blob: 847ee43c28068a0fb744fe124c1d8afa429d0719 [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);
Kirill A. Shutemove1534ae2016-01-15 16:53:46 -0800207 VM_BUG_ON_PAGE(page_mapped(page), 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
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800334static int __filemap_fdatawait_range(struct address_space *mapping,
335 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200337 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
338 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 struct pagevec pvec;
340 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800341 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200343 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700344 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 while ((index <= end) &&
348 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
349 PAGECACHE_TAG_WRITEBACK,
350 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
351 unsigned i;
352
353 for (i = 0; i < nr_pages; i++) {
354 struct page *page = pvec.pages[i];
355
356 /* until radix tree lookup accepts end_index */
357 if (page->index > end)
358 continue;
359
360 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800361 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 ret = -EIO;
363 }
364 pagevec_release(&pvec);
365 cond_resched();
366 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700367out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800368 return ret;
369}
370
371/**
372 * filemap_fdatawait_range - wait for writeback to complete
373 * @mapping: address space structure to wait for
374 * @start_byte: offset in bytes where the range starts
375 * @end_byte: offset in bytes where the range ends (inclusive)
376 *
377 * Walk the list of under-writeback pages of the given address space
378 * in the given range and wait for all of them. Check error status of
379 * the address space and return it.
380 *
381 * Since the error status of the address space is cleared by this function,
382 * callers are responsible for checking the return value and handling and/or
383 * reporting the error.
384 */
385int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
386 loff_t end_byte)
387{
388 int ret, ret2;
389
390 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700391 ret2 = filemap_check_errors(mapping);
392 if (!ret)
393 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 return ret;
396}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200397EXPORT_SYMBOL(filemap_fdatawait_range);
398
399/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800400 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
401 * @mapping: address space structure to wait for
402 *
403 * Walk the list of under-writeback pages of the given address space
404 * and wait for all of them. Unlike filemap_fdatawait(), this function
405 * does not clear error status of the address space.
406 *
407 * Use this function if callers don't handle errors themselves. Expected
408 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
409 * fsfreeze(8)
410 */
411void filemap_fdatawait_keep_errors(struct address_space *mapping)
412{
413 loff_t i_size = i_size_read(mapping->host);
414
415 if (i_size == 0)
416 return;
417
418 __filemap_fdatawait_range(mapping, 0, i_size - 1);
419}
420
421/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700422 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700424 *
425 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800426 * and wait for all of them. Check error status of the address space
427 * and return it.
428 *
429 * Since the error status of the address space is cleared by this function,
430 * callers are responsible for checking the return value and handling and/or
431 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 */
433int filemap_fdatawait(struct address_space *mapping)
434{
435 loff_t i_size = i_size_read(mapping->host);
436
437 if (i_size == 0)
438 return 0;
439
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200440 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442EXPORT_SYMBOL(filemap_fdatawait);
443
444int filemap_write_and_wait(struct address_space *mapping)
445{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800446 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
448 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800449 err = filemap_fdatawrite(mapping);
450 /*
451 * Even if the above returned error, the pages may be
452 * written partially (e.g. -ENOSPC), so we wait for it.
453 * But the -EIO is special case, it may indicate the worst
454 * thing (e.g. bug) happened, so we avoid waiting for it.
455 */
456 if (err != -EIO) {
457 int err2 = filemap_fdatawait(mapping);
458 if (!err)
459 err = err2;
460 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700461 } else {
462 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800464 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800466EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Randy Dunlap485bb992006-06-23 02:03:49 -0700468/**
469 * filemap_write_and_wait_range - write out & wait on a file range
470 * @mapping: the address_space for the pages
471 * @lstart: offset in bytes where the range starts
472 * @lend: offset in bytes where the range ends (inclusive)
473 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800474 * Write out and wait upon file offsets lstart->lend, inclusive.
475 *
476 * Note that `lend' is inclusive (describes the last byte to be written) so
477 * that this function can be used to write to the very end-of-file (end = -1).
478 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479int filemap_write_and_wait_range(struct address_space *mapping,
480 loff_t lstart, loff_t lend)
481{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800482 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800485 err = __filemap_fdatawrite_range(mapping, lstart, lend,
486 WB_SYNC_ALL);
487 /* See comment of filemap_write_and_wait() */
488 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200489 int err2 = filemap_fdatawait_range(mapping,
490 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800491 if (!err)
492 err = err2;
493 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700494 } else {
495 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800497 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
Chris Masonf6995582009-04-15 13:22:37 -0400499EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Randy Dunlap485bb992006-06-23 02:03:49 -0700501/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700502 * replace_page_cache_page - replace a pagecache page with a new one
503 * @old: page to be replaced
504 * @new: page to replace with
505 * @gfp_mask: allocation mode
506 *
507 * This function replaces a page in the pagecache with a new one. On
508 * success it acquires the pagecache reference for the new page and
509 * drops it for the old page. Both the old and new pages must be
510 * locked. This function does not add the new page to the LRU, the
511 * caller must do that.
512 *
513 * The remove + add is atomic. The only way this function can fail is
514 * memory allocation failure.
515 */
516int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
517{
518 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700519
Sasha Levin309381fea2014-01-23 15:52:54 -0800520 VM_BUG_ON_PAGE(!PageLocked(old), old);
521 VM_BUG_ON_PAGE(!PageLocked(new), new);
522 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700523
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700524 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
525 if (!error) {
526 struct address_space *mapping = old->mapping;
527 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400528 struct mem_cgroup *memcg;
529 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700530
531 pgoff_t offset = old->index;
532 freepage = mapping->a_ops->freepage;
533
534 page_cache_get(new);
535 new->mapping = mapping;
536 new->index = offset;
537
Greg Thelenc4843a72015-05-22 17:13:16 -0400538 memcg = mem_cgroup_begin_page_stat(old);
539 spin_lock_irqsave(&mapping->tree_lock, flags);
540 __delete_from_page_cache(old, NULL, memcg);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700541 error = radix_tree_insert(&mapping->page_tree, offset, new);
542 BUG_ON(error);
543 mapping->nrpages++;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700544
545 /*
546 * hugetlb pages do not participate in page cache accounting.
547 */
548 if (!PageHuge(new))
549 __inc_zone_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700550 if (PageSwapBacked(new))
551 __inc_zone_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400552 spin_unlock_irqrestore(&mapping->tree_lock, flags);
553 mem_cgroup_end_page_stat(memcg);
Hugh Dickins45637ba2015-11-05 18:49:40 -0800554 mem_cgroup_replace_page(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700555 radix_tree_preload_end();
556 if (freepage)
557 freepage(old);
558 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700559 }
560
561 return error;
562}
563EXPORT_SYMBOL_GPL(replace_page_cache_page);
564
Johannes Weiner0cd61442014-04-03 14:47:46 -0700565static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700566 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700567{
Johannes Weiner449dd692014-04-03 14:47:56 -0700568 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700569 void **slot;
570 int error;
571
Johannes Weiner449dd692014-04-03 14:47:56 -0700572 error = __radix_tree_create(&mapping->page_tree, page->index,
573 &node, &slot);
574 if (error)
575 return error;
576 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700577 void *p;
578
579 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
580 if (!radix_tree_exceptional_entry(p))
581 return -EEXIST;
Johannes Weinera5289102014-04-03 14:47:51 -0700582 if (shadowp)
583 *shadowp = p;
Johannes Weiner449dd692014-04-03 14:47:56 -0700584 mapping->nrshadows--;
585 if (node)
586 workingset_node_shadows_dec(node);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700587 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700588 radix_tree_replace_slot(slot, page);
589 mapping->nrpages++;
590 if (node) {
591 workingset_node_pages_inc(node);
592 /*
593 * Don't track node that contains actual pages.
594 *
595 * Avoid acquiring the list_lru lock if already
596 * untracked. The list_empty() test is safe as
597 * node->private_list is protected by
598 * mapping->tree_lock.
