blob: a8c69c8c0a90a1e50cfdc822ce2cbc3854e0c8e8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080014#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Johannes Weiner00501b52014-08-08 14:19:20 -070034#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060036#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070037#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080038#include "internal.h"
39
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070040#define CREATE_TRACE_POINTS
41#include <trace/events/filemap.h>
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * FIXME: remove all knowledge of the buffer layer from the core VM
45 */
Jan Kara148f9482009-08-17 19:52:36 +020046#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <asm/mman.h>
49
50/*
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * though.
53 *
54 * Shared mappings now work. 15.8.1995 Bruno.
55 *
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 *
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 */
61
62/*
63 * Lock ordering:
64 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080065 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070067 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080070 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080071 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 *
73 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080074 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070075 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
Al Viroccad2362014-02-11 22:36:48 -050081 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070082 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060084 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110085 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * ->mapping->tree_lock (__sync_single_inode)
87 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080088 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080099 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700104 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800109 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700110 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
112
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700113static void page_cache_tree_delete(struct address_space *mapping,
114 struct page *page, void *shadow)
115{
Johannes Weiner449dd692014-04-03 14:47:56 -0700116 struct radix_tree_node *node;
117 unsigned long index;
118 unsigned int offset;
119 unsigned int tag;
120 void **slot;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700121
Johannes Weiner449dd692014-04-03 14:47:56 -0700122 VM_BUG_ON(!PageLocked(page));
123
124 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
125
126 if (shadow) {
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800127 mapping->nrexceptional++;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700128 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800129 * Make sure the nrexceptional update is committed before
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700130 * the nrpages update so that final truncate racing
131 * with reclaim does not see both counters 0 at the
132 * same time and miss a shadow entry.
133 */
134 smp_wmb();
Johannes Weiner449dd692014-04-03 14:47:56 -0700135 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700136 mapping->nrpages--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700137
138 if (!node) {
139 /* Clear direct pointer tags in root node */
140 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
141 radix_tree_replace_slot(slot, shadow);
142 return;
143 }
144
145 /* Clear tree tags for the removed page */
146 index = page->index;
147 offset = index & RADIX_TREE_MAP_MASK;
148 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
149 if (test_bit(offset, node->tags[tag]))
150 radix_tree_tag_clear(&mapping->page_tree, index, tag);
151 }
152
153 /* Delete page, swap shadow entry */
154 radix_tree_replace_slot(slot, shadow);
155 workingset_node_pages_dec(node);
156 if (shadow)
157 workingset_node_shadows_inc(node);
158 else
159 if (__radix_tree_delete_node(&mapping->page_tree, node))
160 return;
161
162 /*
163 * Track node that only contains shadow entries.
164 *
165 * Avoid acquiring the list_lru lock if already tracked. The
166 * list_empty() test is safe as node->private_list is
167 * protected by mapping->tree_lock.
168 */
169 if (!workingset_node_pages(node) &&
170 list_empty(&node->private_list)) {
171 node->private_data = mapping;
172 list_lru_add(&workingset_shadow_nodes, &node->private_list);
173 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700174}
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176/*
Minchan Kime64a7822011-03-22 16:32:44 -0700177 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 * sure the page is locked and that nobody else uses it - or that usage
Johannes Weinerfdf1cdb2016-03-15 14:57:25 -0700179 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700181void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
183 struct address_space *mapping = page->mapping;
184
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700185 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600186 /*
187 * if we're uptodate, flush out into the cleancache, otherwise
188 * invalidate any existing cleancache entries. We can't leave
189 * stale data around in the cleancache once our page is gone
190 */
191 if (PageUptodate(page) && PageMappedToDisk(page))
192 cleancache_put_page(page);
193 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400194 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600195
Hugh Dickins06b241f2016-03-09 14:08:07 -0800196 VM_BUG_ON_PAGE(page_mapped(page), page);
197 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
198 int mapcount;
199
200 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
201 current->comm, page_to_pfn(page));
202 dump_page(page, "still mapped when deleted");
203 dump_stack();
204 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
205
206 mapcount = page_mapcount(page);
207 if (mapping_exiting(mapping) &&
208 page_count(page) >= mapcount + 2) {
209 /*
210 * All vmas have already been torn down, so it's
211 * a good bet that actually the page is unmapped,
212 * and we'd prefer not to leak it: if we're wrong,
213 * some other bad page check should catch it later.
214 */
215 page_mapcount_reset(page);
216 atomic_sub(mapcount, &page->_count);
217 }
218 }
219
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700220 page_cache_tree_delete(mapping, page, shadow);
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700223 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700224
Michal Hocko4165b9b2015-06-24 16:57:24 -0700225 /* hugetlb pages do not participate in page cache accounting. */
226 if (!PageHuge(page))
227 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700228 if (PageSwapBacked(page))
229 __dec_zone_page_state(page, NR_SHMEM);
Linus Torvalds3a692792007-12-19 14:05:13 -0800230
231 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700232 * At this point page must be either written or cleaned by truncate.
233 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800234 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700235 * This fixes dirty accounting after removing the page entirely but
236 * leaves PageDirty set: it has no effect for truncated page and
237 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800238 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700239 if (WARN_ON_ONCE(PageDirty(page)))
Johannes Weiner62cccb82016-03-15 14:57:22 -0700240 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Minchan Kim702cfbf2011-03-22 16:32:43 -0700243/**
244 * delete_from_page_cache - delete page from page cache
245 * @page: the page which the kernel is trying to remove from page cache
246 *
247 * This must be called only on pages that have been verified to be in the page
248 * cache and locked. It will never put the page into the free list, the caller
249 * has a reference on the page.
250 */
251void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
253 struct address_space *mapping = page->mapping;
Greg Thelenc4843a72015-05-22 17:13:16 -0400254 unsigned long flags;
255
Linus Torvalds6072d132010-12-01 13:35:19 -0500256 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Matt Mackallcd7619d2005-05-01 08:59:01 -0700258 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Linus Torvalds6072d132010-12-01 13:35:19 -0500260 freepage = mapping->a_ops->freepage;
Greg Thelenc4843a72015-05-22 17:13:16 -0400261
Greg Thelenc4843a72015-05-22 17:13:16 -0400262 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700263 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400264 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500265
266 if (freepage)
267 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700268 page_cache_release(page);
269}
270EXPORT_SYMBOL(delete_from_page_cache);
271
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700272static int filemap_check_errors(struct address_space *mapping)
273{
274 int ret = 0;
275 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700276 if (test_bit(AS_ENOSPC, &mapping->flags) &&
277 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700278 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700279 if (test_bit(AS_EIO, &mapping->flags) &&
280 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700281 ret = -EIO;
282 return ret;
283}
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700286 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700287 * @mapping: address space structure to write
288 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800289 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700290 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700292 * Start writeback against all of a mapping's dirty pages that lie
293 * within the byte offsets <start, end> inclusive.
294 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700296 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 * these two operations is that if a dirty page/buffer is encountered, it must
298 * be waited upon, and not just skipped over.
299 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800300int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
301 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 int ret;
304 struct writeback_control wbc = {
305 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800306 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700307 .range_start = start,
308 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 };
310
311 if (!mapping_cap_writeback_dirty(mapping))
312 return 0;
313
Tejun Heob16b1de2015-06-02 08:39:48 -0600314 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600316 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 return ret;
318}
319
320static inline int __filemap_fdatawrite(struct address_space *mapping,
321 int sync_mode)
322{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700323 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
326int filemap_fdatawrite(struct address_space *mapping)
327{
328 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
329}
330EXPORT_SYMBOL(filemap_fdatawrite);
331
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400332int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800333 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
336}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400337EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Randy Dunlap485bb992006-06-23 02:03:49 -0700339/**
340 * filemap_flush - mostly a non-blocking flush
341 * @mapping: target address_space
342 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 * This is a mostly non-blocking flush. Not suitable for data-integrity
344 * purposes - I/O may not be started against all dirty pages.
345 */
346int filemap_flush(struct address_space *mapping)
347{
348 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
349}
350EXPORT_SYMBOL(filemap_flush);
351
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800352static int __filemap_fdatawait_range(struct address_space *mapping,
353 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200355 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
356 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 struct pagevec pvec;
358 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800359 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200361 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700362 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 while ((index <= end) &&
366 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
367 PAGECACHE_TAG_WRITEBACK,
368 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
369 unsigned i;
370
371 for (i = 0; i < nr_pages; i++) {
372 struct page *page = pvec.pages[i];
373
374 /* until radix tree lookup accepts end_index */
375 if (page->index > end)
376 continue;
377
378 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800379 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 ret = -EIO;
381 }
382 pagevec_release(&pvec);
383 cond_resched();
384 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700385out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800386 return ret;
387}
388
389/**
390 * filemap_fdatawait_range - wait for writeback to complete
391 * @mapping: address space structure to wait for
392 * @start_byte: offset in bytes where the range starts
393 * @end_byte: offset in bytes where the range ends (inclusive)
394 *
395 * Walk the list of under-writeback pages of the given address space
396 * in the given range and wait for all of them. Check error status of
397 * the address space and return it.
398 *
399 * Since the error status of the address space is cleared by this function,
400 * callers are responsible for checking the return value and handling and/or
401 * reporting the error.
402 */
403int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
404 loff_t end_byte)
405{
406 int ret, ret2;
407
408 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700409 ret2 = filemap_check_errors(mapping);
410 if (!ret)
411 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 return ret;
414}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200415EXPORT_SYMBOL(filemap_fdatawait_range);
416
417/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800418 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
419 * @mapping: address space structure to wait for
420 *
421 * Walk the list of under-writeback pages of the given address space
422 * and wait for all of them. Unlike filemap_fdatawait(), this function
423 * does not clear error status of the address space.
