IS_ERR(page))
2420 return page;
2421 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 if (PageUptodate(page))
2424 goto out;
2425
Mel Gormanebded022016-03-15 14:55:39 -07002426 /*
2427 * Page is not up to date and may be locked due one of the following
2428 * case a: Page is being filled and the page lock is held
2429 * case b: Read/write error clearing the page uptodate status
2430 * case c: Truncation in progress (page locked)
2431 * case d: Reclaim in progress
2432 *
2433 * Case a, the page will be up to date when the page is unlocked.
2434 * There is no need to serialise on the page lock here as the page
2435 * is pinned so the lock gives no additional protection. Even if the
2436 * the page is truncated, the data is still valid if PageUptodate as
2437 * it's a race vs truncate race.
2438 * Case b, the page will not be up to date
2439 * Case c, the page may be truncated but in itself, the data may still
2440 * be valid after IO completes as it's a read vs truncate race. The
2441 * operation must restart if the page is not uptodate on unlock but
2442 * otherwise serialising on page lock to stabilise the mapping gives
2443 * no additional guarantees to the caller as the page lock is
2444 * released before return.
2445 * Case d, similar to truncation. If reclaim holds the page lock, it
2446 * will be a race with remove_mapping that determines if the mapping
2447 * is valid on unlock but otherwise the data is valid and there is
2448 * no need to serialise with page lock.
2449 *
2450 * As the page lock gives no additional guarantee, we optimistically
2451 * wait on the page to be unlocked and check if it's up to date and
2452 * use the page if it is. Otherwise, the page lock is required to
2453 * distinguish between the different cases. The motivation is that we
2454 * avoid spurious serialisations and wakeups when multiple processes
2455 * wait on the same page for IO to complete.
2456 */
2457 wait_on_page_locked(page);
2458 if (PageUptodate(page))
2459 goto out;
2460
2461 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07002463
2464 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (!page->mapping) {
2466 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002467 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002468 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 }
Mel Gormanebded022016-03-15 14:55:39 -07002470
2471 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 if (PageUptodate(page)) {
2473 unlock_page(page);
2474 goto out;
2475 }
Mel Gorman32b63522016-03-15 14:55:36 -07002476 goto filler;
2477
David Howellsc855ff32007-05-09 13:42:20 +01002478out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002479 mark_page_accessed(page);
2480 return page;
2481}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002482
2483/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002484 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002485 * @mapping: the page's address_space
2486 * @index: the page index
2487 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002488 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002489 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002490 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002491 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002492 *
2493 * If the page does not get brought uptodate, return -EIO.
2494 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002495struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002496 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002497 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002498 void *data)
2499{
2500 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2501}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002502EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002503
2504/**
2505 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2506 * @mapping: the page's address_space
2507 * @index: the page index
2508 * @gfp: the page allocator flags to use if allocating
2509 *
2510 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002511 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002512 *
2513 * If the page does not get brought uptodate, return -EIO.
2514 */
2515struct page *read_cache_page_gfp(struct address_space *mapping,
2516 pgoff_t index,
2517 gfp_t gfp)
2518{
2519 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2520
Sasha Levin67f9fd92014-04-03 14:48:18 -07002521 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002522}
2523EXPORT_SYMBOL(read_cache_page_gfp);
2524
Nick Piggin2f718ff2007-10-16 01:24:59 -07002525/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 * Performs necessary checks before doing a write
2527 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002528 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 * Returns appropriate error code that caller should return or
2530 * zero in case that write should be allowed.
2531 */
Al Viro3309dd02015-04-09 12:55:47 -04002532inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533{
Al Viro3309dd02015-04-09 12:55:47 -04002534 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002536 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002537 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Al Viro3309dd02015-04-09 12:55:47 -04002539 if (!iov_iter_count(from))
2540 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
Al Viro0fa6b002015-04-04 04:05:48 -04002542 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002543 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002544 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Al Viro3309dd02015-04-09 12:55:47 -04002546 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
Al Viro0fa6b002015-04-04 04:05:48 -04002548 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002549 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002550 send_sig(SIGXFSZ, current, 0);
2551 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 }
Al Viro3309dd02015-04-09 12:55:47 -04002553 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 }
2555
2556 /*
2557 * LFS rule
2558 */
Al Viro3309dd02015-04-09 12:55:47 -04002559 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002561 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002563 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 }
2565
2566 /*
2567 * Are we about to exceed the fs block limit ?
