blob: ad7242043bdb8b74872e536b61d01ca05a1de6b3 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Johannes Weiner00501b52014-08-08 14:19:20 -070034#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060036#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070037#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080038#include "internal.h"
39
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070040#define CREATE_TRACE_POINTS
41#include <trace/events/filemap.h>
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * FIXME: remove all knowledge of the buffer layer from the core VM
45 */
Jan Kara148f9482009-08-17 19:52:36 +020046#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <asm/mman.h>
49
50/*
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * though.
53 *
54 * Shared mappings now work. 15.8.1995 Bruno.
55 *
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 *
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 */
61
62/*
63 * Lock ordering:
64 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080065 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070067 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080070 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080071 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 *
73 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080074 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070075 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
Al Viroccad2362014-02-11 22:36:48 -050081 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070082 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060084 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110085 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * ->mapping->tree_lock (__sync_single_inode)
87 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080088 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080099 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800108 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700109 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 */
111
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700112static void page_cache_tree_delete(struct address_space *mapping,
113 struct page *page, void *shadow)
114{
Johannes Weiner449dd692014-04-03 14:47:56 -0700115 struct radix_tree_node *node;
116 unsigned long index;
117 unsigned int offset;
118 unsigned int tag;
119 void **slot;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700120
Johannes Weiner449dd692014-04-03 14:47:56 -0700121 VM_BUG_ON(!PageLocked(page));
122
123 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
124
125 if (shadow) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700126 mapping->nrshadows++;
127 /*
128 * Make sure the nrshadows update is committed before
129 * the nrpages update so that final truncate racing
130 * with reclaim does not see both counters 0 at the
131 * same time and miss a shadow entry.
132 */
133 smp_wmb();
Johannes Weiner449dd692014-04-03 14:47:56 -0700134 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700135 mapping->nrpages--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700136
137 if (!node) {
138 /* Clear direct pointer tags in root node */
139 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
140 radix_tree_replace_slot(slot, shadow);
141 return;
142 }
143
144 /* Clear tree tags for the removed page */
145 index = page->index;
146 offset = index & RADIX_TREE_MAP_MASK;
147 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
148 if (test_bit(offset, node->tags[tag]))
149 radix_tree_tag_clear(&mapping->page_tree, index, tag);
150 }
151
152 /* Delete page, swap shadow entry */
153 radix_tree_replace_slot(slot, shadow);
154 workingset_node_pages_dec(node);
155 if (shadow)
156 workingset_node_shadows_inc(node);
157 else
158 if (__radix_tree_delete_node(&mapping->page_tree, node))
159 return;
160
161 /*
162 * Track node that only contains shadow entries.
163 *
164 * Avoid acquiring the list_lru lock if already tracked. The
165 * list_empty() test is safe as node->private_list is
166 * protected by mapping->tree_lock.
167 */
168 if (!workingset_node_pages(node) &&
169 list_empty(&node->private_list)) {
170 node->private_data = mapping;
171 list_lru_add(&workingset_shadow_nodes, &node->private_list);
172 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700173}
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175/*
Minchan Kime64a7822011-03-22 16:32:44 -0700176 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700178 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700180void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
182 struct address_space *mapping = page->mapping;
183
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700184 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600185 /*
186 * if we're uptodate, flush out into the cleancache, otherwise
187 * invalidate any existing cleancache entries. We can't leave
188 * stale data around in the cleancache once our page is gone
189 */
190 if (PageUptodate(page) && PageMappedToDisk(page))
191 cleancache_put_page(page);
192 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400193 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600194
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700195 page_cache_tree_delete(mapping, page, shadow);
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700198 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700199
Christoph Lameter347ce432006-06-30 01:55:35 -0700200 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700201 if (PageSwapBacked(page))
202 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700203 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800204
205 /*
206 * Some filesystems seem to re-dirty the page even after
207 * the VM has canceled the dirty bit (eg ext3 journaling).
208 *
209 * Fix it up by doing a final dirty accounting check after
210 * having removed the page entirely.
211 */
212 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
213 dec_zone_page_state(page, NR_FILE_DIRTY);
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100214 dec_bdi_stat(inode_to_bdi(mapping->host), BDI_RECLAIMABLE);
Linus Torvalds3a692792007-12-19 14:05:13 -0800215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
Minchan Kim702cfbf2011-03-22 16:32:43 -0700218/**
219 * delete_from_page_cache - delete page from page cache
220 * @page: the page which the kernel is trying to remove from page cache
221 *
222 * This must be called only on pages that have been verified to be in the page
223 * cache and locked. It will never put the page into the free list, the caller
224 * has a reference on the page.
225 */
226void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
228 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500229 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Matt Mackallcd7619d2005-05-01 08:59:01 -0700231 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Linus Torvalds6072d132010-12-01 13:35:19 -0500233 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700234 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700235 __delete_from_page_cache(page, NULL);
Nick Piggin19fd6232008-07-25 19:45:32 -0700236 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds6072d132010-12-01 13:35:19 -0500237
238 if (freepage)
239 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700240 page_cache_release(page);
241}
242EXPORT_SYMBOL(delete_from_page_cache);
243
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700244static int filemap_check_errors(struct address_space *mapping)
245{
246 int ret = 0;
247 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700248 if (test_bit(AS_ENOSPC, &mapping->flags) &&
249 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700250 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700251 if (test_bit(AS_EIO, &mapping->flags) &&
252 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700253 ret = -EIO;
254 return ret;
255}
256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700258 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700259 * @mapping: address space structure to write
260 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800261 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700262 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700264 * Start writeback against all of a mapping's dirty pages that lie
265 * within the byte offsets <start, end> inclusive.
266 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700268 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 * these two operations is that if a dirty page/buffer is encountered, it must
270 * be waited upon, and not just skipped over.
271 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800272int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
273 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
275 int ret;
276 struct writeback_control wbc = {
277 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800278 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700279 .range_start = start,
280 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 };
282
283 if (!mapping_cap_writeback_dirty(mapping))
284 return 0;
285
286 ret = do_writepages(mapping, &wbc);
287 return ret;
288}
289
290static inline int __filemap_fdatawrite(struct address_space *mapping,
291 int sync_mode)
292{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700293 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
296int filemap_fdatawrite(struct address_space *mapping)
297{
298 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
299}
300EXPORT_SYMBOL(filemap_fdatawrite);
301
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400302int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800303 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
305 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
306}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400307EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Randy Dunlap485bb992006-06-23 02:03:49 -0700309/**
310 * filemap_flush - mostly a non-blocking flush
311 * @mapping: target address_space
312 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 * This is a mostly non-blocking flush. Not suitable for data-integrity
314 * purposes - I/O may not be started against all dirty pages.
315 */
316int filemap_flush(struct address_space *mapping)
317{
318 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
319}
320EXPORT_SYMBOL(filemap_flush);
321
Randy Dunlap485bb992006-06-23 02:03:49 -0700322/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200323 * filemap_fdatawait_range - wait for writeback to complete
324 * @mapping: address space structure to wait for
325 * @start_byte: offset in bytes where the range starts
326 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700327 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200328 * Walk the list of under-writeback pages of the given address space
329 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200331int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
332 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200334 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
335 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 struct pagevec pvec;
337 int nr_pages;
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700338 int ret2, ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200340 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700341 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
343 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 while ((index <= end) &&
345 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
346 PAGECACHE_TAG_WRITEBACK,
347 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
348 unsigned i;
349
350 for (i = 0; i < nr_pages; i++) {
351 struct page *page = pvec.pages[i];
352
353 /* until radix tree lookup accepts end_index */
354 if (page->index > end)
355 continue;
356
357 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800358 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 ret = -EIO;
360 }
361 pagevec_release(&pvec);
362 cond_resched();
363 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700364out:
365 ret2 = filemap_check_errors(mapping);
366 if (!ret)
367 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
369 return ret;
370}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200371EXPORT_SYMBOL(filemap_fdatawait_range);
372
373/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700374 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700376 *
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
380int filemap_fdatawait(struct address_space *mapping)
381{
382 loff_t i_size = i_size_read(mapping->host);
383
384 if (i_size == 0)
385 return 0;
386
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200387 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389EXPORT_SYMBOL(filemap_fdatawait);
390
391int filemap_write_and_wait(struct address_space *mapping)
392{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800393 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800396 err = filemap_fdatawrite(mapping);
397 /*
398 * Even if the above returned error, the pages may be
399 * written partially (e.g. -ENOSPC), so we wait for it.
400 * But the -EIO is special case, it may indicate the worst
401 * thing (e.g. bug) happened, so we avoid waiting for it.
402 */
403 if (err != -EIO) {
404 int err2 = filemap_fdatawait(mapping);
405 if (!err)
406 err = err2;
407 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700408 } else {
409 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800411 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800413EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Randy Dunlap485bb992006-06-23 02:03:49 -0700415/**
416 * filemap_write_and_wait_range - write out & wait on a file range
417 * @mapping: the address_space for the pages
418 * @lstart: offset in bytes where the range starts
419 * @lend: offset in bytes where the range ends (inclusive)
420 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800421 * Write out and wait upon file offsets lstart->lend, inclusive.
422 *
423 * Note that `lend' is inclusive (describes the last byte to be written) so
424 * that this function can be used to write to the very end-of-file (end = -1).
