blob: a82fbe4c9e8e1c1d5a3eed5e2649ec87a7bfd16d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
14#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070015#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080034#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060035#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070036#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070039#define CREATE_TRACE_POINTS
40#include <trace/events/filemap.h>
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * FIXME: remove all knowledge of the buffer layer from the core VM
44 */
Jan Kara148f9482009-08-17 19:52:36 +020045#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <asm/mman.h>
48
49/*
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * though.
52 *
53 * Shared mappings now work. 15.8.1995 Bruno.
54 *
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 *
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
59 */
60
61/*
62 * Lock ordering:
63 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070064 * ->i_mmap_mutex (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070066 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080069 * ->i_mutex
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070070 * ->i_mmap_mutex (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 *
72 * ->mmap_sem
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070073 * ->i_mmap_mutex
Hugh Dickinsb8072f02005-10-29 18:16:41 -070074 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 *
77 * ->mmap_sem
78 * ->lock_page (access_process_vm)
79 *
Al Viroccad2362014-02-11 22:36:48 -050080 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070081 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060083 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110084 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * ->mapping->tree_lock (__sync_single_inode)
86 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070087 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * ->anon_vma.lock (vma_adjust)
89 *
90 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070091 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070094 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080098 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600103 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100104 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
106 *
Andi Kleen9a3c5312012-03-21 16:34:09 -0700107 * ->i_mmap_mutex
108 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 */
110
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700111static void page_cache_tree_delete(struct address_space *mapping,
112 struct page *page, void *shadow)
113{
Johannes Weiner449dd692014-04-03 14:47:56 -0700114 struct radix_tree_node *node;
115 unsigned long index;
116 unsigned int offset;
117 unsigned int tag;
118 void **slot;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700119
Johannes Weiner449dd692014-04-03 14:47:56 -0700120 VM_BUG_ON(!PageLocked(page));
121
122 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
123
124 if (shadow) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700125 mapping->nrshadows++;
126 /*
127 * Make sure the nrshadows update is committed before
128 * the nrpages update so that final truncate racing
129 * with reclaim does not see both counters 0 at the
130 * same time and miss a shadow entry.
131 */
132 smp_wmb();
Johannes Weiner449dd692014-04-03 14:47:56 -0700133 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700134 mapping->nrpages--;
Johannes Weiner449dd692014-04-03 14:47:56 -0700135
136 if (!node) {
137 /* Clear direct pointer tags in root node */
138 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
139 radix_tree_replace_slot(slot, shadow);
140 return;
141 }
142
143 /* Clear tree tags for the removed page */
144 index = page->index;
145 offset = index & RADIX_TREE_MAP_MASK;
146 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
147 if (test_bit(offset, node->tags[tag]))
148 radix_tree_tag_clear(&mapping->page_tree, index, tag);
149 }
150
151 /* Delete page, swap shadow entry */
152 radix_tree_replace_slot(slot, shadow);
153 workingset_node_pages_dec(node);
154 if (shadow)
155 workingset_node_shadows_inc(node);
156 else
157 if (__radix_tree_delete_node(&mapping->page_tree, node))
158 return;
159
160 /*
161 * Track node that only contains shadow entries.
162 *
163 * Avoid acquiring the list_lru lock if already tracked. The
164 * list_empty() test is safe as node->private_list is
165 * protected by mapping->tree_lock.
166 */
167 if (!workingset_node_pages(node) &&
168 list_empty(&node->private_list)) {
169 node->private_data = mapping;
170 list_lru_add(&workingset_shadow_nodes, &node->private_list);
171 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700172}
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174/*
Minchan Kime64a7822011-03-22 16:32:44 -0700175 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700177 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700179void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
181 struct address_space *mapping = page->mapping;
182
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700183 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600184 /*
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
188 */
189 if (PageUptodate(page) && PageMappedToDisk(page))
190 cleancache_put_page(page);
191 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400192 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600193
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700194 page_cache_tree_delete(mapping, page, shadow);
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700197 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700198
Christoph Lameter347ce432006-06-30 01:55:35 -0700199 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700200 if (PageSwapBacked(page))
201 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700202 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800203
204 /*
205 * Some filesystems seem to re-dirty the page even after
206 * the VM has canceled the dirty bit (eg ext3 journaling).
207 *
208 * Fix it up by doing a final dirty accounting check after
209 * having removed the page entirely.
210 */
211 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
212 dec_zone_page_state(page, NR_FILE_DIRTY);
213 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
Minchan Kim702cfbf2011-03-22 16:32:43 -0700217/**
218 * delete_from_page_cache - delete page from page cache
219 * @page: the page which the kernel is trying to remove from page cache
220 *
221 * This must be called only on pages that have been verified to be in the page
222 * cache and locked. It will never put the page into the free list, the caller
223 * has a reference on the page.
224 */
225void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226{
227 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500228 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Matt Mackallcd7619d2005-05-01 08:59:01 -0700230 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Linus Torvalds6072d132010-12-01 13:35:19 -0500232 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700233 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700234 __delete_from_page_cache(page, NULL);
Nick Piggin19fd6232008-07-25 19:45:32 -0700235 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700236 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500237
238 if (freepage)
239 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700240 page_cache_release(page);
241}
242EXPORT_SYMBOL(delete_from_page_cache);
243
Jens Axboe7eaceac2011-03-10 08:52:07 +0100244static int sleep_on_page(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 io_schedule();
247 return 0;
248}
249
Jens Axboe7eaceac2011-03-10 08:52:07 +0100250static int sleep_on_page_killable(void *word)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500251{
Jens Axboe7eaceac2011-03-10 08:52:07 +0100252 sleep_on_page(word);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500253 return fatal_signal_pending(current) ? -EINTR : 0;
254}
255
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700256static int filemap_check_errors(struct address_space *mapping)
257{
258 int ret = 0;
259 /* Check for outstanding write errors */
260 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
261 ret = -ENOSPC;
262 if (test_and_clear_bit(AS_EIO, &mapping->flags))
263 ret = -EIO;
264 return ret;
265}
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700268 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700269 * @mapping: address space structure to write
270 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800271 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700272 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700274 * Start writeback against all of a mapping's dirty pages that lie
275 * within the byte offsets <start, end> inclusive.
276 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700278 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 * these two operations is that if a dirty page/buffer is encountered, it must
280 * be waited upon, and not just skipped over.
281 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800282int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
283 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
285 int ret;
286 struct writeback_control wbc = {
287 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800288 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700289 .range_start = start,
290 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 };
292
293 if (!mapping_cap_writeback_dirty(mapping))
294 return 0;
295
296 ret = do_writepages(mapping, &wbc);
297 return ret;
298}
299
300static inline int __filemap_fdatawrite(struct address_space *mapping,
301 int sync_mode)
302{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700303 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
306int filemap_fdatawrite(struct address_space *mapping)
307{
308 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
309}
310EXPORT_SYMBOL(filemap_fdatawrite);
311
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400312int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800313 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
315 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
316}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400317EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Randy Dunlap485bb992006-06-23 02:03:49 -0700319/**
320 * filemap_flush - mostly a non-blocking flush
321 * @mapping: target address_space
322 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 * This is a mostly non-blocking flush. Not suitable for data-integrity
324 * purposes - I/O may not be started against all dirty pages.
325 */
326int filemap_flush(struct address_space *mapping)
327{
328 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
329}
330EXPORT_SYMBOL(filemap_flush);
331
Randy Dunlap485bb992006-06-23 02:03:49 -0700332/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200333 * filemap_fdatawait_range - wait for writeback to complete
334 * @mapping: address space structure to wait for
335 * @start_byte: offset in bytes where the range starts
336 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700337 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200338 * Walk the list of under-writeback pages of the given address space
339 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200341int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
342 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200344 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
345 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 struct pagevec pvec;
347 int nr_pages;
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700348 int ret2, ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200350 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700351 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 while ((index <= end) &&
355 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
356 PAGECACHE_TAG_WRITEBACK,
357 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
358 unsigned i;
359
360 for (i = 0; i < nr_pages; i++) {
361 struct page *page = pvec.pages[i];
362
363 /* until radix tree lookup accepts end_index */
364 if (page->index > end)
365 continue;
366
367 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800368 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 ret = -EIO;
370 }
371 pagevec_release(&pvec);
372 cond_resched();
373 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700374out:
375 ret2 = filemap_check_errors(mapping);
376 if (!ret)
377 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379 return ret;
380}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200381EXPORT_SYMBOL(filemap_fdatawait_range);
382
383/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700384 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700386 *
387 * Walk the list of under-writeback pages of the given address space
388 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
390int filemap_fdatawait(struct address_space *mapping)
391{
392 loff_t i_size = i_size_read(mapping->host);
393
394 if (i_size == 0)
395 return 0;
396
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200397 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399EXPORT_SYMBOL(filemap_fdatawait);
400
401int filemap_write_and_wait(struct address_space *mapping)
402{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800403 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800406 err = filemap_fdatawrite(mapping);
407 /*
408 * Even if the above returned error, the pages may be
409 * written partially (e.g. -ENOSPC), so we wait for it.
410 * But the -EIO is special case, it may indicate the worst
411 * thing (e.g. bug) happened, so we avoid waiting for it.
412 */
413 if (err != -EIO) {
414 int err2 = filemap_fdatawait(mapping);
415 if (!err)
416 err = err2;
417 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700418 } else {
419 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800421 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800423EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Randy Dunlap485bb992006-06-23 02:03:49 -0700425/**
426 * filemap_write_and_wait_range - write out & wait on a file range
427 * @mapping: the address_space for the pages
428 * @lstart: offset in bytes where the range starts
429 * @lend: offset in bytes where the range ends (inclusive)
430 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800431 * Write out and wait upon file offsets lstart->lend, inclusive.