2568 *
2569 * If we have written data it becomes a short write. If we have
2570 * exceeded without writing data we send a signal and return EFBIG.
2571 * Linus frestrict idea will clean these up nicely..
2572 */
Al Viro3309dd02015-04-09 12:55:47 -04002573 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2574 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
Al Viro3309dd02015-04-09 12:55:47 -04002576 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2577 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578}
2579EXPORT_SYMBOL(generic_write_checks);
2580
Nick Pigginafddba42007-10-16 01:25:01 -07002581int pagecache_write_begin(struct file *file, struct address_space *mapping,
2582 loff_t pos, unsigned len, unsigned flags,
2583 struct page **pagep, void **fsdata)
2584{
2585 const struct address_space_operations *aops = mapping->a_ops;
2586
Nick Piggin4e02ed42008-10-29 14:00:55 -07002587 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002588 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002589}
2590EXPORT_SYMBOL(pagecache_write_begin);
2591
2592int pagecache_write_end(struct file *file, struct address_space *mapping,
2593 loff_t pos, unsigned len, unsigned copied,
2594 struct page *page, void *fsdata)
2595{
2596 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002597
Nick Piggin4e02ed42008-10-29 14:00:55 -07002598 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002599}
2600EXPORT_SYMBOL(pagecache_write_end);
2601
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002603generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604{
2605 struct file *file = iocb->ki_filp;
2606 struct address_space *mapping = file->f_mapping;
2607 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002608 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002610 size_t write_len;
2611 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002612 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Al Viro0c949332014-03-22 06:51:37 -04002614 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002615 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002616
Nick Piggin48b47c52009-01-06 14:40:22 -08002617 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002618 if (written)
2619 goto out;
2620
2621 /*
2622 * After a write we want buffered reads to be sure to go to disk to get
2623 * the new data. We invalidate clean cached page from the region we're
2624 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002625 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002626 */
2627 if (mapping->nrpages) {
2628 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002629 pos >> PAGE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002630 /*
2631 * If a page can not be invalidated, return 0 to fall back
2632 * to buffered write.
2633 */
2634 if (written) {
2635 if (written == -EBUSY)
2636 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002637 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002638 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002639 }
2640
Al Viro26978b82014-03-10 14:08:45 -04002641 data = *from;
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07002642 written = mapping->a_ops->direct_IO(iocb, &data);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002643
2644 /*
2645 * Finally, try again to invalidate clean pages which might have been
2646 * cached by non-direct readahead, or faulted in by get_user_pages()
2647 * if the source of the write was an mmap'ed region of the file
2648 * we're writing. Either one is a pretty crazy thing to do,
2649 * so we don't support it 100%. If this invalidation
2650 * fails, tough, the write still worked...
2651 */
2652 if (mapping->nrpages) {
2653 invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002654 pos >> PAGE_SHIFT, end);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002655 }
2656
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002658 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002659 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002660 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2661 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 mark_inode_dirty(inode);
2663 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002664 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002666out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 return written;
2668}
2669EXPORT_SYMBOL(generic_file_direct_write);
2670
Nick Piggineb2be182007-10-16 01:24:57 -07002671/*
2672 * Find or create a page at the given pagecache position. Return the locked
2673 * page. This function is specifically for buffered writes.
