blob: a603c4d7d3c9c7ff2659bbdd273d20fbb5e90f6e [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>
Nick Piggin0f8053a2006-03-22 00:08:33 -080036#include "internal.h"
37
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070038#define CREATE_TRACE_POINTS
39#include <trace/events/filemap.h>
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 * FIXME: remove all knowledge of the buffer layer from the core VM
43 */
Jan Kara148f9482009-08-17 19:52:36 +020044#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <asm/mman.h>
47
48/*
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
50 * though.
51 *
52 * Shared mappings now work. 15.8.1995 Bruno.
53 *
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 *
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
58 */
59
60/*
61 * Lock ordering:
62 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070063 * ->i_mmap_mutex (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070065 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080068 * ->i_mutex
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070069 * ->i_mmap_mutex (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 *
71 * ->mmap_sem
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070072 * ->i_mmap_mutex
Hugh Dickinsb8072f02005-10-29 18:16:41 -070073 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
75 *
76 * ->mmap_sem
77 * ->lock_page (access_process_vm)
78 *
Nick Piggin82591e62006-10-19 23:29:10 -070079 * ->i_mutex (generic_file_buffered_write)
80 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060082 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110083 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * ->mapping->tree_lock (__sync_single_inode)
85 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070086 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * ->anon_vma.lock (vma_adjust)
88 *
89 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070090 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070093 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080097 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 * ->private_lock (page_remove_rmap->set_page_dirty)
99 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600100 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100101 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600102 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100103 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 *
Andi Kleen9a3c5312012-03-21 16:34:09 -0700106 * ->i_mmap_mutex
107 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 */
109
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700110static void page_cache_tree_delete(struct address_space *mapping,
111 struct page *page, void *shadow)
112{
113 if (shadow) {
114 void **slot;
115
116 slot = radix_tree_lookup_slot(&mapping->page_tree, page->index);
117 radix_tree_replace_slot(slot, shadow);
118 mapping->nrshadows++;
119 /*
120 * Make sure the nrshadows update is committed before
121 * the nrpages update so that final truncate racing
122 * with reclaim does not see both counters 0 at the
123 * same time and miss a shadow entry.
124 */
125 smp_wmb();
126 } else
127 radix_tree_delete(&mapping->page_tree, page->index);
128 mapping->nrpages--;
129}
130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131/*
Minchan Kime64a7822011-03-22 16:32:44 -0700132 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700134 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700136void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
138 struct address_space *mapping = page->mapping;
139
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700140 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600141 /*
142 * if we're uptodate, flush out into the cleancache, otherwise
143 * invalidate any existing cleancache entries. We can't leave
144 * stale data around in the cleancache once our page is gone
145 */
146 if (PageUptodate(page) && PageMappedToDisk(page))
147 cleancache_put_page(page);
148 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400149 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600150
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700151 page_cache_tree_delete(mapping, page, shadow);
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700154 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700155
Christoph Lameter347ce432006-06-30 01:55:35 -0700156 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700157 if (PageSwapBacked(page))
158 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700159 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800160
161 /*
162 * Some filesystems seem to re-dirty the page even after
163 * the VM has canceled the dirty bit (eg ext3 journaling).
164 *
165 * Fix it up by doing a final dirty accounting check after
166 * having removed the page entirely.
167 */
168 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
169 dec_zone_page_state(page, NR_FILE_DIRTY);
170 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172}
173
Minchan Kim702cfbf2011-03-22 16:32:43 -0700174/**
175 * delete_from_page_cache - delete page from page cache
176 * @page: the page which the kernel is trying to remove from page cache
177 *
178 * This must be called only on pages that have been verified to be in the page
179 * cache and locked. It will never put the page into the free list, the caller
180 * has a reference on the page.
181 */
182void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
184 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500185 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Matt Mackallcd7619d2005-05-01 08:59:01 -0700187 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Linus Torvalds6072d132010-12-01 13:35:19 -0500189 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700190 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700191 __delete_from_page_cache(page, NULL);
Nick Piggin19fd6232008-07-25 19:45:32 -0700192 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700193 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500194
195 if (freepage)
196 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700197 page_cache_release(page);
198}
199EXPORT_SYMBOL(delete_from_page_cache);
200
Jens Axboe7eaceac2011-03-10 08:52:07 +0100201static int sleep_on_page(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 io_schedule();
204 return 0;
205}
206
Jens Axboe7eaceac2011-03-10 08:52:07 +0100207static int sleep_on_page_killable(void *word)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500208{
Jens Axboe7eaceac2011-03-10 08:52:07 +0100209 sleep_on_page(word);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500210 return fatal_signal_pending(current) ? -EINTR : 0;
211}
212
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700213static int filemap_check_errors(struct address_space *mapping)
214{
215 int ret = 0;
216 /* Check for outstanding write errors */
217 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
218 ret = -ENOSPC;
219 if (test_and_clear_bit(AS_EIO, &mapping->flags))
220 ret = -EIO;
221 return ret;
222}
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700225 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700226 * @mapping: address space structure to write
227 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800228 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700229 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700231 * Start writeback against all of a mapping's dirty pages that lie
232 * within the byte offsets <start, end> inclusive.
233 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700235 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 * these two operations is that if a dirty page/buffer is encountered, it must
237 * be waited upon, and not just skipped over.
238 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800239int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
240 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
242 int ret;
243 struct writeback_control wbc = {
244 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800245 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700246 .range_start = start,
247 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 };
249
250 if (!mapping_cap_writeback_dirty(mapping))
251 return 0;
252
253 ret = do_writepages(mapping, &wbc);
254 return ret;
255}
256
257static inline int __filemap_fdatawrite(struct address_space *mapping,
258 int sync_mode)
259{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700260 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
263int filemap_fdatawrite(struct address_space *mapping)
264{
265 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
266}
267EXPORT_SYMBOL(filemap_fdatawrite);
268
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400269int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800270 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
272 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
273}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400274EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Randy Dunlap485bb992006-06-23 02:03:49 -0700276/**
277 * filemap_flush - mostly a non-blocking flush
278 * @mapping: target address_space
279 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 * This is a mostly non-blocking flush. Not suitable for data-integrity
281 * purposes - I/O may not be started against all dirty pages.
282 */
283int filemap_flush(struct address_space *mapping)
284{
285 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
286}
287EXPORT_SYMBOL(filemap_flush);
288
Randy Dunlap485bb992006-06-23 02:03:49 -0700289/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200290 * filemap_fdatawait_range - wait for writeback to complete
291 * @mapping: address space structure to wait for
292 * @start_byte: offset in bytes where the range starts
293 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700294 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200295 * Walk the list of under-writeback pages of the given address space
296 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200298int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
299 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200301 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
302 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 struct pagevec pvec;
304 int nr_pages;
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700305 int ret2, ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200307 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700308 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
310 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 while ((index <= end) &&
312 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
313 PAGECACHE_TAG_WRITEBACK,
314 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
315 unsigned i;
316
317 for (i = 0; i < nr_pages; i++) {
318 struct page *page = pvec.pages[i];
319
320 /* until radix tree lookup accepts end_index */
321 if (page->index > end)
322 continue;
323
324 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800325 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 ret = -EIO;
327 }
328 pagevec_release(&pvec);
329 cond_resched();
330 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700331out:
332 ret2 = filemap_check_errors(mapping);
333 if (!ret)
334 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 return ret;
337}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200338EXPORT_SYMBOL(filemap_fdatawait_range);
339
340/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700341 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700343 *
344 * Walk the list of under-writeback pages of the given address space
345 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 */
347int filemap_fdatawait(struct address_space *mapping)
348{
349 loff_t i_size = i_size_read(mapping->host);
350
351 if (i_size == 0)
352 return 0;
353
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200354 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
356EXPORT_SYMBOL(filemap_fdatawait);
357
358int filemap_write_and_wait(struct address_space *mapping)
359{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800360 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800363 err = filemap_fdatawrite(mapping);
364 /*
365 * Even if the above returned error, the pages may be
366 * written partially (e.g. -ENOSPC), so we wait for it.
367 * But the -EIO is special case, it may indicate the worst
368 * thing (e.g. bug) happened, so we avoid waiting for it.
369 */
370 if (err != -EIO) {
371 int err2 = filemap_fdatawait(mapping);
372 if (!err)
373 err = err2;
374 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700375 } else {
376 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800378 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800380EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Randy Dunlap485bb992006-06-23 02:03:49 -0700382/**
383 * filemap_write_and_wait_range - write out & wait on a file range
384 * @mapping: the address_space for the pages
385 * @lstart: offset in bytes where the range starts
386 * @lend: offset in bytes where the range ends (inclusive)
387 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800388 * Write out and wait upon file offsets lstart->lend, inclusive.
389 *
390 * Note that `lend' is inclusive (describes the last byte to be written) so
391 * that this function can be used to write to the very end-of-file (end = -1).
392 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393int filemap_write_and_wait_range(struct address_space *mapping,
394 loff_t lstart, loff_t lend)
395{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800396 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800399 err = __filemap_fdatawrite_range(mapping, lstart, lend,
400 WB_SYNC_ALL);
401 /* See comment of filemap_write_and_wait() */
402 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200403 int err2 = filemap_fdatawait_range(mapping,
404 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800405 if (!err)
406 err = err2;
407 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700408 } else {
409 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800411 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
Chris Masonf6995582009-04-15 13:22:37 -0400413EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Randy Dunlap485bb992006-06-23 02:03:49 -0700415/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700416 * replace_page_cache_page - replace a pagecache page with a new one
417 * @old: page to be replaced
418 * @new: page to replace with
419 * @gfp_mask: allocation mode
420 *
421 * This function replaces a page in the pagecache with a new one. On
422 * success it acquires the pagecache reference for the new page and
423 * drops it for the old page. Both the old and new pages must be
424 * locked. This function does not add the new page to the LRU, the
425 * caller must do that.
426 *
427 * The remove + add is atomic. The only way this function can fail is
428 * memory allocation failure.
