blob: bc943867d68c68dab4109fe715c8d37d6c72757a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080014#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Johannes Weiner00501b52014-08-08 14:19:20 -070034#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060036#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070037#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080038#include "internal.h"
39
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070040#define CREATE_TRACE_POINTS
41#include <trace/events/filemap.h>
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * FIXME: remove all knowledge of the buffer layer from the core VM
45 */
Jan Kara148f9482009-08-17 19:52:36 +020046#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <asm/mman.h>
49
50/*
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * though.
53 *
54 * Shared mappings now work. 15.8.1995 Bruno.
55 *
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 *
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 */
61
62/*
63 * Lock ordering:
64 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080065 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070067 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080070 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080071 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 *
73 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080074 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070075 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
Al Viroccad2362014-02-11 22:36:48 -050081 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070082 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060084 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110085 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * ->mapping->tree_lock (__sync_single_inode)
87 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080088 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
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)
Greg Thelenc4843a72015-05-22 17:13:16 -0400104 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800109 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700110 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
112
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700113static void page_cache_tree_delete(struct address_space *mapping,
114 struct page *page, void *shadow)
115{
Johannes Weiner449dd692014-04-03 14:47:56 -0700116 struct radix_tree_node *node;
117 unsigned long index;
118 unsigned int offset;
119 unsigned int tag;
120 void **slot;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700121
Johannes Weiner449dd692014-04-03 14:47:56 -0700122 VM_BUG_ON(!PageLocked(page));
123
124 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
125
126 if (shadow) {
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800127 mapping->nrexceptional++;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700128 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800129 * Make sure the nrexceptional update is committed before
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700130 * the nrpages update so that final truncate racing
131 * with reclaim does not see both counters 0 at the
132 * same time and miss a shadow entry.
133 */
134 smp_wmb();
Johannes Weiner449dd692014-04-03 14:47:56 -0700135 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700136 mapping->nrpages--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700137
138 if (!node) {
139 /* Clear direct pointer tags in root node */
140 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
141 radix_tree_replace_slot(slot, shadow);
142 return;
143 }
144
145 /* Clear tree tags for the removed page */
146 index = page->index;
147 offset = index & RADIX_TREE_MAP_MASK;
148 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
149 if (test_bit(offset, node->tags[tag]))
150 radix_tree_tag_clear(&mapping->page_tree, index, tag);
151 }
152
153 /* Delete page, swap shadow entry */
154 radix_tree_replace_slot(slot, shadow);
155 workingset_node_pages_dec(node);
156 if (shadow)
157 workingset_node_shadows_inc(node);
158 else
159 if (__radix_tree_delete_node(&mapping->page_tree, node))
160 return;
161
162 /*
163 * Track node that only contains shadow entries.
164 *
165 * Avoid acquiring the list_lru lock if already tracked. The
166 * list_empty() test is safe as node->private_list is
167 * protected by mapping->tree_lock.
168 */
169 if (!workingset_node_pages(node) &&
170 list_empty(&node->private_list)) {
171 node->private_data = mapping;
172 list_lru_add(&workingset_shadow_nodes, &node->private_list);
173 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700174}
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176/*
Minchan Kime64a7822011-03-22 16:32:44 -0700177 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 * sure the page is locked and that nobody else uses it - or that usage
Greg Thelenc4843a72015-05-22 17:13:16 -0400179 * is safe. The caller must hold the mapping's tree_lock and
180 * mem_cgroup_begin_page_stat().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 */
Greg Thelenc4843a72015-05-22 17:13:16 -0400182void __delete_from_page_cache(struct page *page, void *shadow,
183 struct mem_cgroup *memcg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
185 struct address_space *mapping = page->mapping;
186
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700187 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600188 /*
189 * if we're uptodate, flush out into the cleancache, otherwise
190 * invalidate any existing cleancache entries. We can't leave
191 * stale data around in the cleancache once our page is gone
192 */
193 if (PageUptodate(page) && PageMappedToDisk(page))
194 cleancache_put_page(page);
195 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400196 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600197
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700198 page_cache_tree_delete(mapping, page, shadow);
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700201 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700202
Michal Hocko4165b9b2015-06-24 16:57:24 -0700203 /* hugetlb pages do not participate in page cache accounting. */
204 if (!PageHuge(page))
205 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700206 if (PageSwapBacked(page))
207 __dec_zone_page_state(page, NR_SHMEM);
Kirill A. Shutemove1534ae2016-01-15 16:53:46 -0800208 VM_BUG_ON_PAGE(page_mapped(page), page);
Linus Torvalds3a692792007-12-19 14:05:13 -0800209
210 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700211 * At this point page must be either written or cleaned by truncate.
212 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800213 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700214 * This fixes dirty accounting after removing the page entirely but
215 * leaves PageDirty set: it has no effect for truncated page and
216 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800217 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700218 if (WARN_ON_ONCE(PageDirty(page)))
Tejun Heo682aa8e2015-05-28 14:50:53 -0400219 account_page_cleaned(page, mapping, memcg,
220 inode_to_wb(mapping->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221}
222
Minchan Kim702cfbf2011-03-22 16:32:43 -0700223/**
224 * delete_from_page_cache - delete page from page cache
225 * @page: the page which the kernel is trying to remove from page cache
226 *
227 * This must be called only on pages that have been verified to be in the page
228 * cache and locked. It will never put the page into the free list, the caller
229 * has a reference on the page.
230 */
231void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
233 struct address_space *mapping = page->mapping;
Greg Thelenc4843a72015-05-22 17:13:16 -0400234 struct mem_cgroup *memcg;
235 unsigned long flags;
236
Linus Torvalds6072d132010-12-01 13:35:19 -0500237 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Matt Mackallcd7619d2005-05-01 08:59:01 -0700239 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Linus Torvalds6072d132010-12-01 13:35:19 -0500241 freepage = mapping->a_ops->freepage;
Greg Thelenc4843a72015-05-22 17:13:16 -0400242
243 memcg = mem_cgroup_begin_page_stat(page);
244 spin_lock_irqsave(&mapping->tree_lock, flags);
245 __delete_from_page_cache(page, NULL, memcg);
246 spin_unlock_irqrestore(&mapping->tree_lock, flags);
247 mem_cgroup_end_page_stat(memcg);
Linus Torvalds6072d132010-12-01 13:35:19 -0500248
249 if (freepage)
250 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700251 page_cache_release(page);
252}
253EXPORT_SYMBOL(delete_from_page_cache);
254
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700255static int filemap_check_errors(struct address_space *mapping)
256{
257 int ret = 0;
258 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700259 if (test_bit(AS_ENOSPC, &mapping->flags) &&
260 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700261 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700262 if (test_bit(AS_EIO, &mapping->flags) &&
263 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700264 ret = -EIO;
265 return ret;
266}
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700269 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700270 * @mapping: address space structure to write
271 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800272 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700273 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700275 * Start writeback against all of a mapping's dirty pages that lie
276 * within the byte offsets <start, end> inclusive.
277 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700279 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 * these two operations is that if a dirty page/buffer is encountered, it must
281 * be waited upon, and not just skipped over.
282 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800283int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
284 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
286 int ret;
287 struct writeback_control wbc = {
288 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800289 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700290 .range_start = start,
291 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 };
293
294 if (!mapping_cap_writeback_dirty(mapping))
295 return 0;
296
Tejun Heob16b1de2015-06-02 08:39:48 -0600297 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600299 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 return ret;
301}
302
303static inline int __filemap_fdatawrite(struct address_space *mapping,
304 int sync_mode)
305{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700306 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
308
309int filemap_fdatawrite(struct address_space *mapping)
310{
311 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
312}
313EXPORT_SYMBOL(filemap_fdatawrite);
314
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400315int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800316 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
318 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
319}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400320EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Randy Dunlap485bb992006-06-23 02:03:49 -0700322/**
323 * filemap_flush - mostly a non-blocking flush
324 * @mapping: target address_space
325 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 * This is a mostly non-blocking flush. Not suitable for data-integrity
327 * purposes - I/O may not be started against all dirty pages.
328 */
329int filemap_flush(struct address_space *mapping)
330{
331 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
332}
333EXPORT_SYMBOL(filemap_flush);
334
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800335static int __filemap_fdatawait_range(struct address_space *mapping,
336 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200338 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
339 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 struct pagevec pvec;
341 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800342 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200344 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700345 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
347 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 while ((index <= end) &&
349 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
350 PAGECACHE_TAG_WRITEBACK,
351 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
352 unsigned i;
353
354 for (i = 0; i < nr_pages; i++) {
355 struct page *page = pvec.pages[i];
356
357 /* until radix tree lookup accepts end_index */
358 if (page->index > end)
359 continue;
360
361 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800362 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 ret = -EIO;
364 }
365 pagevec_release(&pvec);
366 cond_resched();
367 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700368out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800369 return ret;
370}
371
372/**
373 * filemap_fdatawait_range - wait for writeback to complete
374 * @mapping: address space structure to wait for
375 * @start_byte: offset in bytes where the range starts
376 * @end_byte: offset in bytes where the range ends (inclusive)
377 *
378 * Walk the list of under-writeback pages of the given address space
379 * in the given range and wait for all of them. Check error status of
380 * the address space and return it.
381 *
382 * Since the error status of the address space is cleared by this function,
383 * callers are responsible for checking the return value and handling and/or
384 * reporting the error.
385 */
386int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
387 loff_t end_byte)
388{
389 int ret, ret2;
390
391 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700392 ret2 = filemap_check_errors(mapping);
393 if (!ret)
394 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396 return ret;
397}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200398EXPORT_SYMBOL(filemap_fdatawait_range);
399
400/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800401 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
402 * @mapping: address space structure to wait for
403 *
404 * Walk the list of under-writeback pages of the given address space
405 * and wait for all of them. Unlike filemap_fdatawait(), this function
406 * does not clear error status of the address space.