425 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426int filemap_write_and_wait_range(struct address_space *mapping,
427 loff_t lstart, loff_t lend)
428{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800429 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800432 err = __filemap_fdatawrite_range(mapping, lstart, lend,
433 WB_SYNC_ALL);
434 /* See comment of filemap_write_and_wait() */
435 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200436 int err2 = filemap_fdatawait_range(mapping,
437 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800438 if (!err)
439 err = err2;
440 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700441 } else {
442 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800444 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
Chris Masonf6995582009-04-15 13:22:37 -0400446EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Randy Dunlap485bb992006-06-23 02:03:49 -0700448/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700449 * replace_page_cache_page - replace a pagecache page with a new one
450 * @old: page to be replaced
451 * @new: page to replace with
452 * @gfp_mask: allocation mode
453 *
454 * This function replaces a page in the pagecache with a new one. On
455 * success it acquires the pagecache reference for the new page and
456 * drops it for the old page. Both the old and new pages must be
457 * locked. This function does not add the new page to the LRU, the
458 * caller must do that.
459 *
460 * The remove + add is atomic. The only way this function can fail is
461 * memory allocation failure.
462 */
463int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
464{
465 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700466
Sasha Levin309381fea2014-01-23 15:52:54 -0800467 VM_BUG_ON_PAGE(!PageLocked(old), old);
468 VM_BUG_ON_PAGE(!PageLocked(new), new);
469 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700470
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700471 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
472 if (!error) {
473 struct address_space *mapping = old->mapping;
474 void (*freepage)(struct page *);
475
476 pgoff_t offset = old->index;
477 freepage = mapping->a_ops->freepage;
478
479 page_cache_get(new);
480 new->mapping = mapping;
481 new->index = offset;
482
483 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700484 __delete_from_page_cache(old, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700485 error = radix_tree_insert(&mapping->page_tree, offset, new);
486 BUG_ON(error);
487 mapping->nrpages++;
488 __inc_zone_page_state(new, NR_FILE_PAGES);
489 if (PageSwapBacked(new))
490 __inc_zone_page_state(new, NR_SHMEM);
491 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700492 mem_cgroup_migrate(old, new, true);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700493 radix_tree_preload_end();
494 if (freepage)
495 freepage(old);
496 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700497 }
498
499 return error;
500}
501EXPORT_SYMBOL_GPL(replace_page_cache_page);
502
Johannes Weiner0cd61442014-04-03 14:47:46 -0700503static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700504 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700505{
Johannes Weiner449dd692014-04-03 14:47:56 -0700506 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700507 void **slot;
508 int error;
509
Johannes Weiner449dd692014-04-03 14:47:56 -0700510 error = __radix_tree_create(&mapping->page_tree, page->index,
511 &node, &slot);
512 if (error)
513 return error;
514 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700515 void *p;
516
517 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
518 if (!radix_tree_exceptional_entry(p))
519 return -EEXIST;
Johannes Weinera5289102014-04-03 14:47:51 -0700520 if (shadowp)
521 *shadowp = p;
Johannes Weiner449dd692014-04-03 14:47:56 -0700522 mapping->nrshadows--;
523 if (node)
524 workingset_node_shadows_dec(node);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700525 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700526 radix_tree_replace_slot(slot, page);
527 mapping->nrpages++;
528 if (node) {
529 workingset_node_pages_inc(node);
530 /*
531 * Don't track node that contains actual pages.
532 *
533 * Avoid acquiring the list_lru lock if already
534 * untracked. The list_empty() test is safe as
535 * node->private_list is protected by
536 * mapping->tree_lock.
537 */
538 if (!list_empty(&node->private_list))
539 list_lru_del(&workingset_shadow_nodes,
540 &node->private_list);
541 }
542 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700543}
544
Johannes Weinera5289102014-04-03 14:47:51 -0700545static int __add_to_page_cache_locked(struct page *page,
546 struct address_space *mapping,
547 pgoff_t offset, gfp_t gfp_mask,
548 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
Johannes Weiner00501b52014-08-08 14:19:20 -0700550 int huge = PageHuge(page);
551 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700552 int error;
553
Sasha Levin309381fea2014-01-23 15:52:54 -0800554 VM_BUG_ON_PAGE(!PageLocked(page), page);
555 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700556
Johannes Weiner00501b52014-08-08 14:19:20 -0700557 if (!huge) {
558 error = mem_cgroup_try_charge(page, current->mm,
559 gfp_mask, &memcg);
560 if (error)
561 return error;
562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Jan Kara5e4c0d972013-09-11 14:26:05 -0700564 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700565 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700566 if (!huge)
567 mem_cgroup_cancel_charge(page, memcg);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700568 return error;
569 }
570
571 page_cache_get(page);
572 page->mapping = mapping;
573 page->index = offset;
574
575 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700576 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700577 radix_tree_preload_end();
578 if (unlikely(error))
579 goto err_insert;
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700580 __inc_zone_page_state(page, NR_FILE_PAGES);
581 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700582 if (!huge)
583 mem_cgroup_commit_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700584 trace_mm_filemap_add_to_page_cache(page);
585 return 0;
586err_insert:
587 page->mapping = NULL;
588 /* Leave page->index set: truncation relies upon it */
589 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700590 if (!huge)
591 mem_cgroup_cancel_charge(page, memcg);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700592 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 return error;
594}
Johannes Weinera5289102014-04-03 14:47:51 -0700595
596/**
597 * add_to_page_cache_locked - add a locked page to the pagecache
598 * @page: page to add
599 * @mapping: the page's address_space
600 * @offset: page index
601 * @gfp_mask: page allocation mode
602 *
603 * This function is used to add a page to the pagecache. It must be locked.
604 * This function does not add the page to the LRU. The caller must do that.
605 */
606int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
607 pgoff_t offset, gfp_t gfp_mask)
608{
609 return __add_to_page_cache_locked(page, mapping, offset,
610 gfp_mask, NULL);
611}
Nick Piggine2867812008-07-25 19:45:30 -0700612EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400615 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Johannes Weinera5289102014-04-03 14:47:51 -0700617 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700618 int ret;
619
Johannes Weinera5289102014-04-03 14:47:51 -0700620 __set_page_locked(page);
621 ret = __add_to_page_cache_locked(page, mapping, offset,
622 gfp_mask, &shadow);
623 if (unlikely(ret))
624 __clear_page_locked(page);
625 else {
626 /*
627 * The page might have been evicted from cache only
628 * recently, in which case it should be activated like
629 * any other repeatedly accessed page.
630 */
631 if (shadow && workingset_refault(shadow)) {
632 SetPageActive(page);
633 workingset_activation(page);
634 } else
635 ClearPageActive(page);
636 lru_cache_add(page);
637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 return ret;
639}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300640EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Paul Jackson44110fe2006-03-24 03:16:04 -0800642#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700643struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800644{
Miao Xiec0ff7452010-05-24 14:32:08 -0700645 int n;
646 struct page *page;
647
Paul Jackson44110fe2006-03-24 03:16:04 -0800648 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700649 unsigned int cpuset_mems_cookie;
650 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700651 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700652 n = cpuset_mem_spread_node();
653 page = alloc_pages_exact_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700654 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700655
Miao Xiec0ff7452010-05-24 14:32:08 -0700656 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800657 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700658 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800659}
Nick Piggin2ae88142006-10-28 10:38:23 -0700660EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800661#endif
662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663/*
664 * In order to wait for pages to become available there must be
665 * waitqueues associated with pages. By using a hash table of
666 * waitqueues where the bucket discipline is to maintain all
667 * waiters on the same queue and wake all when any of the pages
668 * become available, and for the woken contexts to check to be
669 * sure the appropriate page became available, this saves space
670 * at a cost of "thundering herd" phenomena during rare hash
671 * collisions.
672 */
NeilBrowna4796e32014-09-24 11:28:32 +1000673wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
675 const struct zone *zone = page_zone(page);
676
677 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
678}
NeilBrowna4796e32014-09-24 11:28:32 +1000679EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Harvey Harrison920c7a52008-02-04 22:29:26 -0800681void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
683 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
684
685 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000686 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 TASK_UNINTERRUPTIBLE);
688}
689EXPORT_SYMBOL(wait_on_page_bit);
690
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700691int wait_on_page_bit_killable(struct page *page, int bit_nr)
692{
693 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
694
695 if (!test_bit(bit_nr, &page->flags))
696 return 0;
697
698 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000699 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700700}
701
NeilBrowncbbce822014-09-25 13:55:19 +1000702int wait_on_page_bit_killable_timeout(struct page *page,
703 int bit_nr, unsigned long timeout)
704{
705 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
706
707 wait.key.timeout = jiffies + timeout;
708 if (!test_bit(bit_nr, &page->flags))
709 return 0;
710 return __wait_on_bit(page_waitqueue(page), &wait,
711 bit_wait_io_timeout, TASK_KILLABLE);
712}
713EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100716 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700717 * @page: Page defining the wait queue of interest
718 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100719 *
720 * Add an arbitrary @waiter to the wait queue for the nominated @page.