429 */
430int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
431{
432 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700433
Sasha Levin309381fea2014-01-23 15:52:54 -0800434 VM_BUG_ON_PAGE(!PageLocked(old), old);
435 VM_BUG_ON_PAGE(!PageLocked(new), new);
436 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700437
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700438 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
439 if (!error) {
440 struct address_space *mapping = old->mapping;
441 void (*freepage)(struct page *);
442
443 pgoff_t offset = old->index;
444 freepage = mapping->a_ops->freepage;
445
446 page_cache_get(new);
447 new->mapping = mapping;
448 new->index = offset;
449
450 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700451 __delete_from_page_cache(old, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700452 error = radix_tree_insert(&mapping->page_tree, offset, new);
453 BUG_ON(error);
454 mapping->nrpages++;
455 __inc_zone_page_state(new, NR_FILE_PAGES);
456 if (PageSwapBacked(new))
457 __inc_zone_page_state(new, NR_SHMEM);
458 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -0800459 /* mem_cgroup codes must not be called under tree_lock */
460 mem_cgroup_replace_page_cache(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700461 radix_tree_preload_end();
462 if (freepage)
463 freepage(old);
464 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700465 }
466
467 return error;
468}
469EXPORT_SYMBOL_GPL(replace_page_cache_page);
470
Johannes Weiner0cd61442014-04-03 14:47:46 -0700471static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700472 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700473{
474 void **slot;
475 int error;
476
477 slot = radix_tree_lookup_slot(&mapping->page_tree, page->index);
478 if (slot) {
479 void *p;
480
481 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
482 if (!radix_tree_exceptional_entry(p))
483 return -EEXIST;
484 radix_tree_replace_slot(slot, page);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700485 mapping->nrshadows--;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700486 mapping->nrpages++;
Johannes Weinera5289102014-04-03 14:47:51 -0700487 if (shadowp)
488 *shadowp = p;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700489 return 0;
490 }
491 error = radix_tree_insert(&mapping->page_tree, page->index, page);
492 if (!error)
493 mapping->nrpages++;
494 return error;
495}
496
Johannes Weinera5289102014-04-03 14:47:51 -0700497static int __add_to_page_cache_locked(struct page *page,
498 struct address_space *mapping,
499 pgoff_t offset, gfp_t gfp_mask,
500 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
Nick Piggine2867812008-07-25 19:45:30 -0700502 int error;
503
Sasha Levin309381fea2014-01-23 15:52:54 -0800504 VM_BUG_ON_PAGE(!PageLocked(page), page);
505 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700506
507 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800508 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800509 if (error)
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700510 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Jan Kara5e4c0d972013-09-11 14:26:05 -0700512 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700513 if (error) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700514 mem_cgroup_uncharge_cache_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700515 return error;
516 }
517
518 page_cache_get(page);
519 page->mapping = mapping;
520 page->index = offset;
521
522 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700523 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700524 radix_tree_preload_end();
525 if (unlikely(error))
526 goto err_insert;
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700527 __inc_zone_page_state(page, NR_FILE_PAGES);
528 spin_unlock_irq(&mapping->tree_lock);
529 trace_mm_filemap_add_to_page_cache(page);
530 return 0;
531err_insert:
532 page->mapping = NULL;
533 /* Leave page->index set: truncation relies upon it */
534 spin_unlock_irq(&mapping->tree_lock);
535 mem_cgroup_uncharge_cache_page(page);
536 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return error;
538}
Johannes Weinera5289102014-04-03 14:47:51 -0700539
540/**
541 * add_to_page_cache_locked - add a locked page to the pagecache
542 * @page: page to add
543 * @mapping: the page's address_space
544 * @offset: page index
545 * @gfp_mask: page allocation mode
546 *
547 * This function is used to add a page to the pagecache. It must be locked.
548 * This function does not add the page to the LRU. The caller must do that.
549 */
550int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
551 pgoff_t offset, gfp_t gfp_mask)
552{
553 return __add_to_page_cache_locked(page, mapping, offset,
554 gfp_mask, NULL);
555}
Nick Piggine2867812008-07-25 19:45:30 -0700556EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400559 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Johannes Weinera5289102014-04-03 14:47:51 -0700561 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700562 int ret;
563
Johannes Weinera5289102014-04-03 14:47:51 -0700564 __set_page_locked(page);
565 ret = __add_to_page_cache_locked(page, mapping, offset,
566 gfp_mask, &shadow);
567 if (unlikely(ret))
568 __clear_page_locked(page);
569 else {
570 /*
571 * The page might have been evicted from cache only
572 * recently, in which case it should be activated like
573 * any other repeatedly accessed page.
574 */
575 if (shadow && workingset_refault(shadow)) {
576 SetPageActive(page);
577 workingset_activation(page);
578 } else
579 ClearPageActive(page);
580 lru_cache_add(page);
581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 return ret;
583}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300584EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Paul Jackson44110fe2006-03-24 03:16:04 -0800586#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700587struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800588{
Miao Xiec0ff7452010-05-24 14:32:08 -0700589 int n;
590 struct page *page;
591
Paul Jackson44110fe2006-03-24 03:16:04 -0800592 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700593 unsigned int cpuset_mems_cookie;
594 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700595 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700596 n = cpuset_mem_spread_node();
597 page = alloc_pages_exact_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700598 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700599
Miao Xiec0ff7452010-05-24 14:32:08 -0700600 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800601 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700602 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800603}
Nick Piggin2ae88142006-10-28 10:38:23 -0700604EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800605#endif
606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607/*
608 * In order to wait for pages to become available there must be
609 * waitqueues associated with pages. By using a hash table of
610 * waitqueues where the bucket discipline is to maintain all
611 * waiters on the same queue and wake all when any of the pages
612 * become available, and for the woken contexts to check to be
613 * sure the appropriate page became available, this saves space
614 * at a cost of "thundering herd" phenomena during rare hash
615 * collisions.
616 */
617static wait_queue_head_t *page_waitqueue(struct page *page)
618{
619 const struct zone *zone = page_zone(page);
620
621 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
622}
623
624static inline void wake_up_page(struct page *page, int bit)
625{
626 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
627}
628
Harvey Harrison920c7a52008-02-04 22:29:26 -0800629void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
631 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
632
633 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100634 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 TASK_UNINTERRUPTIBLE);
636}
637EXPORT_SYMBOL(wait_on_page_bit);
638
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700639int wait_on_page_bit_killable(struct page *page, int bit_nr)
640{
641 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
642
643 if (!test_bit(bit_nr, &page->flags))
644 return 0;
645
646 return __wait_on_bit(page_waitqueue(page), &wait,
647 sleep_on_page_killable, TASK_KILLABLE);
648}
649
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100651 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700652 * @page: Page defining the wait queue of interest
653 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100654 *
655 * Add an arbitrary @waiter to the wait queue for the nominated @page.
656 */
657void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
658{
659 wait_queue_head_t *q = page_waitqueue(page);
660 unsigned long flags;
661
662 spin_lock_irqsave(&q->lock, flags);
663 __add_wait_queue(q, waiter);
664 spin_unlock_irqrestore(&q->lock, flags);
665}
666EXPORT_SYMBOL_GPL(add_page_wait_queue);
667
668/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700669 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 * @page: the page
671 *
672 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
673 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
674 * mechananism between PageLocked pages and PageWriteback pages is shared.
675 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
676 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700677 * The mb is necessary to enforce ordering between the clear_bit and the read
678 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800680void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Sasha Levin309381fea2014-01-23 15:52:54 -0800682 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700683 clear_bit_unlock(PG_locked, &page->flags);
684 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 wake_up_page(page, PG_locked);
686}
687EXPORT_SYMBOL(unlock_page);
688
Randy Dunlap485bb992006-06-23 02:03:49 -0700689/**
690 * end_page_writeback - end writeback against a page
691 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 */
693void end_page_writeback(struct page *page)
694{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700695 if (TestClearPageReclaim(page))
696 rotate_reclaimable_page(page);
697
698 if (!test_clear_page_writeback(page))
699 BUG();
700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 smp_mb__after_clear_bit();
702 wake_up_page(page, PG_writeback);
703}
704EXPORT_SYMBOL(end_page_writeback);
705
Randy Dunlap485bb992006-06-23 02:03:49 -0700706/**
707 * __lock_page - get a lock on the page, assuming we need to sleep to get it
708 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800710void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
712 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
713
Jens Axboe7eaceac2011-03-10 08:52:07 +0100714 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 TASK_UNINTERRUPTIBLE);
716}
717EXPORT_SYMBOL(__lock_page);
718
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800719int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500720{
721 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
722
723 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100724 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500725}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300726EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500727
Michel Lespinassed065bd82010-10-26 14:21:57 -0700728int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
729 unsigned int flags)
730{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700731 if (flags & FAULT_FLAG_ALLOW_RETRY) {
732 /*
733 * CAUTION! In this case, mmap_sem is not released
734 * even though return 0.
735 */
736 if (flags & FAULT_FLAG_RETRY_NOWAIT)
737 return 0;
738
739 up_read(&mm->mmap_sem);
740 if (flags & FAULT_FLAG_KILLABLE)
741 wait_on_page_locked_killable(page);
742 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700743 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700744 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700745 } else {
746 if (flags & FAULT_FLAG_KILLABLE) {
747 int ret;
748
749 ret = __lock_page_killable(page);
750 if (ret) {
751 up_read(&mm->mmap_sem);
752 return 0;
753 }
754 } else
755 __lock_page(page);
756 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700757 }
758}
759
Randy Dunlap485bb992006-06-23 02:03:49 -0700760/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700761 * page_cache_next_hole - find the next hole (not-present entry)
762 * @mapping: mapping
763 * @index: index
764 * @max_scan: maximum range to search
765 *
766 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
767 * lowest indexed hole.
768 *
769 * Returns: the index of the hole if found, otherwise returns an index
770 * outside of the set specified (in which case 'return - index >=
771 * max_scan' will be true). In rare cases of index wrap-around, 0 will
772 * be returned.