407 *
408 * Use this function if callers don't handle errors themselves. Expected
409 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
410 * fsfreeze(8)
411 */
412void filemap_fdatawait_keep_errors(struct address_space *mapping)
413{
414 loff_t i_size = i_size_read(mapping->host);
415
416 if (i_size == 0)
417 return;
418
419 __filemap_fdatawait_range(mapping, 0, i_size - 1);
420}
421
422/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700423 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700425 *
426 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800427 * and wait for all of them. Check error status of the address space
428 * and return it.
429 *
430 * Since the error status of the address space is cleared by this function,
431 * callers are responsible for checking the return value and handling and/or
432 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 */
434int filemap_fdatawait(struct address_space *mapping)
435{
436 loff_t i_size = i_size_read(mapping->host);
437
438 if (i_size == 0)
439 return 0;
440
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200441 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443EXPORT_SYMBOL(filemap_fdatawait);
444
445int filemap_write_and_wait(struct address_space *mapping)
446{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800447 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800450 err = filemap_fdatawrite(mapping);
451 /*
452 * Even if the above returned error, the pages may be
453 * written partially (e.g. -ENOSPC), so we wait for it.
454 * But the -EIO is special case, it may indicate the worst
455 * thing (e.g. bug) happened, so we avoid waiting for it.
456 */
457 if (err != -EIO) {
458 int err2 = filemap_fdatawait(mapping);
459 if (!err)
460 err = err2;
461 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700462 } else {
463 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800465 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800467EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Randy Dunlap485bb992006-06-23 02:03:49 -0700469/**
470 * filemap_write_and_wait_range - write out & wait on a file range
471 * @mapping: the address_space for the pages
472 * @lstart: offset in bytes where the range starts
473 * @lend: offset in bytes where the range ends (inclusive)
474 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800475 * Write out and wait upon file offsets lstart->lend, inclusive.
476 *
477 * Note that `lend' is inclusive (describes the last byte to be written) so
478 * that this function can be used to write to the very end-of-file (end = -1).
479 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480int filemap_write_and_wait_range(struct address_space *mapping,
481 loff_t lstart, loff_t lend)
482{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800483 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Ross Zwisler9973c982016-01-22 15:10:47 -0800485 if (dax_mapping(mapping) && mapping->nrexceptional) {
486 err = dax_writeback_mapping_range(mapping, lstart, lend);
487 if (err)
488 return err;
489 }
490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800492 err = __filemap_fdatawrite_range(mapping, lstart, lend,
493 WB_SYNC_ALL);
494 /* See comment of filemap_write_and_wait() */
495 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200496 int err2 = filemap_fdatawait_range(mapping,
497 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800498 if (!err)
499 err = err2;
500 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700501 } else {
502 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800504 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
Chris Masonf6995582009-04-15 13:22:37 -0400506EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Randy Dunlap485bb992006-06-23 02:03:49 -0700508/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700509 * replace_page_cache_page - replace a pagecache page with a new one
510 * @old: page to be replaced
511 * @new: page to replace with
512 * @gfp_mask: allocation mode
513 *
514 * This function replaces a page in the pagecache with a new one. On
515 * success it acquires the pagecache reference for the new page and
516 * drops it for the old page. Both the old and new pages must be
517 * locked. This function does not add the new page to the LRU, the
518 * caller must do that.
519 *
520 * The remove + add is atomic. The only way this function can fail is
521 * memory allocation failure.
522 */
523int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
524{
525 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700526
Sasha Levin309381fea2014-01-23 15:52:54 -0800527 VM_BUG_ON_PAGE(!PageLocked(old), old);
528 VM_BUG_ON_PAGE(!PageLocked(new), new);
529 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700530
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700531 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
532 if (!error) {
533 struct address_space *mapping = old->mapping;
534 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400535 struct mem_cgroup *memcg;
536 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700537
538 pgoff_t offset = old->index;
539 freepage = mapping->a_ops->freepage;
540
541 page_cache_get(new);
542 new->mapping = mapping;
543 new->index = offset;
544
Greg Thelenc4843a72015-05-22 17:13:16 -0400545 memcg = mem_cgroup_begin_page_stat(old);
546 spin_lock_irqsave(&mapping->tree_lock, flags);
547 __delete_from_page_cache(old, NULL, memcg);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700548 error = radix_tree_insert(&mapping->page_tree, offset, new);
549 BUG_ON(error);
550 mapping->nrpages++;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700551
552 /*
553 * hugetlb pages do not participate in page cache accounting.
554 */
555 if (!PageHuge(new))
556 __inc_zone_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700557 if (PageSwapBacked(new))
558 __inc_zone_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400559 spin_unlock_irqrestore(&mapping->tree_lock, flags);
560 mem_cgroup_end_page_stat(memcg);
Hugh Dickins45637ba2015-11-05 18:49:40 -0800561 mem_cgroup_replace_page(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700562 radix_tree_preload_end();
563 if (freepage)
564 freepage(old);
565 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700566 }
567
568 return error;
569}
570EXPORT_SYMBOL_GPL(replace_page_cache_page);
571
Johannes Weiner0cd61442014-04-03 14:47:46 -0700572static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700573 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700574{
Johannes Weiner449dd692014-04-03 14:47:56 -0700575 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700576 void **slot;
577 int error;
578
Johannes Weiner449dd692014-04-03 14:47:56 -0700579 error = __radix_tree_create(&mapping->page_tree, page->index,
580 &node, &slot);
581 if (error)
582 return error;
583 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700584 void *p;
585
586 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
587 if (!radix_tree_exceptional_entry(p))
588 return -EEXIST;
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800589
590 if (WARN_ON(dax_mapping(mapping)))
591 return -EINVAL;
592
Johannes Weinera5289102014-04-03 14:47:51 -0700593 if (shadowp)
594 *shadowp = p;
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800595 mapping->nrexceptional--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700596 if (node)
597 workingset_node_shadows_dec(node);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700598 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700599 radix_tree_replace_slot(slot, page);
600 mapping->nrpages++;
601 if (node) {
602 workingset_node_pages_inc(node);
603 /*
604 * Don't track node that contains actual pages.
605 *
606 * Avoid acquiring the list_lru lock if already
607 * untracked. The list_empty() test is safe as
608 * node->private_list is protected by
609 * mapping->tree_lock.
610 */
611 if (!list_empty(&node->private_list))
612 list_lru_del(&workingset_shadow_nodes,
613 &node->private_list);
614 }
615 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700616}
617
Johannes Weinera5289102014-04-03 14:47:51 -0700618static int __add_to_page_cache_locked(struct page *page,
619 struct address_space *mapping,
620 pgoff_t offset, gfp_t gfp_mask,
621 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Johannes Weiner00501b52014-08-08 14:19:20 -0700623 int huge = PageHuge(page);
624 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700625 int error;
626
Sasha Levin309381fea2014-01-23 15:52:54 -0800627 VM_BUG_ON_PAGE(!PageLocked(page), page);
628 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700629
Johannes Weiner00501b52014-08-08 14:19:20 -0700630 if (!huge) {
631 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800632 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700633 if (error)
634 return error;
635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Jan Kara5e4c0d972013-09-11 14:26:05 -0700637 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700638 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700639 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800640 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700641 return error;
642 }
643
644 page_cache_get(page);
645 page->mapping = mapping;
646 page->index = offset;
647
648 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700649 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700650 radix_tree_preload_end();
651 if (unlikely(error))
652 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700653
654 /* hugetlb pages do not participate in page cache accounting. */
655 if (!huge)
656 __inc_zone_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700657 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700658 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800659 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700660 trace_mm_filemap_add_to_page_cache(page);
661 return 0;
662err_insert:
663 page->mapping = NULL;
664 /* Leave page->index set: truncation relies upon it */
665 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700666 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800667 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700668 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 return error;
670}
Johannes Weinera5289102014-04-03 14:47:51 -0700671
672/**
673 * add_to_page_cache_locked - add a locked page to the pagecache
674 * @page: page to add
675 * @mapping: the page's address_space
676 * @offset: page index
677 * @gfp_mask: page allocation mode
678 *
679 * This function is used to add a page to the pagecache. It must be locked.
680 * This function does not add the page to the LRU. The caller must do that.
681 */
682int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
683 pgoff_t offset, gfp_t gfp_mask)
684{
685 return __add_to_page_cache_locked(page, mapping, offset,
686 gfp_mask, NULL);
687}
Nick Piggine2867812008-07-25 19:45:30 -0700688EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400691 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Johannes Weinera5289102014-04-03 14:47:51 -0700693 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700694 int ret;
695
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800696 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700697 ret = __add_to_page_cache_locked(page, mapping, offset,
698 gfp_mask, &shadow);
699 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800700 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700701 else {
702 /*
703 * The page might have been evicted from cache only
704 * recently, in which case it should be activated like
705 * any other repeatedly accessed page.
706 */
707 if (shadow && workingset_refault(shadow)) {
708 SetPageActive(page);
709 workingset_activation(page);
710 } else
711 ClearPageActive(page);
712 lru_cache_add(page);
713 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 return ret;
715}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300716EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Paul Jackson44110fe2006-03-24 03:16:04 -0800718#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700719struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800720{
Miao Xiec0ff7452010-05-24 14:32:08 -0700721 int n;
722 struct page *page;
723
Paul Jackson44110fe2006-03-24 03:16:04 -0800724 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700725 unsigned int cpuset_mems_cookie;
726 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700727 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700728 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700729 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700730 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700731
Miao Xiec0ff7452010-05-24 14:32:08 -0700732 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800733 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700734 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800735}
Nick Piggin2ae88142006-10-28 10:38:23 -0700736EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800737#endif
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739/*
740 * In order to wait for pages to become available there must be
741 * waitqueues associated with pages. By using a hash table of
742 * waitqueues where the bucket discipline is to maintain all
743 * waiters on the same queue and wake all when any of the pages
744 * become available, and for the woken contexts to check to be
745 * sure the appropriate page became available, this saves space
746 * at a cost of "thundering herd" phenomena during rare hash
747 * collisions.