432 *
433 * Note that `lend' is inclusive (describes the last byte to be written) so
434 * that this function can be used to write to the very end-of-file (end = -1).
435 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436int filemap_write_and_wait_range(struct address_space *mapping,
437 loff_t lstart, loff_t lend)
438{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800439 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
441 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800442 err = __filemap_fdatawrite_range(mapping, lstart, lend,
443 WB_SYNC_ALL);
444 /* See comment of filemap_write_and_wait() */
445 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200446 int err2 = filemap_fdatawait_range(mapping,
447 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800448 if (!err)
449 err = err2;
450 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700451 } else {
452 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800454 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
Chris Masonf6995582009-04-15 13:22:37 -0400456EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Randy Dunlap485bb992006-06-23 02:03:49 -0700458/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700459 * replace_page_cache_page - replace a pagecache page with a new one
460 * @old: page to be replaced
461 * @new: page to replace with
462 * @gfp_mask: allocation mode
463 *
464 * This function replaces a page in the pagecache with a new one. On
465 * success it acquires the pagecache reference for the new page and
466 * drops it for the old page. Both the old and new pages must be
467 * locked. This function does not add the new page to the LRU, the
468 * caller must do that.
469 *
470 * The remove + add is atomic. The only way this function can fail is
471 * memory allocation failure.
472 */
473int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
474{
475 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700476
Sasha Levin309381fea2014-01-23 15:52:54 -0800477 VM_BUG_ON_PAGE(!PageLocked(old), old);
478 VM_BUG_ON_PAGE(!PageLocked(new), new);
479 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700480
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700481 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
482 if (!error) {
483 struct address_space *mapping = old->mapping;
484 void (*freepage)(struct page *);
485
486 pgoff_t offset = old->index;
487 freepage = mapping->a_ops->freepage;
488
489 page_cache_get(new);
490 new->mapping = mapping;
491 new->index = offset;
492
493 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700494 __delete_from_page_cache(old, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700495 error = radix_tree_insert(&mapping->page_tree, offset, new);
496 BUG_ON(error);
497 mapping->nrpages++;
498 __inc_zone_page_state(new, NR_FILE_PAGES);
499 if (PageSwapBacked(new))
500 __inc_zone_page_state(new, NR_SHMEM);
501 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -0800502 /* mem_cgroup codes must not be called under tree_lock */
503 mem_cgroup_replace_page_cache(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700504 radix_tree_preload_end();
505 if (freepage)
506 freepage(old);
507 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700508 }
509
510 return error;
511}
512EXPORT_SYMBOL_GPL(replace_page_cache_page);
513
Johannes Weiner0cd61442014-04-03 14:47:46 -0700514static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700515 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700516{
Johannes Weiner449dd692014-04-03 14:47:56 -0700517 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700518 void **slot;
519 int error;
520
Johannes Weiner449dd692014-04-03 14:47:56 -0700521 error = __radix_tree_create(&mapping->page_tree, page->index,
522 &node, &slot);
523 if (error)
524 return error;
525 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700526 void *p;
527
528 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
529 if (!radix_tree_exceptional_entry(p))
530 return -EEXIST;
Johannes Weinera5289102014-04-03 14:47:51 -0700531 if (shadowp)
532 *shadowp = p;
Johannes Weiner449dd692014-04-03 14:47:56 -0700533 mapping->nrshadows--;
534 if (node)
535 workingset_node_shadows_dec(node);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700536 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700537 radix_tree_replace_slot(slot, page);
538 mapping->nrpages++;
539 if (node) {
540 workingset_node_pages_inc(node);
541 /*
542 * Don't track node that contains actual pages.
543 *
544 * Avoid acquiring the list_lru lock if already
545 * untracked. The list_empty() test is safe as
546 * node->private_list is protected by
547 * mapping->tree_lock.
548 */
549 if (!list_empty(&node->private_list))
550 list_lru_del(&workingset_shadow_nodes,
551 &node->private_list);
552 }
553 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700554}
555
Johannes Weinera5289102014-04-03 14:47:51 -0700556static int __add_to_page_cache_locked(struct page *page,
557 struct address_space *mapping,
558 pgoff_t offset, gfp_t gfp_mask,
559 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Nick Piggine2867812008-07-25 19:45:30 -0700561 int error;
562
Sasha Levin309381fea2014-01-23 15:52:54 -0800563 VM_BUG_ON_PAGE(!PageLocked(page), page);
564 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700565
Michal Hockod715ae02014-04-07 15:37:46 -0700566 error = mem_cgroup_charge_file(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800567 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800568 if (error)
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700569 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Jan Kara5e4c0d972013-09-11 14:26:05 -0700571 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700572 if (error) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700573 mem_cgroup_uncharge_cache_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700574 return error;
575 }
576
577 page_cache_get(page);
578 page->mapping = mapping;
579 page->index = offset;
580
581 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700582 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700583 radix_tree_preload_end();
584 if (unlikely(error))
585 goto err_insert;
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700586 __inc_zone_page_state(page, NR_FILE_PAGES);
587 spin_unlock_irq(&mapping->tree_lock);
588 trace_mm_filemap_add_to_page_cache(page);
589 return 0;
590err_insert:
591 page->mapping = NULL;
592 /* Leave page->index set: truncation relies upon it */
593 spin_unlock_irq(&mapping->tree_lock);
594 mem_cgroup_uncharge_cache_page(page);
595 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 return error;
597}
Johannes Weinera5289102014-04-03 14:47:51 -0700598
599/**
600 * add_to_page_cache_locked - add a locked page to the pagecache
601 * @page: page to add
602 * @mapping: the page's address_space
603 * @offset: page index
604 * @gfp_mask: page allocation mode
605 *
606 * This function is used to add a page to the pagecache. It must be locked.
607 * This function does not add the page to the LRU. The caller must do that.
608 */
609int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
610 pgoff_t offset, gfp_t gfp_mask)
611{
612 return __add_to_page_cache_locked(page, mapping, offset,
613 gfp_mask, NULL);
614}
Nick Piggine2867812008-07-25 19:45:30 -0700615EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400618 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Johannes Weinera5289102014-04-03 14:47:51 -0700620 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700621 int ret;
622
Johannes Weinera5289102014-04-03 14:47:51 -0700623 __set_page_locked(page);
624 ret = __add_to_page_cache_locked(page, mapping, offset,
625 gfp_mask, &shadow);
626 if (unlikely(ret))
627 __clear_page_locked(page);
628 else {
629 /*
630 * The page might have been evicted from cache only
631 * recently, in which case it should be activated like
632 * any other repeatedly accessed page.
633 */
634 if (shadow && workingset_refault(shadow)) {
635 SetPageActive(page);
636 workingset_activation(page);
637 } else
638 ClearPageActive(page);
639 lru_cache_add(page);
640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 return ret;
642}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300643EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Paul Jackson44110fe2006-03-24 03:16:04 -0800645#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700646struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800647{
Miao Xiec0ff7452010-05-24 14:32:08 -0700648 int n;
649 struct page *page;
650
Paul Jackson44110fe2006-03-24 03:16:04 -0800651 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700652 unsigned int cpuset_mems_cookie;
653 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700654 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700655 n = cpuset_mem_spread_node();
656 page = alloc_pages_exact_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700657 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700658
Miao Xiec0ff7452010-05-24 14:32:08 -0700659 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800660 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700661 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800662}
Nick Piggin2ae88142006-10-28 10:38:23 -0700663EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800664#endif
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666/*
667 * In order to wait for pages to become available there must be
668 * waitqueues associated with pages. By using a hash table of
669 * waitqueues where the bucket discipline is to maintain all
670 * waiters on the same queue and wake all when any of the pages
671 * become available, and for the woken contexts to check to be
672 * sure the appropriate page became available, this saves space
673 * at a cost of "thundering herd" phenomena during rare hash
674 * collisions.
675 */
676static wait_queue_head_t *page_waitqueue(struct page *page)
677{
678 const struct zone *zone = page_zone(page);
679
680 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
681}
682
683static inline void wake_up_page(struct page *page, int bit)
684{
685 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
686}
687
Harvey Harrison920c7a52008-02-04 22:29:26 -0800688void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
690 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
691
692 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100693 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 TASK_UNINTERRUPTIBLE);
695}
696EXPORT_SYMBOL(wait_on_page_bit);
697
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700698int wait_on_page_bit_killable(struct page *page, int bit_nr)
699{
700 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
701
702 if (!test_bit(bit_nr, &page->flags))
703 return 0;
704
705 return __wait_on_bit(page_waitqueue(page), &wait,
706 sleep_on_page_killable, TASK_KILLABLE);
707}
708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100710 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700711 * @page: Page defining the wait queue of interest
712 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100713 *
714 * Add an arbitrary @waiter to the wait queue for the nominated @page.
715 */
716void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
717{
718 wait_queue_head_t *q = page_waitqueue(page);
719 unsigned long flags;
720
721 spin_lock_irqsave(&q->lock, flags);
722 __add_wait_queue(q, waiter);
723 spin_unlock_irqrestore(&q->lock, flags);
724}
725EXPORT_SYMBOL_GPL(add_page_wait_queue);
726
727/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700728 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 * @page: the page
730 *
731 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
732 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
733 * mechananism between PageLocked pages and PageWriteback pages is shared.
