blob: 72940fb38666811b80c146bc085a1c84fc0e7ecc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
14#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070015#include <linux/uaccess.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080016#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090018#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mm.h>
20#include <linux/swap.h>
21#include <linux/mman.h>
22#include <linux/pagemap.h>
23#include <linux/file.h>
24#include <linux/uio.h>
25#include <linux/hash.h>
26#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070027#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080031#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070032#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Johannes Weiner00501b52014-08-08 14:19:20 -070033#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080034#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060035#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070036#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070039#define CREATE_TRACE_POINTS
40#include <trace/events/filemap.h>
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * FIXME: remove all knowledge of the buffer layer from the core VM
44 */
Jan Kara148f9482009-08-17 19:52:36 +020045#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <asm/mman.h>
48
49/*
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * though.
52 *
53 * Shared mappings now work. 15.8.1995 Bruno.
54 *
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 *
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
59 */
60
61/*
62 * Lock ordering:
63 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080064 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070066 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080069 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080070 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 *
72 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080073 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070074 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 *
77 * ->mmap_sem
78 * ->lock_page (access_process_vm)
79 *
Al Viroccad2362014-02-11 22:36:48 -050080 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070081 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060083 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110084 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * ->mapping->tree_lock (__sync_single_inode)
86 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080087 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * ->anon_vma.lock (vma_adjust)
89 *
90 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070091 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070094 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080098 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Greg Thelenc4843a72015-05-22 17:13:16 -0400103 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800108 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700109 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 */
111
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700112static void page_cache_tree_delete(struct address_space *mapping,
113 struct page *page, void *shadow)
114{
Johannes Weiner449dd692014-04-03 14:47:56 -0700115 struct radix_tree_node *node;
116 unsigned long index;
117 unsigned int offset;
118 unsigned int tag;
119 void **slot;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700120
Johannes Weiner449dd692014-04-03 14:47:56 -0700121 VM_BUG_ON(!PageLocked(page));
122
123 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
124
125 if (shadow) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700126 mapping->nrshadows++;
127 /*
128 * Make sure the nrshadows update is committed before
129 * the nrpages update so that final truncate racing
130 * with reclaim does not see both counters 0 at the
131 * same time and miss a shadow entry.
132 */
133 smp_wmb();
Johannes Weiner449dd692014-04-03 14:47:56 -0700134 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700135 mapping->nrpages--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700136
137 if (!node) {
138 /* Clear direct pointer tags in root node */
139 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
140 radix_tree_replace_slot(slot, shadow);
141 return;
142 }
143
144 /* Clear tree tags for the removed page */
145 index = page->index;
146 offset = index & RADIX_TREE_MAP_MASK;
147 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
148 if (test_bit(offset, node->tags[tag]))
149 radix_tree_tag_clear(&mapping->page_tree, index, tag);
150 }
151
152 /* Delete page, swap shadow entry */
153 radix_tree_replace_slot(slot, shadow);
154 workingset_node_pages_dec(node);
155 if (shadow)
156 workingset_node_shadows_inc(node);
157 else
158 if (__radix_tree_delete_node(&mapping->page_tree, node))
159 return;
160
161 /*
162 * Track node that only contains shadow entries.
163 *
164 * Avoid acquiring the list_lru lock if already tracked. The
165 * list_empty() test is safe as node->private_list is
166 * protected by mapping->tree_lock.
167 */
168 if (!workingset_node_pages(node) &&
169 list_empty(&node->private_list)) {
170 node->private_data = mapping;
171 list_lru_add(&workingset_shadow_nodes, &node->private_list);
172 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700173}
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175/*
Minchan Kime64a7822011-03-22 16:32:44 -0700176 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 * sure the page is locked and that nobody else uses it - or that usage
Greg Thelenc4843a72015-05-22 17:13:16 -0400178 * is safe. The caller must hold the mapping's tree_lock and
179 * mem_cgroup_begin_page_stat().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
Greg Thelenc4843a72015-05-22 17:13:16 -0400181void __delete_from_page_cache(struct page *page, void *shadow,
182 struct mem_cgroup *memcg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
184 struct address_space *mapping = page->mapping;
185
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700186 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600187 /*
188 * if we're uptodate, flush out into the cleancache, otherwise
189 * invalidate any existing cleancache entries. We can't leave
190 * stale data around in the cleancache once our page is gone
191 */
192 if (PageUptodate(page) && PageMappedToDisk(page))
193 cleancache_put_page(page);
194 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400195 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600196
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700197 page_cache_tree_delete(mapping, page, shadow);
198
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700200 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700201
Michal Hocko4165b9b2015-06-24 16:57:24 -0700202 /* hugetlb pages do not participate in page cache accounting. */
203 if (!PageHuge(page))
204 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700205 if (PageSwapBacked(page))
206 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700207 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800208
209 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700210 * At this point page must be either written or cleaned by truncate.
211 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800212 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700213 * This fixes dirty accounting after removing the page entirely but
214 * leaves PageDirty set: it has no effect for truncated page and
215 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800216 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700217 if (WARN_ON_ONCE(PageDirty(page)))
Tejun Heo682aa8e2015-05-28 14:50:53 -0400218 account_page_cleaned(page, mapping, memcg,
219 inode_to_wb(mapping->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
221
Minchan Kim702cfbf2011-03-22 16:32:43 -0700222/**
223 * delete_from_page_cache - delete page from page cache
224 * @page: the page which the kernel is trying to remove from page cache
225 *
226 * This must be called only on pages that have been verified to be in the page
227 * cache and locked. It will never put the page into the free list, the caller
228 * has a reference on the page.
229 */
230void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 struct address_space *mapping = page->mapping;
Greg Thelenc4843a72015-05-22 17:13:16 -0400233 struct mem_cgroup *memcg;
234 unsigned long flags;
235
Linus Torvalds6072d132010-12-01 13:35:19 -0500236 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Matt Mackallcd7619d2005-05-01 08:59:01 -0700238 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Linus Torvalds6072d132010-12-01 13:35:19 -0500240 freepage = mapping->a_ops->freepage;
Greg Thelenc4843a72015-05-22 17:13:16 -0400241
242 memcg = mem_cgroup_begin_page_stat(page);
243 spin_lock_irqsave(&mapping->tree_lock, flags);
244 __delete_from_page_cache(page, NULL, memcg);
245 spin_unlock_irqrestore(&mapping->tree_lock, flags);
246 mem_cgroup_end_page_stat(memcg);
Linus Torvalds6072d132010-12-01 13:35:19 -0500247
248 if (freepage)
249 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700250 page_cache_release(page);
251}
252EXPORT_SYMBOL(delete_from_page_cache);
253
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700254static int filemap_check_errors(struct address_space *mapping)
255{
256 int ret = 0;
257 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700258 if (test_bit(AS_ENOSPC, &mapping->flags) &&
259 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700260 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700261 if (test_bit(AS_EIO, &mapping->flags) &&
262 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700263 ret = -EIO;
264 return ret;
265}
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700268 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700269 * @mapping: address space structure to write
270 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800271 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700272 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700274 * Start writeback against all of a mapping's dirty pages that lie
275 * within the byte offsets <start, end> inclusive.
276 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700278 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 * these two operations is that if a dirty page/buffer is encountered, it must
280 * be waited upon, and not just skipped over.
281 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800282int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
283 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
285 int ret;
286 struct writeback_control wbc = {
287 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800288 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700289 .range_start = start,
290 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 };
292
293 if (!mapping_cap_writeback_dirty(mapping))
294 return 0;
295
Tejun Heob16b1de2015-06-02 08:39:48 -0600296 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600298 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 return ret;
300}
301
302static inline int __filemap_fdatawrite(struct address_space *mapping,
303 int sync_mode)
304{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700305 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
308int filemap_fdatawrite(struct address_space *mapping)
309{
310 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
311}
312EXPORT_SYMBOL(filemap_fdatawrite);
313
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400314int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800315 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
317 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
318}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400319EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Randy Dunlap485bb992006-06-23 02:03:49 -0700321/**
322 * filemap_flush - mostly a non-blocking flush
323 * @mapping: target address_space
324 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 * This is a mostly non-blocking flush. Not suitable for data-integrity
326 * purposes - I/O may not be started against all dirty pages.
327 */
328int filemap_flush(struct address_space *mapping)
329{
330 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
331}
332EXPORT_SYMBOL(filemap_flush);
333
Randy Dunlap485bb992006-06-23 02:03:49 -0700334/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200335 * filemap_fdatawait_range - wait for writeback to complete
336 * @mapping: address space structure to wait for
337 * @start_byte: offset in bytes where the range starts
338 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700339 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200340 * Walk the list of under-writeback pages of the given address space
341 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200343int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
344 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200346 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
347 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 struct pagevec pvec;
349 int nr_pages;
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700350 int ret2, ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200352 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700353 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 while ((index <= end) &&
357 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
358 PAGECACHE_TAG_WRITEBACK,
359 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
360 unsigned i;
361
362 for (i = 0; i < nr_pages; i++) {
363 struct page *page = pvec.pages[i];
364
365 /* until radix tree lookup accepts end_index */
366 if (page->index > end)
367 continue;
368
369 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800370 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 ret = -EIO;
372 }
373 pagevec_release(&pvec);
374 cond_resched();
375 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700376out:
377 ret2 = filemap_check_errors(mapping);
378 if (!ret)
379 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
381 return ret;
382}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200383EXPORT_SYMBOL(filemap_fdatawait_range);
384
385/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700386 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700388 *
389 * Walk the list of under-writeback pages of the given address space
390 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 */
392int filemap_fdatawait(struct address_space *mapping)
393{
394 loff_t i_size = i_size_read(mapping->host);
395
396 if (i_size == 0)
397 return 0;
398
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200399 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
401EXPORT_SYMBOL(filemap_fdatawait);
402
403int filemap_write_and_wait(struct address_space *mapping)
404{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800405 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800408 err = filemap_fdatawrite(mapping);
409 /*
410 * Even if the above returned error, the pages may be
411 * written partially (e.g. -ENOSPC), so we wait for it.