599 */
600 if (!list_empty(&node->private_list))
601 list_lru_del(&workingset_shadow_nodes,
602 &node->private_list);
603 }
604 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700605}
606
Johannes Weinera5289102014-04-03 14:47:51 -0700607static int __add_to_page_cache_locked(struct page *page,
608 struct address_space *mapping,
609 pgoff_t offset, gfp_t gfp_mask,
610 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Johannes Weiner00501b52014-08-08 14:19:20 -0700612 int huge = PageHuge(page);
613 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700614 int error;
615
Sasha Levin309381fea2014-01-23 15:52:54 -0800616 VM_BUG_ON_PAGE(!PageLocked(page), page);
617 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700618
Johannes Weiner00501b52014-08-08 14:19:20 -0700619 if (!huge) {
620 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800621 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700622 if (error)
623 return error;
624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Jan Kara5e4c0d972013-09-11 14:26:05 -0700626 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700627 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700628 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800629 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700630 return error;
631 }
632
633 page_cache_get(page);
634 page->mapping = mapping;
635 page->index = offset;
636
637 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700638 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700639 radix_tree_preload_end();
640 if (unlikely(error))
641 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700642
643 /* hugetlb pages do not participate in page cache accounting. */
644 if (!huge)
645 __inc_zone_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700646 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700647 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800648 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700649 trace_mm_filemap_add_to_page_cache(page);
650 return 0;
651err_insert:
652 page->mapping = NULL;
653 /* Leave page->index set: truncation relies upon it */
654 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700655 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800656 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700657 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return error;
659}
Johannes Weinera5289102014-04-03 14:47:51 -0700660
661/**
662 * add_to_page_cache_locked - add a locked page to the pagecache
663 * @page: page to add
664 * @mapping: the page's address_space
665 * @offset: page index
666 * @gfp_mask: page allocation mode
667 *
668 * This function is used to add a page to the pagecache. It must be locked.
669 * This function does not add the page to the LRU. The caller must do that.
670 */
671int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
672 pgoff_t offset, gfp_t gfp_mask)
673{
674 return __add_to_page_cache_locked(page, mapping, offset,
675 gfp_mask, NULL);
676}
Nick Piggine2867812008-07-25 19:45:30 -0700677EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400680 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Johannes Weinera5289102014-04-03 14:47:51 -0700682 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700683 int ret;
684
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800685 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700686 ret = __add_to_page_cache_locked(page, mapping, offset,
687 gfp_mask, &shadow);
688 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800689 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700690 else {
691 /*
692 * The page might have been evicted from cache only
693 * recently, in which case it should be activated like
694 * any other repeatedly accessed page.
695 */
696 if (shadow && workingset_refault(shadow)) {
697 SetPageActive(page);
698 workingset_activation(page);
699 } else
700 ClearPageActive(page);
701 lru_cache_add(page);
702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return ret;
704}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300705EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Paul Jackson44110fe2006-03-24 03:16:04 -0800707#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700708struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800709{
Miao Xiec0ff7452010-05-24 14:32:08 -0700710 int n;
711 struct page *page;
712
Paul Jackson44110fe2006-03-24 03:16:04 -0800713 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700714 unsigned int cpuset_mems_cookie;
715 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700716 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700717 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700718 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700719 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700720
Miao Xiec0ff7452010-05-24 14:32:08 -0700721 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800722 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700723 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800724}
Nick Piggin2ae88142006-10-28 10:38:23 -0700725EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800726#endif
727
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728/*
729 * In order to wait for pages to become available there must be
730 * waitqueues associated with pages. By using a hash table of
731 * waitqueues where the bucket discipline is to maintain all
732 * waiters on the same queue and wake all when any of the pages
733 * become available, and for the woken contexts to check to be
734 * sure the appropriate page became available, this saves space
735 * at a cost of "thundering herd" phenomena during rare hash
736 * collisions.
737 */
NeilBrowna4796e32014-09-24 11:28:32 +1000738wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739{
740 const struct zone *zone = page_zone(page);
741
742 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
743}
NeilBrowna4796e32014-09-24 11:28:32 +1000744EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
Harvey Harrison920c7a52008-02-04 22:29:26 -0800746void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
748 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
749
750 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000751 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 TASK_UNINTERRUPTIBLE);
753}
754EXPORT_SYMBOL(wait_on_page_bit);
755
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700756int wait_on_page_bit_killable(struct page *page, int bit_nr)
757{
758 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
759
760 if (!test_bit(bit_nr, &page->flags))
761 return 0;
762
763 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000764 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700765}
766
NeilBrowncbbce822014-09-25 13:55:19 +1000767int wait_on_page_bit_killable_timeout(struct page *page,
768 int bit_nr, unsigned long timeout)
769{
770 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
771
772 wait.key.timeout = jiffies + timeout;
773 if (!test_bit(bit_nr, &page->flags))
774 return 0;
775 return __wait_on_bit(page_waitqueue(page), &wait,
776 bit_wait_io_timeout, TASK_KILLABLE);
777}
778EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100781 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700782 * @page: Page defining the wait queue of interest
783 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100784 *
785 * Add an arbitrary @waiter to the wait queue for the nominated @page.
786 */
787void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
788{
789 wait_queue_head_t *q = page_waitqueue(page);
790 unsigned long flags;
791
792 spin_lock_irqsave(&q->lock, flags);
793 __add_wait_queue(q, waiter);
794 spin_unlock_irqrestore(&q->lock, flags);
795}
796EXPORT_SYMBOL_GPL(add_page_wait_queue);
797
798/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700799 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * @page: the page
801 *
802 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
803 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900804 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
806 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700807 * The mb is necessary to enforce ordering between the clear_bit and the read
808 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800810void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800812 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800813 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700814 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100815 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 wake_up_page(page, PG_locked);
817}
818EXPORT_SYMBOL(unlock_page);
819
Randy Dunlap485bb992006-06-23 02:03:49 -0700820/**
821 * end_page_writeback - end writeback against a page
822 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 */
824void end_page_writeback(struct page *page)
825{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700826 /*
827 * TestClearPageReclaim could be used here but it is an atomic
828 * operation and overkill in this particular case. Failing to
829 * shuffle a page marked for immediate reclaim is too mild to
830 * justify taking an atomic operation penalty at the end of
831 * ever page writeback.
832 */
833 if (PageReclaim(page)) {
834 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700835 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700836 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700837
838 if (!test_clear_page_writeback(page))
839 BUG();
840
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100841 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 wake_up_page(page, PG_writeback);
843}
844EXPORT_SYMBOL(end_page_writeback);
845
Matthew Wilcox57d99842014-06-04 16:07:45 -0700846/*
847 * After completing I/O on a page, call this routine to update the page
848 * flags appropriately
849 */
850void page_endio(struct page *page, int rw, int err)
851{
852 if (rw == READ) {
853 if (!err) {
854 SetPageUptodate(page);
855 } else {
856 ClearPageUptodate(page);
857 SetPageError(page);
858 }
859 unlock_page(page);
860 } else { /* rw == WRITE */
861 if (err) {
862 SetPageError(page);
863 if (page->mapping)
864 mapping_set_error(page->mapping, err);
865 }
866 end_page_writeback(page);
867 }
868}
869EXPORT_SYMBOL_GPL(page_endio);
870
Randy Dunlap485bb992006-06-23 02:03:49 -0700871/**
872 * __lock_page - get a lock on the page, assuming we need to sleep to get it
873 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800875void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800877 struct page *page_head = compound_head(page);
878 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800880 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 TASK_UNINTERRUPTIBLE);
882}
883EXPORT_SYMBOL(__lock_page);
884
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800885int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500886{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800887 struct page *page_head = compound_head(page);
888 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500889
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800890 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000891 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500892}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300893EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500894
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700895/*
896 * Return values:
897 * 1 - page is locked; mmap_sem is still held.
898 * 0 - page is not locked.
899 * mmap_sem has been released (up_read()), unless flags had both
900 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
901 * which case mmap_sem is still held.
902 *
903 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
904 * with the page locked and the mmap_sem unperturbed.
905 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700906int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
907 unsigned int flags)
908{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700909 if (flags & FAULT_FLAG_ALLOW_RETRY) {
910 /*
911 * CAUTION! In this case, mmap_sem is not released
912 * even though return 0.
913 */
914 if (flags & FAULT_FLAG_RETRY_NOWAIT)
915 return 0;
916
917 up_read(&mm->mmap_sem);
918 if (flags & FAULT_FLAG_KILLABLE)
919 wait_on_page_locked_killable(page);
920 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700921 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700922 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700923 } else {
924 if (flags & FAULT_FLAG_KILLABLE) {
925 int ret;
926
927 ret = __lock_page_killable(page);
928 if (ret) {
929 up_read(&mm->mmap_sem);
930 return 0;
931 }
932 } else
933 __lock_page(page);
934 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700935 }
936}
937
Randy Dunlap485bb992006-06-23 02:03:49 -0700938/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700939 * page_cache_next_hole - find the next hole (not-present entry)
940 * @mapping: mapping
941 * @index: index
942 * @max_scan: maximum range to search
943 *
944 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
945 * lowest indexed hole.
