blob: 49bc9720fb00bf22f126a0d28ed149f7ecdedb1d [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>
Ingo Molnar3f07c012017-02-08 18:51:30 +010016#include <linux/sched/signal.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070017#include <linux/uaccess.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mm.h>
22#include <linux/swap.h>
23#include <linux/mman.h>
24#include <linux/pagemap.h>
25#include <linux/file.h>
26#include <linux/uio.h>
27#include <linux/hash.h>
28#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070029#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Johannes Weiner00501b52014-08-08 14:19:20 -070035#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080036#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060037#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070038#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080039#include "internal.h"
40
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070041#define CREATE_TRACE_POINTS
42#include <trace/events/filemap.h>
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 * FIXME: remove all knowledge of the buffer layer from the core VM
46 */
Jan Kara148f9482009-08-17 19:52:36 +020047#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <asm/mman.h>
50
51/*
52 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * though.
54 *
55 * Shared mappings now work. 15.8.1995 Bruno.
56 *
57 * finished 'unifying' the page and buffer cache and SMP-threaded the
58 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
59 *
60 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
61 */
62
63/*
64 * Lock ordering:
65 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080066 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070068 * ->swap_lock (exclusive_swap_page, others)
69 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080071 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080072 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 *
74 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080075 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070076 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
78 *
79 * ->mmap_sem
80 * ->lock_page (access_process_vm)
81 *
Al Viroccad2362014-02-11 22:36:48 -050082 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070083 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060085 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110086 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * ->mapping->tree_lock (__sync_single_inode)
88 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080089 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 * ->anon_vma.lock (vma_adjust)
91 *
92 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070095 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070096 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * ->private_lock (try_to_unmap_one)
98 * ->tree_lock (try_to_unmap_one)
Mel Gormana52633d2016-07-28 15:45:28 -070099 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
100 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 * ->private_lock (page_remove_rmap->set_page_dirty)
102 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600103 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100104 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700105 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600106 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100107 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
109 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800110 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700111 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 */
113
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700114static int page_cache_tree_insert(struct address_space *mapping,
115 struct page *page, void **shadowp)
116{
117 struct radix_tree_node *node;
118 void **slot;
119 int error;
120
121 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
122 &node, &slot);
123 if (error)
124 return error;
125 if (*slot) {
126 void *p;
127
128 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
129 if (!radix_tree_exceptional_entry(p))
130 return -EEXIST;
131
132 mapping->nrexceptional--;
133 if (!dax_mapping(mapping)) {
134 if (shadowp)
135 *shadowp = p;
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700136 } else {
137 /* DAX can replace empty locked entry with a hole */
138 WARN_ON_ONCE(p !=
Ross Zwisler642261a2016-11-08 11:34:45 +1100139 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700140 /* Wakeup waiters for exceptional entry lock */
Ross Zwisler63e95b52016-11-08 11:32:20 +1100141 dax_wake_mapping_entry_waiter(mapping, page->index, p,
Ross Zwisler965d0042017-01-10 16:57:15 -0800142 true);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700143 }
144 }
Johannes Weiner14b46872016-12-12 16:43:52 -0800145 __radix_tree_replace(&mapping->page_tree, node, slot, page,
146 workingset_update_node, mapping);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700147 mapping->nrpages++;
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700148 return 0;
149}
150
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700151static void page_cache_tree_delete(struct address_space *mapping,
152 struct page *page, void *shadow)
153{
Kirill A. Shutemovc70b6472016-12-12 16:43:17 -0800154 int i, nr;
155
156 /* hugetlb pages are represented by one entry in the radix tree */
157 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700158
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700159 VM_BUG_ON_PAGE(!PageLocked(page), page);
160 VM_BUG_ON_PAGE(PageTail(page), page);
161 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700162
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700163 for (i = 0; i < nr; i++) {
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200164 struct radix_tree_node *node;
165 void **slot;
166
167 __radix_tree_lookup(&mapping->page_tree, page->index + i,
168 &node, &slot);
169
Johannes Weinerdbc446b2016-12-12 16:43:55 -0800170 VM_BUG_ON_PAGE(!node && nr != 1, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700171
Johannes Weiner14b46872016-12-12 16:43:52 -0800172 radix_tree_clear_tags(&mapping->page_tree, node, slot);
173 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
174 workingset_update_node, mapping);
Johannes Weiner449dd692014-04-03 14:47:56 -0700175 }
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200176
177 if (shadow) {
178 mapping->nrexceptional += nr;
179 /*
180 * Make sure the nrexceptional update is committed before
181 * the nrpages update so that final truncate racing
182 * with reclaim does not see both counters 0 at the
183 * same time and miss a shadow entry.
184 */
185 smp_wmb();
186 }
187 mapping->nrpages -= nr;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/*
Minchan Kime64a7822011-03-22 16:32:44 -0700191 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 * sure the page is locked and that nobody else uses it - or that usage
Johannes Weinerfdf1cdb2016-03-15 14:57:25 -0700193 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700195void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
197 struct address_space *mapping = page->mapping;
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700198 int nr = hpage_nr_pages(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700200 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600201 /*
202 * if we're uptodate, flush out into the cleancache, otherwise
203 * invalidate any existing cleancache entries. We can't leave
204 * stale data around in the cleancache once our page is gone
205 */
206 if (PageUptodate(page) && PageMappedToDisk(page))
207 cleancache_put_page(page);
208 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400209 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600210
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700211 VM_BUG_ON_PAGE(PageTail(page), page);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800212 VM_BUG_ON_PAGE(page_mapped(page), page);
213 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
214 int mapcount;
215
216 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
217 current->comm, page_to_pfn(page));
218 dump_page(page, "still mapped when deleted");
219 dump_stack();
220 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
221
222 mapcount = page_mapcount(page);
223 if (mapping_exiting(mapping) &&
224 page_count(page) >= mapcount + 2) {
225 /*
226 * All vmas have already been torn down, so it's
227 * a good bet that actually the page is unmapped,
228 * and we'd prefer not to leak it: if we're wrong,
229 * some other bad page check should catch it later.
230 */
231 page_mapcount_reset(page);
Joonsoo Kim6d061f92016-05-19 17:10:46 -0700232 page_ref_sub(page, mapcount);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800233 }
234 }
235
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700236 page_cache_tree_delete(mapping, page, shadow);
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700239 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700240
Michal Hocko4165b9b2015-06-24 16:57:24 -0700241 /* hugetlb pages do not participate in page cache accounting. */
242 if (!PageHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700243 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700244 if (PageSwapBacked(page)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700245 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700246 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700247 __dec_node_page_state(page, NR_SHMEM_THPS);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700248 } else {
249 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
250 }
Linus Torvalds3a692792007-12-19 14:05:13 -0800251
252 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700253 * At this point page must be either written or cleaned by truncate.
254 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800255 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700256 * This fixes dirty accounting after removing the page entirely but
257 * leaves PageDirty set: it has no effect for truncated page and
258 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800259 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700260 if (WARN_ON_ONCE(PageDirty(page)))
Johannes Weiner62cccb82016-03-15 14:57:22 -0700261 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
263
Minchan Kim702cfbf2011-03-22 16:32:43 -0700264/**
265 * delete_from_page_cache - delete page from page cache
266 * @page: the page which the kernel is trying to remove from page cache
267 *
268 * This must be called only on pages that have been verified to be in the page
269 * cache and locked. It will never put the page into the free list, the caller
270 * has a reference on the page.
271 */
272void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700274 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400275 unsigned long flags;
Linus Torvalds6072d132010-12-01 13:35:19 -0500276 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
Matt Mackallcd7619d2005-05-01 08:59:01 -0700278 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Linus Torvalds6072d132010-12-01 13:35:19 -0500280 freepage = mapping->a_ops->freepage;
Greg Thelenc4843a72015-05-22 17:13:16 -0400281
Greg Thelenc4843a72015-05-22 17:13:16 -0400282 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700283 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400284 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500285
286 if (freepage)
287 freepage(page);
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700288
289 if (PageTransHuge(page) && !PageHuge(page)) {
290 page_ref_sub(page, HPAGE_PMD_NR);
291 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
292 } else {
293 put_page(page);
294 }
Minchan Kim97cecb52011-03-22 16:30:53 -0700295}
296EXPORT_SYMBOL(delete_from_page_cache);
297
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200298int filemap_check_errors(struct address_space *mapping)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700299{
300 int ret = 0;
301 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700302 if (test_bit(AS_ENOSPC, &mapping->flags) &&
303 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700304 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700305 if (test_bit(AS_EIO, &mapping->flags) &&
306 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700307 ret = -EIO;
308 return ret;
309}
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200310EXPORT_SYMBOL(filemap_check_errors);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700311
Jeff Layton76341ca2017-07-06 07:02:22 -0400312static int filemap_check_and_keep_errors(struct address_space *mapping)
313{
314 /* Check for outstanding write errors */
315 if (test_bit(AS_EIO, &mapping->flags))
316 return -EIO;
317 if (test_bit(AS_ENOSPC, &mapping->flags))
318 return -ENOSPC;
319 return 0;
320}
321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700323 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700324 * @mapping: address space structure to write
325 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800326 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700327 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700329 * Start writeback against all of a mapping's dirty pages that lie
330 * within the byte offsets <start, end> inclusive.
331 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700333 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 * these two operations is that if a dirty page/buffer is encountered, it must
335 * be waited upon, and not just skipped over.
336 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800337int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
338 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
340 int ret;
341 struct writeback_control wbc = {
342 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800343 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700344 .range_start = start,
345 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 };
347
348 if (!mapping_cap_writeback_dirty(mapping))
349 return 0;
350
Tejun Heob16b1de2015-06-02 08:39:48 -0600351 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600353 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 return ret;
355}
356
357static inline int __filemap_fdatawrite(struct address_space *mapping,
358 int sync_mode)
359{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700360 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
363int filemap_fdatawrite(struct address_space *mapping)
364{
365 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
366}
367EXPORT_SYMBOL(filemap_fdatawrite);
368
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400369int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800370 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
372 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
373}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400374EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Randy Dunlap485bb992006-06-23 02:03:49 -0700376/**
377 * filemap_flush - mostly a non-blocking flush
378 * @mapping: target address_space
379 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 * This is a mostly non-blocking flush. Not suitable for data-integrity
381 * purposes - I/O may not be started against all dirty pages.
382 */
383int filemap_flush(struct address_space *mapping)
384{
385 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
386}
387EXPORT_SYMBOL(filemap_flush);
388
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800389static int __filemap_fdatawait_range(struct address_space *mapping,
390 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300392 pgoff_t index = start_byte >> PAGE_SHIFT;
393 pgoff_t end = end_byte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 struct pagevec pvec;
395 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800396 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200398 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700399 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
401 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 while ((index <= end) &&
403 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
404 PAGECACHE_TAG_WRITEBACK,
405 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
406 unsigned i;
407
408 for (i = 0; i < nr_pages; i++) {
409 struct page *page = pvec.pages[i];
410
411 /* until radix tree lookup accepts end_index */
412 if (page->index > end)
413 continue;
414
415 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800416 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 ret = -EIO;
418 }
419 pagevec_release(&pvec);
420 cond_resched();
421 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700422out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800423 return ret;
424}
425
426/**
427 * filemap_fdatawait_range - wait for writeback to complete
428 * @mapping: address space structure to wait for
429 * @start_byte: offset in bytes where the range starts
430 * @end_byte: offset in bytes where the range ends (inclusive)
431 *
432 * Walk the list of under-writeback pages of the given address space
433 * in the given range and wait for all of them. Check error status of
434 * the address space and return it.