424 *
425 * Use this function if callers don't handle errors themselves. Expected
426 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
427 * fsfreeze(8)
428 */
429void filemap_fdatawait_keep_errors(struct address_space *mapping)
430{
431 loff_t i_size = i_size_read(mapping->host);
432
433 if (i_size == 0)
434 return;
435
436 __filemap_fdatawait_range(mapping, 0, i_size - 1);
437}
438
439/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700440 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700442 *
443 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800444 * and wait for all of them. Check error status of the address space
445 * and return it.
446 *
447 * Since the error status of the address space is cleared by this function,
448 * callers are responsible for checking the return value and handling and/or
449 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 */
451int filemap_fdatawait(struct address_space *mapping)
452{
453 loff_t i_size = i_size_read(mapping->host);
454
455 if (i_size == 0)
456 return 0;
457
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200458 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
460EXPORT_SYMBOL(filemap_fdatawait);
461
462int filemap_write_and_wait(struct address_space *mapping)
463{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800464 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800466 if ((!dax_mapping(mapping) && mapping->nrpages) ||
467 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800468 err = filemap_fdatawrite(mapping);
469 /*
470 * Even if the above returned error, the pages may be
471 * written partially (e.g. -ENOSPC), so we wait for it.
472 * But the -EIO is special case, it may indicate the worst
473 * thing (e.g. bug) happened, so we avoid waiting for it.
474 */
475 if (err != -EIO) {
476 int err2 = filemap_fdatawait(mapping);
477 if (!err)
478 err = err2;
479 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700480 } else {
481 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800483 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800485EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Randy Dunlap485bb992006-06-23 02:03:49 -0700487/**
488 * filemap_write_and_wait_range - write out & wait on a file range
489 * @mapping: the address_space for the pages
490 * @lstart: offset in bytes where the range starts
491 * @lend: offset in bytes where the range ends (inclusive)
492 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800493 * Write out and wait upon file offsets lstart->lend, inclusive.
494 *
495 * Note that `lend' is inclusive (describes the last byte to be written) so
496 * that this function can be used to write to the very end-of-file (end = -1).
497 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498int filemap_write_and_wait_range(struct address_space *mapping,
499 loff_t lstart, loff_t lend)
500{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800501 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800503 if ((!dax_mapping(mapping) && mapping->nrpages) ||
504 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800505 err = __filemap_fdatawrite_range(mapping, lstart, lend,
506 WB_SYNC_ALL);
507 /* See comment of filemap_write_and_wait() */
508 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200509 int err2 = filemap_fdatawait_range(mapping,
510 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800511 if (!err)
512 err = err2;
513 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700514 } else {
515 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800517 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518}
Chris Masonf6995582009-04-15 13:22:37 -0400519EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Randy Dunlap485bb992006-06-23 02:03:49 -0700521/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700522 * replace_page_cache_page - replace a pagecache page with a new one
523 * @old: page to be replaced
524 * @new: page to replace with
525 * @gfp_mask: allocation mode
526 *
527 * This function replaces a page in the pagecache with a new one. On
528 * success it acquires the pagecache reference for the new page and
529 * drops it for the old page. Both the old and new pages must be
530 * locked. This function does not add the new page to the LRU, the
531 * caller must do that.
532 *
533 * The remove + add is atomic. The only way this function can fail is
534 * memory allocation failure.
535 */
536int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
537{
538 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700539
Sasha Levin309381fea2014-01-23 15:52:54 -0800540 VM_BUG_ON_PAGE(!PageLocked(old), old);
541 VM_BUG_ON_PAGE(!PageLocked(new), new);
542 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700543
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700544 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
545 if (!error) {
546 struct address_space *mapping = old->mapping;
547 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400548 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700549
550 pgoff_t offset = old->index;
551 freepage = mapping->a_ops->freepage;
552
553 page_cache_get(new);
554 new->mapping = mapping;
555 new->index = offset;
556
Greg Thelenc4843a72015-05-22 17:13:16 -0400557 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700558 __delete_from_page_cache(old, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700559 error = radix_tree_insert(&mapping->page_tree, offset, new);
560 BUG_ON(error);
561 mapping->nrpages++;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700562
563 /*
564 * hugetlb pages do not participate in page cache accounting.
565 */
566 if (!PageHuge(new))
567 __inc_zone_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700568 if (PageSwapBacked(new))
569 __inc_zone_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400570 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700571 mem_cgroup_migrate(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700572 radix_tree_preload_end();
573 if (freepage)
574 freepage(old);
575 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700576 }
577
578 return error;
579}
580EXPORT_SYMBOL_GPL(replace_page_cache_page);
581
Johannes Weiner0cd61442014-04-03 14:47:46 -0700582static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700583 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700584{
Johannes Weiner449dd692014-04-03 14:47:56 -0700585 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700586 void **slot;
587 int error;
588
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700589 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
Johannes Weiner449dd692014-04-03 14:47:56 -0700590 &node, &slot);
591 if (error)
592 return error;
593 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700594 void *p;
595
596 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
597 if (!radix_tree_exceptional_entry(p))
598 return -EEXIST;
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800599
600 if (WARN_ON(dax_mapping(mapping)))
601 return -EINVAL;
602
Johannes Weinera5289102014-04-03 14:47:51 -0700603 if (shadowp)
604 *shadowp = p;
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800605 mapping->nrexceptional--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700606 if (node)
607 workingset_node_shadows_dec(node);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700608 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700609 radix_tree_replace_slot(slot, page);
610 mapping->nrpages++;
611 if (node) {
612 workingset_node_pages_inc(node);
613 /*
614 * Don't track node that contains actual pages.
615 *
616 * Avoid acquiring the list_lru lock if already
617 * untracked. The list_empty() test is safe as
618 * node->private_list is protected by
619 * mapping->tree_lock.
620 */
621 if (!list_empty(&node->private_list))
622 list_lru_del(&workingset_shadow_nodes,
623 &node->private_list);
624 }
625 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700626}
627
Johannes Weinera5289102014-04-03 14:47:51 -0700628static int __add_to_page_cache_locked(struct page *page,
629 struct address_space *mapping,
630 pgoff_t offset, gfp_t gfp_mask,
631 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Johannes Weiner00501b52014-08-08 14:19:20 -0700633 int huge = PageHuge(page);
634 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700635 int error;
636
Sasha Levin309381fea2014-01-23 15:52:54 -0800637 VM_BUG_ON_PAGE(!PageLocked(page), page);
638 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700639
Johannes Weiner00501b52014-08-08 14:19:20 -0700640 if (!huge) {
641 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800642 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700643 if (error)
644 return error;
645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Jan Kara5e4c0d972013-09-11 14:26:05 -0700647 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700648 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700649 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800650 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700651 return error;
652 }
653
654 page_cache_get(page);
655 page->mapping = mapping;
656 page->index = offset;
657
658 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700659 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700660 radix_tree_preload_end();
661 if (unlikely(error))
662 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700663
664 /* hugetlb pages do not participate in page cache accounting. */
665 if (!huge)
666 __inc_zone_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700667 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700668 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800669 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700670 trace_mm_filemap_add_to_page_cache(page);
671 return 0;
672err_insert:
673 page->mapping = NULL;
674 /* Leave page->index set: truncation relies upon it */
675 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700676 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800677 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700678 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 return error;
680}
Johannes Weinera5289102014-04-03 14:47:51 -0700681
682/**
683 * add_to_page_cache_locked - add a locked page to the pagecache
684 * @page: page to add
685 * @mapping: the page's address_space
686 * @offset: page index
687 * @gfp_mask: page allocation mode
688 *
689 * This function is used to add a page to the pagecache. It must be locked.
690 * This function does not add the page to the LRU. The caller must do that.
691 */
692int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
693 pgoff_t offset, gfp_t gfp_mask)
694{
695 return __add_to_page_cache_locked(page, mapping, offset,
696 gfp_mask, NULL);
697}
Nick Piggine2867812008-07-25 19:45:30 -0700698EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
700int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400701 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
Johannes Weinera5289102014-04-03 14:47:51 -0700703 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700704 int ret;
705
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800706 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700707 ret = __add_to_page_cache_locked(page, mapping, offset,
708 gfp_mask, &shadow);
709 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800710 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700711 else {
712 /*
713 * The page might have been evicted from cache only
714 * recently, in which case it should be activated like
715 * any other repeatedly accessed page.
716 */
717 if (shadow && workingset_refault(shadow)) {
718 SetPageActive(page);
719 workingset_activation(page);
720 } else
721 ClearPageActive(page);
722 lru_cache_add(page);
723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 return ret;
725}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300726EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Paul Jackson44110fe2006-03-24 03:16:04 -0800728#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700729struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800730{
Miao Xiec0ff7452010-05-24 14:32:08 -0700731 int n;
732 struct page *page;
733
Paul Jackson44110fe2006-03-24 03:16:04 -0800734 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700735 unsigned int cpuset_mems_cookie;
736 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700737 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700738 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700739 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700740 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700741
Miao Xiec0ff7452010-05-24 14:32:08 -0700742 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800743 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700744 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800745}
Nick Piggin2ae88142006-10-28 10:38:23 -0700746EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800747#endif
748
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749/*
750 * In order to wait for pages to become available there must be
751 * waitqueues associated with pages. By using a hash table of
752 * waitqueues where the bucket discipline is to maintain all
753 * waiters on the same queue and wake all when any of the pages
754 * become available, and for the woken contexts to check to be
755 * sure the appropriate page became available, this saves space
756 * at a cost of "thundering herd" phenomena during rare hash
757 * collisions.