946 *
947 * Returns: the index of the hole if found, otherwise returns an index
948 * outside of the set specified (in which case 'return - index >=
949 * max_scan' will be true). In rare cases of index wrap-around, 0 will
950 * be returned.
951 *
952 * page_cache_next_hole may be called under rcu_read_lock. However,
953 * like radix_tree_gang_lookup, this will not atomically search a
954 * snapshot of the tree at a single point in time. For example, if a
955 * hole is created at index 5, then subsequently a hole is created at
956 * index 10, page_cache_next_hole covering both indexes may return 10
957 * if called under rcu_read_lock.
958 */
959pgoff_t page_cache_next_hole(struct address_space *mapping,
960 pgoff_t index, unsigned long max_scan)
961{
962 unsigned long i;
963
964 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700965 struct page *page;
966
967 page = radix_tree_lookup(&mapping->page_tree, index);
968 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700969 break;
970 index++;
971 if (index == 0)
972 break;
973 }
974
975 return index;
976}
977EXPORT_SYMBOL(page_cache_next_hole);
978
979/**
980 * page_cache_prev_hole - find the prev hole (not-present entry)
981 * @mapping: mapping
982 * @index: index
983 * @max_scan: maximum range to search
984 *
985 * Search backwards in the range [max(index-max_scan+1, 0), index] for
986 * the first hole.
987 *
988 * Returns: the index of the hole if found, otherwise returns an index
989 * outside of the set specified (in which case 'index - return >=
990 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
991 * will be returned.
992 *
993 * page_cache_prev_hole may be called under rcu_read_lock. However,
994 * like radix_tree_gang_lookup, this will not atomically search a
995 * snapshot of the tree at a single point in time. For example, if a
996 * hole is created at index 10, then subsequently a hole is created at
997 * index 5, page_cache_prev_hole covering both indexes may return 5 if
998 * called under rcu_read_lock.
999 */
1000pgoff_t page_cache_prev_hole(struct address_space *mapping,
1001 pgoff_t index, unsigned long max_scan)
1002{
1003 unsigned long i;
1004
1005 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001006 struct page *page;
1007
1008 page = radix_tree_lookup(&mapping->page_tree, index);
1009 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001010 break;
1011 index--;
1012 if (index == ULONG_MAX)
1013 break;
1014 }
1015
1016 return index;
1017}
1018EXPORT_SYMBOL(page_cache_prev_hole);
1019
1020/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001021 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001022 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001023 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001024 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001025 * Looks up the page cache slot at @mapping & @offset. If there is a
1026 * page cache page, it is returned with an increased refcount.
1027 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001028 * If the slot holds a shadow entry of a previously evicted page, or a
1029 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001030 *
1031 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001033struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Nick Piggina60637c2008-07-25 19:45:31 -07001035 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 struct page *page;
1037
Nick Piggina60637c2008-07-25 19:45:31 -07001038 rcu_read_lock();
1039repeat:
1040 page = NULL;
1041 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1042 if (pagep) {
1043 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001044 if (unlikely(!page))
1045 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001046 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001047 if (radix_tree_deref_retry(page))
1048 goto repeat;
1049 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001050 * A shadow entry of a recently evicted page,
1051 * or a swap entry from shmem/tmpfs. Return
1052 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001053 */
1054 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001055 }
Nick Piggina60637c2008-07-25 19:45:31 -07001056 if (!page_cache_get_speculative(page))
1057 goto repeat;
1058
1059 /*
1060 * Has the page moved?
1061 * This is part of the lockless pagecache protocol. See
1062 * include/linux/pagemap.h for details.
1063 */
1064 if (unlikely(page != *pagep)) {
1065 page_cache_release(page);
1066 goto repeat;
1067 }
1068 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001069out:
Nick Piggina60637c2008-07-25 19:45:31 -07001070 rcu_read_unlock();
1071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 return page;
1073}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001074EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
Randy Dunlap485bb992006-06-23 02:03:49 -07001076/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001077 * find_lock_entry - locate, pin and lock a page cache entry
1078 * @mapping: the address_space to search
1079 * @offset: the page cache index
1080 *
1081 * Looks up the page cache slot at @mapping & @offset. If there is a
1082 * page cache page, it is returned locked and with an increased
1083 * refcount.
1084 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001085 * If the slot holds a shadow entry of a previously evicted page, or a
1086 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001087 *
1088 * Otherwise, %NULL is returned.
1089 *
1090 * find_lock_entry() may sleep.
1091 */
1092struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093{
1094 struct page *page;
1095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001097 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001098 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001099 lock_page(page);
1100 /* Has the page been truncated? */
1101 if (unlikely(page->mapping != mapping)) {
1102 unlock_page(page);
1103 page_cache_release(page);
1104 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001106 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 return page;
1109}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001110EXPORT_SYMBOL(find_lock_entry);
1111
1112/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001113 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001114 * @mapping: the address_space to search
1115 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001116 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001117 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001118 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001119 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001120 *
Randy Dunlap75325182014-07-30 16:08:37 -07001121 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001122 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001123 * FGP_ACCESSED: the page will be marked accessed
1124 * FGP_LOCK: Page is return locked
1125 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001126 * @gfp_mask and added to the page cache and the VM's LRU
1127 * list. The page is returned locked and with an increased
1128 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001129 *
1130 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1131 * if the GFP flags specified for FGP_CREAT are atomic.
1132 *
1133 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001134 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001135struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001136 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137{
Nick Piggineb2be182007-10-16 01:24:57 -07001138 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001139
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001141 page = find_get_entry(mapping, offset);
1142 if (radix_tree_exceptional_entry(page))
1143 page = NULL;
1144 if (!page)
1145 goto no_page;
1146
1147 if (fgp_flags & FGP_LOCK) {
1148 if (fgp_flags & FGP_NOWAIT) {
1149 if (!trylock_page(page)) {
1150 page_cache_release(page);
1151 return NULL;
1152 }
1153 } else {
1154 lock_page(page);
1155 }
1156
1157 /* Has the page been truncated? */
1158 if (unlikely(page->mapping != mapping)) {
1159 unlock_page(page);
1160 page_cache_release(page);
1161 goto repeat;
1162 }
1163 VM_BUG_ON_PAGE(page->index != offset, page);
1164 }
1165
1166 if (page && (fgp_flags & FGP_ACCESSED))
1167 mark_page_accessed(page);
1168
1169no_page:
1170 if (!page && (fgp_flags & FGP_CREAT)) {
1171 int err;
1172 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001173 gfp_mask |= __GFP_WRITE;
1174 if (fgp_flags & FGP_NOFS)
1175 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001176
Michal Hocko45f87de2014-12-29 20:30:35 +01001177 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001178 if (!page)
1179 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001180
1181 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1182 fgp_flags |= FGP_LOCK;
1183
Hugh Dickinseb39d612014-08-06 16:06:43 -07001184 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001185 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001186 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001187
Michal Hocko45f87de2014-12-29 20:30:35 +01001188 err = add_to_page_cache_lru(page, mapping, offset,
1189 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001190 if (unlikely(err)) {
1191 page_cache_release(page);
1192 page = NULL;
1193 if (err == -EEXIST)
1194 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001197
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 return page;
1199}
Mel Gorman2457aec2014-06-04 16:10:31 -07001200EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001203 * find_get_entries - gang pagecache lookup
1204 * @mapping: The address_space to search
1205 * @start: The starting page cache index
1206 * @nr_entries: The maximum number of entries
1207 * @entries: Where the resulting entries are placed
1208 * @indices: The cache indices corresponding to the entries in @entries
1209 *
1210 * find_get_entries() will search for and return a group of up to
1211 * @nr_entries entries in the mapping. The entries are placed at
1212 * @entries. find_get_entries() takes a reference against any actual
1213 * pages it returns.
1214 *
1215 * The search returns a group of mapping-contiguous page cache entries
1216 * with ascending indexes. There may be holes in the indices due to
1217 * not-present pages.
1218 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001219 * Any shadow entries of evicted pages, or swap entries from
1220 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001221 *
1222 * find_get_entries() returns the number of pages and shadow entries
1223 * which were found.