2674 */
Nick Piggin54566b22009-01-04 12:00:53 -08002675struct page *grab_cache_page_write_begin(struct address_space *mapping,
2676 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002677{
Nick Piggineb2be182007-10-16 01:24:57 -07002678 struct page *page;
Johannes Weinerbbddabe2016-05-20 16:56:28 -07002679 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002680
Nick Piggin54566b22009-01-04 12:00:53 -08002681 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002682 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002683
Mel Gorman2457aec2014-06-04 16:10:31 -07002684 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002685 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002686 if (page)
2687 wait_for_stable_page(page);
2688
Nick Piggineb2be182007-10-16 01:24:57 -07002689 return page;
2690}
Nick Piggin54566b22009-01-04 12:00:53 -08002691EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002692
Al Viro3b93f912014-02-11 21:34:08 -05002693ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002694 struct iov_iter *i, loff_t pos)
2695{
2696 struct address_space *mapping = file->f_mapping;
2697 const struct address_space_operations *a_ops = mapping->a_ops;
2698 long status = 0;
2699 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002700 unsigned int flags = 0;
2701
2702 /*
2703 * Copies from kernel address space cannot fail (NFSD is a big user).
2704 */
Al Viro777eda22014-12-17 04:46:46 -05002705 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002706 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002707
2708 do {
2709 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002710 unsigned long offset; /* Offset into pagecache page */
2711 unsigned long bytes; /* Bytes to write to page */
2712 size_t copied; /* Bytes copied from user */
2713 void *fsdata;
2714
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002715 offset = (pos & (PAGE_SIZE - 1));
2716 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002717 iov_iter_count(i));
2718
2719again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002720 /*
2721 * Bring in the user page that we will copy from _first_.
2722 * Otherwise there's a nasty deadlock on copying from the
2723 * same page as we're writing to, without it being marked
2724 * up-to-date.
2725 *
2726 * Not only is this an optimisation, but it is also required
2727 * to check that the address is actually valid, when atomic
2728 * usercopies are used, below.
2729 */
2730 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2731 status = -EFAULT;
2732 break;
2733 }
2734
Jan Kara296291c2015-10-22 13:32:21 -07002735 if (fatal_signal_pending(current)) {
2736 status = -EINTR;
2737 break;
2738 }
2739
Nick Piggin674b8922007-10-16 01:25:03 -07002740 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002741 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002742 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002743 break;
2744
anfei zhou931e80e2010-02-02 13:44:02 -08002745 if (mapping_writably_mapped(mapping))
2746 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002747
Nick Pigginafddba42007-10-16 01:25:01 -07002748 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002749 flush_dcache_page(page);
2750
2751 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2752 page, fsdata);
2753 if (unlikely(status < 0))
2754 break;
2755 copied = status;
2756
2757 cond_resched();
2758
Nick Piggin124d3b72008-02-02 15:01:17 +01002759 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002760 if (unlikely(copied == 0)) {
2761 /*
2762 * If we were unable to copy any data at all, we must
2763 * fall back to a single segment length write.
2764 *
2765 * If we didn't fallback here, we could livelock
2766 * because not all segments in the iov can be copied at
2767 * once without a pagefault.
2768 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002769 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002770 iov_iter_single_seg_count(i));
2771 goto again;
2772 }
Nick Pigginafddba42007-10-16 01:25:01 -07002773 pos += copied;
2774 written += copied;
2775
2776 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002777 } while (iov_iter_count(i));
2778
2779 return written ? written : status;
2780}
Al Viro3b93f912014-02-11 21:34:08 -05002781EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782
Jan Karae4dd9de2009-08-17 18:10:06 +02002783/**
Al Viro81742022014-04-03 03:17:43 -04002784 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002785 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002786 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002787 *
2788 * This function does all the work needed for actually writing data to a
2789 * file. It does all basic checks, removes SUID from the file, updates
2790 * modification times and calls proper subroutines depending on whether we
2791 * do direct IO or a standard buffered write.
2792 *
2793 * It expects i_mutex to be grabbed unless we work on a block device or similar
2794 * object which does not need locking at all.
2795 *
2796 * This function does *not* take care of syncing data in case of O_SYNC write.
2797 * A caller has to handle it. This is mainly due to the fact that we want to
2798 * avoid syncing under i_mutex.