758 */
NeilBrowna4796e32014-09-24 11:28:32 +1000759wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
761 const struct zone *zone = page_zone(page);
762
763 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
764}
NeilBrowna4796e32014-09-24 11:28:32 +1000765EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
Harvey Harrison920c7a52008-02-04 22:29:26 -0800767void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
769 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
770
771 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000772 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 TASK_UNINTERRUPTIBLE);
774}
775EXPORT_SYMBOL(wait_on_page_bit);
776
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700777int wait_on_page_bit_killable(struct page *page, int bit_nr)
778{
779 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
780
781 if (!test_bit(bit_nr, &page->flags))
782 return 0;
783
784 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000785 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700786}
787
NeilBrowncbbce822014-09-25 13:55:19 +1000788int wait_on_page_bit_killable_timeout(struct page *page,
789 int bit_nr, unsigned long timeout)
790{
791 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
792
793 wait.key.timeout = jiffies + timeout;
794 if (!test_bit(bit_nr, &page->flags))
795 return 0;
796 return __wait_on_bit(page_waitqueue(page), &wait,
797 bit_wait_io_timeout, TASK_KILLABLE);
798}
799EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
800
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100802 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700803 * @page: Page defining the wait queue of interest
804 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100805 *
806 * Add an arbitrary @waiter to the wait queue for the nominated @page.
807 */
808void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
809{
810 wait_queue_head_t *q = page_waitqueue(page);
811 unsigned long flags;
812
813 spin_lock_irqsave(&q->lock, flags);
814 __add_wait_queue(q, waiter);
815 spin_unlock_irqrestore(&q->lock, flags);
816}
817EXPORT_SYMBOL_GPL(add_page_wait_queue);
818
819/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700820 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 * @page: the page
822 *
823 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
824 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900825 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
827 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700828 * The mb is necessary to enforce ordering between the clear_bit and the read
829 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800831void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800833 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800834 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700835 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100836 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 wake_up_page(page, PG_locked);
838}
839EXPORT_SYMBOL(unlock_page);
840
Randy Dunlap485bb992006-06-23 02:03:49 -0700841/**
842 * end_page_writeback - end writeback against a page
843 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 */
845void end_page_writeback(struct page *page)
846{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700847 /*
848 * TestClearPageReclaim could be used here but it is an atomic
849 * operation and overkill in this particular case. Failing to
850 * shuffle a page marked for immediate reclaim is too mild to
851 * justify taking an atomic operation penalty at the end of
852 * ever page writeback.
853 */
854 if (PageReclaim(page)) {
855 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700856 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700857 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700858
859 if (!test_clear_page_writeback(page))
860 BUG();
861
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100862 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 wake_up_page(page, PG_writeback);
864}
865EXPORT_SYMBOL(end_page_writeback);
866
Matthew Wilcox57d99842014-06-04 16:07:45 -0700867/*
868 * After completing I/O on a page, call this routine to update the page
869 * flags appropriately
870 */
871void page_endio(struct page *page, int rw, int err)
872{
873 if (rw == READ) {
874 if (!err) {
875 SetPageUptodate(page);
876 } else {
877 ClearPageUptodate(page);
878 SetPageError(page);
879 }
880 unlock_page(page);
881 } else { /* rw == WRITE */
882 if (err) {
883 SetPageError(page);
884 if (page->mapping)
885 mapping_set_error(page->mapping, err);
886 }
887 end_page_writeback(page);
888 }
889}
890EXPORT_SYMBOL_GPL(page_endio);
891
Randy Dunlap485bb992006-06-23 02:03:49 -0700892/**
893 * __lock_page - get a lock on the page, assuming we need to sleep to get it
894 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800896void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800898 struct page *page_head = compound_head(page);
899 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800901 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 TASK_UNINTERRUPTIBLE);
903}
904EXPORT_SYMBOL(__lock_page);
905
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800906int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500907{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800908 struct page *page_head = compound_head(page);
909 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500910
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800911 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000912 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500913}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300914EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500915
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700916/*
917 * Return values:
918 * 1 - page is locked; mmap_sem is still held.
919 * 0 - page is not locked.
920 * mmap_sem has been released (up_read()), unless flags had both
921 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
922 * which case mmap_sem is still held.
923 *
924 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
925 * with the page locked and the mmap_sem unperturbed.
926 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700927int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
928 unsigned int flags)
929{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700930 if (flags & FAULT_FLAG_ALLOW_RETRY) {
931 /*
932 * CAUTION! In this case, mmap_sem is not released
933 * even though return 0.
934 */
935 if (flags & FAULT_FLAG_RETRY_NOWAIT)
936 return 0;
937
938 up_read(&mm->mmap_sem);
939 if (flags & FAULT_FLAG_KILLABLE)
940 wait_on_page_locked_killable(page);
941 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700942 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700943 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700944 } else {
945 if (flags & FAULT_FLAG_KILLABLE) {
946 int ret;
947
948 ret = __lock_page_killable(page);
949 if (ret) {
950 up_read(&mm->mmap_sem);
951 return 0;
952 }
953 } else
954 __lock_page(page);
955 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700956 }
957}
958
Randy Dunlap485bb992006-06-23 02:03:49 -0700959/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700960 * page_cache_next_hole - find the next hole (not-present entry)
961 * @mapping: mapping
962 * @index: index
963 * @max_scan: maximum range to search
964 *
965 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
966 * lowest indexed hole.
967 *
968 * Returns: the index of the hole if found, otherwise returns an index
969 * outside of the set specified (in which case 'return - index >=
970 * max_scan' will be true). In rare cases of index wrap-around, 0 will
971 * be returned.
972 *
973 * page_cache_next_hole may be called under rcu_read_lock. However,
974 * like radix_tree_gang_lookup, this will not atomically search a
975 * snapshot of the tree at a single point in time. For example, if a
976 * hole is created at index 5, then subsequently a hole is created at
977 * index 10, page_cache_next_hole covering both indexes may return 10
978 * if called under rcu_read_lock.
979 */
980pgoff_t page_cache_next_hole(struct address_space *mapping,
981 pgoff_t index, unsigned long max_scan)
982{
983 unsigned long i;
984
985 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700986 struct page *page;
987
988 page = radix_tree_lookup(&mapping->page_tree, index);
989 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700990 break;
991 index++;
992 if (index == 0)
993 break;
994 }
995
996 return index;
997}
998EXPORT_SYMBOL(page_cache_next_hole);
999
1000/**
1001 * page_cache_prev_hole - find the prev hole (not-present entry)
1002 * @mapping: mapping
1003 * @index: index
1004 * @max_scan: maximum range to search
1005 *
1006 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1007 * the first hole.
1008 *
1009 * Returns: the index of the hole if found, otherwise returns an index
1010 * outside of the set specified (in which case 'index - return >=
1011 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1012 * will be returned.
1013 *
1014 * page_cache_prev_hole may be called under rcu_read_lock. However,
1015 * like radix_tree_gang_lookup, this will not atomically search a
1016 * snapshot of the tree at a single point in time. For example, if a
1017 * hole is created at index 10, then subsequently a hole is created at
1018 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1019 * called under rcu_read_lock.
1020 */
1021pgoff_t page_cache_prev_hole(struct address_space *mapping,
1022 pgoff_t index, unsigned long max_scan)
1023{
1024 unsigned long i;
1025
1026 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001027 struct page *page;
1028
1029 page = radix_tree_lookup(&mapping->page_tree, index);
1030 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001031 break;
1032 index--;
1033 if (index == ULONG_MAX)
1034 break;
1035 }
1036
1037 return index;
1038}
1039EXPORT_SYMBOL(page_cache_prev_hole);
1040
1041/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001042 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001043 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001044 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001045 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001046 * Looks up the page cache slot at @mapping & @offset. If there is a
1047 * page cache page, it is returned with an increased refcount.
1048 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001049 * If the slot holds a shadow entry of a previously evicted page, or a
1050 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001051 *
1052 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001054struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055{
Nick Piggina60637c2008-07-25 19:45:31 -07001056 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 struct page *page;
1058
Nick Piggina60637c2008-07-25 19:45:31 -07001059 rcu_read_lock();
1060repeat:
1061 page = NULL;
1062 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1063 if (pagep) {
1064 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001065 if (unlikely(!page))
1066 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001067 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001068 if (radix_tree_deref_retry(page))
1069 goto repeat;
1070 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001071 * A shadow entry of a recently evicted page,
1072 * or a swap entry from shmem/tmpfs. Return
1073 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001074 */
1075 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001076 }
Nick Piggina60637c2008-07-25 19:45:31 -07001077 if (!page_cache_get_speculative(page))
1078 goto repeat;
1079
1080 /*
1081 * Has the page moved?
1082 * This is part of the lockless pagecache protocol. See
1083 * include/linux/pagemap.h for details.
1084 */
1085 if (unlikely(page != *pagep)) {
1086 page_cache_release(page);
1087 goto repeat;
1088 }
1089 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001090out:
Nick Piggina60637c2008-07-25 19:45:31 -07001091 rcu_read_unlock();
1092
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 return page;
1094}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001095EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Randy Dunlap485bb992006-06-23 02:03:49 -07001097/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001098 * find_lock_entry - locate, pin and lock a page cache entry
1099 * @mapping: the address_space to search
1100 * @offset: the page cache index
1101 *
1102 * Looks up the page cache slot at @mapping & @offset. If there is a
1103 * page cache page, it is returned locked and with an increased
1104 * refcount.
1105 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001106 * If the slot holds a shadow entry of a previously evicted page, or a
1107 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001108 *
1109 * Otherwise, %NULL is returned.
1110 *
1111 * find_lock_entry() may sleep.