748 */
NeilBrowna4796e32014-09-24 11:28:32 +1000749wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 const struct zone *zone = page_zone(page);
752
753 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
754}
NeilBrowna4796e32014-09-24 11:28:32 +1000755EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Harvey Harrison920c7a52008-02-04 22:29:26 -0800757void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
759 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
760
761 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000762 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 TASK_UNINTERRUPTIBLE);
764}
765EXPORT_SYMBOL(wait_on_page_bit);
766
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700767int wait_on_page_bit_killable(struct page *page, int bit_nr)
768{
769 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
770
771 if (!test_bit(bit_nr, &page->flags))
772 return 0;
773
774 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000775 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700776}
777
NeilBrowncbbce822014-09-25 13:55:19 +1000778int wait_on_page_bit_killable_timeout(struct page *page,
779 int bit_nr, unsigned long timeout)
780{
781 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
782
783 wait.key.timeout = jiffies + timeout;
784 if (!test_bit(bit_nr, &page->flags))
785 return 0;
786 return __wait_on_bit(page_waitqueue(page), &wait,
787 bit_wait_io_timeout, TASK_KILLABLE);
788}
789EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100792 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700793 * @page: Page defining the wait queue of interest
794 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100795 *
796 * Add an arbitrary @waiter to the wait queue for the nominated @page.
797 */
798void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
799{
800 wait_queue_head_t *q = page_waitqueue(page);
801 unsigned long flags;
802
803 spin_lock_irqsave(&q->lock, flags);
804 __add_wait_queue(q, waiter);
805 spin_unlock_irqrestore(&q->lock, flags);
806}
807EXPORT_SYMBOL_GPL(add_page_wait_queue);
808
809/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700810 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 * @page: the page
812 *
813 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
814 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900815 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
817 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700818 * The mb is necessary to enforce ordering between the clear_bit and the read
819 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800821void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800823 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800824 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700825 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100826 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 wake_up_page(page, PG_locked);
828}
829EXPORT_SYMBOL(unlock_page);
830
Randy Dunlap485bb992006-06-23 02:03:49 -0700831/**
832 * end_page_writeback - end writeback against a page
833 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 */
835void end_page_writeback(struct page *page)
836{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700837 /*
838 * TestClearPageReclaim could be used here but it is an atomic
839 * operation and overkill in this particular case. Failing to
840 * shuffle a page marked for immediate reclaim is too mild to
841 * justify taking an atomic operation penalty at the end of
842 * ever page writeback.
843 */
844 if (PageReclaim(page)) {
845 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700846 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700847 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700848
849 if (!test_clear_page_writeback(page))
850 BUG();
851
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100852 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 wake_up_page(page, PG_writeback);
854}
855EXPORT_SYMBOL(end_page_writeback);
856
Matthew Wilcox57d99842014-06-04 16:07:45 -0700857/*
858 * After completing I/O on a page, call this routine to update the page
859 * flags appropriately
860 */
861void page_endio(struct page *page, int rw, int err)
862{
863 if (rw == READ) {
864 if (!err) {
865 SetPageUptodate(page);
866 } else {
867 ClearPageUptodate(page);
868 SetPageError(page);
869 }
870 unlock_page(page);
871 } else { /* rw == WRITE */
872 if (err) {
873 SetPageError(page);
874 if (page->mapping)
875 mapping_set_error(page->mapping, err);
876 }
877 end_page_writeback(page);
878 }
879}
880EXPORT_SYMBOL_GPL(page_endio);
881
Randy Dunlap485bb992006-06-23 02:03:49 -0700882/**
883 * __lock_page - get a lock on the page, assuming we need to sleep to get it
884 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800886void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800888 struct page *page_head = compound_head(page);
889 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800891 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 TASK_UNINTERRUPTIBLE);
893}
894EXPORT_SYMBOL(__lock_page);
895
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800896int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500897{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800898 struct page *page_head = compound_head(page);
899 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500900
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800901 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000902 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500903}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300904EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500905
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700906/*
907 * Return values:
908 * 1 - page is locked; mmap_sem is still held.
909 * 0 - page is not locked.
910 * mmap_sem has been released (up_read()), unless flags had both
911 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
912 * which case mmap_sem is still held.
913 *
914 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
915 * with the page locked and the mmap_sem unperturbed.
916 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700917int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
918 unsigned int flags)
919{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700920 if (flags & FAULT_FLAG_ALLOW_RETRY) {
921 /*
922 * CAUTION! In this case, mmap_sem is not released
923 * even though return 0.
924 */
925 if (flags & FAULT_FLAG_RETRY_NOWAIT)
926 return 0;
927
928 up_read(&mm->mmap_sem);
929 if (flags & FAULT_FLAG_KILLABLE)
930 wait_on_page_locked_killable(page);
931 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700932 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700933 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700934 } else {
935 if (flags & FAULT_FLAG_KILLABLE) {
936 int ret;
937
938 ret = __lock_page_killable(page);
939 if (ret) {
940 up_read(&mm->mmap_sem);
941 return 0;
942 }
943 } else
944 __lock_page(page);
945 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700946 }
947}
948
Randy Dunlap485bb992006-06-23 02:03:49 -0700949/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700950 * page_cache_next_hole - find the next hole (not-present entry)
951 * @mapping: mapping
952 * @index: index
953 * @max_scan: maximum range to search
954 *
955 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
956 * lowest indexed hole.
957 *
958 * Returns: the index of the hole if found, otherwise returns an index
959 * outside of the set specified (in which case 'return - index >=
960 * max_scan' will be true). In rare cases of index wrap-around, 0 will
961 * be returned.
962 *
963 * page_cache_next_hole may be called under rcu_read_lock. However,
964 * like radix_tree_gang_lookup, this will not atomically search a
965 * snapshot of the tree at a single point in time. For example, if a
966 * hole is created at index 5, then subsequently a hole is created at
967 * index 10, page_cache_next_hole covering both indexes may return 10
968 * if called under rcu_read_lock.
969 */
970pgoff_t page_cache_next_hole(struct address_space *mapping,
971 pgoff_t index, unsigned long max_scan)
972{
973 unsigned long i;
974
975 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700976 struct page *page;
977
978 page = radix_tree_lookup(&mapping->page_tree, index);
979 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700980 break;
981 index++;
982 if (index == 0)
983 break;
984 }
985
986 return index;
987}
988EXPORT_SYMBOL(page_cache_next_hole);
989
990/**
991 * page_cache_prev_hole - find the prev hole (not-present entry)
992 * @mapping: mapping
993 * @index: index
994 * @max_scan: maximum range to search
995 *
996 * Search backwards in the range [max(index-max_scan+1, 0), index] for
997 * the first hole.
998 *
999 * Returns: the index of the hole if found, otherwise returns an index
1000 * outside of the set specified (in which case 'index - return >=
1001 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1002 * will be returned.
1003 *
1004 * page_cache_prev_hole may be called under rcu_read_lock. However,
1005 * like radix_tree_gang_lookup, this will not atomically search a
1006 * snapshot of the tree at a single point in time. For example, if a
1007 * hole is created at index 10, then subsequently a hole is created at
1008 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1009 * called under rcu_read_lock.
1010 */
1011pgoff_t page_cache_prev_hole(struct address_space *mapping,
1012 pgoff_t index, unsigned long max_scan)
1013{
1014 unsigned long i;
1015
1016 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001017 struct page *page;
1018
1019 page = radix_tree_lookup(&mapping->page_tree, index);
1020 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001021 break;
1022 index--;
1023 if (index == ULONG_MAX)
1024 break;
1025 }
1026
1027 return index;
1028}
1029EXPORT_SYMBOL(page_cache_prev_hole);
1030
1031/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001032 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001033 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001034 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001035 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001036 * Looks up the page cache slot at @mapping & @offset. If there is a
1037 * page cache page, it is returned with an increased refcount.
1038 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001039 * If the slot holds a shadow entry of a previously evicted page, or a
1040 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001041 *
1042 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001044struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
Nick Piggina60637c2008-07-25 19:45:31 -07001046 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 struct page *page;
1048
Nick Piggina60637c2008-07-25 19:45:31 -07001049 rcu_read_lock();
1050repeat:
1051 page = NULL;
1052 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1053 if (pagep) {
1054 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001055 if (unlikely(!page))
1056 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001057 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001058 if (radix_tree_deref_retry(page))
1059 goto repeat;
1060 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001061 * A shadow entry of a recently evicted page,
1062 * or a swap entry from shmem/tmpfs. Return
1063 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001064 */
1065 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001066 }
Nick Piggina60637c2008-07-25 19:45:31 -07001067 if (!page_cache_get_speculative(page))
1068 goto repeat;
1069
1070 /*
1071 * Has the page moved?
1072 * This is part of the lockless pagecache protocol. See
1073 * include/linux/pagemap.h for details.
