blob: 126d3973b3d1f3be7a3f5e4f42630725c796f31f [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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
13#include <linux/slab.h>
14#include <linux/compiler.h>
15#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kernel_stat.h>
20#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>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036#include <linux/mm_inline.h> /* for page_is_file_cache() */
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
42#include <linux/buffer_head.h> /* for generic_osync_inode */
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
Adrian Bunk5ce78522005-09-10 00:26:28 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/*
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
49 * though.
50 *
51 * Shared mappings now work. 15.8.1995 Bruno.
52 *
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 *
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
57 */
58
59/*
60 * Lock ordering:
61 *
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070064 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080067 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * ->i_mmap_lock (truncate->unmap_mapping_range)
69 *
70 * ->mmap_sem
71 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070072 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
74 *
75 * ->mmap_sem
76 * ->lock_page (access_process_vm)
77 *
Nick Piggin82591e62006-10-19 23:29:10 -070078 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080081 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * ->i_alloc_sem (various)
83 *
84 * ->inode_lock
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
87 *
88 * ->i_mmap_lock
89 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080099 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 *
106 * ->task->proc_lock
107 * ->dcache_lock (proc_pid_lookup)
108 */
109
110/*
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700113 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
115void __remove_from_page_cache(struct page *page)
116{
117 struct address_space *mapping = page->mapping;
118
119 radix_tree_delete(&mapping->page_tree, page->index);
120 page->mapping = NULL;
121 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700122 __dec_zone_page_state(page, NR_FILE_PAGES);
Nick Piggin45426812007-07-15 23:38:12 -0700123 BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -0700124 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds3a692792007-12-19 14:05:13 -0800125
126 /*
127 * Some filesystems seem to re-dirty the page even after
128 * the VM has canceled the dirty bit (eg ext3 journaling).
129 *
130 * Fix it up by doing a final dirty accounting check after
131 * having removed the page entirely.
132 */
133 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
134 dec_zone_page_state(page, NR_FILE_DIRTY);
135 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
139void remove_from_page_cache(struct page *page)
140{
141 struct address_space *mapping = page->mapping;
142
Matt Mackallcd7619d2005-05-01 08:59:01 -0700143 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Nick Piggin19fd6232008-07-25 19:45:32 -0700145 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700147 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
150static int sync_page(void *word)
151{
152 struct address_space *mapping;
153 struct page *page;
154
Andi Kleen07808b72005-11-05 17:25:53 +0100155 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700158 * page_mapping() is being called without PG_locked held.
159 * Some knowledge of the state and use of the page is used to
160 * reduce the requirements down to a memory barrier.
161 * The danger here is of a stale page_mapping() return value
162 * indicating a struct address_space different from the one it's
163 * associated with when it is associated with one.
164 * After smp_mb(), it's either the correct page_mapping() for
165 * the page, or an old page_mapping() and the page's own
166 * page_mapping() has gone NULL.
167 * The ->sync_page() address_space operation must tolerate
168 * page_mapping() going NULL. By an amazing coincidence,
169 * this comes about because none of the users of the page
170 * in the ->sync_page() methods make essential use of the
171 * page_mapping(), merely passing the page down to the backing
172 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700173 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700174 * of interest. When page_mapping() does go NULL, the entire
175 * call stack gracefully ignores the page and returns.
176 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 */
178 smp_mb();
179 mapping = page_mapping(page);
180 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
181 mapping->a_ops->sync_page(page);
182 io_schedule();
183 return 0;
184}
185
Matthew Wilcox2687a352007-12-06 11:18:49 -0500186static int sync_page_killable(void *word)
187{
188 sync_page(word);
189 return fatal_signal_pending(current) ? -EINTR : 0;
190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700194 * @mapping: address space structure to write
195 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800196 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700197 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700199 * Start writeback against all of a mapping's dirty pages that lie
200 * within the byte offsets <start, end> inclusive.
201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700203 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * these two operations is that if a dirty page/buffer is encountered, it must
205 * be waited upon, and not just skipped over.
206 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800207int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
208 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 int ret;
211 struct writeback_control wbc = {
212 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800213 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700214 .range_start = start,
215 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 };
217
218 if (!mapping_cap_writeback_dirty(mapping))
219 return 0;
220
221 ret = do_writepages(mapping, &wbc);
222 return ret;
223}
224
225static inline int __filemap_fdatawrite(struct address_space *mapping,
226 int sync_mode)
227{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229}
230
231int filemap_fdatawrite(struct address_space *mapping)
232{
233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
234}
235EXPORT_SYMBOL(filemap_fdatawrite);
236
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400237int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800238 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
241}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400242EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Randy Dunlap485bb992006-06-23 02:03:49 -0700244/**
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
247 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
250 */
251int filemap_flush(struct address_space *mapping)
252{
253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
254}
255EXPORT_SYMBOL(filemap_flush);
256
Randy Dunlap485bb992006-06-23 02:03:49 -0700257/**
258 * wait_on_page_writeback_range - wait for writeback to complete
259 * @mapping: target address_space
260 * @start: beginning page index
261 * @end: ending page index
262 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 * Wait for writeback to complete against pages indexed by start->end
264 * inclusive
265 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800266int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 pgoff_t start, pgoff_t end)
268{
269 struct pagevec pvec;
270 int nr_pages;
271 int ret = 0;
272 pgoff_t index;
273
274 if (end < start)
275 return 0;
276
277 pagevec_init(&pvec, 0);
278 index = start;
279 while ((index <= end) &&
280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
281 PAGECACHE_TAG_WRITEBACK,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
283 unsigned i;
284
285 for (i = 0; i < nr_pages; i++) {
286 struct page *page = pvec.pages[i];
287
288 /* until radix tree lookup accepts end_index */
289 if (page->index > end)
290 continue;
291
292 wait_on_page_writeback(page);
293 if (PageError(page))
294 ret = -EIO;
295 }
296 pagevec_release(&pvec);
297 cond_resched();
298 }
299
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
302 ret = -ENOSPC;
303 if (test_and_clear_bit(AS_EIO, &mapping->flags))
304 ret = -EIO;
305
306 return ret;
307}
308
Randy Dunlap485bb992006-06-23 02:03:49 -0700309/**
310 * sync_page_range - write and wait on all pages in the passed range
311 * @inode: target inode
312 * @mapping: target address_space
313 * @pos: beginning offset in pages to write
314 * @count: number of bytes to write
315 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * Write and wait upon all the pages in the passed range. This is a "data
317 * integrity" operation. It waits upon in-flight writeout before starting and
318 * waiting upon new writeout. If there was an IO error, return it.
319 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800320 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * it is otherwise livelockable.