1112 */
1113struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114{
1115 struct page *page;
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001118 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001119 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001120 lock_page(page);
1121 /* Has the page been truncated? */
1122 if (unlikely(page->mapping != mapping)) {
1123 unlock_page(page);
1124 page_cache_release(page);
1125 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001127 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 return page;
1130}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001131EXPORT_SYMBOL(find_lock_entry);
1132
1133/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001134 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001135 * @mapping: the address_space to search
1136 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001137 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001138 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001139 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001140 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001141 *
Randy Dunlap75325182014-07-30 16:08:37 -07001142 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001143 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001144 * FGP_ACCESSED: the page will be marked accessed
1145 * FGP_LOCK: Page is return locked
1146 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001147 * @gfp_mask and added to the page cache and the VM's LRU
1148 * list. The page is returned locked and with an increased
1149 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001150 *
1151 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1152 * if the GFP flags specified for FGP_CREAT are atomic.
1153 *
1154 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001155 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001156struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001157 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
Nick Piggineb2be182007-10-16 01:24:57 -07001159 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001162 page = find_get_entry(mapping, offset);
1163 if (radix_tree_exceptional_entry(page))
1164 page = NULL;
1165 if (!page)
1166 goto no_page;
1167
1168 if (fgp_flags & FGP_LOCK) {
1169 if (fgp_flags & FGP_NOWAIT) {
1170 if (!trylock_page(page)) {
1171 page_cache_release(page);
1172 return NULL;
1173 }
1174 } else {
1175 lock_page(page);
1176 }
1177
1178 /* Has the page been truncated? */
1179 if (unlikely(page->mapping != mapping)) {
1180 unlock_page(page);
1181 page_cache_release(page);
1182 goto repeat;
1183 }
1184 VM_BUG_ON_PAGE(page->index != offset, page);
1185 }
1186
1187 if (page && (fgp_flags & FGP_ACCESSED))
1188 mark_page_accessed(page);
1189
1190no_page:
1191 if (!page && (fgp_flags & FGP_CREAT)) {
1192 int err;
1193 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001194 gfp_mask |= __GFP_WRITE;
1195 if (fgp_flags & FGP_NOFS)
1196 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001197
Michal Hocko45f87de2014-12-29 20:30:35 +01001198 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001199 if (!page)
1200 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001201
1202 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1203 fgp_flags |= FGP_LOCK;
1204
Hugh Dickinseb39d612014-08-06 16:06:43 -07001205 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001206 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001207 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001208
Michal Hocko45f87de2014-12-29 20:30:35 +01001209 err = add_to_page_cache_lru(page, mapping, offset,
1210 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001211 if (unlikely(err)) {
1212 page_cache_release(page);
1213 page = NULL;
1214 if (err == -EEXIST)
1215 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 return page;
1220}
Mel Gorman2457aec2014-06-04 16:10:31 -07001221EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
1223/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001224 * find_get_entries - gang pagecache lookup
1225 * @mapping: The address_space to search
1226 * @start: The starting page cache index
1227 * @nr_entries: The maximum number of entries
1228 * @entries: Where the resulting entries are placed
1229 * @indices: The cache indices corresponding to the entries in @entries
1230 *
1231 * find_get_entries() will search for and return a group of up to
1232 * @nr_entries entries in the mapping. The entries are placed at
1233 * @entries. find_get_entries() takes a reference against any actual
1234 * pages it returns.
1235 *
1236 * The search returns a group of mapping-contiguous page cache entries
1237 * with ascending indexes. There may be holes in the indices due to
1238 * not-present pages.
1239 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001240 * Any shadow entries of evicted pages, or swap entries from
1241 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001242 *
1243 * find_get_entries() returns the number of pages and shadow entries
1244 * which were found.
1245 */
1246unsigned find_get_entries(struct address_space *mapping,
1247 pgoff_t start, unsigned int nr_entries,
1248 struct page **entries, pgoff_t *indices)
1249{
1250 void **slot;
1251 unsigned int ret = 0;
1252 struct radix_tree_iter iter;
1253
1254 if (!nr_entries)
1255 return 0;
1256
1257 rcu_read_lock();
Johannes Weiner0cd61442014-04-03 14:47:46 -07001258 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1259 struct page *page;
1260repeat:
1261 page = radix_tree_deref_slot(slot);
1262 if (unlikely(!page))
1263 continue;
1264 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001265 if (radix_tree_deref_retry(page)) {
1266 slot = radix_tree_iter_retry(&iter);
1267 continue;
1268 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001269 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -08001270 * A shadow entry of a recently evicted page, a swap
1271 * entry from shmem/tmpfs or a DAX entry. Return it
1272 * without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001273 */
1274 goto export;
1275 }
1276 if (!page_cache_get_speculative(page))
1277 goto repeat;
1278
1279 /* Has the page moved? */
1280 if (unlikely(page != *slot)) {
1281 page_cache_release(page);
1282 goto repeat;
1283 }
1284export:
1285 indices[ret] = iter.index;
1286 entries[ret] = page;
1287 if (++ret == nr_entries)
1288 break;
1289 }
1290 rcu_read_unlock();
1291 return ret;
1292}
1293
1294/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 * find_get_pages - gang pagecache lookup
1296 * @mapping: The address_space to search
1297 * @start: The starting page index
1298 * @nr_pages: The maximum number of pages
1299 * @pages: Where the resulting pages are placed
1300 *
1301 * find_get_pages() will search for and return a group of up to
1302 * @nr_pages pages in the mapping. The pages are placed at @pages.
1303 * find_get_pages() takes a reference against the returned pages.
1304 *
1305 * The search returns a group of mapping-contiguous pages with ascending
1306 * indexes. There may be holes in the indices due to not-present pages.
1307 *
1308 * find_get_pages() returns the number of pages which were found.
1309 */
1310unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1311 unsigned int nr_pages, struct page **pages)
1312{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001313 struct radix_tree_iter iter;
1314 void **slot;
1315 unsigned ret = 0;
1316
1317 if (unlikely(!nr_pages))
1318 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Nick Piggina60637c2008-07-25 19:45:31 -07001320 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001321 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001322 struct page *page;
1323repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001324 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001325 if (unlikely(!page))
1326 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001327
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001328 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001329 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001330 slot = radix_tree_iter_retry(&iter);
1331 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001332 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001333 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001334 * A shadow entry of a recently evicted page,
1335 * or a swap entry from shmem/tmpfs. Skip
1336 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001337 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001338 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001339 }
Nick Piggina60637c2008-07-25 19:45:31 -07001340
1341 if (!page_cache_get_speculative(page))
1342 goto repeat;
1343
1344 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001345 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001346 page_cache_release(page);
1347 goto repeat;
1348 }
1349
1350 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001351 if (++ret == nr_pages)
1352 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001353 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001354
Nick Piggina60637c2008-07-25 19:45:31 -07001355 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 return ret;
1357}
1358
Jens Axboeebf43502006-04-27 08:46:01 +02001359/**
1360 * find_get_pages_contig - gang contiguous pagecache lookup
1361 * @mapping: The address_space to search
1362 * @index: The starting page index
1363 * @nr_pages: The maximum number of pages
1364 * @pages: Where the resulting pages are placed
1365 *
1366 * find_get_pages_contig() works exactly like find_get_pages(), except
1367 * that the returned number of pages are guaranteed to be contiguous.
1368 *
1369 * find_get_pages_contig() returns the number of pages which were found.
1370 */
1371unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1372 unsigned int nr_pages, struct page **pages)
1373{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001374 struct radix_tree_iter iter;
1375 void **slot;
1376 unsigned int ret = 0;
1377
1378 if (unlikely(!nr_pages))
1379 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001380
Nick Piggina60637c2008-07-25 19:45:31 -07001381 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001382 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001383 struct page *page;
1384repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001385 page = radix_tree_deref_slot(slot);
1386 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001387 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001388 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001389
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001390 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001391 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001392 slot = radix_tree_iter_retry(&iter);
1393 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001394 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001395 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001396 * A shadow entry of a recently evicted page,
1397 * or a swap entry from shmem/tmpfs. Stop
1398 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001399 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001400 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001401 }
Nick Piggina60637c2008-07-25 19:45:31 -07001402
Nick Piggina60637c2008-07-25 19:45:31 -07001403 if (!page_cache_get_speculative(page))
1404 goto repeat;
1405
1406 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001407 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001408 page_cache_release(page);
1409 goto repeat;
1410 }
1411
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001412 /*
1413 * must check mapping and index after taking the ref.
1414 * otherwise we can get both false positives and false
1415 * negatives, which is just confusing to the caller.
1416 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001417 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001418 page_cache_release(page);
1419 break;
1420 }
1421
Nick Piggina60637c2008-07-25 19:45:31 -07001422 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001423 if (++ret == nr_pages)
1424 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001425 }
Nick Piggina60637c2008-07-25 19:45:31 -07001426 rcu_read_unlock();
1427 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001428}
David Howellsef71c152007-05-09 02:33:44 -07001429EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001430
Randy Dunlap485bb992006-06-23 02:03:49 -07001431/**
1432 * find_get_pages_tag - find and return pages that match @tag
1433 * @mapping: the address_space to search
1434 * @index: the starting page index
1435 * @tag: the tag index
1436 * @nr_pages: the maximum number of pages
1437 * @pages: where the resulting pages are placed
1438 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001440 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 */
1442unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1443 int tag, unsigned int nr_pages, struct page **pages)
1444{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001445 struct radix_tree_iter iter;
1446 void **slot;
1447 unsigned ret = 0;
1448
1449 if (unlikely(!nr_pages))
1450 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Nick Piggina60637c2008-07-25 19:45:31 -07001452 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001453 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1454 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001455 struct page *page;
1456repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001457 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001458 if (unlikely(!page))
1459 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001460
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001461 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001462 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001463 slot = radix_tree_iter_retry(&iter);
1464 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001465 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001466 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001467 * A shadow entry of a recently evicted page.
1468 *
1469 * Those entries should never be tagged, but
1470 * this tree walk is lockless and the tags are
1471 * looked up in bulk, one radix tree node at a
1472 * time, so there is a sizable window for page
1473 * reclaim to evict a page we saw tagged.