435 *
436 * Since the error status of the address space is cleared by this function,
437 * callers are responsible for checking the return value and handling and/or
438 * reporting the error.
439 */
440int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
441 loff_t end_byte)
442{
443 int ret, ret2;
444
445 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700446 ret2 = filemap_check_errors(mapping);
447 if (!ret)
448 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
450 return ret;
451}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200452EXPORT_SYMBOL(filemap_fdatawait_range);
453
454/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800455 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
456 * @mapping: address space structure to wait for
457 *
458 * Walk the list of under-writeback pages of the given address space
459 * and wait for all of them. Unlike filemap_fdatawait(), this function
460 * does not clear error status of the address space.
461 *
462 * Use this function if callers don't handle errors themselves. Expected
463 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
464 * fsfreeze(8)
465 */
Jeff Layton76341ca2017-07-06 07:02:22 -0400466int filemap_fdatawait_keep_errors(struct address_space *mapping)
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800467{
468 loff_t i_size = i_size_read(mapping->host);
469
470 if (i_size == 0)
Jeff Layton76341ca2017-07-06 07:02:22 -0400471 return 0;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800472
473 __filemap_fdatawait_range(mapping, 0, i_size - 1);
Jeff Layton76341ca2017-07-06 07:02:22 -0400474 return filemap_check_and_keep_errors(mapping);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800475}
Jeff Layton76341ca2017-07-06 07:02:22 -0400476EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800477
478/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700479 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700481 *
482 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800483 * and wait for all of them. Check error status of the address space
484 * and return it.
485 *
486 * Since the error status of the address space is cleared by this function,
487 * callers are responsible for checking the return value and handling and/or
488 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 */
490int filemap_fdatawait(struct address_space *mapping)
491{
492 loff_t i_size = i_size_read(mapping->host);
493
494 if (i_size == 0)
495 return 0;
496
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200497 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
499EXPORT_SYMBOL(filemap_fdatawait);
500
501int filemap_write_and_wait(struct address_space *mapping)
502{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800503 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800505 if ((!dax_mapping(mapping) && mapping->nrpages) ||
506 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800507 err = filemap_fdatawrite(mapping);
508 /*
509 * Even if the above returned error, the pages may be
510 * written partially (e.g. -ENOSPC), so we wait for it.
511 * But the -EIO is special case, it may indicate the worst
512 * thing (e.g. bug) happened, so we avoid waiting for it.
513 */
514 if (err != -EIO) {
515 int err2 = filemap_fdatawait(mapping);
516 if (!err)
517 err = err2;
Jeff Laytoncbeaf952017-07-06 07:02:23 -0400518 } else {
519 /* Clear any previously stored errors */
520 filemap_check_errors(mapping);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800521 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700522 } else {
523 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800525 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800527EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Randy Dunlap485bb992006-06-23 02:03:49 -0700529/**
530 * filemap_write_and_wait_range - write out & wait on a file range
531 * @mapping: the address_space for the pages
532 * @lstart: offset in bytes where the range starts
533 * @lend: offset in bytes where the range ends (inclusive)
534 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800535 * Write out and wait upon file offsets lstart->lend, inclusive.
536 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -0300537 * Note that @lend is inclusive (describes the last byte to be written) so
Andrew Morton469eb4d2006-03-24 03:17:45 -0800538 * that this function can be used to write to the very end-of-file (end = -1).
539 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540int filemap_write_and_wait_range(struct address_space *mapping,
541 loff_t lstart, loff_t lend)
542{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800543 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800545 if ((!dax_mapping(mapping) && mapping->nrpages) ||
546 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800547 err = __filemap_fdatawrite_range(mapping, lstart, lend,
548 WB_SYNC_ALL);
549 /* See comment of filemap_write_and_wait() */
550 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200551 int err2 = filemap_fdatawait_range(mapping,
552 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800553 if (!err)
554 err = err2;
Jeff Laytoncbeaf952017-07-06 07:02:23 -0400555 } else {
556 /* Clear any previously stored errors */
557 filemap_check_errors(mapping);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800558 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700559 } else {
560 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800562 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563}
Chris Masonf6995582009-04-15 13:22:37 -0400564EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Randy Dunlap485bb992006-06-23 02:03:49 -0700566/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700567 * replace_page_cache_page - replace a pagecache page with a new one
568 * @old: page to be replaced
569 * @new: page to replace with
570 * @gfp_mask: allocation mode
571 *
572 * This function replaces a page in the pagecache with a new one. On
573 * success it acquires the pagecache reference for the new page and
574 * drops it for the old page. Both the old and new pages must be
575 * locked. This function does not add the new page to the LRU, the
576 * caller must do that.
577 *
578 * The remove + add is atomic. The only way this function can fail is
579 * memory allocation failure.
580 */
581int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
582{
583 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700584
Sasha Levin309381fea2014-01-23 15:52:54 -0800585 VM_BUG_ON_PAGE(!PageLocked(old), old);
586 VM_BUG_ON_PAGE(!PageLocked(new), new);
587 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700588
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700589 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
590 if (!error) {
591 struct address_space *mapping = old->mapping;
592 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400593 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700594
595 pgoff_t offset = old->index;
596 freepage = mapping->a_ops->freepage;
597
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300598 get_page(new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700599 new->mapping = mapping;
600 new->index = offset;
601
Greg Thelenc4843a72015-05-22 17:13:16 -0400602 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700603 __delete_from_page_cache(old, NULL);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700604 error = page_cache_tree_insert(mapping, new, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700605 BUG_ON(error);
Michal Hocko4165b9b2015-06-24 16:57:24 -0700606
607 /*
608 * hugetlb pages do not participate in page cache accounting.
609 */
610 if (!PageHuge(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700611 __inc_node_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700612 if (PageSwapBacked(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700613 __inc_node_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400614 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700615 mem_cgroup_migrate(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700616 radix_tree_preload_end();
617 if (freepage)
618 freepage(old);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300619 put_page(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700620 }
621
622 return error;
623}
624EXPORT_SYMBOL_GPL(replace_page_cache_page);
625
Johannes Weinera5289102014-04-03 14:47:51 -0700626static int __add_to_page_cache_locked(struct page *page,
627 struct address_space *mapping,
628 pgoff_t offset, gfp_t gfp_mask,
629 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Johannes Weiner00501b52014-08-08 14:19:20 -0700631 int huge = PageHuge(page);
632 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700633 int error;
634
Sasha Levin309381fea2014-01-23 15:52:54 -0800635 VM_BUG_ON_PAGE(!PageLocked(page), page);
636 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700637
Johannes Weiner00501b52014-08-08 14:19:20 -0700638 if (!huge) {
639 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800640 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700641 if (error)
642 return error;
643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Jan Kara5e4c0d972013-09-11 14:26:05 -0700645 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700646 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700647 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800648 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700649 return error;
650 }
651
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300652 get_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700653 page->mapping = mapping;
654 page->index = offset;
655
656 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700657 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700658 radix_tree_preload_end();
659 if (unlikely(error))
660 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700661
662 /* hugetlb pages do not participate in page cache accounting. */
663 if (!huge)
Mel Gorman11fb9982016-07-28 15:46:20 -0700664 __inc_node_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700665 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700666 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800667 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700668 trace_mm_filemap_add_to_page_cache(page);
669 return 0;
670err_insert:
671 page->mapping = NULL;
672 /* Leave page->index set: truncation relies upon it */
673 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700674 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800675 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300676 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 return error;
678}
Johannes Weinera5289102014-04-03 14:47:51 -0700679
680/**
681 * add_to_page_cache_locked - add a locked page to the pagecache
682 * @page: page to add
683 * @mapping: the page's address_space
684 * @offset: page index
685 * @gfp_mask: page allocation mode
686 *
687 * This function is used to add a page to the pagecache. It must be locked.
688 * This function does not add the page to the LRU. The caller must do that.
689 */
690int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
691 pgoff_t offset, gfp_t gfp_mask)
692{
693 return __add_to_page_cache_locked(page, mapping, offset,
694 gfp_mask, NULL);
695}
Nick Piggine2867812008-07-25 19:45:30 -0700696EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400699 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
Johannes Weinera5289102014-04-03 14:47:51 -0700701 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700702 int ret;
703
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800704 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700705 ret = __add_to_page_cache_locked(page, mapping, offset,
706 gfp_mask, &shadow);
707 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800708 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700709 else {
710 /*
711 * The page might have been evicted from cache only
712 * recently, in which case it should be activated like
713 * any other repeatedly accessed page.
Rik van Rielf0281a02016-05-20 16:56:25 -0700714 * The exception is pages getting rewritten; evicting other
715 * data from the working set, only to cache data that will
716 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700717 */
Rik van Rielf0281a02016-05-20 16:56:25 -0700718 if (!(gfp_mask & __GFP_WRITE) &&
719 shadow && workingset_refault(shadow)) {
Johannes Weinera5289102014-04-03 14:47:51 -0700720 SetPageActive(page);
721 workingset_activation(page);
722 } else
723 ClearPageActive(page);
724 lru_cache_add(page);
725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 return ret;
727}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300728EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Paul Jackson44110fe2006-03-24 03:16:04 -0800730#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700731struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800732{
Miao Xiec0ff7452010-05-24 14:32:08 -0700733 int n;
734 struct page *page;
735
Paul Jackson44110fe2006-03-24 03:16:04 -0800736 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700737 unsigned int cpuset_mems_cookie;
738 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700739 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700740 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700741 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700742 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700743
Miao Xiec0ff7452010-05-24 14:32:08 -0700744 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800745 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700746 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800747}
Nick Piggin2ae88142006-10-28 10:38:23 -0700748EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800749#endif
750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751/*
752 * In order to wait for pages to become available there must be
753 * waitqueues associated with pages. By using a hash table of
754 * waitqueues where the bucket discipline is to maintain all
755 * waiters on the same queue and wake all when any of the pages
756 * become available, and for the woken contexts to check to be
757 * sure the appropriate page became available, this saves space
758 * at a cost of "thundering herd" phenomena during rare hash
759 * collisions.