773 *
774 * page_cache_next_hole may be called under rcu_read_lock. However,
775 * like radix_tree_gang_lookup, this will not atomically search a
776 * snapshot of the tree at a single point in time. For example, if a
777 * hole is created at index 5, then subsequently a hole is created at
778 * index 10, page_cache_next_hole covering both indexes may return 10
779 * if called under rcu_read_lock.
780 */
781pgoff_t page_cache_next_hole(struct address_space *mapping,
782 pgoff_t index, unsigned long max_scan)
783{
784 unsigned long i;
785
786 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700787 struct page *page;
788
789 page = radix_tree_lookup(&mapping->page_tree, index);
790 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700791 break;
792 index++;
793 if (index == 0)
794 break;
795 }
796
797 return index;
798}
799EXPORT_SYMBOL(page_cache_next_hole);
800
801/**
802 * page_cache_prev_hole - find the prev hole (not-present entry)
803 * @mapping: mapping
804 * @index: index
805 * @max_scan: maximum range to search
806 *
807 * Search backwards in the range [max(index-max_scan+1, 0), index] for
808 * the first hole.
809 *
810 * Returns: the index of the hole if found, otherwise returns an index
811 * outside of the set specified (in which case 'index - return >=
812 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
813 * will be returned.
814 *
815 * page_cache_prev_hole may be called under rcu_read_lock. However,
816 * like radix_tree_gang_lookup, this will not atomically search a
817 * snapshot of the tree at a single point in time. For example, if a
818 * hole is created at index 10, then subsequently a hole is created at
819 * index 5, page_cache_prev_hole covering both indexes may return 5 if
820 * called under rcu_read_lock.
821 */
822pgoff_t page_cache_prev_hole(struct address_space *mapping,
823 pgoff_t index, unsigned long max_scan)
824{
825 unsigned long i;
826
827 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700828 struct page *page;
829
830 page = radix_tree_lookup(&mapping->page_tree, index);
831 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -0700832 break;
833 index--;
834 if (index == ULONG_MAX)
835 break;
836 }
837
838 return index;
839}
840EXPORT_SYMBOL(page_cache_prev_hole);
841
842/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700843 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -0700844 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -0700845 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -0700846 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700847 * Looks up the page cache slot at @mapping & @offset. If there is a
848 * page cache page, it is returned with an increased refcount.
849 *
850 * If the slot holds a shadow entry of a previously evicted page, it
851 * is returned.
852 *
853 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700855struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856{
Nick Piggina60637c2008-07-25 19:45:31 -0700857 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 struct page *page;
859
Nick Piggina60637c2008-07-25 19:45:31 -0700860 rcu_read_lock();
861repeat:
862 page = NULL;
863 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
864 if (pagep) {
865 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800866 if (unlikely(!page))
867 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700868 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700869 if (radix_tree_deref_retry(page))
870 goto repeat;
871 /*
872 * Otherwise, shmem/tmpfs must be storing a swap entry
873 * here as an exceptional entry: so return it without
874 * attempting to raise page count.
875 */
876 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700877 }
Nick Piggina60637c2008-07-25 19:45:31 -0700878 if (!page_cache_get_speculative(page))
879 goto repeat;
880
881 /*
882 * Has the page moved?
883 * This is part of the lockless pagecache protocol. See
884 * include/linux/pagemap.h for details.
885 */
886 if (unlikely(page != *pagep)) {
887 page_cache_release(page);
888 goto repeat;
889 }
890 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800891out:
Nick Piggina60637c2008-07-25 19:45:31 -0700892 rcu_read_unlock();
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 return page;
895}
Johannes Weiner0cd61442014-04-03 14:47:46 -0700896EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
Randy Dunlap485bb992006-06-23 02:03:49 -0700898/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700899 * find_get_page - find and get a page reference
Martin Waitz67be2dd2005-05-01 08:59:26 -0700900 * @mapping: the address_space to search
901 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700903 * Looks up the page cache slot at @mapping & @offset. If there is a
904 * page cache page, it is returned with an increased refcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700906 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700908struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
909{
910 struct page *page = find_get_entry(mapping, offset);
911
912 if (radix_tree_exceptional_entry(page))
913 page = NULL;
914 return page;
915}
916EXPORT_SYMBOL(find_get_page);
917
918/**
919 * find_lock_entry - locate, pin and lock a page cache entry
920 * @mapping: the address_space to search
921 * @offset: the page cache index
922 *
923 * Looks up the page cache slot at @mapping & @offset. If there is a
924 * page cache page, it is returned locked and with an increased
925 * refcount.
926 *
927 * If the slot holds a shadow entry of a previously evicted page, it
928 * is returned.
929 *
930 * Otherwise, %NULL is returned.
931 *
932 * find_lock_entry() may sleep.
933 */
934struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
936 struct page *page;
937
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -0700939 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700940 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700941 lock_page(page);
942 /* Has the page been truncated? */
943 if (unlikely(page->mapping != mapping)) {
944 unlock_page(page);
945 page_cache_release(page);
946 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800948 VM_BUG_ON_PAGE(page->index != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 return page;
951}
Johannes Weiner0cd61442014-04-03 14:47:46 -0700952EXPORT_SYMBOL(find_lock_entry);
953
954/**
955 * find_lock_page - locate, pin and lock a pagecache page
956 * @mapping: the address_space to search
957 * @offset: the page index
958 *
959 * Looks up the page cache slot at @mapping & @offset. If there is a
960 * page cache page, it is returned locked and with an increased
961 * refcount.
962 *
963 * Otherwise, %NULL is returned.
964 *
965 * find_lock_page() may sleep.
966 */
967struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
968{
969 struct page *page = find_lock_entry(mapping, offset);
970
971 if (radix_tree_exceptional_entry(page))
972 page = NULL;
973 return page;
974}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975EXPORT_SYMBOL(find_lock_page);
976
977/**
978 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700979 * @mapping: the page's address_space
980 * @index: the page's index into the mapping
981 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700983 * Looks up the page cache slot at @mapping & @offset. If there is a
984 * page cache page, it is returned locked and with an increased
985 * refcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700987 * If the page is not present, a new page is allocated using @gfp_mask
988 * and added to the page cache and the VM's LRU list. The page is
989 * returned locked and with an increased refcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 *
Johannes Weiner0cd61442014-04-03 14:47:46 -0700991 * On memory exhaustion, %NULL is returned.
992 *
993 * find_or_create_page() may sleep, even if @gfp_flags specifies an
994 * atomic allocation!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 */
996struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700997 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
Nick Piggineb2be182007-10-16 01:24:57 -0700999 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 int err;
1001repeat:
1002 page = find_lock_page(mapping, index);
1003 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001004 page = __page_cache_alloc(gfp_mask);
1005 if (!page)
1006 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -08001007 /*
1008 * We want a regular kernel memory (not highmem or DMA etc)
1009 * allocation for the radix tree nodes, but we need to honour
1010 * the context-specific requirements the caller has asked for.
1011 * GFP_RECLAIM_MASK collects those requirements.
1012 */
1013 err = add_to_page_cache_lru(page, mapping, index,
1014 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -07001015 if (unlikely(err)) {
1016 page_cache_release(page);
1017 page = NULL;
1018 if (err == -EEXIST)
1019 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 return page;
1023}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024EXPORT_SYMBOL(find_or_create_page);
1025
1026/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001027 * find_get_entries - gang pagecache lookup
1028 * @mapping: The address_space to search
1029 * @start: The starting page cache index
1030 * @nr_entries: The maximum number of entries
1031 * @entries: Where the resulting entries are placed
1032 * @indices: The cache indices corresponding to the entries in @entries
1033 *
1034 * find_get_entries() will search for and return a group of up to
1035 * @nr_entries entries in the mapping. The entries are placed at
1036 * @entries. find_get_entries() takes a reference against any actual
1037 * pages it returns.
1038 *
1039 * The search returns a group of mapping-contiguous page cache entries
1040 * with ascending indexes. There may be holes in the indices due to
1041 * not-present pages.
1042 *
1043 * Any shadow entries of evicted pages are included in the returned
1044 * array.
1045 *
1046 * find_get_entries() returns the number of pages and shadow entries
1047 * which were found.
1048 */
1049unsigned find_get_entries(struct address_space *mapping,
1050 pgoff_t start, unsigned int nr_entries,
1051 struct page **entries, pgoff_t *indices)
1052{
1053 void **slot;
1054 unsigned int ret = 0;
1055 struct radix_tree_iter iter;
1056
1057 if (!nr_entries)
1058 return 0;
1059
1060 rcu_read_lock();
1061restart:
1062 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1063 struct page *page;
1064repeat:
1065 page = radix_tree_deref_slot(slot);
1066 if (unlikely(!page))
1067 continue;
1068 if (radix_tree_exception(page)) {
1069 if (radix_tree_deref_retry(page))
1070 goto restart;
1071 /*
1072 * Otherwise, we must be storing a swap entry
1073 * here as an exceptional entry: so return it
1074 * without attempting to raise page count.
1075 */
1076 goto export;
1077 }
1078 if (!page_cache_get_speculative(page))
1079 goto repeat;
1080
1081 /* Has the page moved? */
1082 if (unlikely(page != *slot)) {
1083 page_cache_release(page);
1084 goto repeat;
1085 }
1086export:
1087 indices[ret] = iter.index;
1088 entries[ret] = page;
1089 if (++ret == nr_entries)
1090 break;
1091 }
1092 rcu_read_unlock();
1093 return ret;
1094}
1095
1096/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 * find_get_pages - gang pagecache lookup
1098 * @mapping: The address_space to search
1099 * @start: The starting page index
1100 * @nr_pages: The maximum number of pages
1101 * @pages: Where the resulting pages are placed
1102 *
1103 * find_get_pages() will search for and return a group of up to
1104 * @nr_pages pages in the mapping. The pages are placed at @pages.
1105 * find_get_pages() takes a reference against the returned pages.