322 */
323int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800324 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
327 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
328 int ret;
329
330 if (!mapping_cap_writeback_dirty(mapping) || !count)
331 return 0;
332 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
333 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800334 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800336 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
338 if (ret == 0)
339 ret = wait_on_page_writeback_range(mapping, start, end);
340 return ret;
341}
342EXPORT_SYMBOL(sync_page_range);
343
Randy Dunlap485bb992006-06-23 02:03:49 -0700344/**
Randy Dunlap76824862008-03-19 17:00:40 -0700345 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
Randy Dunlap485bb992006-06-23 02:03:49 -0700346 * @inode: target inode
347 * @mapping: target address_space
348 * @pos: beginning offset in pages to write
349 * @count: number of bytes to write
350 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800351 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 * as it forces O_SYNC writers to different parts of the same file
353 * to be serialised right until io completion.
354 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800355int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
356 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
358 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
359 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
360 int ret;
361
362 if (!mapping_cap_writeback_dirty(mapping) || !count)
363 return 0;
364 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
365 if (ret == 0)
366 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
367 if (ret == 0)
368 ret = wait_on_page_writeback_range(mapping, start, end);
369 return ret;
370}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800371EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700374 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700376 *
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
380int filemap_fdatawait(struct address_space *mapping)
381{
382 loff_t i_size = i_size_read(mapping->host);
383
384 if (i_size == 0)
385 return 0;
386
387 return wait_on_page_writeback_range(mapping, 0,
388 (i_size - 1) >> PAGE_CACHE_SHIFT);
389}
390EXPORT_SYMBOL(filemap_fdatawait);
391
392int filemap_write_and_wait(struct address_space *mapping)
393{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800394 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800397 err = filemap_fdatawrite(mapping);
398 /*
399 * Even if the above returned error, the pages may be
400 * written partially (e.g. -ENOSPC), so we wait for it.
401 * But the -EIO is special case, it may indicate the worst
402 * thing (e.g. bug) happened, so we avoid waiting for it.
403 */
404 if (err != -EIO) {
405 int err2 = filemap_fdatawait(mapping);
406 if (!err)
407 err = err2;
408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800410 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800412EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Randy Dunlap485bb992006-06-23 02:03:49 -0700414/**
415 * filemap_write_and_wait_range - write out & wait on a file range
416 * @mapping: the address_space for the pages
417 * @lstart: offset in bytes where the range starts
418 * @lend: offset in bytes where the range ends (inclusive)
419 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800420 * Write out and wait upon file offsets lstart->lend, inclusive.
421 *
422 * Note that `lend' is inclusive (describes the last byte to be written) so
423 * that this function can be used to write to the very end-of-file (end = -1).
424 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425int filemap_write_and_wait_range(struct address_space *mapping,
426 loff_t lstart, loff_t lend)
427{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800428 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800431 err = __filemap_fdatawrite_range(mapping, lstart, lend,
432 WB_SYNC_ALL);
433 /* See comment of filemap_write_and_wait() */
434 if (err != -EIO) {
435 int err2 = wait_on_page_writeback_range(mapping,
436 lstart >> PAGE_CACHE_SHIFT,
437 lend >> PAGE_CACHE_SHIFT);
438 if (!err)
439 err = err2;
440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800442 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
Randy Dunlap485bb992006-06-23 02:03:49 -0700445/**
Nick Piggine2867812008-07-25 19:45:30 -0700446 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700447 * @page: page to add
448 * @mapping: the page's address_space
449 * @offset: page index
450 * @gfp_mask: page allocation mode
451 *
Nick Piggine2867812008-07-25 19:45:30 -0700452 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * This function does not add the page to the LRU. The caller must do that.
454 */
Nick Piggine2867812008-07-25 19:45:30 -0700455int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400456 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Nick Piggine2867812008-07-25 19:45:30 -0700458 int error;
459
460 VM_BUG_ON(!PageLocked(page));
461
462 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800463 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800464 if (error)
465 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Balbir Singh35c754d2008-02-07 00:14:05 -0800467 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700469 page_cache_get(page);
470 page->mapping = mapping;
471 page->index = offset;
472
Nick Piggin19fd6232008-07-25 19:45:32 -0700473 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700475 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700477 __inc_zone_page_state(page, NR_FILE_PAGES);
Nick Piggine2867812008-07-25 19:45:30 -0700478 } else {
479 page->mapping = NULL;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700480 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700481 page_cache_release(page);
482 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800483
Nick Piggin19fd6232008-07-25 19:45:32 -0700484 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800486 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700487 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800488out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 return error;
490}
Nick Piggine2867812008-07-25 19:45:30 -0700491EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400494 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700496 int ret;
497
498 /*
499 * Splice_read and readahead add shmem/tmpfs pages into the page cache
500 * before shmem_readpage has a chance to mark them as SwapBacked: they
501 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
502 * (called in add_to_page_cache) needs to know where they're going too.
503 */
504 if (mapping_cap_swap_backed(mapping))
505 SetPageSwapBacked(page);
506
507 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
508 if (ret == 0) {
509 if (page_is_file_cache(page))
510 lru_cache_add_file(page);
511 else
512 lru_cache_add_active_anon(page);
513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return ret;
515}
516
Paul Jackson44110fe2006-03-24 03:16:04 -0800517#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700518struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800519{
520 if (cpuset_do_page_mem_spread()) {
521 int n = cpuset_mem_spread_node();
Nick Piggin2ae88142006-10-28 10:38:23 -0700522 return alloc_pages_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800523 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700524 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800525}
Nick Piggin2ae88142006-10-28 10:38:23 -0700526EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800527#endif
528
Nick Piggindb376482006-09-25 23:31:24 -0700529static int __sleep_on_page_lock(void *word)
530{
531 io_schedule();
532 return 0;
533}
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535/*
536 * In order to wait for pages to become available there must be
537 * waitqueues associated with pages. By using a hash table of
538 * waitqueues where the bucket discipline is to maintain all
539 * waiters on the same queue and wake all when any of the pages
540 * become available, and for the woken contexts to check to be
541 * sure the appropriate page became available, this saves space
542 * at a cost of "thundering herd" phenomena during rare hash
543 * collisions.
544 */
545static wait_queue_head_t *page_waitqueue(struct page *page)
546{
547 const struct zone *zone = page_zone(page);
548
549 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
550}
551
552static inline void wake_up_page(struct page *page, int bit)
553{
554 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
555}
556
Harvey Harrison920c7a52008-02-04 22:29:26 -0800557void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
559 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
560
561 if (test_bit(bit_nr, &page->flags))
562 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
563 TASK_UNINTERRUPTIBLE);
564}
565EXPORT_SYMBOL(wait_on_page_bit);
566
567/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700568 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 * @page: the page
570 *
571 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
572 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
573 * mechananism between PageLocked pages and PageWriteback pages is shared.