760 */
Nicholas Piggin62906022016-12-25 13:00:30 +1000761#define PAGE_WAIT_TABLE_BITS 8
762#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
763static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
764
765static wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
Nicholas Piggin62906022016-12-25 13:00:30 +1000767 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768}
Nicholas Piggin62906022016-12-25 13:00:30 +1000769
770void __init pagecache_init(void)
771{
772 int i;
773
774 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
775 init_waitqueue_head(&page_wait_table[i]);
776
777 page_writeback_init();
778}
779
780struct wait_page_key {
781 struct page *page;
782 int bit_nr;
783 int page_match;
784};
785
786struct wait_page_queue {
787 struct page *page;
788 int bit_nr;
789 wait_queue_t wait;
790};
791
792static int wake_page_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
793{
794 struct wait_page_key *key = arg;
795 struct wait_page_queue *wait_page
796 = container_of(wait, struct wait_page_queue, wait);
797
798 if (wait_page->page != key->page)
799 return 0;
800 key->page_match = 1;
801
802 if (wait_page->bit_nr != key->bit_nr)
803 return 0;
804 if (test_bit(key->bit_nr, &key->page->flags))
805 return 0;
806
807 return autoremove_wake_function(wait, mode, sync, key);
808}
809
Nicholas Piggin74d81bf2017-02-22 15:44:41 -0800810static void wake_up_page_bit(struct page *page, int bit_nr)
Nicholas Piggin62906022016-12-25 13:00:30 +1000811{
812 wait_queue_head_t *q = page_waitqueue(page);
813 struct wait_page_key key;
814 unsigned long flags;
815
816 key.page = page;
817 key.bit_nr = bit_nr;
818 key.page_match = 0;
819
820 spin_lock_irqsave(&q->lock, flags);
821 __wake_up_locked_key(q, TASK_NORMAL, &key);
822 /*
823 * It is possible for other pages to have collided on the waitqueue
824 * hash, so in that case check for a page match. That prevents a long-
825 * term waiter
826 *
827 * It is still possible to miss a case here, when we woke page waiters
828 * and removed them from the waitqueue, but there are still other
829 * page waiters.
830 */
831 if (!waitqueue_active(q) || !key.page_match) {
832 ClearPageWaiters(page);
833 /*
834 * It's possible to miss clearing Waiters here, when we woke
835 * our page waiters, but the hashed waitqueue has waiters for
836 * other pages on it.
837 *
838 * That's okay, it's a rare case. The next waker will clear it.
839 */
840 }
841 spin_unlock_irqrestore(&q->lock, flags);
842}
Nicholas Piggin74d81bf2017-02-22 15:44:41 -0800843
844static void wake_up_page(struct page *page, int bit)
845{
846 if (!PageWaiters(page))
847 return;
848 wake_up_page_bit(page, bit);
849}
Nicholas Piggin62906022016-12-25 13:00:30 +1000850
851static inline int wait_on_page_bit_common(wait_queue_head_t *q,
852 struct page *page, int bit_nr, int state, bool lock)
853{
854 struct wait_page_queue wait_page;
855 wait_queue_t *wait = &wait_page.wait;
856 int ret = 0;
857
858 init_wait(wait);
859 wait->func = wake_page_function;
860 wait_page.page = page;
861 wait_page.bit_nr = bit_nr;
862
863 for (;;) {
864 spin_lock_irq(&q->lock);
865
866 if (likely(list_empty(&wait->task_list))) {
867 if (lock)
868 __add_wait_queue_tail_exclusive(q, wait);
869 else
870 __add_wait_queue(q, wait);
871 SetPageWaiters(page);
872 }
873
874 set_current_state(state);
875
876 spin_unlock_irq(&q->lock);
877
878 if (likely(test_bit(bit_nr, &page->flags))) {
879 io_schedule();
880 if (unlikely(signal_pending_state(state, current))) {
881 ret = -EINTR;
882 break;
883 }
884 }
885
886 if (lock) {
887 if (!test_and_set_bit_lock(bit_nr, &page->flags))
888 break;
889 } else {
890 if (!test_bit(bit_nr, &page->flags))
891 break;
892 }
893 }
894
895 finish_wait(q, wait);
896
897 /*
898 * A signal could leave PageWaiters set. Clearing it here if
899 * !waitqueue_active would be possible (by open-coding finish_wait),
900 * but still fail to catch it in the case of wait hash collision. We
901 * already can fail to clear wait hash collision cases, so don't
902 * bother with signals either.
903 */
904
905 return ret;
906}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Harvey Harrison920c7a52008-02-04 22:29:26 -0800908void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
Nicholas Piggin62906022016-12-25 13:00:30 +1000910 wait_queue_head_t *q = page_waitqueue(page);
911 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912}
913EXPORT_SYMBOL(wait_on_page_bit);
914
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700915int wait_on_page_bit_killable(struct page *page, int bit_nr)
916{
Nicholas Piggin62906022016-12-25 13:00:30 +1000917 wait_queue_head_t *q = page_waitqueue(page);
918 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700919}
920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100922 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700923 * @page: Page defining the wait queue of interest
924 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100925 *
926 * Add an arbitrary @waiter to the wait queue for the nominated @page.
927 */
928void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
929{
930 wait_queue_head_t *q = page_waitqueue(page);
931 unsigned long flags;
932
933 spin_lock_irqsave(&q->lock, flags);
934 __add_wait_queue(q, waiter);
Nicholas Piggin62906022016-12-25 13:00:30 +1000935 SetPageWaiters(page);
David Howells385e1ca5f2009-04-03 16:42:39 +0100936 spin_unlock_irqrestore(&q->lock, flags);
937}
938EXPORT_SYMBOL_GPL(add_page_wait_queue);
939
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800940#ifndef clear_bit_unlock_is_negative_byte
941
942/*
943 * PG_waiters is the high bit in the same byte as PG_lock.
944 *
945 * On x86 (and on many other architectures), we can clear PG_lock and
946 * test the sign bit at the same time. But if the architecture does
947 * not support that special operation, we just do this all by hand
948 * instead.
949 *
950 * The read of PG_waiters has to be after (or concurrently with) PG_locked
951 * being cleared, but a memory barrier should be unneccssary since it is
952 * in the same byte as PG_locked.
953 */
954static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
955{
956 clear_bit_unlock(nr, mem);
957 /* smp_mb__after_atomic(); */
Olof Johansson98473f92016-12-29 14:16:07 -0800958 return test_bit(PG_waiters, mem);
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800959}
960
961#endif
962
David Howells385e1ca5f2009-04-03 16:42:39 +0100963/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700964 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 * @page: the page
966 *
967 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
968 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900969 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
971 *
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800972 * Note that this depends on PG_waiters being the sign bit in the byte
973 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
974 * clear the PG_locked bit and test PG_waiters at the same time fairly
975 * portably (architectures that do LL/SC can test any bit, while x86 can
976 * test the sign bit).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800978void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800980 BUILD_BUG_ON(PG_waiters != 7);
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800981 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800982 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800983 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
984 wake_up_page_bit(page, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985}
986EXPORT_SYMBOL(unlock_page);
987
Randy Dunlap485bb992006-06-23 02:03:49 -0700988/**
989 * end_page_writeback - end writeback against a page
990 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 */
992void end_page_writeback(struct page *page)
993{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700994 /*
995 * TestClearPageReclaim could be used here but it is an atomic
996 * operation and overkill in this particular case. Failing to
997 * shuffle a page marked for immediate reclaim is too mild to
998 * justify taking an atomic operation penalty at the end of
999 * ever page writeback.
1000 */
1001 if (PageReclaim(page)) {
1002 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001003 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -07001004 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001005
1006 if (!test_clear_page_writeback(page))
1007 BUG();
1008
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001009 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 wake_up_page(page, PG_writeback);
1011}
1012EXPORT_SYMBOL(end_page_writeback);
1013
Matthew Wilcox57d99842014-06-04 16:07:45 -07001014/*
1015 * After completing I/O on a page, call this routine to update the page
1016 * flags appropriately
1017 */
Jens Axboec11f0c02016-08-05 08:11:04 -06001018void page_endio(struct page *page, bool is_write, int err)
Matthew Wilcox57d99842014-06-04 16:07:45 -07001019{
Jens Axboec11f0c02016-08-05 08:11:04 -06001020 if (!is_write) {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001021 if (!err) {
1022 SetPageUptodate(page);
1023 } else {
1024 ClearPageUptodate(page);
1025 SetPageError(page);
1026 }
1027 unlock_page(page);
Mike Christieabf54542016-08-04 14:23:34 -06001028 } else {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001029 if (err) {
Minchan Kimdd8416c2017-02-24 14:59:59 -08001030 struct address_space *mapping;
1031
Matthew Wilcox57d99842014-06-04 16:07:45 -07001032 SetPageError(page);
Minchan Kimdd8416c2017-02-24 14:59:59 -08001033 mapping = page_mapping(page);
1034 if (mapping)
1035 mapping_set_error(mapping, err);
Matthew Wilcox57d99842014-06-04 16:07:45 -07001036 }
1037 end_page_writeback(page);
1038 }
1039}
1040EXPORT_SYMBOL_GPL(page_endio);
1041
Randy Dunlap485bb992006-06-23 02:03:49 -07001042/**
1043 * __lock_page - get a lock on the page, assuming we need to sleep to get it
Randy Dunlap87066752017-02-22 15:44:44 -08001044 * @__page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 */
Nicholas Piggin62906022016-12-25 13:00:30 +10001046void __lock_page(struct page *__page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
Nicholas Piggin62906022016-12-25 13:00:30 +10001048 struct page *page = compound_head(__page);
1049 wait_queue_head_t *q = page_waitqueue(page);
1050 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052EXPORT_SYMBOL(__lock_page);
1053
Nicholas Piggin62906022016-12-25 13:00:30 +10001054int __lock_page_killable(struct page *__page)
Matthew Wilcox2687a352007-12-06 11:18:49 -05001055{
Nicholas Piggin62906022016-12-25 13:00:30 +10001056 struct page *page = compound_head(__page);
1057 wait_queue_head_t *q = page_waitqueue(page);
1058 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001059}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +03001060EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001061
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001062/*
1063 * Return values:
1064 * 1 - page is locked; mmap_sem is still held.
1065 * 0 - page is not locked.
1066 * mmap_sem has been released (up_read()), unless flags had both
1067 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1068 * which case mmap_sem is still held.
1069 *
1070 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1071 * with the page locked and the mmap_sem unperturbed.
1072 */
Michel Lespinassed065bd82010-10-26 14:21:57 -07001073int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1074 unsigned int flags)
1075{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001076 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1077 /*
1078 * CAUTION! In this case, mmap_sem is not released
1079 * even though return 0.
1080 */
1081 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1082 return 0;
1083
1084 up_read(&mm->mmap_sem);
1085 if (flags & FAULT_FLAG_KILLABLE)
1086 wait_on_page_locked_killable(page);
1087 else
Gleb Natapov318b2752011-03-22 16:30:51 -07001088 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001089 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001090 } else {
1091 if (flags & FAULT_FLAG_KILLABLE) {
1092 int ret;
1093
1094 ret = __lock_page_killable(page);
1095 if (ret) {
1096 up_read(&mm->mmap_sem);
1097 return 0;
1098 }
1099 } else
1100 __lock_page(page);
1101 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -07001102 }
1103}
1104
Randy Dunlap485bb992006-06-23 02:03:49 -07001105/**
Johannes Weinere7b563b2014-04-03 14:47:44 -07001106 * page_cache_next_hole - find the next hole (not-present entry)
1107 * @mapping: mapping
1108 * @index: index
1109 * @max_scan: maximum range to search
1110 *
1111 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1112 * lowest indexed hole.
1113 *
1114 * Returns: the index of the hole if found, otherwise returns an index
1115 * outside of the set specified (in which case 'return - index >=
1116 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1117 * be returned.
1118 *
1119 * page_cache_next_hole may be called under rcu_read_lock. However,
1120 * like radix_tree_gang_lookup, this will not atomically search a
1121 * snapshot of the tree at a single point in time. For example, if a
1122 * hole is created at index 5, then subsequently a hole is created at
1123 * index 10, page_cache_next_hole covering both indexes may return 10
1124 * if called under rcu_read_lock.
