blob: 6bf5e42d560a46eea8e4916bd017d855033b8d65 [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>
14#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070015#include <linux/uaccess.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080016#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090018#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mm.h>
20#include <linux/swap.h>
21#include <linux/mman.h>
22#include <linux/pagemap.h>
23#include <linux/file.h>
24#include <linux/uio.h>
25#include <linux/hash.h>
26#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070027#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080031#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070032#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Johannes Weiner00501b52014-08-08 14:19:20 -070033#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080034#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060035#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070036#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070039#define CREATE_TRACE_POINTS
40#include <trace/events/filemap.h>
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * FIXME: remove all knowledge of the buffer layer from the core VM
44 */
Jan Kara148f9482009-08-17 19:52:36 +020045#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <asm/mman.h>
48
49/*
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * though.
52 *
53 * Shared mappings now work. 15.8.1995 Bruno.
54 *
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 *
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
59 */
60
61/*
62 * Lock ordering:
63 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080064 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070066 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080069 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080070 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 *
72 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080073 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070074 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 *
77 * ->mmap_sem
78 * ->lock_page (access_process_vm)
79 *
Al Viroccad2362014-02-11 22:36:48 -050080 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070081 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060083 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110084 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * ->mapping->tree_lock (__sync_single_inode)
86 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080087 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * ->anon_vma.lock (vma_adjust)
89 *
90 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070091 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070094 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080098 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600103 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100104 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
106 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800107 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700108 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 */
110
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700111static void page_cache_tree_delete(struct address_space *mapping,
112 struct page *page, void *shadow)
113{
Johannes Weiner449dd692014-04-03 14:47:56 -0700114 struct radix_tree_node *node;
115 unsigned long index;
116 unsigned int offset;
117 unsigned int tag;
118 void **slot;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700119
Johannes Weiner449dd692014-04-03 14:47:56 -0700120 VM_BUG_ON(!PageLocked(page));
121
122 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
123
124 if (shadow) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700125 mapping->nrshadows++;
126 /*
127 * Make sure the nrshadows update is committed before
128 * the nrpages update so that final truncate racing
129 * with reclaim does not see both counters 0 at the
130 * same time and miss a shadow entry.
131 */
132 smp_wmb();
Johannes Weiner449dd692014-04-03 14:47:56 -0700133 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700134 mapping->nrpages--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700135
136 if (!node) {
137 /* Clear direct pointer tags in root node */
138 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
139 radix_tree_replace_slot(slot, shadow);
140 return;
141 }
142
143 /* Clear tree tags for the removed page */
144 index = page->index;
145 offset = index & RADIX_TREE_MAP_MASK;
146 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
147 if (test_bit(offset, node->tags[tag]))
148 radix_tree_tag_clear(&mapping->page_tree, index, tag);
149 }
150
151 /* Delete page, swap shadow entry */
152 radix_tree_replace_slot(slot, shadow);
153 workingset_node_pages_dec(node);
154 if (shadow)
155 workingset_node_shadows_inc(node);
156 else
157 if (__radix_tree_delete_node(&mapping->page_tree, node))
158 return;
159
160 /*
161 * Track node that only contains shadow entries.
162 *
163 * Avoid acquiring the list_lru lock if already tracked. The
164 * list_empty() test is safe as node->private_list is
165 * protected by mapping->tree_lock.
166 */
167 if (!workingset_node_pages(node) &&
168 list_empty(&node->private_list)) {
169 node->private_data = mapping;
170 list_lru_add(&workingset_shadow_nodes, &node->private_list);
171 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700172}
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174/*
Minchan Kime64a7822011-03-22 16:32:44 -0700175 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700177 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700179void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
181 struct address_space *mapping = page->mapping;
182
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700183 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600184 /*
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
188 */
189 if (PageUptodate(page) && PageMappedToDisk(page))
190 cleancache_put_page(page);
191 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400192 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600193
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700194 page_cache_tree_delete(mapping, page, shadow);
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700197 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700198
Christoph Lameter347ce432006-06-30 01:55:35 -0700199 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700200 if (PageSwapBacked(page))
201 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700202 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800203
204 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700205 * At this point page must be either written or cleaned by truncate.
206 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800207 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700208 * This fixes dirty accounting after removing the page entirely but
209 * leaves PageDirty set: it has no effect for truncated page and
210 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800211 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700212 if (WARN_ON_ONCE(PageDirty(page)))
213 account_page_cleaned(page, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
Minchan Kim702cfbf2011-03-22 16:32:43 -0700216/**
217 * delete_from_page_cache - delete page from page cache
218 * @page: the page which the kernel is trying to remove from page cache
219 *
220 * This must be called only on pages that have been verified to be in the page
221 * cache and locked. It will never put the page into the free list, the caller
222 * has a reference on the page.
223 */
224void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500227 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Matt Mackallcd7619d2005-05-01 08:59:01 -0700229 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Linus Torvalds6072d132010-12-01 13:35:19 -0500231 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700232 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700233 __delete_from_page_cache(page, NULL);
Nick Piggin19fd6232008-07-25 19:45:32 -0700234 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds6072d132010-12-01 13:35:19 -0500235
236 if (freepage)
237 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700238 page_cache_release(page);
239}
240EXPORT_SYMBOL(delete_from_page_cache);
241
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700242static int filemap_check_errors(struct address_space *mapping)
243{
244 int ret = 0;
245 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700246 if (test_bit(AS_ENOSPC, &mapping->flags) &&
247 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700248 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700249 if (test_bit(AS_EIO, &mapping->flags) &&
250 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700251 ret = -EIO;
252 return ret;
253}
254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700256 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700257 * @mapping: address space structure to write
258 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800259 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700260 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700262 * Start writeback against all of a mapping's dirty pages that lie
263 * within the byte offsets <start, end> inclusive.
264 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700266 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 * these two operations is that if a dirty page/buffer is encountered, it must
268 * be waited upon, and not just skipped over.
269 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800270int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
271 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 int ret;
274 struct writeback_control wbc = {
275 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800276 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700277 .range_start = start,
278 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 };
280
281 if (!mapping_cap_writeback_dirty(mapping))
282 return 0;
283
284 ret = do_writepages(mapping, &wbc);
285 return ret;
286}
287
288static inline int __filemap_fdatawrite(struct address_space *mapping,
289 int sync_mode)
290{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700291 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292}
293
294int filemap_fdatawrite(struct address_space *mapping)
295{
296 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
297}
298EXPORT_SYMBOL(filemap_fdatawrite);
299
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400300int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800301 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
304}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400305EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Randy Dunlap485bb992006-06-23 02:03:49 -0700307/**
308 * filemap_flush - mostly a non-blocking flush
309 * @mapping: target address_space
310 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 * This is a mostly non-blocking flush. Not suitable for data-integrity
312 * purposes - I/O may not be started against all dirty pages.
313 */
314int filemap_flush(struct address_space *mapping)
315{
316 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
317}
318EXPORT_SYMBOL(filemap_flush);
319
Randy Dunlap485bb992006-06-23 02:03:49 -0700320/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200321 * filemap_fdatawait_range - wait for writeback to complete
322 * @mapping: address space structure to wait for
323 * @start_byte: offset in bytes where the range starts
324 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700325 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200326 * Walk the list of under-writeback pages of the given address space
327 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200329int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
330 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200332 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
333 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 struct pagevec pvec;
335 int nr_pages;
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700336 int ret2, ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200338 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700339 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
341 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 while ((index <= end) &&
343 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
344 PAGECACHE_TAG_WRITEBACK,
345 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
346 unsigned i;
347
348 for (i = 0; i < nr_pages; i++) {
349 struct page *page = pvec.pages[i];
350
351 /* until radix tree lookup accepts end_index */
352 if (page->index > end)
353 continue;
354
355 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800356 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 ret = -EIO;
358 }
359 pagevec_release(&pvec);
360 cond_resched();
361 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700362out:
363 ret2 = filemap_check_errors(mapping);
364 if (!ret)
365 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 return ret;
368}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200369EXPORT_SYMBOL(filemap_fdatawait_range);
370
371/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700372 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700374 *
375 * Walk the list of under-writeback pages of the given address space
376 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 */
378int filemap_fdatawait(struct address_space *mapping)
379{
380 loff_t i_size = i_size_read(mapping->host);
381
382 if (i_size == 0)
383 return 0;
384
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200385 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
387EXPORT_SYMBOL(filemap_fdatawait);
388
389int filemap_write_and_wait(struct address_space *mapping)
390{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800391 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
393 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800394 err = filemap_fdatawrite(mapping);
395 /*
396 * Even if the above returned error, the pages may be
397 * written partially (e.g. -ENOSPC), so we wait for it.
398 * But the -EIO is special case, it may indicate the worst
399 * thing (e.g. bug) happened, so we avoid waiting for it.
400 */
401 if (err != -EIO) {
402 int err2 = filemap_fdatawait(mapping);
403 if (!err)
404 err = err2;
405 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700406 } else {
407 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800409 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800411EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Randy Dunlap485bb992006-06-23 02:03:49 -0700413/**
414 * filemap_write_and_wait_range - write out & wait on a file range
415 * @mapping: the address_space for the pages
416 * @lstart: offset in bytes where the range starts
417 * @lend: offset in bytes where the range ends (inclusive)
418 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800419 * Write out and wait upon file offsets lstart->lend, inclusive.