1224 */
1225unsigned find_get_entries(struct address_space *mapping,
1226 pgoff_t start, unsigned int nr_entries,
1227 struct page **entries, pgoff_t *indices)
1228{
1229 void **slot;
1230 unsigned int ret = 0;
1231 struct radix_tree_iter iter;
1232
1233 if (!nr_entries)
1234 return 0;
1235
1236 rcu_read_lock();
1237restart:
1238 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1239 struct page *page;
1240repeat:
1241 page = radix_tree_deref_slot(slot);
1242 if (unlikely(!page))
1243 continue;
1244 if (radix_tree_exception(page)) {
1245 if (radix_tree_deref_retry(page))
1246 goto restart;
1247 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001248 * A shadow entry of a recently evicted page,
1249 * or a swap entry from shmem/tmpfs. Return
1250 * it without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001251 */
1252 goto export;
1253 }
1254 if (!page_cache_get_speculative(page))
1255 goto repeat;
1256
1257 /* Has the page moved? */
1258 if (unlikely(page != *slot)) {
1259 page_cache_release(page);
1260 goto repeat;
1261 }
1262export:
1263 indices[ret] = iter.index;
1264 entries[ret] = page;
1265 if (++ret == nr_entries)
1266 break;
1267 }
1268 rcu_read_unlock();
1269 return ret;
1270}
1271
1272/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 * find_get_pages - gang pagecache lookup
1274 * @mapping: The address_space to search
1275 * @start: The starting page index
1276 * @nr_pages: The maximum number of pages
1277 * @pages: Where the resulting pages are placed
1278 *
1279 * find_get_pages() will search for and return a group of up to
1280 * @nr_pages pages in the mapping. The pages are placed at @pages.
1281 * find_get_pages() takes a reference against the returned pages.
1282 *
1283 * The search returns a group of mapping-contiguous pages with ascending
1284 * indexes. There may be holes in the indices due to not-present pages.
1285 *
1286 * find_get_pages() returns the number of pages which were found.
1287 */
1288unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1289 unsigned int nr_pages, struct page **pages)
1290{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001291 struct radix_tree_iter iter;
1292 void **slot;
1293 unsigned ret = 0;
1294
1295 if (unlikely(!nr_pages))
1296 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Nick Piggina60637c2008-07-25 19:45:31 -07001298 rcu_read_lock();
1299restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001300 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001301 struct page *page;
1302repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001303 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001304 if (unlikely(!page))
1305 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001306
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001307 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001308 if (radix_tree_deref_retry(page)) {
1309 /*
1310 * Transient condition which can only trigger
1311 * when entry at index 0 moves out of or back
1312 * to root: none yet gotten, safe to restart.
1313 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001314 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001315 goto restart;
1316 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001317 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001318 * A shadow entry of a recently evicted page,
1319 * or a swap entry from shmem/tmpfs. Skip
1320 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001321 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001322 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001323 }
Nick Piggina60637c2008-07-25 19:45:31 -07001324
1325 if (!page_cache_get_speculative(page))
1326 goto repeat;
1327
1328 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001329 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001330 page_cache_release(page);
1331 goto repeat;
1332 }
1333
1334 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001335 if (++ret == nr_pages)
1336 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001337 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001338
Nick Piggina60637c2008-07-25 19:45:31 -07001339 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 return ret;
1341}
1342
Jens Axboeebf43502006-04-27 08:46:01 +02001343/**
1344 * find_get_pages_contig - gang contiguous pagecache lookup
1345 * @mapping: The address_space to search
1346 * @index: The starting page index
1347 * @nr_pages: The maximum number of pages
1348 * @pages: Where the resulting pages are placed
1349 *
1350 * find_get_pages_contig() works exactly like find_get_pages(), except
1351 * that the returned number of pages are guaranteed to be contiguous.
1352 *
1353 * find_get_pages_contig() returns the number of pages which were found.
1354 */
1355unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1356 unsigned int nr_pages, struct page **pages)
1357{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001358 struct radix_tree_iter iter;
1359 void **slot;
1360 unsigned int ret = 0;
1361
1362 if (unlikely(!nr_pages))
1363 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001364
Nick Piggina60637c2008-07-25 19:45:31 -07001365 rcu_read_lock();
1366restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001367 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001368 struct page *page;
1369repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001370 page = radix_tree_deref_slot(slot);
1371 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001372 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001373 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001374
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001375 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001376 if (radix_tree_deref_retry(page)) {
1377 /*
1378 * Transient condition which can only trigger
1379 * when entry at index 0 moves out of or back
1380 * to root: none yet gotten, safe to restart.
1381 */
1382 goto restart;
1383 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001384 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001385 * A shadow entry of a recently evicted page,
1386 * or a swap entry from shmem/tmpfs. Stop
1387 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001388 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001389 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001390 }
Nick Piggina60637c2008-07-25 19:45:31 -07001391
Nick Piggina60637c2008-07-25 19:45:31 -07001392 if (!page_cache_get_speculative(page))
1393 goto repeat;
1394
1395 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001396 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001397 page_cache_release(page);
1398 goto repeat;
1399 }
1400
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001401 /*
1402 * must check mapping and index after taking the ref.
1403 * otherwise we can get both false positives and false
1404 * negatives, which is just confusing to the caller.
1405 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001406 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001407 page_cache_release(page);
1408 break;
1409 }
1410
Nick Piggina60637c2008-07-25 19:45:31 -07001411 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001412 if (++ret == nr_pages)
1413 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001414 }
Nick Piggina60637c2008-07-25 19:45:31 -07001415 rcu_read_unlock();
1416 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001417}
David Howellsef71c152007-05-09 02:33:44 -07001418EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001419
Randy Dunlap485bb992006-06-23 02:03:49 -07001420/**
1421 * find_get_pages_tag - find and return pages that match @tag
1422 * @mapping: the address_space to search
1423 * @index: the starting page index
1424 * @tag: the tag index
1425 * @nr_pages: the maximum number of pages
1426 * @pages: where the resulting pages are placed
1427 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001429 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 */
1431unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1432 int tag, unsigned int nr_pages, struct page **pages)
1433{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001434 struct radix_tree_iter iter;
1435 void **slot;
1436 unsigned ret = 0;
1437
1438 if (unlikely(!nr_pages))
1439 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Nick Piggina60637c2008-07-25 19:45:31 -07001441 rcu_read_lock();
1442restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001443 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1444 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001445 struct page *page;
1446repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001447 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001448 if (unlikely(!page))
1449 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001450
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001451 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001452 if (radix_tree_deref_retry(page)) {
1453 /*
1454 * Transient condition which can only trigger
1455 * when entry at index 0 moves out of or back
1456 * to root: none yet gotten, safe to restart.
1457 */
1458 goto restart;
1459 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001460 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001461 * A shadow entry of a recently evicted page.
1462 *
1463 * Those entries should never be tagged, but
1464 * this tree walk is lockless and the tags are
1465 * looked up in bulk, one radix tree node at a
1466 * time, so there is a sizable window for page
1467 * reclaim to evict a page we saw tagged.
1468 *
1469 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001470 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001471 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001472 }
Nick Piggina60637c2008-07-25 19:45:31 -07001473
1474 if (!page_cache_get_speculative(page))
1475 goto repeat;
1476
1477 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001478 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001479 page_cache_release(page);
1480 goto repeat;
1481 }
1482
1483 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001484 if (++ret == nr_pages)
1485 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001486 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001487
Nick Piggina60637c2008-07-25 19:45:31 -07001488 rcu_read_unlock();
1489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 if (ret)
1491 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return ret;
1494}
David Howellsef71c152007-05-09 02:33:44 -07001495EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001497/*
1498 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1499 * a _large_ part of the i/o request. Imagine the worst scenario:
1500 *
1501 * ---R__________________________________________B__________
1502 * ^ reading here ^ bad block(assume 4k)
1503 *
1504 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1505 * => failing the whole request => read(R) => read(R+1) =>
1506 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1507 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1508 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1509 *
1510 * It is going insane. Fix it by quickly scaling down the readahead size.