1106 *
1107 * The search returns a group of mapping-contiguous pages with ascending
1108 * indexes. There may be holes in the indices due to not-present pages.
1109 *
1110 * find_get_pages() returns the number of pages which were found.
1111 */
1112unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1113 unsigned int nr_pages, struct page **pages)
1114{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001115 struct radix_tree_iter iter;
1116 void **slot;
1117 unsigned ret = 0;
1118
1119 if (unlikely(!nr_pages))
1120 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Nick Piggina60637c2008-07-25 19:45:31 -07001122 rcu_read_lock();
1123restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001124 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -07001125 struct page *page;
1126repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001127 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001128 if (unlikely(!page))
1129 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001130
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001131 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001132 if (radix_tree_deref_retry(page)) {
1133 /*
1134 * Transient condition which can only trigger
1135 * when entry at index 0 moves out of or back
1136 * to root: none yet gotten, safe to restart.
1137 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001138 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001139 goto restart;
1140 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001141 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001142 * Otherwise, shmem/tmpfs must be storing a swap entry
1143 * here as an exceptional entry: so skip over it -
1144 * we only reach this from invalidate_mapping_pages().
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001145 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001146 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001147 }
Nick Piggina60637c2008-07-25 19:45:31 -07001148
1149 if (!page_cache_get_speculative(page))
1150 goto repeat;
1151
1152 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001153 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001154 page_cache_release(page);
1155 goto repeat;
1156 }
1157
1158 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001159 if (++ret == nr_pages)
1160 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001161 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001162
Nick Piggina60637c2008-07-25 19:45:31 -07001163 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 return ret;
1165}
1166
Jens Axboeebf43502006-04-27 08:46:01 +02001167/**
1168 * find_get_pages_contig - gang contiguous pagecache lookup
1169 * @mapping: The address_space to search
1170 * @index: The starting page index
1171 * @nr_pages: The maximum number of pages
1172 * @pages: Where the resulting pages are placed
1173 *
1174 * find_get_pages_contig() works exactly like find_get_pages(), except
1175 * that the returned number of pages are guaranteed to be contiguous.
1176 *
1177 * find_get_pages_contig() returns the number of pages which were found.
1178 */
1179unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1180 unsigned int nr_pages, struct page **pages)
1181{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001182 struct radix_tree_iter iter;
1183 void **slot;
1184 unsigned int ret = 0;
1185
1186 if (unlikely(!nr_pages))
1187 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001188
Nick Piggina60637c2008-07-25 19:45:31 -07001189 rcu_read_lock();
1190restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001191 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -07001192 struct page *page;
1193repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001194 page = radix_tree_deref_slot(slot);
1195 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001196 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001197 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001198
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001199 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001200 if (radix_tree_deref_retry(page)) {
1201 /*
1202 * Transient condition which can only trigger
1203 * when entry at index 0 moves out of or back
1204 * to root: none yet gotten, safe to restart.
1205 */
1206 goto restart;
1207 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001208 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001209 * Otherwise, shmem/tmpfs must be storing a swap entry
1210 * here as an exceptional entry: so stop looking for
1211 * contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001212 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001213 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001214 }
Nick Piggina60637c2008-07-25 19:45:31 -07001215
Nick Piggina60637c2008-07-25 19:45:31 -07001216 if (!page_cache_get_speculative(page))
1217 goto repeat;
1218
1219 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001220 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001221 page_cache_release(page);
1222 goto repeat;
1223 }
1224
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001225 /*
1226 * must check mapping and index after taking the ref.
1227 * otherwise we can get both false positives and false
1228 * negatives, which is just confusing to the caller.
1229 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001230 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001231 page_cache_release(page);
1232 break;
1233 }
1234
Nick Piggina60637c2008-07-25 19:45:31 -07001235 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001236 if (++ret == nr_pages)
1237 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001238 }
Nick Piggina60637c2008-07-25 19:45:31 -07001239 rcu_read_unlock();
1240 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001241}
David Howellsef71c152007-05-09 02:33:44 -07001242EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001243
Randy Dunlap485bb992006-06-23 02:03:49 -07001244/**
1245 * find_get_pages_tag - find and return pages that match @tag
1246 * @mapping: the address_space to search
1247 * @index: the starting page index
1248 * @tag: the tag index
1249 * @nr_pages: the maximum number of pages
1250 * @pages: where the resulting pages are placed
1251 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001253 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 */
1255unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1256 int tag, unsigned int nr_pages, struct page **pages)
1257{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001258 struct radix_tree_iter iter;
1259 void **slot;
1260 unsigned ret = 0;
1261
1262 if (unlikely(!nr_pages))
1263 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Nick Piggina60637c2008-07-25 19:45:31 -07001265 rcu_read_lock();
1266restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001267 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1268 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -07001269 struct page *page;
1270repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001271 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001272 if (unlikely(!page))
1273 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001274
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001275 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001276 if (radix_tree_deref_retry(page)) {
1277 /*
1278 * Transient condition which can only trigger
1279 * when entry at index 0 moves out of or back
1280 * to root: none yet gotten, safe to restart.
1281 */
1282 goto restart;
1283 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001284 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001285 * This function is never used on a shmem/tmpfs
1286 * mapping, so a swap entry won't be found here.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001287 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001288 BUG();
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001289 }
Nick Piggina60637c2008-07-25 19:45:31 -07001290
1291 if (!page_cache_get_speculative(page))
1292 goto repeat;
1293
1294 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001295 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001296 page_cache_release(page);
1297 goto repeat;
1298 }
1299
1300 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001301 if (++ret == nr_pages)
1302 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001303 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001304
Nick Piggina60637c2008-07-25 19:45:31 -07001305 rcu_read_unlock();
1306
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 if (ret)
1308 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 return ret;
1311}
David Howellsef71c152007-05-09 02:33:44 -07001312EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Randy Dunlap485bb992006-06-23 02:03:49 -07001314/**
1315 * grab_cache_page_nowait - returns locked page at given index in given cache
1316 * @mapping: target address_space
1317 * @index: the page index
1318 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001319 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 * This is intended for speculative data generators, where the data can
1321 * be regenerated if the page couldn't be grabbed. This routine should
1322 * be safe to call while holding the lock for another page.
1323 *
1324 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1325 * and deadlock against the caller's locked page.
1326 */
1327struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001328grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
1330 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
1332 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001333 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 return page;
1335 page_cache_release(page);
1336 return NULL;
1337 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001338 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001339 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 page_cache_release(page);
1341 page = NULL;
1342 }
1343 return page;
1344}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345EXPORT_SYMBOL(grab_cache_page_nowait);
1346
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001347/*
1348 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1349 * a _large_ part of the i/o request. Imagine the worst scenario:
1350 *
1351 * ---R__________________________________________B__________
1352 * ^ reading here ^ bad block(assume 4k)
1353 *
1354 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1355 * => failing the whole request => read(R) => read(R+1) =>
1356 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1357 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1358 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1359 *
1360 * It is going insane. Fix it by quickly scaling down the readahead size.
1361 */
1362static void shrink_readahead_size_eio(struct file *filp,
1363 struct file_ra_state *ra)
1364{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001365 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001366}
1367
Randy Dunlap485bb992006-06-23 02:03:49 -07001368/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001369 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001370 * @filp: the file to read
1371 * @ppos: current file position
1372 * @desc: read_descriptor
Randy Dunlap485bb992006-06-23 02:03:49 -07001373 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001375 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 *
1377 * This is really ugly. But the goto's actually try to clarify some
1378 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001380static void do_generic_file_read(struct file *filp, loff_t *ppos,
Kirill A. Shutemovb77d88d2013-11-14 14:30:40 -08001381 read_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001383 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001385 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001386 pgoff_t index;
1387 pgoff_t last_index;
1388 pgoff_t prev_index;
1389 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001390 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001394 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1395 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1397 offset = *ppos & ~PAGE_CACHE_MASK;
1398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 for (;;) {
1400 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001401 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001402 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 unsigned long nr, ret;
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406find_page:
1407 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001408 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001409 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001410 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001411 index, last_index - index);
1412 page = find_get_page(mapping, index);
1413 if (unlikely(page == NULL))
1414 goto no_cached_page;
1415 }
1416 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001417 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001418 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001419 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001421 if (!PageUptodate(page)) {
1422 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1423 !mapping->a_ops->is_partially_uptodate)
1424 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001425 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001426 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001427 /* Did it get truncated before we got the lock? */
1428 if (!page->mapping)
1429 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001430 if (!mapping->a_ops->is_partially_uptodate(page,
1431 desc, offset))
1432 goto page_not_up_to_date_locked;
1433 unlock_page(page);
1434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001436 /*
1437 * i_size must be checked after we know the page is Uptodate.
1438 *
1439 * Checking i_size after the check allows us to calculate
1440 * the correct value for "nr", which means the zero-filled
1441 * part of the page is not copied back to userspace (unless
1442 * another truncate extends the file - this is desired though).
1443 */
1444
1445 isize = i_size_read(inode);
1446 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1447 if (unlikely(!isize || index > end_index)) {
1448 page_cache_release(page);
1449 goto out;
1450 }
1451
1452 /* nr is the maximum number of bytes to copy from this page */
1453 nr = PAGE_CACHE_SIZE;
1454 if (index == end_index) {
1455 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1456 if (nr <= offset) {
1457 page_cache_release(page);
1458 goto out;
1459 }
1460 }
1461 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
1463 /* If users can be writing to this page using arbitrary
1464 * virtual addresses, take care about potential aliasing
1465 * before reading the page on the kernel side.
1466 */
1467 if (mapping_writably_mapped(mapping))
1468 flush_dcache_page(page);
1469
1470 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001471 * When a sequential read accesses a page several times,
1472 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 */
Jan Karaec0f1632007-05-06 14:49:25 -07001474 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 mark_page_accessed(page);
1476 prev_index = index;
1477
1478 /*
1479 * Ok, we have the page, and it's up-to-date, so
1480 * now we can copy it to user space...