1125 */
1126pgoff_t page_cache_next_hole(struct address_space *mapping,
1127 pgoff_t index, unsigned long max_scan)
1128{
1129 unsigned long i;
1130
1131 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001132 struct page *page;
1133
1134 page = radix_tree_lookup(&mapping->page_tree, index);
1135 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001136 break;
1137 index++;
1138 if (index == 0)
1139 break;
1140 }
1141
1142 return index;
1143}
1144EXPORT_SYMBOL(page_cache_next_hole);
1145
1146/**
1147 * page_cache_prev_hole - find the prev hole (not-present entry)
1148 * @mapping: mapping
1149 * @index: index
1150 * @max_scan: maximum range to search
1151 *
1152 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1153 * the first hole.
1154 *
1155 * Returns: the index of the hole if found, otherwise returns an index
1156 * outside of the set specified (in which case 'index - return >=
1157 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1158 * will be returned.
1159 *
1160 * page_cache_prev_hole may be called under rcu_read_lock. However,
1161 * like radix_tree_gang_lookup, this will not atomically search a
1162 * snapshot of the tree at a single point in time. For example, if a
1163 * hole is created at index 10, then subsequently a hole is created at
1164 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1165 * called under rcu_read_lock.
1166 */
1167pgoff_t page_cache_prev_hole(struct address_space *mapping,
1168 pgoff_t index, unsigned long max_scan)
1169{
1170 unsigned long i;
1171
1172 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001173 struct page *page;
1174
1175 page = radix_tree_lookup(&mapping->page_tree, index);
1176 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001177 break;
1178 index--;
1179 if (index == ULONG_MAX)
1180 break;
1181 }
1182
1183 return index;
1184}
1185EXPORT_SYMBOL(page_cache_prev_hole);
1186
1187/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001188 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001189 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001190 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001191 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001192 * Looks up the page cache slot at @mapping & @offset. If there is a
1193 * page cache page, it is returned with an increased refcount.
1194 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001195 * If the slot holds a shadow entry of a previously evicted page, or a
1196 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001197 *
1198 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001200struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201{
Nick Piggina60637c2008-07-25 19:45:31 -07001202 void **pagep;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001203 struct page *head, *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Nick Piggina60637c2008-07-25 19:45:31 -07001205 rcu_read_lock();
1206repeat:
1207 page = NULL;
1208 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1209 if (pagep) {
1210 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001211 if (unlikely(!page))
1212 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001213 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001214 if (radix_tree_deref_retry(page))
1215 goto repeat;
1216 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001217 * A shadow entry of a recently evicted page,
1218 * or a swap entry from shmem/tmpfs. Return
1219 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001220 */
1221 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001222 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001223
1224 head = compound_head(page);
1225 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001226 goto repeat;
1227
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001228 /* The page was split under us? */
1229 if (compound_head(page) != head) {
1230 put_page(head);
1231 goto repeat;
1232 }
1233
Nick Piggina60637c2008-07-25 19:45:31 -07001234 /*
1235 * Has the page moved?
1236 * This is part of the lockless pagecache protocol. See
1237 * include/linux/pagemap.h for details.
1238 */
1239 if (unlikely(page != *pagep)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001240 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001241 goto repeat;
1242 }
1243 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001244out:
Nick Piggina60637c2008-07-25 19:45:31 -07001245 rcu_read_unlock();
1246
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 return page;
1248}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001249EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Randy Dunlap485bb992006-06-23 02:03:49 -07001251/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001252 * find_lock_entry - locate, pin and lock a page cache entry
1253 * @mapping: the address_space to search
1254 * @offset: the page cache index
1255 *
1256 * Looks up the page cache slot at @mapping & @offset. If there is a
1257 * page cache page, it is returned locked and with an increased
1258 * refcount.
1259 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001260 * If the slot holds a shadow entry of a previously evicted page, or a
1261 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001262 *
1263 * Otherwise, %NULL is returned.
1264 *
1265 * find_lock_entry() may sleep.
1266 */
1267struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268{
1269 struct page *page;
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001272 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001273 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001274 lock_page(page);
1275 /* Has the page been truncated? */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001276 if (unlikely(page_mapping(page) != mapping)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001277 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001278 put_page(page);
Nick Piggina60637c2008-07-25 19:45:31 -07001279 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001281 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 return page;
1284}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001285EXPORT_SYMBOL(find_lock_entry);
1286
1287/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001288 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001289 * @mapping: the address_space to search
1290 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001291 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001292 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001293 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001294 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001295 *
Randy Dunlap75325182014-07-30 16:08:37 -07001296 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001297 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -03001298 * @fgp_flags can be:
1299 *
1300 * - FGP_ACCESSED: the page will be marked accessed
1301 * - FGP_LOCK: Page is return locked
1302 * - FGP_CREAT: If page is not present then a new page is allocated using
1303 * @gfp_mask and added to the page cache and the VM's LRU
1304 * list. The page is returned locked and with an increased
1305 * refcount. Otherwise, NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001306 *
1307 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1308 * if the GFP flags specified for FGP_CREAT are atomic.
1309 *
1310 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001311 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001312struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001313 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314{
Nick Piggineb2be182007-10-16 01:24:57 -07001315 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001318 page = find_get_entry(mapping, offset);
1319 if (radix_tree_exceptional_entry(page))
1320 page = NULL;
1321 if (!page)
1322 goto no_page;
1323
1324 if (fgp_flags & FGP_LOCK) {
1325 if (fgp_flags & FGP_NOWAIT) {
1326 if (!trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001327 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001328 return NULL;
1329 }
1330 } else {
1331 lock_page(page);
1332 }
1333
1334 /* Has the page been truncated? */
1335 if (unlikely(page->mapping != mapping)) {
1336 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001337 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001338 goto repeat;
1339 }
1340 VM_BUG_ON_PAGE(page->index != offset, page);
1341 }
1342
1343 if (page && (fgp_flags & FGP_ACCESSED))
1344 mark_page_accessed(page);
1345
1346no_page:
1347 if (!page && (fgp_flags & FGP_CREAT)) {
1348 int err;
1349 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001350 gfp_mask |= __GFP_WRITE;
1351 if (fgp_flags & FGP_NOFS)
1352 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001353
Michal Hocko45f87de2014-12-29 20:30:35 +01001354 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001355 if (!page)
1356 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001357
1358 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1359 fgp_flags |= FGP_LOCK;
1360
Hugh Dickinseb39d612014-08-06 16:06:43 -07001361 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001362 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001363 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001364
Michal Hocko45f87de2014-12-29 20:30:35 +01001365 err = add_to_page_cache_lru(page, mapping, offset,
1366 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001367 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001368 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07001369 page = NULL;
1370 if (err == -EEXIST)
1371 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 return page;
1376}
Mel Gorman2457aec2014-06-04 16:10:31 -07001377EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
1379/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001380 * find_get_entries - gang pagecache lookup
1381 * @mapping: The address_space to search
1382 * @start: The starting page cache index
1383 * @nr_entries: The maximum number of entries
1384 * @entries: Where the resulting entries are placed
1385 * @indices: The cache indices corresponding to the entries in @entries
1386 *
1387 * find_get_entries() will search for and return a group of up to
1388 * @nr_entries entries in the mapping. The entries are placed at
1389 * @entries. find_get_entries() takes a reference against any actual
1390 * pages it returns.
1391 *
1392 * The search returns a group of mapping-contiguous page cache entries
1393 * with ascending indexes. There may be holes in the indices due to
1394 * not-present pages.
1395 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001396 * Any shadow entries of evicted pages, or swap entries from
1397 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001398 *
1399 * find_get_entries() returns the number of pages and shadow entries
1400 * which were found.
1401 */
1402unsigned find_get_entries(struct address_space *mapping,
1403 pgoff_t start, unsigned int nr_entries,
1404 struct page **entries, pgoff_t *indices)
1405{
1406 void **slot;
1407 unsigned int ret = 0;
1408 struct radix_tree_iter iter;
1409
1410 if (!nr_entries)
1411 return 0;
1412
1413 rcu_read_lock();
Johannes Weiner0cd61442014-04-03 14:47:46 -07001414 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001415 struct page *head, *page;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001416repeat:
1417 page = radix_tree_deref_slot(slot);
1418 if (unlikely(!page))
1419 continue;
1420 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001421 if (radix_tree_deref_retry(page)) {
1422 slot = radix_tree_iter_retry(&iter);
1423 continue;
1424 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001425 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -08001426 * A shadow entry of a recently evicted page, a swap
1427 * entry from shmem/tmpfs or a DAX entry. Return it
1428 * without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001429 */
1430 goto export;
1431 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001432
1433 head = compound_head(page);
1434 if (!page_cache_get_speculative(head))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001435 goto repeat;
1436
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001437 /* The page was split under us? */
1438 if (compound_head(page) != head) {
1439 put_page(head);
1440 goto repeat;
1441 }
1442
Johannes Weiner0cd61442014-04-03 14:47:46 -07001443 /* Has the page moved? */
1444 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001445 put_page(head);
Johannes Weiner0cd61442014-04-03 14:47:46 -07001446 goto repeat;
1447 }
1448export:
1449 indices[ret] = iter.index;
1450 entries[ret] = page;
1451 if (++ret == nr_entries)
1452 break;
1453 }
1454 rcu_read_unlock();
1455 return ret;
1456}
1457
1458/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 * find_get_pages - gang pagecache lookup
1460 * @mapping: The address_space to search
1461 * @start: The starting page index
1462 * @nr_pages: The maximum number of pages
1463 * @pages: Where the resulting pages are placed
1464 *
1465 * find_get_pages() will search for and return a group of up to
1466 * @nr_pages pages in the mapping. The pages are placed at @pages.
1467 * find_get_pages() takes a reference against the returned pages.
1468 *
1469 * The search returns a group of mapping-contiguous pages with ascending
1470 * indexes. There may be holes in the indices due to not-present pages.
1471 *
1472 * find_get_pages() returns the number of pages which were found.
1473 */
1474unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1475 unsigned int nr_pages, struct page **pages)
1476{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001477 struct radix_tree_iter iter;
1478 void **slot;
1479 unsigned ret = 0;
1480
1481 if (unlikely(!nr_pages))
1482 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483
Nick Piggina60637c2008-07-25 19:45:31 -07001484 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001485 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001486 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001487repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001488 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001489 if (unlikely(!page))
1490 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001491
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001492 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001493 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001494 slot = radix_tree_iter_retry(&iter);
1495 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001496 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001497 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001498 * A shadow entry of a recently evicted page,
1499 * or a swap entry from shmem/tmpfs. Skip
1500 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001501 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001502 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001503 }
Nick Piggina60637c2008-07-25 19:45:31 -07001504
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001505 head = compound_head(page);
1506 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001507 goto repeat;
1508
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001509 /* The page was split under us? */
1510 if (compound_head(page) != head) {
1511 put_page(head);
1512 goto repeat;
1513 }
1514
Nick Piggina60637c2008-07-25 19:45:31 -07001515 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001516 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001517 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001518 goto repeat;
1519 }
1520
1521 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001522 if (++ret == nr_pages)
1523 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001524 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001525
Nick Piggina60637c2008-07-25 19:45:31 -07001526 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 return ret;
1528}
1529
Jens Axboeebf43502006-04-27 08:46:01 +02001530/**
1531 * find_get_pages_contig - gang contiguous pagecache lookup
1532 * @mapping: The address_space to search
1533 * @index: The starting page index
1534 * @nr_pages: The maximum number of pages
1535 * @pages: Where the resulting pages are placed
1536 *
1537 * find_get_pages_contig() works exactly like find_get_pages(), except
1538 * that the returned number of pages are guaranteed to be contiguous.