1511 */
1512static void shrink_readahead_size_eio(struct file *filp,
1513 struct file_ra_state *ra)
1514{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001515 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001516}
1517
Randy Dunlap485bb992006-06-23 02:03:49 -07001518/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001519 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001520 * @filp: the file to read
1521 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001522 * @iter: data destination
1523 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001524 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001526 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 *
1528 * This is really ugly. But the goto's actually try to clarify some
1529 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 */
Al Viro6e58e792014-02-03 17:07:03 -05001531static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1532 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001534 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001536 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001537 pgoff_t index;
1538 pgoff_t last_index;
1539 pgoff_t prev_index;
1540 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001541 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001542 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001545 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1546 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Al Viro6e58e792014-02-03 17:07:03 -05001547 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 offset = *ppos & ~PAGE_CACHE_MASK;
1549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 for (;;) {
1551 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001552 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001553 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 unsigned long nr, ret;
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557find_page:
1558 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001559 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001560 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001561 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001562 index, last_index - index);
1563 page = find_get_page(mapping, index);
1564 if (unlikely(page == NULL))
1565 goto no_cached_page;
1566 }
1567 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001568 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001569 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001570 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001572 if (!PageUptodate(page)) {
1573 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1574 !mapping->a_ops->is_partially_uptodate)
1575 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001576 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001577 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001578 /* Did it get truncated before we got the lock? */
1579 if (!page->mapping)
1580 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001581 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001582 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001583 goto page_not_up_to_date_locked;
1584 unlock_page(page);
1585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001587 /*
1588 * i_size must be checked after we know the page is Uptodate.
1589 *
1590 * Checking i_size after the check allows us to calculate
1591 * the correct value for "nr", which means the zero-filled
1592 * part of the page is not copied back to userspace (unless
1593 * another truncate extends the file - this is desired though).
1594 */
1595
1596 isize = i_size_read(inode);
1597 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1598 if (unlikely(!isize || index > end_index)) {
1599 page_cache_release(page);
1600 goto out;
1601 }
1602
1603 /* nr is the maximum number of bytes to copy from this page */
1604 nr = PAGE_CACHE_SIZE;
1605 if (index == end_index) {
1606 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1607 if (nr <= offset) {
1608 page_cache_release(page);
1609 goto out;
1610 }
1611 }
1612 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
1614 /* If users can be writing to this page using arbitrary
1615 * virtual addresses, take care about potential aliasing
1616 * before reading the page on the kernel side.
1617 */
1618 if (mapping_writably_mapped(mapping))
1619 flush_dcache_page(page);
1620
1621 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001622 * When a sequential read accesses a page several times,
1623 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 */
Jan Karaec0f1632007-05-06 14:49:25 -07001625 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 mark_page_accessed(page);
1627 prev_index = index;
1628
1629 /*
1630 * Ok, we have the page, and it's up-to-date, so
1631 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 */
Al Viro6e58e792014-02-03 17:07:03 -05001633
1634 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 offset += ret;
1636 index += offset >> PAGE_CACHE_SHIFT;
1637 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001638 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
1640 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001641 written += ret;
1642 if (!iov_iter_count(iter))
1643 goto out;
1644 if (ret < nr) {
1645 error = -EFAULT;
1646 goto out;
1647 }
1648 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
1650page_not_up_to_date:
1651 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001652 error = lock_page_killable(page);
1653 if (unlikely(error))
1654 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001656page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001657 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 if (!page->mapping) {
1659 unlock_page(page);
1660 page_cache_release(page);
1661 continue;
1662 }
1663
1664 /* Did somebody else fill it already? */
1665 if (PageUptodate(page)) {
1666 unlock_page(page);
1667 goto page_ok;
1668 }
1669
1670readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001671 /*
1672 * A previous I/O error may have been due to temporary
1673 * failures, eg. multipath errors.
1674 * PG_error will be set again if readpage fails.
1675 */
1676 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 /* Start the actual read. The read will unlock the page. */
1678 error = mapping->a_ops->readpage(filp, page);
1679
Zach Brown994fc28c2005-12-15 14:28:17 -08001680 if (unlikely(error)) {
1681 if (error == AOP_TRUNCATED_PAGE) {
1682 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001683 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001684 goto find_page;
1685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001690 error = lock_page_killable(page);
1691 if (unlikely(error))
1692 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 if (!PageUptodate(page)) {
1694 if (page->mapping == NULL) {
1695 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001696 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 */
1698 unlock_page(page);
1699 page_cache_release(page);
1700 goto find_page;
1701 }
1702 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001703 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001704 error = -EIO;
1705 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 }
1707 unlock_page(page);
1708 }
1709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 goto page_ok;
1711
1712readpage_error:
1713 /* UHHUH! A synchronous read error occurred. Report it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 page_cache_release(page);
1715 goto out;
1716
1717no_cached_page:
1718 /*
1719 * Ok, it wasn't cached, so we need to create a new
1720 * page..
1721 */
Nick Piggineb2be182007-10-16 01:24:57 -07001722 page = page_cache_alloc_cold(mapping);
1723 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001724 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001725 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001727 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08001728 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001730 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001731 if (error == -EEXIST) {
1732 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 goto out;
1736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 goto readpage;
1738 }
1739
1740out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001741 ra->prev_pos = prev_index;
1742 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1743 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001745 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001746 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001747 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748}
1749
Randy Dunlap485bb992006-06-23 02:03:49 -07001750/**
Al Viro6abd2322014-04-04 14:20:57 -04001751 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001752 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001753 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001754 *
Al Viro6abd2322014-04-04 14:20:57 -04001755 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 * that can use the page cache directly.
1757 */
1758ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001759generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
Al Viroed978a82014-03-05 22:53:04 -05001761 struct file *file = iocb->ki_filp;
1762 ssize_t retval = 0;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001763 loff_t *ppos = &iocb->ki_pos;
Al Viroed978a82014-03-05 22:53:04 -05001764 loff_t pos = *ppos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
Al Viro2ba48ce2015-04-09 13:52:01 -04001766 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001767 struct address_space *mapping = file->f_mapping;
1768 struct inode *inode = mapping->host;
1769 size_t count = iov_iter_count(iter);
Badari Pulavarty543ade12006-09-30 23:28:48 -07001770 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 if (!count)
1773 goto out; /* skip atime */
1774 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001775 retval = filemap_write_and_wait_range(mapping, pos,
Al Viroed978a82014-03-05 22:53:04 -05001776 pos + count - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001777 if (!retval) {
Al Viroed978a82014-03-05 22:53:04 -05001778 struct iov_iter data = *iter;
Omar Sandoval22c61862015-03-16 04:33:53 -07001779 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001780 }
Al Viroed978a82014-03-05 22:53:04 -05001781
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001782 if (retval > 0) {
1783 *ppos = pos + retval;
Al Viroed978a82014-03-05 22:53:04 -05001784 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001785 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001786
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001787 /*
1788 * Btrfs can have a short DIO read if we encounter
1789 * compressed extents, so if there was an error, or if
1790 * we've already read everything we wanted to, or if
1791 * there was a short read because we hit EOF, go ahead
1792 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001793 * the rest of the read. Buffered reads will not work for
1794 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001795 */
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001796 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1797 IS_DAX(inode)) {
Al Viroed978a82014-03-05 22:53:04 -05001798 file_accessed(file);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001799 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001800 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 }
1802
Al Viroed978a82014-03-05 22:53:04 -05001803 retval = do_generic_file_read(file, ppos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804out:
1805 return retval;
1806}
Al Viroed978a82014-03-05 22:53:04 -05001807EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001810/**
1811 * page_cache_read - adds requested page to the page cache if not already there
1812 * @file: file to read
1813 * @offset: page index
1814 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 * This adds the requested page to the page cache if it isn't already there,
1816 * and schedules an I/O to read in its contents from disk.
1817 */
Michal Hockoc20cd452016-01-14 15:20:12 -08001818static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
1820 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001821 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001822 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Zach Brown994fc28c2005-12-15 14:28:17 -08001824 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08001825 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08001826 if (!page)
1827 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
Michal Hockoc20cd452016-01-14 15:20:12 -08001829 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
Zach Brown994fc28c2005-12-15 14:28:17 -08001830 if (ret == 0)
1831 ret = mapping->a_ops->readpage(file, page);
1832 else if (ret == -EEXIST)
1833 ret = 0; /* losing race to add is OK */
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Zach Brown994fc28c2005-12-15 14:28:17 -08001837 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07001838
Zach Brown994fc28c2005-12-15 14:28:17 -08001839 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840}
1841
1842#define MMAP_LOTSAMISS (100)
1843
Linus Torvaldsef00e082009-06-16 15:31:25 -07001844/*
1845 * Synchronous readahead happens when we don't even find
1846 * a page in the page cache at all.