1074 */
1075 if (unlikely(page != *pagep)) {
1076 page_cache_release(page);
1077 goto repeat;
1078 }
1079 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001080out:
Nick Piggina60637c2008-07-25 19:45:31 -07001081 rcu_read_unlock();
1082
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 return page;
1084}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001085EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Randy Dunlap485bb992006-06-23 02:03:49 -07001087/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001088 * find_lock_entry - locate, pin and lock a page cache entry
1089 * @mapping: the address_space to search
1090 * @offset: the page cache index
1091 *
1092 * Looks up the page cache slot at @mapping & @offset. If there is a
1093 * page cache page, it is returned locked and with an increased
1094 * refcount.
1095 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001096 * If the slot holds a shadow entry of a previously evicted page, or a
1097 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001098 *
1099 * Otherwise, %NULL is returned.
1100 *
1101 * find_lock_entry() may sleep.
1102 */
1103struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104{
1105 struct page *page;
1106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001108 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001109 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001110 lock_page(page);
1111 /* Has the page been truncated? */
1112 if (unlikely(page->mapping != mapping)) {
1113 unlock_page(page);
1114 page_cache_release(page);
1115 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001117 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 return page;
1120}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001121EXPORT_SYMBOL(find_lock_entry);
1122
1123/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001124 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001125 * @mapping: the address_space to search
1126 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001127 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001128 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001129 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001130 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001131 *
Randy Dunlap75325182014-07-30 16:08:37 -07001132 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001133 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001134 * FGP_ACCESSED: the page will be marked accessed
1135 * FGP_LOCK: Page is return locked
1136 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001137 * @gfp_mask and added to the page cache and the VM's LRU
1138 * list. The page is returned locked and with an increased
1139 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001140 *
1141 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1142 * if the GFP flags specified for FGP_CREAT are atomic.
1143 *
1144 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001145 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001146struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001147 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148{
Nick Piggineb2be182007-10-16 01:24:57 -07001149 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001150
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001152 page = find_get_entry(mapping, offset);
1153 if (radix_tree_exceptional_entry(page))
1154 page = NULL;
1155 if (!page)
1156 goto no_page;
1157
1158 if (fgp_flags & FGP_LOCK) {
1159 if (fgp_flags & FGP_NOWAIT) {
1160 if (!trylock_page(page)) {
1161 page_cache_release(page);
1162 return NULL;
1163 }
1164 } else {
1165 lock_page(page);
1166 }
1167
1168 /* Has the page been truncated? */
1169 if (unlikely(page->mapping != mapping)) {
1170 unlock_page(page);
1171 page_cache_release(page);
1172 goto repeat;
1173 }
1174 VM_BUG_ON_PAGE(page->index != offset, page);
1175 }
1176
1177 if (page && (fgp_flags & FGP_ACCESSED))
1178 mark_page_accessed(page);
1179
1180no_page:
1181 if (!page && (fgp_flags & FGP_CREAT)) {
1182 int err;
1183 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001184 gfp_mask |= __GFP_WRITE;
1185 if (fgp_flags & FGP_NOFS)
1186 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001187
Michal Hocko45f87de2014-12-29 20:30:35 +01001188 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001189 if (!page)
1190 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001191
1192 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1193 fgp_flags |= FGP_LOCK;
1194
Hugh Dickinseb39d612014-08-06 16:06:43 -07001195 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001196 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001197 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001198
Michal Hocko45f87de2014-12-29 20:30:35 +01001199 err = add_to_page_cache_lru(page, mapping, offset,
1200 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001201 if (unlikely(err)) {
1202 page_cache_release(page);
1203 page = NULL;
1204 if (err == -EEXIST)
1205 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 return page;
1210}
Mel Gorman2457aec2014-06-04 16:10:31 -07001211EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001214 * find_get_entries - gang pagecache lookup
1215 * @mapping: The address_space to search
1216 * @start: The starting page cache index
1217 * @nr_entries: The maximum number of entries
1218 * @entries: Where the resulting entries are placed
1219 * @indices: The cache indices corresponding to the entries in @entries
1220 *
1221 * find_get_entries() will search for and return a group of up to
1222 * @nr_entries entries in the mapping. The entries are placed at
1223 * @entries. find_get_entries() takes a reference against any actual
1224 * pages it returns.
1225 *
1226 * The search returns a group of mapping-contiguous page cache entries
1227 * with ascending indexes. There may be holes in the indices due to
1228 * not-present pages.
1229 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001230 * Any shadow entries of evicted pages, or swap entries from
1231 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001232 *
1233 * find_get_entries() returns the number of pages and shadow entries
1234 * which were found.
1235 */
1236unsigned find_get_entries(struct address_space *mapping,
1237 pgoff_t start, unsigned int nr_entries,
1238 struct page **entries, pgoff_t *indices)
1239{
1240 void **slot;
1241 unsigned int ret = 0;
1242 struct radix_tree_iter iter;
1243
1244 if (!nr_entries)
1245 return 0;
1246
1247 rcu_read_lock();
1248restart:
1249 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1250 struct page *page;
1251repeat:
1252 page = radix_tree_deref_slot(slot);
1253 if (unlikely(!page))
1254 continue;
1255 if (radix_tree_exception(page)) {
1256 if (radix_tree_deref_retry(page))
1257 goto restart;
1258 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -08001259 * A shadow entry of a recently evicted page, a swap
1260 * entry from shmem/tmpfs or a DAX entry. Return it
1261 * without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001262 */
1263 goto export;
1264 }
1265 if (!page_cache_get_speculative(page))
1266 goto repeat;
1267
1268 /* Has the page moved? */
1269 if (unlikely(page != *slot)) {
1270 page_cache_release(page);
1271 goto repeat;
1272 }
1273export:
1274 indices[ret] = iter.index;
1275 entries[ret] = page;
1276 if (++ret == nr_entries)
1277 break;
1278 }
1279 rcu_read_unlock();
1280 return ret;
1281}
1282
1283/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 * find_get_pages - gang pagecache lookup
1285 * @mapping: The address_space to search
1286 * @start: The starting page index
1287 * @nr_pages: The maximum number of pages
1288 * @pages: Where the resulting pages are placed
1289 *
1290 * find_get_pages() will search for and return a group of up to
1291 * @nr_pages pages in the mapping. The pages are placed at @pages.
1292 * find_get_pages() takes a reference against the returned pages.
1293 *
1294 * The search returns a group of mapping-contiguous pages with ascending
1295 * indexes. There may be holes in the indices due to not-present pages.
1296 *
1297 * find_get_pages() returns the number of pages which were found.
1298 */
1299unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1300 unsigned int nr_pages, struct page **pages)
1301{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001302 struct radix_tree_iter iter;
1303 void **slot;
1304 unsigned ret = 0;
1305
1306 if (unlikely(!nr_pages))
1307 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Nick Piggina60637c2008-07-25 19:45:31 -07001309 rcu_read_lock();
1310restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001311 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001312 struct page *page;
1313repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001314 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001315 if (unlikely(!page))
1316 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001317
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001318 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001319 if (radix_tree_deref_retry(page)) {
1320 /*
1321 * Transient condition which can only trigger
1322 * when entry at index 0 moves out of or back
1323 * to root: none yet gotten, safe to restart.
1324 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001325 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001326 goto restart;
1327 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001328 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001329 * A shadow entry of a recently evicted page,
1330 * or a swap entry from shmem/tmpfs. Skip
1331 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001332 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001333 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001334 }
Nick Piggina60637c2008-07-25 19:45:31 -07001335
1336 if (!page_cache_get_speculative(page))
1337 goto repeat;
1338
1339 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001340 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001341 page_cache_release(page);
1342 goto repeat;
1343 }
1344
1345 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001346 if (++ret == nr_pages)
1347 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001348 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001349
Nick Piggina60637c2008-07-25 19:45:31 -07001350 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return ret;
1352}
1353
Jens Axboeebf43502006-04-27 08:46:01 +02001354/**
1355 * find_get_pages_contig - gang contiguous pagecache lookup
1356 * @mapping: The address_space to search
1357 * @index: The starting page index
1358 * @nr_pages: The maximum number of pages
1359 * @pages: Where the resulting pages are placed
1360 *
1361 * find_get_pages_contig() works exactly like find_get_pages(), except
1362 * that the returned number of pages are guaranteed to be contiguous.
1363 *
1364 * find_get_pages_contig() returns the number of pages which were found.
1365 */
1366unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1367 unsigned int nr_pages, struct page **pages)
1368{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001369 struct radix_tree_iter iter;
1370 void **slot;
1371 unsigned int ret = 0;
1372
1373 if (unlikely(!nr_pages))
1374 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001375
Nick Piggina60637c2008-07-25 19:45:31 -07001376 rcu_read_lock();
1377restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001378 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001379 struct page *page;
1380repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001381 page = radix_tree_deref_slot(slot);
1382 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001383 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001384 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001385
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001386 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001387 if (radix_tree_deref_retry(page)) {
1388 /*
1389 * Transient condition which can only trigger
1390 * when entry at index 0 moves out of or back
1391 * to root: none yet gotten, safe to restart.
1392 */
1393 goto restart;
1394 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001395 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001396 * A shadow entry of a recently evicted page,
1397 * or a swap entry from shmem/tmpfs. Stop
1398 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001399 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001400 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001401 }
Nick Piggina60637c2008-07-25 19:45:31 -07001402
Nick Piggina60637c2008-07-25 19:45:31 -07001403 if (!page_cache_get_speculative(page))
1404 goto repeat;
1405
1406 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001407 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001408 page_cache_release(page);
1409 goto repeat;
1410 }
1411
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001412 /*
1413 * must check mapping and index after taking the ref.
1414 * otherwise we can get both false positives and false
1415 * negatives, which is just confusing to the caller.