721 */
722void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
723{
724 wait_queue_head_t *q = page_waitqueue(page);
725 unsigned long flags;
726
727 spin_lock_irqsave(&q->lock, flags);
728 __add_wait_queue(q, waiter);
729 spin_unlock_irqrestore(&q->lock, flags);
730}
731EXPORT_SYMBOL_GPL(add_page_wait_queue);
732
733/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700734 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 * @page: the page
736 *
737 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
738 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900739 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
741 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700742 * The mb is necessary to enforce ordering between the clear_bit and the read
743 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800745void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
Sasha Levin309381fea2014-01-23 15:52:54 -0800747 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700748 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100749 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 wake_up_page(page, PG_locked);
751}
752EXPORT_SYMBOL(unlock_page);
753
Randy Dunlap485bb992006-06-23 02:03:49 -0700754/**
755 * end_page_writeback - end writeback against a page
756 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 */
758void end_page_writeback(struct page *page)
759{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700760 /*
761 * TestClearPageReclaim could be used here but it is an atomic
762 * operation and overkill in this particular case. Failing to
763 * shuffle a page marked for immediate reclaim is too mild to
764 * justify taking an atomic operation penalty at the end of
765 * ever page writeback.
766 */
767 if (PageReclaim(page)) {
768 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700769 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700770 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700771
772 if (!test_clear_page_writeback(page))
773 BUG();
774
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100775 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 wake_up_page(page, PG_writeback);
777}
778EXPORT_SYMBOL(end_page_writeback);
779
Matthew Wilcox57d99842014-06-04 16:07:45 -0700780/*
781 * After completing I/O on a page, call this routine to update the page
782 * flags appropriately
783 */
784void page_endio(struct page *page, int rw, int err)
785{
786 if (rw == READ) {
787 if (!err) {
788 SetPageUptodate(page);
789 } else {
790 ClearPageUptodate(page);
791 SetPageError(page);
792 }
793 unlock_page(page);
794 } else { /* rw == WRITE */
795 if (err) {
796 SetPageError(page);
797 if (page->mapping)
798 mapping_set_error(page->mapping, err);
799 }
800 end_page_writeback(page);
801 }
802}
803EXPORT_SYMBOL_GPL(page_endio);
804
Randy Dunlap485bb992006-06-23 02:03:49 -0700805/**
806 * __lock_page - get a lock on the page, assuming we need to sleep to get it
807 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800809void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
811 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
812
NeilBrown74316202014-07-07 15:16:04 +1000813 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 TASK_UNINTERRUPTIBLE);
815}
816EXPORT_SYMBOL(__lock_page);
817
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800818int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500819{
820 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
821
822 return __wait_on_bit_lock(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000823 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500824}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300825EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500826
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700827/*
828 * Return values:
829 * 1 - page is locked; mmap_sem is still held.
830 * 0 - page is not locked.
831 * mmap_sem has been released (up_read()), unless flags had both
832 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
833 * which case mmap_sem is still held.
834 *
835 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
836 * with the page locked and the mmap_sem unperturbed.
837 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700838int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
839 unsigned int flags)
840{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700841 if (flags & FAULT_FLAG_ALLOW_RETRY) {
842 /*
843 * CAUTION! In this case, mmap_sem is not released
844 * even though return 0.
845 */
846 if (flags & FAULT_FLAG_RETRY_NOWAIT)
847 return 0;
848
849 up_read(&mm->mmap_sem);
850 if (flags & FAULT_FLAG_KILLABLE)
851 wait_on_page_locked_killable(page);
852 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700853 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700854 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700855 } else {
856 if (flags & FAULT_FLAG_KILLABLE) {
857 int ret;
858
859 ret = __lock_page_killable(page);
860 if (ret) {
861 up_read(&mm->mmap_sem);
862 return 0;
863 }
864 } else
865 __lock_page(page);
866 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700867 }
868}
869
Randy Dunlap485bb992006-06-23 02:03:49 -0700870/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700871 * page_cache_next_hole - find the next hole (not-present entry)
872 * @mapping: mapping
873 * @index: index
874 * @max_scan: maximum range to search
875 *
876 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
877 * lowest indexed hole.
878 *
879 * Returns: the index of the hole if found, otherwise returns an index
880 * outside of the set specified (in which case 'return - index >=
881 * max_scan' will be true). In rare cases of index wrap-around, 0 will
882 * be returned.
883 *
884 * page_cache_next_hole may be called under rcu_read_lock. However,
885 * like radix_tree_gang_lookup, this will not atomically search a
886 * snapshot of the tree at a single point in time. For example, if a
887 * hole is created at index 5, then subsequently a hole is created at
888 * index 10, page_cache_next_hole covering both indexes may return 10
889 * if called under rcu_read_lock.
890 */
891pgoff_t page_cache_next_hole(struct address_space *mapping,
892 pgoff_t index, unsigned long max_scan)
893{
894 unsigned long i;
895
896 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700897 struct page *page;
898
899 page = radix_tree_lookup(&mapping->page_tree, index);
900 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700901 break;
902 index++;
903 if (index == 0)
904 break;
905 }
906
907 return index;
908}
909EXPORT_SYMBOL(page_cache_next_hole);
910
911/**
912 * page_cache_prev_hole - find the prev hole (not-present entry)
913 * @mapping: mapping
914 * @index: index
915 * @max_scan: maximum range to search
916 *
917 * Search backwards in the range [max(index-max_scan+1, 0), index] for
918 * the first hole.
919 *
920 * Returns: the index of the hole if found, otherwise returns an index
921 * outside of the set specified (in which case 'index - return >=
922 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
923 * will be returned.
924 *
925 * page_cache_prev_hole may be called under rcu_read_lock. However,
926 * like radix_tree_gang_lookup, this will not atomically search a
927 * snapshot of the tree at a single point in time. For example, if a
928 * hole is created at index 10, then subsequently a hole is created at
929 * index 5, page_cache_prev_hole covering both indexes may return 5 if
930 * called under rcu_read_lock.
931 */
932pgoff_t page_cache_prev_hole(struct address_space *mapping,
933 pgoff_t index, unsigned long max_scan)
934{
935 unsigned long i;
936
937 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700938 struct page *page;
939
940 page = radix_tree_lookup(&mapping->page_tree, index);
941 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700942 break;
943 index--;
944 if (index == ULONG_MAX)
945 break;
946 }
947
948 return index;
949}
950EXPORT_SYMBOL(page_cache_prev_hole);
951
952/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700953 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -0700954 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -0700955 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -0700956 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700957 * Looks up the page cache slot at @mapping & @offset. If there is a
958 * page cache page, it is returned with an increased refcount.
959 *
Johannes Weiner139b6a62014-05-06 12:50:05 -0700960 * If the slot holds a shadow entry of a previously evicted page, or a
961 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -0700962 *
963 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700965struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966{
Nick Piggina60637c2008-07-25 19:45:31 -0700967 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 struct page *page;
969
Nick Piggina60637c2008-07-25 19:45:31 -0700970 rcu_read_lock();
971repeat:
972 page = NULL;
973 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
974 if (pagep) {
975 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800976 if (unlikely(!page))
977 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700978 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700979 if (radix_tree_deref_retry(page))
980 goto repeat;
981 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -0700982 * A shadow entry of a recently evicted page,
983 * or a swap entry from shmem/tmpfs. Return
984 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700985 */
986 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700987 }
Nick Piggina60637c2008-07-25 19:45:31 -0700988 if (!page_cache_get_speculative(page))
989 goto repeat;
990
991 /*
992 * Has the page moved?
993 * This is part of the lockless pagecache protocol. See
994 * include/linux/pagemap.h for details.
995 */
996 if (unlikely(page != *pagep)) {
997 page_cache_release(page);
998 goto repeat;
999 }
1000 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001001out:
Nick Piggina60637c2008-07-25 19:45:31 -07001002 rcu_read_unlock();
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 return page;
1005}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001006EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Randy Dunlap485bb992006-06-23 02:03:49 -07001008/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001009 * find_lock_entry - locate, pin and lock a page cache entry
1010 * @mapping: the address_space to search
1011 * @offset: the page cache index
1012 *
1013 * Looks up the page cache slot at @mapping & @offset. If there is a
1014 * page cache page, it is returned locked and with an increased
1015 * refcount.
1016 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001017 * If the slot holds a shadow entry of a previously evicted page, or a
1018 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001019 *
1020 * Otherwise, %NULL is returned.
1021 *
1022 * find_lock_entry() may sleep.
1023 */
1024struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
1026 struct page *page;
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001029 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001030 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001031 lock_page(page);
1032 /* Has the page been truncated? */
1033 if (unlikely(page->mapping != mapping)) {
1034 unlock_page(page);
1035 page_cache_release(page);
1036 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001038 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 return page;
1041}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001042EXPORT_SYMBOL(find_lock_entry);
1043
1044/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001045 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001046 * @mapping: the address_space to search
1047 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001048 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001049 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001050 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001051 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001052 *
Randy Dunlap75325182014-07-30 16:08:37 -07001053 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001054 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001055 * FGP_ACCESSED: the page will be marked accessed
1056 * FGP_LOCK: Page is return locked
1057 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001058 * @gfp_mask and added to the page cache and the VM's LRU
1059 * list. The page is returned locked and with an increased
1060 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001061 *
1062 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1063 * if the GFP flags specified for FGP_CREAT are atomic.