412 * But the -EIO is special case, it may indicate the worst
413 * thing (e.g. bug) happened, so we avoid waiting for it.
414 */
415 if (err != -EIO) {
416 int err2 = filemap_fdatawait(mapping);
417 if (!err)
418 err = err2;
419 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700420 } else {
421 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800423 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800425EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
Randy Dunlap485bb992006-06-23 02:03:49 -0700427/**
428 * filemap_write_and_wait_range - write out & wait on a file range
429 * @mapping: the address_space for the pages
430 * @lstart: offset in bytes where the range starts
431 * @lend: offset in bytes where the range ends (inclusive)
432 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800433 * Write out and wait upon file offsets lstart->lend, inclusive.
434 *
435 * Note that `lend' is inclusive (describes the last byte to be written) so
436 * that this function can be used to write to the very end-of-file (end = -1).
437 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438int filemap_write_and_wait_range(struct address_space *mapping,
439 loff_t lstart, loff_t lend)
440{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800441 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
443 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800444 err = __filemap_fdatawrite_range(mapping, lstart, lend,
445 WB_SYNC_ALL);
446 /* See comment of filemap_write_and_wait() */
447 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200448 int err2 = filemap_fdatawait_range(mapping,
449 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800450 if (!err)
451 err = err2;
452 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700453 } else {
454 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800456 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457}
Chris Masonf6995582009-04-15 13:22:37 -0400458EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Randy Dunlap485bb992006-06-23 02:03:49 -0700460/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700461 * replace_page_cache_page - replace a pagecache page with a new one
462 * @old: page to be replaced
463 * @new: page to replace with
464 * @gfp_mask: allocation mode
465 *
466 * This function replaces a page in the pagecache with a new one. On
467 * success it acquires the pagecache reference for the new page and
468 * drops it for the old page. Both the old and new pages must be
469 * locked. This function does not add the new page to the LRU, the
470 * caller must do that.
471 *
472 * The remove + add is atomic. The only way this function can fail is
473 * memory allocation failure.
474 */
475int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
476{
477 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700478
Sasha Levin309381fea2014-01-23 15:52:54 -0800479 VM_BUG_ON_PAGE(!PageLocked(old), old);
480 VM_BUG_ON_PAGE(!PageLocked(new), new);
481 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700482
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700483 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
484 if (!error) {
485 struct address_space *mapping = old->mapping;
486 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400487 struct mem_cgroup *memcg;
488 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700489
490 pgoff_t offset = old->index;
491 freepage = mapping->a_ops->freepage;
492
493 page_cache_get(new);
494 new->mapping = mapping;
495 new->index = offset;
496
Greg Thelenc4843a72015-05-22 17:13:16 -0400497 memcg = mem_cgroup_begin_page_stat(old);
498 spin_lock_irqsave(&mapping->tree_lock, flags);
499 __delete_from_page_cache(old, NULL, memcg);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700500 error = radix_tree_insert(&mapping->page_tree, offset, new);
501 BUG_ON(error);
502 mapping->nrpages++;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700503
504 /*
505 * hugetlb pages do not participate in page cache accounting.
506 */
507 if (!PageHuge(new))
508 __inc_zone_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700509 if (PageSwapBacked(new))
510 __inc_zone_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400511 spin_unlock_irqrestore(&mapping->tree_lock, flags);
512 mem_cgroup_end_page_stat(memcg);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700513 mem_cgroup_migrate(old, new, true);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700514 radix_tree_preload_end();
515 if (freepage)
516 freepage(old);
517 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700518 }
519
520 return error;
521}
522EXPORT_SYMBOL_GPL(replace_page_cache_page);
523
Johannes Weiner0cd61442014-04-03 14:47:46 -0700524static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700525 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700526{
Johannes Weiner449dd692014-04-03 14:47:56 -0700527 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700528 void **slot;
529 int error;
530
Johannes Weiner449dd692014-04-03 14:47:56 -0700531 error = __radix_tree_create(&mapping->page_tree, page->index,
532 &node, &slot);
533 if (error)
534 return error;
535 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700536 void *p;
537
538 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
539 if (!radix_tree_exceptional_entry(p))
540 return -EEXIST;
Johannes Weinera5289102014-04-03 14:47:51 -0700541 if (shadowp)
542 *shadowp = p;
Johannes Weiner449dd692014-04-03 14:47:56 -0700543 mapping->nrshadows--;
544 if (node)
545 workingset_node_shadows_dec(node);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700546 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700547 radix_tree_replace_slot(slot, page);
548 mapping->nrpages++;
549 if (node) {
550 workingset_node_pages_inc(node);
551 /*
552 * Don't track node that contains actual pages.
553 *
554 * Avoid acquiring the list_lru lock if already
555 * untracked. The list_empty() test is safe as
556 * node->private_list is protected by
557 * mapping->tree_lock.
558 */
559 if (!list_empty(&node->private_list))
560 list_lru_del(&workingset_shadow_nodes,
561 &node->private_list);
562 }
563 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700564}
565
Johannes Weinera5289102014-04-03 14:47:51 -0700566static int __add_to_page_cache_locked(struct page *page,
567 struct address_space *mapping,
568 pgoff_t offset, gfp_t gfp_mask,
569 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570{
Johannes Weiner00501b52014-08-08 14:19:20 -0700571 int huge = PageHuge(page);
572 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700573 int error;
574
Sasha Levin309381fea2014-01-23 15:52:54 -0800575 VM_BUG_ON_PAGE(!PageLocked(page), page);
576 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700577
Johannes Weiner00501b52014-08-08 14:19:20 -0700578 if (!huge) {
579 error = mem_cgroup_try_charge(page, current->mm,
580 gfp_mask, &memcg);
581 if (error)
582 return error;
583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Jan Kara5e4c0d972013-09-11 14:26:05 -0700585 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700586 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700587 if (!huge)
588 mem_cgroup_cancel_charge(page, memcg);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700589 return error;
590 }
591
592 page_cache_get(page);
593 page->mapping = mapping;
594 page->index = offset;
595
596 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700597 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700598 radix_tree_preload_end();
599 if (unlikely(error))
600 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700601
602 /* hugetlb pages do not participate in page cache accounting. */
603 if (!huge)
604 __inc_zone_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700605 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700606 if (!huge)
607 mem_cgroup_commit_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700608 trace_mm_filemap_add_to_page_cache(page);
609 return 0;
610err_insert:
611 page->mapping = NULL;
612 /* Leave page->index set: truncation relies upon it */
613 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700614 if (!huge)
615 mem_cgroup_cancel_charge(page, memcg);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700616 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 return error;
618}
Johannes Weinera5289102014-04-03 14:47:51 -0700619
620/**
621 * add_to_page_cache_locked - add a locked page to the pagecache
622 * @page: page to add
623 * @mapping: the page's address_space
624 * @offset: page index
625 * @gfp_mask: page allocation mode
626 *
627 * This function is used to add a page to the pagecache. It must be locked.
628 * This function does not add the page to the LRU. The caller must do that.
629 */
630int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
631 pgoff_t offset, gfp_t gfp_mask)
632{
633 return __add_to_page_cache_locked(page, mapping, offset,
634 gfp_mask, NULL);
635}
Nick Piggine2867812008-07-25 19:45:30 -0700636EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
638int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400639 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Johannes Weinera5289102014-04-03 14:47:51 -0700641 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700642 int ret;
643
Johannes Weinera5289102014-04-03 14:47:51 -0700644 __set_page_locked(page);
645 ret = __add_to_page_cache_locked(page, mapping, offset,
646 gfp_mask, &shadow);
647 if (unlikely(ret))
648 __clear_page_locked(page);
649 else {
650 /*
651 * The page might have been evicted from cache only
652 * recently, in which case it should be activated like
653 * any other repeatedly accessed page.
654 */
655 if (shadow && workingset_refault(shadow)) {
656 SetPageActive(page);
657 workingset_activation(page);
658 } else
659 ClearPageActive(page);
660 lru_cache_add(page);
661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return ret;
663}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300664EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Paul Jackson44110fe2006-03-24 03:16:04 -0800666#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700667struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800668{
Miao Xiec0ff7452010-05-24 14:32:08 -0700669 int n;
670 struct page *page;
671
Paul Jackson44110fe2006-03-24 03:16:04 -0800672 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700673 unsigned int cpuset_mems_cookie;
674 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700675 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700676 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700677 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700678 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700679
Miao Xiec0ff7452010-05-24 14:32:08 -0700680 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800681 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700682 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800683}
Nick Piggin2ae88142006-10-28 10:38:23 -0700684EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800685#endif
686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687/*
688 * In order to wait for pages to become available there must be
689 * waitqueues associated with pages. By using a hash table of
690 * waitqueues where the bucket discipline is to maintain all
691 * waiters on the same queue and wake all when any of the pages
692 * become available, and for the woken contexts to check to be
693 * sure the appropriate page became available, this saves space
694 * at a cost of "thundering herd" phenomena during rare hash
695 * collisions.