1416 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001417 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001418 page_cache_release(page);
1419 break;
1420 }
1421
Nick Piggina60637c2008-07-25 19:45:31 -07001422 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001423 if (++ret == nr_pages)
1424 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001425 }
Nick Piggina60637c2008-07-25 19:45:31 -07001426 rcu_read_unlock();
1427 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001428}
David Howellsef71c152007-05-09 02:33:44 -07001429EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001430
Randy Dunlap485bb992006-06-23 02:03:49 -07001431/**
1432 * find_get_pages_tag - find and return pages that match @tag
1433 * @mapping: the address_space to search
1434 * @index: the starting page index
1435 * @tag: the tag index
1436 * @nr_pages: the maximum number of pages
1437 * @pages: where the resulting pages are placed
1438 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001440 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 */
1442unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1443 int tag, unsigned int nr_pages, struct page **pages)
1444{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001445 struct radix_tree_iter iter;
1446 void **slot;
1447 unsigned ret = 0;
1448
1449 if (unlikely(!nr_pages))
1450 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Nick Piggina60637c2008-07-25 19:45:31 -07001452 rcu_read_lock();
1453restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001454 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1455 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001456 struct page *page;
1457repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001458 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001459 if (unlikely(!page))
1460 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001461
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001462 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001463 if (radix_tree_deref_retry(page)) {
1464 /*
1465 * Transient condition which can only trigger
1466 * when entry at index 0 moves out of or back
1467 * to root: none yet gotten, safe to restart.
1468 */
1469 goto restart;
1470 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001471 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001472 * A shadow entry of a recently evicted page.
1473 *
1474 * Those entries should never be tagged, but
1475 * this tree walk is lockless and the tags are
1476 * looked up in bulk, one radix tree node at a
1477 * time, so there is a sizable window for page
1478 * reclaim to evict a page we saw tagged.
1479 *
1480 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001481 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001482 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001483 }
Nick Piggina60637c2008-07-25 19:45:31 -07001484
1485 if (!page_cache_get_speculative(page))
1486 goto repeat;
1487
1488 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001489 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001490 page_cache_release(page);
1491 goto repeat;
1492 }
1493
1494 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001495 if (++ret == nr_pages)
1496 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001497 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001498
Nick Piggina60637c2008-07-25 19:45:31 -07001499 rcu_read_unlock();
1500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 if (ret)
1502 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 return ret;
1505}
David Howellsef71c152007-05-09 02:33:44 -07001506EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001508/**
1509 * find_get_entries_tag - find and return entries that match @tag
1510 * @mapping: the address_space to search
1511 * @start: the starting page cache index
1512 * @tag: the tag index
1513 * @nr_entries: the maximum number of entries
1514 * @entries: where the resulting entries are placed
1515 * @indices: the cache indices corresponding to the entries in @entries
1516 *
1517 * Like find_get_entries, except we only return entries which are tagged with
1518 * @tag.
1519 */
1520unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1521 int tag, unsigned int nr_entries,
1522 struct page **entries, pgoff_t *indices)
1523{
1524 void **slot;
1525 unsigned int ret = 0;
1526 struct radix_tree_iter iter;
1527
1528 if (!nr_entries)
1529 return 0;
1530
1531 rcu_read_lock();
1532restart:
1533 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1534 &iter, start, tag) {
1535 struct page *page;
1536repeat:
1537 page = radix_tree_deref_slot(slot);
1538 if (unlikely(!page))
1539 continue;
1540 if (radix_tree_exception(page)) {
1541 if (radix_tree_deref_retry(page)) {
1542 /*
1543 * Transient condition which can only trigger
1544 * when entry at index 0 moves out of or back
1545 * to root: none yet gotten, safe to restart.
1546 */
1547 goto restart;
1548 }
1549
1550 /*
1551 * A shadow entry of a recently evicted page, a swap
1552 * entry from shmem/tmpfs or a DAX entry. Return it
1553 * without attempting to raise page count.
1554 */
1555 goto export;
1556 }
1557 if (!page_cache_get_speculative(page))
1558 goto repeat;
1559
1560 /* Has the page moved? */
1561 if (unlikely(page != *slot)) {
1562 page_cache_release(page);
1563 goto repeat;
1564 }
1565export:
1566 indices[ret] = iter.index;
1567 entries[ret] = page;
1568 if (++ret == nr_entries)
1569 break;
1570 }
1571 rcu_read_unlock();
1572 return ret;
1573}
1574EXPORT_SYMBOL(find_get_entries_tag);
1575
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001576/*
1577 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1578 * a _large_ part of the i/o request. Imagine the worst scenario:
1579 *
1580 * ---R__________________________________________B__________
1581 * ^ reading here ^ bad block(assume 4k)
1582 *
1583 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1584 * => failing the whole request => read(R) => read(R+1) =>
1585 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1586 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1587 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1588 *
1589 * It is going insane. Fix it by quickly scaling down the readahead size.
1590 */
1591static void shrink_readahead_size_eio(struct file *filp,
1592 struct file_ra_state *ra)
1593{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001594 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001595}
1596
Randy Dunlap485bb992006-06-23 02:03:49 -07001597/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001598 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001599 * @filp: the file to read
1600 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001601 * @iter: data destination
1602 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001603 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001605 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 *
1607 * This is really ugly. But the goto's actually try to clarify some
1608 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 */
Al Viro6e58e792014-02-03 17:07:03 -05001610static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1611 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001613 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001615 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001616 pgoff_t index;
1617 pgoff_t last_index;
1618 pgoff_t prev_index;
1619 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001620 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001621 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001624 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1625 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Al Viro6e58e792014-02-03 17:07:03 -05001626 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 offset = *ppos & ~PAGE_CACHE_MASK;
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 for (;;) {
1630 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001631 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001632 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 unsigned long nr, ret;
1634
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636find_page:
1637 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001638 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001639 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001640 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001641 index, last_index - index);
1642 page = find_get_page(mapping, index);
1643 if (unlikely(page == NULL))
1644 goto no_cached_page;
1645 }
1646 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001647 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001648 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001649 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001651 if (!PageUptodate(page)) {
1652 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1653 !mapping->a_ops->is_partially_uptodate)
1654 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001655 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001656 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001657 /* Did it get truncated before we got the lock? */
1658 if (!page->mapping)
1659 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001660 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001661 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001662 goto page_not_up_to_date_locked;
1663 unlock_page(page);
1664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001666 /*
1667 * i_size must be checked after we know the page is Uptodate.
1668 *
1669 * Checking i_size after the check allows us to calculate
1670 * the correct value for "nr", which means the zero-filled
1671 * part of the page is not copied back to userspace (unless
1672 * another truncate extends the file - this is desired though).
1673 */
1674
1675 isize = i_size_read(inode);
1676 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1677 if (unlikely(!isize || index > end_index)) {
1678 page_cache_release(page);
1679 goto out;
1680 }
1681
1682 /* nr is the maximum number of bytes to copy from this page */
1683 nr = PAGE_CACHE_SIZE;
1684 if (index == end_index) {
1685 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1686 if (nr <= offset) {
1687 page_cache_release(page);
1688 goto out;
1689 }
1690 }
1691 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693 /* If users can be writing to this page using arbitrary
1694 * virtual addresses, take care about potential aliasing
1695 * before reading the page on the kernel side.
1696 */
1697 if (mapping_writably_mapped(mapping))
1698 flush_dcache_page(page);
1699
1700 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001701 * When a sequential read accesses a page several times,
1702 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 */
Jan Karaec0f1632007-05-06 14:49:25 -07001704 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 mark_page_accessed(page);
1706 prev_index = index;
1707
1708 /*
1709 * Ok, we have the page, and it's up-to-date, so
1710 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 */
Al Viro6e58e792014-02-03 17:07:03 -05001712
1713 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 offset += ret;
1715 index += offset >> PAGE_CACHE_SHIFT;
1716 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001717 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001720 written += ret;
1721 if (!iov_iter_count(iter))
1722 goto out;
1723 if (ret < nr) {
1724 error = -EFAULT;
1725 goto out;
1726 }
1727 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729page_not_up_to_date:
1730 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001731 error = lock_page_killable(page);
1732 if (unlikely(error))
1733 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001735page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001736 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 if (!page->mapping) {
1738 unlock_page(page);
1739 page_cache_release(page);
1740 continue;
1741 }
1742
1743 /* Did somebody else fill it already? */
1744 if (PageUptodate(page)) {
1745 unlock_page(page);
1746 goto page_ok;
1747 }
1748
1749readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001750 /*
1751 * A previous I/O error may have been due to temporary
1752 * failures, eg. multipath errors.
1753 * PG_error will be set again if readpage fails.
1754 */
1755 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 /* Start the actual read. The read will unlock the page. */
1757 error = mapping->a_ops->readpage(filp, page);
1758
Zach Brown994fc28c2005-12-15 14:28:17 -08001759 if (unlikely(error)) {
1760 if (error == AOP_TRUNCATED_PAGE) {
1761 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001762 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001763 goto find_page;
1764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
1768 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001769 error = lock_page_killable(page);
1770 if (unlikely(error))
1771 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 if (!PageUptodate(page)) {
1773 if (page->mapping == NULL) {
1774 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001775 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 */
1777 unlock_page(page);
1778 page_cache_release(page);
1779 goto find_page;
1780 }
1781 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001782 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001783 error = -EIO;
1784 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 }
1786 unlock_page(page);
1787 }
1788
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 goto page_ok;
1790
1791readpage_error:
1792 /* UHHUH! A synchronous read error occurred. Report it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 page_cache_release(page);
1794 goto out;
1795
1796no_cached_page:
1797 /*
1798 * Ok, it wasn't cached, so we need to create a new
1799 * page..