1064 *
1065 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001066 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001067struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001068 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069{
Nick Piggineb2be182007-10-16 01:24:57 -07001070 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001073 page = find_get_entry(mapping, offset);
1074 if (radix_tree_exceptional_entry(page))
1075 page = NULL;
1076 if (!page)
1077 goto no_page;
1078
1079 if (fgp_flags & FGP_LOCK) {
1080 if (fgp_flags & FGP_NOWAIT) {
1081 if (!trylock_page(page)) {
1082 page_cache_release(page);
1083 return NULL;
1084 }
1085 } else {
1086 lock_page(page);
1087 }
1088
1089 /* Has the page been truncated? */
1090 if (unlikely(page->mapping != mapping)) {
1091 unlock_page(page);
1092 page_cache_release(page);
1093 goto repeat;
1094 }
1095 VM_BUG_ON_PAGE(page->index != offset, page);
1096 }
1097
1098 if (page && (fgp_flags & FGP_ACCESSED))
1099 mark_page_accessed(page);
1100
1101no_page:
1102 if (!page && (fgp_flags & FGP_CREAT)) {
1103 int err;
1104 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001105 gfp_mask |= __GFP_WRITE;
1106 if (fgp_flags & FGP_NOFS)
1107 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001108
Michal Hocko45f87de2014-12-29 20:30:35 +01001109 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001110 if (!page)
1111 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001112
1113 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1114 fgp_flags |= FGP_LOCK;
1115
Hugh Dickinseb39d612014-08-06 16:06:43 -07001116 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001117 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001118 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001119
Michal Hocko45f87de2014-12-29 20:30:35 +01001120 err = add_to_page_cache_lru(page, mapping, offset,
1121 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001122 if (unlikely(err)) {
1123 page_cache_release(page);
1124 page = NULL;
1125 if (err == -EEXIST)
1126 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 return page;
1131}
Mel Gorman2457aec2014-06-04 16:10:31 -07001132EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
1134/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001135 * find_get_entries - gang pagecache lookup
1136 * @mapping: The address_space to search
1137 * @start: The starting page cache index
1138 * @nr_entries: The maximum number of entries
1139 * @entries: Where the resulting entries are placed
1140 * @indices: The cache indices corresponding to the entries in @entries
1141 *
1142 * find_get_entries() will search for and return a group of up to
1143 * @nr_entries entries in the mapping. The entries are placed at
1144 * @entries. find_get_entries() takes a reference against any actual
1145 * pages it returns.
1146 *
1147 * The search returns a group of mapping-contiguous page cache entries
1148 * with ascending indexes. There may be holes in the indices due to
1149 * not-present pages.
1150 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001151 * Any shadow entries of evicted pages, or swap entries from
1152 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001153 *
1154 * find_get_entries() returns the number of pages and shadow entries
1155 * which were found.
1156 */
1157unsigned find_get_entries(struct address_space *mapping,
1158 pgoff_t start, unsigned int nr_entries,
1159 struct page **entries, pgoff_t *indices)
1160{
1161 void **slot;
1162 unsigned int ret = 0;
1163 struct radix_tree_iter iter;
1164
1165 if (!nr_entries)
1166 return 0;
1167
1168 rcu_read_lock();
1169restart:
1170 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1171 struct page *page;
1172repeat:
1173 page = radix_tree_deref_slot(slot);
1174 if (unlikely(!page))
1175 continue;
1176 if (radix_tree_exception(page)) {
1177 if (radix_tree_deref_retry(page))
1178 goto restart;
1179 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001180 * A shadow entry of a recently evicted page,
1181 * or a swap entry from shmem/tmpfs. Return
1182 * it without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001183 */
1184 goto export;
1185 }
1186 if (!page_cache_get_speculative(page))
1187 goto repeat;
1188
1189 /* Has the page moved? */
1190 if (unlikely(page != *slot)) {
1191 page_cache_release(page);
1192 goto repeat;
1193 }
1194export:
1195 indices[ret] = iter.index;
1196 entries[ret] = page;
1197 if (++ret == nr_entries)
1198 break;
1199 }
1200 rcu_read_unlock();
1201 return ret;
1202}
1203
1204/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 * find_get_pages - gang pagecache lookup
1206 * @mapping: The address_space to search
1207 * @start: The starting page index
1208 * @nr_pages: The maximum number of pages
1209 * @pages: Where the resulting pages are placed
1210 *
1211 * find_get_pages() will search for and return a group of up to
1212 * @nr_pages pages in the mapping. The pages are placed at @pages.
1213 * find_get_pages() takes a reference against the returned pages.
1214 *
1215 * The search returns a group of mapping-contiguous pages with ascending
1216 * indexes. There may be holes in the indices due to not-present pages.
1217 *
1218 * find_get_pages() returns the number of pages which were found.
1219 */
1220unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1221 unsigned int nr_pages, struct page **pages)
1222{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001223 struct radix_tree_iter iter;
1224 void **slot;
1225 unsigned ret = 0;
1226
1227 if (unlikely(!nr_pages))
1228 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
Nick Piggina60637c2008-07-25 19:45:31 -07001230 rcu_read_lock();
1231restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001232 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001233 struct page *page;
1234repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001235 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001236 if (unlikely(!page))
1237 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001238
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001239 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001240 if (radix_tree_deref_retry(page)) {
1241 /*
1242 * Transient condition which can only trigger
1243 * when entry at index 0 moves out of or back
1244 * to root: none yet gotten, safe to restart.
1245 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001246 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001247 goto restart;
1248 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001249 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001250 * A shadow entry of a recently evicted page,
1251 * or a swap entry from shmem/tmpfs. Skip
1252 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001253 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001254 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001255 }
Nick Piggina60637c2008-07-25 19:45:31 -07001256
1257 if (!page_cache_get_speculative(page))
1258 goto repeat;
1259
1260 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001261 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001262 page_cache_release(page);
1263 goto repeat;
1264 }
1265
1266 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001267 if (++ret == nr_pages)
1268 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001269 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001270
Nick Piggina60637c2008-07-25 19:45:31 -07001271 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 return ret;
1273}
1274
Jens Axboeebf43502006-04-27 08:46:01 +02001275/**
1276 * find_get_pages_contig - gang contiguous pagecache lookup
1277 * @mapping: The address_space to search
1278 * @index: The starting page index
1279 * @nr_pages: The maximum number of pages
1280 * @pages: Where the resulting pages are placed
1281 *
1282 * find_get_pages_contig() works exactly like find_get_pages(), except
1283 * that the returned number of pages are guaranteed to be contiguous.
1284 *
1285 * find_get_pages_contig() returns the number of pages which were found.
1286 */
1287unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1288 unsigned int nr_pages, struct page **pages)
1289{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001290 struct radix_tree_iter iter;
1291 void **slot;
1292 unsigned int ret = 0;
1293
1294 if (unlikely(!nr_pages))
1295 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001296
Nick Piggina60637c2008-07-25 19:45:31 -07001297 rcu_read_lock();
1298restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001299 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001300 struct page *page;
1301repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001302 page = radix_tree_deref_slot(slot);
1303 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001304 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001305 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001306
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001307 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001308 if (radix_tree_deref_retry(page)) {
1309 /*
1310 * Transient condition which can only trigger
1311 * when entry at index 0 moves out of or back
1312 * to root: none yet gotten, safe to restart.
1313 */
1314 goto restart;
1315 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001316 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001317 * A shadow entry of a recently evicted page,
1318 * or a swap entry from shmem/tmpfs. Stop
1319 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001320 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001321 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001322 }
Nick Piggina60637c2008-07-25 19:45:31 -07001323
Nick Piggina60637c2008-07-25 19:45:31 -07001324 if (!page_cache_get_speculative(page))
1325 goto repeat;
1326
1327 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001328 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001329 page_cache_release(page);
1330 goto repeat;
1331 }
1332
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001333 /*
1334 * must check mapping and index after taking the ref.
1335 * otherwise we can get both false positives and false
1336 * negatives, which is just confusing to the caller.
1337 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001338 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001339 page_cache_release(page);
1340 break;
1341 }
1342
Nick Piggina60637c2008-07-25 19:45:31 -07001343 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001344 if (++ret == nr_pages)
1345 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001346 }
Nick Piggina60637c2008-07-25 19:45:31 -07001347 rcu_read_unlock();
1348 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001349}
David Howellsef71c152007-05-09 02:33:44 -07001350EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001351
Randy Dunlap485bb992006-06-23 02:03:49 -07001352/**
1353 * find_get_pages_tag - find and return pages that match @tag
1354 * @mapping: the address_space to search
1355 * @index: the starting page index
1356 * @tag: the tag index
1357 * @nr_pages: the maximum number of pages
1358 * @pages: where the resulting pages are placed
1359 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001361 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 */
1363unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1364 int tag, unsigned int nr_pages, struct page **pages)
1365{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001366 struct radix_tree_iter iter;
1367 void **slot;
1368 unsigned ret = 0;
1369
1370 if (unlikely(!nr_pages))
1371 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Nick Piggina60637c2008-07-25 19:45:31 -07001373 rcu_read_lock();
1374restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001375 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1376 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001377 struct page *page;
1378repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001379 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001380 if (unlikely(!page))
1381 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001382
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001383 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001384 if (radix_tree_deref_retry(page)) {
1385 /*
1386 * Transient condition which can only trigger
1387 * when entry at index 0 moves out of or back
1388 * to root: none yet gotten, safe to restart.