1474 *
1475 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001476 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001477 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001478 }
Nick Piggina60637c2008-07-25 19:45:31 -07001479
1480 if (!page_cache_get_speculative(page))
1481 goto repeat;
1482
1483 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001484 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001485 page_cache_release(page);
1486 goto repeat;
1487 }
1488
1489 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001490 if (++ret == nr_pages)
1491 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001492 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001493
Nick Piggina60637c2008-07-25 19:45:31 -07001494 rcu_read_unlock();
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 if (ret)
1497 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 return ret;
1500}
David Howellsef71c152007-05-09 02:33:44 -07001501EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001503/**
1504 * find_get_entries_tag - find and return entries that match @tag
1505 * @mapping: the address_space to search
1506 * @start: the starting page cache index
1507 * @tag: the tag index
1508 * @nr_entries: the maximum number of entries
1509 * @entries: where the resulting entries are placed
1510 * @indices: the cache indices corresponding to the entries in @entries
1511 *
1512 * Like find_get_entries, except we only return entries which are tagged with
1513 * @tag.
1514 */
1515unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1516 int tag, unsigned int nr_entries,
1517 struct page **entries, pgoff_t *indices)
1518{
1519 void **slot;
1520 unsigned int ret = 0;
1521 struct radix_tree_iter iter;
1522
1523 if (!nr_entries)
1524 return 0;
1525
1526 rcu_read_lock();
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001527 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1528 &iter, start, tag) {
1529 struct page *page;
1530repeat:
1531 page = radix_tree_deref_slot(slot);
1532 if (unlikely(!page))
1533 continue;
1534 if (radix_tree_exception(page)) {
1535 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001536 slot = radix_tree_iter_retry(&iter);
1537 continue;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001538 }
1539
1540 /*
1541 * A shadow entry of a recently evicted page, a swap
1542 * entry from shmem/tmpfs or a DAX entry. Return it
1543 * without attempting to raise page count.
1544 */
1545 goto export;
1546 }
1547 if (!page_cache_get_speculative(page))
1548 goto repeat;
1549
1550 /* Has the page moved? */
1551 if (unlikely(page != *slot)) {
1552 page_cache_release(page);
1553 goto repeat;
1554 }
1555export:
1556 indices[ret] = iter.index;
1557 entries[ret] = page;
1558 if (++ret == nr_entries)
1559 break;
1560 }
1561 rcu_read_unlock();
1562 return ret;
1563}
1564EXPORT_SYMBOL(find_get_entries_tag);
1565
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001566/*
1567 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1568 * a _large_ part of the i/o request. Imagine the worst scenario:
1569 *
1570 * ---R__________________________________________B__________
1571 * ^ reading here ^ bad block(assume 4k)
1572 *
1573 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1574 * => failing the whole request => read(R) => read(R+1) =>
1575 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1576 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1577 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1578 *
1579 * It is going insane. Fix it by quickly scaling down the readahead size.
1580 */
1581static void shrink_readahead_size_eio(struct file *filp,
1582 struct file_ra_state *ra)
1583{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001584 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001585}
1586
Randy Dunlap485bb992006-06-23 02:03:49 -07001587/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001588 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001589 * @filp: the file to read
1590 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001591 * @iter: data destination
1592 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001593 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001595 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 *
1597 * This is really ugly. But the goto's actually try to clarify some
1598 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 */
Al Viro6e58e792014-02-03 17:07:03 -05001600static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1601 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001603 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001605 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001606 pgoff_t index;
1607 pgoff_t last_index;
1608 pgoff_t prev_index;
1609 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001610 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001611 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001614 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1615 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Al Viro6e58e792014-02-03 17:07:03 -05001616 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 offset = *ppos & ~PAGE_CACHE_MASK;
1618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 for (;;) {
1620 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001621 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001622 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 unsigned long nr, ret;
1624
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626find_page:
1627 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001628 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001629 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001630 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001631 index, last_index - index);
1632 page = find_get_page(mapping, index);
1633 if (unlikely(page == NULL))
1634 goto no_cached_page;
1635 }
1636 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001637 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001638 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001639 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001641 if (!PageUptodate(page)) {
Mel Gormanebded022016-03-15 14:55:39 -07001642 /*
1643 * See comment in do_read_cache_page on why
1644 * wait_on_page_locked is used to avoid unnecessarily
1645 * serialisations and why it's safe.
1646 */
1647 wait_on_page_locked_killable(page);
1648 if (PageUptodate(page))
1649 goto page_ok;
1650
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001651 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1652 !mapping->a_ops->is_partially_uptodate)
1653 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001654 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001655 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001656 /* Did it get truncated before we got the lock? */
1657 if (!page->mapping)
1658 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001659 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001660 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001661 goto page_not_up_to_date_locked;
1662 unlock_page(page);
1663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001665 /*
1666 * i_size must be checked after we know the page is Uptodate.
1667 *
1668 * Checking i_size after the check allows us to calculate
1669 * the correct value for "nr", which means the zero-filled
1670 * part of the page is not copied back to userspace (unless
1671 * another truncate extends the file - this is desired though).
1672 */
1673
1674 isize = i_size_read(inode);
1675 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1676 if (unlikely(!isize || index > end_index)) {
1677 page_cache_release(page);
1678 goto out;
1679 }
1680
1681 /* nr is the maximum number of bytes to copy from this page */
1682 nr = PAGE_CACHE_SIZE;
1683 if (index == end_index) {
1684 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1685 if (nr <= offset) {
1686 page_cache_release(page);
1687 goto out;
1688 }
1689 }
1690 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691
1692 /* If users can be writing to this page using arbitrary
1693 * virtual addresses, take care about potential aliasing
1694 * before reading the page on the kernel side.
1695 */
1696 if (mapping_writably_mapped(mapping))
1697 flush_dcache_page(page);
1698
1699 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001700 * When a sequential read accesses a page several times,
1701 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 */
Jan Karaec0f1632007-05-06 14:49:25 -07001703 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 mark_page_accessed(page);
1705 prev_index = index;
1706
1707 /*
1708 * Ok, we have the page, and it's up-to-date, so
1709 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 */
Al Viro6e58e792014-02-03 17:07:03 -05001711
1712 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 offset += ret;
1714 index += offset >> PAGE_CACHE_SHIFT;
1715 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001716 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001719 written += ret;
1720 if (!iov_iter_count(iter))
1721 goto out;
1722 if (ret < nr) {
1723 error = -EFAULT;
1724 goto out;
1725 }
1726 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
1728page_not_up_to_date:
1729 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001730 error = lock_page_killable(page);
1731 if (unlikely(error))
1732 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001734page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001735 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 if (!page->mapping) {
1737 unlock_page(page);
1738 page_cache_release(page);
1739 continue;
1740 }
1741
1742 /* Did somebody else fill it already? */
1743 if (PageUptodate(page)) {
1744 unlock_page(page);
1745 goto page_ok;
1746 }
1747
1748readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001749 /*
1750 * A previous I/O error may have been due to temporary
1751 * failures, eg. multipath errors.
1752 * PG_error will be set again if readpage fails.
1753 */
1754 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 /* Start the actual read. The read will unlock the page. */
1756 error = mapping->a_ops->readpage(filp, page);
1757
Zach Brown994fc28c2005-12-15 14:28:17 -08001758 if (unlikely(error)) {
1759 if (error == AOP_TRUNCATED_PAGE) {
1760 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001761 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001762 goto find_page;
1763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
1767 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001768 error = lock_page_killable(page);
1769 if (unlikely(error))
1770 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 if (!PageUptodate(page)) {
1772 if (page->mapping == NULL) {
1773 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001774 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 */
1776 unlock_page(page);
1777 page_cache_release(page);
1778 goto find_page;
1779 }
1780 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001781 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001782 error = -EIO;
1783 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 }
1785 unlock_page(page);
1786 }
1787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 goto page_ok;
1789
1790readpage_error:
1791 /* UHHUH! A synchronous read error occurred. Report it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 page_cache_release(page);
1793 goto out;
1794
1795no_cached_page:
1796 /*
1797 * Ok, it wasn't cached, so we need to create a new
1798 * page..
1799 */
Nick Piggineb2be182007-10-16 01:24:57 -07001800 page = page_cache_alloc_cold(mapping);
1801 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001802 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001803 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001805 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08001806 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001808 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001809 if (error == -EEXIST) {
1810 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 goto out;
1814 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 goto readpage;
1816 }
1817
1818out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001819 ra->prev_pos = prev_index;
1820 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1821 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001823 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001824 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001825 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826}
1827
Randy Dunlap485bb992006-06-23 02:03:49 -07001828/**
Al Viro6abd2322014-04-04 14:20:57 -04001829 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001830 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001831 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001832 *
Al Viro6abd2322014-04-04 14:20:57 -04001833 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 * that can use the page cache directly.