1481 *
Kirill A. Shutemovb77d88d2013-11-14 14:30:40 -08001482 * The file_read_actor routine returns how many bytes were
1483 * actually used..
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 * NOTE! This may not be the same as how much of a user buffer
1485 * we filled up (we may be padding etc), so we can only update
1486 * "pos" here (the actor routine has to update the user buffer
1487 * pointers and the remaining count).
1488 */
Kirill A. Shutemovb77d88d2013-11-14 14:30:40 -08001489 ret = file_read_actor(desc, page, offset, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 offset += ret;
1491 index += offset >> PAGE_CACHE_SHIFT;
1492 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001493 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 page_cache_release(page);
1496 if (ret == nr && desc->count)
1497 continue;
1498 goto out;
1499
1500page_not_up_to_date:
1501 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001502 error = lock_page_killable(page);
1503 if (unlikely(error))
1504 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001506page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001507 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (!page->mapping) {
1509 unlock_page(page);
1510 page_cache_release(page);
1511 continue;
1512 }
1513
1514 /* Did somebody else fill it already? */
1515 if (PageUptodate(page)) {
1516 unlock_page(page);
1517 goto page_ok;
1518 }
1519
1520readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001521 /*
1522 * A previous I/O error may have been due to temporary
1523 * failures, eg. multipath errors.
1524 * PG_error will be set again if readpage fails.
1525 */
1526 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 /* Start the actual read. The read will unlock the page. */
1528 error = mapping->a_ops->readpage(filp, page);
1529
Zach Brown994fc28c2005-12-15 14:28:17 -08001530 if (unlikely(error)) {
1531 if (error == AOP_TRUNCATED_PAGE) {
1532 page_cache_release(page);
1533 goto find_page;
1534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
1538 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001539 error = lock_page_killable(page);
1540 if (unlikely(error))
1541 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 if (!PageUptodate(page)) {
1543 if (page->mapping == NULL) {
1544 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001545 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 */
1547 unlock_page(page);
1548 page_cache_release(page);
1549 goto find_page;
1550 }
1551 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001552 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001553 error = -EIO;
1554 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 }
1556 unlock_page(page);
1557 }
1558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 goto page_ok;
1560
1561readpage_error:
1562 /* UHHUH! A synchronous read error occurred. Report it */
1563 desc->error = error;
1564 page_cache_release(page);
1565 goto out;
1566
1567no_cached_page:
1568 /*
1569 * Ok, it wasn't cached, so we need to create a new
1570 * page..
1571 */
Nick Piggineb2be182007-10-16 01:24:57 -07001572 page = page_cache_alloc_cold(mapping);
1573 if (!page) {
1574 desc->error = -ENOMEM;
1575 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 }
Nick Piggineb2be182007-10-16 01:24:57 -07001577 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 index, GFP_KERNEL);
1579 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001580 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 if (error == -EEXIST)
1582 goto find_page;
1583 desc->error = error;
1584 goto out;
1585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 goto readpage;
1587 }
1588
1589out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001590 ra->prev_pos = prev_index;
1591 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1592 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001594 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001595 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
1598int file_read_actor(read_descriptor_t *desc, struct page *page,
1599 unsigned long offset, unsigned long size)
1600{
1601 char *kaddr;
1602 unsigned long left, count = desc->count;
1603
1604 if (size > count)
1605 size = count;
1606
1607 /*
1608 * Faults on the destination of a read are common, so do it before
1609 * taking the kmap.
1610 */
1611 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001612 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 left = __copy_to_user_inatomic(desc->arg.buf,
1614 kaddr + offset, size);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001615 kunmap_atomic(kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 if (left == 0)
1617 goto success;
1618 }
1619
1620 /* Do it the slow way */
1621 kaddr = kmap(page);
1622 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1623 kunmap(page);
1624
1625 if (left) {
1626 size -= left;
1627 desc->error = -EFAULT;
1628 }
1629success:
1630 desc->count = count - size;
1631 desc->written += size;
1632 desc->arg.buf += size;
1633 return size;
1634}
1635
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001636/*
1637 * Performs necessary checks before doing a write
1638 * @iov: io vector request
1639 * @nr_segs: number of segments in the iovec
1640 * @count: number of bytes to write
1641 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1642 *
1643 * Adjust number of segments and amount of bytes to write (nr_segs should be
1644 * properly initialized first). Returns appropriate error code that caller
1645 * should return or zero in case that write should be allowed.
1646 */
1647int generic_segment_checks(const struct iovec *iov,
1648 unsigned long *nr_segs, size_t *count, int access_flags)
1649{
1650 unsigned long seg;
1651 size_t cnt = 0;
1652 for (seg = 0; seg < *nr_segs; seg++) {
1653 const struct iovec *iv = &iov[seg];
1654
1655 /*
1656 * If any segment has a negative length, or the cumulative
1657 * length ever wraps negative then return -EINVAL.
1658 */
1659 cnt += iv->iov_len;
1660 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1661 return -EINVAL;
1662 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1663 continue;
1664 if (seg == 0)
1665 return -EFAULT;
1666 *nr_segs = seg;
1667 cnt -= iv->iov_len; /* This segment is no good */
1668 break;
1669 }
1670 *count = cnt;
1671 return 0;
1672}
1673EXPORT_SYMBOL(generic_segment_checks);
1674
Randy Dunlap485bb992006-06-23 02:03:49 -07001675/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001676 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001677 * @iocb: kernel I/O control block
1678 * @iov: io vector request
1679 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001680 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001681 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 * This is the "read()" routine for all filesystems
1683 * that can use the page cache directly.
1684 */
1685ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001686generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1687 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688{
1689 struct file *filp = iocb->ki_filp;
1690 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001691 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001693 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
1695 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001696 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1697 if (retval)
1698 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
1700 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1701 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001702 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 struct address_space *mapping;
1704 struct inode *inode;
1705
1706 mapping = filp->f_mapping;
1707 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 if (!count)
1709 goto out; /* skip atime */
1710 size = i_size_read(inode);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001711 retval = filemap_write_and_wait_range(mapping, pos,
Nick Piggin48b47c52009-01-06 14:40:22 -08001712 pos + iov_length(iov, nr_segs) - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001713 if (!retval) {
1714 retval = mapping->a_ops->direct_IO(READ, iocb,
1715 iov, pos, nr_segs);
1716 }
1717 if (retval > 0) {
1718 *ppos = pos + retval;
1719 count -= retval;
1720 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001721
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001722 /*
1723 * Btrfs can have a short DIO read if we encounter
1724 * compressed extents, so if there was an error, or if
1725 * we've already read everything we wanted to, or if
1726 * there was a short read because we hit EOF, go ahead
1727 * and return. Otherwise fallthrough to buffered io for
1728 * the rest of the read.
1729 */
1730 if (retval < 0 || !count || *ppos >= size) {
1731 file_accessed(filp);
1732 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001733 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 }
1735
Josef Bacik66f998f2010-05-23 11:00:54 -04001736 count = retval;
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001737 for (seg = 0; seg < nr_segs; seg++) {
1738 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001739 loff_t offset = 0;
1740
1741 /*
1742 * If we did a short DIO read we need to skip the section of the
1743 * iov that we've already read data into.
1744 */
1745 if (count) {
1746 if (count > iov[seg].iov_len) {
1747 count -= iov[seg].iov_len;
1748 continue;
1749 }
1750 offset = count;
1751 count = 0;
1752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001754 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001755 desc.arg.buf = iov[seg].iov_base + offset;
1756 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001757 if (desc.count == 0)
1758 continue;
1759 desc.error = 0;
Kirill A. Shutemovb77d88d2013-11-14 14:30:40 -08001760 do_generic_file_read(filp, ppos, &desc);
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001761 retval += desc.written;
1762 if (desc.error) {
1763 retval = retval ?: desc.error;
1764 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 }
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001766 if (desc.count > 0)
1767 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 }
1769out:
1770 return retval;
1771}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772EXPORT_SYMBOL(generic_file_aio_read);
1773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001775/**
1776 * page_cache_read - adds requested page to the page cache if not already there
1777 * @file: file to read
1778 * @offset: page index
1779 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 * This adds the requested page to the page cache if it isn't already there,
1781 * and schedules an I/O to read in its contents from disk.
1782 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001783static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784{
1785 struct address_space *mapping = file->f_mapping;
1786 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001787 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Zach Brown994fc28c2005-12-15 14:28:17 -08001789 do {
1790 page = page_cache_alloc_cold(mapping);
1791 if (!page)
1792 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
Zach Brown994fc28c2005-12-15 14:28:17 -08001794 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1795 if (ret == 0)
1796 ret = mapping->a_ops->readpage(file, page);
1797 else if (ret == -EEXIST)
1798 ret = 0; /* losing race to add is OK */
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Zach Brown994fc28c2005-12-15 14:28:17 -08001802 } while (ret == AOP_TRUNCATED_PAGE);
1803
1804 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805}
1806
1807#define MMAP_LOTSAMISS (100)
1808
Linus Torvaldsef00e082009-06-16 15:31:25 -07001809/*
1810 * Synchronous readahead happens when we don't even find
1811 * a page in the page cache at all.
1812 */
1813static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1814 struct file_ra_state *ra,
1815 struct file *file,
1816 pgoff_t offset)
1817{
1818 unsigned long ra_pages;
1819 struct address_space *mapping = file->f_mapping;
1820
1821 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001822 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001823 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001824 if (!ra->ra_pages)
1825 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001826
Joe Perches64363aa2013-07-08 16:00:18 -07001827 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001828 page_cache_sync_readahead(mapping, ra, file, offset,
1829 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001830 return;
1831 }
1832
Andi Kleen207d04b2011-05-24 17:12:29 -07001833 /* Avoid banging the cache line if not needed */
1834 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001835 ra->mmap_miss++;
1836
1837 /*
1838 * Do we miss much more than hit in this file? If so,
1839 * stop bothering with read-ahead. It will only hurt.