696 */
NeilBrowna4796e32014-09-24 11:28:32 +1000697wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
699 const struct zone *zone = page_zone(page);
700
701 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
702}
NeilBrowna4796e32014-09-24 11:28:32 +1000703EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Harvey Harrison920c7a52008-02-04 22:29:26 -0800705void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
708
709 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000710 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 TASK_UNINTERRUPTIBLE);
712}
713EXPORT_SYMBOL(wait_on_page_bit);
714
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700715int wait_on_page_bit_killable(struct page *page, int bit_nr)
716{
717 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
718
719 if (!test_bit(bit_nr, &page->flags))
720 return 0;
721
722 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000723 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700724}
725
NeilBrowncbbce822014-09-25 13:55:19 +1000726int wait_on_page_bit_killable_timeout(struct page *page,
727 int bit_nr, unsigned long timeout)
728{
729 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
730
731 wait.key.timeout = jiffies + timeout;
732 if (!test_bit(bit_nr, &page->flags))
733 return 0;
734 return __wait_on_bit(page_waitqueue(page), &wait,
735 bit_wait_io_timeout, TASK_KILLABLE);
736}
737EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100740 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700741 * @page: Page defining the wait queue of interest
742 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100743 *
744 * Add an arbitrary @waiter to the wait queue for the nominated @page.
745 */
746void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
747{
748 wait_queue_head_t *q = page_waitqueue(page);
749 unsigned long flags;
750
751 spin_lock_irqsave(&q->lock, flags);
752 __add_wait_queue(q, waiter);
753 spin_unlock_irqrestore(&q->lock, flags);
754}
755EXPORT_SYMBOL_GPL(add_page_wait_queue);
756
757/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700758 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 * @page: the page
760 *
761 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
762 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900763 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
765 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700766 * The mb is necessary to enforce ordering between the clear_bit and the read
767 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800769void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
Sasha Levin309381fea2014-01-23 15:52:54 -0800771 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700772 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100773 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 wake_up_page(page, PG_locked);
775}
776EXPORT_SYMBOL(unlock_page);
777
Randy Dunlap485bb992006-06-23 02:03:49 -0700778/**
779 * end_page_writeback - end writeback against a page
780 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 */
782void end_page_writeback(struct page *page)
783{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700784 /*
785 * TestClearPageReclaim could be used here but it is an atomic
786 * operation and overkill in this particular case. Failing to
787 * shuffle a page marked for immediate reclaim is too mild to
788 * justify taking an atomic operation penalty at the end of
789 * ever page writeback.
790 */
791 if (PageReclaim(page)) {
792 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700793 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700794 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700795
796 if (!test_clear_page_writeback(page))
797 BUG();
798
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100799 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 wake_up_page(page, PG_writeback);
801}
802EXPORT_SYMBOL(end_page_writeback);
803
Matthew Wilcox57d99842014-06-04 16:07:45 -0700804/*
805 * After completing I/O on a page, call this routine to update the page
806 * flags appropriately
807 */
808void page_endio(struct page *page, int rw, int err)
809{
810 if (rw == READ) {
811 if (!err) {
812 SetPageUptodate(page);
813 } else {
814 ClearPageUptodate(page);
815 SetPageError(page);
816 }
817 unlock_page(page);
818 } else { /* rw == WRITE */
819 if (err) {
820 SetPageError(page);
821 if (page->mapping)
822 mapping_set_error(page->mapping, err);
823 }
824 end_page_writeback(page);
825 }
826}
827EXPORT_SYMBOL_GPL(page_endio);
828
Randy Dunlap485bb992006-06-23 02:03:49 -0700829/**
830 * __lock_page - get a lock on the page, assuming we need to sleep to get it
831 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800833void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
835 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
836
NeilBrown74316202014-07-07 15:16:04 +1000837 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 TASK_UNINTERRUPTIBLE);
839}
840EXPORT_SYMBOL(__lock_page);
841
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800842int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500843{
844 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
845
846 return __wait_on_bit_lock(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000847 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500848}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300849EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500850
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700851/*
852 * Return values:
853 * 1 - page is locked; mmap_sem is still held.
854 * 0 - page is not locked.
855 * mmap_sem has been released (up_read()), unless flags had both
856 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
857 * which case mmap_sem is still held.
858 *
859 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
860 * with the page locked and the mmap_sem unperturbed.
861 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700862int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
863 unsigned int flags)
864{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700865 if (flags & FAULT_FLAG_ALLOW_RETRY) {
866 /*
867 * CAUTION! In this case, mmap_sem is not released
868 * even though return 0.
869 */
870 if (flags & FAULT_FLAG_RETRY_NOWAIT)
871 return 0;
872
873 up_read(&mm->mmap_sem);
874 if (flags & FAULT_FLAG_KILLABLE)
875 wait_on_page_locked_killable(page);
876 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700877 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700878 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700879 } else {
880 if (flags & FAULT_FLAG_KILLABLE) {
881 int ret;
882
883 ret = __lock_page_killable(page);
884 if (ret) {
885 up_read(&mm->mmap_sem);
886 return 0;
887 }
888 } else
889 __lock_page(page);
890 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700891 }
892}
893
Randy Dunlap485bb992006-06-23 02:03:49 -0700894/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700895 * page_cache_next_hole - find the next hole (not-present entry)
896 * @mapping: mapping
897 * @index: index
898 * @max_scan: maximum range to search
899 *
900 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
901 * lowest indexed hole.
902 *
903 * Returns: the index of the hole if found, otherwise returns an index
904 * outside of the set specified (in which case 'return - index >=
905 * max_scan' will be true). In rare cases of index wrap-around, 0 will
906 * be returned.
907 *
908 * page_cache_next_hole may be called under rcu_read_lock. However,
909 * like radix_tree_gang_lookup, this will not atomically search a
910 * snapshot of the tree at a single point in time. For example, if a
911 * hole is created at index 5, then subsequently a hole is created at
912 * index 10, page_cache_next_hole covering both indexes may return 10
913 * if called under rcu_read_lock.
914 */
915pgoff_t page_cache_next_hole(struct address_space *mapping,
916 pgoff_t index, unsigned long max_scan)
917{
918 unsigned long i;
919
920 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700921 struct page *page;
922
923 page = radix_tree_lookup(&mapping->page_tree, index);
924 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700925 break;
926 index++;
927 if (index == 0)
928 break;
929 }
930
931 return index;
932}
933EXPORT_SYMBOL(page_cache_next_hole);
934
935/**
936 * page_cache_prev_hole - find the prev hole (not-present entry)
937 * @mapping: mapping
938 * @index: index
939 * @max_scan: maximum range to search
940 *
941 * Search backwards in the range [max(index-max_scan+1, 0), index] for
942 * the first hole.
943 *
944 * Returns: the index of the hole if found, otherwise returns an index
945 * outside of the set specified (in which case 'index - return >=
946 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
947 * will be returned.
948 *
949 * page_cache_prev_hole may be called under rcu_read_lock. However,
950 * like radix_tree_gang_lookup, this will not atomically search a
951 * snapshot of the tree at a single point in time. For example, if a
952 * hole is created at index 10, then subsequently a hole is created at
953 * index 5, page_cache_prev_hole covering both indexes may return 5 if
954 * called under rcu_read_lock.
955 */
956pgoff_t page_cache_prev_hole(struct address_space *mapping,
957 pgoff_t index, unsigned long max_scan)
958{
959 unsigned long i;
960
961 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700962 struct page *page;
963
964 page = radix_tree_lookup(&mapping->page_tree, index);
965 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700966 break;
967 index--;
968 if (index == ULONG_MAX)
969 break;
970 }
971
972 return index;
973}
974EXPORT_SYMBOL(page_cache_prev_hole);
975
976/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700977 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -0700978 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -0700979 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -0700980 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700981 * Looks up the page cache slot at @mapping & @offset. If there is a
982 * page cache page, it is returned with an increased refcount.
983 *
Johannes Weiner139b6a62014-05-06 12:50:05 -0700984 * If the slot holds a shadow entry of a previously evicted page, or a
985 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -0700986 *
987 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700989struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990{
Nick Piggina60637c2008-07-25 19:45:31 -0700991 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 struct page *page;
993
Nick Piggina60637c2008-07-25 19:45:31 -0700994 rcu_read_lock();
995repeat:
996 page = NULL;
997 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
998 if (pagep) {
999 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001000 if (unlikely(!page))
1001 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001002 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001003 if (radix_tree_deref_retry(page))
1004 goto repeat;
1005 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001006 * A shadow entry of a recently evicted page,
1007 * or a swap entry from shmem/tmpfs. Return
1008 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001009 */
1010 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001011 }
Nick Piggina60637c2008-07-25 19:45:31 -07001012 if (!page_cache_get_speculative(page))
1013 goto repeat;
1014
1015 /*
1016 * Has the page moved?
1017 * This is part of the lockless pagecache protocol. See
1018 * include/linux/pagemap.h for details.