420 *
421 * Note that `lend' is inclusive (describes the last byte to be written) so
422 * that this function can be used to write to the very end-of-file (end = -1).
423 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424int filemap_write_and_wait_range(struct address_space *mapping,
425 loff_t lstart, loff_t lend)
426{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800427 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800430 err = __filemap_fdatawrite_range(mapping, lstart, lend,
431 WB_SYNC_ALL);
432 /* See comment of filemap_write_and_wait() */
433 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200434 int err2 = filemap_fdatawait_range(mapping,
435 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800436 if (!err)
437 err = err2;
438 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700439 } else {
440 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800442 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
Chris Masonf6995582009-04-15 13:22:37 -0400444EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Randy Dunlap485bb992006-06-23 02:03:49 -0700446/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700447 * replace_page_cache_page - replace a pagecache page with a new one
448 * @old: page to be replaced
449 * @new: page to replace with
450 * @gfp_mask: allocation mode
451 *
452 * This function replaces a page in the pagecache with a new one. On
453 * success it acquires the pagecache reference for the new page and
454 * drops it for the old page. Both the old and new pages must be
455 * locked. This function does not add the new page to the LRU, the
456 * caller must do that.
457 *
458 * The remove + add is atomic. The only way this function can fail is
459 * memory allocation failure.
460 */
461int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
462{
463 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700464
Sasha Levin309381fea2014-01-23 15:52:54 -0800465 VM_BUG_ON_PAGE(!PageLocked(old), old);
466 VM_BUG_ON_PAGE(!PageLocked(new), new);
467 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700468
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700469 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
470 if (!error) {
471 struct address_space *mapping = old->mapping;
472 void (*freepage)(struct page *);
473
474 pgoff_t offset = old->index;
475 freepage = mapping->a_ops->freepage;
476
477 page_cache_get(new);
478 new->mapping = mapping;
479 new->index = offset;
480
481 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700482 __delete_from_page_cache(old, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700483 error = radix_tree_insert(&mapping->page_tree, offset, new);
484 BUG_ON(error);
485 mapping->nrpages++;
486 __inc_zone_page_state(new, NR_FILE_PAGES);
487 if (PageSwapBacked(new))
488 __inc_zone_page_state(new, NR_SHMEM);
489 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700490 mem_cgroup_migrate(old, new, true);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700491 radix_tree_preload_end();
492 if (freepage)
493 freepage(old);
494 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700495 }
496
497 return error;
498}
499EXPORT_SYMBOL_GPL(replace_page_cache_page);
500
Johannes Weiner0cd61442014-04-03 14:47:46 -0700501static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700502 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700503{
Johannes Weiner449dd692014-04-03 14:47:56 -0700504 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700505 void **slot;
506 int error;
507
Johannes Weiner449dd692014-04-03 14:47:56 -0700508 error = __radix_tree_create(&mapping->page_tree, page->index,
509 &node, &slot);
510 if (error)
511 return error;
512 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700513 void *p;
514
515 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
516 if (!radix_tree_exceptional_entry(p))
517 return -EEXIST;
Johannes Weinera5289102014-04-03 14:47:51 -0700518 if (shadowp)
519 *shadowp = p;
Johannes Weiner449dd692014-04-03 14:47:56 -0700520 mapping->nrshadows--;
521 if (node)
522 workingset_node_shadows_dec(node);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700523 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700524 radix_tree_replace_slot(slot, page);
525 mapping->nrpages++;
526 if (node) {
527 workingset_node_pages_inc(node);
528 /*
529 * Don't track node that contains actual pages.
530 *
531 * Avoid acquiring the list_lru lock if already
532 * untracked. The list_empty() test is safe as
533 * node->private_list is protected by
534 * mapping->tree_lock.
535 */
536 if (!list_empty(&node->private_list))
537 list_lru_del(&workingset_shadow_nodes,
538 &node->private_list);
539 }
540 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700541}
542
Johannes Weinera5289102014-04-03 14:47:51 -0700543static int __add_to_page_cache_locked(struct page *page,
544 struct address_space *mapping,
545 pgoff_t offset, gfp_t gfp_mask,
546 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
Johannes Weiner00501b52014-08-08 14:19:20 -0700548 int huge = PageHuge(page);
549 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700550 int error;
551
Sasha Levin309381fea2014-01-23 15:52:54 -0800552 VM_BUG_ON_PAGE(!PageLocked(page), page);
553 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700554
Johannes Weiner00501b52014-08-08 14:19:20 -0700555 if (!huge) {
556 error = mem_cgroup_try_charge(page, current->mm,
557 gfp_mask, &memcg);
558 if (error)
559 return error;
560 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Jan Kara5e4c0d972013-09-11 14:26:05 -0700562 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700563 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700564 if (!huge)
565 mem_cgroup_cancel_charge(page, memcg);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700566 return error;
567 }
568
569 page_cache_get(page);
570 page->mapping = mapping;
571 page->index = offset;
572
573 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700574 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700575 radix_tree_preload_end();
576 if (unlikely(error))
577 goto err_insert;
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700578 __inc_zone_page_state(page, NR_FILE_PAGES);
579 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700580 if (!huge)
581 mem_cgroup_commit_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700582 trace_mm_filemap_add_to_page_cache(page);
583 return 0;
584err_insert:
585 page->mapping = NULL;
586 /* Leave page->index set: truncation relies upon it */
587 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700588 if (!huge)
589 mem_cgroup_cancel_charge(page, memcg);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700590 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 return error;
592}
Johannes Weinera5289102014-04-03 14:47:51 -0700593
594/**
595 * add_to_page_cache_locked - add a locked page to the pagecache
596 * @page: page to add
597 * @mapping: the page's address_space
598 * @offset: page index
599 * @gfp_mask: page allocation mode
600 *
601 * This function is used to add a page to the pagecache. It must be locked.
602 * This function does not add the page to the LRU. The caller must do that.
603 */
604int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
605 pgoff_t offset, gfp_t gfp_mask)
606{
607 return __add_to_page_cache_locked(page, mapping, offset,
608 gfp_mask, NULL);
609}
Nick Piggine2867812008-07-25 19:45:30 -0700610EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400613 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
Johannes Weinera5289102014-04-03 14:47:51 -0700615 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700616 int ret;
617
Johannes Weinera5289102014-04-03 14:47:51 -0700618 __set_page_locked(page);
619 ret = __add_to_page_cache_locked(page, mapping, offset,
620 gfp_mask, &shadow);
621 if (unlikely(ret))
622 __clear_page_locked(page);
623 else {
624 /*
625 * The page might have been evicted from cache only
626 * recently, in which case it should be activated like
627 * any other repeatedly accessed page.
628 */
629 if (shadow && workingset_refault(shadow)) {
630 SetPageActive(page);
631 workingset_activation(page);
632 } else
633 ClearPageActive(page);
634 lru_cache_add(page);
635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 return ret;
637}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300638EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Paul Jackson44110fe2006-03-24 03:16:04 -0800640#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700641struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800642{
Miao Xiec0ff7452010-05-24 14:32:08 -0700643 int n;
644 struct page *page;
645
Paul Jackson44110fe2006-03-24 03:16:04 -0800646 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700647 unsigned int cpuset_mems_cookie;
648 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700649 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700650 n = cpuset_mem_spread_node();
651 page = alloc_pages_exact_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700652 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700653
Miao Xiec0ff7452010-05-24 14:32:08 -0700654 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800655 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700656 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800657}
Nick Piggin2ae88142006-10-28 10:38:23 -0700658EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800659#endif
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661/*
662 * In order to wait for pages to become available there must be
663 * waitqueues associated with pages. By using a hash table of
664 * waitqueues where the bucket discipline is to maintain all
665 * waiters on the same queue and wake all when any of the pages
666 * become available, and for the woken contexts to check to be
667 * sure the appropriate page became available, this saves space
668 * at a cost of "thundering herd" phenomena during rare hash
669 * collisions.
670 */
NeilBrowna4796e32014-09-24 11:28:32 +1000671wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
673 const struct zone *zone = page_zone(page);
674
675 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
676}
NeilBrowna4796e32014-09-24 11:28:32 +1000677EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Harvey Harrison920c7a52008-02-04 22:29:26 -0800679void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
682
683 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000684 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 TASK_UNINTERRUPTIBLE);
686}
687EXPORT_SYMBOL(wait_on_page_bit);
688
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700689int wait_on_page_bit_killable(struct page *page, int bit_nr)
690{
691 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
692
693 if (!test_bit(bit_nr, &page->flags))
694 return 0;
695
696 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000697 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700698}
699
NeilBrowncbbce822014-09-25 13:55:19 +1000700int wait_on_page_bit_killable_timeout(struct page *page,
701 int bit_nr, unsigned long timeout)
702{
703 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
704
705 wait.key.timeout = jiffies + timeout;
706 if (!test_bit(bit_nr, &page->flags))
707 return 0;
708 return __wait_on_bit(page_waitqueue(page), &wait,
709 bit_wait_io_timeout, TASK_KILLABLE);
710}
711EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100714 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700715 * @page: Page defining the wait queue of interest
716 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100717 *
718 * Add an arbitrary @waiter to the wait queue for the nominated @page.