1847 */
1848static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1849 struct file_ra_state *ra,
1850 struct file *file,
1851 pgoff_t offset)
1852{
Linus Torvaldsef00e082009-06-16 15:31:25 -07001853 struct address_space *mapping = file->f_mapping;
1854
1855 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001856 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001857 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001858 if (!ra->ra_pages)
1859 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001860
Joe Perches64363aa2013-07-08 16:00:18 -07001861 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001862 page_cache_sync_readahead(mapping, ra, file, offset,
1863 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001864 return;
1865 }
1866
Andi Kleen207d04b2011-05-24 17:12:29 -07001867 /* Avoid banging the cache line if not needed */
1868 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001869 ra->mmap_miss++;
1870
1871 /*
1872 * Do we miss much more than hit in this file? If so,
1873 * stop bothering with read-ahead. It will only hurt.
1874 */
1875 if (ra->mmap_miss > MMAP_LOTSAMISS)
1876 return;
1877
Wu Fengguangd30a1102009-06-16 15:31:30 -07001878 /*
1879 * mmap read-around
1880 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08001881 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1882 ra->size = ra->ra_pages;
1883 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001884 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001885}
1886
1887/*
1888 * Asynchronous readahead happens when we find the page and PG_readahead,
1889 * so we want to possibly extend the readahead further..
1890 */
1891static void do_async_mmap_readahead(struct vm_area_struct *vma,
1892 struct file_ra_state *ra,
1893 struct file *file,
1894 struct page *page,
1895 pgoff_t offset)
1896{
1897 struct address_space *mapping = file->f_mapping;
1898
1899 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001900 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001901 return;
1902 if (ra->mmap_miss > 0)
1903 ra->mmap_miss--;
1904 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001905 page_cache_async_readahead(mapping, ra, file,
1906 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001907}
1908
Randy Dunlap485bb992006-06-23 02:03:49 -07001909/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001910 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001911 * @vma: vma in which the fault was taken
1912 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001913 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001914 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 * mapped memory region to read in file data during a page fault.
1916 *
1917 * The goto's are kind of ugly, but this streamlines the normal case of having
1918 * it in the page cache, and handles the special cases reasonably without
1919 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001920 *
1921 * vma->vm_mm->mmap_sem must be held on entry.
1922 *
1923 * If our return value has VM_FAULT_RETRY set, it's because
1924 * lock_page_or_retry() returned 0.
1925 * The mmap_sem has usually been released in this case.
1926 * See __lock_page_or_retry() for the exception.
1927 *
1928 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1929 * has not been released.
1930 *
1931 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001933int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
1935 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001936 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 struct address_space *mapping = file->f_mapping;
1938 struct file_ra_state *ra = &file->f_ra;
1939 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001940 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001942 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001943 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001945 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1946 if (offset >= size >> PAGE_CACHE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001947 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001950 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001952 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001953 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001955 * We found the page, so try async readahead before
1956 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001958 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001959 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001960 /* No page in the page cache at all */
1961 do_sync_mmap_readahead(vma, ra, file, offset);
1962 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001963 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001964 ret = VM_FAULT_MAJOR;
1965retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001966 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (!page)
1968 goto no_cached_page;
1969 }
1970
Michel Lespinassed88c0922010-11-02 13:05:18 -07001971 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1972 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001973 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001974 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001975
1976 /* Did it get truncated? */
1977 if (unlikely(page->mapping != mapping)) {
1978 unlock_page(page);
1979 put_page(page);
1980 goto retry_find;
1981 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001982 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07001983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001985 * We have a locked page in the page cache, now we need to check
1986 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 */
Nick Piggind00806b2007-07-19 01:46:57 -07001988 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 goto page_not_uptodate;
1990
Linus Torvaldsef00e082009-06-16 15:31:25 -07001991 /*
1992 * Found the page and have a reference on it.
1993 * We must recheck i_size under page lock.
1994 */
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001995 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1996 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001997 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001998 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001999 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002000 }
2001
Nick Piggind0217ac2007-07-19 01:47:03 -07002002 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002003 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005no_cached_page:
2006 /*
2007 * We're only likely to ever get here if MADV_RANDOM is in
2008 * effect.
2009 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002010 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
2012 /*
2013 * The page we want has now been added to the page cache.
2014 * In the unlikely event that someone removed it in the
2015 * meantime, we'll just come back here and read it again.
2016 */
2017 if (error >= 0)
2018 goto retry_find;
2019
2020 /*
2021 * An error return from page_cache_read can result if the
2022 * system is low on memory, or a problem occurs while trying
2023 * to schedule I/O.
2024 */
2025 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002026 return VM_FAULT_OOM;
2027 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
2029page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 /*
2031 * Umm, take care of errors if the page isn't up-to-date.
2032 * Try to re-read it _once_. We do this synchronously,
2033 * because there really aren't any performance issues here
2034 * and we need to check for errors.
2035 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002037 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002038 if (!error) {
2039 wait_on_page_locked(page);
2040 if (!PageUptodate(page))
2041 error = -EIO;
2042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002044
2045 if (!error || error == AOP_TRUNCATED_PAGE)
2046 goto retry_find;
2047
2048 /* Things didn't work out. Return zero to tell the mm layer so. */
2049 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002050 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002051}
2052EXPORT_SYMBOL(filemap_fault);
2053
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002054void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2055{
2056 struct radix_tree_iter iter;
2057 void **slot;
2058 struct file *file = vma->vm_file;
2059 struct address_space *mapping = file->f_mapping;
2060 loff_t size;
2061 struct page *page;
2062 unsigned long address = (unsigned long) vmf->virtual_address;
2063 unsigned long addr;
2064 pte_t *pte;
2065
2066 rcu_read_lock();
2067 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2068 if (iter.index > vmf->max_pgoff)
2069 break;
2070repeat:
2071 page = radix_tree_deref_slot(slot);
2072 if (unlikely(!page))
2073 goto next;
2074 if (radix_tree_exception(page)) {
2075 if (radix_tree_deref_retry(page))
2076 break;
2077 else
2078 goto next;
2079 }
2080
2081 if (!page_cache_get_speculative(page))
2082 goto repeat;
2083
2084 /* Has the page moved? */
2085 if (unlikely(page != *slot)) {
2086 page_cache_release(page);
2087 goto repeat;
2088 }
2089
2090 if (!PageUptodate(page) ||
2091 PageReadahead(page) ||
2092 PageHWPoison(page))
2093 goto skip;
2094 if (!trylock_page(page))
2095 goto skip;
2096
2097 if (page->mapping != mapping || !PageUptodate(page))
2098 goto unlock;
2099
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002100 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2101 if (page->index >= size >> PAGE_CACHE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002102 goto unlock;
2103
2104 pte = vmf->pte + page->index - vmf->pgoff;
2105 if (!pte_none(*pte))
2106 goto unlock;
2107
2108 if (file->f_ra.mmap_miss > 0)
2109 file->f_ra.mmap_miss--;
2110 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2111 do_set_pte(vma, addr, page, pte, false, false);
2112 unlock_page(page);
2113 goto next;
2114unlock:
2115 unlock_page(page);
2116skip:
2117 page_cache_release(page);
2118next:
2119 if (iter.index == vmf->max_pgoff)
2120 break;
2121 }
2122 rcu_read_unlock();
2123}
2124EXPORT_SYMBOL(filemap_map_pages);
2125
Jan Kara4fcf1c62012-06-12 16:20:29 +02002126int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2127{
2128 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002129 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002130 int ret = VM_FAULT_LOCKED;
2131
Jan Kara14da9202012-06-12 16:20:37 +02002132 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002133 file_update_time(vma->vm_file);
2134 lock_page(page);
2135 if (page->mapping != inode->i_mapping) {
2136 unlock_page(page);
2137 ret = VM_FAULT_NOPAGE;
2138 goto out;
2139 }
Jan Kara14da9202012-06-12 16:20:37 +02002140 /*
2141 * We mark the page dirty already here so that when freeze is in
2142 * progress, we are guaranteed that writeback during freezing will
2143 * see the dirty page and writeprotect it again.