1389 */
1390 goto restart;
1391 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001392 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001393 * A shadow entry of a recently evicted page.
1394 *
1395 * Those entries should never be tagged, but
1396 * this tree walk is lockless and the tags are
1397 * looked up in bulk, one radix tree node at a
1398 * time, so there is a sizable window for page
1399 * reclaim to evict a page we saw tagged.
1400 *
1401 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001402 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001403 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001404 }
Nick Piggina60637c2008-07-25 19:45:31 -07001405
1406 if (!page_cache_get_speculative(page))
1407 goto repeat;
1408
1409 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001410 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001411 page_cache_release(page);
1412 goto repeat;
1413 }
1414
1415 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001416 if (++ret == nr_pages)
1417 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001418 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001419
Nick Piggina60637c2008-07-25 19:45:31 -07001420 rcu_read_unlock();
1421
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 if (ret)
1423 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 return ret;
1426}
David Howellsef71c152007-05-09 02:33:44 -07001427EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001429/*
1430 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1431 * a _large_ part of the i/o request. Imagine the worst scenario:
1432 *
1433 * ---R__________________________________________B__________
1434 * ^ reading here ^ bad block(assume 4k)
1435 *
1436 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1437 * => failing the whole request => read(R) => read(R+1) =>
1438 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1439 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1440 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1441 *
1442 * It is going insane. Fix it by quickly scaling down the readahead size.
1443 */
1444static void shrink_readahead_size_eio(struct file *filp,
1445 struct file_ra_state *ra)
1446{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001447 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001448}
1449
Randy Dunlap485bb992006-06-23 02:03:49 -07001450/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001451 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001452 * @filp: the file to read
1453 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001454 * @iter: data destination
1455 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001456 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001458 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 *
1460 * This is really ugly. But the goto's actually try to clarify some
1461 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 */
Al Viro6e58e792014-02-03 17:07:03 -05001463static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1464 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001466 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001468 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001469 pgoff_t index;
1470 pgoff_t last_index;
1471 pgoff_t prev_index;
1472 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001473 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001474 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001477 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1478 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Al Viro6e58e792014-02-03 17:07:03 -05001479 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 offset = *ppos & ~PAGE_CACHE_MASK;
1481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 for (;;) {
1483 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001484 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001485 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 unsigned long nr, ret;
1487
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489find_page:
1490 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001491 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001492 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001493 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001494 index, last_index - index);
1495 page = find_get_page(mapping, index);
1496 if (unlikely(page == NULL))
1497 goto no_cached_page;
1498 }
1499 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001500 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001501 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001502 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001504 if (!PageUptodate(page)) {
1505 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1506 !mapping->a_ops->is_partially_uptodate)
1507 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001508 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001509 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001510 /* Did it get truncated before we got the lock? */
1511 if (!page->mapping)
1512 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001513 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001514 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001515 goto page_not_up_to_date_locked;
1516 unlock_page(page);
1517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001519 /*
1520 * i_size must be checked after we know the page is Uptodate.
1521 *
1522 * Checking i_size after the check allows us to calculate
1523 * the correct value for "nr", which means the zero-filled
1524 * part of the page is not copied back to userspace (unless
1525 * another truncate extends the file - this is desired though).
1526 */
1527
1528 isize = i_size_read(inode);
1529 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1530 if (unlikely(!isize || index > end_index)) {
1531 page_cache_release(page);
1532 goto out;
1533 }
1534
1535 /* nr is the maximum number of bytes to copy from this page */
1536 nr = PAGE_CACHE_SIZE;
1537 if (index == end_index) {
1538 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1539 if (nr <= offset) {
1540 page_cache_release(page);
1541 goto out;
1542 }
1543 }
1544 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
1546 /* If users can be writing to this page using arbitrary
1547 * virtual addresses, take care about potential aliasing
1548 * before reading the page on the kernel side.
1549 */
1550 if (mapping_writably_mapped(mapping))
1551 flush_dcache_page(page);
1552
1553 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001554 * When a sequential read accesses a page several times,
1555 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 */
Jan Karaec0f1632007-05-06 14:49:25 -07001557 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 mark_page_accessed(page);
1559 prev_index = index;
1560
1561 /*
1562 * Ok, we have the page, and it's up-to-date, so
1563 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 */
Al Viro6e58e792014-02-03 17:07:03 -05001565
1566 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 offset += ret;
1568 index += offset >> PAGE_CACHE_SHIFT;
1569 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001570 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001573 written += ret;
1574 if (!iov_iter_count(iter))
1575 goto out;
1576 if (ret < nr) {
1577 error = -EFAULT;
1578 goto out;
1579 }
1580 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582page_not_up_to_date:
1583 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001584 error = lock_page_killable(page);
1585 if (unlikely(error))
1586 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001588page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001589 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 if (!page->mapping) {
1591 unlock_page(page);
1592 page_cache_release(page);
1593 continue;
1594 }
1595
1596 /* Did somebody else fill it already? */
1597 if (PageUptodate(page)) {
1598 unlock_page(page);
1599 goto page_ok;
1600 }
1601
1602readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001603 /*
1604 * A previous I/O error may have been due to temporary
1605 * failures, eg. multipath errors.
1606 * PG_error will be set again if readpage fails.
1607 */
1608 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 /* Start the actual read. The read will unlock the page. */
1610 error = mapping->a_ops->readpage(filp, page);
1611
Zach Brown994fc28c2005-12-15 14:28:17 -08001612 if (unlikely(error)) {
1613 if (error == AOP_TRUNCATED_PAGE) {
1614 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001615 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001616 goto find_page;
1617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
1621 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001622 error = lock_page_killable(page);
1623 if (unlikely(error))
1624 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 if (!PageUptodate(page)) {
1626 if (page->mapping == NULL) {
1627 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001628 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 */
1630 unlock_page(page);
1631 page_cache_release(page);
1632 goto find_page;
1633 }
1634 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001635 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001636 error = -EIO;
1637 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639 unlock_page(page);
1640 }
1641
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 goto page_ok;
1643
1644readpage_error:
1645 /* UHHUH! A synchronous read error occurred. Report it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 page_cache_release(page);
1647 goto out;
1648
1649no_cached_page:
1650 /*
1651 * Ok, it wasn't cached, so we need to create a new
1652 * page..
1653 */
Nick Piggineb2be182007-10-16 01:24:57 -07001654 page = page_cache_alloc_cold(mapping);
1655 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001656 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001657 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 }
Nick Piggineb2be182007-10-16 01:24:57 -07001659 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 index, GFP_KERNEL);
1661 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001662 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001663 if (error == -EEXIST) {
1664 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 goto out;
1668 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 goto readpage;
1670 }
1671
1672out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001673 ra->prev_pos = prev_index;
1674 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1675 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001677 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001678 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001679 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680}
1681
Randy Dunlap485bb992006-06-23 02:03:49 -07001682/**
Al Viro6abd2322014-04-04 14:20:57 -04001683 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001684 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001685 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001686 *
Al Viro6abd2322014-04-04 14:20:57 -04001687 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 * that can use the page cache directly.
1689 */
1690ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001691generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
Al Viroed978a82014-03-05 22:53:04 -05001693 struct file *file = iocb->ki_filp;
1694 ssize_t retval = 0;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001695 loff_t *ppos = &iocb->ki_pos;
Al Viroed978a82014-03-05 22:53:04 -05001696 loff_t pos = *ppos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Matthew Wilcoxd475c632015-02-16 15:58:56 -08001698 if (io_is_direct(file)) {
Al Viroed978a82014-03-05 22:53:04 -05001699 struct address_space *mapping = file->f_mapping;
1700 struct inode *inode = mapping->host;
1701 size_t count = iov_iter_count(iter);
Badari Pulavarty543ade12006-09-30 23:28:48 -07001702 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 if (!count)
1705 goto out; /* skip atime */
1706 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001707 retval = filemap_write_and_wait_range(mapping, pos,
Al Viroed978a82014-03-05 22:53:04 -05001708 pos + count - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001709 if (!retval) {
Al Viroed978a82014-03-05 22:53:04 -05001710 struct iov_iter data = *iter;
1711 retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001712 }
Al Viroed978a82014-03-05 22:53:04 -05001713
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001714 if (retval > 0) {
1715 *ppos = pos + retval;
Al Viroed978a82014-03-05 22:53:04 -05001716 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001717 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001718
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001719 /*
1720 * Btrfs can have a short DIO read if we encounter
1721 * compressed extents, so if there was an error, or if
1722 * we've already read everything we wanted to, or if
1723 * there was a short read because we hit EOF, go ahead
1724 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001725 * the rest of the read. Buffered reads will not work for
1726 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001727 */
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001728 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1729 IS_DAX(inode)) {
Al Viroed978a82014-03-05 22:53:04 -05001730 file_accessed(file);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001731 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 }
1734
Al Viroed978a82014-03-05 22:53:04 -05001735 retval = do_generic_file_read(file, ppos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736out:
1737 return retval;
1738}
Al Viroed978a82014-03-05 22:53:04 -05001739EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001742/**
1743 * page_cache_read - adds requested page to the page cache if not already there
1744 * @file: file to read
1745 * @offset: page index
1746 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 * This adds the requested page to the page cache if it isn't already there,
1748 * and schedules an I/O to read in its contents from disk.