1840 */
1841 if (ra->mmap_miss > MMAP_LOTSAMISS)
1842 return;
1843
Wu Fengguangd30a1102009-06-16 15:31:30 -07001844 /*
1845 * mmap read-around
1846 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001847 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001848 ra->start = max_t(long, 0, offset - ra_pages / 2);
1849 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001850 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001851 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001852}
1853
1854/*
1855 * Asynchronous readahead happens when we find the page and PG_readahead,
1856 * so we want to possibly extend the readahead further..
1857 */
1858static void do_async_mmap_readahead(struct vm_area_struct *vma,
1859 struct file_ra_state *ra,
1860 struct file *file,
1861 struct page *page,
1862 pgoff_t offset)
1863{
1864 struct address_space *mapping = file->f_mapping;
1865
1866 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001867 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001868 return;
1869 if (ra->mmap_miss > 0)
1870 ra->mmap_miss--;
1871 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001872 page_cache_async_readahead(mapping, ra, file,
1873 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001874}
1875
Randy Dunlap485bb992006-06-23 02:03:49 -07001876/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001877 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001878 * @vma: vma in which the fault was taken
1879 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001880 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001881 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 * mapped memory region to read in file data during a page fault.
1883 *
1884 * The goto's are kind of ugly, but this streamlines the normal case of having
1885 * it in the page cache, and handles the special cases reasonably without
1886 * having a lot of duplicated code.
1887 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001888int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
1890 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001891 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 struct address_space *mapping = file->f_mapping;
1893 struct file_ra_state *ra = &file->f_ra;
1894 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001895 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001897 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001898 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001901 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001902 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 /*
Johannes Weiner494264202013-10-16 13:46:59 -07001905 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001907 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001908 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001910 * We found the page, so try async readahead before
1911 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001913 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001914 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001915 /* No page in the page cache at all */
1916 do_sync_mmap_readahead(vma, ra, file, offset);
1917 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001918 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001919 ret = VM_FAULT_MAJOR;
1920retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001921 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (!page)
1923 goto no_cached_page;
1924 }
1925
Michel Lespinassed88c0922010-11-02 13:05:18 -07001926 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1927 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001928 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001929 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001930
1931 /* Did it get truncated? */
1932 if (unlikely(page->mapping != mapping)) {
1933 unlock_page(page);
1934 put_page(page);
1935 goto retry_find;
1936 }
Sasha Levin309381fea2014-01-23 15:52:54 -08001937 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07001938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001940 * We have a locked page in the page cache, now we need to check
1941 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 */
Nick Piggind00806b2007-07-19 01:46:57 -07001943 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 goto page_not_uptodate;
1945
Linus Torvaldsef00e082009-06-16 15:31:25 -07001946 /*
1947 * Found the page and have a reference on it.
1948 * We must recheck i_size under page lock.
1949 */
Nick Piggind00806b2007-07-19 01:46:57 -07001950 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001951 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001952 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001953 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001954 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001955 }
1956
Nick Piggind0217ac2007-07-19 01:47:03 -07001957 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001958 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960no_cached_page:
1961 /*
1962 * We're only likely to ever get here if MADV_RANDOM is in
1963 * effect.
1964 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001965 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967 /*
1968 * The page we want has now been added to the page cache.
1969 * In the unlikely event that someone removed it in the
1970 * meantime, we'll just come back here and read it again.
1971 */
1972 if (error >= 0)
1973 goto retry_find;
1974
1975 /*
1976 * An error return from page_cache_read can result if the
1977 * system is low on memory, or a problem occurs while trying
1978 * to schedule I/O.
1979 */
1980 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001981 return VM_FAULT_OOM;
1982 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
1984page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 /*
1986 * Umm, take care of errors if the page isn't up-to-date.
1987 * Try to re-read it _once_. We do this synchronously,
1988 * because there really aren't any performance issues here
1989 * and we need to check for errors.
1990 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001992 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001993 if (!error) {
1994 wait_on_page_locked(page);
1995 if (!PageUptodate(page))
1996 error = -EIO;
1997 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001999
2000 if (!error || error == AOP_TRUNCATED_PAGE)
2001 goto retry_find;
2002
2003 /* Things didn't work out. Return zero to tell the mm layer so. */
2004 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002005 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002006}
2007EXPORT_SYMBOL(filemap_fault);
2008
Jan Kara4fcf1c62012-06-12 16:20:29 +02002009int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2010{
2011 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002012 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002013 int ret = VM_FAULT_LOCKED;
2014
Jan Kara14da9202012-06-12 16:20:37 +02002015 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002016 file_update_time(vma->vm_file);
2017 lock_page(page);
2018 if (page->mapping != inode->i_mapping) {
2019 unlock_page(page);
2020 ret = VM_FAULT_NOPAGE;
2021 goto out;
2022 }
Jan Kara14da9202012-06-12 16:20:37 +02002023 /*
2024 * We mark the page dirty already here so that when freeze is in
2025 * progress, we are guaranteed that writeback during freezing will
2026 * see the dirty page and writeprotect it again.
2027 */
2028 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002029 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002030out:
Jan Kara14da9202012-06-12 16:20:37 +02002031 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002032 return ret;
2033}
2034EXPORT_SYMBOL(filemap_page_mkwrite);
2035
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04002036const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002037 .fault = filemap_fault,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002038 .page_mkwrite = filemap_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07002039 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040};
2041
2042/* This is used for a general mmap of a disk file */
2043
2044int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2045{
2046 struct address_space *mapping = file->f_mapping;
2047
2048 if (!mapping->a_ops->readpage)
2049 return -ENOEXEC;
2050 file_accessed(file);
2051 vma->vm_ops = &generic_file_vm_ops;
2052 return 0;
2053}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
2055/*
2056 * This is for filesystems which do not implement ->writepage.
2057 */
2058int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2059{
2060 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2061 return -EINVAL;
2062 return generic_file_mmap(file, vma);
2063}
2064#else
2065int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2066{
2067 return -ENOSYS;
2068}
2069int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2070{
2071 return -ENOSYS;
2072}
2073#endif /* CONFIG_MMU */
2074
2075EXPORT_SYMBOL(generic_file_mmap);
2076EXPORT_SYMBOL(generic_file_readonly_mmap);
2077
Nick Piggin6fe69002007-05-06 14:49:04 -07002078static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002079 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002080 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002081 void *data,
2082 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
Nick Piggineb2be182007-10-16 01:24:57 -07002084 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 int err;
2086repeat:
2087 page = find_get_page(mapping, index);
2088 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002089 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002090 if (!page)
2091 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002092 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002093 if (unlikely(err)) {
2094 page_cache_release(page);
2095 if (err == -EEXIST)
2096 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return ERR_PTR(err);
2099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 err = filler(data, page);
2101 if (err < 0) {
2102 page_cache_release(page);
2103 page = ERR_PTR(err);
2104 }
2105 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return page;
2107}
2108
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002109static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002110 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002111 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002112 void *data,
2113 gfp_t gfp)
2114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115{
2116 struct page *page;
2117 int err;
2118
2119retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002120 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01002122 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 if (PageUptodate(page))
2124 goto out;
2125
2126 lock_page(page);
2127 if (!page->mapping) {
2128 unlock_page(page);
2129 page_cache_release(page);
2130 goto retry;
2131 }
2132 if (PageUptodate(page)) {
2133 unlock_page(page);
2134 goto out;
2135 }
2136 err = filler(data, page);
2137 if (err < 0) {
2138 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01002139 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 }
David Howellsc855ff32007-05-09 13:42:20 +01002141out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002142 mark_page_accessed(page);
2143 return page;
2144}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002145
2146/**
2147 * read_cache_page_async - read into page cache, fill it if needed
2148 * @mapping: the page's address_space
2149 * @index: the page index
2150 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002151 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002152 *
2153 * Same as read_cache_page, but don't wait for page to become unlocked
2154 * after submitting it to the filler.
2155 *
2156 * Read into the page cache. If a page already exists, and PageUptodate() is
2157 * not set, try to fill the page but don't wait for it to become unlocked.
2158 *
2159 * If the page does not get brought uptodate, return -EIO.
2160 */
2161struct page *read_cache_page_async(struct address_space *mapping,
2162 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002163 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002164 void *data)
2165{
2166 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2167}
Nick Piggin6fe69002007-05-06 14:49:04 -07002168EXPORT_SYMBOL(read_cache_page_async);
2169
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002170static struct page *wait_on_page_read(struct page *page)
2171{
2172 if (!IS_ERR(page)) {
2173 wait_on_page_locked(page);
2174 if (!PageUptodate(page)) {
2175 page_cache_release(page);
2176 page = ERR_PTR(-EIO);
2177 }
2178 }
2179 return page;
2180}
2181
2182/**
2183 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2184 * @mapping: the page's address_space
2185 * @index: the page index
2186 * @gfp: the page allocator flags to use if allocating
2187 *
2188 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002189 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002190 *
2191 * If the page does not get brought uptodate, return -EIO.
2192 */
2193struct page *read_cache_page_gfp(struct address_space *mapping,
2194 pgoff_t index,
2195 gfp_t gfp)
2196{
2197 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2198
2199 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
2200}
2201EXPORT_SYMBOL(read_cache_page_gfp);
2202
Nick Piggin6fe69002007-05-06 14:49:04 -07002203/**
2204 * read_cache_page - read into page cache, fill it if needed
2205 * @mapping: the page's address_space
2206 * @index: the page index
2207 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002208 * @data: first arg to filler(data, page) function, often left as NULL
Nick Piggin6fe69002007-05-06 14:49:04 -07002209 *
2210 * Read into the page cache. If a page already exists, and PageUptodate() is
2211 * not set, try to fill the page then wait for it to become unlocked.
2212 *
2213 * If the page does not get brought uptodate, return -EIO.