2144 */
2145 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002146 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002147out:
Jan Kara14da9202012-06-12 16:20:37 +02002148 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002149 return ret;
2150}
2151EXPORT_SYMBOL(filemap_page_mkwrite);
2152
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002153const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002154 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002155 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002156 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157};
2158
2159/* This is used for a general mmap of a disk file */
2160
2161int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2162{
2163 struct address_space *mapping = file->f_mapping;
2164
2165 if (!mapping->a_ops->readpage)
2166 return -ENOEXEC;
2167 file_accessed(file);
2168 vma->vm_ops = &generic_file_vm_ops;
2169 return 0;
2170}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
2172/*
2173 * This is for filesystems which do not implement ->writepage.
2174 */
2175int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2176{
2177 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2178 return -EINVAL;
2179 return generic_file_mmap(file, vma);
2180}
2181#else
2182int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2183{
2184 return -ENOSYS;
2185}
2186int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2187{
2188 return -ENOSYS;
2189}
2190#endif /* CONFIG_MMU */
2191
2192EXPORT_SYMBOL(generic_file_mmap);
2193EXPORT_SYMBOL(generic_file_readonly_mmap);
2194
Sasha Levin67f9fd92014-04-03 14:48:18 -07002195static struct page *wait_on_page_read(struct page *page)
2196{
2197 if (!IS_ERR(page)) {
2198 wait_on_page_locked(page);
2199 if (!PageUptodate(page)) {
2200 page_cache_release(page);
2201 page = ERR_PTR(-EIO);
2202 }
2203 }
2204 return page;
2205}
2206
Nick Piggin6fe69002007-05-06 14:49:04 -07002207static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002208 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002209 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002210 void *data,
2211 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
Nick Piggineb2be182007-10-16 01:24:57 -07002213 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 int err;
2215repeat:
2216 page = find_get_page(mapping, index);
2217 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002218 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002219 if (!page)
2220 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002221 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002222 if (unlikely(err)) {
2223 page_cache_release(page);
2224 if (err == -EEXIST)
2225 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 return ERR_PTR(err);
2228 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 err = filler(data, page);
2230 if (err < 0) {
2231 page_cache_release(page);
2232 page = ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002233 } else {
2234 page = wait_on_page_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 }
2236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 return page;
2238}
2239
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002240static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002241 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002242 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002243 void *data,
2244 gfp_t gfp)
2245
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246{
2247 struct page *page;
2248 int err;
2249
2250retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002251 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01002253 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 if (PageUptodate(page))
2255 goto out;
2256
2257 lock_page(page);
2258 if (!page->mapping) {
2259 unlock_page(page);
2260 page_cache_release(page);
2261 goto retry;
2262 }
2263 if (PageUptodate(page)) {
2264 unlock_page(page);
2265 goto out;
2266 }
2267 err = filler(data, page);
2268 if (err < 0) {
2269 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01002270 return ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002271 } else {
2272 page = wait_on_page_read(page);
2273 if (IS_ERR(page))
2274 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 }
David Howellsc855ff32007-05-09 13:42:20 +01002276out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002277 mark_page_accessed(page);
2278 return page;
2279}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002280
2281/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002282 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002283 * @mapping: the page's address_space
2284 * @index: the page index
2285 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002286 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002287 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002288 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002289 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002290 *
2291 * If the page does not get brought uptodate, return -EIO.
2292 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002293struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002294 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002295 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002296 void *data)
2297{
2298 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2299}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002300EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002301
2302/**
2303 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2304 * @mapping: the page's address_space
2305 * @index: the page index
2306 * @gfp: the page allocator flags to use if allocating
2307 *
2308 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002309 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002310 *
2311 * If the page does not get brought uptodate, return -EIO.
2312 */
2313struct page *read_cache_page_gfp(struct address_space *mapping,
2314 pgoff_t index,
2315 gfp_t gfp)
2316{
2317 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2318
Sasha Levin67f9fd92014-04-03 14:48:18 -07002319 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002320}
2321EXPORT_SYMBOL(read_cache_page_gfp);
2322
Nick Piggin2f718ff2007-10-16 01:24:59 -07002323/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 * Performs necessary checks before doing a write
2325 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002326 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 * Returns appropriate error code that caller should return or
2328 * zero in case that write should be allowed.
2329 */
Al Viro3309dd02015-04-09 12:55:47 -04002330inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331{
Al Viro3309dd02015-04-09 12:55:47 -04002332 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002334 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002335 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
Al Viro3309dd02015-04-09 12:55:47 -04002337 if (!iov_iter_count(from))
2338 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Al Viro0fa6b002015-04-04 04:05:48 -04002340 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002341 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002342 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
Al Viro3309dd02015-04-09 12:55:47 -04002344 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
Al Viro0fa6b002015-04-04 04:05:48 -04002346 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002347 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002348 send_sig(SIGXFSZ, current, 0);
2349 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 }
Al Viro3309dd02015-04-09 12:55:47 -04002351 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 }
2353
2354 /*
2355 * LFS rule
2356 */
Al Viro3309dd02015-04-09 12:55:47 -04002357 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002359 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002361 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 }
2363
2364 /*
2365 * Are we about to exceed the fs block limit ?
2366 *
2367 * If we have written data it becomes a short write. If we have
2368 * exceeded without writing data we send a signal and return EFBIG.
2369 * Linus frestrict idea will clean these up nicely..
2370 */
Al Viro3309dd02015-04-09 12:55:47 -04002371 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2372 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Al Viro3309dd02015-04-09 12:55:47 -04002374 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2375 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376}
2377EXPORT_SYMBOL(generic_write_checks);
2378
Nick Pigginafddba42007-10-16 01:25:01 -07002379int pagecache_write_begin(struct file *file, struct address_space *mapping,
2380 loff_t pos, unsigned len, unsigned flags,
2381 struct page **pagep, void **fsdata)
2382{
2383 const struct address_space_operations *aops = mapping->a_ops;
2384
Nick Piggin4e02ed42008-10-29 14:00:55 -07002385 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002386 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002387}
2388EXPORT_SYMBOL(pagecache_write_begin);
2389
2390int pagecache_write_end(struct file *file, struct address_space *mapping,
2391 loff_t pos, unsigned len, unsigned copied,
2392 struct page *page, void *fsdata)
2393{
2394 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002395
Nick Piggin4e02ed42008-10-29 14:00:55 -07002396 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002397}
2398EXPORT_SYMBOL(pagecache_write_end);
2399
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400ssize_t
Al Viro0c949332014-03-22 06:51:37 -04002401generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402{
2403 struct file *file = iocb->ki_filp;
2404 struct address_space *mapping = file->f_mapping;
2405 struct inode *inode = mapping->host;
2406 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002407 size_t write_len;
2408 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002409 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
Al Viro0c949332014-03-22 06:51:37 -04002411 write_len = iov_iter_count(from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002412 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002413
Nick Piggin48b47c52009-01-06 14:40:22 -08002414 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002415 if (written)
2416 goto out;
2417
2418 /*
2419 * After a write we want buffered reads to be sure to go to disk to get
2420 * the new data. We invalidate clean cached page from the region we're
2421 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002422 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002423 */
2424 if (mapping->nrpages) {
2425 written = invalidate_inode_pages2_range(mapping,
2426 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002427 /*
2428 * If a page can not be invalidated, return 0 to fall back
2429 * to buffered write.
2430 */
2431 if (written) {
2432 if (written == -EBUSY)
2433 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002434 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002435 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002436 }
2437
Al Viro26978b82014-03-10 14:08:45 -04002438 data = *from;
Omar Sandoval22c61862015-03-16 04:33:53 -07002439 written = mapping->a_ops->direct_IO(iocb, &data, pos);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002440
2441 /*
2442 * Finally, try again to invalidate clean pages which might have been
2443 * cached by non-direct readahead, or faulted in by get_user_pages()
2444 * if the source of the write was an mmap'ed region of the file
2445 * we're writing. Either one is a pretty crazy thing to do,
2446 * so we don't support it 100%. If this invalidation
2447 * fails, tough, the write still worked...
2448 */
2449 if (mapping->nrpages) {
2450 invalidate_inode_pages2_range(mapping,
2451 pos >> PAGE_CACHE_SHIFT, end);
2452 }
2453
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002455 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002456 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002457 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2458 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 mark_inode_dirty(inode);
2460 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002461 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002463out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 return written;
2465}
2466EXPORT_SYMBOL(generic_file_direct_write);
2467
Nick Piggineb2be182007-10-16 01:24:57 -07002468/*
2469 * Find or create a page at the given pagecache position. Return the locked
2470 * page. This function is specifically for buffered writes.