719 */
720void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
721{
722 wait_queue_head_t *q = page_waitqueue(page);
723 unsigned long flags;
724
725 spin_lock_irqsave(&q->lock, flags);
726 __add_wait_queue(q, waiter);
727 spin_unlock_irqrestore(&q->lock, flags);
728}
729EXPORT_SYMBOL_GPL(add_page_wait_queue);
730
731/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700732 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 * @page: the page
734 *
735 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
736 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900737 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
739 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700740 * The mb is necessary to enforce ordering between the clear_bit and the read
741 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800743void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
Sasha Levin309381fea2014-01-23 15:52:54 -0800745 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700746 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100747 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 wake_up_page(page, PG_locked);
749}
750EXPORT_SYMBOL(unlock_page);
751
Randy Dunlap485bb992006-06-23 02:03:49 -0700752/**
753 * end_page_writeback - end writeback against a page
754 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 */
756void end_page_writeback(struct page *page)
757{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700758 /*
759 * TestClearPageReclaim could be used here but it is an atomic
760 * operation and overkill in this particular case. Failing to
761 * shuffle a page marked for immediate reclaim is too mild to
762 * justify taking an atomic operation penalty at the end of
763 * ever page writeback.
764 */
765 if (PageReclaim(page)) {
766 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700767 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700768 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700769
770 if (!test_clear_page_writeback(page))
771 BUG();
772
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100773 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 wake_up_page(page, PG_writeback);
775}
776EXPORT_SYMBOL(end_page_writeback);
777
Matthew Wilcox57d99842014-06-04 16:07:45 -0700778/*
779 * After completing I/O on a page, call this routine to update the page
780 * flags appropriately
781 */
782void page_endio(struct page *page, int rw, int err)
783{
784 if (rw == READ) {
785 if (!err) {
786 SetPageUptodate(page);
787 } else {
788 ClearPageUptodate(page);
789 SetPageError(page);
790 }
791 unlock_page(page);
792 } else { /* rw == WRITE */
793 if (err) {
794 SetPageError(page);
795 if (page->mapping)
796 mapping_set_error(page->mapping, err);
797 }
798 end_page_writeback(page);
799 }
800}
801EXPORT_SYMBOL_GPL(page_endio);
802
Randy Dunlap485bb992006-06-23 02:03:49 -0700803/**
804 * __lock_page - get a lock on the page, assuming we need to sleep to get it
805 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800807void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
809 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
810
NeilBrown74316202014-07-07 15:16:04 +1000811 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 TASK_UNINTERRUPTIBLE);
813}
814EXPORT_SYMBOL(__lock_page);
815
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800816int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500817{
818 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
819
820 return __wait_on_bit_lock(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000821 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500822}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300823EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500824
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700825/*
826 * Return values:
827 * 1 - page is locked; mmap_sem is still held.
828 * 0 - page is not locked.
829 * mmap_sem has been released (up_read()), unless flags had both
830 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
831 * which case mmap_sem is still held.
832 *
833 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
834 * with the page locked and the mmap_sem unperturbed.
835 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700836int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
837 unsigned int flags)
838{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700839 if (flags & FAULT_FLAG_ALLOW_RETRY) {
840 /*
841 * CAUTION! In this case, mmap_sem is not released
842 * even though return 0.
843 */
844 if (flags & FAULT_FLAG_RETRY_NOWAIT)
845 return 0;
846
847 up_read(&mm->mmap_sem);
848 if (flags & FAULT_FLAG_KILLABLE)
849 wait_on_page_locked_killable(page);
850 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700851 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700852 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700853 } else {
854 if (flags & FAULT_FLAG_KILLABLE) {
855 int ret;
856
857 ret = __lock_page_killable(page);
858 if (ret) {
859 up_read(&mm->mmap_sem);
860 return 0;
861 }
862 } else
863 __lock_page(page);
864 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700865 }
866}
867
Randy Dunlap485bb992006-06-23 02:03:49 -0700868/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700869 * page_cache_next_hole - find the next hole (not-present entry)
870 * @mapping: mapping
871 * @index: index
872 * @max_scan: maximum range to search
873 *
874 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
875 * lowest indexed hole.
876 *
877 * Returns: the index of the hole if found, otherwise returns an index
878 * outside of the set specified (in which case 'return - index >=
879 * max_scan' will be true). In rare cases of index wrap-around, 0 will
880 * be returned.
881 *
882 * page_cache_next_hole may be called under rcu_read_lock. However,
883 * like radix_tree_gang_lookup, this will not atomically search a
884 * snapshot of the tree at a single point in time. For example, if a
885 * hole is created at index 5, then subsequently a hole is created at
886 * index 10, page_cache_next_hole covering both indexes may return 10
887 * if called under rcu_read_lock.
888 */
889pgoff_t page_cache_next_hole(struct address_space *mapping,
890 pgoff_t index, unsigned long max_scan)
891{
892 unsigned long i;
893
894 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700895 struct page *page;
896
897 page = radix_tree_lookup(&mapping->page_tree, index);
898 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700899 break;
900 index++;
901 if (index == 0)
902 break;
903 }
904
905 return index;
906}
907EXPORT_SYMBOL(page_cache_next_hole);
908
909/**
910 * page_cache_prev_hole - find the prev hole (not-present entry)
911 * @mapping: mapping
912 * @index: index
913 * @max_scan: maximum range to search
914 *
915 * Search backwards in the range [max(index-max_scan+1, 0), index] for
916 * the first hole.
917 *
918 * Returns: the index of the hole if found, otherwise returns an index
919 * outside of the set specified (in which case 'index - return >=
920 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
921 * will be returned.
922 *
923 * page_cache_prev_hole may be called under rcu_read_lock. However,
924 * like radix_tree_gang_lookup, this will not atomically search a
925 * snapshot of the tree at a single point in time. For example, if a
926 * hole is created at index 10, then subsequently a hole is created at
927 * index 5, page_cache_prev_hole covering both indexes may return 5 if
928 * called under rcu_read_lock.
929 */
930pgoff_t page_cache_prev_hole(struct address_space *mapping,
931 pgoff_t index, unsigned long max_scan)
932{
933 unsigned long i;
934
935 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700936 struct page *page;
937
938 page = radix_tree_lookup(&mapping->page_tree, index);
939 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700940 break;
941 index--;
942 if (index == ULONG_MAX)
943 break;
944 }
945
946 return index;
947}
948EXPORT_SYMBOL(page_cache_prev_hole);
949
950/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700951 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -0700952 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -0700953 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -0700954 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700955 * Looks up the page cache slot at @mapping & @offset. If there is a
956 * page cache page, it is returned with an increased refcount.
957 *
Johannes Weiner139b6a62014-05-06 12:50:05 -0700958 * If the slot holds a shadow entry of a previously evicted page, or a
959 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -0700960 *
961 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700963struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
Nick Piggina60637c2008-07-25 19:45:31 -0700965 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 struct page *page;
967
Nick Piggina60637c2008-07-25 19:45:31 -0700968 rcu_read_lock();
969repeat:
970 page = NULL;
971 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
972 if (pagep) {
973 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800974 if (unlikely(!page))
975 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700976 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700977 if (radix_tree_deref_retry(page))
978 goto repeat;
979 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -0700980 * A shadow entry of a recently evicted page,
981 * or a swap entry from shmem/tmpfs. Return
982 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700983 */
984 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700985 }
Nick Piggina60637c2008-07-25 19:45:31 -0700986 if (!page_cache_get_speculative(page))
987 goto repeat;
988
989 /*
990 * Has the page moved?
991 * This is part of the lockless pagecache protocol. See
992 * include/linux/pagemap.h for details.
993 */
994 if (unlikely(page != *pagep)) {
995 page_cache_release(page);
996 goto repeat;
997 }
998 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800999out:
Nick Piggina60637c2008-07-25 19:45:31 -07001000 rcu_read_unlock();
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 return page;
1003}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001004EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Randy Dunlap485bb992006-06-23 02:03:49 -07001006/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001007 * find_lock_entry - locate, pin and lock a page cache entry
1008 * @mapping: the address_space to search
1009 * @offset: the page cache index
1010 *
1011 * Looks up the page cache slot at @mapping & @offset. If there is a
1012 * page cache page, it is returned locked and with an increased
1013 * refcount.
1014 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001015 * If the slot holds a shadow entry of a previously evicted page, or a
1016 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001017 *
1018 * Otherwise, %NULL is returned.
1019 *
1020 * find_lock_entry() may sleep.