1749 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001750static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751{
1752 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001753 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001754 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
Zach Brown994fc28c2005-12-15 14:28:17 -08001756 do {
1757 page = page_cache_alloc_cold(mapping);
1758 if (!page)
1759 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Zach Brown994fc28c2005-12-15 14:28:17 -08001761 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1762 if (ret == 0)
1763 ret = mapping->a_ops->readpage(file, page);
1764 else if (ret == -EEXIST)
1765 ret = 0; /* losing race to add is OK */
1766
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Zach Brown994fc28c2005-12-15 14:28:17 -08001769 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07001770
Zach Brown994fc28c2005-12-15 14:28:17 -08001771 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772}
1773
1774#define MMAP_LOTSAMISS (100)
1775
Linus Torvaldsef00e082009-06-16 15:31:25 -07001776/*
1777 * Synchronous readahead happens when we don't even find
1778 * a page in the page cache at all.
1779 */
1780static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1781 struct file_ra_state *ra,
1782 struct file *file,
1783 pgoff_t offset)
1784{
1785 unsigned long ra_pages;
1786 struct address_space *mapping = file->f_mapping;
1787
1788 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001789 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001790 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001791 if (!ra->ra_pages)
1792 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001793
Joe Perches64363aa2013-07-08 16:00:18 -07001794 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001795 page_cache_sync_readahead(mapping, ra, file, offset,
1796 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001797 return;
1798 }
1799
Andi Kleen207d04b2011-05-24 17:12:29 -07001800 /* Avoid banging the cache line if not needed */
1801 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001802 ra->mmap_miss++;
1803
1804 /*
1805 * Do we miss much more than hit in this file? If so,
1806 * stop bothering with read-ahead. It will only hurt.
1807 */
1808 if (ra->mmap_miss > MMAP_LOTSAMISS)
1809 return;
1810
Wu Fengguangd30a1102009-06-16 15:31:30 -07001811 /*
1812 * mmap read-around
1813 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001814 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001815 ra->start = max_t(long, 0, offset - ra_pages / 2);
1816 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001817 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001818 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001819}
1820
1821/*
1822 * Asynchronous readahead happens when we find the page and PG_readahead,
1823 * so we want to possibly extend the readahead further..
1824 */
1825static void do_async_mmap_readahead(struct vm_area_struct *vma,
1826 struct file_ra_state *ra,
1827 struct file *file,
1828 struct page *page,
1829 pgoff_t offset)
1830{
1831 struct address_space *mapping = file->f_mapping;
1832
1833 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001834 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001835 return;
1836 if (ra->mmap_miss > 0)
1837 ra->mmap_miss--;
1838 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001839 page_cache_async_readahead(mapping, ra, file,
1840 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001841}
1842
Randy Dunlap485bb992006-06-23 02:03:49 -07001843/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001844 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001845 * @vma: vma in which the fault was taken
1846 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001847 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001848 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 * mapped memory region to read in file data during a page fault.
1850 *
1851 * The goto's are kind of ugly, but this streamlines the normal case of having
1852 * it in the page cache, and handles the special cases reasonably without
1853 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001854 *
1855 * vma->vm_mm->mmap_sem must be held on entry.
1856 *
1857 * If our return value has VM_FAULT_RETRY set, it's because
1858 * lock_page_or_retry() returned 0.
1859 * The mmap_sem has usually been released in this case.
1860 * See __lock_page_or_retry() for the exception.
1861 *
1862 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1863 * has not been released.
1864 *
1865 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001867int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
1869 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001870 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 struct address_space *mapping = file->f_mapping;
1872 struct file_ra_state *ra = &file->f_ra;
1873 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001874 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001876 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001877 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001879 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1880 if (offset >= size >> PAGE_CACHE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001881 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001884 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001886 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001887 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001889 * We found the page, so try async readahead before
1890 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001892 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001893 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001894 /* No page in the page cache at all */
1895 do_sync_mmap_readahead(vma, ra, file, offset);
1896 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001897 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001898 ret = VM_FAULT_MAJOR;
1899retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001900 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 if (!page)
1902 goto no_cached_page;
1903 }
1904
Michel Lespinassed88c0922010-11-02 13:05:18 -07001905 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1906 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001907 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001908 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001909
1910 /* Did it get truncated? */
1911 if (unlikely(page->mapping != mapping)) {
1912 unlock_page(page);
1913 put_page(page);
1914 goto retry_find;
1915 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001916 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07001917
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001919 * We have a locked page in the page cache, now we need to check
1920 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 */
Nick Piggind00806b2007-07-19 01:46:57 -07001922 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 goto page_not_uptodate;
1924
Linus Torvaldsef00e082009-06-16 15:31:25 -07001925 /*
1926 * Found the page and have a reference on it.
1927 * We must recheck i_size under page lock.
1928 */
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001929 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1930 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001931 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001932 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001933 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001934 }
1935
Nick Piggind0217ac2007-07-19 01:47:03 -07001936 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001937 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939no_cached_page:
1940 /*
1941 * We're only likely to ever get here if MADV_RANDOM is in
1942 * effect.
1943 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001944 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946 /*
1947 * The page we want has now been added to the page cache.
1948 * In the unlikely event that someone removed it in the
1949 * meantime, we'll just come back here and read it again.
1950 */
1951 if (error >= 0)
1952 goto retry_find;
1953
1954 /*
1955 * An error return from page_cache_read can result if the
1956 * system is low on memory, or a problem occurs while trying
1957 * to schedule I/O.
1958 */
1959 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001960 return VM_FAULT_OOM;
1961 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 /*
1965 * Umm, take care of errors if the page isn't up-to-date.
1966 * Try to re-read it _once_. We do this synchronously,
1967 * because there really aren't any performance issues here
1968 * and we need to check for errors.
1969 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001971 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001972 if (!error) {
1973 wait_on_page_locked(page);
1974 if (!PageUptodate(page))
1975 error = -EIO;
1976 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001978
1979 if (!error || error == AOP_TRUNCATED_PAGE)
1980 goto retry_find;
1981
1982 /* Things didn't work out. Return zero to tell the mm layer so. */
1983 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001984 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001985}
1986EXPORT_SYMBOL(filemap_fault);
1987
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001988void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1989{
1990 struct radix_tree_iter iter;
1991 void **slot;
1992 struct file *file = vma->vm_file;
1993 struct address_space *mapping = file->f_mapping;
1994 loff_t size;
1995 struct page *page;
1996 unsigned long address = (unsigned long) vmf->virtual_address;
1997 unsigned long addr;
1998 pte_t *pte;
1999
2000 rcu_read_lock();
2001 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2002 if (iter.index > vmf->max_pgoff)
2003 break;
2004repeat:
2005 page = radix_tree_deref_slot(slot);
2006 if (unlikely(!page))
2007 goto next;
2008 if (radix_tree_exception(page)) {
2009 if (radix_tree_deref_retry(page))
2010 break;
2011 else
2012 goto next;
2013 }
2014
2015 if (!page_cache_get_speculative(page))
2016 goto repeat;
2017
2018 /* Has the page moved? */
2019 if (unlikely(page != *slot)) {
2020 page_cache_release(page);
2021 goto repeat;
2022 }
2023
2024 if (!PageUptodate(page) ||
2025 PageReadahead(page) ||
2026 PageHWPoison(page))
2027 goto skip;
2028 if (!trylock_page(page))
2029 goto skip;
2030
2031 if (page->mapping != mapping || !PageUptodate(page))
2032 goto unlock;
2033
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002034 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2035 if (page->index >= size >> PAGE_CACHE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002036 goto unlock;
2037
2038 pte = vmf->pte + page->index - vmf->pgoff;
2039 if (!pte_none(*pte))
2040 goto unlock;
2041
2042 if (file->f_ra.mmap_miss > 0)
2043 file->f_ra.mmap_miss--;
2044 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2045 do_set_pte(vma, addr, page, pte, false, false);
2046 unlock_page(page);
2047 goto next;
2048unlock:
2049 unlock_page(page);
2050skip:
2051 page_cache_release(page);
2052next:
2053 if (iter.index == vmf->max_pgoff)
2054 break;
2055 }
2056 rcu_read_unlock();
2057}
2058EXPORT_SYMBOL(filemap_map_pages);
2059
Jan Kara4fcf1c62012-06-12 16:20:29 +02002060int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2061{
2062 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002063 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002064 int ret = VM_FAULT_LOCKED;
2065
Jan Kara14da9202012-06-12 16:20:37 +02002066 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002067 file_update_time(vma->vm_file);
2068 lock_page(page);
2069 if (page->mapping != inode->i_mapping) {
2070 unlock_page(page);
2071 ret = VM_FAULT_NOPAGE;
2072 goto out;
2073 }
Jan Kara14da9202012-06-12 16:20:37 +02002074 /*
2075 * We mark the page dirty already here so that when freeze is in
2076 * progress, we are guaranteed that writeback during freezing will
2077 * see the dirty page and writeprotect it again.