734 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
735 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700736 * The mb is necessary to enforce ordering between the clear_bit and the read
737 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800739void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
Sasha Levin309381fea2014-01-23 15:52:54 -0800741 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700742 clear_bit_unlock(PG_locked, &page->flags);
743 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 wake_up_page(page, PG_locked);
745}
746EXPORT_SYMBOL(unlock_page);
747
Randy Dunlap485bb992006-06-23 02:03:49 -0700748/**
749 * end_page_writeback - end writeback against a page
750 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 */
752void end_page_writeback(struct page *page)
753{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700754 if (TestClearPageReclaim(page))
755 rotate_reclaimable_page(page);
756
757 if (!test_clear_page_writeback(page))
758 BUG();
759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 smp_mb__after_clear_bit();
761 wake_up_page(page, PG_writeback);
762}
763EXPORT_SYMBOL(end_page_writeback);
764
Randy Dunlap485bb992006-06-23 02:03:49 -0700765/**
766 * __lock_page - get a lock on the page, assuming we need to sleep to get it
767 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800769void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
771 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
772
Jens Axboe7eaceac2011-03-10 08:52:07 +0100773 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 TASK_UNINTERRUPTIBLE);
775}
776EXPORT_SYMBOL(__lock_page);
777
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800778int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500779{
780 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
781
782 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100783 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500784}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300785EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500786
Michel Lespinassed065bd82010-10-26 14:21:57 -0700787int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
788 unsigned int flags)
789{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700790 if (flags & FAULT_FLAG_ALLOW_RETRY) {
791 /*
792 * CAUTION! In this case, mmap_sem is not released
793 * even though return 0.
794 */
795 if (flags & FAULT_FLAG_RETRY_NOWAIT)
796 return 0;
797
798 up_read(&mm->mmap_sem);
799 if (flags & FAULT_FLAG_KILLABLE)
800 wait_on_page_locked_killable(page);
801 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700802 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700803 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700804 } else {
805 if (flags & FAULT_FLAG_KILLABLE) {
806 int ret;
807
808 ret = __lock_page_killable(page);
809 if (ret) {
810 up_read(&mm->mmap_sem);
811 return 0;
812 }
813 } else
814 __lock_page(page);
815 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700816 }
817}
818
Randy Dunlap485bb992006-06-23 02:03:49 -0700819/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700820 * page_cache_next_hole - find the next hole (not-present entry)
821 * @mapping: mapping
822 * @index: index
823 * @max_scan: maximum range to search
824 *
825 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
826 * lowest indexed hole.
827 *
828 * Returns: the index of the hole if found, otherwise returns an index
829 * outside of the set specified (in which case 'return - index >=
830 * max_scan' will be true). In rare cases of index wrap-around, 0 will
831 * be returned.
832 *
833 * page_cache_next_hole may be called under rcu_read_lock. However,
834 * like radix_tree_gang_lookup, this will not atomically search a
835 * snapshot of the tree at a single point in time. For example, if a
836 * hole is created at index 5, then subsequently a hole is created at
837 * index 10, page_cache_next_hole covering both indexes may return 10
838 * if called under rcu_read_lock.
839 */
840pgoff_t page_cache_next_hole(struct address_space *mapping,
841 pgoff_t index, unsigned long max_scan)
842{
843 unsigned long i;
844
845 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700846 struct page *page;
847
848 page = radix_tree_lookup(&mapping->page_tree, index);
849 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700850 break;
851 index++;
852 if (index == 0)
853 break;
854 }
855
856 return index;
857}
858EXPORT_SYMBOL(page_cache_next_hole);
859
860/**
861 * page_cache_prev_hole - find the prev hole (not-present entry)
862 * @mapping: mapping
863 * @index: index
864 * @max_scan: maximum range to search
865 *
866 * Search backwards in the range [max(index-max_scan+1, 0), index] for
867 * the first hole.
868 *
869 * Returns: the index of the hole if found, otherwise returns an index
870 * outside of the set specified (in which case 'index - return >=
871 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
872 * will be returned.
873 *
874 * page_cache_prev_hole may be called under rcu_read_lock. However,
875 * like radix_tree_gang_lookup, this will not atomically search a
876 * snapshot of the tree at a single point in time. For example, if a
877 * hole is created at index 10, then subsequently a hole is created at
878 * index 5, page_cache_prev_hole covering both indexes may return 5 if
879 * called under rcu_read_lock.
880 */
881pgoff_t page_cache_prev_hole(struct address_space *mapping,
882 pgoff_t index, unsigned long max_scan)
883{
884 unsigned long i;
885
886 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700887 struct page *page;
888
889 page = radix_tree_lookup(&mapping->page_tree, index);
890 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700891 break;
892 index--;
893 if (index == ULONG_MAX)
894 break;
895 }
896
897 return index;
898}
899EXPORT_SYMBOL(page_cache_prev_hole);
900
901/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700902 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -0700903 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -0700904 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -0700905 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700906 * Looks up the page cache slot at @mapping & @offset. If there is a
907 * page cache page, it is returned with an increased refcount.
908 *
909 * If the slot holds a shadow entry of a previously evicted page, it
910 * is returned.
911 *
912 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700914struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915{
Nick Piggina60637c2008-07-25 19:45:31 -0700916 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 struct page *page;
918
Nick Piggina60637c2008-07-25 19:45:31 -0700919 rcu_read_lock();
920repeat:
921 page = NULL;
922 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
923 if (pagep) {
924 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800925 if (unlikely(!page))
926 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700927 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700928 if (radix_tree_deref_retry(page))
929 goto repeat;
930 /*
931 * Otherwise, shmem/tmpfs must be storing a swap entry
932 * here as an exceptional entry: so return it without
933 * attempting to raise page count.
934 */
935 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700936 }
Nick Piggina60637c2008-07-25 19:45:31 -0700937 if (!page_cache_get_speculative(page))
938 goto repeat;
939
940 /*
941 * Has the page moved?
942 * This is part of the lockless pagecache protocol. See
943 * include/linux/pagemap.h for details.
944 */
945 if (unlikely(page != *pagep)) {
946 page_cache_release(page);
947 goto repeat;
948 }
949 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800950out:
Nick Piggina60637c2008-07-25 19:45:31 -0700951 rcu_read_unlock();
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 return page;
954}
Johannes Weiner0cd61442014-04-03 14:47:46 -0700955EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Randy Dunlap485bb992006-06-23 02:03:49 -0700957/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700958 * find_get_page - find and get a page reference
Martin Waitz67be2dd2005-05-01 08:59:26 -0700959 * @mapping: the address_space to search
960 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700962 * Looks up the page cache slot at @mapping & @offset. If there is a
963 * page cache page, it is returned with an increased refcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700965 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700967struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
968{
969 struct page *page = find_get_entry(mapping, offset);
970
971 if (radix_tree_exceptional_entry(page))
972 page = NULL;
973 return page;
974}
975EXPORT_SYMBOL(find_get_page);
976
977/**
978 * find_lock_entry - locate, pin and lock a page cache entry
979 * @mapping: the address_space to search
980 * @offset: the page cache index
981 *
982 * Looks up the page cache slot at @mapping & @offset. If there is a
983 * page cache page, it is returned locked and with an increased
984 * refcount.
985 *
986 * If the slot holds a shadow entry of a previously evicted page, it
987 * is returned.
988 *
989 * Otherwise, %NULL is returned.
990 *
991 * find_lock_entry() may sleep.
992 */
993struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 struct page *page;
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -0700998 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700999 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001000 lock_page(page);
1001 /* Has the page been truncated? */
1002 if (unlikely(page->mapping != mapping)) {
1003 unlock_page(page);
1004 page_cache_release(page);
1005 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001007 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 return page;
1010}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001011EXPORT_SYMBOL(find_lock_entry);
1012
1013/**
1014 * find_lock_page - locate, pin and lock a pagecache page
1015 * @mapping: the address_space to search
1016 * @offset: the page index
1017 *
1018 * Looks up the page cache slot at @mapping & @offset. If there is a
1019 * page cache page, it is returned locked and with an increased
1020 * refcount.
1021 *
1022 * Otherwise, %NULL is returned.
1023 *
1024 * find_lock_page() may sleep.
1025 */
1026struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
1027{
1028 struct page *page = find_lock_entry(mapping, offset);
1029
1030 if (radix_tree_exceptional_entry(page))
1031 page = NULL;
1032 return page;
1033}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034EXPORT_SYMBOL(find_lock_page);
1035
1036/**
1037 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -07001038 * @mapping: the page's address_space
1039 * @index: the page's index into the mapping
1040 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001042 * Looks up the page cache slot at @mapping & @offset. If there is a
1043 * page cache page, it is returned locked and with an increased
1044 * refcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001046 * If the page is not present, a new page is allocated using @gfp_mask
1047 * and added to the page cache and the VM's LRU list. The page is
1048 * returned locked and with an increased refcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001050 * On memory exhaustion, %NULL is returned.
1051 *
1052 * find_or_create_page() may sleep, even if @gfp_flags specifies an
1053 * atomic allocation!
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 */
1055struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001056 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057{
Nick Piggineb2be182007-10-16 01:24:57 -07001058 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 int err;
1060repeat:
1061 page = find_lock_page(mapping, index);
1062 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001063 page = __page_cache_alloc(gfp_mask);
1064 if (!page)
1065 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -08001066 /*
1067 * We want a regular kernel memory (not highmem or DMA etc)
1068 * allocation for the radix tree nodes, but we need to honour
1069 * the context-specific requirements the caller has asked for.
1070 * GFP_RECLAIM_MASK collects those requirements.