1021 */
1022struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
1024 struct page *page;
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001027 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001028 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001029 lock_page(page);
1030 /* Has the page been truncated? */
1031 if (unlikely(page->mapping != mapping)) {
1032 unlock_page(page);
1033 page_cache_release(page);
1034 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001036 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 return page;
1039}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001040EXPORT_SYMBOL(find_lock_entry);
1041
1042/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001043 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001044 * @mapping: the address_space to search
1045 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001046 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001047 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001048 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001049 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001050 *
Randy Dunlap75325182014-07-30 16:08:37 -07001051 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001052 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001053 * FGP_ACCESSED: the page will be marked accessed
1054 * FGP_LOCK: Page is return locked
1055 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001056 * @gfp_mask and added to the page cache and the VM's LRU
1057 * list. The page is returned locked and with an increased
1058 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001059 *
1060 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1061 * if the GFP flags specified for FGP_CREAT are atomic.
1062 *
1063 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001064 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001065struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001066 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
Nick Piggineb2be182007-10-16 01:24:57 -07001068 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001071 page = find_get_entry(mapping, offset);
1072 if (radix_tree_exceptional_entry(page))
1073 page = NULL;
1074 if (!page)
1075 goto no_page;
1076
1077 if (fgp_flags & FGP_LOCK) {
1078 if (fgp_flags & FGP_NOWAIT) {
1079 if (!trylock_page(page)) {
1080 page_cache_release(page);
1081 return NULL;
1082 }
1083 } else {
1084 lock_page(page);
1085 }
1086
1087 /* Has the page been truncated? */
1088 if (unlikely(page->mapping != mapping)) {
1089 unlock_page(page);
1090 page_cache_release(page);
1091 goto repeat;
1092 }
1093 VM_BUG_ON_PAGE(page->index != offset, page);
1094 }
1095
1096 if (page && (fgp_flags & FGP_ACCESSED))
1097 mark_page_accessed(page);
1098
1099no_page:
1100 if (!page && (fgp_flags & FGP_CREAT)) {
1101 int err;
1102 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001103 gfp_mask |= __GFP_WRITE;
1104 if (fgp_flags & FGP_NOFS)
1105 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001106
Michal Hocko45f87de2014-12-29 20:30:35 +01001107 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001108 if (!page)
1109 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001110
1111 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1112 fgp_flags |= FGP_LOCK;
1113
Hugh Dickinseb39d612014-08-06 16:06:43 -07001114 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001115 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001116 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001117
Michal Hocko45f87de2014-12-29 20:30:35 +01001118 err = add_to_page_cache_lru(page, mapping, offset,
1119 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001120 if (unlikely(err)) {
1121 page_cache_release(page);
1122 page = NULL;
1123 if (err == -EEXIST)
1124 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 return page;
1129}
Mel Gorman2457aec2014-06-04 16:10:31 -07001130EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001133 * find_get_entries - gang pagecache lookup
1134 * @mapping: The address_space to search
1135 * @start: The starting page cache index
1136 * @nr_entries: The maximum number of entries
1137 * @entries: Where the resulting entries are placed
1138 * @indices: The cache indices corresponding to the entries in @entries
1139 *
1140 * find_get_entries() will search for and return a group of up to
1141 * @nr_entries entries in the mapping. The entries are placed at
1142 * @entries. find_get_entries() takes a reference against any actual
1143 * pages it returns.
1144 *
1145 * The search returns a group of mapping-contiguous page cache entries
1146 * with ascending indexes. There may be holes in the indices due to
1147 * not-present pages.
1148 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001149 * Any shadow entries of evicted pages, or swap entries from
1150 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001151 *
1152 * find_get_entries() returns the number of pages and shadow entries
1153 * which were found.
1154 */
1155unsigned find_get_entries(struct address_space *mapping,
1156 pgoff_t start, unsigned int nr_entries,
1157 struct page **entries, pgoff_t *indices)
1158{
1159 void **slot;
1160 unsigned int ret = 0;
1161 struct radix_tree_iter iter;
1162
1163 if (!nr_entries)
1164 return 0;
1165
1166 rcu_read_lock();
1167restart:
1168 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1169 struct page *page;
1170repeat:
1171 page = radix_tree_deref_slot(slot);
1172 if (unlikely(!page))
1173 continue;
1174 if (radix_tree_exception(page)) {
1175 if (radix_tree_deref_retry(page))
1176 goto restart;
1177 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001178 * A shadow entry of a recently evicted page,
1179 * or a swap entry from shmem/tmpfs. Return
1180 * it without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001181 */
1182 goto export;
1183 }
1184 if (!page_cache_get_speculative(page))
1185 goto repeat;
1186
1187 /* Has the page moved? */
1188 if (unlikely(page != *slot)) {
1189 page_cache_release(page);
1190 goto repeat;
1191 }
1192export:
1193 indices[ret] = iter.index;
1194 entries[ret] = page;
1195 if (++ret == nr_entries)
1196 break;
1197 }
1198 rcu_read_unlock();
1199 return ret;
1200}
1201
1202/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 * find_get_pages - gang pagecache lookup
1204 * @mapping: The address_space to search
1205 * @start: The starting page index
1206 * @nr_pages: The maximum number of pages
1207 * @pages: Where the resulting pages are placed
1208 *
1209 * find_get_pages() will search for and return a group of up to
1210 * @nr_pages pages in the mapping. The pages are placed at @pages.
1211 * find_get_pages() takes a reference against the returned pages.
1212 *
1213 * The search returns a group of mapping-contiguous pages with ascending
1214 * indexes. There may be holes in the indices due to not-present pages.
1215 *
1216 * find_get_pages() returns the number of pages which were found.
1217 */
1218unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1219 unsigned int nr_pages, struct page **pages)
1220{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001221 struct radix_tree_iter iter;
1222 void **slot;
1223 unsigned ret = 0;
1224
1225 if (unlikely(!nr_pages))
1226 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227
Nick Piggina60637c2008-07-25 19:45:31 -07001228 rcu_read_lock();
1229restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001230 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001231 struct page *page;
1232repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001233 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001234 if (unlikely(!page))
1235 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001236
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001237 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001238 if (radix_tree_deref_retry(page)) {
1239 /*
1240 * Transient condition which can only trigger
1241 * when entry at index 0 moves out of or back
1242 * to root: none yet gotten, safe to restart.
1243 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001244 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001245 goto restart;
1246 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001247 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001248 * A shadow entry of a recently evicted page,
1249 * or a swap entry from shmem/tmpfs. Skip
1250 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001251 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001252 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001253 }
Nick Piggina60637c2008-07-25 19:45:31 -07001254
1255 if (!page_cache_get_speculative(page))
1256 goto repeat;
1257
1258 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001259 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001260 page_cache_release(page);
1261 goto repeat;
1262 }
1263
1264 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001265 if (++ret == nr_pages)
1266 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001267 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001268
Nick Piggina60637c2008-07-25 19:45:31 -07001269 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 return ret;
1271}
1272
Jens Axboeebf43502006-04-27 08:46:01 +02001273/**
1274 * find_get_pages_contig - gang contiguous pagecache lookup
1275 * @mapping: The address_space to search
1276 * @index: The starting page index
1277 * @nr_pages: The maximum number of pages
1278 * @pages: Where the resulting pages are placed
1279 *
1280 * find_get_pages_contig() works exactly like find_get_pages(), except
1281 * that the returned number of pages are guaranteed to be contiguous.
1282 *
1283 * find_get_pages_contig() returns the number of pages which were found.
1284 */
1285unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1286 unsigned int nr_pages, struct page **pages)
1287{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001288 struct radix_tree_iter iter;
1289 void **slot;
1290 unsigned int ret = 0;
1291
1292 if (unlikely(!nr_pages))
1293 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001294
Nick Piggina60637c2008-07-25 19:45:31 -07001295 rcu_read_lock();
1296restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001297 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001298 struct page *page;
1299repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001300 page = radix_tree_deref_slot(slot);
1301 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001302 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001303 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001304
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001305 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001306 if (radix_tree_deref_retry(page)) {
1307 /*
1308 * Transient condition which can only trigger
1309 * when entry at index 0 moves out of or back
1310 * to root: none yet gotten, safe to restart.
1311 */
1312 goto restart;
1313 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001314 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001315 * A shadow entry of a recently evicted page,
1316 * or a swap entry from shmem/tmpfs. Stop
1317 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001318 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001319 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001320 }
Nick Piggina60637c2008-07-25 19:45:31 -07001321
Nick Piggina60637c2008-07-25 19:45:31 -07001322 if (!page_cache_get_speculative(page))
1323 goto repeat;
1324
1325 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001326 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001327 page_cache_release(page);
1328 goto repeat;
1329 }
1330
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001331 /*
1332 * must check mapping and index after taking the ref.
1333 * otherwise we can get both false positives and false
1334 * negatives, which is just confusing to the caller.