1539 *
1540 * find_get_pages_contig() returns the number of pages which were found.
1541 */
1542unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1543 unsigned int nr_pages, struct page **pages)
1544{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001545 struct radix_tree_iter iter;
1546 void **slot;
1547 unsigned int ret = 0;
1548
1549 if (unlikely(!nr_pages))
1550 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001551
Nick Piggina60637c2008-07-25 19:45:31 -07001552 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001553 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001554 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001555repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001556 page = radix_tree_deref_slot(slot);
1557 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001558 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001559 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001560
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001561 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001562 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001563 slot = radix_tree_iter_retry(&iter);
1564 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001565 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001566 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001567 * A shadow entry of a recently evicted page,
1568 * or a swap entry from shmem/tmpfs. Stop
1569 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001570 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001571 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001572 }
Nick Piggina60637c2008-07-25 19:45:31 -07001573
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001574 head = compound_head(page);
1575 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001576 goto repeat;
1577
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001578 /* The page was split under us? */
1579 if (compound_head(page) != head) {
1580 put_page(head);
1581 goto repeat;
1582 }
1583
Nick Piggina60637c2008-07-25 19:45:31 -07001584 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001585 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001586 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001587 goto repeat;
1588 }
1589
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001590 /*
1591 * must check mapping and index after taking the ref.
1592 * otherwise we can get both false positives and false
1593 * negatives, which is just confusing to the caller.
1594 */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001595 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001596 put_page(page);
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001597 break;
1598 }
1599
Nick Piggina60637c2008-07-25 19:45:31 -07001600 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001601 if (++ret == nr_pages)
1602 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001603 }
Nick Piggina60637c2008-07-25 19:45:31 -07001604 rcu_read_unlock();
1605 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001606}
David Howellsef71c152007-05-09 02:33:44 -07001607EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001608
Randy Dunlap485bb992006-06-23 02:03:49 -07001609/**
1610 * find_get_pages_tag - find and return pages that match @tag
1611 * @mapping: the address_space to search
1612 * @index: the starting page index
1613 * @tag: the tag index
1614 * @nr_pages: the maximum number of pages
1615 * @pages: where the resulting pages are placed
1616 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001618 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 */
1620unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1621 int tag, unsigned int nr_pages, struct page **pages)
1622{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001623 struct radix_tree_iter iter;
1624 void **slot;
1625 unsigned ret = 0;
1626
1627 if (unlikely(!nr_pages))
1628 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629
Nick Piggina60637c2008-07-25 19:45:31 -07001630 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001631 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1632 &iter, *index, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001633 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001634repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001635 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001636 if (unlikely(!page))
1637 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001638
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001639 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001640 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001641 slot = radix_tree_iter_retry(&iter);
1642 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001643 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001644 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001645 * A shadow entry of a recently evicted page.
1646 *
1647 * Those entries should never be tagged, but
1648 * this tree walk is lockless and the tags are
1649 * looked up in bulk, one radix tree node at a
1650 * time, so there is a sizable window for page
1651 * reclaim to evict a page we saw tagged.
1652 *
1653 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001654 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001655 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001656 }
Nick Piggina60637c2008-07-25 19:45:31 -07001657
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001658 head = compound_head(page);
1659 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001660 goto repeat;
1661
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001662 /* The page was split under us? */
1663 if (compound_head(page) != head) {
1664 put_page(head);
1665 goto repeat;
1666 }
1667
Nick Piggina60637c2008-07-25 19:45:31 -07001668 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001669 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001670 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001671 goto repeat;
1672 }
1673
1674 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001675 if (++ret == nr_pages)
1676 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001677 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001678
Nick Piggina60637c2008-07-25 19:45:31 -07001679 rcu_read_unlock();
1680
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 if (ret)
1682 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 return ret;
1685}
David Howellsef71c152007-05-09 02:33:44 -07001686EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001688/**
1689 * find_get_entries_tag - find and return entries that match @tag
1690 * @mapping: the address_space to search
1691 * @start: the starting page cache index
1692 * @tag: the tag index
1693 * @nr_entries: the maximum number of entries
1694 * @entries: where the resulting entries are placed
1695 * @indices: the cache indices corresponding to the entries in @entries
1696 *
1697 * Like find_get_entries, except we only return entries which are tagged with
1698 * @tag.
1699 */
1700unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1701 int tag, unsigned int nr_entries,
1702 struct page **entries, pgoff_t *indices)
1703{
1704 void **slot;
1705 unsigned int ret = 0;
1706 struct radix_tree_iter iter;
1707
1708 if (!nr_entries)
1709 return 0;
1710
1711 rcu_read_lock();
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001712 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1713 &iter, start, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001714 struct page *head, *page;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001715repeat:
1716 page = radix_tree_deref_slot(slot);
1717 if (unlikely(!page))
1718 continue;
1719 if (radix_tree_exception(page)) {
1720 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001721 slot = radix_tree_iter_retry(&iter);
1722 continue;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001723 }
1724
1725 /*
1726 * A shadow entry of a recently evicted page, a swap
1727 * entry from shmem/tmpfs or a DAX entry. Return it
1728 * without attempting to raise page count.
1729 */
1730 goto export;
1731 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001732
1733 head = compound_head(page);
1734 if (!page_cache_get_speculative(head))
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001735 goto repeat;
1736
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001737 /* The page was split under us? */
1738 if (compound_head(page) != head) {
1739 put_page(head);
1740 goto repeat;
1741 }
1742
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001743 /* Has the page moved? */
1744 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001745 put_page(head);
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001746 goto repeat;
1747 }
1748export:
1749 indices[ret] = iter.index;
1750 entries[ret] = page;
1751 if (++ret == nr_entries)
1752 break;
1753 }
1754 rcu_read_unlock();
1755 return ret;
1756}
1757EXPORT_SYMBOL(find_get_entries_tag);
1758
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001759/*
1760 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1761 * a _large_ part of the i/o request. Imagine the worst scenario:
1762 *
1763 * ---R__________________________________________B__________
1764 * ^ reading here ^ bad block(assume 4k)
1765 *
1766 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1767 * => failing the whole request => read(R) => read(R+1) =>
1768 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1769 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1770 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1771 *
1772 * It is going insane. Fix it by quickly scaling down the readahead size.
1773 */
1774static void shrink_readahead_size_eio(struct file *filp,
1775 struct file_ra_state *ra)
1776{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001777 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001778}
1779
Randy Dunlap485bb992006-06-23 02:03:49 -07001780/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001781 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001782 * @filp: the file to read
1783 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001784 * @iter: data destination
1785 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001786 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001788 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 *
1790 * This is really ugly. But the goto's actually try to clarify some
1791 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 */
Al Viro6e58e792014-02-03 17:07:03 -05001793static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1794 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001796 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001798 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001799 pgoff_t index;
1800 pgoff_t last_index;
1801 pgoff_t prev_index;
1802 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001803 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001804 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
Wei Fangc2a97372016-10-07 17:01:52 -07001806 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
Linus Torvaldsd05c5f72016-12-14 12:45:25 -08001807 return 0;
Wei Fangc2a97372016-10-07 17:01:52 -07001808 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1809
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001810 index = *ppos >> PAGE_SHIFT;
1811 prev_index = ra->prev_pos >> PAGE_SHIFT;
1812 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1813 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1814 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 for (;;) {
1817 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001818 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001819 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 unsigned long nr, ret;
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823find_page:
Michal Hocko5abf1862017-02-03 13:13:29 -08001824 if (fatal_signal_pending(current)) {
1825 error = -EINTR;
1826 goto out;
1827 }
1828
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001830 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001831 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001832 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001833 index, last_index - index);
1834 page = find_get_page(mapping, index);
1835 if (unlikely(page == NULL))
1836 goto no_cached_page;
1837 }
1838 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001839 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001840 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001841 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001843 if (!PageUptodate(page)) {
Mel Gormanebded022016-03-15 14:55:39 -07001844 /*
1845 * See comment in do_read_cache_page on why
1846 * wait_on_page_locked is used to avoid unnecessarily
1847 * serialisations and why it's safe.
1848 */
Bart Van Asschec4b209a2016-10-07 16:58:33 -07001849 error = wait_on_page_locked_killable(page);
1850 if (unlikely(error))
1851 goto readpage_error;
Mel Gormanebded022016-03-15 14:55:39 -07001852 if (PageUptodate(page))
1853 goto page_ok;
1854
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001855 if (inode->i_blkbits == PAGE_SHIFT ||
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001856 !mapping->a_ops->is_partially_uptodate)
1857 goto page_not_up_to_date;
Eryu Guan6d6d36b2016-11-01 15:43:07 +08001858 /* pipes can't handle partially uptodate pages */
1859 if (unlikely(iter->type & ITER_PIPE))
1860 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001861 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001862 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001863 /* Did it get truncated before we got the lock? */
1864 if (!page->mapping)
1865 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001866 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001867 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001868 goto page_not_up_to_date_locked;
1869 unlock_page(page);
1870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001872 /*
1873 * i_size must be checked after we know the page is Uptodate.
1874 *
1875 * Checking i_size after the check allows us to calculate
1876 * the correct value for "nr", which means the zero-filled
1877 * part of the page is not copied back to userspace (unless
1878 * another truncate extends the file - this is desired though).
1879 */
1880
1881 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001882 end_index = (isize - 1) >> PAGE_SHIFT;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001883 if (unlikely(!isize || index > end_index)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001884 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001885 goto out;
1886 }
1887
1888 /* nr is the maximum number of bytes to copy from this page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001889 nr = PAGE_SIZE;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001890 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001891 nr = ((isize - 1) & ~PAGE_MASK) + 1;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001892 if (nr <= offset) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001893 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001894 goto out;
1895 }
1896 }
1897 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
1899 /* If users can be writing to this page using arbitrary
1900 * virtual addresses, take care about potential aliasing
1901 * before reading the page on the kernel side.
1902 */
1903 if (mapping_writably_mapped(mapping))
1904 flush_dcache_page(page);
1905
1906 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001907 * When a sequential read accesses a page several times,
1908 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 */
Jan Karaec0f1632007-05-06 14:49:25 -07001910 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 mark_page_accessed(page);
1912 prev_index = index;
1913
1914 /*
1915 * Ok, we have the page, and it's up-to-date, so
1916 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 */
Al Viro6e58e792014-02-03 17:07:03 -05001918
1919 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001921 index += offset >> PAGE_SHIFT;
1922 offset &= ~PAGE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001923 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001925 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001926 written += ret;
1927 if (!iov_iter_count(iter))
1928 goto out;
1929 if (ret < nr) {
1930 error = -EFAULT;
1931 goto out;
1932 }
1933 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
1935page_not_up_to_date:
1936 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001937 error = lock_page_killable(page);
1938 if (unlikely(error))
1939 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001941page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001942 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 if (!page->mapping) {
1944 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001945 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 continue;
1947 }
1948
1949 /* Did somebody else fill it already? */
1950 if (PageUptodate(page)) {
1951 unlock_page(page);
1952 goto page_ok;
1953 }
1954
1955readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001956 /*
1957 * A previous I/O error may have been due to temporary
1958 * failures, eg. multipath errors.