1071 */
1072 err = add_to_page_cache_lru(page, mapping, index,
1073 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -07001074 if (unlikely(err)) {
1075 page_cache_release(page);
1076 page = NULL;
1077 if (err == -EEXIST)
1078 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return page;
1082}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083EXPORT_SYMBOL(find_or_create_page);
1084
1085/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001086 * find_get_entries - gang pagecache lookup
1087 * @mapping: The address_space to search
1088 * @start: The starting page cache index
1089 * @nr_entries: The maximum number of entries
1090 * @entries: Where the resulting entries are placed
1091 * @indices: The cache indices corresponding to the entries in @entries
1092 *
1093 * find_get_entries() will search for and return a group of up to
1094 * @nr_entries entries in the mapping. The entries are placed at
1095 * @entries. find_get_entries() takes a reference against any actual
1096 * pages it returns.
1097 *
1098 * The search returns a group of mapping-contiguous page cache entries
1099 * with ascending indexes. There may be holes in the indices due to
1100 * not-present pages.
1101 *
1102 * Any shadow entries of evicted pages are included in the returned
1103 * array.
1104 *
1105 * find_get_entries() returns the number of pages and shadow entries
1106 * which were found.
1107 */
1108unsigned find_get_entries(struct address_space *mapping,
1109 pgoff_t start, unsigned int nr_entries,
1110 struct page **entries, pgoff_t *indices)
1111{
1112 void **slot;
1113 unsigned int ret = 0;
1114 struct radix_tree_iter iter;
1115
1116 if (!nr_entries)
1117 return 0;
1118
1119 rcu_read_lock();
1120restart:
1121 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1122 struct page *page;
1123repeat:
1124 page = radix_tree_deref_slot(slot);
1125 if (unlikely(!page))
1126 continue;
1127 if (radix_tree_exception(page)) {
1128 if (radix_tree_deref_retry(page))
1129 goto restart;
1130 /*
1131 * Otherwise, we must be storing a swap entry
1132 * here as an exceptional entry: so return it
1133 * without attempting to raise page count.
1134 */
1135 goto export;
1136 }
1137 if (!page_cache_get_speculative(page))
1138 goto repeat;
1139
1140 /* Has the page moved? */
1141 if (unlikely(page != *slot)) {
1142 page_cache_release(page);
1143 goto repeat;
1144 }
1145export:
1146 indices[ret] = iter.index;
1147 entries[ret] = page;
1148 if (++ret == nr_entries)
1149 break;
1150 }
1151 rcu_read_unlock();
1152 return ret;
1153}
1154
1155/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 * find_get_pages - gang pagecache lookup
1157 * @mapping: The address_space to search
1158 * @start: The starting page index
1159 * @nr_pages: The maximum number of pages
1160 * @pages: Where the resulting pages are placed
1161 *
1162 * find_get_pages() will search for and return a group of up to
1163 * @nr_pages pages in the mapping. The pages are placed at @pages.
1164 * find_get_pages() takes a reference against the returned pages.
1165 *
1166 * The search returns a group of mapping-contiguous pages with ascending
1167 * indexes. There may be holes in the indices due to not-present pages.
1168 *
1169 * find_get_pages() returns the number of pages which were found.
1170 */
1171unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1172 unsigned int nr_pages, struct page **pages)
1173{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001174 struct radix_tree_iter iter;
1175 void **slot;
1176 unsigned ret = 0;
1177
1178 if (unlikely(!nr_pages))
1179 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Nick Piggina60637c2008-07-25 19:45:31 -07001181 rcu_read_lock();
1182restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001183 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001184 struct page *page;
1185repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001186 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001187 if (unlikely(!page))
1188 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001189
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001190 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001191 if (radix_tree_deref_retry(page)) {
1192 /*
1193 * Transient condition which can only trigger
1194 * when entry at index 0 moves out of or back
1195 * to root: none yet gotten, safe to restart.
1196 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001197 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001198 goto restart;
1199 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001200 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001201 * Otherwise, shmem/tmpfs must be storing a swap entry
1202 * here as an exceptional entry: so skip over it -
1203 * we only reach this from invalidate_mapping_pages().
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001204 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001205 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001206 }
Nick Piggina60637c2008-07-25 19:45:31 -07001207
1208 if (!page_cache_get_speculative(page))
1209 goto repeat;
1210
1211 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001212 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001213 page_cache_release(page);
1214 goto repeat;
1215 }
1216
1217 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001218 if (++ret == nr_pages)
1219 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001220 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001221
Nick Piggina60637c2008-07-25 19:45:31 -07001222 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 return ret;
1224}
1225
Jens Axboeebf43502006-04-27 08:46:01 +02001226/**
1227 * find_get_pages_contig - gang contiguous pagecache lookup
1228 * @mapping: The address_space to search
1229 * @index: The starting page index
1230 * @nr_pages: The maximum number of pages
1231 * @pages: Where the resulting pages are placed
1232 *
1233 * find_get_pages_contig() works exactly like find_get_pages(), except
1234 * that the returned number of pages are guaranteed to be contiguous.
1235 *
1236 * find_get_pages_contig() returns the number of pages which were found.
1237 */
1238unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1239 unsigned int nr_pages, struct page **pages)
1240{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001241 struct radix_tree_iter iter;
1242 void **slot;
1243 unsigned int ret = 0;
1244
1245 if (unlikely(!nr_pages))
1246 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001247
Nick Piggina60637c2008-07-25 19:45:31 -07001248 rcu_read_lock();
1249restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001250 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001251 struct page *page;
1252repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001253 page = radix_tree_deref_slot(slot);
1254 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001255 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001256 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001257
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001258 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001259 if (radix_tree_deref_retry(page)) {
1260 /*
1261 * Transient condition which can only trigger
1262 * when entry at index 0 moves out of or back
1263 * to root: none yet gotten, safe to restart.
1264 */
1265 goto restart;
1266 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001267 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001268 * Otherwise, shmem/tmpfs must be storing a swap entry
1269 * here as an exceptional entry: so stop looking for
1270 * contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001271 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001272 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001273 }
Nick Piggina60637c2008-07-25 19:45:31 -07001274
Nick Piggina60637c2008-07-25 19:45:31 -07001275 if (!page_cache_get_speculative(page))
1276 goto repeat;
1277
1278 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001279 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001280 page_cache_release(page);
1281 goto repeat;
1282 }
1283
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001284 /*
1285 * must check mapping and index after taking the ref.
1286 * otherwise we can get both false positives and false
1287 * negatives, which is just confusing to the caller.
1288 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001289 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001290 page_cache_release(page);
1291 break;
1292 }
1293
Nick Piggina60637c2008-07-25 19:45:31 -07001294 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001295 if (++ret == nr_pages)
1296 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001297 }
Nick Piggina60637c2008-07-25 19:45:31 -07001298 rcu_read_unlock();
1299 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001300}
David Howellsef71c152007-05-09 02:33:44 -07001301EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001302
Randy Dunlap485bb992006-06-23 02:03:49 -07001303/**
1304 * find_get_pages_tag - find and return pages that match @tag
1305 * @mapping: the address_space to search
1306 * @index: the starting page index
1307 * @tag: the tag index
1308 * @nr_pages: the maximum number of pages
1309 * @pages: where the resulting pages are placed
1310 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001312 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 */
1314unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1315 int tag, unsigned int nr_pages, struct page **pages)
1316{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001317 struct radix_tree_iter iter;
1318 void **slot;
1319 unsigned ret = 0;
1320
1321 if (unlikely(!nr_pages))
1322 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Nick Piggina60637c2008-07-25 19:45:31 -07001324 rcu_read_lock();
1325restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001326 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1327 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001328 struct page *page;
1329repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001330 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001331 if (unlikely(!page))
1332 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001333
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001334 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001335 if (radix_tree_deref_retry(page)) {
1336 /*
1337 * Transient condition which can only trigger
1338 * when entry at index 0 moves out of or back
1339 * to root: none yet gotten, safe to restart.
1340 */
1341 goto restart;
1342 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001343 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001344 * This function is never used on a shmem/tmpfs
1345 * mapping, so a swap entry won't be found here.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001346 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001347 BUG();
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001348 }
Nick Piggina60637c2008-07-25 19:45:31 -07001349
1350 if (!page_cache_get_speculative(page))
1351 goto repeat;
1352
1353 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001354 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001355 page_cache_release(page);
1356 goto repeat;
1357 }
1358
1359 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001360 if (++ret == nr_pages)
1361 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001362 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001363
Nick Piggina60637c2008-07-25 19:45:31 -07001364 rcu_read_unlock();
1365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (ret)
1367 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 return ret;
1370}
David Howellsef71c152007-05-09 02:33:44 -07001371EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Randy Dunlap485bb992006-06-23 02:03:49 -07001373/**
1374 * grab_cache_page_nowait - returns locked page at given index in given cache
1375 * @mapping: target address_space
1376 * @index: the page index
1377 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001378 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 * This is intended for speculative data generators, where the data can
1380 * be regenerated if the page couldn't be grabbed. This routine should
1381 * be safe to call while holding the lock for another page.
1382 *
1383 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1384 * and deadlock against the caller's locked page.
1385 */
1386struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001387grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
1389 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001392 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 return page;
1394 page_cache_release(page);
1395 return NULL;
1396 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001397 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001398 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 page_cache_release(page);
1400 page = NULL;
1401 }
1402 return page;
1403}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404EXPORT_SYMBOL(grab_cache_page_nowait);
1405
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001406/*
1407 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1408 * a _large_ part of the i/o request. Imagine the worst scenario:
1409 *
1410 * ---R__________________________________________B__________
1411 * ^ reading here ^ bad block(assume 4k)
1412 *
1413 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1414 * => failing the whole request => read(R) => read(R+1) =>
1415 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1416 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1417 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1418 *
1419 * It is going insane. Fix it by quickly scaling down the readahead size.