1019 */
1020 if (unlikely(page != *pagep)) {
1021 page_cache_release(page);
1022 goto repeat;
1023 }
1024 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001025out:
Nick Piggina60637c2008-07-25 19:45:31 -07001026 rcu_read_unlock();
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 return page;
1029}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001030EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
Randy Dunlap485bb992006-06-23 02:03:49 -07001032/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001033 * find_lock_entry - locate, pin and lock a page cache entry
1034 * @mapping: the address_space to search
1035 * @offset: the page cache index
1036 *
1037 * Looks up the page cache slot at @mapping & @offset. If there is a
1038 * page cache page, it is returned locked and with an increased
1039 * refcount.
1040 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001041 * If the slot holds a shadow entry of a previously evicted page, or a
1042 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001043 *
1044 * Otherwise, %NULL is returned.
1045 *
1046 * find_lock_entry() may sleep.
1047 */
1048struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049{
1050 struct page *page;
1051
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001053 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001054 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001055 lock_page(page);
1056 /* Has the page been truncated? */
1057 if (unlikely(page->mapping != mapping)) {
1058 unlock_page(page);
1059 page_cache_release(page);
1060 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001062 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 return page;
1065}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001066EXPORT_SYMBOL(find_lock_entry);
1067
1068/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001069 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001070 * @mapping: the address_space to search
1071 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001072 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001073 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001074 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001075 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001076 *
Randy Dunlap75325182014-07-30 16:08:37 -07001077 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001078 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001079 * FGP_ACCESSED: the page will be marked accessed
1080 * FGP_LOCK: Page is return locked
1081 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001082 * @gfp_mask and added to the page cache and the VM's LRU
1083 * list. The page is returned locked and with an increased
1084 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001085 *
1086 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1087 * if the GFP flags specified for FGP_CREAT are atomic.
1088 *
1089 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001090 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001091struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001092 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093{
Nick Piggineb2be182007-10-16 01:24:57 -07001094 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001097 page = find_get_entry(mapping, offset);
1098 if (radix_tree_exceptional_entry(page))
1099 page = NULL;
1100 if (!page)
1101 goto no_page;
1102
1103 if (fgp_flags & FGP_LOCK) {
1104 if (fgp_flags & FGP_NOWAIT) {
1105 if (!trylock_page(page)) {
1106 page_cache_release(page);
1107 return NULL;
1108 }
1109 } else {
1110 lock_page(page);
1111 }
1112
1113 /* Has the page been truncated? */
1114 if (unlikely(page->mapping != mapping)) {
1115 unlock_page(page);
1116 page_cache_release(page);
1117 goto repeat;
1118 }
1119 VM_BUG_ON_PAGE(page->index != offset, page);
1120 }
1121
1122 if (page && (fgp_flags & FGP_ACCESSED))
1123 mark_page_accessed(page);
1124
1125no_page:
1126 if (!page && (fgp_flags & FGP_CREAT)) {
1127 int err;
1128 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001129 gfp_mask |= __GFP_WRITE;
1130 if (fgp_flags & FGP_NOFS)
1131 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001132
Michal Hocko45f87de2014-12-29 20:30:35 +01001133 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001134 if (!page)
1135 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001136
1137 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1138 fgp_flags |= FGP_LOCK;
1139
Hugh Dickinseb39d612014-08-06 16:06:43 -07001140 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001141 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001142 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001143
Michal Hocko45f87de2014-12-29 20:30:35 +01001144 err = add_to_page_cache_lru(page, mapping, offset,
1145 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001146 if (unlikely(err)) {
1147 page_cache_release(page);
1148 page = NULL;
1149 if (err == -EEXIST)
1150 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001153
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 return page;
1155}
Mel Gorman2457aec2014-06-04 16:10:31 -07001156EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
1158/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001159 * find_get_entries - gang pagecache lookup
1160 * @mapping: The address_space to search
1161 * @start: The starting page cache index
1162 * @nr_entries: The maximum number of entries
1163 * @entries: Where the resulting entries are placed
1164 * @indices: The cache indices corresponding to the entries in @entries
1165 *
1166 * find_get_entries() will search for and return a group of up to
1167 * @nr_entries entries in the mapping. The entries are placed at
1168 * @entries. find_get_entries() takes a reference against any actual
1169 * pages it returns.
1170 *
1171 * The search returns a group of mapping-contiguous page cache entries
1172 * with ascending indexes. There may be holes in the indices due to
1173 * not-present pages.
1174 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001175 * Any shadow entries of evicted pages, or swap entries from
1176 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001177 *
1178 * find_get_entries() returns the number of pages and shadow entries
1179 * which were found.
1180 */
1181unsigned find_get_entries(struct address_space *mapping,
1182 pgoff_t start, unsigned int nr_entries,
1183 struct page **entries, pgoff_t *indices)
1184{
1185 void **slot;
1186 unsigned int ret = 0;
1187 struct radix_tree_iter iter;
1188
1189 if (!nr_entries)
1190 return 0;
1191
1192 rcu_read_lock();
1193restart:
1194 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1195 struct page *page;
1196repeat:
1197 page = radix_tree_deref_slot(slot);
1198 if (unlikely(!page))
1199 continue;
1200 if (radix_tree_exception(page)) {
1201 if (radix_tree_deref_retry(page))
1202 goto restart;
1203 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001204 * A shadow entry of a recently evicted page,
1205 * or a swap entry from shmem/tmpfs. Return
1206 * it without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001207 */
1208 goto export;
1209 }
1210 if (!page_cache_get_speculative(page))
1211 goto repeat;
1212
1213 /* Has the page moved? */
1214 if (unlikely(page != *slot)) {
1215 page_cache_release(page);
1216 goto repeat;
1217 }
1218export:
1219 indices[ret] = iter.index;
1220 entries[ret] = page;
1221 if (++ret == nr_entries)
1222 break;
1223 }
1224 rcu_read_unlock();
1225 return ret;
1226}
1227
1228/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 * find_get_pages - gang pagecache lookup
1230 * @mapping: The address_space to search
1231 * @start: The starting page index
1232 * @nr_pages: The maximum number of pages
1233 * @pages: Where the resulting pages are placed
1234 *
1235 * find_get_pages() will search for and return a group of up to
1236 * @nr_pages pages in the mapping. The pages are placed at @pages.
1237 * find_get_pages() takes a reference against the returned pages.
1238 *
1239 * The search returns a group of mapping-contiguous pages with ascending
1240 * indexes. There may be holes in the indices due to not-present pages.
1241 *
1242 * find_get_pages() returns the number of pages which were found.
1243 */
1244unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1245 unsigned int nr_pages, struct page **pages)
1246{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001247 struct radix_tree_iter iter;
1248 void **slot;
1249 unsigned ret = 0;
1250
1251 if (unlikely(!nr_pages))
1252 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Nick Piggina60637c2008-07-25 19:45:31 -07001254 rcu_read_lock();
1255restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001256 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001257 struct page *page;
1258repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001259 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001260 if (unlikely(!page))
1261 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001262
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001263 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001264 if (radix_tree_deref_retry(page)) {
1265 /*
1266 * Transient condition which can only trigger
1267 * when entry at index 0 moves out of or back
1268 * to root: none yet gotten, safe to restart.
1269 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001270 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001271 goto restart;
1272 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001273 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001274 * A shadow entry of a recently evicted page,
1275 * or a swap entry from shmem/tmpfs. Skip
1276 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001277 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001278 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001279 }
Nick Piggina60637c2008-07-25 19:45:31 -07001280
1281 if (!page_cache_get_speculative(page))
1282 goto repeat;
1283
1284 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001285 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001286 page_cache_release(page);
1287 goto repeat;
1288 }
1289
1290 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001291 if (++ret == nr_pages)
1292 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001293 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001294
Nick Piggina60637c2008-07-25 19:45:31 -07001295 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 return ret;
1297}
1298
Jens Axboeebf43502006-04-27 08:46:01 +02001299/**
1300 * find_get_pages_contig - gang contiguous pagecache lookup
1301 * @mapping: The address_space to search
1302 * @index: The starting page index
1303 * @nr_pages: The maximum number of pages
1304 * @pages: Where the resulting pages are placed
1305 *
1306 * find_get_pages_contig() works exactly like find_get_pages(), except
1307 * that the returned number of pages are guaranteed to be contiguous.
1308 *
1309 * find_get_pages_contig() returns the number of pages which were found.
1310 */
1311unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1312 unsigned int nr_pages, struct page **pages)
1313{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001314 struct radix_tree_iter iter;
1315 void **slot;
1316 unsigned int ret = 0;
1317
1318 if (unlikely(!nr_pages))
1319 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001320
Nick Piggina60637c2008-07-25 19:45:31 -07001321 rcu_read_lock();
1322restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001323 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001324 struct page *page;
1325repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001326 page = radix_tree_deref_slot(slot);
1327 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001328 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001329 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001330
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001331 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001332 if (radix_tree_deref_retry(page)) {
1333 /*
1334 * Transient condition which can only trigger
1335 * when entry at index 0 moves out of or back
1336 * to root: none yet gotten, safe to restart.
1337 */
1338 goto restart;
1339 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001340 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001341 * A shadow entry of a recently evicted page,
1342 * or a swap entry from shmem/tmpfs. Stop
1343 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001344 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001345 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001346 }
Nick Piggina60637c2008-07-25 19:45:31 -07001347
Nick Piggina60637c2008-07-25 19:45:31 -07001348 if (!page_cache_get_speculative(page))
1349 goto repeat;
1350
1351 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001352 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001353 page_cache_release(page);
1354 goto repeat;
1355 }
1356
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001357 /*
1358 * must check mapping and index after taking the ref.