1959 * PG_error will be set again if readpage fails.
1960 */
1961 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 /* Start the actual read. The read will unlock the page. */
1963 error = mapping->a_ops->readpage(filp, page);
1964
Zach Brown994fc28c2005-12-15 14:28:17 -08001965 if (unlikely(error)) {
1966 if (error == AOP_TRUNCATED_PAGE) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001967 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001968 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001969 goto find_page;
1970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001975 error = lock_page_killable(page);
1976 if (unlikely(error))
1977 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 if (!PageUptodate(page)) {
1979 if (page->mapping == NULL) {
1980 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001981 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 */
1983 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001984 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 goto find_page;
1986 }
1987 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001988 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001989 error = -EIO;
1990 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 }
1992 unlock_page(page);
1993 }
1994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 goto page_ok;
1996
1997readpage_error:
1998 /* UHHUH! A synchronous read error occurred. Report it */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001999 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 goto out;
2001
2002no_cached_page:
2003 /*
2004 * Ok, it wasn't cached, so we need to create a new
2005 * page..
2006 */
Nick Piggineb2be182007-10-16 01:24:57 -07002007 page = page_cache_alloc_cold(mapping);
2008 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05002009 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07002010 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 }
Michal Hocko6afdb852015-06-24 16:58:06 -07002012 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08002013 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 if (error) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002015 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05002016 if (error == -EEXIST) {
2017 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05002019 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 goto out;
2021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 goto readpage;
2023 }
2024
2025out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07002026 ra->prev_pos = prev_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002027 ra->prev_pos <<= PAGE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07002028 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002030 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07002031 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05002032 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033}
2034
Randy Dunlap485bb992006-06-23 02:03:49 -07002035/**
Al Viro6abd2322014-04-04 14:20:57 -04002036 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07002037 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04002038 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07002039 *
Al Viro6abd2322014-04-04 14:20:57 -04002040 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 * that can use the page cache directly.
2042 */
2043ssize_t
Al Viroed978a82014-03-05 22:53:04 -05002044generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045{
Al Viroed978a82014-03-05 22:53:04 -05002046 struct file *file = iocb->ki_filp;
2047 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07002048 size_t count = iov_iter_count(iter);
2049
2050 if (!count)
2051 goto out; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Al Viro2ba48ce2015-04-09 13:52:01 -04002053 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05002054 struct address_space *mapping = file->f_mapping;
2055 struct inode *inode = mapping->host;
Badari Pulavarty543ade12006-09-30 23:28:48 -07002056 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 size = i_size_read(inode);
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002059 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
2060 iocb->ki_pos + count - 1);
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002061 if (retval < 0)
2062 goto out;
Al Viroed978a82014-03-05 22:53:04 -05002063
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002064 file_accessed(file);
2065
Al Viro5ecda132017-04-13 14:13:36 -04002066 retval = mapping->a_ops->direct_IO(iocb, iter);
Al Viroc3a69022016-10-10 13:26:27 -04002067 if (retval >= 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002068 iocb->ki_pos += retval;
Al Viro5ecda132017-04-13 14:13:36 -04002069 count -= retval;
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002070 }
Al Viro5b47d592017-05-08 13:54:47 -04002071 iov_iter_revert(iter, count - iov_iter_count(iter));
Josef Bacik66f998f2010-05-23 11:00:54 -04002072
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002073 /*
2074 * Btrfs can have a short DIO read if we encounter
2075 * compressed extents, so if there was an error, or if
2076 * we've already read everything we wanted to, or if
2077 * there was a short read because we hit EOF, go ahead
2078 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002079 * the rest of the read. Buffered reads will not work for
2080 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002081 */
Al Viro5ecda132017-04-13 14:13:36 -04002082 if (retval < 0 || !count || iocb->ki_pos >= size ||
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002083 IS_DAX(inode))
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002084 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 }
2086
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002087 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088out:
2089 return retval;
2090}
Al Viroed978a82014-03-05 22:53:04 -05002091EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07002094/**
2095 * page_cache_read - adds requested page to the page cache if not already there
2096 * @file: file to read
2097 * @offset: page index
Randy Dunlap62eb3202016-02-11 16:12:58 -08002098 * @gfp_mask: memory allocation flags
Randy Dunlap485bb992006-06-23 02:03:49 -07002099 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 * This adds the requested page to the page cache if it isn't already there,
2101 * and schedules an I/O to read in its contents from disk.
2102 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002103static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
2105 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07002106 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08002107 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
Zach Brown994fc28c2005-12-15 14:28:17 -08002109 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08002110 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08002111 if (!page)
2112 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Michal Hockoc20cd452016-01-14 15:20:12 -08002114 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
Zach Brown994fc28c2005-12-15 14:28:17 -08002115 if (ret == 0)
2116 ret = mapping->a_ops->readpage(file, page);
2117 else if (ret == -EEXIST)
2118 ret = 0; /* losing race to add is OK */
2119
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002120 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
Zach Brown994fc28c2005-12-15 14:28:17 -08002122 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07002123
Zach Brown994fc28c2005-12-15 14:28:17 -08002124 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125}
2126
2127#define MMAP_LOTSAMISS (100)
2128
Linus Torvaldsef00e082009-06-16 15:31:25 -07002129/*
2130 * Synchronous readahead happens when we don't even find
2131 * a page in the page cache at all.
2132 */
2133static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2134 struct file_ra_state *ra,
2135 struct file *file,
2136 pgoff_t offset)
2137{
Linus Torvaldsef00e082009-06-16 15:31:25 -07002138 struct address_space *mapping = file->f_mapping;
2139
2140 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002141 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002142 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002143 if (!ra->ra_pages)
2144 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002145
Joe Perches64363aa2013-07-08 16:00:18 -07002146 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07002147 page_cache_sync_readahead(mapping, ra, file, offset,
2148 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002149 return;
2150 }
2151
Andi Kleen207d04b2011-05-24 17:12:29 -07002152 /* Avoid banging the cache line if not needed */
2153 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002154 ra->mmap_miss++;
2155
2156 /*
2157 * Do we miss much more than hit in this file? If so,
2158 * stop bothering with read-ahead. It will only hurt.
2159 */
2160 if (ra->mmap_miss > MMAP_LOTSAMISS)
2161 return;
2162
Wu Fengguangd30a1102009-06-16 15:31:30 -07002163 /*
2164 * mmap read-around
2165 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08002166 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2167 ra->size = ra->ra_pages;
2168 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002169 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002170}
2171
2172/*
2173 * Asynchronous readahead happens when we find the page and PG_readahead,
2174 * so we want to possibly extend the readahead further..
2175 */
2176static void do_async_mmap_readahead(struct vm_area_struct *vma,
2177 struct file_ra_state *ra,
2178 struct file *file,
2179 struct page *page,
2180 pgoff_t offset)
2181{
2182 struct address_space *mapping = file->f_mapping;
2183
2184 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002185 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002186 return;
2187 if (ra->mmap_miss > 0)
2188 ra->mmap_miss--;
2189 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07002190 page_cache_async_readahead(mapping, ra, file,
2191 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002192}
2193
Randy Dunlap485bb992006-06-23 02:03:49 -07002194/**
Nick Piggin54cb8822007-07-19 01:46:59 -07002195 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07002196 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07002197 *
Nick Piggin54cb8822007-07-19 01:46:59 -07002198 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 * mapped memory region to read in file data during a page fault.
2200 *
2201 * The goto's are kind of ugly, but this streamlines the normal case of having
2202 * it in the page cache, and handles the special cases reasonably without
2203 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002204 *
2205 * vma->vm_mm->mmap_sem must be held on entry.
2206 *
2207 * If our return value has VM_FAULT_RETRY set, it's because
2208 * lock_page_or_retry() returned 0.
2209 * The mmap_sem has usually been released in this case.
2210 * See __lock_page_or_retry() for the exception.
2211 *
2212 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2213 * has not been released.
2214 *
2215 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 */
Dave Jiang11bac802017-02-24 14:56:41 -08002217int filemap_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218{
2219 int error;
Dave Jiang11bac802017-02-24 14:56:41 -08002220 struct file *file = vmf->vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 struct address_space *mapping = file->f_mapping;
2222 struct file_ra_state *ra = &file->f_ra;
2223 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002224 pgoff_t offset = vmf->pgoff;
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002225 pgoff_t max_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 struct page *page;
Nick Piggin83c54072007-07-19 01:47:05 -07002227 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002229 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2230 if (unlikely(offset >= max_off))
Linus Torvalds5307cc12007-10-31 09:19:46 -07002231 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002234 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002236 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002237 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002239 * We found the page, so try async readahead before
2240 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 */
Dave Jiang11bac802017-02-24 14:56:41 -08002242 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002243 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002244 /* No page in the page cache at all */
Dave Jiang11bac802017-02-24 14:56:41 -08002245 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002246 count_vm_event(PGMAJFAULT);
Dave Jiang11bac802017-02-24 14:56:41 -08002247 mem_cgroup_count_vm_event(vmf->vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002248 ret = VM_FAULT_MAJOR;
2249retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07002250 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 if (!page)
2252 goto no_cached_page;
2253 }
2254
Dave Jiang11bac802017-02-24 14:56:41 -08002255 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002256 put_page(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002257 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07002258 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07002259
2260 /* Did it get truncated? */
2261 if (unlikely(page->mapping != mapping)) {
2262 unlock_page(page);
2263 put_page(page);
2264 goto retry_find;
2265 }
Sasha Levin309381fea2014-01-23 15:52:54 -08002266 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002267
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002269 * We have a locked page in the page cache, now we need to check
2270 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 */
Nick Piggind00806b2007-07-19 01:46:57 -07002272 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 goto page_not_uptodate;
2274
Linus Torvaldsef00e082009-06-16 15:31:25 -07002275 /*
2276 * Found the page and have a reference on it.
2277 * We must recheck i_size under page lock.
2278 */
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002279 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2280 if (unlikely(offset >= max_off)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002281 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002282 put_page(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002283 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002284 }
2285
Nick Piggind0217ac2007-07-19 01:47:03 -07002286 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002287 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289no_cached_page:
2290 /*
2291 * We're only likely to ever get here if MADV_RANDOM is in
2292 * effect.
2293 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002294 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
2296 /*
2297 * The page we want has now been added to the page cache.
2298 * In the unlikely event that someone removed it in the
2299 * meantime, we'll just come back here and read it again.
2300 */
2301 if (error >= 0)
2302 goto retry_find;
2303
2304 /*
2305 * An error return from page_cache_read can result if the
2306 * system is low on memory, or a problem occurs while trying
2307 * to schedule I/O.
2308 */
2309 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002310 return VM_FAULT_OOM;
2311 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
2313page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 /*
2315 * Umm, take care of errors if the page isn't up-to-date.