1800 */
Nick Piggineb2be182007-10-16 01:24:57 -07001801 page = page_cache_alloc_cold(mapping);
1802 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001803 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001804 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001806 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08001807 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001809 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001810 if (error == -EEXIST) {
1811 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001813 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 goto out;
1815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 goto readpage;
1817 }
1818
1819out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001820 ra->prev_pos = prev_index;
1821 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1822 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001824 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001825 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001826 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827}
1828
Randy Dunlap485bb992006-06-23 02:03:49 -07001829/**
Al Viro6abd2322014-04-04 14:20:57 -04001830 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001831 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001832 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001833 *
Al Viro6abd2322014-04-04 14:20:57 -04001834 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 * that can use the page cache directly.
1836 */
1837ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001838generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839{
Al Viroed978a82014-03-05 22:53:04 -05001840 struct file *file = iocb->ki_filp;
1841 ssize_t retval = 0;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001842 loff_t *ppos = &iocb->ki_pos;
Al Viroed978a82014-03-05 22:53:04 -05001843 loff_t pos = *ppos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844
Al Viro2ba48ce2015-04-09 13:52:01 -04001845 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001846 struct address_space *mapping = file->f_mapping;
1847 struct inode *inode = mapping->host;
1848 size_t count = iov_iter_count(iter);
Badari Pulavarty543ade12006-09-30 23:28:48 -07001849 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 if (!count)
1852 goto out; /* skip atime */
1853 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001854 retval = filemap_write_and_wait_range(mapping, pos,
Al Viroed978a82014-03-05 22:53:04 -05001855 pos + count - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001856 if (!retval) {
Al Viroed978a82014-03-05 22:53:04 -05001857 struct iov_iter data = *iter;
Omar Sandoval22c61862015-03-16 04:33:53 -07001858 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001859 }
Al Viroed978a82014-03-05 22:53:04 -05001860
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001861 if (retval > 0) {
1862 *ppos = pos + retval;
Al Viroed978a82014-03-05 22:53:04 -05001863 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001864 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001865
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001866 /*
1867 * Btrfs can have a short DIO read if we encounter
1868 * compressed extents, so if there was an error, or if
1869 * we've already read everything we wanted to, or if
1870 * there was a short read because we hit EOF, go ahead
1871 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001872 * the rest of the read. Buffered reads will not work for
1873 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001874 */
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001875 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1876 IS_DAX(inode)) {
Al Viroed978a82014-03-05 22:53:04 -05001877 file_accessed(file);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001878 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 }
1881
Al Viroed978a82014-03-05 22:53:04 -05001882 retval = do_generic_file_read(file, ppos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883out:
1884 return retval;
1885}
Al Viroed978a82014-03-05 22:53:04 -05001886EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001889/**
1890 * page_cache_read - adds requested page to the page cache if not already there
1891 * @file: file to read
1892 * @offset: page index
1893 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 * This adds the requested page to the page cache if it isn't already there,
1895 * and schedules an I/O to read in its contents from disk.
1896 */
Michal Hockoc20cd452016-01-14 15:20:12 -08001897static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
1899 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001900 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001901 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Zach Brown994fc28c2005-12-15 14:28:17 -08001903 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08001904 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08001905 if (!page)
1906 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
Michal Hockoc20cd452016-01-14 15:20:12 -08001908 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
Zach Brown994fc28c2005-12-15 14:28:17 -08001909 if (ret == 0)
1910 ret = mapping->a_ops->readpage(file, page);
1911 else if (ret == -EEXIST)
1912 ret = 0; /* losing race to add is OK */
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
Zach Brown994fc28c2005-12-15 14:28:17 -08001916 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07001917
Zach Brown994fc28c2005-12-15 14:28:17 -08001918 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919}
1920
1921#define MMAP_LOTSAMISS (100)
1922
Linus Torvaldsef00e082009-06-16 15:31:25 -07001923/*
1924 * Synchronous readahead happens when we don't even find
1925 * a page in the page cache at all.
1926 */
1927static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1928 struct file_ra_state *ra,
1929 struct file *file,
1930 pgoff_t offset)
1931{
Linus Torvaldsef00e082009-06-16 15:31:25 -07001932 struct address_space *mapping = file->f_mapping;
1933
1934 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001935 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001936 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001937 if (!ra->ra_pages)
1938 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001939
Joe Perches64363aa2013-07-08 16:00:18 -07001940 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001941 page_cache_sync_readahead(mapping, ra, file, offset,
1942 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001943 return;
1944 }
1945
Andi Kleen207d04b2011-05-24 17:12:29 -07001946 /* Avoid banging the cache line if not needed */
1947 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001948 ra->mmap_miss++;
1949
1950 /*
1951 * Do we miss much more than hit in this file? If so,
1952 * stop bothering with read-ahead. It will only hurt.
1953 */
1954 if (ra->mmap_miss > MMAP_LOTSAMISS)
1955 return;
1956
Wu Fengguangd30a1102009-06-16 15:31:30 -07001957 /*
1958 * mmap read-around
1959 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08001960 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1961 ra->size = ra->ra_pages;
1962 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001963 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001964}
1965
1966/*
1967 * Asynchronous readahead happens when we find the page and PG_readahead,
1968 * so we want to possibly extend the readahead further..
1969 */
1970static void do_async_mmap_readahead(struct vm_area_struct *vma,
1971 struct file_ra_state *ra,
1972 struct file *file,
1973 struct page *page,
1974 pgoff_t offset)
1975{
1976 struct address_space *mapping = file->f_mapping;
1977
1978 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001979 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001980 return;
1981 if (ra->mmap_miss > 0)
1982 ra->mmap_miss--;
1983 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001984 page_cache_async_readahead(mapping, ra, file,
1985 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001986}
1987
Randy Dunlap485bb992006-06-23 02:03:49 -07001988/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001989 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001990 * @vma: vma in which the fault was taken
1991 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001992 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001993 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 * mapped memory region to read in file data during a page fault.
1995 *
1996 * The goto's are kind of ugly, but this streamlines the normal case of having
1997 * it in the page cache, and handles the special cases reasonably without
1998 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001999 *
2000 * vma->vm_mm->mmap_sem must be held on entry.
2001 *
2002 * If our return value has VM_FAULT_RETRY set, it's because
2003 * lock_page_or_retry() returned 0.
2004 * The mmap_sem has usually been released in this case.
2005 * See __lock_page_or_retry() for the exception.
2006 *
2007 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2008 * has not been released.
2009 *
2010 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002012int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013{
2014 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07002015 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 struct address_space *mapping = file->f_mapping;
2017 struct file_ra_state *ra = &file->f_ra;
2018 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002019 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002021 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07002022 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002024 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2025 if (offset >= size >> PAGE_CACHE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07002026 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002029 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002031 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002032 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002034 * We found the page, so try async readahead before
2035 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002037 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002038 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002039 /* No page in the page cache at all */
2040 do_sync_mmap_readahead(vma, ra, file, offset);
2041 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002042 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002043 ret = VM_FAULT_MAJOR;
2044retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07002045 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 if (!page)
2047 goto no_cached_page;
2048 }
2049
Michel Lespinassed88c0922010-11-02 13:05:18 -07002050 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2051 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002052 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07002053 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07002054
2055 /* Did it get truncated? */
2056 if (unlikely(page->mapping != mapping)) {
2057 unlock_page(page);
2058 put_page(page);
2059 goto retry_find;
2060 }
Sasha Levin309381fea2014-01-23 15:52:54 -08002061 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002064 * We have a locked page in the page cache, now we need to check
2065 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 */
Nick Piggind00806b2007-07-19 01:46:57 -07002067 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 goto page_not_uptodate;
2069
Linus Torvaldsef00e082009-06-16 15:31:25 -07002070 /*
2071 * Found the page and have a reference on it.
2072 * We must recheck i_size under page lock.
2073 */
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002074 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2075 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002076 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07002077 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002078 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002079 }
2080
Nick Piggind0217ac2007-07-19 01:47:03 -07002081 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002082 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084no_cached_page:
2085 /*
2086 * We're only likely to ever get here if MADV_RANDOM is in
2087 * effect.
2088 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002089 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090
2091 /*
2092 * The page we want has now been added to the page cache.
2093 * In the unlikely event that someone removed it in the
2094 * meantime, we'll just come back here and read it again.
2095 */
2096 if (error >= 0)
2097 goto retry_find;
2098
2099 /*
2100 * An error return from page_cache_read can result if the
2101 * system is low on memory, or a problem occurs while trying
2102 * to schedule I/O.
2103 */
2104 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002105 return VM_FAULT_OOM;
2106 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
2108page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 /*
2110 * Umm, take care of errors if the page isn't up-to-date.
2111 * Try to re-read it _once_. We do this synchronously,
2112 * because there really aren't any performance issues here
2113 * and we need to check for errors.