574 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
575 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700576 * The mb is necessary to enforce ordering between the clear_bit and the read
577 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800579void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Nick Piggin8413ac92008-10-18 20:26:59 -0700581 VM_BUG_ON(!PageLocked(page));
582 clear_bit_unlock(PG_locked, &page->flags);
583 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 wake_up_page(page, PG_locked);
585}
586EXPORT_SYMBOL(unlock_page);
587
Randy Dunlap485bb992006-06-23 02:03:49 -0700588/**
589 * end_page_writeback - end writeback against a page
590 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 */
592void end_page_writeback(struct page *page)
593{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700594 if (TestClearPageReclaim(page))
595 rotate_reclaimable_page(page);
596
597 if (!test_clear_page_writeback(page))
598 BUG();
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 smp_mb__after_clear_bit();
601 wake_up_page(page, PG_writeback);
602}
603EXPORT_SYMBOL(end_page_writeback);
604
Randy Dunlap485bb992006-06-23 02:03:49 -0700605/**
606 * __lock_page - get a lock on the page, assuming we need to sleep to get it
607 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700609 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
611 * chances are that on the second loop, the block layer's plug list is empty,
612 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
613 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800614void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
616 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
617
618 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
619 TASK_UNINTERRUPTIBLE);
620}
621EXPORT_SYMBOL(__lock_page);
622
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800623int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500624{
625 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
626
627 return __wait_on_bit_lock(page_waitqueue(page), &wait,
628 sync_page_killable, TASK_KILLABLE);
629}
630
Randy Dunlap76824862008-03-19 17:00:40 -0700631/**
632 * __lock_page_nosync - get a lock on the page, without calling sync_page()
633 * @page: the page to lock
634 *
Nick Piggindb376482006-09-25 23:31:24 -0700635 * Variant of lock_page that does not require the caller to hold a reference
636 * on the page's mapping.
637 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800638void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700639{
640 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
641 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
642 TASK_UNINTERRUPTIBLE);
643}
644
Randy Dunlap485bb992006-06-23 02:03:49 -0700645/**
646 * find_get_page - find and get a page reference
647 * @mapping: the address_space to search
648 * @offset: the page index
649 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700650 * Is there a pagecache struct page at the given (mapping, offset) tuple?
651 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 */
Nick Piggina60637c2008-07-25 19:45:31 -0700653struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
Nick Piggina60637c2008-07-25 19:45:31 -0700655 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 struct page *page;
657
Nick Piggina60637c2008-07-25 19:45:31 -0700658 rcu_read_lock();
659repeat:
660 page = NULL;
661 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
662 if (pagep) {
663 page = radix_tree_deref_slot(pagep);
664 if (unlikely(!page || page == RADIX_TREE_RETRY))
665 goto repeat;
666
667 if (!page_cache_get_speculative(page))
668 goto repeat;
669
670 /*
671 * Has the page moved?
672 * This is part of the lockless pagecache protocol. See
673 * include/linux/pagemap.h for details.
674 */
675 if (unlikely(page != *pagep)) {
676 page_cache_release(page);
677 goto repeat;
678 }
679 }
680 rcu_read_unlock();
681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 return page;
683}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684EXPORT_SYMBOL(find_get_page);
685
Randy Dunlap485bb992006-06-23 02:03:49 -0700686/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700688 * @mapping: the address_space to search
689 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 *
691 * Locates the desired pagecache page, locks it, increments its reference
692 * count and returns its address.
693 *
694 * Returns zero if the page was not present. find_lock_page() may sleep.
695 */
Nick Piggina60637c2008-07-25 19:45:31 -0700696struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
698 struct page *page;
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700701 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700703 lock_page(page);
704 /* Has the page been truncated? */
705 if (unlikely(page->mapping != mapping)) {
706 unlock_page(page);
707 page_cache_release(page);
708 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
Nick Piggina60637c2008-07-25 19:45:31 -0700710 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 return page;
713}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714EXPORT_SYMBOL(find_lock_page);
715
716/**
717 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700718 * @mapping: the page's address_space
719 * @index: the page's index into the mapping
720 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 *
722 * Locates a page in the pagecache. If the page is not present, a new page
723 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
724 * LRU list. The returned page is locked and has its reference count
725 * incremented.
726 *
727 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
728 * allocation!
729 *
730 * find_or_create_page() returns the desired page's address, or zero on
731 * memory exhaustion.
732 */
733struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700734 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Nick Piggineb2be182007-10-16 01:24:57 -0700736 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 int err;
738repeat:
739 page = find_lock_page(mapping, index);
740 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700741 page = __page_cache_alloc(gfp_mask);
742 if (!page)
743 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800744 /*
745 * We want a regular kernel memory (not highmem or DMA etc)
746 * allocation for the radix tree nodes, but we need to honour
747 * the context-specific requirements the caller has asked for.
748 * GFP_RECLAIM_MASK collects those requirements.
749 */
750 err = add_to_page_cache_lru(page, mapping, index,
751 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700752 if (unlikely(err)) {
753 page_cache_release(page);
754 page = NULL;
755 if (err == -EEXIST)
756 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return page;
760}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761EXPORT_SYMBOL(find_or_create_page);
762
763/**
764 * find_get_pages - gang pagecache lookup
765 * @mapping: The address_space to search
766 * @start: The starting page index
767 * @nr_pages: The maximum number of pages
768 * @pages: Where the resulting pages are placed
769 *
770 * find_get_pages() will search for and return a group of up to
771 * @nr_pages pages in the mapping. The pages are placed at @pages.
772 * find_get_pages() takes a reference against the returned pages.
773 *
774 * The search returns a group of mapping-contiguous pages with ascending
775 * indexes. There may be holes in the indices due to not-present pages.
776 *
777 * find_get_pages() returns the number of pages which were found.
778 */
779unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
780 unsigned int nr_pages, struct page **pages)
781{
782 unsigned int i;
783 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700784 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Nick Piggina60637c2008-07-25 19:45:31 -0700786 rcu_read_lock();
787restart:
788 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
789 (void ***)pages, start, nr_pages);
790 ret = 0;
791 for (i = 0; i < nr_found; i++) {
792 struct page *page;
793repeat:
794 page = radix_tree_deref_slot((void **)pages[i]);
795 if (unlikely(!page))
796 continue;
797 /*
798 * this can only trigger if nr_found == 1, making livelock
799 * a non issue.
800 */
801 if (unlikely(page == RADIX_TREE_RETRY))
802 goto restart;
803
804 if (!page_cache_get_speculative(page))
805 goto repeat;
806
807 /* Has the page moved? */
808 if (unlikely(page != *((void **)pages[i]))) {
809 page_cache_release(page);
810 goto repeat;
811 }
812
813 pages[ret] = page;
814 ret++;
815 }
816 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 return ret;
818}
819
Jens Axboeebf43502006-04-27 08:46:01 +0200820/**
821 * find_get_pages_contig - gang contiguous pagecache lookup
822 * @mapping: The address_space to search
823 * @index: The starting page index
824 * @nr_pages: The maximum number of pages
825 * @pages: Where the resulting pages are placed
826 *
827 * find_get_pages_contig() works exactly like find_get_pages(), except
828 * that the returned number of pages are guaranteed to be contiguous.