1335 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001336 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001337 page_cache_release(page);
1338 break;
1339 }
1340
Nick Piggina60637c2008-07-25 19:45:31 -07001341 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001342 if (++ret == nr_pages)
1343 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001344 }
Nick Piggina60637c2008-07-25 19:45:31 -07001345 rcu_read_unlock();
1346 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001347}
David Howellsef71c152007-05-09 02:33:44 -07001348EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001349
Randy Dunlap485bb992006-06-23 02:03:49 -07001350/**
1351 * find_get_pages_tag - find and return pages that match @tag
1352 * @mapping: the address_space to search
1353 * @index: the starting page index
1354 * @tag: the tag index
1355 * @nr_pages: the maximum number of pages
1356 * @pages: where the resulting pages are placed
1357 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001359 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 */
1361unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1362 int tag, unsigned int nr_pages, struct page **pages)
1363{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001364 struct radix_tree_iter iter;
1365 void **slot;
1366 unsigned ret = 0;
1367
1368 if (unlikely(!nr_pages))
1369 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Nick Piggina60637c2008-07-25 19:45:31 -07001371 rcu_read_lock();
1372restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001373 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1374 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001375 struct page *page;
1376repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001377 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001378 if (unlikely(!page))
1379 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001380
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001381 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001382 if (radix_tree_deref_retry(page)) {
1383 /*
1384 * Transient condition which can only trigger
1385 * when entry at index 0 moves out of or back
1386 * to root: none yet gotten, safe to restart.
1387 */
1388 goto restart;
1389 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001390 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001391 * A shadow entry of a recently evicted page.
1392 *
1393 * Those entries should never be tagged, but
1394 * this tree walk is lockless and the tags are
1395 * looked up in bulk, one radix tree node at a
1396 * time, so there is a sizable window for page
1397 * reclaim to evict a page we saw tagged.
1398 *
1399 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001400 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001401 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001402 }
Nick Piggina60637c2008-07-25 19:45:31 -07001403
1404 if (!page_cache_get_speculative(page))
1405 goto repeat;
1406
1407 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001408 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001409 page_cache_release(page);
1410 goto repeat;
1411 }
1412
1413 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001414 if (++ret == nr_pages)
1415 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001416 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001417
Nick Piggina60637c2008-07-25 19:45:31 -07001418 rcu_read_unlock();
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 if (ret)
1421 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 return ret;
1424}
David Howellsef71c152007-05-09 02:33:44 -07001425EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001427/*
1428 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1429 * a _large_ part of the i/o request. Imagine the worst scenario:
1430 *
1431 * ---R__________________________________________B__________
1432 * ^ reading here ^ bad block(assume 4k)
1433 *
1434 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1435 * => failing the whole request => read(R) => read(R+1) =>
1436 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1437 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1438 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1439 *
1440 * It is going insane. Fix it by quickly scaling down the readahead size.
1441 */
1442static void shrink_readahead_size_eio(struct file *filp,
1443 struct file_ra_state *ra)
1444{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001445 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001446}
1447
Randy Dunlap485bb992006-06-23 02:03:49 -07001448/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001449 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001450 * @filp: the file to read
1451 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001452 * @iter: data destination
1453 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001454 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001456 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 *
1458 * This is really ugly. But the goto's actually try to clarify some
1459 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 */
Al Viro6e58e792014-02-03 17:07:03 -05001461static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1462 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001464 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001466 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001467 pgoff_t index;
1468 pgoff_t last_index;
1469 pgoff_t prev_index;
1470 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001471 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001472 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001475 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1476 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Al Viro6e58e792014-02-03 17:07:03 -05001477 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 offset = *ppos & ~PAGE_CACHE_MASK;
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 for (;;) {
1481 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001482 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001483 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 unsigned long nr, ret;
1485
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487find_page:
1488 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001489 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001490 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001491 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001492 index, last_index - index);
1493 page = find_get_page(mapping, index);
1494 if (unlikely(page == NULL))
1495 goto no_cached_page;
1496 }
1497 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001498 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001499 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001500 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001502 if (!PageUptodate(page)) {
1503 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1504 !mapping->a_ops->is_partially_uptodate)
1505 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001506 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001507 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001508 /* Did it get truncated before we got the lock? */
1509 if (!page->mapping)
1510 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001511 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001512 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001513 goto page_not_up_to_date_locked;
1514 unlock_page(page);
1515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001517 /*
1518 * i_size must be checked after we know the page is Uptodate.
1519 *
1520 * Checking i_size after the check allows us to calculate
1521 * the correct value for "nr", which means the zero-filled
1522 * part of the page is not copied back to userspace (unless
1523 * another truncate extends the file - this is desired though).
1524 */
1525
1526 isize = i_size_read(inode);
1527 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1528 if (unlikely(!isize || index > end_index)) {
1529 page_cache_release(page);
1530 goto out;
1531 }
1532
1533 /* nr is the maximum number of bytes to copy from this page */
1534 nr = PAGE_CACHE_SIZE;
1535 if (index == end_index) {
1536 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1537 if (nr <= offset) {
1538 page_cache_release(page);
1539 goto out;
1540 }
1541 }
1542 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
1544 /* If users can be writing to this page using arbitrary
1545 * virtual addresses, take care about potential aliasing
1546 * before reading the page on the kernel side.
1547 */
1548 if (mapping_writably_mapped(mapping))
1549 flush_dcache_page(page);
1550
1551 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001552 * When a sequential read accesses a page several times,
1553 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 */
Jan Karaec0f1632007-05-06 14:49:25 -07001555 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 mark_page_accessed(page);
1557 prev_index = index;
1558
1559 /*
1560 * Ok, we have the page, and it's up-to-date, so
1561 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 */
Al Viro6e58e792014-02-03 17:07:03 -05001563
1564 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 offset += ret;
1566 index += offset >> PAGE_CACHE_SHIFT;
1567 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001568 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
1570 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001571 written += ret;
1572 if (!iov_iter_count(iter))
1573 goto out;
1574 if (ret < nr) {
1575 error = -EFAULT;
1576 goto out;
1577 }
1578 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580page_not_up_to_date:
1581 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001582 error = lock_page_killable(page);
1583 if (unlikely(error))
1584 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001586page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001587 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 if (!page->mapping) {
1589 unlock_page(page);
1590 page_cache_release(page);
1591 continue;
1592 }
1593
1594 /* Did somebody else fill it already? */
1595 if (PageUptodate(page)) {
1596 unlock_page(page);
1597 goto page_ok;
1598 }
1599
1600readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001601 /*
1602 * A previous I/O error may have been due to temporary
1603 * failures, eg. multipath errors.
1604 * PG_error will be set again if readpage fails.
1605 */
1606 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 /* Start the actual read. The read will unlock the page. */
1608 error = mapping->a_ops->readpage(filp, page);
1609
Zach Brown994fc28c2005-12-15 14:28:17 -08001610 if (unlikely(error)) {
1611 if (error == AOP_TRUNCATED_PAGE) {
1612 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001613 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001614 goto find_page;
1615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
1619 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001620 error = lock_page_killable(page);
1621 if (unlikely(error))
1622 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 if (!PageUptodate(page)) {
1624 if (page->mapping == NULL) {
1625 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001626 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 */
1628 unlock_page(page);
1629 page_cache_release(page);
1630 goto find_page;
1631 }
1632 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001633 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001634 error = -EIO;
1635 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 }
1637 unlock_page(page);
1638 }
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 goto page_ok;
1641
1642readpage_error:
1643 /* UHHUH! A synchronous read error occurred. Report it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 page_cache_release(page);
1645 goto out;
1646
1647no_cached_page:
1648 /*
1649 * Ok, it wasn't cached, so we need to create a new
1650 * page..
1651 */
Nick Piggineb2be182007-10-16 01:24:57 -07001652 page = page_cache_alloc_cold(mapping);
1653 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001654 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001655 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 }
Nick Piggineb2be182007-10-16 01:24:57 -07001657 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 index, GFP_KERNEL);
1659 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001660 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001661 if (error == -EEXIST) {
1662 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 goto out;
1666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 goto readpage;
1668 }
1669
1670out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001671 ra->prev_pos = prev_index;
1672 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1673 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001675 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001676 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001677 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678}
1679
Randy Dunlap485bb992006-06-23 02:03:49 -07001680/**
Al Viro6abd2322014-04-04 14:20:57 -04001681 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001682 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001683 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001684 *
Al Viro6abd2322014-04-04 14:20:57 -04001685 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 * that can use the page cache directly.
1687 */
1688ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001689generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690{
Al Viroed978a82014-03-05 22:53:04 -05001691 struct file *file = iocb->ki_filp;
1692 ssize_t retval = 0;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001693 loff_t *ppos = &iocb->ki_pos;
Al Viroed978a82014-03-05 22:53:04 -05001694 loff_t pos = *ppos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
Al Viro2ba48ce2015-04-09 13:52:01 -04001696 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001697 struct address_space *mapping = file->f_mapping;
1698 struct inode *inode = mapping->host;
1699 size_t count = iov_iter_count(iter);
Badari Pulavarty543ade12006-09-30 23:28:48 -07001700 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 if (!count)
1703 goto out; /* skip atime */
1704 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001705 retval = filemap_write_and_wait_range(mapping, pos,
Al Viroed978a82014-03-05 22:53:04 -05001706 pos + count - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001707 if (!retval) {
Al Viroed978a82014-03-05 22:53:04 -05001708 struct iov_iter data = *iter;
Omar Sandoval22c61862015-03-16 04:33:53 -07001709 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001710 }
Al Viroed978a82014-03-05 22:53:04 -05001711
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001712 if (retval > 0) {
1713 *ppos = pos + retval;
Al Viroed978a82014-03-05 22:53:04 -05001714 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001715 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001716
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001717 /*
1718 * Btrfs can have a short DIO read if we encounter
1719 * compressed extents, so if there was an error, or if
1720 * we've already read everything we wanted to, or if
1721 * there was a short read because we hit EOF, go ahead
1722 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001723 * the rest of the read. Buffered reads will not work for
1724 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001725 */
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001726 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1727 IS_DAX(inode)) {
Al Viroed978a82014-03-05 22:53:04 -05001728 file_accessed(file);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001729 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 }
1732
Al Viroed978a82014-03-05 22:53:04 -05001733 retval = do_generic_file_read(file, ppos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734out:
1735 return retval;
1736}
Al Viroed978a82014-03-05 22:53:04 -05001737EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001740/**
1741 * page_cache_read - adds requested page to the page cache if not already there
1742 * @file: file to read
1743 * @offset: page index
1744 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 * This adds the requested page to the page cache if it isn't already there,
1746 * and schedules an I/O to read in its contents from disk.