1420 */
1421static void shrink_readahead_size_eio(struct file *filp,
1422 struct file_ra_state *ra)
1423{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001424 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001425}
1426
Randy Dunlap485bb992006-06-23 02:03:49 -07001427/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001428 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001429 * @filp: the file to read
1430 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001431 * @iter: data destination
1432 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001433 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001435 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 *
1437 * This is really ugly. But the goto's actually try to clarify some
1438 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 */
Al Viro6e58e792014-02-03 17:07:03 -05001440static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1441 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001443 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001445 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001446 pgoff_t index;
1447 pgoff_t last_index;
1448 pgoff_t prev_index;
1449 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001450 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001451 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001454 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1455 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Al Viro6e58e792014-02-03 17:07:03 -05001456 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 offset = *ppos & ~PAGE_CACHE_MASK;
1458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 for (;;) {
1460 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001461 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001462 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 unsigned long nr, ret;
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466find_page:
1467 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001468 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001469 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001470 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001471 index, last_index - index);
1472 page = find_get_page(mapping, index);
1473 if (unlikely(page == NULL))
1474 goto no_cached_page;
1475 }
1476 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001477 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001478 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001479 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001481 if (!PageUptodate(page)) {
1482 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1483 !mapping->a_ops->is_partially_uptodate)
1484 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001485 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001486 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001487 /* Did it get truncated before we got the lock? */
1488 if (!page->mapping)
1489 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001490 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001491 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001492 goto page_not_up_to_date_locked;
1493 unlock_page(page);
1494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001496 /*
1497 * i_size must be checked after we know the page is Uptodate.
1498 *
1499 * Checking i_size after the check allows us to calculate
1500 * the correct value for "nr", which means the zero-filled
1501 * part of the page is not copied back to userspace (unless
1502 * another truncate extends the file - this is desired though).
1503 */
1504
1505 isize = i_size_read(inode);
1506 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1507 if (unlikely(!isize || index > end_index)) {
1508 page_cache_release(page);
1509 goto out;
1510 }
1511
1512 /* nr is the maximum number of bytes to copy from this page */
1513 nr = PAGE_CACHE_SIZE;
1514 if (index == end_index) {
1515 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1516 if (nr <= offset) {
1517 page_cache_release(page);
1518 goto out;
1519 }
1520 }
1521 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
1523 /* If users can be writing to this page using arbitrary
1524 * virtual addresses, take care about potential aliasing
1525 * before reading the page on the kernel side.
1526 */
1527 if (mapping_writably_mapped(mapping))
1528 flush_dcache_page(page);
1529
1530 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001531 * When a sequential read accesses a page several times,
1532 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 */
Jan Karaec0f1632007-05-06 14:49:25 -07001534 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 mark_page_accessed(page);
1536 prev_index = index;
1537
1538 /*
1539 * Ok, we have the page, and it's up-to-date, so
1540 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 */
Al Viro6e58e792014-02-03 17:07:03 -05001542
1543 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 offset += ret;
1545 index += offset >> PAGE_CACHE_SHIFT;
1546 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001547 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001550 written += ret;
1551 if (!iov_iter_count(iter))
1552 goto out;
1553 if (ret < nr) {
1554 error = -EFAULT;
1555 goto out;
1556 }
1557 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
1559page_not_up_to_date:
1560 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001561 error = lock_page_killable(page);
1562 if (unlikely(error))
1563 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001565page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001566 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 if (!page->mapping) {
1568 unlock_page(page);
1569 page_cache_release(page);
1570 continue;
1571 }
1572
1573 /* Did somebody else fill it already? */
1574 if (PageUptodate(page)) {
1575 unlock_page(page);
1576 goto page_ok;
1577 }
1578
1579readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001580 /*
1581 * A previous I/O error may have been due to temporary
1582 * failures, eg. multipath errors.
1583 * PG_error will be set again if readpage fails.
1584 */
1585 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 /* Start the actual read. The read will unlock the page. */
1587 error = mapping->a_ops->readpage(filp, page);
1588
Zach Brown994fc28c2005-12-15 14:28:17 -08001589 if (unlikely(error)) {
1590 if (error == AOP_TRUNCATED_PAGE) {
1591 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001592 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001593 goto find_page;
1594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
1598 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001599 error = lock_page_killable(page);
1600 if (unlikely(error))
1601 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 if (!PageUptodate(page)) {
1603 if (page->mapping == NULL) {
1604 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001605 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 */
1607 unlock_page(page);
1608 page_cache_release(page);
1609 goto find_page;
1610 }
1611 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001612 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001613 error = -EIO;
1614 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
1616 unlock_page(page);
1617 }
1618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 goto page_ok;
1620
1621readpage_error:
1622 /* UHHUH! A synchronous read error occurred. Report it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 page_cache_release(page);
1624 goto out;
1625
1626no_cached_page:
1627 /*
1628 * Ok, it wasn't cached, so we need to create a new
1629 * page..
1630 */
Nick Piggineb2be182007-10-16 01:24:57 -07001631 page = page_cache_alloc_cold(mapping);
1632 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001633 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001634 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
Nick Piggineb2be182007-10-16 01:24:57 -07001636 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 index, GFP_KERNEL);
1638 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001639 page_cache_release(page);
Al Viro6e58e792014-02-03 17:07:03 -05001640 if (error == -EEXIST) {
1641 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 goto out;
1645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 goto readpage;
1647 }
1648
1649out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001650 ra->prev_pos = prev_index;
1651 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1652 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001654 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001655 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001656 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657}
1658
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001659/*
1660 * Performs necessary checks before doing a write
1661 * @iov: io vector request
1662 * @nr_segs: number of segments in the iovec
1663 * @count: number of bytes to write
1664 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1665 *
1666 * Adjust number of segments and amount of bytes to write (nr_segs should be
1667 * properly initialized first). Returns appropriate error code that caller
1668 * should return or zero in case that write should be allowed.
1669 */
1670int generic_segment_checks(const struct iovec *iov,
1671 unsigned long *nr_segs, size_t *count, int access_flags)
1672{
1673 unsigned long seg;
1674 size_t cnt = 0;
1675 for (seg = 0; seg < *nr_segs; seg++) {
1676 const struct iovec *iv = &iov[seg];
1677
1678 /*
1679 * If any segment has a negative length, or the cumulative
1680 * length ever wraps negative then return -EINVAL.
1681 */
1682 cnt += iv->iov_len;
1683 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1684 return -EINVAL;
1685 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1686 continue;
1687 if (seg == 0)
1688 return -EFAULT;
1689 *nr_segs = seg;
1690 cnt -= iv->iov_len; /* This segment is no good */
1691 break;
1692 }
1693 *count = cnt;
1694 return 0;
1695}
1696EXPORT_SYMBOL(generic_segment_checks);
1697
Randy Dunlap485bb992006-06-23 02:03:49 -07001698/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001699 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001700 * @iocb: kernel I/O control block
1701 * @iov: io vector request
1702 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001703 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001704 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 * This is the "read()" routine for all filesystems
1706 * that can use the page cache directly.
1707 */
1708ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001709generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1710 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711{
1712 struct file *filp = iocb->ki_filp;
1713 ssize_t retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001715 loff_t *ppos = &iocb->ki_pos;
Al Viro6e58e792014-02-03 17:07:03 -05001716 struct iov_iter i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001719 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1720 if (retval)
1721 return retval;
Al Viro6e58e792014-02-03 17:07:03 -05001722 iov_iter_init(&i, iov, nr_segs, count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
1724 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1725 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001726 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 struct address_space *mapping;
1728 struct inode *inode;
1729
1730 mapping = filp->f_mapping;
1731 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 if (!count)
1733 goto out; /* skip atime */
1734 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001735 retval = filemap_write_and_wait_range(mapping, pos,
Nick Piggin48b47c52009-01-06 14:40:22 -08001736 pos + iov_length(iov, nr_segs) - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001737 if (!retval) {
1738 retval = mapping->a_ops->direct_IO(READ, iocb,
1739 iov, pos, nr_segs);
1740 }
1741 if (retval > 0) {
1742 *ppos = pos + retval;
1743 count -= retval;
Al Viro6e58e792014-02-03 17:07:03 -05001744 /*
1745 * If we did a short DIO read we need to skip the
1746 * section of the iov that we've already read data into.
1747 */
1748 iov_iter_advance(&i, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001749 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001750
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001751 /*
1752 * Btrfs can have a short DIO read if we encounter
1753 * compressed extents, so if there was an error, or if
1754 * we've already read everything we wanted to, or if
1755 * there was a short read because we hit EOF, go ahead
1756 * and return. Otherwise fallthrough to buffered io for
1757 * the rest of the read.
1758 */
1759 if (retval < 0 || !count || *ppos >= size) {
1760 file_accessed(filp);
1761 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
1764
Al Viro6e58e792014-02-03 17:07:03 -05001765 retval = do_generic_file_read(filp, ppos, &i, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766out:
1767 return retval;
1768}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769EXPORT_SYMBOL(generic_file_aio_read);
1770
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001772/**
1773 * page_cache_read - adds requested page to the page cache if not already there
1774 * @file: file to read
1775 * @offset: page index
1776 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 * This adds the requested page to the page cache if it isn't already there,
1778 * and schedules an I/O to read in its contents from disk.