2078 */
2079 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002080 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002081out:
Jan Kara14da9202012-06-12 16:20:37 +02002082 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002083 return ret;
2084}
2085EXPORT_SYMBOL(filemap_page_mkwrite);
2086
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002087const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002088 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002089 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002090 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091};
2092
2093/* This is used for a general mmap of a disk file */
2094
2095int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2096{
2097 struct address_space *mapping = file->f_mapping;
2098
2099 if (!mapping->a_ops->readpage)
2100 return -ENOEXEC;
2101 file_accessed(file);
2102 vma->vm_ops = &generic_file_vm_ops;
2103 return 0;
2104}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106/*
2107 * This is for filesystems which do not implement ->writepage.
2108 */
2109int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2110{
2111 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2112 return -EINVAL;
2113 return generic_file_mmap(file, vma);
2114}
2115#else
2116int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2117{
2118 return -ENOSYS;
2119}
2120int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2121{
2122 return -ENOSYS;
2123}
2124#endif /* CONFIG_MMU */
2125
2126EXPORT_SYMBOL(generic_file_mmap);
2127EXPORT_SYMBOL(generic_file_readonly_mmap);
2128
Sasha Levin67f9fd92014-04-03 14:48:18 -07002129static struct page *wait_on_page_read(struct page *page)
2130{
2131 if (!IS_ERR(page)) {
2132 wait_on_page_locked(page);
2133 if (!PageUptodate(page)) {
2134 page_cache_release(page);
2135 page = ERR_PTR(-EIO);
2136 }
2137 }
2138 return page;
2139}
2140
Nick Piggin6fe69002007-05-06 14:49:04 -07002141static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002142 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002143 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002144 void *data,
2145 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
Nick Piggineb2be182007-10-16 01:24:57 -07002147 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 int err;
2149repeat:
2150 page = find_get_page(mapping, index);
2151 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002152 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002153 if (!page)
2154 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002155 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002156 if (unlikely(err)) {
2157 page_cache_release(page);
2158 if (err == -EEXIST)
2159 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 return ERR_PTR(err);
2162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 err = filler(data, page);
2164 if (err < 0) {
2165 page_cache_release(page);
2166 page = ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002167 } else {
2168 page = wait_on_page_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 }
2170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 return page;
2172}
2173
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002174static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002175 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002176 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002177 void *data,
2178 gfp_t gfp)
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180{
2181 struct page *page;
2182 int err;
2183
2184retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002185 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01002187 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 if (PageUptodate(page))
2189 goto out;
2190
2191 lock_page(page);
2192 if (!page->mapping) {
2193 unlock_page(page);
2194 page_cache_release(page);
2195 goto retry;
2196 }
2197 if (PageUptodate(page)) {
2198 unlock_page(page);
2199 goto out;
2200 }
2201 err = filler(data, page);
2202 if (err < 0) {
2203 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01002204 return ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002205 } else {
2206 page = wait_on_page_read(page);
2207 if (IS_ERR(page))
2208 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 }
David Howellsc855ff32007-05-09 13:42:20 +01002210out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002211 mark_page_accessed(page);
2212 return page;
2213}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002214
2215/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002216 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002217 * @mapping: the page's address_space
2218 * @index: the page index
2219 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002220 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002221 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002222 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002223 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002224 *
2225 * If the page does not get brought uptodate, return -EIO.
2226 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002227struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002228 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002229 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002230 void *data)
2231{
2232 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2233}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002234EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002235
2236/**
2237 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2238 * @mapping: the page's address_space
2239 * @index: the page index
2240 * @gfp: the page allocator flags to use if allocating
2241 *
2242 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002243 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002244 *
2245 * If the page does not get brought uptodate, return -EIO.
2246 */
2247struct page *read_cache_page_gfp(struct address_space *mapping,
2248 pgoff_t index,
2249 gfp_t gfp)
2250{
2251 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2252
Sasha Levin67f9fd92014-04-03 14:48:18 -07002253 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002254}
2255EXPORT_SYMBOL(read_cache_page_gfp);
2256
Nick Piggin2f718ff2007-10-16 01:24:59 -07002257/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 * Performs necessary checks before doing a write
2259 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002260 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 * Returns appropriate error code that caller should return or
2262 * zero in case that write should be allowed.
2263 */
2264inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2265{
2266 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002267 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
2269 if (unlikely(*pos < 0))
2270 return -EINVAL;
2271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 if (!isblk) {
2273 /* FIXME: this is for backwards compatibility with 2.4 */
2274 if (file->f_flags & O_APPEND)
2275 *pos = i_size_read(inode);
2276
2277 if (limit != RLIM_INFINITY) {
2278 if (*pos >= limit) {
2279 send_sig(SIGXFSZ, current, 0);
2280 return -EFBIG;
2281 }
2282 if (*count > limit - (typeof(limit))*pos) {
2283 *count = limit - (typeof(limit))*pos;
2284 }
2285 }
2286 }
2287
2288 /*
2289 * LFS rule
2290 */
2291 if (unlikely(*pos + *count > MAX_NON_LFS &&
2292 !(file->f_flags & O_LARGEFILE))) {
2293 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 return -EFBIG;
2295 }
2296 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2297 *count = MAX_NON_LFS - (unsigned long)*pos;
2298 }
2299 }
2300
2301 /*
2302 * Are we about to exceed the fs block limit ?
2303 *
2304 * If we have written data it becomes a short write. If we have
2305 * exceeded without writing data we send a signal and return EFBIG.
2306 * Linus frestrict idea will clean these up nicely..
2307 */
2308 if (likely(!isblk)) {
2309 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2310 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 return -EFBIG;
2312 }
2313 /* zero-length writes at ->s_maxbytes are OK */
2314 }
2315
2316 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2317 *count = inode->i_sb->s_maxbytes - *pos;
2318 } else {
David Howells93614012006-09-30 20:45:40 +02002319#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 loff_t isize;
2321 if (bdev_read_only(I_BDEV(inode)))
2322 return -EPERM;
2323 isize = i_size_read(inode);
2324 if (*pos >= isize) {
2325 if (*count || *pos > isize)
2326 return -ENOSPC;
2327 }
2328
2329 if (*pos + *count > isize)
2330 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002331#else
2332 return -EPERM;
2333#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 }
2335 return 0;
2336}
2337EXPORT_SYMBOL(generic_write_checks);
2338
Nick Pigginafddba42007-10-16 01:25:01 -07002339int pagecache_write_begin(struct file *file, struct address_space *mapping,
2340 loff_t pos, unsigned len, unsigned flags,
2341 struct page **pagep, void **fsdata)
2342{
2343 const struct address_space_operations *aops = mapping->a_ops;
2344
Nick Piggin4e02ed42008-10-29 14:00:55 -07002345 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002346 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002347}
2348EXPORT_SYMBOL(pagecache_write_begin);
2349
2350int pagecache_write_end(struct file *file, struct address_space *mapping,
2351 loff_t pos, unsigned len, unsigned copied,
2352 struct page *page, void *fsdata)
2353{
2354 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002355
Nick Piggin4e02ed42008-10-29 14:00:55 -07002356 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002357}
2358EXPORT_SYMBOL(pagecache_write_end);
2359
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360ssize_t
Al Viro0c949332014-03-22 06:51:37 -04002361generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362{
2363 struct file *file = iocb->ki_filp;
2364 struct address_space *mapping = file->f_mapping;
2365 struct inode *inode = mapping->host;
2366 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002367 size_t write_len;
2368 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002369 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370
Al Viro0c949332014-03-22 06:51:37 -04002371 write_len = iov_iter_count(from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002372 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002373
Nick Piggin48b47c52009-01-06 14:40:22 -08002374 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002375 if (written)
2376 goto out;
2377
2378 /*
2379 * After a write we want buffered reads to be sure to go to disk to get
2380 * the new data. We invalidate clean cached page from the region we're
2381 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002382 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002383 */
2384 if (mapping->nrpages) {
2385 written = invalidate_inode_pages2_range(mapping,
2386 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002387 /*
2388 * If a page can not be invalidated, return 0 to fall back
2389 * to buffered write.
2390 */
2391 if (written) {
2392 if (written == -EBUSY)
2393 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002394 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002395 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002396 }
2397
Al Viro26978b82014-03-10 14:08:45 -04002398 data = *from;
2399 written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002400
2401 /*
2402 * Finally, try again to invalidate clean pages which might have been
2403 * cached by non-direct readahead, or faulted in by get_user_pages()
2404 * if the source of the write was an mmap'ed region of the file
2405 * we're writing. Either one is a pretty crazy thing to do,
2406 * so we don't support it 100%. If this invalidation
2407 * fails, tough, the write still worked...
2408 */
2409 if (mapping->nrpages) {
2410 invalidate_inode_pages2_range(mapping,
2411 pos >> PAGE_CACHE_SHIFT, end);
2412 }
2413
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002415 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002416 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002417 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2418 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 mark_inode_dirty(inode);
2420 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002421 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002423out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 return written;
2425}
2426EXPORT_SYMBOL(generic_file_direct_write);
2427
Nick Piggineb2be182007-10-16 01:24:57 -07002428/*
2429 * Find or create a page at the given pagecache position. Return the locked
2430 * page. This function is specifically for buffered writes.