1747 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001748static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749{
1750 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001751 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001752 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
Zach Brown994fc28c2005-12-15 14:28:17 -08001754 do {
1755 page = page_cache_alloc_cold(mapping);
1756 if (!page)
1757 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Zach Brown994fc28c2005-12-15 14:28:17 -08001759 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1760 if (ret == 0)
1761 ret = mapping->a_ops->readpage(file, page);
1762 else if (ret == -EEXIST)
1763 ret = 0; /* losing race to add is OK */
1764
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
Zach Brown994fc28c2005-12-15 14:28:17 -08001767 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07001768
Zach Brown994fc28c2005-12-15 14:28:17 -08001769 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770}
1771
1772#define MMAP_LOTSAMISS (100)
1773
Linus Torvaldsef00e082009-06-16 15:31:25 -07001774/*
1775 * Synchronous readahead happens when we don't even find
1776 * a page in the page cache at all.
1777 */
1778static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1779 struct file_ra_state *ra,
1780 struct file *file,
1781 pgoff_t offset)
1782{
1783 unsigned long ra_pages;
1784 struct address_space *mapping = file->f_mapping;
1785
1786 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001787 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001788 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001789 if (!ra->ra_pages)
1790 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001791
Joe Perches64363aa2013-07-08 16:00:18 -07001792 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001793 page_cache_sync_readahead(mapping, ra, file, offset,
1794 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001795 return;
1796 }
1797
Andi Kleen207d04b2011-05-24 17:12:29 -07001798 /* Avoid banging the cache line if not needed */
1799 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001800 ra->mmap_miss++;
1801
1802 /*
1803 * Do we miss much more than hit in this file? If so,
1804 * stop bothering with read-ahead. It will only hurt.
1805 */
1806 if (ra->mmap_miss > MMAP_LOTSAMISS)
1807 return;
1808
Wu Fengguangd30a1102009-06-16 15:31:30 -07001809 /*
1810 * mmap read-around
1811 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001812 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001813 ra->start = max_t(long, 0, offset - ra_pages / 2);
1814 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001815 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001816 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001817}
1818
1819/*
1820 * Asynchronous readahead happens when we find the page and PG_readahead,
1821 * so we want to possibly extend the readahead further..
1822 */
1823static void do_async_mmap_readahead(struct vm_area_struct *vma,
1824 struct file_ra_state *ra,
1825 struct file *file,
1826 struct page *page,
1827 pgoff_t offset)
1828{
1829 struct address_space *mapping = file->f_mapping;
1830
1831 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001832 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001833 return;
1834 if (ra->mmap_miss > 0)
1835 ra->mmap_miss--;
1836 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001837 page_cache_async_readahead(mapping, ra, file,
1838 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001839}
1840
Randy Dunlap485bb992006-06-23 02:03:49 -07001841/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001842 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001843 * @vma: vma in which the fault was taken
1844 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001845 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001846 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 * mapped memory region to read in file data during a page fault.
1848 *
1849 * The goto's are kind of ugly, but this streamlines the normal case of having
1850 * it in the page cache, and handles the special cases reasonably without
1851 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001852 *
1853 * vma->vm_mm->mmap_sem must be held on entry.
1854 *
1855 * If our return value has VM_FAULT_RETRY set, it's because
1856 * lock_page_or_retry() returned 0.
1857 * The mmap_sem has usually been released in this case.
1858 * See __lock_page_or_retry() for the exception.
1859 *
1860 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1861 * has not been released.
1862 *
1863 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001865int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866{
1867 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001868 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 struct address_space *mapping = file->f_mapping;
1870 struct file_ra_state *ra = &file->f_ra;
1871 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001872 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001874 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001875 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001877 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1878 if (offset >= size >> PAGE_CACHE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001879 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001882 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001884 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001885 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001887 * We found the page, so try async readahead before
1888 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001890 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001891 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001892 /* No page in the page cache at all */
1893 do_sync_mmap_readahead(vma, ra, file, offset);
1894 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001895 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001896 ret = VM_FAULT_MAJOR;
1897retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001898 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (!page)
1900 goto no_cached_page;
1901 }
1902
Michel Lespinassed88c0922010-11-02 13:05:18 -07001903 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1904 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001905 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001906 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001907
1908 /* Did it get truncated? */
1909 if (unlikely(page->mapping != mapping)) {
1910 unlock_page(page);
1911 put_page(page);
1912 goto retry_find;
1913 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001914 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07001915
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001917 * We have a locked page in the page cache, now we need to check
1918 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 */
Nick Piggind00806b2007-07-19 01:46:57 -07001920 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 goto page_not_uptodate;
1922
Linus Torvaldsef00e082009-06-16 15:31:25 -07001923 /*
1924 * Found the page and have a reference on it.
1925 * We must recheck i_size under page lock.
1926 */
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001927 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1928 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001929 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001930 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001931 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001932 }
1933
Nick Piggind0217ac2007-07-19 01:47:03 -07001934 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001935 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937no_cached_page:
1938 /*
1939 * We're only likely to ever get here if MADV_RANDOM is in
1940 * effect.
1941 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001942 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944 /*
1945 * The page we want has now been added to the page cache.
1946 * In the unlikely event that someone removed it in the
1947 * meantime, we'll just come back here and read it again.
1948 */
1949 if (error >= 0)
1950 goto retry_find;
1951
1952 /*
1953 * An error return from page_cache_read can result if the
1954 * system is low on memory, or a problem occurs while trying
1955 * to schedule I/O.
1956 */
1957 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001958 return VM_FAULT_OOM;
1959 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
1961page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 /*
1963 * Umm, take care of errors if the page isn't up-to-date.
1964 * Try to re-read it _once_. We do this synchronously,
1965 * because there really aren't any performance issues here
1966 * and we need to check for errors.
1967 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001969 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001970 if (!error) {
1971 wait_on_page_locked(page);
1972 if (!PageUptodate(page))
1973 error = -EIO;
1974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001976
1977 if (!error || error == AOP_TRUNCATED_PAGE)
1978 goto retry_find;
1979
1980 /* Things didn't work out. Return zero to tell the mm layer so. */
1981 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001982 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001983}
1984EXPORT_SYMBOL(filemap_fault);
1985
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001986void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1987{
1988 struct radix_tree_iter iter;
1989 void **slot;
1990 struct file *file = vma->vm_file;
1991 struct address_space *mapping = file->f_mapping;
1992 loff_t size;
1993 struct page *page;
1994 unsigned long address = (unsigned long) vmf->virtual_address;
1995 unsigned long addr;
1996 pte_t *pte;
1997
1998 rcu_read_lock();
1999 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2000 if (iter.index > vmf->max_pgoff)
2001 break;
2002repeat:
2003 page = radix_tree_deref_slot(slot);
2004 if (unlikely(!page))
2005 goto next;
2006 if (radix_tree_exception(page)) {
2007 if (radix_tree_deref_retry(page))
2008 break;
2009 else
2010 goto next;
2011 }
2012
2013 if (!page_cache_get_speculative(page))
2014 goto repeat;
2015
2016 /* Has the page moved? */
2017 if (unlikely(page != *slot)) {
2018 page_cache_release(page);
2019 goto repeat;
2020 }
2021
2022 if (!PageUptodate(page) ||
2023 PageReadahead(page) ||
2024 PageHWPoison(page))
2025 goto skip;
2026 if (!trylock_page(page))
2027 goto skip;
2028
2029 if (page->mapping != mapping || !PageUptodate(page))
2030 goto unlock;
2031
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002032 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2033 if (page->index >= size >> PAGE_CACHE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002034 goto unlock;
2035
2036 pte = vmf->pte + page->index - vmf->pgoff;
2037 if (!pte_none(*pte))
2038 goto unlock;
2039
2040 if (file->f_ra.mmap_miss > 0)
2041 file->f_ra.mmap_miss--;
2042 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2043 do_set_pte(vma, addr, page, pte, false, false);
2044 unlock_page(page);
2045 goto next;
2046unlock:
2047 unlock_page(page);
2048skip:
2049 page_cache_release(page);
2050next:
2051 if (iter.index == vmf->max_pgoff)
2052 break;
2053 }
2054 rcu_read_unlock();
2055}
2056EXPORT_SYMBOL(filemap_map_pages);
2057
Jan Kara4fcf1c62012-06-12 16:20:29 +02002058int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2059{
2060 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002061 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002062 int ret = VM_FAULT_LOCKED;
2063
Jan Kara14da9202012-06-12 16:20:37 +02002064 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002065 file_update_time(vma->vm_file);
2066 lock_page(page);
2067 if (page->mapping != inode->i_mapping) {
2068 unlock_page(page);
2069 ret = VM_FAULT_NOPAGE;
2070 goto out;
2071 }
Jan Kara14da9202012-06-12 16:20:37 +02002072 /*
2073 * We mark the page dirty already here so that when freeze is in
2074 * progress, we are guaranteed that writeback during freezing will
2075 * see the dirty page and writeprotect it again.