1779 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001780static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
1782 struct address_space *mapping = file->f_mapping;
1783 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001784 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
Zach Brown994fc28c2005-12-15 14:28:17 -08001786 do {
1787 page = page_cache_alloc_cold(mapping);
1788 if (!page)
1789 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790
Zach Brown994fc28c2005-12-15 14:28:17 -08001791 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1792 if (ret == 0)
1793 ret = mapping->a_ops->readpage(file, page);
1794 else if (ret == -EEXIST)
1795 ret = 0; /* losing race to add is OK */
1796
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Zach Brown994fc28c2005-12-15 14:28:17 -08001799 } while (ret == AOP_TRUNCATED_PAGE);
1800
1801 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802}
1803
1804#define MMAP_LOTSAMISS (100)
1805
Linus Torvaldsef00e082009-06-16 15:31:25 -07001806/*
1807 * Synchronous readahead happens when we don't even find
1808 * a page in the page cache at all.
1809 */
1810static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1811 struct file_ra_state *ra,
1812 struct file *file,
1813 pgoff_t offset)
1814{
1815 unsigned long ra_pages;
1816 struct address_space *mapping = file->f_mapping;
1817
1818 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001819 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001820 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001821 if (!ra->ra_pages)
1822 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001823
Joe Perches64363aa2013-07-08 16:00:18 -07001824 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001825 page_cache_sync_readahead(mapping, ra, file, offset,
1826 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001827 return;
1828 }
1829
Andi Kleen207d04b2011-05-24 17:12:29 -07001830 /* Avoid banging the cache line if not needed */
1831 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001832 ra->mmap_miss++;
1833
1834 /*
1835 * Do we miss much more than hit in this file? If so,
1836 * stop bothering with read-ahead. It will only hurt.
1837 */
1838 if (ra->mmap_miss > MMAP_LOTSAMISS)
1839 return;
1840
Wu Fengguangd30a1102009-06-16 15:31:30 -07001841 /*
1842 * mmap read-around
1843 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001844 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001845 ra->start = max_t(long, 0, offset - ra_pages / 2);
1846 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001847 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001848 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001849}
1850
1851/*
1852 * Asynchronous readahead happens when we find the page and PG_readahead,
1853 * so we want to possibly extend the readahead further..
1854 */
1855static void do_async_mmap_readahead(struct vm_area_struct *vma,
1856 struct file_ra_state *ra,
1857 struct file *file,
1858 struct page *page,
1859 pgoff_t offset)
1860{
1861 struct address_space *mapping = file->f_mapping;
1862
1863 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001864 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001865 return;
1866 if (ra->mmap_miss > 0)
1867 ra->mmap_miss--;
1868 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001869 page_cache_async_readahead(mapping, ra, file,
1870 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001871}
1872
Randy Dunlap485bb992006-06-23 02:03:49 -07001873/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001874 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001875 * @vma: vma in which the fault was taken
1876 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001877 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001878 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 * mapped memory region to read in file data during a page fault.
1880 *
1881 * The goto's are kind of ugly, but this streamlines the normal case of having
1882 * it in the page cache, and handles the special cases reasonably without
1883 * having a lot of duplicated code.
1884 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001885int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886{
1887 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001888 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 struct address_space *mapping = file->f_mapping;
1890 struct file_ra_state *ra = &file->f_ra;
1891 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001892 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001894 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001895 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001897 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1898 if (offset >= size >> PAGE_CACHE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001899 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001902 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001904 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001905 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001907 * We found the page, so try async readahead before
1908 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001910 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001911 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001912 /* No page in the page cache at all */
1913 do_sync_mmap_readahead(vma, ra, file, offset);
1914 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001915 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001916 ret = VM_FAULT_MAJOR;
1917retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001918 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 if (!page)
1920 goto no_cached_page;
1921 }
1922
Michel Lespinassed88c0922010-11-02 13:05:18 -07001923 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1924 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001925 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001926 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001927
1928 /* Did it get truncated? */
1929 if (unlikely(page->mapping != mapping)) {
1930 unlock_page(page);
1931 put_page(page);
1932 goto retry_find;
1933 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001934 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07001935
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001937 * We have a locked page in the page cache, now we need to check
1938 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 */
Nick Piggind00806b2007-07-19 01:46:57 -07001940 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 goto page_not_uptodate;
1942
Linus Torvaldsef00e082009-06-16 15:31:25 -07001943 /*
1944 * Found the page and have a reference on it.
1945 * We must recheck i_size under page lock.
1946 */
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07001947 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1948 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001949 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001950 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001951 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001952 }
1953
Nick Piggind0217ac2007-07-19 01:47:03 -07001954 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001955 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957no_cached_page:
1958 /*
1959 * We're only likely to ever get here if MADV_RANDOM is in
1960 * effect.
1961 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001962 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
1964 /*
1965 * The page we want has now been added to the page cache.
1966 * In the unlikely event that someone removed it in the
1967 * meantime, we'll just come back here and read it again.
1968 */
1969 if (error >= 0)
1970 goto retry_find;
1971
1972 /*
1973 * An error return from page_cache_read can result if the
1974 * system is low on memory, or a problem occurs while trying
1975 * to schedule I/O.
1976 */
1977 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001978 return VM_FAULT_OOM;
1979 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
1981page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 /*
1983 * Umm, take care of errors if the page isn't up-to-date.
1984 * Try to re-read it _once_. We do this synchronously,
1985 * because there really aren't any performance issues here
1986 * and we need to check for errors.
1987 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001989 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001990 if (!error) {
1991 wait_on_page_locked(page);
1992 if (!PageUptodate(page))
1993 error = -EIO;
1994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001996
1997 if (!error || error == AOP_TRUNCATED_PAGE)
1998 goto retry_find;
1999
2000 /* Things didn't work out. Return zero to tell the mm layer so. */
2001 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002002 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002003}
2004EXPORT_SYMBOL(filemap_fault);
2005
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002006void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2007{
2008 struct radix_tree_iter iter;
2009 void **slot;
2010 struct file *file = vma->vm_file;
2011 struct address_space *mapping = file->f_mapping;
2012 loff_t size;
2013 struct page *page;
2014 unsigned long address = (unsigned long) vmf->virtual_address;
2015 unsigned long addr;
2016 pte_t *pte;
2017
2018 rcu_read_lock();
2019 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2020 if (iter.index > vmf->max_pgoff)
2021 break;
2022repeat:
2023 page = radix_tree_deref_slot(slot);
2024 if (unlikely(!page))
2025 goto next;
2026 if (radix_tree_exception(page)) {
2027 if (radix_tree_deref_retry(page))
2028 break;
2029 else
2030 goto next;
2031 }
2032
2033 if (!page_cache_get_speculative(page))
2034 goto repeat;
2035
2036 /* Has the page moved? */
2037 if (unlikely(page != *slot)) {
2038 page_cache_release(page);
2039 goto repeat;
2040 }
2041
2042 if (!PageUptodate(page) ||
2043 PageReadahead(page) ||
2044 PageHWPoison(page))
2045 goto skip;
2046 if (!trylock_page(page))
2047 goto skip;
2048
2049 if (page->mapping != mapping || !PageUptodate(page))
2050 goto unlock;
2051
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002052 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2053 if (page->index >= size >> PAGE_CACHE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002054 goto unlock;
2055
2056 pte = vmf->pte + page->index - vmf->pgoff;
2057 if (!pte_none(*pte))
2058 goto unlock;
2059
2060 if (file->f_ra.mmap_miss > 0)
2061 file->f_ra.mmap_miss--;
2062 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2063 do_set_pte(vma, addr, page, pte, false, false);
2064 unlock_page(page);
2065 goto next;
2066unlock:
2067 unlock_page(page);
2068skip:
2069 page_cache_release(page);
2070next:
2071 if (iter.index == vmf->max_pgoff)
2072 break;
2073 }
2074 rcu_read_unlock();
2075}
2076EXPORT_SYMBOL(filemap_map_pages);
2077
Jan Kara4fcf1c62012-06-12 16:20:29 +02002078int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2079{
2080 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002081 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002082 int ret = VM_FAULT_LOCKED;
2083
Jan Kara14da9202012-06-12 16:20:37 +02002084 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002085 file_update_time(vma->vm_file);
2086 lock_page(page);
2087 if (page->mapping != inode->i_mapping) {
2088 unlock_page(page);
2089 ret = VM_FAULT_NOPAGE;
2090 goto out;
2091 }
Jan Kara14da9202012-06-12 16:20:37 +02002092 /*
2093 * We mark the page dirty already here so that when freeze is in
2094 * progress, we are guaranteed that writeback during freezing will
2095 * see the dirty page and writeprotect it again.
2096 */
2097 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002098 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002099out:
Jan Kara14da9202012-06-12 16:20:37 +02002100 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002101 return ret;
2102}
2103EXPORT_SYMBOL(filemap_page_mkwrite);
2104
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002105const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002106 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002107 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002108 .page_mkwrite = filemap_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07002109 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110};
2111
2112/* This is used for a general mmap of a disk file */
2113
2114int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2115{
2116 struct address_space *mapping = file->f_mapping;
2117
2118 if (!mapping->a_ops->readpage)
2119 return -ENOEXEC;
2120 file_accessed(file);
2121 vma->vm_ops = &generic_file_vm_ops;
2122 return 0;
2123}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
2125/*
2126 * This is for filesystems which do not implement ->writepage.