2214 */
2215struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002216 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002217 int (*filler)(void *, struct page *),
Nick Piggin6fe69002007-05-06 14:49:04 -07002218 void *data)
2219{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002220 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222EXPORT_SYMBOL(read_cache_page);
2223
Nick Piggin2f718ff2007-10-16 01:24:59 -07002224static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 const struct iovec *iov, size_t base, size_t bytes)
2226{
Ingo Molnarf1800532009-03-02 11:00:57 +01002227 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
2229 while (bytes) {
2230 char __user *buf = iov->iov_base + base;
2231 int copy = min(bytes, iov->iov_len - base);
2232
2233 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01002234 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 copied += copy;
2236 bytes -= copy;
2237 vaddr += copy;
2238 iov++;
2239
NeilBrown01408c42006-06-25 05:47:58 -07002240 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 }
2243 return copied - left;
2244}
2245
2246/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07002247 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002248 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07002249 * bytes which were copied.
2250 */
2251size_t iov_iter_copy_from_user_atomic(struct page *page,
2252 struct iov_iter *i, unsigned long offset, size_t bytes)
2253{
2254 char *kaddr;
2255 size_t copied;
2256
2257 BUG_ON(!in_atomic());
Cong Wang9b04c5f2011-11-25 23:14:39 +08002258 kaddr = kmap_atomic(page);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002259 if (likely(i->nr_segs == 1)) {
2260 int left;
2261 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002262 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002263 copied = bytes - left;
2264 } else {
2265 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2266 i->iov, i->iov_offset, bytes);
2267 }
Cong Wang9b04c5f2011-11-25 23:14:39 +08002268 kunmap_atomic(kaddr);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002269
2270 return copied;
2271}
Nick Piggin89e10782007-10-16 01:25:07 -07002272EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002273
2274/*
2275 * This has the same sideeffects and return value as
2276 * iov_iter_copy_from_user_atomic().
2277 * The difference is that it attempts to resolve faults.
2278 * Page must not be locked.
2279 */
2280size_t iov_iter_copy_from_user(struct page *page,
2281 struct iov_iter *i, unsigned long offset, size_t bytes)
2282{
2283 char *kaddr;
2284 size_t copied;
2285
2286 kaddr = kmap(page);
2287 if (likely(i->nr_segs == 1)) {
2288 int left;
2289 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002290 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002291 copied = bytes - left;
2292 } else {
2293 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2294 i->iov, i->iov_offset, bytes);
2295 }
2296 kunmap(page);
2297 return copied;
2298}
Nick Piggin89e10782007-10-16 01:25:07 -07002299EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002300
Nick Pigginf7009262008-03-10 11:43:59 -07002301void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002302{
Nick Pigginf7009262008-03-10 11:43:59 -07002303 BUG_ON(i->count < bytes);
2304
Nick Piggin2f718ff2007-10-16 01:24:59 -07002305 if (likely(i->nr_segs == 1)) {
2306 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07002307 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002308 } else {
2309 const struct iovec *iov = i->iov;
2310 size_t base = i->iov_offset;
Jeff Layton39be79c2011-10-27 23:53:08 +02002311 unsigned long nr_segs = i->nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002312
Nick Piggin124d3b72008-02-02 15:01:17 +01002313 /*
2314 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07002315 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01002316 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07002317 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07002318 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002319
Nick Pigginf7009262008-03-10 11:43:59 -07002320 copy = min(bytes, iov->iov_len - base);
2321 BUG_ON(!i->count || i->count < copy);
2322 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002323 bytes -= copy;
2324 base += copy;
2325 if (iov->iov_len == base) {
2326 iov++;
Jeff Layton39be79c2011-10-27 23:53:08 +02002327 nr_segs--;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002328 base = 0;
2329 }
2330 }
2331 i->iov = iov;
2332 i->iov_offset = base;
Jeff Layton39be79c2011-10-27 23:53:08 +02002333 i->nr_segs = nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002334 }
2335}
Nick Piggin89e10782007-10-16 01:25:07 -07002336EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002337
Nick Pigginafddba42007-10-16 01:25:01 -07002338/*
2339 * Fault in the first iovec of the given iov_iter, to a maximum length
2340 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2341 * accessed (ie. because it is an invalid address).
2342 *
2343 * writev-intensive code may want this to prefault several iovecs -- that
2344 * would be possible (callers must not rely on the fact that _only_ the
2345 * first iovec will be faulted with the current implementation).
2346 */
2347int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002348{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002349 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002350 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2351 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002352}
Nick Piggin89e10782007-10-16 01:25:07 -07002353EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002354
2355/*
2356 * Return the count of just the current iov_iter segment.
2357 */
Maxim Patlasovd28574e2012-10-26 19:50:04 +04002358size_t iov_iter_single_seg_count(const struct iov_iter *i)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002359{
2360 const struct iovec *iov = i->iov;
2361 if (i->nr_segs == 1)
2362 return i->count;
2363 else
2364 return min(i->count, iov->iov_len - i->iov_offset);
2365}
Nick Piggin89e10782007-10-16 01:25:07 -07002366EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002367
2368/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 * Performs necessary checks before doing a write
2370 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002371 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 * Returns appropriate error code that caller should return or
2373 * zero in case that write should be allowed.
2374 */
2375inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2376{
2377 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002378 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
2380 if (unlikely(*pos < 0))
2381 return -EINVAL;
2382
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 if (!isblk) {
2384 /* FIXME: this is for backwards compatibility with 2.4 */
2385 if (file->f_flags & O_APPEND)
2386 *pos = i_size_read(inode);
2387
2388 if (limit != RLIM_INFINITY) {
2389 if (*pos >= limit) {
2390 send_sig(SIGXFSZ, current, 0);
2391 return -EFBIG;
2392 }
2393 if (*count > limit - (typeof(limit))*pos) {
2394 *count = limit - (typeof(limit))*pos;
2395 }
2396 }
2397 }
2398
2399 /*
2400 * LFS rule
2401 */
2402 if (unlikely(*pos + *count > MAX_NON_LFS &&
2403 !(file->f_flags & O_LARGEFILE))) {
2404 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 return -EFBIG;
2406 }
2407 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2408 *count = MAX_NON_LFS - (unsigned long)*pos;
2409 }
2410 }
2411
2412 /*
2413 * Are we about to exceed the fs block limit ?
2414 *
2415 * If we have written data it becomes a short write. If we have
2416 * exceeded without writing data we send a signal and return EFBIG.
2417 * Linus frestrict idea will clean these up nicely..
2418 */
2419 if (likely(!isblk)) {
2420 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2421 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 return -EFBIG;
2423 }
2424 /* zero-length writes at ->s_maxbytes are OK */
2425 }
2426
2427 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2428 *count = inode->i_sb->s_maxbytes - *pos;
2429 } else {
David Howells93614012006-09-30 20:45:40 +02002430#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 loff_t isize;
2432 if (bdev_read_only(I_BDEV(inode)))
2433 return -EPERM;
2434 isize = i_size_read(inode);
2435 if (*pos >= isize) {
2436 if (*count || *pos > isize)
2437 return -ENOSPC;
2438 }
2439
2440 if (*pos + *count > isize)
2441 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002442#else
2443 return -EPERM;
2444#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 }
2446 return 0;
2447}
2448EXPORT_SYMBOL(generic_write_checks);
2449
Nick Pigginafddba42007-10-16 01:25:01 -07002450int pagecache_write_begin(struct file *file, struct address_space *mapping,
2451 loff_t pos, unsigned len, unsigned flags,
2452 struct page **pagep, void **fsdata)
2453{
2454 const struct address_space_operations *aops = mapping->a_ops;
2455
Nick Piggin4e02ed42008-10-29 14:00:55 -07002456 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002457 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002458}
2459EXPORT_SYMBOL(pagecache_write_begin);
2460
2461int pagecache_write_end(struct file *file, struct address_space *mapping,
2462 loff_t pos, unsigned len, unsigned copied,
2463 struct page *page, void *fsdata)
2464{
2465 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002466
Nick Piggin4e02ed42008-10-29 14:00:55 -07002467 mark_page_accessed(page);
2468 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002469}
2470EXPORT_SYMBOL(pagecache_write_end);
2471
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472ssize_t
2473generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2474 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2475 size_t count, size_t ocount)
2476{
2477 struct file *file = iocb->ki_filp;
2478 struct address_space *mapping = file->f_mapping;
2479 struct inode *inode = mapping->host;
2480 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002481 size_t write_len;
2482 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
2484 if (count != ocount)
2485 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2486
Christoph Hellwiga969e902008-07-23 21:27:04 -07002487 write_len = iov_length(iov, *nr_segs);
2488 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002489
Nick Piggin48b47c52009-01-06 14:40:22 -08002490 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002491 if (written)
2492 goto out;
2493
2494 /*
2495 * After a write we want buffered reads to be sure to go to disk to get
2496 * the new data. We invalidate clean cached page from the region we're
2497 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002498 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002499 */
2500 if (mapping->nrpages) {
2501 written = invalidate_inode_pages2_range(mapping,
2502 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002503 /*
2504 * If a page can not be invalidated, return 0 to fall back
2505 * to buffered write.
2506 */
2507 if (written) {
2508 if (written == -EBUSY)
2509 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002510 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002511 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002512 }
2513
2514 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2515
2516 /*
2517 * Finally, try again to invalidate clean pages which might have been
2518 * cached by non-direct readahead, or faulted in by get_user_pages()
2519 * if the source of the write was an mmap'ed region of the file
2520 * we're writing. Either one is a pretty crazy thing to do,
2521 * so we don't support it 100%. If this invalidation
2522 * fails, tough, the write still worked...
2523 */
2524 if (mapping->nrpages) {
2525 invalidate_inode_pages2_range(mapping,
2526 pos >> PAGE_CACHE_SHIFT, end);
2527 }
2528
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002530 pos += written;
2531 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2532 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 mark_inode_dirty(inode);
2534 }
Namhyung Kim01166512010-10-26 14:21:58 -07002535 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002537out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 return written;
2539}
2540EXPORT_SYMBOL(generic_file_direct_write);
2541
Nick Piggineb2be182007-10-16 01:24:57 -07002542/*
2543 * Find or create a page at the given pagecache position. Return the locked
2544 * page. This function is specifically for buffered writes.