829 *
830 * find_get_pages_contig() returns the number of pages which were found.
831 */
832unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
833 unsigned int nr_pages, struct page **pages)
834{
835 unsigned int i;
836 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700837 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200838
Nick Piggina60637c2008-07-25 19:45:31 -0700839 rcu_read_lock();
840restart:
841 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
842 (void ***)pages, index, nr_pages);
843 ret = 0;
844 for (i = 0; i < nr_found; i++) {
845 struct page *page;
846repeat:
847 page = radix_tree_deref_slot((void **)pages[i]);
848 if (unlikely(!page))
849 continue;
850 /*
851 * this can only trigger if nr_found == 1, making livelock
852 * a non issue.
853 */
854 if (unlikely(page == RADIX_TREE_RETRY))
855 goto restart;
856
857 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200858 break;
859
Nick Piggina60637c2008-07-25 19:45:31 -0700860 if (!page_cache_get_speculative(page))
861 goto repeat;
862
863 /* Has the page moved? */
864 if (unlikely(page != *((void **)pages[i]))) {
865 page_cache_release(page);
866 goto repeat;
867 }
868
869 pages[ret] = page;
870 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200871 index++;
872 }
Nick Piggina60637c2008-07-25 19:45:31 -0700873 rcu_read_unlock();
874 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200875}
David Howellsef71c152007-05-09 02:33:44 -0700876EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200877
Randy Dunlap485bb992006-06-23 02:03:49 -0700878/**
879 * find_get_pages_tag - find and return pages that match @tag
880 * @mapping: the address_space to search
881 * @index: the starting page index
882 * @tag: the tag index
883 * @nr_pages: the maximum number of pages
884 * @pages: where the resulting pages are placed
885 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700887 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 */
889unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
890 int tag, unsigned int nr_pages, struct page **pages)
891{
892 unsigned int i;
893 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700894 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
Nick Piggina60637c2008-07-25 19:45:31 -0700896 rcu_read_lock();
897restart:
898 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
899 (void ***)pages, *index, nr_pages, tag);
900 ret = 0;
901 for (i = 0; i < nr_found; i++) {
902 struct page *page;
903repeat:
904 page = radix_tree_deref_slot((void **)pages[i]);
905 if (unlikely(!page))
906 continue;
907 /*
908 * this can only trigger if nr_found == 1, making livelock
909 * a non issue.
910 */
911 if (unlikely(page == RADIX_TREE_RETRY))
912 goto restart;
913
914 if (!page_cache_get_speculative(page))
915 goto repeat;
916
917 /* Has the page moved? */
918 if (unlikely(page != *((void **)pages[i]))) {
919 page_cache_release(page);
920 goto repeat;
921 }
922
923 pages[ret] = page;
924 ret++;
925 }
926 rcu_read_unlock();
927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (ret)
929 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700930
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 return ret;
932}
David Howellsef71c152007-05-09 02:33:44 -0700933EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Randy Dunlap485bb992006-06-23 02:03:49 -0700935/**
936 * grab_cache_page_nowait - returns locked page at given index in given cache
937 * @mapping: target address_space
938 * @index: the page index
939 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800940 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 * This is intended for speculative data generators, where the data can
942 * be regenerated if the page couldn't be grabbed. This routine should
943 * be safe to call while holding the lock for another page.
944 *
945 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
946 * and deadlock against the caller's locked page.
947 */
948struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700949grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950{
951 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
953 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200954 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 return page;
956 page_cache_release(page);
957 return NULL;
958 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700959 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800960 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 page_cache_release(page);
962 page = NULL;
963 }
964 return page;
965}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966EXPORT_SYMBOL(grab_cache_page_nowait);
967
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700968/*
969 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
970 * a _large_ part of the i/o request. Imagine the worst scenario:
971 *
972 * ---R__________________________________________B__________
973 * ^ reading here ^ bad block(assume 4k)
974 *
975 * read(R) => miss => readahead(R...B) => media error => frustrating retries
976 * => failing the whole request => read(R) => read(R+1) =>
977 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
978 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
979 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
980 *
981 * It is going insane. Fix it by quickly scaling down the readahead size.
982 */
983static void shrink_readahead_size_eio(struct file *filp,
984 struct file_ra_state *ra)
985{
986 if (!ra->ra_pages)
987 return;
988
989 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700990}
991
Randy Dunlap485bb992006-06-23 02:03:49 -0700992/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800993 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700994 * @filp: the file to read
995 * @ppos: current file position
996 * @desc: read_descriptor
997 * @actor: read method
998 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001000 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 *
1002 * This is really ugly. But the goto's actually try to clarify some
1003 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001005static void do_generic_file_read(struct file *filp, loff_t *ppos,
1006 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001008 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001010 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001011 pgoff_t index;
1012 pgoff_t last_index;
1013 pgoff_t prev_index;
1014 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001015 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001019 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1020 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1022 offset = *ppos & ~PAGE_CACHE_MASK;
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 for (;;) {
1025 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001026 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001027 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 unsigned long nr, ret;
1029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031find_page:
1032 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001033 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001034 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001035 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001036 index, last_index - index);
1037 page = find_get_page(mapping, index);
1038 if (unlikely(page == NULL))
1039 goto no_cached_page;
1040 }
1041 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001042 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001043 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001044 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001046 if (!PageUptodate(page)) {
1047 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1048 !mapping->a_ops->is_partially_uptodate)
1049 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001050 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001051 goto page_not_up_to_date;
1052 if (!mapping->a_ops->is_partially_uptodate(page,
1053 desc, offset))
1054 goto page_not_up_to_date_locked;
1055 unlock_page(page);
1056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001058 /*
1059 * i_size must be checked after we know the page is Uptodate.
1060 *
1061 * Checking i_size after the check allows us to calculate
1062 * the correct value for "nr", which means the zero-filled
1063 * part of the page is not copied back to userspace (unless
1064 * another truncate extends the file - this is desired though).
1065 */
1066
1067 isize = i_size_read(inode);
1068 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1069 if (unlikely(!isize || index > end_index)) {
1070 page_cache_release(page);
1071 goto out;
1072 }
1073
1074 /* nr is the maximum number of bytes to copy from this page */
1075 nr = PAGE_CACHE_SIZE;
1076 if (index == end_index) {
1077 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1078 if (nr <= offset) {
1079 page_cache_release(page);
1080 goto out;
1081 }
1082 }
1083 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
1085 /* If users can be writing to this page using arbitrary
1086 * virtual addresses, take care about potential aliasing
1087 * before reading the page on the kernel side.
1088 */
1089 if (mapping_writably_mapped(mapping))
1090 flush_dcache_page(page);
1091
1092 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001093 * When a sequential read accesses a page several times,
1094 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 */
Jan Karaec0f1632007-05-06 14:49:25 -07001096 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 mark_page_accessed(page);
1098 prev_index = index;
1099
1100 /*
1101 * Ok, we have the page, and it's up-to-date, so
1102 * now we can copy it to user space...