1359 * otherwise we can get both false positives and false
1360 * negatives, which is just confusing to the caller.
1361 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001362 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001363 page_cache_release(page);
1364 break;
1365 }
1366
Nick Piggina60637c2008-07-25 19:45:31 -07001367 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001368 if (++ret == nr_pages)
1369 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001370 }
Nick Piggina60637c2008-07-25 19:45:31 -07001371 rcu_read_unlock();
1372 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001373}
David Howellsef71c152007-05-09 02:33:44 -07001374EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001375
Randy Dunlap485bb992006-06-23 02:03:49 -07001376/**
1377 * find_get_pages_tag - find and return pages that match @tag
1378 * @mapping: the address_space to search
1379 * @index: the starting page index
1380 * @tag: the tag index
1381 * @nr_pages: the maximum number of pages
1382 * @pages: where the resulting pages are placed
1383 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001385 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 */
1387unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1388 int tag, unsigned int nr_pages, struct page **pages)
1389{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001390 struct radix_tree_iter iter;
1391 void **slot;
1392 unsigned ret = 0;
1393
1394 if (unlikely(!nr_pages))
1395 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Nick Piggina60637c2008-07-25 19:45:31 -07001397 rcu_read_lock();
1398restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001399 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1400 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001401 struct page *page;
1402repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001403 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001404 if (unlikely(!page))
1405 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001406
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001407 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001408 if (radix_tree_deref_retry(page)) {
1409 /*
1410 * Transient condition which can only trigger
1411 * when entry at index 0 moves out of or back
1412 * to root: none yet gotten, safe to restart.
1413 */
1414 goto restart;
1415 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001416 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001417 * A shadow entry of a recently evicted page.
1418 *
1419 * Those entries should never be tagged, but
1420 * this tree walk is lockless and the tags are
1421 * looked up in bulk, one radix tree node at a
1422 * time, so there is a sizable window for page
1423 * reclaim to evict a page we saw tagged.
1424 *
1425 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001426 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001427 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001428 }
Nick Piggina60637c2008-07-25 19:45:31 -07001429
1430 if (!page_cache_get_speculative(page))
1431 goto repeat;
1432
1433 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001434 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001435 page_cache_release(page);
1436 goto repeat;
1437 }
1438
1439 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001440 if (++ret == nr_pages)
1441 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001442 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001443
Nick Piggina60637c2008-07-25 19:45:31 -07001444 rcu_read_unlock();
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 if (ret)
1447 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return ret;
1450}
David Howellsef71c152007-05-09 02:33:44 -07001451EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001453/*
1454 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1455 * a _large_ part of the i/o request. Imagine the worst scenario:
1456 *
1457 * ---R__________________________________________B__________
1458 * ^ reading here ^ bad block(assume 4k)
1459 *
1460 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1461 * => failing the whole request => read(R) => read(R+1) =>
1462 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1463 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1464 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1465 *
1466 * It is going insane. Fix it by quickly scaling down the readahead size.
1467 */
1468static void shrink_readahead_size_eio(struct file *filp,
1469 struct file_ra_state *ra)
1470{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001471 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001472}
1473
Randy Dunlap485bb992006-06-23 02:03:49 -07001474/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001475 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001476 * @filp: the file to read
1477 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001478 * @iter: data destination
1479 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001480 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001482 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 *
1484 * This is really ugly. But the goto's actually try to clarify some
1485 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 */
Al Viro6e58e792014-02-03 17:07:03 -05001487static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1488 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001490 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001492 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001493 pgoff_t index;
1494 pgoff_t last_index;
1495 pgoff_t prev_index;
1496 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001497 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001498 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001501 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1502 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Al Viro6e58e792014-02-03 17:07:03 -05001503 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 offset = *ppos & ~PAGE_CACHE_MASK;
1505
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 for (;;) {
1507 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001508 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001509 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 unsigned long nr, ret;
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513find_page:
1514 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001515 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001516 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001517 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001518 index, last_index - index);
1519 page = find_get_page(mapping, index);
1520 if (unlikely(page == NULL))
1521 goto no_cached_page;
1522 }
1523 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001524 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001525 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001526 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001528 if (!PageUptodate(page)) {
1529 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1530 !mapping->a_ops->is_partially_uptodate)
1531 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001532 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001533 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001534 /* Did it get truncated before we got the lock? */
1535 if (!page->mapping)
1536 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001537 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001538 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001539 goto page_not_up_to_date_locked;
1540 unlock_page(page);
1541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001543 /*
1544 * i_size must be checked after we know the page is Uptodate.
1545 *
1546 * Checking i_size after the check allows us to calculate
1547 * the correct value for "nr", which means the zero-filled
1548 * part of the page is not copied back to userspace (unless
1549 * another truncate extends the file - this is desired though).
1550 */
1551
1552 isize = i_size_read(inode);
1553 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1554 if (unlikely(!isize || index > end_index)) {
1555 page_cache_release(page);
1556 goto out;
1557 }
1558
1559 /* nr is the maximum number of bytes to copy from this page */
1560 nr = PAGE_CACHE_SIZE;
1561 if (index == end_index) {
1562 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1563 if (nr <= offset) {
1564 page_cache_release(page);
1565 goto out;
1566 }
1567 }
1568 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
1570 /* If users can be writing to this page using arbitrary
1571 * virtual addresses, take care about potential aliasing
1572 * before reading the page on the kernel side.
1573 */
1574 if (mapping_writably_mapped(mapping))
1575 flush_dcache_page(page);
1576
1577 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001578 * When a sequential read accesses a page several times,
1579 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 */
Jan Karaec0f1632007-05-06 14:49:25 -07001581 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 mark_page_accessed(page);
1583 prev_index = index;
1584
1585 /*
1586 * Ok, we have the page, and it's up-to-date, so
1587 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 */
Al Viro6e58e792014-02-03 17:07:03 -05001589
1590 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 offset += ret;
1592 index += offset >> PAGE_CACHE_SHIFT;
1593 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001594 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
1596 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001597 written += ret;
1598 if (!iov_iter_count(iter))
1599 goto out;
1600 if (ret < nr) {
1601 error = -EFAULT;
1602 goto out;
1603 }
1604 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
1606page_not_up_to_date:
1607 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001608 error = lock_page_killable(page);
1609 if (unlikely(error))
1610 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001612page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001613 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 if (!page->mapping) {
1615 unlock_page(page);
1616 page_cache_release(page);
1617 continue;
1618 }
1619
1620 /* Did somebody else fill it already? */
1621 if (PageUptodate(page)) {
1622 unlock_page(page);
1623 goto page_ok;
1624 }
1625
1626readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001627 /*
1628 * A previous I/O error may have been due to temporary
1629 * failures, eg. multipath errors.
1630 * PG_error will be set again if readpage fails.
1631 */
1632 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 /* Start the actual read. The read will unlock the page. */
1634 error = mapping->a_ops->readpage(filp, page);
1635
Zach Brown994fc28c2005-12-15 14:28:17 -08001636 if (unlikely(error)) {
1637 if (error == AOP_TRUNCATED_PAGE) {
1638 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001639 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001640 goto find_page;
1641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
1645 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001646 error = lock_page_killable(page);
1647 if (unlikely(error))
1648 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 if (!PageUptodate(page)) {
1650 if (page->mapping == NULL) {
1651 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001652 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 */
1654 unlock_page(page);
1655 page_cache_release(page);
1656 goto find_page;
1657 }
1658 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001659 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001660 error = -EIO;
1661 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 }
1663 unlock_page(page);
1664 }
1665
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 goto page_ok;
1667
1668readpage_error:
1669 /* UHHUH! A synchronous read error occurred. Report it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 page_cache_release(page);
1671 goto out;
1672
1673no_cached_page:
1674 /*
1675 * Ok, it wasn't cached, so we need to create a new
1676 * page..
1677 */
Nick Piggineb2be182007-10-16 01:24:57 -07001678 page = page_cache_alloc_cold(mapping);
1679 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001680 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001681 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001683 error = add_to_page_cache_lru(page, mapping, index,
1684 GFP_KERNEL & mapping_gfp_mask(mapping));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001686 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001687 if (error == -EEXIST) {
1688 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 goto out;
1692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 goto readpage;
1694 }
1695
1696out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001697 ra->prev_pos = prev_index;
1698 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1699 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001701 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001702 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001703 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704}
1705
Randy Dunlap485bb992006-06-23 02:03:49 -07001706/**
Al Viro6abd2322014-04-04 14:20:57 -04001707 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001708 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001709 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001710 *
Al Viro6abd2322014-04-04 14:20:57 -04001711 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 * that can use the page cache directly.