2076 */
2077 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002078 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002079out:
Jan Kara14da9202012-06-12 16:20:37 +02002080 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002081 return ret;
2082}
2083EXPORT_SYMBOL(filemap_page_mkwrite);
2084
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002085const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002086 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002087 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002088 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089};
2090
2091/* This is used for a general mmap of a disk file */
2092
2093int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2094{
2095 struct address_space *mapping = file->f_mapping;
2096
2097 if (!mapping->a_ops->readpage)
2098 return -ENOEXEC;
2099 file_accessed(file);
2100 vma->vm_ops = &generic_file_vm_ops;
2101 return 0;
2102}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
2104/*
2105 * This is for filesystems which do not implement ->writepage.
2106 */
2107int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2108{
2109 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2110 return -EINVAL;
2111 return generic_file_mmap(file, vma);
2112}
2113#else
2114int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2115{
2116 return -ENOSYS;
2117}
2118int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2119{
2120 return -ENOSYS;
2121}
2122#endif /* CONFIG_MMU */
2123
2124EXPORT_SYMBOL(generic_file_mmap);
2125EXPORT_SYMBOL(generic_file_readonly_mmap);
2126
Sasha Levin67f9fd92014-04-03 14:48:18 -07002127static struct page *wait_on_page_read(struct page *page)
2128{
2129 if (!IS_ERR(page)) {
2130 wait_on_page_locked(page);
2131 if (!PageUptodate(page)) {
2132 page_cache_release(page);
2133 page = ERR_PTR(-EIO);
2134 }
2135 }
2136 return page;
2137}
2138
Nick Piggin6fe69002007-05-06 14:49:04 -07002139static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002140 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002141 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002142 void *data,
2143 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144{
Nick Piggineb2be182007-10-16 01:24:57 -07002145 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 int err;
2147repeat:
2148 page = find_get_page(mapping, index);
2149 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002150 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002151 if (!page)
2152 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002153 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002154 if (unlikely(err)) {
2155 page_cache_release(page);
2156 if (err == -EEXIST)
2157 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 return ERR_PTR(err);
2160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 err = filler(data, page);
2162 if (err < 0) {
2163 page_cache_release(page);
2164 page = ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002165 } else {
2166 page = wait_on_page_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 }
2168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 return page;
2170}
2171
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002172static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002173 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002174 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002175 void *data,
2176 gfp_t gfp)
2177
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178{
2179 struct page *page;
2180 int err;
2181
2182retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002183 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01002185 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 if (PageUptodate(page))
2187 goto out;
2188
2189 lock_page(page);
2190 if (!page->mapping) {
2191 unlock_page(page);
2192 page_cache_release(page);
2193 goto retry;
2194 }
2195 if (PageUptodate(page)) {
2196 unlock_page(page);
2197 goto out;
2198 }
2199 err = filler(data, page);
2200 if (err < 0) {
2201 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01002202 return ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002203 } else {
2204 page = wait_on_page_read(page);
2205 if (IS_ERR(page))
2206 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 }
David Howellsc855ff32007-05-09 13:42:20 +01002208out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002209 mark_page_accessed(page);
2210 return page;
2211}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002212
2213/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002214 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002215 * @mapping: the page's address_space
2216 * @index: the page index
2217 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002218 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002219 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002220 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002221 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002222 *
2223 * If the page does not get brought uptodate, return -EIO.
2224 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002225struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002226 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002227 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002228 void *data)
2229{
2230 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2231}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002232EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002233
2234/**
2235 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2236 * @mapping: the page's address_space
2237 * @index: the page index
2238 * @gfp: the page allocator flags to use if allocating
2239 *
2240 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002241 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002242 *
2243 * If the page does not get brought uptodate, return -EIO.
2244 */
2245struct page *read_cache_page_gfp(struct address_space *mapping,
2246 pgoff_t index,
2247 gfp_t gfp)
2248{
2249 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2250
Sasha Levin67f9fd92014-04-03 14:48:18 -07002251 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002252}
2253EXPORT_SYMBOL(read_cache_page_gfp);
2254
Nick Piggin2f718ff2007-10-16 01:24:59 -07002255/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 * Performs necessary checks before doing a write
2257 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002258 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 * Returns appropriate error code that caller should return or
2260 * zero in case that write should be allowed.
2261 */
Al Viro3309dd02015-04-09 12:55:47 -04002262inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263{
Al Viro3309dd02015-04-09 12:55:47 -04002264 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002266 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002267 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
Al Viro3309dd02015-04-09 12:55:47 -04002269 if (!iov_iter_count(from))
2270 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
Al Viro0fa6b002015-04-04 04:05:48 -04002272 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002273 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002274 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
Al Viro3309dd02015-04-09 12:55:47 -04002276 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Al Viro0fa6b002015-04-04 04:05:48 -04002278 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002279 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002280 send_sig(SIGXFSZ, current, 0);
2281 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 }
Al Viro3309dd02015-04-09 12:55:47 -04002283 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 }
2285
2286 /*
2287 * LFS rule
2288 */
Al Viro3309dd02015-04-09 12:55:47 -04002289 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002291 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002293 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 }
2295
2296 /*
2297 * Are we about to exceed the fs block limit ?
2298 *
2299 * If we have written data it becomes a short write. If we have
2300 * exceeded without writing data we send a signal and return EFBIG.
2301 * Linus frestrict idea will clean these up nicely..
2302 */
Al Viro3309dd02015-04-09 12:55:47 -04002303 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2304 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305
Al Viro3309dd02015-04-09 12:55:47 -04002306 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2307 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308}
2309EXPORT_SYMBOL(generic_write_checks);
2310
Nick Pigginafddba42007-10-16 01:25:01 -07002311int pagecache_write_begin(struct file *file, struct address_space *mapping,
2312 loff_t pos, unsigned len, unsigned flags,
2313 struct page **pagep, void **fsdata)
2314{
2315 const struct address_space_operations *aops = mapping->a_ops;
2316
Nick Piggin4e02ed42008-10-29 14:00:55 -07002317 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002318 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002319}
2320EXPORT_SYMBOL(pagecache_write_begin);
2321
2322int pagecache_write_end(struct file *file, struct address_space *mapping,
2323 loff_t pos, unsigned len, unsigned copied,
2324 struct page *page, void *fsdata)
2325{
2326 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002327
Nick Piggin4e02ed42008-10-29 14:00:55 -07002328 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002329}
2330EXPORT_SYMBOL(pagecache_write_end);
2331
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332ssize_t
Al Viro0c949332014-03-22 06:51:37 -04002333generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334{
2335 struct file *file = iocb->ki_filp;
2336 struct address_space *mapping = file->f_mapping;
2337 struct inode *inode = mapping->host;
2338 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002339 size_t write_len;
2340 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002341 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
Al Viro0c949332014-03-22 06:51:37 -04002343 write_len = iov_iter_count(from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002344 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002345
Nick Piggin48b47c52009-01-06 14:40:22 -08002346 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002347 if (written)
2348 goto out;
2349
2350 /*
2351 * After a write we want buffered reads to be sure to go to disk to get
2352 * the new data. We invalidate clean cached page from the region we're
2353 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002354 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002355 */
2356 if (mapping->nrpages) {
2357 written = invalidate_inode_pages2_range(mapping,
2358 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002359 /*
2360 * If a page can not be invalidated, return 0 to fall back
2361 * to buffered write.
2362 */
2363 if (written) {
2364 if (written == -EBUSY)
2365 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002366 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002367 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002368 }
2369
Al Viro26978b82014-03-10 14:08:45 -04002370 data = *from;
Omar Sandoval22c61862015-03-16 04:33:53 -07002371 written = mapping->a_ops->direct_IO(iocb, &data, pos);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002372
2373 /*
2374 * Finally, try again to invalidate clean pages which might have been
2375 * cached by non-direct readahead, or faulted in by get_user_pages()
2376 * if the source of the write was an mmap'ed region of the file
2377 * we're writing. Either one is a pretty crazy thing to do,
2378 * so we don't support it 100%. If this invalidation
2379 * fails, tough, the write still worked...