1103 *
1104 * The actor routine returns how many bytes were actually used..
1105 * NOTE! This may not be the same as how much of a user buffer
1106 * we filled up (we may be padding etc), so we can only update
1107 * "pos" here (the actor routine has to update the user buffer
1108 * pointers and the remaining count).
1109 */
1110 ret = actor(desc, page, offset, nr);
1111 offset += ret;
1112 index += offset >> PAGE_CACHE_SHIFT;
1113 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001114 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
1116 page_cache_release(page);
1117 if (ret == nr && desc->count)
1118 continue;
1119 goto out;
1120
1121page_not_up_to_date:
1122 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001123 error = lock_page_killable(page);
1124 if (unlikely(error))
1125 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001127page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001128 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 if (!page->mapping) {
1130 unlock_page(page);
1131 page_cache_release(page);
1132 continue;
1133 }
1134
1135 /* Did somebody else fill it already? */
1136 if (PageUptodate(page)) {
1137 unlock_page(page);
1138 goto page_ok;
1139 }
1140
1141readpage:
1142 /* Start the actual read. The read will unlock the page. */
1143 error = mapping->a_ops->readpage(filp, page);
1144
Zach Brown994fc28c2005-12-15 14:28:17 -08001145 if (unlikely(error)) {
1146 if (error == AOP_TRUNCATED_PAGE) {
1147 page_cache_release(page);
1148 goto find_page;
1149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
1153 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001154 error = lock_page_killable(page);
1155 if (unlikely(error))
1156 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (!PageUptodate(page)) {
1158 if (page->mapping == NULL) {
1159 /*
1160 * invalidate_inode_pages got it
1161 */
1162 unlock_page(page);
1163 page_cache_release(page);
1164 goto find_page;
1165 }
1166 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001167 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001168 error = -EIO;
1169 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 }
1171 unlock_page(page);
1172 }
1173
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 goto page_ok;
1175
1176readpage_error:
1177 /* UHHUH! A synchronous read error occurred. Report it */
1178 desc->error = error;
1179 page_cache_release(page);
1180 goto out;
1181
1182no_cached_page:
1183 /*
1184 * Ok, it wasn't cached, so we need to create a new
1185 * page..
1186 */
Nick Piggineb2be182007-10-16 01:24:57 -07001187 page = page_cache_alloc_cold(mapping);
1188 if (!page) {
1189 desc->error = -ENOMEM;
1190 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 }
Nick Piggineb2be182007-10-16 01:24:57 -07001192 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 index, GFP_KERNEL);
1194 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001195 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 if (error == -EEXIST)
1197 goto find_page;
1198 desc->error = error;
1199 goto out;
1200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 goto readpage;
1202 }
1203
1204out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001205 ra->prev_pos = prev_index;
1206 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1207 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001209 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001210 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213int file_read_actor(read_descriptor_t *desc, struct page *page,
1214 unsigned long offset, unsigned long size)
1215{
1216 char *kaddr;
1217 unsigned long left, count = desc->count;
1218
1219 if (size > count)
1220 size = count;
1221
1222 /*
1223 * Faults on the destination of a read are common, so do it before
1224 * taking the kmap.
1225 */
1226 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1227 kaddr = kmap_atomic(page, KM_USER0);
1228 left = __copy_to_user_inatomic(desc->arg.buf,
1229 kaddr + offset, size);
1230 kunmap_atomic(kaddr, KM_USER0);
1231 if (left == 0)
1232 goto success;
1233 }
1234
1235 /* Do it the slow way */
1236 kaddr = kmap(page);
1237 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1238 kunmap(page);
1239
1240 if (left) {
1241 size -= left;
1242 desc->error = -EFAULT;
1243 }
1244success:
1245 desc->count = count - size;
1246 desc->written += size;
1247 desc->arg.buf += size;
1248 return size;
1249}
1250
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001251/*
1252 * Performs necessary checks before doing a write
1253 * @iov: io vector request
1254 * @nr_segs: number of segments in the iovec
1255 * @count: number of bytes to write
1256 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1257 *
1258 * Adjust number of segments and amount of bytes to write (nr_segs should be
1259 * properly initialized first). Returns appropriate error code that caller
1260 * should return or zero in case that write should be allowed.
1261 */
1262int generic_segment_checks(const struct iovec *iov,
1263 unsigned long *nr_segs, size_t *count, int access_flags)
1264{
1265 unsigned long seg;
1266 size_t cnt = 0;
1267 for (seg = 0; seg < *nr_segs; seg++) {
1268 const struct iovec *iv = &iov[seg];
1269
1270 /*
1271 * If any segment has a negative length, or the cumulative
1272 * length ever wraps negative then return -EINVAL.
1273 */
1274 cnt += iv->iov_len;
1275 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1276 return -EINVAL;
1277 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1278 continue;
1279 if (seg == 0)
1280 return -EFAULT;
1281 *nr_segs = seg;
1282 cnt -= iv->iov_len; /* This segment is no good */
1283 break;
1284 }
1285 *count = cnt;
1286 return 0;
1287}
1288EXPORT_SYMBOL(generic_segment_checks);
1289
Randy Dunlap485bb992006-06-23 02:03:49 -07001290/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001291 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001292 * @iocb: kernel I/O control block
1293 * @iov: io vector request
1294 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001295 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001296 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 * This is the "read()" routine for all filesystems
1298 * that can use the page cache directly.
1299 */
1300ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001301generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1302 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303{
1304 struct file *filp = iocb->ki_filp;
1305 ssize_t retval;
1306 unsigned long seg;
1307 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001308 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001311 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1312 if (retval)
1313 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
1315 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1316 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001317 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 struct address_space *mapping;
1319 struct inode *inode;
1320
1321 mapping = filp->f_mapping;
1322 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 if (!count)
1324 goto out; /* skip atime */
1325 size = i_size_read(inode);
1326 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001327 retval = filemap_write_and_wait_range(mapping, pos,
1328 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001329 if (!retval) {
1330 retval = mapping->a_ops->direct_IO(READ, iocb,
1331 iov, pos, nr_segs);
1332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 if (retval > 0)
1334 *ppos = pos + retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001335 if (retval) {
1336 file_accessed(filp);
1337 goto out;
1338 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001339 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 }
1341
Hugh Dickins11fa9772008-07-23 21:27:34 -07001342 for (seg = 0; seg < nr_segs; seg++) {
1343 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Hugh Dickins11fa9772008-07-23 21:27:34 -07001345 desc.written = 0;
1346 desc.arg.buf = iov[seg].iov_base;
1347 desc.count = iov[seg].iov_len;
1348 if (desc.count == 0)
1349 continue;
1350 desc.error = 0;
1351 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1352 retval += desc.written;
1353 if (desc.error) {
1354 retval = retval ?: desc.error;
1355 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001357 if (desc.count > 0)
1358 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 }
1360out:
1361 return retval;
1362}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363EXPORT_SYMBOL(generic_file_aio_read);
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365static ssize_t
1366do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001367 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368{
1369 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1370 return -EINVAL;
1371
1372 force_page_cache_readahead(mapping, filp, index,
1373 max_sane_readahead(nr));
1374 return 0;
1375}
1376
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001377SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378{
1379 ssize_t ret;
1380 struct file *file;
1381
1382 ret = -EBADF;
1383 file = fget(fd);
1384 if (file) {
1385 if (file->f_mode & FMODE_READ) {
1386 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001387 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1388 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 unsigned long len = end - start + 1;
1390 ret = do_readahead(mapping, file, start, len);
1391 }
1392 fput(file);
1393 }
1394 return ret;
1395}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001396#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1397asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1398{
1399 return SYSC_readahead((int) fd, offset, (size_t) count);
1400}
1401SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1402#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
1404#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001405/**
1406 * page_cache_read - adds requested page to the page cache if not already there
1407 * @file: file to read
1408 * @offset: page index
1409 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 * This adds the requested page to the page cache if it isn't already there,
1411 * and schedules an I/O to read in its contents from disk.