2127 */
2128int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2129{
2130 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2131 return -EINVAL;
2132 return generic_file_mmap(file, vma);
2133}
2134#else
2135int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2136{
2137 return -ENOSYS;
2138}
2139int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2140{
2141 return -ENOSYS;
2142}
2143#endif /* CONFIG_MMU */
2144
2145EXPORT_SYMBOL(generic_file_mmap);
2146EXPORT_SYMBOL(generic_file_readonly_mmap);
2147
Sasha Levin67f9fd92014-04-03 14:48:18 -07002148static struct page *wait_on_page_read(struct page *page)
2149{
2150 if (!IS_ERR(page)) {
2151 wait_on_page_locked(page);
2152 if (!PageUptodate(page)) {
2153 page_cache_release(page);
2154 page = ERR_PTR(-EIO);
2155 }
2156 }
2157 return page;
2158}
2159
Nick Piggin6fe69002007-05-06 14:49:04 -07002160static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002161 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002162 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002163 void *data,
2164 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165{
Nick Piggineb2be182007-10-16 01:24:57 -07002166 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 int err;
2168repeat:
2169 page = find_get_page(mapping, index);
2170 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002171 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002172 if (!page)
2173 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002174 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002175 if (unlikely(err)) {
2176 page_cache_release(page);
2177 if (err == -EEXIST)
2178 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 return ERR_PTR(err);
2181 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 err = filler(data, page);
2183 if (err < 0) {
2184 page_cache_release(page);
2185 page = ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002186 } else {
2187 page = wait_on_page_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 }
2189 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 return page;
2191}
2192
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002193static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002194 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002195 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002196 void *data,
2197 gfp_t gfp)
2198
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199{
2200 struct page *page;
2201 int err;
2202
2203retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002204 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01002206 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 if (PageUptodate(page))
2208 goto out;
2209
2210 lock_page(page);
2211 if (!page->mapping) {
2212 unlock_page(page);
2213 page_cache_release(page);
2214 goto retry;
2215 }
2216 if (PageUptodate(page)) {
2217 unlock_page(page);
2218 goto out;
2219 }
2220 err = filler(data, page);
2221 if (err < 0) {
2222 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01002223 return ERR_PTR(err);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002224 } else {
2225 page = wait_on_page_read(page);
2226 if (IS_ERR(page))
2227 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 }
David Howellsc855ff32007-05-09 13:42:20 +01002229out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002230 mark_page_accessed(page);
2231 return page;
2232}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002233
2234/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002235 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002236 * @mapping: the page's address_space
2237 * @index: the page index
2238 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002239 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002240 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002241 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002242 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002243 *
2244 * If the page does not get brought uptodate, return -EIO.
2245 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002246struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002247 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002248 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002249 void *data)
2250{
2251 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2252}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002253EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002254
2255/**
2256 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2257 * @mapping: the page's address_space
2258 * @index: the page index
2259 * @gfp: the page allocator flags to use if allocating
2260 *
2261 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002262 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002263 *
2264 * If the page does not get brought uptodate, return -EIO.
2265 */
2266struct page *read_cache_page_gfp(struct address_space *mapping,
2267 pgoff_t index,
2268 gfp_t gfp)
2269{
2270 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2271
Sasha Levin67f9fd92014-04-03 14:48:18 -07002272 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002273}
2274EXPORT_SYMBOL(read_cache_page_gfp);
2275
Nick Piggin2f718ff2007-10-16 01:24:59 -07002276/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 * Performs necessary checks before doing a write
2278 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002279 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 * Returns appropriate error code that caller should return or
2281 * zero in case that write should be allowed.
2282 */
2283inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2284{
2285 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002286 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
2288 if (unlikely(*pos < 0))
2289 return -EINVAL;
2290
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 if (!isblk) {
2292 /* FIXME: this is for backwards compatibility with 2.4 */
2293 if (file->f_flags & O_APPEND)
2294 *pos = i_size_read(inode);
2295
2296 if (limit != RLIM_INFINITY) {
2297 if (*pos >= limit) {
2298 send_sig(SIGXFSZ, current, 0);
2299 return -EFBIG;
2300 }
2301 if (*count > limit - (typeof(limit))*pos) {
2302 *count = limit - (typeof(limit))*pos;
2303 }
2304 }
2305 }
2306
2307 /*
2308 * LFS rule
2309 */
2310 if (unlikely(*pos + *count > MAX_NON_LFS &&
2311 !(file->f_flags & O_LARGEFILE))) {
2312 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 return -EFBIG;
2314 }
2315 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2316 *count = MAX_NON_LFS - (unsigned long)*pos;
2317 }
2318 }
2319
2320 /*
2321 * Are we about to exceed the fs block limit ?
2322 *
2323 * If we have written data it becomes a short write. If we have
2324 * exceeded without writing data we send a signal and return EFBIG.
2325 * Linus frestrict idea will clean these up nicely..
2326 */
2327 if (likely(!isblk)) {
2328 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2329 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 return -EFBIG;
2331 }
2332 /* zero-length writes at ->s_maxbytes are OK */
2333 }
2334
2335 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2336 *count = inode->i_sb->s_maxbytes - *pos;
2337 } else {
David Howells93614012006-09-30 20:45:40 +02002338#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 loff_t isize;
2340 if (bdev_read_only(I_BDEV(inode)))
2341 return -EPERM;
2342 isize = i_size_read(inode);
2343 if (*pos >= isize) {
2344 if (*count || *pos > isize)
2345 return -ENOSPC;
2346 }
2347
2348 if (*pos + *count > isize)
2349 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002350#else
2351 return -EPERM;
2352#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 }
2354 return 0;
2355}
2356EXPORT_SYMBOL(generic_write_checks);
2357
Nick Pigginafddba42007-10-16 01:25:01 -07002358int pagecache_write_begin(struct file *file, struct address_space *mapping,
2359 loff_t pos, unsigned len, unsigned flags,
2360 struct page **pagep, void **fsdata)
2361{
2362 const struct address_space_operations *aops = mapping->a_ops;
2363
Nick Piggin4e02ed42008-10-29 14:00:55 -07002364 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002365 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002366}
2367EXPORT_SYMBOL(pagecache_write_begin);
2368
2369int pagecache_write_end(struct file *file, struct address_space *mapping,
2370 loff_t pos, unsigned len, unsigned copied,
2371 struct page *page, void *fsdata)
2372{
2373 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002374
Nick Piggin4e02ed42008-10-29 14:00:55 -07002375 mark_page_accessed(page);
2376 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002377}
2378EXPORT_SYMBOL(pagecache_write_end);
2379
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380ssize_t
2381generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
Al Viro5cb6c6c2014-02-11 20:58:20 -05002382 unsigned long *nr_segs, loff_t pos,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 size_t count, size_t ocount)
2384{
2385 struct file *file = iocb->ki_filp;
2386 struct address_space *mapping = file->f_mapping;
2387 struct inode *inode = mapping->host;
2388 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002389 size_t write_len;
2390 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391
2392 if (count != ocount)
2393 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2394
Christoph Hellwiga969e902008-07-23 21:27:04 -07002395 write_len = iov_length(iov, *nr_segs);
2396 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002397
Nick Piggin48b47c52009-01-06 14:40:22 -08002398 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002399 if (written)
2400 goto out;
2401
2402 /*
2403 * After a write we want buffered reads to be sure to go to disk to get
2404 * the new data. We invalidate clean cached page from the region we're
2405 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002406 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002407 */
2408 if (mapping->nrpages) {
2409 written = invalidate_inode_pages2_range(mapping,
2410 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002411 /*
2412 * If a page can not be invalidated, return 0 to fall back
2413 * to buffered write.
2414 */
2415 if (written) {
2416 if (written == -EBUSY)
2417 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002418 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002419 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002420 }
2421
2422 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2423
2424 /*
2425 * Finally, try again to invalidate clean pages which might have been
2426 * cached by non-direct readahead, or faulted in by get_user_pages()
2427 * if the source of the write was an mmap'ed region of the file
2428 * we're writing. Either one is a pretty crazy thing to do,
2429 * so we don't support it 100%. If this invalidation
2430 * fails, tough, the write still worked...
2431 */
2432 if (mapping->nrpages) {
2433 invalidate_inode_pages2_range(mapping,
2434 pos >> PAGE_CACHE_SHIFT, end);
2435 }
2436
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002438 pos += written;
2439 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2440 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 mark_inode_dirty(inode);
2442 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002443 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002445out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 return written;
2447}
2448EXPORT_SYMBOL(generic_file_direct_write);
2449
Nick Piggineb2be182007-10-16 01:24:57 -07002450/*
2451 * Find or create a page at the given pagecache position. Return the locked
2452 * page. This function is specifically for buffered writes.
2453 */
Nick Piggin54566b22009-01-04 12:00:53 -08002454struct page *grab_cache_page_write_begin(struct address_space *mapping,
2455 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002456{
2457 int status;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002458 gfp_t gfp_mask;
Nick Piggineb2be182007-10-16 01:24:57 -07002459 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002460 gfp_t gfp_notmask = 0;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002461
Fengguang Wu1010bb12012-03-21 16:34:08 -07002462 gfp_mask = mapping_gfp_mask(mapping);
2463 if (mapping_cap_account_dirty(mapping))
2464 gfp_mask |= __GFP_WRITE;
Nick Piggin54566b22009-01-04 12:00:53 -08002465 if (flags & AOP_FLAG_NOFS)
2466 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002467repeat:
2468 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002469 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002470 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002471
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002472 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002473 if (!page)
2474 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002475 status = add_to_page_cache_lru(page, mapping, index,
2476 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002477 if (unlikely(status)) {
2478 page_cache_release(page);
2479 if (status == -EEXIST)
2480 goto repeat;
2481 return NULL;
2482 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002483found:
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002484 wait_for_stable_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002485 return page;
2486}
Nick Piggin54566b22009-01-04 12:00:53 -08002487EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002488
Al Viro3b93f912014-02-11 21:34:08 -05002489ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002490 struct iov_iter *i, loff_t pos)
2491{
2492 struct address_space *mapping = file->f_mapping;
2493 const struct address_space_operations *a_ops = mapping->a_ops;
2494 long status = 0;
2495 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002496 unsigned int flags = 0;
2497
2498 /*
2499 * Copies from kernel address space cannot fail (NFSD is a big user).