1713 */
1714ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001715generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716{
Al Viroed978a82014-03-05 22:53:04 -05001717 struct file *file = iocb->ki_filp;
1718 ssize_t retval = 0;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001719 loff_t *ppos = &iocb->ki_pos;
Al Viroed978a82014-03-05 22:53:04 -05001720 loff_t pos = *ppos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
Al Viro2ba48ce2015-04-09 13:52:01 -04001722 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001723 struct address_space *mapping = file->f_mapping;
1724 struct inode *inode = mapping->host;
1725 size_t count = iov_iter_count(iter);
Badari Pulavarty543ade12006-09-30 23:28:48 -07001726 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 if (!count)
1729 goto out; /* skip atime */
1730 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001731 retval = filemap_write_and_wait_range(mapping, pos,
Al Viroed978a82014-03-05 22:53:04 -05001732 pos + count - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001733 if (!retval) {
Al Viroed978a82014-03-05 22:53:04 -05001734 struct iov_iter data = *iter;
Omar Sandoval22c61862015-03-16 04:33:53 -07001735 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001736 }
Al Viroed978a82014-03-05 22:53:04 -05001737
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001738 if (retval > 0) {
1739 *ppos = pos + retval;
Al Viroed978a82014-03-05 22:53:04 -05001740 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001741 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001742
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001743 /*
1744 * Btrfs can have a short DIO read if we encounter
1745 * compressed extents, so if there was an error, or if
1746 * we've already read everything we wanted to, or if
1747 * there was a short read because we hit EOF, go ahead
1748 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001749 * the rest of the read. Buffered reads will not work for
1750 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001751 */
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001752 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1753 IS_DAX(inode)) {
Al Viroed978a82014-03-05 22:53:04 -05001754 file_accessed(file);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001755 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001756 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 }
1758
Al Viroed978a82014-03-05 22:53:04 -05001759 retval = do_generic_file_read(file, ppos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760out:
1761 return retval;
1762}
Al Viroed978a82014-03-05 22:53:04 -05001763EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001766/**
1767 * page_cache_read - adds requested page to the page cache if not already there
1768 * @file: file to read
1769 * @offset: page index
1770 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 * This adds the requested page to the page cache if it isn't already there,
1772 * and schedules an I/O to read in its contents from disk.
1773 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001774static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775{
1776 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001777 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001778 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779
Zach Brown994fc28c2005-12-15 14:28:17 -08001780 do {
1781 page = page_cache_alloc_cold(mapping);
1782 if (!page)
1783 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
Michal Hocko6afdb852015-06-24 16:58:06 -07001785 ret = add_to_page_cache_lru(page, mapping, offset,
1786 GFP_KERNEL & mapping_gfp_mask(mapping));
Zach Brown994fc28c2005-12-15 14:28:17 -08001787 if (ret == 0)
1788 ret = mapping->a_ops->readpage(file, page);
1789 else if (ret == -EEXIST)
1790 ret = 0; /* losing race to add is OK */
1791
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
Zach Brown994fc28c2005-12-15 14:28:17 -08001794 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07001795
Zach Brown994fc28c2005-12-15 14:28:17 -08001796 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797}
1798
1799#define MMAP_LOTSAMISS (100)
1800
Linus Torvaldsef00e082009-06-16 15:31:25 -07001801/*
1802 * Synchronous readahead happens when we don't even find
1803 * a page in the page cache at all.
1804 */
1805static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1806 struct file_ra_state *ra,
1807 struct file *file,
1808 pgoff_t offset)
1809{
1810 unsigned long ra_pages;
1811 struct address_space *mapping = file->f_mapping;
1812
1813 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001814 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001815 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001816 if (!ra->ra_pages)
1817 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001818
Joe Perches64363aa2013-07-08 16:00:18 -07001819 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001820 page_cache_sync_readahead(mapping, ra, file, offset,
1821 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001822 return;
1823 }
1824
Andi Kleen207d04b2011-05-24 17:12:29 -07001825 /* Avoid banging the cache line if not needed */
1826 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001827 ra->mmap_miss++;
1828
1829 /*
1830 * Do we miss much more than hit in this file? If so,
1831 * stop bothering with read-ahead. It will only hurt.
1832 */
1833 if (ra->mmap_miss > MMAP_LOTSAMISS)
1834 return;
1835
Wu Fengguangd30a1102009-06-16 15:31:30 -07001836 /*
1837 * mmap read-around
1838 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001839 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001840 ra->start = max_t(long, 0, offset - ra_pages / 2);
1841 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001842 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001843 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001844}
1845
1846/*
1847 * Asynchronous readahead happens when we find the page and PG_readahead,
1848 * so we want to possibly extend the readahead further..
1849 */
1850static void do_async_mmap_readahead(struct vm_area_struct *vma,
1851 struct file_ra_state *ra,
1852 struct file *file,
1853 struct page *page,
1854 pgoff_t offset)
1855{
1856 struct address_space *mapping = file->f_mapping;
1857
1858 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001859 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001860 return;
1861 if (ra->mmap_miss > 0)
1862 ra->mmap_miss--;
1863 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001864 page_cache_async_readahead(mapping, ra, file,
1865 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001866}
1867
Randy Dunlap485bb992006-06-23 02:03:49 -07001868/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001869 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001870 * @vma: vma in which the fault was taken
1871 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001872 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001873 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 * mapped memory region to read in file data during a page fault.
1875 *
1876 * The goto's are kind of ugly, but this streamlines the normal case of having
1877 * it in the page cache, and handles the special cases reasonably without
1878 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001879 *
1880 * vma->vm_mm->mmap_sem must be held on entry.
1881 *
1882 * If our return value has VM_FAULT_RETRY set, it's because
1883 * lock_page_or_retry() returned 0.
1884 * The mmap_sem has usually been released in this case.
1885 * See __lock_page_or_retry() for the exception.
1886 *
1887 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1888 * has not been released.
1889 *
1890 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001892int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
1894 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001895 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 struct address_space *mapping = file->f_mapping;
1897 struct file_ra_state *ra = &file->f_ra;
1898 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001899 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001901 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001902 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001904 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1905 if (offset >= size >> PAGE_CACHE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001906 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001909 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001911 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001912 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001914 * We found the page, so try async readahead before
1915 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001917 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001918 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001919 /* No page in the page cache at all */
1920 do_sync_mmap_readahead(vma, ra, file, offset);
1921 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001922 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001923 ret = VM_FAULT_MAJOR;
1924retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001925 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 if (!page)
1927 goto no_cached_page;
1928 }
1929
Michel Lespinassed88c0922010-11-02 13:05:18 -07001930 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1931 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001932 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001933 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001934
1935 /* Did it get truncated? */
1936 if (unlikely(page->mapping != mapping)) {
1937 unlock_page(page);
1938 put_page(page);
1939 goto retry_find;
1940 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001941 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07001942
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001944 * We have a locked page in the page cache, now we need to check
1945 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 */
Nick Piggind00806b2007-07-19 01:46:57 -07001947 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 goto page_not_uptodate;
1949
Linus Torvaldsef00e082009-06-16 15:31:25 -07001950 /*
1951 * Found the page and have a reference on it.
1952 * We must recheck i_size under page lock.
1953 */
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001954 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1955 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001956 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001957 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001958 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001959 }
1960
Nick Piggind0217ac2007-07-19 01:47:03 -07001961 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001962 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964no_cached_page:
1965 /*
1966 * We're only likely to ever get here if MADV_RANDOM is in
1967 * effect.
1968 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001969 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
1971 /*
1972 * The page we want has now been added to the page cache.
1973 * In the unlikely event that someone removed it in the
1974 * meantime, we'll just come back here and read it again.
1975 */
1976 if (error >= 0)
1977 goto retry_find;
1978
1979 /*
1980 * An error return from page_cache_read can result if the
1981 * system is low on memory, or a problem occurs while trying
1982 * to schedule I/O.
1983 */
1984 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001985 return VM_FAULT_OOM;
1986 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
1988page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 /*
1990 * Umm, take care of errors if the page isn't up-to-date.
1991 * Try to re-read it _once_. We do this synchronously,
1992 * because there really aren't any performance issues here
1993 * and we need to check for errors.
1994 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001996 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001997 if (!error) {
1998 wait_on_page_locked(page);
1999 if (!PageUptodate(page))
2000 error = -EIO;
2001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002003
2004 if (!error || error == AOP_TRUNCATED_PAGE)
2005 goto retry_find;
2006
2007 /* Things didn't work out. Return zero to tell the mm layer so. */
2008 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002009 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002010}
2011EXPORT_SYMBOL(filemap_fault);
2012
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002013void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2014{
2015 struct radix_tree_iter iter;
2016 void **slot;
2017 struct file *file = vma->vm_file;
2018 struct address_space *mapping = file->f_mapping;
2019 loff_t size;
2020 struct page *page;
2021 unsigned long address = (unsigned long) vmf->virtual_address;
2022 unsigned long addr;
2023 pte_t *pte;
2024
2025 rcu_read_lock();
2026 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2027 if (iter.index > vmf->max_pgoff)
2028 break;
2029repeat:
2030 page = radix_tree_deref_slot(slot);
2031 if (unlikely(!page))
2032 goto next;
2033 if (radix_tree_exception(page)) {
2034 if (radix_tree_deref_retry(page))
2035 break;
2036 else
2037 goto next;
2038 }
2039
2040 if (!page_cache_get_speculative(page))
2041 goto repeat;
2042
2043 /* Has the page moved? */
2044 if (unlikely(page != *slot)) {
2045 page_cache_release(page);
2046 goto repeat;
2047 }
2048
2049 if (!PageUptodate(page) ||
2050 PageReadahead(page) ||
2051 PageHWPoison(page))
2052 goto skip;
2053 if (!trylock_page(page))
2054 goto skip;
2055
2056 if (page->mapping != mapping || !PageUptodate(page))
2057 goto unlock;
2058
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002059 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2060 if (page->index >= size >> PAGE_CACHE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002061 goto unlock;
2062
2063 pte = vmf->pte + page->index - vmf->pgoff;
2064 if (!pte_none(*pte))
2065 goto unlock;
2066
2067 if (file->f_ra.mmap_miss > 0)
2068 file->f_ra.mmap_miss--;
2069 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2070 do_set_pte(vma, addr, page, pte, false, false);
2071 unlock_page(page);
2072 goto next;
2073unlock:
2074 unlock_page(page);
2075skip:
2076 page_cache_release(page);
2077next:
2078 if (iter.index == vmf->max_pgoff)
2079 break;
2080 }
2081 rcu_read_unlock();
2082}
2083EXPORT_SYMBOL(filemap_map_pages);
2084
Jan Kara4fcf1c62012-06-12 16:20:29 +02002085int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2086{
2087 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002088 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002089 int ret = VM_FAULT_LOCKED;
2090
Jan Kara14da9202012-06-12 16:20:37 +02002091 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002092 file_update_time(vma->vm_file);
2093 lock_page(page);
2094 if (page->mapping != inode->i_mapping) {
2095 unlock_page(page);
2096 ret = VM_FAULT_NOPAGE;
2097 goto out;
2098 }
Jan Kara14da9202012-06-12 16:20:37 +02002099 /*
2100 * We mark the page dirty already here so that when freeze is in
2101 * progress, we are guaranteed that writeback during freezing will
2102 * see the dirty page and writeprotect it again.