1835 */
1836ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001837generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838{
Al Viroed978a82014-03-05 22:53:04 -05001839 struct file *file = iocb->ki_filp;
1840 ssize_t retval = 0;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001841 loff_t *ppos = &iocb->ki_pos;
Al Viroed978a82014-03-05 22:53:04 -05001842 loff_t pos = *ppos;
Nicolai Stangee7080a42016-03-25 14:22:14 -07001843 size_t count = iov_iter_count(iter);
1844
1845 if (!count)
1846 goto out; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Al Viro2ba48ce2015-04-09 13:52:01 -04001848 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001849 struct address_space *mapping = file->f_mapping;
1850 struct inode *inode = mapping->host;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001851 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001854 retval = filemap_write_and_wait_range(mapping, pos,
Al Viroed978a82014-03-05 22:53:04 -05001855 pos + count - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001856 if (!retval) {
Al Viroed978a82014-03-05 22:53:04 -05001857 struct iov_iter data = *iter;
Omar Sandoval22c61862015-03-16 04:33:53 -07001858 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001859 }
Al Viroed978a82014-03-05 22:53:04 -05001860
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001861 if (retval > 0) {
1862 *ppos = pos + retval;
Al Viroed978a82014-03-05 22:53:04 -05001863 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001864 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001865
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001866 /*
1867 * Btrfs can have a short DIO read if we encounter
1868 * compressed extents, so if there was an error, or if
1869 * we've already read everything we wanted to, or if
1870 * there was a short read because we hit EOF, go ahead
1871 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001872 * the rest of the read. Buffered reads will not work for
1873 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001874 */
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001875 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1876 IS_DAX(inode)) {
Al Viroed978a82014-03-05 22:53:04 -05001877 file_accessed(file);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001878 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 }
1881
Al Viroed978a82014-03-05 22:53:04 -05001882 retval = do_generic_file_read(file, ppos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883out:
1884 return retval;
1885}
Al Viroed978a82014-03-05 22:53:04 -05001886EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001889/**
1890 * page_cache_read - adds requested page to the page cache if not already there
1891 * @file: file to read
1892 * @offset: page index
Randy Dunlap62eb3202016-02-11 16:12:58 -08001893 * @gfp_mask: memory allocation flags
Randy Dunlap485bb992006-06-23 02:03:49 -07001894 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 * This adds the requested page to the page cache if it isn't already there,
1896 * and schedules an I/O to read in its contents from disk.
1897 */
Michal Hockoc20cd452016-01-14 15:20:12 -08001898static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899{
1900 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001901 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001902 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
Zach Brown994fc28c2005-12-15 14:28:17 -08001904 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08001905 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08001906 if (!page)
1907 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Michal Hockoc20cd452016-01-14 15:20:12 -08001909 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
Zach Brown994fc28c2005-12-15 14:28:17 -08001910 if (ret == 0)
1911 ret = mapping->a_ops->readpage(file, page);
1912 else if (ret == -EEXIST)
1913 ret = 0; /* losing race to add is OK */
1914
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Zach Brown994fc28c2005-12-15 14:28:17 -08001917 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07001918
Zach Brown994fc28c2005-12-15 14:28:17 -08001919 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920}
1921
1922#define MMAP_LOTSAMISS (100)
1923
Linus Torvaldsef00e082009-06-16 15:31:25 -07001924/*
1925 * Synchronous readahead happens when we don't even find
1926 * a page in the page cache at all.
1927 */
1928static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1929 struct file_ra_state *ra,
1930 struct file *file,
1931 pgoff_t offset)
1932{
Linus Torvaldsef00e082009-06-16 15:31:25 -07001933 struct address_space *mapping = file->f_mapping;
1934
1935 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001936 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001937 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001938 if (!ra->ra_pages)
1939 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001940
Joe Perches64363aa2013-07-08 16:00:18 -07001941 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001942 page_cache_sync_readahead(mapping, ra, file, offset,
1943 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001944 return;
1945 }
1946
Andi Kleen207d04b2011-05-24 17:12:29 -07001947 /* Avoid banging the cache line if not needed */
1948 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001949 ra->mmap_miss++;
1950
1951 /*
1952 * Do we miss much more than hit in this file? If so,
1953 * stop bothering with read-ahead. It will only hurt.
1954 */
1955 if (ra->mmap_miss > MMAP_LOTSAMISS)
1956 return;
1957
Wu Fengguangd30a1102009-06-16 15:31:30 -07001958 /*
1959 * mmap read-around
1960 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08001961 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1962 ra->size = ra->ra_pages;
1963 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001964 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001965}
1966
1967/*
1968 * Asynchronous readahead happens when we find the page and PG_readahead,
1969 * so we want to possibly extend the readahead further..
1970 */
1971static void do_async_mmap_readahead(struct vm_area_struct *vma,
1972 struct file_ra_state *ra,
1973 struct file *file,
1974 struct page *page,
1975 pgoff_t offset)
1976{
1977 struct address_space *mapping = file->f_mapping;
1978
1979 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001980 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001981 return;
1982 if (ra->mmap_miss > 0)
1983 ra->mmap_miss--;
1984 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001985 page_cache_async_readahead(mapping, ra, file,
1986 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001987}
1988
Randy Dunlap485bb992006-06-23 02:03:49 -07001989/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001990 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001991 * @vma: vma in which the fault was taken
1992 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001993 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001994 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 * mapped memory region to read in file data during a page fault.
1996 *
1997 * The goto's are kind of ugly, but this streamlines the normal case of having
1998 * it in the page cache, and handles the special cases reasonably without
1999 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002000 *
2001 * vma->vm_mm->mmap_sem must be held on entry.
2002 *
2003 * If our return value has VM_FAULT_RETRY set, it's because
2004 * lock_page_or_retry() returned 0.
2005 * The mmap_sem has usually been released in this case.
2006 * See __lock_page_or_retry() for the exception.
2007 *
2008 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2009 * has not been released.
2010 *
2011 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002013int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014{
2015 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07002016 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 struct address_space *mapping = file->f_mapping;
2018 struct file_ra_state *ra = &file->f_ra;
2019 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002020 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002022 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07002023 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002025 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2026 if (offset >= size >> PAGE_CACHE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07002027 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002030 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002032 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002033 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002035 * We found the page, so try async readahead before
2036 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002038 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002039 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002040 /* No page in the page cache at all */
2041 do_sync_mmap_readahead(vma, ra, file, offset);
2042 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002043 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002044 ret = VM_FAULT_MAJOR;
2045retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07002046 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 if (!page)
2048 goto no_cached_page;
2049 }
2050
Michel Lespinassed88c0922010-11-02 13:05:18 -07002051 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2052 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002053 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07002054 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07002055
2056 /* Did it get truncated? */
2057 if (unlikely(page->mapping != mapping)) {
2058 unlock_page(page);
2059 put_page(page);
2060 goto retry_find;
2061 }
Sasha Levin309381fea2014-01-23 15:52:54 -08002062 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002063
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002065 * We have a locked page in the page cache, now we need to check
2066 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 */
Nick Piggind00806b2007-07-19 01:46:57 -07002068 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 goto page_not_uptodate;
2070
Linus Torvaldsef00e082009-06-16 15:31:25 -07002071 /*
2072 * Found the page and have a reference on it.
2073 * We must recheck i_size under page lock.
2074 */
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002075 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2076 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002077 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07002078 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002079 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002080 }
2081
Nick Piggind0217ac2007-07-19 01:47:03 -07002082 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002083 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085no_cached_page:
2086 /*
2087 * We're only likely to ever get here if MADV_RANDOM is in
2088 * effect.
2089 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002090 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
2092 /*
2093 * The page we want has now been added to the page cache.
2094 * In the unlikely event that someone removed it in the
2095 * meantime, we'll just come back here and read it again.
2096 */
2097 if (error >= 0)
2098 goto retry_find;
2099
2100 /*
2101 * An error return from page_cache_read can result if the
2102 * system is low on memory, or a problem occurs while trying
2103 * to schedule I/O.
2104 */
2105 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002106 return VM_FAULT_OOM;
2107 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
2109page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 /*
2111 * Umm, take care of errors if the page isn't up-to-date.
2112 * Try to re-read it _once_. We do this synchronously,
2113 * because there really aren't any performance issues here
2114 * and we need to check for errors.
2115 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002117 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002118 if (!error) {
2119 wait_on_page_locked(page);
2120 if (!PageUptodate(page))
2121 error = -EIO;
2122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002124
2125 if (!error || error == AOP_TRUNCATED_PAGE)
2126 goto retry_find;
2127
2128 /* Things didn't work out. Return zero to tell the mm layer so. */
2129 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002130 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002131}
2132EXPORT_SYMBOL(filemap_fault);
2133
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002134void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2135{
2136 struct radix_tree_iter iter;
2137 void **slot;
2138 struct file *file = vma->vm_file;
2139 struct address_space *mapping = file->f_mapping;
2140 loff_t size;
2141 struct page *page;
2142 unsigned long address = (unsigned long) vmf->virtual_address;
2143 unsigned long addr;
2144 pte_t *pte;
2145
2146 rcu_read_lock();
2147 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2148 if (iter.index > vmf->max_pgoff)
2149 break;
2150repeat:
2151 page = radix_tree_deref_slot(slot);
2152 if (unlikely(!page))
2153 goto next;
2154 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002155 if (radix_tree_deref_retry(page)) {
2156 slot = radix_tree_iter_retry(&iter);
2157 continue;
2158 }
2159 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002160 }
2161
2162 if (!page_cache_get_speculative(page))
2163 goto repeat;
2164
2165 /* Has the page moved? */
2166 if (unlikely(page != *slot)) {
2167 page_cache_release(page);
2168 goto repeat;
2169 }
2170
2171 if (!PageUptodate(page) ||
2172 PageReadahead(page) ||
2173 PageHWPoison(page))
2174 goto skip;
2175 if (!trylock_page(page))
2176 goto skip;
2177
2178 if (page->mapping != mapping || !PageUptodate(page))
2179 goto unlock;
2180
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002181 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2182 if (page->index >= size >> PAGE_CACHE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002183 goto unlock;
2184
2185 pte = vmf->pte + page->index - vmf->pgoff;
2186 if (!pte_none(*pte))
2187 goto unlock;
2188
2189 if (file->f_ra.mmap_miss > 0)
2190 file->f_ra.mmap_miss--;
2191 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2192 do_set_pte(vma, addr, page, pte, false, false);
2193 unlock_page(page);
2194 goto next;
2195unlock:
2196 unlock_page(page);
2197skip:
2198 page_cache_release(page);
2199next:
2200 if (iter.index == vmf->max_pgoff)
2201 break;
2202 }
2203 rcu_read_unlock();
2204}
2205EXPORT_SYMBOL(filemap_map_pages);
2206
Jan Kara4fcf1c62012-06-12 16:20:29 +02002207int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2208{
2209 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002210 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002211 int ret = VM_FAULT_LOCKED;
2212
Jan Kara14da9202012-06-12 16:20:37 +02002213 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002214 file_update_time(vma->vm_file);
2215 lock_page(page);
2216 if (page->mapping != inode->i_mapping) {
2217 unlock_page(page);
2218 ret = VM_FAULT_NOPAGE;
2219 goto out;
2220 }
Jan Kara14da9202012-06-12 16:20:37 +02002221 /*
2222 * We mark the page dirty already here so that when freeze is in
2223 * progress, we are guaranteed that writeback during freezing will
2224 * see the dirty page and writeprotect it again.