2316 * Try to re-read it _once_. We do this synchronously,
2317 * because there really aren't any performance issues here
2318 * and we need to check for errors.
2319 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002321 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002322 if (!error) {
2323 wait_on_page_locked(page);
2324 if (!PageUptodate(page))
2325 error = -EIO;
2326 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002327 put_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002328
2329 if (!error || error == AOP_TRUNCATED_PAGE)
2330 goto retry_find;
2331
2332 /* Things didn't work out. Return zero to tell the mm layer so. */
2333 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002334 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002335}
2336EXPORT_SYMBOL(filemap_fault);
2337
Jan Kara82b0f8c2016-12-14 15:06:58 -08002338void filemap_map_pages(struct vm_fault *vmf,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002339 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002340{
2341 struct radix_tree_iter iter;
2342 void **slot;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002343 struct file *file = vmf->vma->vm_file;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002344 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002345 pgoff_t last_pgoff = start_pgoff;
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002346 unsigned long max_idx;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002347 struct page *head, *page;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002348
2349 rcu_read_lock();
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002350 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2351 start_pgoff) {
2352 if (iter.index > end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002353 break;
2354repeat:
2355 page = radix_tree_deref_slot(slot);
2356 if (unlikely(!page))
2357 goto next;
2358 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002359 if (radix_tree_deref_retry(page)) {
2360 slot = radix_tree_iter_retry(&iter);
2361 continue;
2362 }
2363 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002364 }
2365
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002366 head = compound_head(page);
2367 if (!page_cache_get_speculative(head))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002368 goto repeat;
2369
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002370 /* The page was split under us? */
2371 if (compound_head(page) != head) {
2372 put_page(head);
2373 goto repeat;
2374 }
2375
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002376 /* Has the page moved? */
2377 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002378 put_page(head);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002379 goto repeat;
2380 }
2381
2382 if (!PageUptodate(page) ||
2383 PageReadahead(page) ||
2384 PageHWPoison(page))
2385 goto skip;
2386 if (!trylock_page(page))
2387 goto skip;
2388
2389 if (page->mapping != mapping || !PageUptodate(page))
2390 goto unlock;
2391
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002392 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2393 if (page->index >= max_idx)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002394 goto unlock;
2395
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002396 if (file->f_ra.mmap_miss > 0)
2397 file->f_ra.mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002398
Jan Kara82b0f8c2016-12-14 15:06:58 -08002399 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2400 if (vmf->pte)
2401 vmf->pte += iter.index - last_pgoff;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002402 last_pgoff = iter.index;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002403 if (alloc_set_pte(vmf, NULL, page))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002404 goto unlock;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002405 unlock_page(page);
2406 goto next;
2407unlock:
2408 unlock_page(page);
2409skip:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002410 put_page(page);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002411next:
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002412 /* Huge page is mapped? No need to proceed. */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002413 if (pmd_trans_huge(*vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002414 break;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002415 if (iter.index == end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002416 break;
2417 }
2418 rcu_read_unlock();
2419}
2420EXPORT_SYMBOL(filemap_map_pages);
2421
Dave Jiang11bac802017-02-24 14:56:41 -08002422int filemap_page_mkwrite(struct vm_fault *vmf)
Jan Kara4fcf1c62012-06-12 16:20:29 +02002423{
2424 struct page *page = vmf->page;
Dave Jiang11bac802017-02-24 14:56:41 -08002425 struct inode *inode = file_inode(vmf->vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002426 int ret = VM_FAULT_LOCKED;
2427
Jan Kara14da9202012-06-12 16:20:37 +02002428 sb_start_pagefault(inode->i_sb);
Dave Jiang11bac802017-02-24 14:56:41 -08002429 file_update_time(vmf->vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002430 lock_page(page);
2431 if (page->mapping != inode->i_mapping) {
2432 unlock_page(page);
2433 ret = VM_FAULT_NOPAGE;
2434 goto out;
2435 }
Jan Kara14da9202012-06-12 16:20:37 +02002436 /*
2437 * We mark the page dirty already here so that when freeze is in
2438 * progress, we are guaranteed that writeback during freezing will
2439 * see the dirty page and writeprotect it again.
2440 */
2441 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002442 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002443out:
Jan Kara14da9202012-06-12 16:20:37 +02002444 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002445 return ret;
2446}
2447EXPORT_SYMBOL(filemap_page_mkwrite);
2448
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04002449const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002450 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002451 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002452 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453};
2454
2455/* This is used for a general mmap of a disk file */
2456
2457int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2458{
2459 struct address_space *mapping = file->f_mapping;
2460
2461 if (!mapping->a_ops->readpage)
2462 return -ENOEXEC;
2463 file_accessed(file);
2464 vma->vm_ops = &generic_file_vm_ops;
2465 return 0;
2466}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
2468/*
2469 * This is for filesystems which do not implement ->writepage.
2470 */
2471int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2472{
2473 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2474 return -EINVAL;
2475 return generic_file_mmap(file, vma);
2476}
2477#else
2478int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2479{
2480 return -ENOSYS;
2481}
2482int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2483{
2484 return -ENOSYS;
2485}
2486#endif /* CONFIG_MMU */
2487
2488EXPORT_SYMBOL(generic_file_mmap);
2489EXPORT_SYMBOL(generic_file_readonly_mmap);
2490
Sasha Levin67f9fd92014-04-03 14:48:18 -07002491static struct page *wait_on_page_read(struct page *page)
2492{
2493 if (!IS_ERR(page)) {
2494 wait_on_page_locked(page);
2495 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002496 put_page(page);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002497 page = ERR_PTR(-EIO);
2498 }
2499 }
2500 return page;
2501}
2502
Mel Gorman32b63522016-03-15 14:55:36 -07002503static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002504 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002505 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002506 void *data,
2507 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508{
Nick Piggineb2be182007-10-16 01:24:57 -07002509 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 int err;
2511repeat:
2512 page = find_get_page(mapping, index);
2513 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002514 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002515 if (!page)
2516 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002517 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002518 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002519 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002520 if (err == -EEXIST)
2521 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 return ERR_PTR(err);
2524 }
Mel Gorman32b63522016-03-15 14:55:36 -07002525
2526filler:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 err = filler(data, page);
2528 if (err < 0) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002529 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002530 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 }
Mel Gorman32b63522016-03-15 14:55:36 -07002532
2533 page = wait_on_page_read(page);
2534 if (IS_ERR(page))
2535 return page;
2536 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 if (PageUptodate(page))
2539 goto out;
2540
Mel Gormanebded022016-03-15 14:55:39 -07002541 /*
2542 * Page is not up to date and may be locked due one of the following
2543 * case a: Page is being filled and the page lock is held
2544 * case b: Read/write error clearing the page uptodate status
2545 * case c: Truncation in progress (page locked)
2546 * case d: Reclaim in progress
2547 *
2548 * Case a, the page will be up to date when the page is unlocked.
2549 * There is no need to serialise on the page lock here as the page
2550 * is pinned so the lock gives no additional protection. Even if the
2551 * the page is truncated, the data is still valid if PageUptodate as
2552 * it's a race vs truncate race.
2553 * Case b, the page will not be up to date
2554 * Case c, the page may be truncated but in itself, the data may still
2555 * be valid after IO completes as it's a read vs truncate race. The
2556 * operation must restart if the page is not uptodate on unlock but
2557 * otherwise serialising on page lock to stabilise the mapping gives
2558 * no additional guarantees to the caller as the page lock is
2559 * released before return.
2560 * Case d, similar to truncation. If reclaim holds the page lock, it
2561 * will be a race with remove_mapping that determines if the mapping
2562 * is valid on unlock but otherwise the data is valid and there is
2563 * no need to serialise with page lock.
2564 *
2565 * As the page lock gives no additional guarantee, we optimistically
2566 * wait on the page to be unlocked and check if it's up to date and
2567 * use the page if it is. Otherwise, the page lock is required to
2568 * distinguish between the different cases. The motivation is that we
2569 * avoid spurious serialisations and wakeups when multiple processes
2570 * wait on the same page for IO to complete.
2571 */
2572 wait_on_page_locked(page);
2573 if (PageUptodate(page))
2574 goto out;
2575
2576 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07002578
2579 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 if (!page->mapping) {
2581 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002582 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002583 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 }
Mel Gormanebded022016-03-15 14:55:39 -07002585
2586 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 if (PageUptodate(page)) {
2588 unlock_page(page);
2589 goto out;
2590 }
Mel Gorman32b63522016-03-15 14:55:36 -07002591 goto filler;
2592
David Howellsc855ff32007-05-09 13:42:20 +01002593out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002594 mark_page_accessed(page);
2595 return page;
2596}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002597
2598/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002599 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002600 * @mapping: the page's address_space
2601 * @index: the page index
2602 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002603 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002604 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002605 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002606 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002607 *
2608 * If the page does not get brought uptodate, return -EIO.
2609 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002610struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002611 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002612 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002613 void *data)
2614{
2615 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2616}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002617EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002618
2619/**
2620 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2621 * @mapping: the page's address_space
2622 * @index: the page index
2623 * @gfp: the page allocator flags to use if allocating
2624 *
2625 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002626 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002627 *
2628 * If the page does not get brought uptodate, return -EIO.
2629 */
2630struct page *read_cache_page_gfp(struct address_space *mapping,
2631 pgoff_t index,
2632 gfp_t gfp)
2633{
2634 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2635
Sasha Levin67f9fd92014-04-03 14:48:18 -07002636 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002637}
2638EXPORT_SYMBOL(read_cache_page_gfp);
2639
Nick Piggin2f718ff2007-10-16 01:24:59 -07002640/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 * Performs necessary checks before doing a write
2642 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002643 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 * Returns appropriate error code that caller should return or
2645 * zero in case that write should be allowed.
2646 */
Al Viro3309dd02015-04-09 12:55:47 -04002647inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648{
Al Viro3309dd02015-04-09 12:55:47 -04002649 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002651 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002652 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653
Al Viro3309dd02015-04-09 12:55:47 -04002654 if (!iov_iter_count(from))
2655 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656
Al Viro0fa6b002015-04-04 04:05:48 -04002657 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002658 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002659 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
Al Viro3309dd02015-04-09 12:55:47 -04002661 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662
Al Viro0fa6b002015-04-04 04:05:48 -04002663 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002664 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002665 send_sig(SIGXFSZ, current, 0);
2666 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 }
Al Viro3309dd02015-04-09 12:55:47 -04002668 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 }
2670
2671 /*
2672 * LFS rule
2673 */
Al Viro3309dd02015-04-09 12:55:47 -04002674 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002676 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002678 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 }
2680
2681 /*
2682 * Are we about to exceed the fs block limit ?
2683 *
2684 * If we have written data it becomes a short write. If we have
2685 * exceeded without writing data we send a signal and return EFBIG.
2686 * Linus frestrict idea will clean these up nicely..