2545 */
Nick Piggin54566b22009-01-04 12:00:53 -08002546struct page *grab_cache_page_write_begin(struct address_space *mapping,
2547 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002548{
2549 int status;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002550 gfp_t gfp_mask;
Nick Piggineb2be182007-10-16 01:24:57 -07002551 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002552 gfp_t gfp_notmask = 0;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002553
Fengguang Wu1010bb12012-03-21 16:34:08 -07002554 gfp_mask = mapping_gfp_mask(mapping);
2555 if (mapping_cap_account_dirty(mapping))
2556 gfp_mask |= __GFP_WRITE;
Nick Piggin54566b22009-01-04 12:00:53 -08002557 if (flags & AOP_FLAG_NOFS)
2558 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002559repeat:
2560 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002561 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002562 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002563
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002564 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002565 if (!page)
2566 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002567 status = add_to_page_cache_lru(page, mapping, index,
2568 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002569 if (unlikely(status)) {
2570 page_cache_release(page);
2571 if (status == -EEXIST)
2572 goto repeat;
2573 return NULL;
2574 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002575found:
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002576 wait_for_stable_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002577 return page;
2578}
Nick Piggin54566b22009-01-04 12:00:53 -08002579EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002580
Nick Pigginafddba42007-10-16 01:25:01 -07002581static ssize_t generic_perform_write(struct file *file,
2582 struct iov_iter *i, loff_t pos)
2583{
2584 struct address_space *mapping = file->f_mapping;
2585 const struct address_space_operations *a_ops = mapping->a_ops;
2586 long status = 0;
2587 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002588 unsigned int flags = 0;
2589
2590 /*
2591 * Copies from kernel address space cannot fail (NFSD is a big user).
2592 */
2593 if (segment_eq(get_fs(), KERNEL_DS))
2594 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002595
2596 do {
2597 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002598 unsigned long offset; /* Offset into pagecache page */
2599 unsigned long bytes; /* Bytes to write to page */
2600 size_t copied; /* Bytes copied from user */
2601 void *fsdata;
2602
2603 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002604 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2605 iov_iter_count(i));
2606
2607again:
Nick Pigginafddba42007-10-16 01:25:01 -07002608 /*
2609 * Bring in the user page that we will copy from _first_.
2610 * Otherwise there's a nasty deadlock on copying from the
2611 * same page as we're writing to, without it being marked
2612 * up-to-date.
2613 *
2614 * Not only is this an optimisation, but it is also required
2615 * to check that the address is actually valid, when atomic
2616 * usercopies are used, below.
2617 */
2618 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2619 status = -EFAULT;
2620 break;
2621 }
2622
Nick Piggin674b8922007-10-16 01:25:03 -07002623 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002624 &page, &fsdata);
2625 if (unlikely(status))
2626 break;
2627
anfei zhou931e80e2010-02-02 13:44:02 -08002628 if (mapping_writably_mapped(mapping))
2629 flush_dcache_page(page);
2630
Nick Pigginafddba42007-10-16 01:25:01 -07002631 pagefault_disable();
2632 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2633 pagefault_enable();
2634 flush_dcache_page(page);
2635
Josef Bacikc8236db2009-07-05 12:08:18 -07002636 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002637 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2638 page, fsdata);
2639 if (unlikely(status < 0))
2640 break;
2641 copied = status;
2642
2643 cond_resched();
2644
Nick Piggin124d3b72008-02-02 15:01:17 +01002645 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002646 if (unlikely(copied == 0)) {
2647 /*
2648 * If we were unable to copy any data at all, we must
2649 * fall back to a single segment length write.
2650 *
2651 * If we didn't fallback here, we could livelock
2652 * because not all segments in the iov can be copied at
2653 * once without a pagefault.
2654 */
2655 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2656 iov_iter_single_seg_count(i));
2657 goto again;
2658 }
Nick Pigginafddba42007-10-16 01:25:01 -07002659 pos += copied;
2660 written += copied;
2661
2662 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002663 if (fatal_signal_pending(current)) {
2664 status = -EINTR;
2665 break;
2666 }
Nick Pigginafddba42007-10-16 01:25:01 -07002667 } while (iov_iter_count(i));
2668
2669 return written ? written : status;
2670}
2671
2672ssize_t
2673generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2674 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2675 size_t count, ssize_t written)
2676{
2677 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002678 ssize_t status;
2679 struct iov_iter i;
2680
2681 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002682 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002683
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002685 written += status;
2686 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 }
2688
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 return written ? written : status;
2690}
2691EXPORT_SYMBOL(generic_file_buffered_write);
2692
Jan Karae4dd9de2009-08-17 18:10:06 +02002693/**
2694 * __generic_file_aio_write - write data to a file
2695 * @iocb: IO state structure (file, offset, etc.)
2696 * @iov: vector with data to write
2697 * @nr_segs: number of segments in the vector
2698 * @ppos: position where to write
2699 *
2700 * This function does all the work needed for actually writing data to a
2701 * file. It does all basic checks, removes SUID from the file, updates
2702 * modification times and calls proper subroutines depending on whether we
2703 * do direct IO or a standard buffered write.
2704 *
2705 * It expects i_mutex to be grabbed unless we work on a block device or similar
2706 * object which does not need locking at all.
2707 *
2708 * This function does *not* take care of syncing data in case of O_SYNC write.
2709 * A caller has to handle it. This is mainly due to the fact that we want to
2710 * avoid syncing under i_mutex.
2711 */
2712ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2713 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714{
2715 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002716 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 size_t ocount; /* original count */
2718 size_t count; /* after file limit checks */
2719 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 loff_t pos;
2721 ssize_t written;
2722 ssize_t err;
2723
2724 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002725 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2726 if (err)
2727 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
2729 count = ocount;
2730 pos = *ppos;
2731
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 /* We can write back this queue in page reclaim */
2733 current->backing_dev_info = mapping->backing_dev_info;
2734 written = 0;
2735
2736 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2737 if (err)
2738 goto out;
2739
2740 if (count == 0)
2741 goto out;
2742
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002743 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 if (err)
2745 goto out;
2746
Josef Bacikc3b2da32012-03-26 09:59:21 -04002747 err = file_update_time(file);
2748 if (err)
2749 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
2751 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2752 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002753 loff_t endbyte;
2754 ssize_t written_buffered;
2755
2756 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2757 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 if (written < 0 || written == count)
2759 goto out;
2760 /*
2761 * direct-io write to a hole: fall through to buffered I/O
2762 * for completing the rest of the request.
2763 */
2764 pos += written;
2765 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002766 written_buffered = generic_file_buffered_write(iocb, iov,
2767 nr_segs, pos, ppos, count,
2768 written);
2769 /*
2770 * If generic_file_buffered_write() retuned a synchronous error
2771 * then we want to return the number of bytes which were
2772 * direct-written, or the error code if that was zero. Note
2773 * that this differs from normal direct-io semantics, which
2774 * will return -EFOO even if some bytes were written.
2775 */
2776 if (written_buffered < 0) {
2777 err = written_buffered;
2778 goto out;
2779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002781 /*
2782 * We need to ensure that the page cache pages are written to
2783 * disk and invalidated to preserve the expected O_DIRECT
2784 * semantics.
2785 */
2786 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002787 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002788 if (err == 0) {
2789 written = written_buffered;
2790 invalidate_mapping_pages(mapping,
2791 pos >> PAGE_CACHE_SHIFT,
2792 endbyte >> PAGE_CACHE_SHIFT);
2793 } else {
2794 /*
2795 * We don't know how much we wrote, so just return
2796 * the number of bytes which were direct-written
2797 */
2798 }
2799 } else {
2800 written = generic_file_buffered_write(iocb, iov, nr_segs,
2801 pos, ppos, count, written);
2802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803out:
2804 current->backing_dev_info = NULL;
2805 return written ? written : err;
2806}
Jan Karae4dd9de2009-08-17 18:10:06 +02002807EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
Jan Karae4dd9de2009-08-17 18:10:06 +02002809/**
2810 * generic_file_aio_write - write data to a file
2811 * @iocb: IO state structure
2812 * @iov: vector with data to write
2813 * @nr_segs: number of segments in the vector
2814 * @pos: position in file where to write
2815 *
2816 * This is a wrapper around __generic_file_aio_write() to be used by most
2817 * filesystems. It takes care of syncing the file in case of O_SYNC file
2818 * and acquires i_mutex as needed.
2819 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002820ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2821 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822{
2823 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002824 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
2827 BUG_ON(iocb->ki_pos != pos);
2828
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002829 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002830 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002831 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832
Christoph Hellwig02afc272013-09-04 15:04:40 +02002833 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 ssize_t err;
2835
Al Virod311d792014-02-09 15:18:09 -05002836 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2837 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 ret = err;
2839 }
2840 return ret;
2841}
2842EXPORT_SYMBOL(generic_file_aio_write);
2843
David Howellscf9a2ae2006-08-29 19:05:54 +01002844/**
2845 * try_to_release_page() - release old fs-specific metadata on a page
2846 *
2847 * @page: the page which the kernel is trying to free
2848 * @gfp_mask: memory allocation flags (and I/O mode)
2849 *
2850 * The address_space is to try to release any data against the page
2851 * (presumably at page->private). If the release was successful, return `1'.
2852 * Otherwise return zero.
2853 *
David Howells266cf652009-04-03 16:42:36 +01002854 * This may also be called if PG_fscache is set on a page, indicating that the
2855 * page is known to the local caching routines.
2856 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002857 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002858 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002859 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002860 */
2861int try_to_release_page(struct page *page, gfp_t gfp_mask)
2862{
2863 struct address_space * const mapping = page->mapping;
2864
2865 BUG_ON(!PageLocked(page));
2866 if (PageWriteback(page))
2867 return 0;
2868
2869 if (mapping && mapping->a_ops->releasepage)
2870 return mapping->a_ops->releasepage(page, gfp_mask);
2871 return try_to_free_buffers(page);
2872}
2873
2874EXPORT_SYMBOL(try_to_release_page);