2103 */
2104 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002105 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002106out:
Jan Kara14da9202012-06-12 16:20:37 +02002107 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002108 return ret;
2109}
2110EXPORT_SYMBOL(filemap_page_mkwrite);
2111
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002112const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002113 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002114 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002115 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116};
2117
2118/* This is used for a general mmap of a disk file */
2119
2120int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2121{
2122 struct address_space *mapping = file->f_mapping;
2123
2124 if (!mapping->a_ops->readpage)
2125 return -ENOEXEC;
2126 file_accessed(file);
2127 vma->vm_ops = &generic_file_vm_ops;
2128 return 0;
2129}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
2131/*
2132 * This is for filesystems which do not implement ->writepage.
2133 */
2134int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2135{
2136 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2137 return -EINVAL;
2138 return generic_file_mmap(file, vma);
2139}
2140#else
2141int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2142{
2143 return -ENOSYS;
2144}
2145int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2146{
2147 return -ENOSYS;
2148}
2149#endif /* CONFIG_MMU */
2150
2151EXPORT_SYMBOL(generic_file_mmap);
2152EXPORT_SYMBOL(generic_file_readonly_mmap);
2153
Sasha Levin67f9fd92014-04-03 14:48:18 -07002154static struct page *wait_on_page_read(struct page *page)
2155{
2156 if (!IS_ERR(page)) {
2157 wait_on_page_locked(page);
2158 if (!PageUptodate(page)) {
2159 page_cache_release(page);
2160 page = ERR_PTR(-EIO);
2161 }
2162 }
2163 return page;
2164}
2165
Nick Piggin6fe69002007-05-06 14:49:04 -07002166static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002167 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002168 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002169 void *data,
2170 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171{
Nick Piggineb2be182007-10-16 01:24:57 -07002172 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 int err;
2174repeat:
2175 page = find_get_page(mapping, index);
2176 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002177 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002178 if (!page)
2179 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002180 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002181 if (unlikely(err)) {
2182 page_cache_release(page);
2183 if (err == -EEXIST)
2184 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 return ERR_PTR(err);
2187 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 err = filler(data, page);
2189 if (err < 0) {
2190 page_cache_release(page);
2191 page = ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002192 } else {
2193 page = wait_on_page_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 }
2195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 return page;
2197}
2198
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002199static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002200 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002201 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002202 void *data,
2203 gfp_t gfp)
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
2206 struct page *page;
2207 int err;
2208
2209retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002210 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01002212 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 if (PageUptodate(page))
2214 goto out;
2215
2216 lock_page(page);
2217 if (!page->mapping) {
2218 unlock_page(page);
2219 page_cache_release(page);
2220 goto retry;
2221 }
2222 if (PageUptodate(page)) {
2223 unlock_page(page);
2224 goto out;
2225 }
2226 err = filler(data, page);
2227 if (err < 0) {
2228 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01002229 return ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002230 } else {
2231 page = wait_on_page_read(page);
2232 if (IS_ERR(page))
2233 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 }
David Howellsc855ff32007-05-09 13:42:20 +01002235out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002236 mark_page_accessed(page);
2237 return page;
2238}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002239
2240/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002241 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002242 * @mapping: the page's address_space
2243 * @index: the page index
2244 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002245 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002246 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002247 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002248 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002249 *
2250 * If the page does not get brought uptodate, return -EIO.
2251 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002252struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002253 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002254 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002255 void *data)
2256{
2257 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2258}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002259EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002260
2261/**
2262 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2263 * @mapping: the page's address_space
2264 * @index: the page index
2265 * @gfp: the page allocator flags to use if allocating
2266 *
2267 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002268 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002269 *
2270 * If the page does not get brought uptodate, return -EIO.
2271 */
2272struct page *read_cache_page_gfp(struct address_space *mapping,
2273 pgoff_t index,
2274 gfp_t gfp)
2275{
2276 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2277
Sasha Levin67f9fd92014-04-03 14:48:18 -07002278 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002279}
2280EXPORT_SYMBOL(read_cache_page_gfp);
2281
Nick Piggin2f718ff2007-10-16 01:24:59 -07002282/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 * Performs necessary checks before doing a write
2284 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002285 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 * Returns appropriate error code that caller should return or
2287 * zero in case that write should be allowed.
2288 */
Al Viro3309dd02015-04-09 12:55:47 -04002289inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290{
Al Viro3309dd02015-04-09 12:55:47 -04002291 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002293 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002294 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Al Viro3309dd02015-04-09 12:55:47 -04002296 if (!iov_iter_count(from))
2297 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
Al Viro0fa6b002015-04-04 04:05:48 -04002299 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002300 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002301 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
Al Viro3309dd02015-04-09 12:55:47 -04002303 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
Al Viro0fa6b002015-04-04 04:05:48 -04002305 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002306 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002307 send_sig(SIGXFSZ, current, 0);
2308 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 }
Al Viro3309dd02015-04-09 12:55:47 -04002310 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 }
2312
2313 /*
2314 * LFS rule
2315 */
Al Viro3309dd02015-04-09 12:55:47 -04002316 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002318 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002320 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 }
2322
2323 /*
2324 * Are we about to exceed the fs block limit ?
2325 *
2326 * If we have written data it becomes a short write. If we have
2327 * exceeded without writing data we send a signal and return EFBIG.
2328 * Linus frestrict idea will clean these up nicely..
2329 */
Al Viro3309dd02015-04-09 12:55:47 -04002330 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2331 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
Al Viro3309dd02015-04-09 12:55:47 -04002333 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2334 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335}
2336EXPORT_SYMBOL(generic_write_checks);
2337
Nick Pigginafddba42007-10-16 01:25:01 -07002338int pagecache_write_begin(struct file *file, struct address_space *mapping,
2339 loff_t pos, unsigned len, unsigned flags,
2340 struct page **pagep, void **fsdata)
2341{
2342 const struct address_space_operations *aops = mapping->a_ops;
2343
Nick Piggin4e02ed42008-10-29 14:00:55 -07002344 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002345 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002346}
2347EXPORT_SYMBOL(pagecache_write_begin);
2348
2349int pagecache_write_end(struct file *file, struct address_space *mapping,
2350 loff_t pos, unsigned len, unsigned copied,
2351 struct page *page, void *fsdata)
2352{
2353 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002354
Nick Piggin4e02ed42008-10-29 14:00:55 -07002355 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002356}
2357EXPORT_SYMBOL(pagecache_write_end);
2358
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359ssize_t
Al Viro0c949332014-03-22 06:51:37 -04002360generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361{
2362 struct file *file = iocb->ki_filp;
2363 struct address_space *mapping = file->f_mapping;
2364 struct inode *inode = mapping->host;
2365 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002366 size_t write_len;
2367 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002368 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Al Viro0c949332014-03-22 06:51:37 -04002370 write_len = iov_iter_count(from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002371 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002372
Nick Piggin48b47c52009-01-06 14:40:22 -08002373 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002374 if (written)
2375 goto out;
2376
2377 /*
2378 * After a write we want buffered reads to be sure to go to disk to get
2379 * the new data. We invalidate clean cached page from the region we're
2380 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002381 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002382 */
2383 if (mapping->nrpages) {
2384 written = invalidate_inode_pages2_range(mapping,
2385 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002386 /*
2387 * If a page can not be invalidated, return 0 to fall back
2388 * to buffered write.
2389 */
2390 if (written) {
2391 if (written == -EBUSY)
2392 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002393 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002394 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002395 }
2396
Al Viro26978b82014-03-10 14:08:45 -04002397 data = *from;
Omar Sandoval22c61862015-03-16 04:33:53 -07002398 written = mapping->a_ops->direct_IO(iocb, &data, pos);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002399
2400 /*
2401 * Finally, try again to invalidate clean pages which might have been
2402 * cached by non-direct readahead, or faulted in by get_user_pages()
2403 * if the source of the write was an mmap'ed region of the file
2404 * we're writing. Either one is a pretty crazy thing to do,
2405 * so we don't support it 100%. If this invalidation
2406 * fails, tough, the write still worked...