2114 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002116 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002117 if (!error) {
2118 wait_on_page_locked(page);
2119 if (!PageUptodate(page))
2120 error = -EIO;
2121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002123
2124 if (!error || error == AOP_TRUNCATED_PAGE)
2125 goto retry_find;
2126
2127 /* Things didn't work out. Return zero to tell the mm layer so. */
2128 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002129 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002130}
2131EXPORT_SYMBOL(filemap_fault);
2132
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002133void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2134{
2135 struct radix_tree_iter iter;
2136 void **slot;
2137 struct file *file = vma->vm_file;
2138 struct address_space *mapping = file->f_mapping;
2139 loff_t size;
2140 struct page *page;
2141 unsigned long address = (unsigned long) vmf->virtual_address;
2142 unsigned long addr;
2143 pte_t *pte;
2144
2145 rcu_read_lock();
2146 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2147 if (iter.index > vmf->max_pgoff)
2148 break;
2149repeat:
2150 page = radix_tree_deref_slot(slot);
2151 if (unlikely(!page))
2152 goto next;
2153 if (radix_tree_exception(page)) {
2154 if (radix_tree_deref_retry(page))
2155 break;
2156 else
2157 goto next;
2158 }
2159
2160 if (!page_cache_get_speculative(page))
2161 goto repeat;
2162
2163 /* Has the page moved? */
2164 if (unlikely(page != *slot)) {
2165 page_cache_release(page);
2166 goto repeat;
2167 }
2168
2169 if (!PageUptodate(page) ||
2170 PageReadahead(page) ||
2171 PageHWPoison(page))
2172 goto skip;
2173 if (!trylock_page(page))
2174 goto skip;
2175
2176 if (page->mapping != mapping || !PageUptodate(page))
2177 goto unlock;
2178
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002179 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2180 if (page->index >= size >> PAGE_CACHE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002181 goto unlock;
2182
2183 pte = vmf->pte + page->index - vmf->pgoff;
2184 if (!pte_none(*pte))
2185 goto unlock;
2186
2187 if (file->f_ra.mmap_miss > 0)
2188 file->f_ra.mmap_miss--;
2189 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2190 do_set_pte(vma, addr, page, pte, false, false);
2191 unlock_page(page);
2192 goto next;
2193unlock:
2194 unlock_page(page);
2195skip:
2196 page_cache_release(page);
2197next:
2198 if (iter.index == vmf->max_pgoff)
2199 break;
2200 }
2201 rcu_read_unlock();
2202}
2203EXPORT_SYMBOL(filemap_map_pages);
2204
Jan Kara4fcf1c62012-06-12 16:20:29 +02002205int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2206{
2207 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002208 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002209 int ret = VM_FAULT_LOCKED;
2210
Jan Kara14da9202012-06-12 16:20:37 +02002211 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002212 file_update_time(vma->vm_file);
2213 lock_page(page);
2214 if (page->mapping != inode->i_mapping) {
2215 unlock_page(page);
2216 ret = VM_FAULT_NOPAGE;
2217 goto out;
2218 }
Jan Kara14da9202012-06-12 16:20:37 +02002219 /*
2220 * We mark the page dirty already here so that when freeze is in
2221 * progress, we are guaranteed that writeback during freezing will
2222 * see the dirty page and writeprotect it again.
2223 */
2224 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002225 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002226out:
Jan Kara14da9202012-06-12 16:20:37 +02002227 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002228 return ret;
2229}
2230EXPORT_SYMBOL(filemap_page_mkwrite);
2231
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002232const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002233 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002234 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002235 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236};
2237
2238/* This is used for a general mmap of a disk file */
2239
2240int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2241{
2242 struct address_space *mapping = file->f_mapping;
2243
2244 if (!mapping->a_ops->readpage)
2245 return -ENOEXEC;
2246 file_accessed(file);
2247 vma->vm_ops = &generic_file_vm_ops;
2248 return 0;
2249}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250
2251/*
2252 * This is for filesystems which do not implement ->writepage.
2253 */
2254int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2255{
2256 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2257 return -EINVAL;
2258 return generic_file_mmap(file, vma);
2259}
2260#else
2261int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2262{
2263 return -ENOSYS;
2264}
2265int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2266{
2267 return -ENOSYS;
2268}
2269#endif /* CONFIG_MMU */
2270
2271EXPORT_SYMBOL(generic_file_mmap);
2272EXPORT_SYMBOL(generic_file_readonly_mmap);
2273
Sasha Levin67f9fd92014-04-03 14:48:18 -07002274static struct page *wait_on_page_read(struct page *page)
2275{
2276 if (!IS_ERR(page)) {
2277 wait_on_page_locked(page);
2278 if (!PageUptodate(page)) {
2279 page_cache_release(page);
2280 page = ERR_PTR(-EIO);
2281 }
2282 }
2283 return page;
2284}
2285
Nick Piggin6fe69002007-05-06 14:49:04 -07002286static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002287 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002288 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002289 void *data,
2290 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291{
Nick Piggineb2be182007-10-16 01:24:57 -07002292 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 int err;
2294repeat:
2295 page = find_get_page(mapping, index);
2296 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002297 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002298 if (!page)
2299 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002300 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002301 if (unlikely(err)) {
2302 page_cache_release(page);
2303 if (err == -EEXIST)
2304 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 return ERR_PTR(err);
2307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 err = filler(data, page);
2309 if (err < 0) {
2310 page_cache_release(page);
2311 page = ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002312 } else {
2313 page = wait_on_page_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 }
2315 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 return page;
2317}
2318
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002319static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002320 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002321 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002322 void *data,
2323 gfp_t gfp)
2324
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325{
2326 struct page *page;
2327 int err;
2328
2329retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002330 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01002332 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 if (PageUptodate(page))
2334 goto out;
2335
2336 lock_page(page);
2337 if (!page->mapping) {
2338 unlock_page(page);
2339 page_cache_release(page);
2340 goto retry;
2341 }
2342 if (PageUptodate(page)) {
2343 unlock_page(page);
2344 goto out;
2345 }
2346 err = filler(data, page);
2347 if (err < 0) {
2348 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01002349 return ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002350 } else {
2351 page = wait_on_page_read(page);
2352 if (IS_ERR(page))
2353 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 }
David Howellsc855ff32007-05-09 13:42:20 +01002355out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002356 mark_page_accessed(page);
2357 return page;
2358}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002359
2360/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002361 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002362 * @mapping: the page's address_space
2363 * @index: the page index
2364 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002365 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002366 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002367 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002368 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002369 *
2370 * If the page does not get brought uptodate, return -EIO.
2371 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002372struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002373 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002374 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002375 void *data)
2376{
2377 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2378}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002379EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002380
2381/**
2382 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2383 * @mapping: the page's address_space
2384 * @index: the page index
2385 * @gfp: the page allocator flags to use if allocating
2386 *
2387 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002388 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002389 *
2390 * If the page does not get brought uptodate, return -EIO.
2391 */
2392struct page *read_cache_page_gfp(struct address_space *mapping,
2393 pgoff_t index,
2394 gfp_t gfp)
2395{
2396 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2397
Sasha Levin67f9fd92014-04-03 14:48:18 -07002398 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002399}
2400EXPORT_SYMBOL(read_cache_page_gfp);
2401
Nick Piggin2f718ff2007-10-16 01:24:59 -07002402/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 * Performs necessary checks before doing a write
2404 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002405 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 * Returns appropriate error code that caller should return or
2407 * zero in case that write should be allowed.
2408 */
Al Viro3309dd02015-04-09 12:55:47 -04002409inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410{
Al Viro3309dd02015-04-09 12:55:47 -04002411 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002413 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002414 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Al Viro3309dd02015-04-09 12:55:47 -04002416 if (!iov_iter_count(from))
2417 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
Al Viro0fa6b002015-04-04 04:05:48 -04002419 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002420 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002421 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
Al Viro3309dd02015-04-09 12:55:47 -04002423 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424
Al Viro0fa6b002015-04-04 04:05:48 -04002425 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002426 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002427 send_sig(SIGXFSZ, current, 0);
2428 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 }
Al Viro3309dd02015-04-09 12:55:47 -04002430 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 }
2432
2433 /*
2434 * LFS rule
2435 */
Al Viro3309dd02015-04-09 12:55:47 -04002436 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002438 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002440 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 }
2442
2443 /*
2444 * Are we about to exceed the fs block limit ?
2445 *
2446 * If we have written data it becomes a short write. If we have
2447 * exceeded without writing data we send a signal and return EFBIG.
2448 * Linus frestrict idea will clean these up nicely..
2449 */
Al Viro3309dd02015-04-09 12:55:47 -04002450 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2451 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
Al Viro3309dd02015-04-09 12:55:47 -04002453 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2454 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455}
2456EXPORT_SYMBOL(generic_write_checks);
2457
Nick Pigginafddba42007-10-16 01:25:01 -07002458int pagecache_write_begin(struct file *file, struct address_space *mapping,
2459 loff_t pos, unsigned len, unsigned flags,
2460 struct page **pagep, void **fsdata)
2461{
2462 const struct address_space_operations *aops = mapping->a_ops;
2463
Nick Piggin4e02ed42008-10-29 14:00:55 -07002464 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002465 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002466}
2467EXPORT_SYMBOL(pagecache_write_begin);
2468
2469int pagecache_write_end(struct file *file, struct address_space *mapping,
2470 loff_t pos, unsigned len, unsigned copied,
2471 struct page *page, void *fsdata)
2472{
2473 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002474
Nick Piggin4e02ed42008-10-29 14:00:55 -07002475 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002476}
2477EXPORT_SYMBOL(pagecache_write_end);
2478
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479ssize_t
Al Viro0c949332014-03-22 06:51:37 -04002480generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481{
2482 struct file *file = iocb->ki_filp;
2483 struct address_space *mapping = file->f_mapping;
2484 struct inode *inode = mapping->host;
2485 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002486 size_t write_len;
2487 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002488 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
Al Viro0c949332014-03-22 06:51:37 -04002490 write_len = iov_iter_count(from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002491 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002492
Nick Piggin48b47c52009-01-06 14:40:22 -08002493 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002494 if (written)
2495 goto out;
2496
2497 /*
2498 * After a write we want buffered reads to be sure to go to disk to get
2499 * the new data. We invalidate clean cached page from the region we're
2500 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002501 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002502 */
2503 if (mapping->nrpages) {
2504 written = invalidate_inode_pages2_range(mapping,
2505 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002506 /*
2507 * If a page can not be invalidated, return 0 to fall back
2508 * to buffered write.