2471 */
Nick Piggin54566b22009-01-04 12:00:53 -08002472struct page *grab_cache_page_write_begin(struct address_space *mapping,
2473 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002474{
Nick Piggineb2be182007-10-16 01:24:57 -07002475 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07002476 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002477
Nick Piggin54566b22009-01-04 12:00:53 -08002478 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002479 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002480
Mel Gorman2457aec2014-06-04 16:10:31 -07002481 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002482 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002483 if (page)
2484 wait_for_stable_page(page);
2485
Nick Piggineb2be182007-10-16 01:24:57 -07002486 return page;
2487}
Nick Piggin54566b22009-01-04 12:00:53 -08002488EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002489
Al Viro3b93f912014-02-11 21:34:08 -05002490ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002491 struct iov_iter *i, loff_t pos)
2492{
2493 struct address_space *mapping = file->f_mapping;
2494 const struct address_space_operations *a_ops = mapping->a_ops;
2495 long status = 0;
2496 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002497 unsigned int flags = 0;
2498
2499 /*
2500 * Copies from kernel address space cannot fail (NFSD is a big user).
2501 */
Al Viro777eda22014-12-17 04:46:46 -05002502 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002503 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002504
2505 do {
2506 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002507 unsigned long offset; /* Offset into pagecache page */
2508 unsigned long bytes; /* Bytes to write to page */
2509 size_t copied; /* Bytes copied from user */
2510 void *fsdata;
2511
2512 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002513 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2514 iov_iter_count(i));
2515
2516again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002517 /*
2518 * Bring in the user page that we will copy from _first_.
2519 * Otherwise there's a nasty deadlock on copying from the
2520 * same page as we're writing to, without it being marked
2521 * up-to-date.
2522 *
2523 * Not only is this an optimisation, but it is also required
2524 * to check that the address is actually valid, when atomic
2525 * usercopies are used, below.
2526 */
2527 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2528 status = -EFAULT;
2529 break;
2530 }
2531
Jan Kara296291c2015-10-22 13:32:21 -07002532 if (fatal_signal_pending(current)) {
2533 status = -EINTR;
2534 break;
2535 }
2536
Nick Piggin674b8922007-10-16 01:25:03 -07002537 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002538 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002539 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002540 break;
2541
anfei zhou931e80e2010-02-02 13:44:02 -08002542 if (mapping_writably_mapped(mapping))
2543 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002544
Nick Pigginafddba42007-10-16 01:25:01 -07002545 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002546 flush_dcache_page(page);
2547
2548 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2549 page, fsdata);
2550 if (unlikely(status < 0))
2551 break;
2552 copied = status;
2553
2554 cond_resched();
2555
Nick Piggin124d3b72008-02-02 15:01:17 +01002556 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002557 if (unlikely(copied == 0)) {
2558 /*
2559 * If we were unable to copy any data at all, we must
2560 * fall back to a single segment length write.
2561 *
2562 * If we didn't fallback here, we could livelock
2563 * because not all segments in the iov can be copied at
2564 * once without a pagefault.
2565 */
2566 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2567 iov_iter_single_seg_count(i));
2568 goto again;
2569 }
Nick Pigginafddba42007-10-16 01:25:01 -07002570 pos += copied;
2571 written += copied;
2572
2573 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002574 } while (iov_iter_count(i));
2575
2576 return written ? written : status;
2577}
Al Viro3b93f912014-02-11 21:34:08 -05002578EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
Jan Karae4dd9de2009-08-17 18:10:06 +02002580/**
Al Viro81742022014-04-03 03:17:43 -04002581 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002582 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002583 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002584 *
2585 * This function does all the work needed for actually writing data to a
2586 * file. It does all basic checks, removes SUID from the file, updates
2587 * modification times and calls proper subroutines depending on whether we
2588 * do direct IO or a standard buffered write.
2589 *
2590 * It expects i_mutex to be grabbed unless we work on a block device or similar
2591 * object which does not need locking at all.
2592 *
2593 * This function does *not* take care of syncing data in case of O_SYNC write.
2594 * A caller has to handle it. This is mainly due to the fact that we want to
2595 * avoid syncing under i_mutex.
2596 */
Al Viro81742022014-04-03 03:17:43 -04002597ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598{
2599 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002600 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002602 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002604 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002607 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002608 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 if (err)
2610 goto out;
2611
Josef Bacikc3b2da32012-03-26 09:59:21 -04002612 err = file_update_time(file);
2613 if (err)
2614 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
Al Viro2ba48ce2015-04-09 13:52:01 -04002616 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002617 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002618
Al Viro0b8def92015-04-07 10:22:53 -04002619 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002620 /*
2621 * If the write stopped short of completing, fall back to
2622 * buffered writes. Some filesystems do this for writes to
2623 * holes, for example. For DAX files, a buffered write will
2624 * not succeed (even if it did, DAX does not handle dirty
2625 * page-cache pages correctly).
2626 */
Al Viro0b8def92015-04-07 10:22:53 -04002627 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002629
Al Viro0b8def92015-04-07 10:22:53 -04002630 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002631 /*
Al Viro3b93f912014-02-11 21:34:08 -05002632 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002633 * then we want to return the number of bytes which were
2634 * direct-written, or the error code if that was zero. Note
2635 * that this differs from normal direct-io semantics, which
2636 * will return -EFOO even if some bytes were written.
2637 */
Al Viro60bb4522014-08-08 12:39:16 -04002638 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002639 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002640 goto out;
2641 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002642 /*
2643 * We need to ensure that the page cache pages are written to
2644 * disk and invalidated to preserve the expected O_DIRECT
2645 * semantics.
2646 */
Al Viro3b93f912014-02-11 21:34:08 -05002647 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002648 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002649 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002650 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002651 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002652 invalidate_mapping_pages(mapping,
2653 pos >> PAGE_CACHE_SHIFT,
2654 endbyte >> PAGE_CACHE_SHIFT);
2655 } else {
2656 /*
2657 * We don't know how much we wrote, so just return
2658 * the number of bytes which were direct-written
2659 */
2660 }
2661 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002662 written = generic_perform_write(file, from, iocb->ki_pos);
2663 if (likely(written > 0))
2664 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002665 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666out:
2667 current->backing_dev_info = NULL;
2668 return written ? written : err;
2669}
Al Viro81742022014-04-03 03:17:43 -04002670EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
Jan Karae4dd9de2009-08-17 18:10:06 +02002672/**
Al Viro81742022014-04-03 03:17:43 -04002673 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002674 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002675 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002676 *
Al Viro81742022014-04-03 03:17:43 -04002677 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002678 * filesystems. It takes care of syncing the file in case of O_SYNC file
2679 * and acquires i_mutex as needed.
2680 */
Al Viro81742022014-04-03 03:17:43 -04002681ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682{
2683 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002684 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002687 mutex_lock(&inode->i_mutex);
Al Viro3309dd02015-04-09 12:55:47 -04002688 ret = generic_write_checks(iocb, from);
2689 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002690 ret = __generic_file_write_iter(iocb, from);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002691 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Christoph Hellwig02afc272013-09-04 15:04:40 +02002693 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 ssize_t err;
2695
Al Virod311d792014-02-09 15:18:09 -05002696 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2697 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 ret = err;
2699 }
2700 return ret;
2701}
Al Viro81742022014-04-03 03:17:43 -04002702EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
David Howellscf9a2ae2006-08-29 19:05:54 +01002704/**
2705 * try_to_release_page() - release old fs-specific metadata on a page
2706 *
2707 * @page: the page which the kernel is trying to free
2708 * @gfp_mask: memory allocation flags (and I/O mode)
2709 *
2710 * The address_space is to try to release any data against the page
2711 * (presumably at page->private). If the release was successful, return `1'.
2712 * Otherwise return zero.
2713 *
David Howells266cf652009-04-03 16:42:36 +01002714 * This may also be called if PG_fscache is set on a page, indicating that the
2715 * page is known to the local caching routines.
2716 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002717 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08002718 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002719 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002720 */
2721int try_to_release_page(struct page *page, gfp_t gfp_mask)
2722{
2723 struct address_space * const mapping = page->mapping;
2724
2725 BUG_ON(!PageLocked(page));
2726 if (PageWriteback(page))
2727 return 0;
2728
2729 if (mapping && mapping->a_ops->releasepage)
2730 return mapping->a_ops->releasepage(page, gfp_mask);
2731 return try_to_free_buffers(page);
2732}
2733
2734EXPORT_SYMBOL(try_to_release_page);