1412 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001413static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414{
1415 struct address_space *mapping = file->f_mapping;
1416 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001417 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Zach Brown994fc28c2005-12-15 14:28:17 -08001419 do {
1420 page = page_cache_alloc_cold(mapping);
1421 if (!page)
1422 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Zach Brown994fc28c2005-12-15 14:28:17 -08001424 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1425 if (ret == 0)
1426 ret = mapping->a_ops->readpage(file, page);
1427 else if (ret == -EEXIST)
1428 ret = 0; /* losing race to add is OK */
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Zach Brown994fc28c2005-12-15 14:28:17 -08001432 } while (ret == AOP_TRUNCATED_PAGE);
1433
1434 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435}
1436
1437#define MMAP_LOTSAMISS (100)
1438
Randy Dunlap485bb992006-06-23 02:03:49 -07001439/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001440 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001441 * @vma: vma in which the fault was taken
1442 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001443 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001444 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 * mapped memory region to read in file data during a page fault.
1446 *
1447 * The goto's are kind of ugly, but this streamlines the normal case of having
1448 * it in the page cache, and handles the special cases reasonably without
1449 * having a lot of duplicated code.
1450 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001451int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452{
1453 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001454 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 struct address_space *mapping = file->f_mapping;
1456 struct file_ra_state *ra = &file->f_ra;
1457 struct inode *inode = mapping->host;
1458 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001459 pgoff_t size;
Nick Piggin54cb8822007-07-19 01:46:59 -07001460 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001461 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001464 if (vmf->pgoff >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001465 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
1467 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001468 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 goto no_cached_page;
1470
1471 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 * Do we have something in the page cache already?
1473 */
1474retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001475 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001476 /*
1477 * For sequential accesses, we use the generic readahead logic.
1478 */
1479 if (VM_SequentialReadHint(vma)) {
1480 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001481 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001482 vmf->pgoff, 1);
1483 page = find_lock_page(mapping, vmf->pgoff);
1484 if (!page)
1485 goto no_cached_page;
1486 }
1487 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001488 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001489 vmf->pgoff, 1);
1490 }
1491 }
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (!page) {
1494 unsigned long ra_pages;
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 ra->mmap_miss++;
1497
1498 /*
1499 * Do we miss much more than hit in this file? If so,
1500 * stop bothering with read-ahead. It will only hurt.
1501 */
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001502 if (ra->mmap_miss > MMAP_LOTSAMISS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 goto no_cached_page;
1504
1505 /*
1506 * To keep the pgmajfault counter straight, we need to
1507 * check did_readaround, as this is an inner loop.
1508 */
1509 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001510 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001511 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
1513 did_readaround = 1;
1514 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1515 if (ra_pages) {
1516 pgoff_t start = 0;
1517
Nick Piggind0217ac2007-07-19 01:47:03 -07001518 if (vmf->pgoff > ra_pages / 2)
1519 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 do_page_cache_readahead(mapping, file, start, ra_pages);
1521 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001522 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 if (!page)
1524 goto no_cached_page;
1525 }
1526
1527 if (!did_readaround)
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001528 ra->mmap_miss--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
1530 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001531 * We have a locked page in the page cache, now we need to check
1532 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 */
Nick Piggind00806b2007-07-19 01:46:57 -07001534 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 goto page_not_uptodate;
1536
Nick Piggind00806b2007-07-19 01:46:57 -07001537 /* Must recheck i_size under page lock */
1538 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001539 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001540 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001541 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001542 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001543 }
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 /*
1546 * Found the page and have a reference on it.
1547 */
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001548 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001549 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001550 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552no_cached_page:
1553 /*
1554 * We're only likely to ever get here if MADV_RANDOM is in
1555 * effect.
1556 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001557 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
1559 /*
1560 * The page we want has now been added to the page cache.
1561 * In the unlikely event that someone removed it in the
1562 * meantime, we'll just come back here and read it again.
1563 */
1564 if (error >= 0)
1565 goto retry_find;
1566
1567 /*
1568 * An error return from page_cache_read can result if the
1569 * system is low on memory, or a problem occurs while trying
1570 * to schedule I/O.
1571 */
1572 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001573 return VM_FAULT_OOM;
1574 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001577 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001579 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001580 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
1583 /*
1584 * Umm, take care of errors if the page isn't up-to-date.
1585 * Try to re-read it _once_. We do this synchronously,
1586 * because there really aren't any performance issues here
1587 * and we need to check for errors.
1588 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001590 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001591 if (!error) {
1592 wait_on_page_locked(page);
1593 if (!PageUptodate(page))
1594 error = -EIO;
1595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001597
1598 if (!error || error == AOP_TRUNCATED_PAGE)
1599 goto retry_find;
1600
1601 /* Things didn't work out. Return zero to tell the mm layer so. */
1602 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001603 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001604}
1605EXPORT_SYMBOL(filemap_fault);
1606
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001608 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609};
1610
1611/* This is used for a general mmap of a disk file */
1612
1613int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1614{
1615 struct address_space *mapping = file->f_mapping;
1616
1617 if (!mapping->a_ops->readpage)
1618 return -ENOEXEC;
1619 file_accessed(file);
1620 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001621 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 return 0;
1623}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625/*
1626 * This is for filesystems which do not implement ->writepage.