2225 */
2226 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002227 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002228out:
Jan Kara14da9202012-06-12 16:20:37 +02002229 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002230 return ret;
2231}
2232EXPORT_SYMBOL(filemap_page_mkwrite);
2233
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002234const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002235 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002236 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002237 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238};
2239
2240/* This is used for a general mmap of a disk file */
2241
2242int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2243{
2244 struct address_space *mapping = file->f_mapping;
2245
2246 if (!mapping->a_ops->readpage)
2247 return -ENOEXEC;
2248 file_accessed(file);
2249 vma->vm_ops = &generic_file_vm_ops;
2250 return 0;
2251}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253/*
2254 * This is for filesystems which do not implement ->writepage.
2255 */
2256int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2257{
2258 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2259 return -EINVAL;
2260 return generic_file_mmap(file, vma);
2261}
2262#else
2263int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2264{
2265 return -ENOSYS;
2266}
2267int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2268{
2269 return -ENOSYS;
2270}
2271#endif /* CONFIG_MMU */
2272
2273EXPORT_SYMBOL(generic_file_mmap);
2274EXPORT_SYMBOL(generic_file_readonly_mmap);
2275
Sasha Levin67f9fd92014-04-03 14:48:18 -07002276static struct page *wait_on_page_read(struct page *page)
2277{
2278 if (!IS_ERR(page)) {
2279 wait_on_page_locked(page);
2280 if (!PageUptodate(page)) {
2281 page_cache_release(page);
2282 page = ERR_PTR(-EIO);
2283 }
2284 }
2285 return page;
2286}
2287
Mel Gorman32b63522016-03-15 14:55:36 -07002288static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002289 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002290 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002291 void *data,
2292 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293{
Nick Piggineb2be182007-10-16 01:24:57 -07002294 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 int err;
2296repeat:
2297 page = find_get_page(mapping, index);
2298 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002299 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002300 if (!page)
2301 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002302 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002303 if (unlikely(err)) {
2304 page_cache_release(page);
2305 if (err == -EEXIST)
2306 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 return ERR_PTR(err);
2309 }
Mel Gorman32b63522016-03-15 14:55:36 -07002310
2311filler:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 err = filler(data, page);
2313 if (err < 0) {
2314 page_cache_release(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002315 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 }
Mel Gorman32b63522016-03-15 14:55:36 -07002317
2318 page = wait_on_page_read(page);
2319 if (IS_ERR(page))
2320 return page;
2321 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 if (PageUptodate(page))
2324 goto out;
2325
Mel Gormanebded022016-03-15 14:55:39 -07002326 /*
2327 * Page is not up to date and may be locked due one of the following
2328 * case a: Page is being filled and the page lock is held
2329 * case b: Read/write error clearing the page uptodate status
2330 * case c: Truncation in progress (page locked)
2331 * case d: Reclaim in progress
2332 *
2333 * Case a, the page will be up to date when the page is unlocked.
2334 * There is no need to serialise on the page lock here as the page
2335 * is pinned so the lock gives no additional protection. Even if the
2336 * the page is truncated, the data is still valid if PageUptodate as
2337 * it's a race vs truncate race.
2338 * Case b, the page will not be up to date
2339 * Case c, the page may be truncated but in itself, the data may still
2340 * be valid after IO completes as it's a read vs truncate race. The
2341 * operation must restart if the page is not uptodate on unlock but
2342 * otherwise serialising on page lock to stabilise the mapping gives
2343 * no additional guarantees to the caller as the page lock is
2344 * released before return.
2345 * Case d, similar to truncation. If reclaim holds the page lock, it
2346 * will be a race with remove_mapping that determines if the mapping
2347 * is valid on unlock but otherwise the data is valid and there is
2348 * no need to serialise with page lock.
2349 *
2350 * As the page lock gives no additional guarantee, we optimistically
2351 * wait on the page to be unlocked and check if it's up to date and
2352 * use the page if it is. Otherwise, the page lock is required to
2353 * distinguish between the different cases. The motivation is that we
2354 * avoid spurious serialisations and wakeups when multiple processes
2355 * wait on the same page for IO to complete.
2356 */
2357 wait_on_page_locked(page);
2358 if (PageUptodate(page))
2359 goto out;
2360
2361 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07002363
2364 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 if (!page->mapping) {
2366 unlock_page(page);
2367 page_cache_release(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002368 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 }
Mel Gormanebded022016-03-15 14:55:39 -07002370
2371 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 if (PageUptodate(page)) {
2373 unlock_page(page);
2374 goto out;
2375 }
Mel Gorman32b63522016-03-15 14:55:36 -07002376 goto filler;
2377
David Howellsc855ff32007-05-09 13:42:20 +01002378out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002379 mark_page_accessed(page);
2380 return page;
2381}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002382
2383/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002384 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002385 * @mapping: the page's address_space
2386 * @index: the page index
2387 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002388 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002389 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002390 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002391 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002392 *
2393 * If the page does not get brought uptodate, return -EIO.
2394 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002395struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002396 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002397 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002398 void *data)
2399{
2400 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2401}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002402EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002403
2404/**
2405 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2406 * @mapping: the page's address_space
2407 * @index: the page index
2408 * @gfp: the page allocator flags to use if allocating
2409 *
2410 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002411 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002412 *
2413 * If the page does not get brought uptodate, return -EIO.
2414 */
2415struct page *read_cache_page_gfp(struct address_space *mapping,
2416 pgoff_t index,
2417 gfp_t gfp)
2418{
2419 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2420
Sasha Levin67f9fd92014-04-03 14:48:18 -07002421 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002422}
2423EXPORT_SYMBOL(read_cache_page_gfp);
2424
Nick Piggin2f718ff2007-10-16 01:24:59 -07002425/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 * Performs necessary checks before doing a write
2427 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002428 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 * Returns appropriate error code that caller should return or
2430 * zero in case that write should be allowed.
2431 */
Al Viro3309dd02015-04-09 12:55:47 -04002432inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433{
Al Viro3309dd02015-04-09 12:55:47 -04002434 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002436 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002437 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Al Viro3309dd02015-04-09 12:55:47 -04002439 if (!iov_iter_count(from))
2440 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Al Viro0fa6b002015-04-04 04:05:48 -04002442 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002443 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002444 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Al Viro3309dd02015-04-09 12:55:47 -04002446 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Al Viro0fa6b002015-04-04 04:05:48 -04002448 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002449 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002450 send_sig(SIGXFSZ, current, 0);
2451 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 }
Al Viro3309dd02015-04-09 12:55:47 -04002453 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 }
2455
2456 /*
2457 * LFS rule
2458 */
Al Viro3309dd02015-04-09 12:55:47 -04002459 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002461 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002463 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 }
2465
2466 /*
2467 * Are we about to exceed the fs block limit ?
2468 *
2469 * If we have written data it becomes a short write. If we have
2470 * exceeded without writing data we send a signal and return EFBIG.
2471 * Linus frestrict idea will clean these up nicely..
2472 */
Al Viro3309dd02015-04-09 12:55:47 -04002473 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2474 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475
Al Viro3309dd02015-04-09 12:55:47 -04002476 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2477 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478}
2479EXPORT_SYMBOL(generic_write_checks);
2480
Nick Pigginafddba42007-10-16 01:25:01 -07002481int pagecache_write_begin(struct file *file, struct address_space *mapping,
2482 loff_t pos, unsigned len, unsigned flags,
2483 struct page **pagep, void **fsdata)
2484{
2485 const struct address_space_operations *aops = mapping->a_ops;
2486
Nick Piggin4e02ed42008-10-29 14:00:55 -07002487 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002488 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002489}
2490EXPORT_SYMBOL(pagecache_write_begin);
2491
2492int pagecache_write_end(struct file *file, struct address_space *mapping,
2493 loff_t pos, unsigned len, unsigned copied,
2494 struct page *page, void *fsdata)
2495{
2496 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002497
Nick Piggin4e02ed42008-10-29 14:00:55 -07002498 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002499}
2500EXPORT_SYMBOL(pagecache_write_end);
2501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502ssize_t
Al Viro0c949332014-03-22 06:51:37 -04002503generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504{
2505 struct file *file = iocb->ki_filp;
2506 struct address_space *mapping = file->f_mapping;
2507 struct inode *inode = mapping->host;
2508 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002509 size_t write_len;
2510 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002511 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512
Al Viro0c949332014-03-22 06:51:37 -04002513 write_len = iov_iter_count(from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002514 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002515
Nick Piggin48b47c52009-01-06 14:40:22 -08002516 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002517 if (written)
2518 goto out;
2519
2520 /*
2521 * After a write we want buffered reads to be sure to go to disk to get
2522 * the new data. We invalidate clean cached page from the region we're
2523 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002524 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002525 */
2526 if (mapping->nrpages) {
2527 written = invalidate_inode_pages2_range(mapping,
2528 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002529 /*
2530 * If a page can not be invalidated, return 0 to fall back
2531 * to buffered write.