2407 */
2408 if (mapping->nrpages) {
2409 invalidate_inode_pages2_range(mapping,
2410 pos >> PAGE_CACHE_SHIFT, end);
2411 }
2412
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002414 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002415 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002416 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2417 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 mark_inode_dirty(inode);
2419 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002420 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002422out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 return written;
2424}
2425EXPORT_SYMBOL(generic_file_direct_write);
2426
Nick Piggineb2be182007-10-16 01:24:57 -07002427/*
2428 * Find or create a page at the given pagecache position. Return the locked
2429 * page. This function is specifically for buffered writes.
2430 */
Nick Piggin54566b22009-01-04 12:00:53 -08002431struct page *grab_cache_page_write_begin(struct address_space *mapping,
2432 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002433{
Nick Piggineb2be182007-10-16 01:24:57 -07002434 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07002435 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002436
Nick Piggin54566b22009-01-04 12:00:53 -08002437 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002438 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002439
Mel Gorman2457aec2014-06-04 16:10:31 -07002440 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002441 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002442 if (page)
2443 wait_for_stable_page(page);
2444
Nick Piggineb2be182007-10-16 01:24:57 -07002445 return page;
2446}
Nick Piggin54566b22009-01-04 12:00:53 -08002447EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002448
Al Viro3b93f912014-02-11 21:34:08 -05002449ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002450 struct iov_iter *i, loff_t pos)
2451{
2452 struct address_space *mapping = file->f_mapping;
2453 const struct address_space_operations *a_ops = mapping->a_ops;
2454 long status = 0;
2455 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002456 unsigned int flags = 0;
2457
2458 /*
2459 * Copies from kernel address space cannot fail (NFSD is a big user).
2460 */
Al Viro777eda22014-12-17 04:46:46 -05002461 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002462 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002463
2464 do {
2465 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002466 unsigned long offset; /* Offset into pagecache page */
2467 unsigned long bytes; /* Bytes to write to page */
2468 size_t copied; /* Bytes copied from user */
2469 void *fsdata;
2470
2471 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002472 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2473 iov_iter_count(i));
2474
2475again:
Nick Piggin674b8922007-10-16 01:25:03 -07002476 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002477 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002478 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002479 break;
2480
anfei zhou931e80e2010-02-02 13:44:02 -08002481 if (mapping_writably_mapped(mapping))
2482 flush_dcache_page(page);
Dave Hansen998ef752015-09-08 15:00:28 -07002483 /*
2484 * 'page' is now locked. If we are trying to copy from a
2485 * mapping of 'page' in userspace, the copy might fault and
2486 * would need PageUptodate() to complete. But, page can not be
2487 * made Uptodate without acquiring the page lock, which we hold.
2488 * Deadlock. Avoid with pagefault_disable(). Fix up below with
2489 * iov_iter_fault_in_readable().
2490 */
2491 pagefault_disable();
Nick Pigginafddba42007-10-16 01:25:01 -07002492 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Dave Hansen998ef752015-09-08 15:00:28 -07002493 pagefault_enable();
Nick Pigginafddba42007-10-16 01:25:01 -07002494 flush_dcache_page(page);
2495
2496 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2497 page, fsdata);
2498 if (unlikely(status < 0))
2499 break;
2500 copied = status;
2501
2502 cond_resched();
2503
Nick Piggin124d3b72008-02-02 15:01:17 +01002504 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002505 if (unlikely(copied == 0)) {
2506 /*
2507 * If we were unable to copy any data at all, we must
2508 * fall back to a single segment length write.
2509 *
2510 * If we didn't fallback here, we could livelock
2511 * because not all segments in the iov can be copied at
2512 * once without a pagefault.
2513 */
2514 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2515 iov_iter_single_seg_count(i));
Dave Hansen998ef752015-09-08 15:00:28 -07002516 /*
2517 * This is the fallback to recover if the copy from
2518 * userspace above faults.
2519 */
2520 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2521 status = -EFAULT;
2522 break;
2523 }
Nick Pigginafddba42007-10-16 01:25:01 -07002524 goto again;
2525 }
Nick Pigginafddba42007-10-16 01:25:01 -07002526 pos += copied;
2527 written += copied;
2528
2529 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002530 if (fatal_signal_pending(current)) {
2531 status = -EINTR;
2532 break;
2533 }
Nick Pigginafddba42007-10-16 01:25:01 -07002534 } while (iov_iter_count(i));
2535
2536 return written ? written : status;
2537}
Al Viro3b93f912014-02-11 21:34:08 -05002538EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Jan Karae4dd9de2009-08-17 18:10:06 +02002540/**
Al Viro81742022014-04-03 03:17:43 -04002541 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002542 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002543 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002544 *
2545 * This function does all the work needed for actually writing data to a
2546 * file. It does all basic checks, removes SUID from the file, updates
2547 * modification times and calls proper subroutines depending on whether we
2548 * do direct IO or a standard buffered write.
2549 *
2550 * It expects i_mutex to be grabbed unless we work on a block device or similar
2551 * object which does not need locking at all.
2552 *
2553 * This function does *not* take care of syncing data in case of O_SYNC write.
2554 * A caller has to handle it. This is mainly due to the fact that we want to
2555 * avoid syncing under i_mutex.
2556 */
Al Viro81742022014-04-03 03:17:43 -04002557ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558{
2559 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002560 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002562 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002564 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002567 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002568 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 if (err)
2570 goto out;
2571
Josef Bacikc3b2da32012-03-26 09:59:21 -04002572 err = file_update_time(file);
2573 if (err)
2574 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
Al Viro2ba48ce2015-04-09 13:52:01 -04002576 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002577 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002578
Al Viro0b8def92015-04-07 10:22:53 -04002579 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002580 /*
2581 * If the write stopped short of completing, fall back to
2582 * buffered writes. Some filesystems do this for writes to
2583 * holes, for example. For DAX files, a buffered write will
2584 * not succeed (even if it did, DAX does not handle dirty
2585 * page-cache pages correctly).
2586 */
Al Viro0b8def92015-04-07 10:22:53 -04002587 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002589
Al Viro0b8def92015-04-07 10:22:53 -04002590 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002591 /*
Al Viro3b93f912014-02-11 21:34:08 -05002592 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002593 * then we want to return the number of bytes which were
2594 * direct-written, or the error code if that was zero. Note
2595 * that this differs from normal direct-io semantics, which
2596 * will return -EFOO even if some bytes were written.
2597 */
Al Viro60bb4522014-08-08 12:39:16 -04002598 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002599 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002600 goto out;
2601 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002602 /*
2603 * We need to ensure that the page cache pages are written to
2604 * disk and invalidated to preserve the expected O_DIRECT
2605 * semantics.
2606 */
Al Viro3b93f912014-02-11 21:34:08 -05002607 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002608 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002609 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002610 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002611 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002612 invalidate_mapping_pages(mapping,
2613 pos >> PAGE_CACHE_SHIFT,
2614 endbyte >> PAGE_CACHE_SHIFT);
2615 } else {
2616 /*
2617 * We don't know how much we wrote, so just return
2618 * the number of bytes which were direct-written
2619 */
2620 }
2621 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002622 written = generic_perform_write(file, from, iocb->ki_pos);
2623 if (likely(written > 0))
2624 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626out:
2627 current->backing_dev_info = NULL;
2628 return written ? written : err;
2629}
Al Viro81742022014-04-03 03:17:43 -04002630EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631
Jan Karae4dd9de2009-08-17 18:10:06 +02002632/**
Al Viro81742022014-04-03 03:17:43 -04002633 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002634 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002635 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002636 *
Al Viro81742022014-04-03 03:17:43 -04002637 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002638 * filesystems. It takes care of syncing the file in case of O_SYNC file
2639 * and acquires i_mutex as needed.
2640 */
Al Viro81742022014-04-03 03:17:43 -04002641ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642{
2643 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002644 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002647 mutex_lock(&inode->i_mutex);
Al Viro3309dd02015-04-09 12:55:47 -04002648 ret = generic_write_checks(iocb, from);
2649 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002650 ret = __generic_file_write_iter(iocb, from);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002651 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
Christoph Hellwig02afc272013-09-04 15:04:40 +02002653 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 ssize_t err;
2655
Al Virod311d792014-02-09 15:18:09 -05002656 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2657 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 ret = err;
2659 }
2660 return ret;
2661}
Al Viro81742022014-04-03 03:17:43 -04002662EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
David Howellscf9a2ae2006-08-29 19:05:54 +01002664/**
2665 * try_to_release_page() - release old fs-specific metadata on a page
2666 *
2667 * @page: the page which the kernel is trying to free
2668 * @gfp_mask: memory allocation flags (and I/O mode)
2669 *
2670 * The address_space is to try to release any data against the page
2671 * (presumably at page->private). If the release was successful, return `1'.
2672 * Otherwise return zero.
2673 *
David Howells266cf652009-04-03 16:42:36 +01002674 * This may also be called if PG_fscache is set on a page, indicating that the
2675 * page is known to the local caching routines.
2676 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002677 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002678 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002679 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002680 */
2681int try_to_release_page(struct page *page, gfp_t gfp_mask)
2682{
2683 struct address_space * const mapping = page->mapping;
2684
2685 BUG_ON(!PageLocked(page));
2686 if (PageWriteback(page))
2687 return 0;
2688
2689 if (mapping && mapping->a_ops->releasepage)
2690 return mapping->a_ops->releasepage(page, gfp_mask);
2691 return try_to_free_buffers(page);
2692}
2693
2694EXPORT_SYMBOL(try_to_release_page);