1627 */
1628int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1629{
1630 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1631 return -EINVAL;
1632 return generic_file_mmap(file, vma);
1633}
1634#else
1635int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1636{
1637 return -ENOSYS;
1638}
1639int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1640{
1641 return -ENOSYS;
1642}
1643#endif /* CONFIG_MMU */
1644
1645EXPORT_SYMBOL(generic_file_mmap);
1646EXPORT_SYMBOL(generic_file_readonly_mmap);
1647
Nick Piggin6fe69002007-05-06 14:49:04 -07001648static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001649 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 int (*filler)(void *,struct page*),
1651 void *data)
1652{
Nick Piggineb2be182007-10-16 01:24:57 -07001653 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 int err;
1655repeat:
1656 page = find_get_page(mapping, index);
1657 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001658 page = page_cache_alloc_cold(mapping);
1659 if (!page)
1660 return ERR_PTR(-ENOMEM);
1661 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1662 if (unlikely(err)) {
1663 page_cache_release(page);
1664 if (err == -EEXIST)
1665 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 return ERR_PTR(err);
1668 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 err = filler(data, page);
1670 if (err < 0) {
1671 page_cache_release(page);
1672 page = ERR_PTR(err);
1673 }
1674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 return page;
1676}
1677
Randy Dunlap76824862008-03-19 17:00:40 -07001678/**
1679 * read_cache_page_async - read into page cache, fill it if needed
1680 * @mapping: the page's address_space
1681 * @index: the page index
1682 * @filler: function to perform the read
1683 * @data: destination for read data
1684 *
Nick Piggin6fe69002007-05-06 14:49:04 -07001685 * Same as read_cache_page, but don't wait for page to become unlocked
1686 * after submitting it to the filler.
Randy Dunlap76824862008-03-19 17:00:40 -07001687 *
1688 * Read into the page cache. If a page already exists, and PageUptodate() is
1689 * not set, try to fill the page but don't wait for it to become unlocked.
1690 *
1691 * If the page does not get brought uptodate, return -EIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001693struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001694 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 int (*filler)(void *,struct page*),
1696 void *data)
1697{
1698 struct page *page;
1699 int err;
1700
1701retry:
1702 page = __read_cache_page(mapping, index, filler, data);
1703 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001704 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 if (PageUptodate(page))
1706 goto out;
1707
1708 lock_page(page);
1709 if (!page->mapping) {
1710 unlock_page(page);
1711 page_cache_release(page);
1712 goto retry;
1713 }
1714 if (PageUptodate(page)) {
1715 unlock_page(page);
1716 goto out;
1717 }
1718 err = filler(data, page);
1719 if (err < 0) {
1720 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001721 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 }
David Howellsc855ff32007-05-09 13:42:20 +01001723out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001724 mark_page_accessed(page);
1725 return page;
1726}
1727EXPORT_SYMBOL(read_cache_page_async);
1728
1729/**
1730 * read_cache_page - read into page cache, fill it if needed
1731 * @mapping: the page's address_space
1732 * @index: the page index
1733 * @filler: function to perform the read
1734 * @data: destination for read data
1735 *
1736 * Read into the page cache. If a page already exists, and PageUptodate() is
1737 * not set, try to fill the page then wait for it to become unlocked.
1738 *
1739 * If the page does not get brought uptodate, return -EIO.
1740 */
1741struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001742 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001743 int (*filler)(void *,struct page*),
1744 void *data)
1745{
1746 struct page *page;
1747
1748 page = read_cache_page_async(mapping, index, filler, data);
1749 if (IS_ERR(page))
1750 goto out;
1751 wait_on_page_locked(page);
1752 if (!PageUptodate(page)) {
1753 page_cache_release(page);
1754 page = ERR_PTR(-EIO);
1755 }
1756 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 return page;
1758}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759EXPORT_SYMBOL(read_cache_page);
1760
1761/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 * The logic we want is
1763 *
1764 * if suid or (sgid and xgrp)
1765 * remove privs
1766 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001767int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768{
1769 mode_t mode = dentry->d_inode->i_mode;
1770 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
1772 /* suid always must be killed */
1773 if (unlikely(mode & S_ISUID))
1774 kill = ATTR_KILL_SUID;
1775
1776 /*
1777 * sgid without any exec bits is just a mandatory locking mark; leave
1778 * it alone. If some exec bits are set, it's a real sgid; kill it.
1779 */
1780 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1781 kill |= ATTR_KILL_SGID;
1782
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001783 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001784 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
Jens Axboe01de85e2006-10-17 19:50:36 +02001786 return 0;
1787}
Mark Fashehd23a1472006-10-17 17:05:18 -07001788EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001789
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001790static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001791{
1792 struct iattr newattrs;
1793
1794 newattrs.ia_valid = ATTR_FORCE | kill;
1795 return notify_change(dentry, &newattrs);
1796}
1797
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001798int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001799{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001800 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001801 int killsuid = should_remove_suid(dentry);
1802 int killpriv = security_inode_need_killpriv(dentry);
1803 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001804
Serge E. Hallynb5376772007-10-16 23:31:36 -07001805 if (killpriv < 0)
1806 return killpriv;
1807 if (killpriv)
1808 error = security_inode_killpriv(dentry);
1809 if (!error && killsuid)
1810 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001811
Serge E. Hallynb5376772007-10-16 23:31:36 -07001812 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001814EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
Nick Piggin2f718ff2007-10-16 01:24:59 -07001816static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 const struct iovec *iov, size_t base, size_t bytes)
1818{
Ingo Molnarf1800532009-03-02 11:00:57 +01001819 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
1821 while (bytes) {
1822 char __user *buf = iov->iov_base + base;
1823 int copy = min(bytes, iov->iov_len - base);
1824
1825 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001826 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 copied += copy;
1828 bytes -= copy;
1829 vaddr += copy;
1830 iov++;
1831
NeilBrown01408c42006-06-25 05:47:58 -07001832 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 }
1835 return copied - left;
1836}
1837
1838/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001839 * Copy as much as we can into the page and return the number of bytes which
1840 * were sucessfully copied. If a fault is encountered then return the number of
1841 * bytes which were copied.
1842 */
1843size_t iov_iter_copy_from_user_atomic(struct page *page,
1844 struct iov_iter *i, unsigned long offset, size_t bytes)
1845{
1846 char *kaddr;
1847 size_t copied;
1848
1849 BUG_ON(!in_atomic());
1850 kaddr = kmap_atomic(page, KM_USER0);
1851 if (likely(i->nr_segs == 1)) {
1852 int left;
1853 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001854 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001855 copied = bytes - left;
1856 } else {
1857 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1858 i->iov, i->iov_offset, bytes);
1859 }
1860 kunmap_atomic(kaddr, KM_USER0);
1861
1862 return copied;
1863}
Nick Piggin89e10782007-10-16 01:25:07 -07001864EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001865
1866/*
1867 * This has the same sideeffects and return value as
1868 * iov_iter_copy_from_user_atomic().
1869 * The difference is that it attempts to resolve faults.
1870 * Page must not be locked.