2380 */
2381 if (mapping->nrpages) {
2382 invalidate_inode_pages2_range(mapping,
2383 pos >> PAGE_CACHE_SHIFT, end);
2384 }
2385
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002387 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002388 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002389 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2390 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 mark_inode_dirty(inode);
2392 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002393 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002395out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 return written;
2397}
2398EXPORT_SYMBOL(generic_file_direct_write);
2399
Nick Piggineb2be182007-10-16 01:24:57 -07002400/*
2401 * Find or create a page at the given pagecache position. Return the locked
2402 * page. This function is specifically for buffered writes.
2403 */
Nick Piggin54566b22009-01-04 12:00:53 -08002404struct page *grab_cache_page_write_begin(struct address_space *mapping,
2405 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002406{
Nick Piggineb2be182007-10-16 01:24:57 -07002407 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07002408 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002409
Nick Piggin54566b22009-01-04 12:00:53 -08002410 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002411 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002412
Mel Gorman2457aec2014-06-04 16:10:31 -07002413 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002414 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002415 if (page)
2416 wait_for_stable_page(page);
2417
Nick Piggineb2be182007-10-16 01:24:57 -07002418 return page;
2419}
Nick Piggin54566b22009-01-04 12:00:53 -08002420EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002421
Al Viro3b93f912014-02-11 21:34:08 -05002422ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002423 struct iov_iter *i, loff_t pos)
2424{
2425 struct address_space *mapping = file->f_mapping;
2426 const struct address_space_operations *a_ops = mapping->a_ops;
2427 long status = 0;
2428 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002429 unsigned int flags = 0;
2430
2431 /*
2432 * Copies from kernel address space cannot fail (NFSD is a big user).
2433 */
Al Viro777eda22014-12-17 04:46:46 -05002434 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002435 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002436
2437 do {
2438 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002439 unsigned long offset; /* Offset into pagecache page */
2440 unsigned long bytes; /* Bytes to write to page */
2441 size_t copied; /* Bytes copied from user */
2442 void *fsdata;
2443
2444 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002445 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2446 iov_iter_count(i));
2447
2448again:
Nick Pigginafddba42007-10-16 01:25:01 -07002449 /*
2450 * Bring in the user page that we will copy from _first_.
2451 * Otherwise there's a nasty deadlock on copying from the
2452 * same page as we're writing to, without it being marked
2453 * up-to-date.
2454 *
2455 * Not only is this an optimisation, but it is also required
2456 * to check that the address is actually valid, when atomic
2457 * usercopies are used, below.
2458 */
2459 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2460 status = -EFAULT;
2461 break;
2462 }
2463
Nick Piggin674b8922007-10-16 01:25:03 -07002464 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002465 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002466 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002467 break;
2468
anfei zhou931e80e2010-02-02 13:44:02 -08002469 if (mapping_writably_mapped(mapping))
2470 flush_dcache_page(page);
2471
Nick Pigginafddba42007-10-16 01:25:01 -07002472 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002473 flush_dcache_page(page);
2474
2475 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2476 page, fsdata);
2477 if (unlikely(status < 0))
2478 break;
2479 copied = status;
2480
2481 cond_resched();
2482
Nick Piggin124d3b72008-02-02 15:01:17 +01002483 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002484 if (unlikely(copied == 0)) {
2485 /*
2486 * If we were unable to copy any data at all, we must
2487 * fall back to a single segment length write.
2488 *
2489 * If we didn't fallback here, we could livelock
2490 * because not all segments in the iov can be copied at
2491 * once without a pagefault.
2492 */
2493 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2494 iov_iter_single_seg_count(i));
2495 goto again;
2496 }
Nick Pigginafddba42007-10-16 01:25:01 -07002497 pos += copied;
2498 written += copied;
2499
2500 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002501 if (fatal_signal_pending(current)) {
2502 status = -EINTR;
2503 break;
2504 }
Nick Pigginafddba42007-10-16 01:25:01 -07002505 } while (iov_iter_count(i));
2506
2507 return written ? written : status;
2508}
Al Viro3b93f912014-02-11 21:34:08 -05002509EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Jan Karae4dd9de2009-08-17 18:10:06 +02002511/**
Al Viro81742022014-04-03 03:17:43 -04002512 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002513 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002514 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002515 *
2516 * This function does all the work needed for actually writing data to a
2517 * file. It does all basic checks, removes SUID from the file, updates
2518 * modification times and calls proper subroutines depending on whether we
2519 * do direct IO or a standard buffered write.
2520 *
2521 * It expects i_mutex to be grabbed unless we work on a block device or similar
2522 * object which does not need locking at all.
2523 *
2524 * This function does *not* take care of syncing data in case of O_SYNC write.
2525 * A caller has to handle it. This is mainly due to the fact that we want to
2526 * avoid syncing under i_mutex.
2527 */
Al Viro81742022014-04-03 03:17:43 -04002528ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529{
2530 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002531 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002533 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002535 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002538 current->backing_dev_info = inode_to_bdi(inode);
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002539 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 if (err)
2541 goto out;
2542
Josef Bacikc3b2da32012-03-26 09:59:21 -04002543 err = file_update_time(file);
2544 if (err)
2545 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Al Viro2ba48ce2015-04-09 13:52:01 -04002547 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002548 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002549
Al Viro0b8def92015-04-07 10:22:53 -04002550 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002551 /*
2552 * If the write stopped short of completing, fall back to
2553 * buffered writes. Some filesystems do this for writes to
2554 * holes, for example. For DAX files, a buffered write will
2555 * not succeed (even if it did, DAX does not handle dirty
2556 * page-cache pages correctly).
2557 */
Al Viro0b8def92015-04-07 10:22:53 -04002558 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002560
Al Viro0b8def92015-04-07 10:22:53 -04002561 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002562 /*
Al Viro3b93f912014-02-11 21:34:08 -05002563 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002564 * then we want to return the number of bytes which were
2565 * direct-written, or the error code if that was zero. Note
2566 * that this differs from normal direct-io semantics, which
2567 * will return -EFOO even if some bytes were written.
2568 */
Al Viro60bb4522014-08-08 12:39:16 -04002569 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002570 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002571 goto out;
2572 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002573 /*
2574 * We need to ensure that the page cache pages are written to
2575 * disk and invalidated to preserve the expected O_DIRECT
2576 * semantics.
2577 */
Al Viro3b93f912014-02-11 21:34:08 -05002578 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002579 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002580 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002581 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002582 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002583 invalidate_mapping_pages(mapping,
2584 pos >> PAGE_CACHE_SHIFT,
2585 endbyte >> PAGE_CACHE_SHIFT);
2586 } else {
2587 /*
2588 * We don't know how much we wrote, so just return
2589 * the number of bytes which were direct-written
2590 */
2591 }
2592 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002593 written = generic_perform_write(file, from, iocb->ki_pos);
2594 if (likely(written > 0))
2595 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597out:
2598 current->backing_dev_info = NULL;
2599 return written ? written : err;
2600}
Al Viro81742022014-04-03 03:17:43 -04002601EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
Jan Karae4dd9de2009-08-17 18:10:06 +02002603/**
Al Viro81742022014-04-03 03:17:43 -04002604 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002605 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002606 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002607 *
Al Viro81742022014-04-03 03:17:43 -04002608 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002609 * filesystems. It takes care of syncing the file in case of O_SYNC file
2610 * and acquires i_mutex as needed.
2611 */
Al Viro81742022014-04-03 03:17:43 -04002612ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613{
2614 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002615 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002618 mutex_lock(&inode->i_mutex);
Al Viro3309dd02015-04-09 12:55:47 -04002619 ret = generic_write_checks(iocb, from);
2620 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002621 ret = __generic_file_write_iter(iocb, from);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002622 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623
Christoph Hellwig02afc272013-09-04 15:04:40 +02002624 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 ssize_t err;
2626
Al Virod311d792014-02-09 15:18:09 -05002627 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2628 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 ret = err;
2630 }
2631 return ret;
2632}
Al Viro81742022014-04-03 03:17:43 -04002633EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
David Howellscf9a2ae2006-08-29 19:05:54 +01002635/**
2636 * try_to_release_page() - release old fs-specific metadata on a page
2637 *
2638 * @page: the page which the kernel is trying to free
2639 * @gfp_mask: memory allocation flags (and I/O mode)
2640 *
2641 * The address_space is to try to release any data against the page
2642 * (presumably at page->private). If the release was successful, return `1'.
2643 * Otherwise return zero.
2644 *
David Howells266cf652009-04-03 16:42:36 +01002645 * This may also be called if PG_fscache is set on a page, indicating that the
2646 * page is known to the local caching routines.
2647 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002648 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002649 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002650 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002651 */
2652int try_to_release_page(struct page *page, gfp_t gfp_mask)
2653{
2654 struct address_space * const mapping = page->mapping;
2655
2656 BUG_ON(!PageLocked(page));
2657 if (PageWriteback(page))
2658 return 0;
2659
2660 if (mapping && mapping->a_ops->releasepage)
2661 return mapping->a_ops->releasepage(page, gfp_mask);
2662 return try_to_free_buffers(page);
2663}
2664
2665EXPORT_SYMBOL(try_to_release_page);