2500 */
2501 if (segment_eq(get_fs(), KERNEL_DS))
2502 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002503
2504 do {
2505 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002506 unsigned long offset; /* Offset into pagecache page */
2507 unsigned long bytes; /* Bytes to write to page */
2508 size_t copied; /* Bytes copied from user */
2509 void *fsdata;
2510
2511 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002512 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2513 iov_iter_count(i));
2514
2515again:
Nick Pigginafddba42007-10-16 01:25:01 -07002516 /*
2517 * Bring in the user page that we will copy from _first_.
2518 * Otherwise there's a nasty deadlock on copying from the
2519 * same page as we're writing to, without it being marked
2520 * up-to-date.
2521 *
2522 * Not only is this an optimisation, but it is also required
2523 * to check that the address is actually valid, when atomic
2524 * usercopies are used, below.
2525 */
2526 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2527 status = -EFAULT;
2528 break;
2529 }
2530
Nick Piggin674b8922007-10-16 01:25:03 -07002531 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002532 &page, &fsdata);
2533 if (unlikely(status))
2534 break;
2535
anfei zhou931e80e2010-02-02 13:44:02 -08002536 if (mapping_writably_mapped(mapping))
2537 flush_dcache_page(page);
2538
Nick Pigginafddba42007-10-16 01:25:01 -07002539 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002540 flush_dcache_page(page);
2541
Josef Bacikc8236db2009-07-05 12:08:18 -07002542 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002543 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2544 page, fsdata);
2545 if (unlikely(status < 0))
2546 break;
2547 copied = status;
2548
2549 cond_resched();
2550
Nick Piggin124d3b72008-02-02 15:01:17 +01002551 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002552 if (unlikely(copied == 0)) {
2553 /*
2554 * If we were unable to copy any data at all, we must
2555 * fall back to a single segment length write.
2556 *
2557 * If we didn't fallback here, we could livelock
2558 * because not all segments in the iov can be copied at
2559 * once without a pagefault.
2560 */
2561 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2562 iov_iter_single_seg_count(i));
2563 goto again;
2564 }
Nick Pigginafddba42007-10-16 01:25:01 -07002565 pos += copied;
2566 written += copied;
2567
2568 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002569 if (fatal_signal_pending(current)) {
2570 status = -EINTR;
2571 break;
2572 }
Nick Pigginafddba42007-10-16 01:25:01 -07002573 } while (iov_iter_count(i));
2574
2575 return written ? written : status;
2576}
Al Viro3b93f912014-02-11 21:34:08 -05002577EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Jan Karae4dd9de2009-08-17 18:10:06 +02002579/**
2580 * __generic_file_aio_write - write data to a file
2581 * @iocb: IO state structure (file, offset, etc.)
2582 * @iov: vector with data to write
2583 * @nr_segs: number of segments in the vector
2584 * @ppos: position where to write
2585 *
2586 * This function does all the work needed for actually writing data to a
2587 * file. It does all basic checks, removes SUID from the file, updates
2588 * modification times and calls proper subroutines depending on whether we
2589 * do direct IO or a standard buffered write.
2590 *
2591 * It expects i_mutex to be grabbed unless we work on a block device or similar
2592 * object which does not need locking at all.
2593 *
2594 * This function does *not* take care of syncing data in case of O_SYNC write.
2595 * A caller has to handle it. This is mainly due to the fact that we want to
2596 * avoid syncing under i_mutex.
2597 */
2598ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
Al Viro41fc56d2014-02-09 12:58:52 -05002599 unsigned long nr_segs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600{
2601 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002602 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 size_t ocount; /* original count */
2604 size_t count; /* after file limit checks */
2605 struct inode *inode = mapping->host;
Al Viro41fc56d2014-02-09 12:58:52 -05002606 loff_t pos = iocb->ki_pos;
Al Viro3b93f912014-02-11 21:34:08 -05002607 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002609 ssize_t status;
2610 struct iov_iter from;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
2612 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002613 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2614 if (err)
2615 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
2617 count = ocount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 /* We can write back this queue in page reclaim */
2620 current->backing_dev_info = mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2622 if (err)
2623 goto out;
2624
2625 if (count == 0)
2626 goto out;
2627
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002628 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 if (err)
2630 goto out;
2631
Josef Bacikc3b2da32012-03-26 09:59:21 -04002632 err = file_update_time(file);
2633 if (err)
2634 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635
Al Viro3b93f912014-02-11 21:34:08 -05002636 iov_iter_init(&from, iov, nr_segs, count, 0);
2637
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2639 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002640 loff_t endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002641
Al Viro3b93f912014-02-11 21:34:08 -05002642 written = generic_file_direct_write(iocb, iov, &from.nr_segs, pos,
Al Viro5cb6c6c2014-02-11 20:58:20 -05002643 count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 if (written < 0 || written == count)
2645 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002646 iov_iter_advance(&from, written);
2647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 /*
2649 * direct-io write to a hole: fall through to buffered I/O
2650 * for completing the rest of the request.
2651 */
2652 pos += written;
2653 count -= written;
Al Viro3b93f912014-02-11 21:34:08 -05002654
2655 status = generic_perform_write(file, &from, pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002656 /*
Al Viro3b93f912014-02-11 21:34:08 -05002657 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002658 * then we want to return the number of bytes which were
2659 * direct-written, or the error code if that was zero. Note
2660 * that this differs from normal direct-io semantics, which
2661 * will return -EFOO even if some bytes were written.
2662 */
Al Viro3b93f912014-02-11 21:34:08 -05002663 if (unlikely(status < 0) && !written) {
2664 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002665 goto out;
2666 }
Al Viro3b93f912014-02-11 21:34:08 -05002667 iocb->ki_pos = pos + status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002668 /*
2669 * We need to ensure that the page cache pages are written to
2670 * disk and invalidated to preserve the expected O_DIRECT
2671 * semantics.
2672 */
Al Viro3b93f912014-02-11 21:34:08 -05002673 endbyte = pos + status - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002674 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002675 if (err == 0) {
Al Viro3b93f912014-02-11 21:34:08 -05002676 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002677 invalidate_mapping_pages(mapping,
2678 pos >> PAGE_CACHE_SHIFT,
2679 endbyte >> PAGE_CACHE_SHIFT);
2680 } else {
2681 /*
2682 * We don't know how much we wrote, so just return
2683 * the number of bytes which were direct-written
2684 */
2685 }
2686 } else {
Al Viro3b93f912014-02-11 21:34:08 -05002687 written = generic_perform_write(file, &from, pos);
2688 if (likely(written >= 0))
2689 iocb->ki_pos = pos + written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691out:
2692 current->backing_dev_info = NULL;
2693 return written ? written : err;
2694}
Jan Karae4dd9de2009-08-17 18:10:06 +02002695EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696
Jan Karae4dd9de2009-08-17 18:10:06 +02002697/**
2698 * generic_file_aio_write - write data to a file
2699 * @iocb: IO state structure
2700 * @iov: vector with data to write
2701 * @nr_segs: number of segments in the vector
2702 * @pos: position in file where to write
2703 *
2704 * This is a wrapper around __generic_file_aio_write() to be used by most
2705 * filesystems. It takes care of syncing the file in case of O_SYNC file
2706 * and acquires i_mutex as needed.
2707 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002708ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2709 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710{
2711 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002712 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
2715 BUG_ON(iocb->ki_pos != pos);
2716
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002717 mutex_lock(&inode->i_mutex);
Al Viro41fc56d2014-02-09 12:58:52 -05002718 ret = __generic_file_aio_write(iocb, iov, nr_segs);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002719 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
Christoph Hellwig02afc272013-09-04 15:04:40 +02002721 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 ssize_t err;
2723
Al Virod311d792014-02-09 15:18:09 -05002724 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2725 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 ret = err;
2727 }
2728 return ret;
2729}
2730EXPORT_SYMBOL(generic_file_aio_write);
2731
David Howellscf9a2ae2006-08-29 19:05:54 +01002732/**
2733 * try_to_release_page() - release old fs-specific metadata on a page
2734 *
2735 * @page: the page which the kernel is trying to free
2736 * @gfp_mask: memory allocation flags (and I/O mode)
2737 *
2738 * The address_space is to try to release any data against the page
2739 * (presumably at page->private). If the release was successful, return `1'.
2740 * Otherwise return zero.
2741 *
David Howells266cf652009-04-03 16:42:36 +01002742 * This may also be called if PG_fscache is set on a page, indicating that the
2743 * page is known to the local caching routines.
2744 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002745 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002746 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002747 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002748 */
2749int try_to_release_page(struct page *page, gfp_t gfp_mask)
2750{
2751 struct address_space * const mapping = page->mapping;
2752
2753 BUG_ON(!PageLocked(page));
2754 if (PageWriteback(page))
2755 return 0;
2756
2757 if (mapping && mapping->a_ops->releasepage)
2758 return mapping->a_ops->releasepage(page, gfp_mask);
2759 return try_to_free_buffers(page);
2760}
2761
2762EXPORT_SYMBOL(try_to_release_page);