2799 */
Al Viro81742022014-04-03 03:17:43 -04002800ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801{
2802 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002803 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002805 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002807 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002810 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002811 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 if (err)
2813 goto out;
2814
Josef Bacikc3b2da32012-03-26 09:59:21 -04002815 err = file_update_time(file);
2816 if (err)
2817 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
Al Viro2ba48ce2015-04-09 13:52:01 -04002819 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002820 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002821
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002822 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002823 /*
2824 * If the write stopped short of completing, fall back to
2825 * buffered writes. Some filesystems do this for writes to
2826 * holes, for example. For DAX files, a buffered write will
2827 * not succeed (even if it did, DAX does not handle dirty
2828 * page-cache pages correctly).
2829 */
Al Viro0b8def92015-04-07 10:22:53 -04002830 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002832
Al Viro0b8def92015-04-07 10:22:53 -04002833 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002834 /*
Al Viro3b93f912014-02-11 21:34:08 -05002835 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002836 * then we want to return the number of bytes which were
2837 * direct-written, or the error code if that was zero. Note
2838 * that this differs from normal direct-io semantics, which
2839 * will return -EFOO even if some bytes were written.
2840 */
Al Viro60bb4522014-08-08 12:39:16 -04002841 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002842 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002843 goto out;
2844 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002845 /*
2846 * We need to ensure that the page cache pages are written to
2847 * disk and invalidated to preserve the expected O_DIRECT
2848 * semantics.
2849 */
Al Viro3b93f912014-02-11 21:34:08 -05002850 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002851 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002852 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002853 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002854 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002855 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002856 pos >> PAGE_SHIFT,
2857 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002858 } else {
2859 /*
2860 * We don't know how much we wrote, so just return
2861 * the number of bytes which were direct-written
2862 */
2863 }
2864 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002865 written = generic_perform_write(file, from, iocb->ki_pos);
2866 if (likely(written > 0))
2867 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869out:
2870 current->backing_dev_info = NULL;
2871 return written ? written : err;
2872}
Al Viro81742022014-04-03 03:17:43 -04002873EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874
Jan Karae4dd9de2009-08-17 18:10:06 +02002875/**
Al Viro81742022014-04-03 03:17:43 -04002876 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002877 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002878 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002879 *
Al Viro81742022014-04-03 03:17:43 -04002880 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002881 * filesystems. It takes care of syncing the file in case of O_SYNC file
2882 * and acquires i_mutex as needed.
2883 */
Al Viro81742022014-04-03 03:17:43 -04002884ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885{
2886 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002887 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
Al Viro59551022016-01-22 15:40:57 -05002890 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04002891 ret = generic_write_checks(iocb, from);
2892 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002893 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05002894 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
Christoph Hellwige2592212016-04-07 08:52:01 -07002896 if (ret > 0)
2897 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 return ret;
2899}
Al Viro81742022014-04-03 03:17:43 -04002900EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901
David Howellscf9a2ae2006-08-29 19:05:54 +01002902/**
2903 * try_to_release_page() - release old fs-specific metadata on a page
2904 *
2905 * @page: the page which the kernel is trying to free
2906 * @gfp_mask: memory allocation flags (and I/O mode)
2907 *
2908 * The address_space is to try to release any data against the page
2909 * (presumably at page->private). If the release was successful, return `1'.
2910 * Otherwise return zero.
2911 *
David Howells266cf652009-04-03 16:42:36 +01002912 * This may also be called if PG_fscache is set on a page, indicating that the
2913 * page is known to the local caching routines.
2914 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002915 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08002916 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002917 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002918 */
2919int try_to_release_page(struct page *page, gfp_t gfp_mask)
2920{
2921 struct address_space * const mapping = page->mapping;
2922
2923 BUG_ON(!PageLocked(page));
2924 if (PageWriteback(page))
2925 return 0;
2926
2927 if (mapping && mapping->a_ops->releasepage)
2928 return mapping->a_ops->releasepage(page, gfp_mask);
2929 return try_to_free_buffers(page);
2930}
2931
2932EXPORT_SYMBOL(try_to_release_page);