2431 */
Nick Piggin54566b22009-01-04 12:00:53 -08002432struct page *grab_cache_page_write_begin(struct address_space *mapping,
2433 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002434{
Nick Piggineb2be182007-10-16 01:24:57 -07002435 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07002436 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002437
Nick Piggin54566b22009-01-04 12:00:53 -08002438 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002439 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002440
Mel Gorman2457aec2014-06-04 16:10:31 -07002441 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002442 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002443 if (page)
2444 wait_for_stable_page(page);
2445
Nick Piggineb2be182007-10-16 01:24:57 -07002446 return page;
2447}
Nick Piggin54566b22009-01-04 12:00:53 -08002448EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002449
Al Viro3b93f912014-02-11 21:34:08 -05002450ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002451 struct iov_iter *i, loff_t pos)
2452{
2453 struct address_space *mapping = file->f_mapping;
2454 const struct address_space_operations *a_ops = mapping->a_ops;
2455 long status = 0;
2456 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002457 unsigned int flags = 0;
2458
2459 /*
2460 * Copies from kernel address space cannot fail (NFSD is a big user).
2461 */
Al Viro777eda22014-12-17 04:46:46 -05002462 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002463 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002464
2465 do {
2466 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002467 unsigned long offset; /* Offset into pagecache page */
2468 unsigned long bytes; /* Bytes to write to page */
2469 size_t copied; /* Bytes copied from user */
2470 void *fsdata;
2471
2472 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002473 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2474 iov_iter_count(i));
2475
2476again:
Nick Pigginafddba42007-10-16 01:25:01 -07002477 /*
2478 * Bring in the user page that we will copy from _first_.
2479 * Otherwise there's a nasty deadlock on copying from the
2480 * same page as we're writing to, without it being marked
2481 * up-to-date.
2482 *
2483 * Not only is this an optimisation, but it is also required
2484 * to check that the address is actually valid, when atomic
2485 * usercopies are used, below.
2486 */
2487 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2488 status = -EFAULT;
2489 break;
2490 }
2491
Nick Piggin674b8922007-10-16 01:25:03 -07002492 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002493 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002494 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002495 break;
2496
anfei zhou931e80e2010-02-02 13:44:02 -08002497 if (mapping_writably_mapped(mapping))
2498 flush_dcache_page(page);
2499
Nick Pigginafddba42007-10-16 01:25:01 -07002500 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002501 flush_dcache_page(page);
2502
2503 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2504 page, fsdata);
2505 if (unlikely(status < 0))
2506 break;
2507 copied = status;
2508
2509 cond_resched();
2510
Nick Piggin124d3b72008-02-02 15:01:17 +01002511 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002512 if (unlikely(copied == 0)) {
2513 /*
2514 * If we were unable to copy any data at all, we must
2515 * fall back to a single segment length write.
2516 *
2517 * If we didn't fallback here, we could livelock
2518 * because not all segments in the iov can be copied at
2519 * once without a pagefault.
2520 */
2521 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2522 iov_iter_single_seg_count(i));
2523 goto again;
2524 }
Nick Pigginafddba42007-10-16 01:25:01 -07002525 pos += copied;
2526 written += copied;
2527
2528 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002529 if (fatal_signal_pending(current)) {
2530 status = -EINTR;
2531 break;
2532 }
Nick Pigginafddba42007-10-16 01:25:01 -07002533 } while (iov_iter_count(i));
2534
2535 return written ? written : status;
2536}
Al Viro3b93f912014-02-11 21:34:08 -05002537EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Jan Karae4dd9de2009-08-17 18:10:06 +02002539/**
Al Viro81742022014-04-03 03:17:43 -04002540 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002541 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002542 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002543 *
2544 * This function does all the work needed for actually writing data to a
2545 * file. It does all basic checks, removes SUID from the file, updates
2546 * modification times and calls proper subroutines depending on whether we
2547 * do direct IO or a standard buffered write.
2548 *
2549 * It expects i_mutex to be grabbed unless we work on a block device or similar
2550 * object which does not need locking at all.
2551 *
2552 * This function does *not* take care of syncing data in case of O_SYNC write.
2553 * A caller has to handle it. This is mainly due to the fact that we want to
2554 * avoid syncing under i_mutex.
2555 */
Al Viro81742022014-04-03 03:17:43 -04002556ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557{
2558 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002559 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 struct inode *inode = mapping->host;
Al Viro41fc56d2014-02-09 12:58:52 -05002561 loff_t pos = iocb->ki_pos;
Al Viro3b93f912014-02-11 21:34:08 -05002562 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002564 ssize_t status;
Al Viro81742022014-04-03 03:17:43 -04002565 size_t count = iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002568 current->backing_dev_info = inode_to_bdi(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2570 if (err)
2571 goto out;
2572
2573 if (count == 0)
2574 goto out;
2575
Al Viro81742022014-04-03 03:17:43 -04002576 iov_iter_truncate(from, count);
Al Viro0c949332014-03-22 06:51:37 -04002577
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002578 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 if (err)
2580 goto out;
2581
Josef Bacikc3b2da32012-03-26 09:59:21 -04002582 err = file_update_time(file);
2583 if (err)
2584 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
Matthew Wilcoxd475c632015-02-16 15:58:56 -08002586 if (io_is_direct(file)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002587 loff_t endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002588
Al Viro81742022014-04-03 03:17:43 -04002589 written = generic_file_direct_write(iocb, from, pos);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002590 /*
2591 * If the write stopped short of completing, fall back to
2592 * buffered writes. Some filesystems do this for writes to
2593 * holes, for example. For DAX files, a buffered write will
2594 * not succeed (even if it did, DAX does not handle dirty
2595 * page-cache pages correctly).
2596 */
2597 if (written < 0 || written == count || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002599
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 pos += written;
2601 count -= written;
Al Viro3b93f912014-02-11 21:34:08 -05002602
Al Viro81742022014-04-03 03:17:43 -04002603 status = generic_perform_write(file, from, pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002604 /*
Al Viro3b93f912014-02-11 21:34:08 -05002605 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002606 * then we want to return the number of bytes which were
2607 * direct-written, or the error code if that was zero. Note
2608 * that this differs from normal direct-io semantics, which
2609 * will return -EFOO even if some bytes were written.
2610 */
Al Viro60bb4522014-08-08 12:39:16 -04002611 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002612 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002613 goto out;
2614 }
Al Viro3b93f912014-02-11 21:34:08 -05002615 iocb->ki_pos = pos + status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002616 /*
2617 * We need to ensure that the page cache pages are written to
2618 * disk and invalidated to preserve the expected O_DIRECT
2619 * semantics.
2620 */
Al Viro3b93f912014-02-11 21:34:08 -05002621 endbyte = pos + status - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002622 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002623 if (err == 0) {
Al Viro3b93f912014-02-11 21:34:08 -05002624 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002625 invalidate_mapping_pages(mapping,
2626 pos >> PAGE_CACHE_SHIFT,
2627 endbyte >> PAGE_CACHE_SHIFT);
2628 } else {
2629 /*
2630 * We don't know how much we wrote, so just return
2631 * the number of bytes which were direct-written
2632 */
2633 }
2634 } else {
Al Viro81742022014-04-03 03:17:43 -04002635 written = generic_perform_write(file, from, pos);
Al Viro3b93f912014-02-11 21:34:08 -05002636 if (likely(written >= 0))
2637 iocb->ki_pos = pos + written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639out:
2640 current->backing_dev_info = NULL;
2641 return written ? written : err;
2642}
Al Viro81742022014-04-03 03:17:43 -04002643EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
Jan Karae4dd9de2009-08-17 18:10:06 +02002645/**
Al Viro81742022014-04-03 03:17:43 -04002646 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002647 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002648 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002649 *
Al Viro81742022014-04-03 03:17:43 -04002650 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002651 * filesystems. It takes care of syncing the file in case of O_SYNC file
2652 * and acquires i_mutex as needed.
2653 */
Al Viro81742022014-04-03 03:17:43 -04002654ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655{
2656 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002657 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002660 mutex_lock(&inode->i_mutex);
Al Viro81742022014-04-03 03:17:43 -04002661 ret = __generic_file_write_iter(iocb, from);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002662 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
Christoph Hellwig02afc272013-09-04 15:04:40 +02002664 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 ssize_t err;
2666
Al Virod311d792014-02-09 15:18:09 -05002667 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2668 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 ret = err;
2670 }
2671 return ret;
2672}
Al Viro81742022014-04-03 03:17:43 -04002673EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674
David Howellscf9a2ae2006-08-29 19:05:54 +01002675/**
2676 * try_to_release_page() - release old fs-specific metadata on a page
2677 *
2678 * @page: the page which the kernel is trying to free
2679 * @gfp_mask: memory allocation flags (and I/O mode)
2680 *
2681 * The address_space is to try to release any data against the page
2682 * (presumably at page->private). If the release was successful, return `1'.
2683 * Otherwise return zero.
2684 *
David Howells266cf652009-04-03 16:42:36 +01002685 * This may also be called if PG_fscache is set on a page, indicating that the
2686 * page is known to the local caching routines.
2687 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002688 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002689 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002690 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002691 */
2692int try_to_release_page(struct page *page, gfp_t gfp_mask)
2693{
2694 struct address_space * const mapping = page->mapping;
2695
2696 BUG_ON(!PageLocked(page));
2697 if (PageWriteback(page))
2698 return 0;
2699
2700 if (mapping && mapping->a_ops->releasepage)
2701 return mapping->a_ops->releasepage(page, gfp_mask);
2702 return try_to_free_buffers(page);
2703}
2704
2705EXPORT_SYMBOL(try_to_release_page);