2509 */
2510 if (written) {
2511 if (written == -EBUSY)
2512 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002513 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002514 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002515 }
2516
Al Viro26978b82014-03-10 14:08:45 -04002517 data = *from;
Omar Sandoval22c61862015-03-16 04:33:53 -07002518 written = mapping->a_ops->direct_IO(iocb, &data, pos);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002519
2520 /*
2521 * Finally, try again to invalidate clean pages which might have been
2522 * cached by non-direct readahead, or faulted in by get_user_pages()
2523 * if the source of the write was an mmap'ed region of the file
2524 * we're writing. Either one is a pretty crazy thing to do,
2525 * so we don't support it 100%. If this invalidation
2526 * fails, tough, the write still worked...
2527 */
2528 if (mapping->nrpages) {
2529 invalidate_inode_pages2_range(mapping,
2530 pos >> PAGE_CACHE_SHIFT, end);
2531 }
2532
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002534 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002535 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002536 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2537 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 mark_inode_dirty(inode);
2539 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002540 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002542out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 return written;
2544}
2545EXPORT_SYMBOL(generic_file_direct_write);
2546
Nick Piggineb2be182007-10-16 01:24:57 -07002547/*
2548 * Find or create a page at the given pagecache position. Return the locked
2549 * page. This function is specifically for buffered writes.
2550 */
Nick Piggin54566b22009-01-04 12:00:53 -08002551struct page *grab_cache_page_write_begin(struct address_space *mapping,
2552 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002553{
Nick Piggineb2be182007-10-16 01:24:57 -07002554 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07002555 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002556
Nick Piggin54566b22009-01-04 12:00:53 -08002557 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002558 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002559
Mel Gorman2457aec2014-06-04 16:10:31 -07002560 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002561 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002562 if (page)
2563 wait_for_stable_page(page);
2564
Nick Piggineb2be182007-10-16 01:24:57 -07002565 return page;
2566}
Nick Piggin54566b22009-01-04 12:00:53 -08002567EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002568
Al Viro3b93f912014-02-11 21:34:08 -05002569ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002570 struct iov_iter *i, loff_t pos)
2571{
2572 struct address_space *mapping = file->f_mapping;
2573 const struct address_space_operations *a_ops = mapping->a_ops;
2574 long status = 0;
2575 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002576 unsigned int flags = 0;
2577
2578 /*
2579 * Copies from kernel address space cannot fail (NFSD is a big user).
2580 */
Al Viro777eda22014-12-17 04:46:46 -05002581 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002582 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002583
2584 do {
2585 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002586 unsigned long offset; /* Offset into pagecache page */
2587 unsigned long bytes; /* Bytes to write to page */
2588 size_t copied; /* Bytes copied from user */
2589 void *fsdata;
2590
2591 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002592 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2593 iov_iter_count(i));
2594
2595again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002596 /*
2597 * Bring in the user page that we will copy from _first_.
2598 * Otherwise there's a nasty deadlock on copying from the
2599 * same page as we're writing to, without it being marked
2600 * up-to-date.
2601 *
2602 * Not only is this an optimisation, but it is also required
2603 * to check that the address is actually valid, when atomic
2604 * usercopies are used, below.
2605 */
2606 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2607 status = -EFAULT;
2608 break;
2609 }
2610
Jan Kara296291c2015-10-22 13:32:21 -07002611 if (fatal_signal_pending(current)) {
2612 status = -EINTR;
2613 break;
2614 }
2615
Nick Piggin674b8922007-10-16 01:25:03 -07002616 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002617 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002618 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002619 break;
2620
anfei zhou931e80e2010-02-02 13:44:02 -08002621 if (mapping_writably_mapped(mapping))
2622 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002623
Nick Pigginafddba42007-10-16 01:25:01 -07002624 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002625 flush_dcache_page(page);
2626
2627 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2628 page, fsdata);
2629 if (unlikely(status < 0))
2630 break;
2631 copied = status;
2632
2633 cond_resched();
2634
Nick Piggin124d3b72008-02-02 15:01:17 +01002635 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002636 if (unlikely(copied == 0)) {
2637 /*
2638 * If we were unable to copy any data at all, we must
2639 * fall back to a single segment length write.
2640 *
2641 * If we didn't fallback here, we could livelock
2642 * because not all segments in the iov can be copied at
2643 * once without a pagefault.
2644 */
2645 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2646 iov_iter_single_seg_count(i));
2647 goto again;
2648 }
Nick Pigginafddba42007-10-16 01:25:01 -07002649 pos += copied;
2650 written += copied;
2651
2652 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002653 } while (iov_iter_count(i));
2654
2655 return written ? written : status;
2656}
Al Viro3b93f912014-02-11 21:34:08 -05002657EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658
Jan Karae4dd9de2009-08-17 18:10:06 +02002659/**
Al Viro81742022014-04-03 03:17:43 -04002660 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002661 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002662 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002663 *
2664 * This function does all the work needed for actually writing data to a
2665 * file. It does all basic checks, removes SUID from the file, updates
2666 * modification times and calls proper subroutines depending on whether we
2667 * do direct IO or a standard buffered write.
2668 *
2669 * It expects i_mutex to be grabbed unless we work on a block device or similar
2670 * object which does not need locking at all.
2671 *
2672 * This function does *not* take care of syncing data in case of O_SYNC write.
2673 * A caller has to handle it. This is mainly due to the fact that we want to
2674 * avoid syncing under i_mutex.
2675 */
Al Viro81742022014-04-03 03:17:43 -04002676ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677{
2678 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002679 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002681 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002683 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002686 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002687 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (err)
2689 goto out;
2690
Josef Bacikc3b2da32012-03-26 09:59:21 -04002691 err = file_update_time(file);
2692 if (err)
2693 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
Al Viro2ba48ce2015-04-09 13:52:01 -04002695 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002696 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002697
Al Viro0b8def92015-04-07 10:22:53 -04002698 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002699 /*
2700 * If the write stopped short of completing, fall back to
2701 * buffered writes. Some filesystems do this for writes to
2702 * holes, for example. For DAX files, a buffered write will
2703 * not succeed (even if it did, DAX does not handle dirty
2704 * page-cache pages correctly).
2705 */
Al Viro0b8def92015-04-07 10:22:53 -04002706 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002708
Al Viro0b8def92015-04-07 10:22:53 -04002709 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002710 /*
Al Viro3b93f912014-02-11 21:34:08 -05002711 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002712 * then we want to return the number of bytes which were
2713 * direct-written, or the error code if that was zero. Note
2714 * that this differs from normal direct-io semantics, which
2715 * will return -EFOO even if some bytes were written.
2716 */
Al Viro60bb4522014-08-08 12:39:16 -04002717 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002718 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002719 goto out;
2720 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002721 /*
2722 * We need to ensure that the page cache pages are written to
2723 * disk and invalidated to preserve the expected O_DIRECT
2724 * semantics.
2725 */
Al Viro3b93f912014-02-11 21:34:08 -05002726 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002727 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002728 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002729 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002730 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002731 invalidate_mapping_pages(mapping,
2732 pos >> PAGE_CACHE_SHIFT,
2733 endbyte >> PAGE_CACHE_SHIFT);
2734 } else {
2735 /*
2736 * We don't know how much we wrote, so just return
2737 * the number of bytes which were direct-written
2738 */
2739 }
2740 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002741 written = generic_perform_write(file, from, iocb->ki_pos);
2742 if (likely(written > 0))
2743 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745out:
2746 current->backing_dev_info = NULL;
2747 return written ? written : err;
2748}
Al Viro81742022014-04-03 03:17:43 -04002749EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
Jan Karae4dd9de2009-08-17 18:10:06 +02002751/**
Al Viro81742022014-04-03 03:17:43 -04002752 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002753 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002754 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002755 *
Al Viro81742022014-04-03 03:17:43 -04002756 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002757 * filesystems. It takes care of syncing the file in case of O_SYNC file
2758 * and acquires i_mutex as needed.
2759 */
Al Viro81742022014-04-03 03:17:43 -04002760ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761{
2762 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002763 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765
Al Viro59551022016-01-22 15:40:57 -05002766 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04002767 ret = generic_write_checks(iocb, from);
2768 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002769 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05002770 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771
Christoph Hellwig02afc272013-09-04 15:04:40 +02002772 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 ssize_t err;
2774
Al Virod311d792014-02-09 15:18:09 -05002775 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2776 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 ret = err;
2778 }
2779 return ret;
2780}
Al Viro81742022014-04-03 03:17:43 -04002781EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782
David Howellscf9a2ae2006-08-29 19:05:54 +01002783/**
2784 * try_to_release_page() - release old fs-specific metadata on a page
2785 *
2786 * @page: the page which the kernel is trying to free
2787 * @gfp_mask: memory allocation flags (and I/O mode)
2788 *
2789 * The address_space is to try to release any data against the page
2790 * (presumably at page->private). If the release was successful, return `1'.
2791 * Otherwise return zero.
2792 *
David Howells266cf652009-04-03 16:42:36 +01002793 * This may also be called if PG_fscache is set on a page, indicating that the
2794 * page is known to the local caching routines.
2795 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002796 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08002797 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002798 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002799 */
2800int try_to_release_page(struct page *page, gfp_t gfp_mask)
2801{
2802 struct address_space * const mapping = page->mapping;
2803
2804 BUG_ON(!PageLocked(page));
2805 if (PageWriteback(page))
2806 return 0;
2807
2808 if (mapping && mapping->a_ops->releasepage)
2809 return mapping->a_ops->releasepage(page, gfp_mask);
2810 return try_to_free_buffers(page);
2811}
2812
2813EXPORT_SYMBOL(try_to_release_page);