2532 */
2533 if (written) {
2534 if (written == -EBUSY)
2535 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002536 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002537 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002538 }
2539
Al Viro26978b82014-03-10 14:08:45 -04002540 data = *from;
Omar Sandoval22c61862015-03-16 04:33:53 -07002541 written = mapping->a_ops->direct_IO(iocb, &data, pos);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002542
2543 /*
2544 * Finally, try again to invalidate clean pages which might have been
2545 * cached by non-direct readahead, or faulted in by get_user_pages()
2546 * if the source of the write was an mmap'ed region of the file
2547 * we're writing. Either one is a pretty crazy thing to do,
2548 * so we don't support it 100%. If this invalidation
2549 * fails, tough, the write still worked...
2550 */
2551 if (mapping->nrpages) {
2552 invalidate_inode_pages2_range(mapping,
2553 pos >> PAGE_CACHE_SHIFT, end);
2554 }
2555
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002557 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002558 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002559 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2560 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 mark_inode_dirty(inode);
2562 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002563 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002565out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 return written;
2567}
2568EXPORT_SYMBOL(generic_file_direct_write);
2569
Nick Piggineb2be182007-10-16 01:24:57 -07002570/*
2571 * Find or create a page at the given pagecache position. Return the locked
2572 * page. This function is specifically for buffered writes.
2573 */
Nick Piggin54566b22009-01-04 12:00:53 -08002574struct page *grab_cache_page_write_begin(struct address_space *mapping,
2575 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002576{
Nick Piggineb2be182007-10-16 01:24:57 -07002577 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07002578 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002579
Nick Piggin54566b22009-01-04 12:00:53 -08002580 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002581 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002582
Mel Gorman2457aec2014-06-04 16:10:31 -07002583 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002584 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002585 if (page)
2586 wait_for_stable_page(page);
2587
Nick Piggineb2be182007-10-16 01:24:57 -07002588 return page;
2589}
Nick Piggin54566b22009-01-04 12:00:53 -08002590EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002591
Al Viro3b93f912014-02-11 21:34:08 -05002592ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002593 struct iov_iter *i, loff_t pos)
2594{
2595 struct address_space *mapping = file->f_mapping;
2596 const struct address_space_operations *a_ops = mapping->a_ops;
2597 long status = 0;
2598 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002599 unsigned int flags = 0;
2600
2601 /*
2602 * Copies from kernel address space cannot fail (NFSD is a big user).
2603 */
Al Viro777eda22014-12-17 04:46:46 -05002604 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002605 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002606
2607 do {
2608 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002609 unsigned long offset; /* Offset into pagecache page */
2610 unsigned long bytes; /* Bytes to write to page */
2611 size_t copied; /* Bytes copied from user */
2612 void *fsdata;
2613
2614 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002615 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2616 iov_iter_count(i));
2617
2618again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002619 /*
2620 * Bring in the user page that we will copy from _first_.
2621 * Otherwise there's a nasty deadlock on copying from the
2622 * same page as we're writing to, without it being marked
2623 * up-to-date.
2624 *
2625 * Not only is this an optimisation, but it is also required
2626 * to check that the address is actually valid, when atomic
2627 * usercopies are used, below.
2628 */
2629 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2630 status = -EFAULT;
2631 break;
2632 }
2633
Jan Kara296291c2015-10-22 13:32:21 -07002634 if (fatal_signal_pending(current)) {
2635 status = -EINTR;
2636 break;
2637 }
2638
Nick Piggin674b8922007-10-16 01:25:03 -07002639 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002640 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002641 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002642 break;
2643
anfei zhou931e80e2010-02-02 13:44:02 -08002644 if (mapping_writably_mapped(mapping))
2645 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002646
Nick Pigginafddba42007-10-16 01:25:01 -07002647 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002648 flush_dcache_page(page);
2649
2650 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2651 page, fsdata);
2652 if (unlikely(status < 0))
2653 break;
2654 copied = status;
2655
2656 cond_resched();
2657
Nick Piggin124d3b72008-02-02 15:01:17 +01002658 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002659 if (unlikely(copied == 0)) {
2660 /*
2661 * If we were unable to copy any data at all, we must
2662 * fall back to a single segment length write.
2663 *
2664 * If we didn't fallback here, we could livelock
2665 * because not all segments in the iov can be copied at
2666 * once without a pagefault.
2667 */
2668 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2669 iov_iter_single_seg_count(i));
2670 goto again;
2671 }
Nick Pigginafddba42007-10-16 01:25:01 -07002672 pos += copied;
2673 written += copied;
2674
2675 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002676 } while (iov_iter_count(i));
2677
2678 return written ? written : status;
2679}
Al Viro3b93f912014-02-11 21:34:08 -05002680EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
Jan Karae4dd9de2009-08-17 18:10:06 +02002682/**
Al Viro81742022014-04-03 03:17:43 -04002683 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002684 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002685 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002686 *
2687 * This function does all the work needed for actually writing data to a
2688 * file. It does all basic checks, removes SUID from the file, updates
2689 * modification times and calls proper subroutines depending on whether we
2690 * do direct IO or a standard buffered write.
2691 *
2692 * It expects i_mutex to be grabbed unless we work on a block device or similar
2693 * object which does not need locking at all.
2694 *
2695 * This function does *not* take care of syncing data in case of O_SYNC write.
2696 * A caller has to handle it. This is mainly due to the fact that we want to
2697 * avoid syncing under i_mutex.
2698 */
Al Viro81742022014-04-03 03:17:43 -04002699ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700{
2701 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002702 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002704 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002706 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002709 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002710 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 if (err)
2712 goto out;
2713
Josef Bacikc3b2da32012-03-26 09:59:21 -04002714 err = file_update_time(file);
2715 if (err)
2716 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
Al Viro2ba48ce2015-04-09 13:52:01 -04002718 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002719 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002720
Al Viro0b8def92015-04-07 10:22:53 -04002721 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002722 /*
2723 * If the write stopped short of completing, fall back to
2724 * buffered writes. Some filesystems do this for writes to
2725 * holes, for example. For DAX files, a buffered write will
2726 * not succeed (even if it did, DAX does not handle dirty
2727 * page-cache pages correctly).
2728 */
Al Viro0b8def92015-04-07 10:22:53 -04002729 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002731
Al Viro0b8def92015-04-07 10:22:53 -04002732 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002733 /*
Al Viro3b93f912014-02-11 21:34:08 -05002734 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002735 * then we want to return the number of bytes which were
2736 * direct-written, or the error code if that was zero. Note
2737 * that this differs from normal direct-io semantics, which
2738 * will return -EFOO even if some bytes were written.
2739 */
Al Viro60bb4522014-08-08 12:39:16 -04002740 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002741 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002742 goto out;
2743 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002744 /*
2745 * We need to ensure that the page cache pages are written to
2746 * disk and invalidated to preserve the expected O_DIRECT
2747 * semantics.
2748 */
Al Viro3b93f912014-02-11 21:34:08 -05002749 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002750 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002751 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002752 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002753 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002754 invalidate_mapping_pages(mapping,
2755 pos >> PAGE_CACHE_SHIFT,
2756 endbyte >> PAGE_CACHE_SHIFT);
2757 } else {
2758 /*
2759 * We don't know how much we wrote, so just return
2760 * the number of bytes which were direct-written
2761 */
2762 }
2763 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002764 written = generic_perform_write(file, from, iocb->ki_pos);
2765 if (likely(written > 0))
2766 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768out:
2769 current->backing_dev_info = NULL;
2770 return written ? written : err;
2771}
Al Viro81742022014-04-03 03:17:43 -04002772EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
Jan Karae4dd9de2009-08-17 18:10:06 +02002774/**
Al Viro81742022014-04-03 03:17:43 -04002775 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002776 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002777 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002778 *
Al Viro81742022014-04-03 03:17:43 -04002779 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002780 * filesystems. It takes care of syncing the file in case of O_SYNC file
2781 * and acquires i_mutex as needed.
2782 */
Al Viro81742022014-04-03 03:17:43 -04002783ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784{
2785 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002786 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
Al Viro59551022016-01-22 15:40:57 -05002789 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04002790 ret = generic_write_checks(iocb, from);
2791 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002792 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05002793 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
Christoph Hellwig02afc272013-09-04 15:04:40 +02002795 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 ssize_t err;
2797
Al Virod311d792014-02-09 15:18:09 -05002798 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2799 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 ret = err;
2801 }
2802 return ret;
2803}
Al Viro81742022014-04-03 03:17:43 -04002804EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805
David Howellscf9a2ae2006-08-29 19:05:54 +01002806/**
2807 * try_to_release_page() - release old fs-specific metadata on a page
2808 *
2809 * @page: the page which the kernel is trying to free
2810 * @gfp_mask: memory allocation flags (and I/O mode)
2811 *
2812 * The address_space is to try to release any data against the page
2813 * (presumably at page->private). If the release was successful, return `1'.
2814 * Otherwise return zero.
2815 *
David Howells266cf652009-04-03 16:42:36 +01002816 * This may also be called if PG_fscache is set on a page, indicating that the
2817 * page is known to the local caching routines.
2818 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002819 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08002820 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002821 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002822 */
2823int try_to_release_page(struct page *page, gfp_t gfp_mask)
2824{
2825 struct address_space * const mapping = page->mapping;
2826
2827 BUG_ON(!PageLocked(page));
2828 if (PageWriteback(page))
2829 return 0;
2830
2831 if (mapping && mapping->a_ops->releasepage)
2832 return mapping->a_ops->releasepage(page, gfp_mask);
2833 return try_to_free_buffers(page);
2834}
2835
2836EXPORT_SYMBOL(try_to_release_page);