2687 */
Al Viro3309dd02015-04-09 12:55:47 -04002688 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2689 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
Al Viro3309dd02015-04-09 12:55:47 -04002691 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2692 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693}
2694EXPORT_SYMBOL(generic_write_checks);
2695
Nick Pigginafddba42007-10-16 01:25:01 -07002696int pagecache_write_begin(struct file *file, struct address_space *mapping,
2697 loff_t pos, unsigned len, unsigned flags,
2698 struct page **pagep, void **fsdata)
2699{
2700 const struct address_space_operations *aops = mapping->a_ops;
2701
Nick Piggin4e02ed42008-10-29 14:00:55 -07002702 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002703 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002704}
2705EXPORT_SYMBOL(pagecache_write_begin);
2706
2707int pagecache_write_end(struct file *file, struct address_space *mapping,
2708 loff_t pos, unsigned len, unsigned copied,
2709 struct page *page, void *fsdata)
2710{
2711 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002712
Nick Piggin4e02ed42008-10-29 14:00:55 -07002713 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002714}
2715EXPORT_SYMBOL(pagecache_write_end);
2716
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002718generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719{
2720 struct file *file = iocb->ki_filp;
2721 struct address_space *mapping = file->f_mapping;
2722 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002723 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002725 size_t write_len;
2726 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
Al Viro0c949332014-03-22 06:51:37 -04002728 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002729 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002730
Nick Piggin48b47c52009-01-06 14:40:22 -08002731 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002732 if (written)
2733 goto out;
2734
2735 /*
2736 * After a write we want buffered reads to be sure to go to disk to get
2737 * the new data. We invalidate clean cached page from the region we're
2738 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002739 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002740 */
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07002741 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002742 pos >> PAGE_SHIFT, end);
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07002743 /*
2744 * If a page can not be invalidated, return 0 to fall back
2745 * to buffered write.
2746 */
2747 if (written) {
2748 if (written == -EBUSY)
2749 return 0;
2750 goto out;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002751 }
2752
Al Viro639a93a52017-04-13 14:10:15 -04002753 written = mapping->a_ops->direct_IO(iocb, from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002754
2755 /*
2756 * Finally, try again to invalidate clean pages which might have been
2757 * cached by non-direct readahead, or faulted in by get_user_pages()
2758 * if the source of the write was an mmap'ed region of the file
2759 * we're writing. Either one is a pretty crazy thing to do,
2760 * so we don't support it 100%. If this invalidation
2761 * fails, tough, the write still worked...
2762 */
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07002763 invalidate_inode_pages2_range(mapping,
2764 pos >> PAGE_SHIFT, end);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002767 pos += written;
Al Viro639a93a52017-04-13 14:10:15 -04002768 write_len -= written;
Namhyung Kim01166512010-10-26 14:21:58 -07002769 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2770 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 mark_inode_dirty(inode);
2772 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002773 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 }
Al Viro639a93a52017-04-13 14:10:15 -04002775 iov_iter_revert(from, write_len - iov_iter_count(from));
Christoph Hellwiga969e902008-07-23 21:27:04 -07002776out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 return written;
2778}
2779EXPORT_SYMBOL(generic_file_direct_write);
2780
Nick Piggineb2be182007-10-16 01:24:57 -07002781/*
2782 * Find or create a page at the given pagecache position. Return the locked
2783 * page. This function is specifically for buffered writes.
2784 */
Nick Piggin54566b22009-01-04 12:00:53 -08002785struct page *grab_cache_page_write_begin(struct address_space *mapping,
2786 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002787{
Nick Piggineb2be182007-10-16 01:24:57 -07002788 struct page *page;
Johannes Weinerbbddabe2016-05-20 16:56:28 -07002789 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002790
Nick Piggin54566b22009-01-04 12:00:53 -08002791 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002792 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002793
Mel Gorman2457aec2014-06-04 16:10:31 -07002794 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002795 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002796 if (page)
2797 wait_for_stable_page(page);
2798
Nick Piggineb2be182007-10-16 01:24:57 -07002799 return page;
2800}
Nick Piggin54566b22009-01-04 12:00:53 -08002801EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002802
Al Viro3b93f912014-02-11 21:34:08 -05002803ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002804 struct iov_iter *i, loff_t pos)
2805{
2806 struct address_space *mapping = file->f_mapping;
2807 const struct address_space_operations *a_ops = mapping->a_ops;
2808 long status = 0;
2809 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002810 unsigned int flags = 0;
2811
Nick Pigginafddba42007-10-16 01:25:01 -07002812 do {
2813 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002814 unsigned long offset; /* Offset into pagecache page */
2815 unsigned long bytes; /* Bytes to write to page */
2816 size_t copied; /* Bytes copied from user */
2817 void *fsdata;
2818
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002819 offset = (pos & (PAGE_SIZE - 1));
2820 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002821 iov_iter_count(i));
2822
2823again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002824 /*
2825 * Bring in the user page that we will copy from _first_.
2826 * Otherwise there's a nasty deadlock on copying from the
2827 * same page as we're writing to, without it being marked
2828 * up-to-date.
2829 *
2830 * Not only is this an optimisation, but it is also required
2831 * to check that the address is actually valid, when atomic
2832 * usercopies are used, below.
2833 */
2834 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2835 status = -EFAULT;
2836 break;
2837 }
2838
Jan Kara296291c2015-10-22 13:32:21 -07002839 if (fatal_signal_pending(current)) {
2840 status = -EINTR;
2841 break;
2842 }
2843
Nick Piggin674b8922007-10-16 01:25:03 -07002844 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002845 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002846 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002847 break;
2848
anfei zhou931e80e2010-02-02 13:44:02 -08002849 if (mapping_writably_mapped(mapping))
2850 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002851
Nick Pigginafddba42007-10-16 01:25:01 -07002852 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002853 flush_dcache_page(page);
2854
2855 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2856 page, fsdata);
2857 if (unlikely(status < 0))
2858 break;
2859 copied = status;
2860
2861 cond_resched();
2862
Nick Piggin124d3b72008-02-02 15:01:17 +01002863 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002864 if (unlikely(copied == 0)) {
2865 /*
2866 * If we were unable to copy any data at all, we must
2867 * fall back to a single segment length write.
2868 *
2869 * If we didn't fallback here, we could livelock
2870 * because not all segments in the iov can be copied at
2871 * once without a pagefault.
2872 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002873 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002874 iov_iter_single_seg_count(i));
2875 goto again;
2876 }
Nick Pigginafddba42007-10-16 01:25:01 -07002877 pos += copied;
2878 written += copied;
2879
2880 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002881 } while (iov_iter_count(i));
2882
2883 return written ? written : status;
2884}
Al Viro3b93f912014-02-11 21:34:08 -05002885EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886
Jan Karae4dd9de2009-08-17 18:10:06 +02002887/**
Al Viro81742022014-04-03 03:17:43 -04002888 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002889 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002890 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002891 *
2892 * This function does all the work needed for actually writing data to a
2893 * file. It does all basic checks, removes SUID from the file, updates
2894 * modification times and calls proper subroutines depending on whether we
2895 * do direct IO or a standard buffered write.
2896 *
2897 * It expects i_mutex to be grabbed unless we work on a block device or similar
2898 * object which does not need locking at all.
2899 *
2900 * This function does *not* take care of syncing data in case of O_SYNC write.
2901 * A caller has to handle it. This is mainly due to the fact that we want to
2902 * avoid syncing under i_mutex.
2903 */
Al Viro81742022014-04-03 03:17:43 -04002904ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905{
2906 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002907 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002909 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002911 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002914 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002915 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 if (err)
2917 goto out;
2918
Josef Bacikc3b2da32012-03-26 09:59:21 -04002919 err = file_update_time(file);
2920 if (err)
2921 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
Al Viro2ba48ce2015-04-09 13:52:01 -04002923 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002924 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002925
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002926 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002927 /*
2928 * If the write stopped short of completing, fall back to
2929 * buffered writes. Some filesystems do this for writes to
2930 * holes, for example. For DAX files, a buffered write will
2931 * not succeed (even if it did, DAX does not handle dirty
2932 * page-cache pages correctly).
2933 */
Al Viro0b8def92015-04-07 10:22:53 -04002934 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002936
Al Viro0b8def92015-04-07 10:22:53 -04002937 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002938 /*
Al Viro3b93f912014-02-11 21:34:08 -05002939 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002940 * then we want to return the number of bytes which were
2941 * direct-written, or the error code if that was zero. Note
2942 * that this differs from normal direct-io semantics, which
2943 * will return -EFOO even if some bytes were written.
2944 */
Al Viro60bb4522014-08-08 12:39:16 -04002945 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002946 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002947 goto out;
2948 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002949 /*
2950 * We need to ensure that the page cache pages are written to
2951 * disk and invalidated to preserve the expected O_DIRECT
2952 * semantics.
2953 */
Al Viro3b93f912014-02-11 21:34:08 -05002954 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002955 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002956 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002957 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002958 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002959 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002960 pos >> PAGE_SHIFT,
2961 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002962 } else {
2963 /*
2964 * We don't know how much we wrote, so just return
2965 * the number of bytes which were direct-written
2966 */
2967 }
2968 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002969 written = generic_perform_write(file, from, iocb->ki_pos);
2970 if (likely(written > 0))
2971 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973out:
2974 current->backing_dev_info = NULL;
2975 return written ? written : err;
2976}
Al Viro81742022014-04-03 03:17:43 -04002977EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978
Jan Karae4dd9de2009-08-17 18:10:06 +02002979/**
Al Viro81742022014-04-03 03:17:43 -04002980 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002981 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002982 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002983 *
Al Viro81742022014-04-03 03:17:43 -04002984 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002985 * filesystems. It takes care of syncing the file in case of O_SYNC file
2986 * and acquires i_mutex as needed.
2987 */
Al Viro81742022014-04-03 03:17:43 -04002988ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989{
2990 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002991 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993
Al Viro59551022016-01-22 15:40:57 -05002994 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04002995 ret = generic_write_checks(iocb, from);
2996 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002997 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05002998 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999
Christoph Hellwige2592212016-04-07 08:52:01 -07003000 if (ret > 0)
3001 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 return ret;
3003}
Al Viro81742022014-04-03 03:17:43 -04003004EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005
David Howellscf9a2ae2006-08-29 19:05:54 +01003006/**
3007 * try_to_release_page() - release old fs-specific metadata on a page
3008 *
3009 * @page: the page which the kernel is trying to free
3010 * @gfp_mask: memory allocation flags (and I/O mode)
3011 *
3012 * The address_space is to try to release any data against the page
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -03003013 * (presumably at page->private). If the release was successful, return '1'.
David Howellscf9a2ae2006-08-29 19:05:54 +01003014 * Otherwise return zero.
3015 *
David Howells266cf652009-04-03 16:42:36 +01003016 * This may also be called if PG_fscache is set on a page, indicating that the
3017 * page is known to the local caching routines.
3018 *
David Howellscf9a2ae2006-08-29 19:05:54 +01003019 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08003020 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01003021 *
David Howellscf9a2ae2006-08-29 19:05:54 +01003022 */
3023int try_to_release_page(struct page *page, gfp_t gfp_mask)
3024{
3025 struct address_space * const mapping = page->mapping;
3026
3027 BUG_ON(!PageLocked(page));
3028 if (PageWriteback(page))
3029 return 0;
3030
3031 if (mapping && mapping->a_ops->releasepage)
3032 return mapping->a_ops->releasepage(page, gfp_mask);
3033 return try_to_free_buffers(page);
3034}
3035
3036EXPORT_SYMBOL(try_to_release_page);