1871 */
1872size_t iov_iter_copy_from_user(struct page *page,
1873 struct iov_iter *i, unsigned long offset, size_t bytes)
1874{
1875 char *kaddr;
1876 size_t copied;
1877
1878 kaddr = kmap(page);
1879 if (likely(i->nr_segs == 1)) {
1880 int left;
1881 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001882 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001883 copied = bytes - left;
1884 } else {
1885 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1886 i->iov, i->iov_offset, bytes);
1887 }
1888 kunmap(page);
1889 return copied;
1890}
Nick Piggin89e10782007-10-16 01:25:07 -07001891EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001892
Nick Pigginf7009262008-03-10 11:43:59 -07001893void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001894{
Nick Pigginf7009262008-03-10 11:43:59 -07001895 BUG_ON(i->count < bytes);
1896
Nick Piggin2f718ff2007-10-16 01:24:59 -07001897 if (likely(i->nr_segs == 1)) {
1898 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001899 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001900 } else {
1901 const struct iovec *iov = i->iov;
1902 size_t base = i->iov_offset;
1903
Nick Piggin124d3b72008-02-02 15:01:17 +01001904 /*
1905 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001906 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001907 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001908 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001909 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001910
Nick Pigginf7009262008-03-10 11:43:59 -07001911 copy = min(bytes, iov->iov_len - base);
1912 BUG_ON(!i->count || i->count < copy);
1913 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001914 bytes -= copy;
1915 base += copy;
1916 if (iov->iov_len == base) {
1917 iov++;
1918 base = 0;
1919 }
1920 }
1921 i->iov = iov;
1922 i->iov_offset = base;
1923 }
1924}
Nick Piggin89e10782007-10-16 01:25:07 -07001925EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001926
Nick Pigginafddba42007-10-16 01:25:01 -07001927/*
1928 * Fault in the first iovec of the given iov_iter, to a maximum length
1929 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1930 * accessed (ie. because it is an invalid address).
1931 *
1932 * writev-intensive code may want this to prefault several iovecs -- that
1933 * would be possible (callers must not rely on the fact that _only_ the
1934 * first iovec will be faulted with the current implementation).
1935 */
1936int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001937{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001938 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001939 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1940 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001941}
Nick Piggin89e10782007-10-16 01:25:07 -07001942EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001943
1944/*
1945 * Return the count of just the current iov_iter segment.
1946 */
1947size_t iov_iter_single_seg_count(struct iov_iter *i)
1948{
1949 const struct iovec *iov = i->iov;
1950 if (i->nr_segs == 1)
1951 return i->count;
1952 else
1953 return min(i->count, iov->iov_len - i->iov_offset);
1954}
Nick Piggin89e10782007-10-16 01:25:07 -07001955EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001956
1957/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 * Performs necessary checks before doing a write
1959 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001960 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 * Returns appropriate error code that caller should return or
1962 * zero in case that write should be allowed.
1963 */
1964inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1965{
1966 struct inode *inode = file->f_mapping->host;
1967 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1968
1969 if (unlikely(*pos < 0))
1970 return -EINVAL;
1971
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 if (!isblk) {
1973 /* FIXME: this is for backwards compatibility with 2.4 */
1974 if (file->f_flags & O_APPEND)
1975 *pos = i_size_read(inode);
1976
1977 if (limit != RLIM_INFINITY) {
1978 if (*pos >= limit) {
1979 send_sig(SIGXFSZ, current, 0);
1980 return -EFBIG;
1981 }
1982 if (*count > limit - (typeof(limit))*pos) {
1983 *count = limit - (typeof(limit))*pos;
1984 }
1985 }
1986 }
1987
1988 /*
1989 * LFS rule
1990 */
1991 if (unlikely(*pos + *count > MAX_NON_LFS &&
1992 !(file->f_flags & O_LARGEFILE))) {
1993 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return -EFBIG;
1995 }
1996 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1997 *count = MAX_NON_LFS - (unsigned long)*pos;
1998 }
1999 }
2000
2001 /*
2002 * Are we about to exceed the fs block limit ?
2003 *
2004 * If we have written data it becomes a short write. If we have
2005 * exceeded without writing data we send a signal and return EFBIG.
2006 * Linus frestrict idea will clean these up nicely..
2007 */
2008 if (likely(!isblk)) {
2009 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2010 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 return -EFBIG;
2012 }
2013 /* zero-length writes at ->s_maxbytes are OK */
2014 }
2015
2016 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2017 *count = inode->i_sb->s_maxbytes - *pos;
2018 } else {
David Howells93614012006-09-30 20:45:40 +02002019#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 loff_t isize;
2021 if (bdev_read_only(I_BDEV(inode)))
2022 return -EPERM;
2023 isize = i_size_read(inode);
2024 if (*pos >= isize) {
2025 if (*count || *pos > isize)
2026 return -ENOSPC;
2027 }
2028
2029 if (*pos + *count > isize)
2030 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002031#else
2032 return -EPERM;
2033#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 }
2035 return 0;
2036}
2037EXPORT_SYMBOL(generic_write_checks);
2038
Nick Pigginafddba42007-10-16 01:25:01 -07002039int pagecache_write_begin(struct file *file, struct address_space *mapping,
2040 loff_t pos, unsigned len, unsigned flags,
2041 struct page **pagep, void **fsdata)
2042{
2043 const struct address_space_operations *aops = mapping->a_ops;
2044
Nick Piggin4e02ed42008-10-29 14:00:55 -07002045 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002046 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002047}
2048EXPORT_SYMBOL(pagecache_write_begin);
2049
2050int pagecache_write_end(struct file *file, struct address_space *mapping,
2051 loff_t pos, unsigned len, unsigned copied,
2052 struct page *page, void *fsdata)
2053{
2054 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002055
Nick Piggin4e02ed42008-10-29 14:00:55 -07002056 mark_page_accessed(page);
2057 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002058}
2059EXPORT_SYMBOL(pagecache_write_end);
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061ssize_t
2062generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2063 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2064 size_t count, size_t ocount)
2065{
2066 struct file *file = iocb->ki_filp;
2067 struct address_space *mapping = file->f_mapping;
2068 struct inode *inode = mapping->host;
2069 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002070 size_t write_len;
2071 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
2073 if (count != ocount)
2074 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2075
Christoph Hellwiga969e902008-07-23 21:27:04 -07002076 write_len = iov_length(iov, *nr_segs);
2077 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002078
Nick Piggin48b47c52009-01-06 14:40:22 -08002079 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002080 if (written)
2081 goto out;
2082
2083 /*
2084 * After a write we want buffered reads to be sure to go to disk to get
2085 * the new data. We invalidate clean cached page from the region we're
2086 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002087 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002088 */
2089 if (mapping->nrpages) {
2090 written = invalidate_inode_pages2_range(mapping,
2091 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002092 /*
2093 * If a page can not be invalidated, return 0 to fall back
2094 * to buffered write.
2095 */
2096 if (written) {
2097 if (written == -EBUSY)
2098 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002099 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002100 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002101 }
2102
2103 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2104
2105 /*
2106 * Finally, try again to invalidate clean pages which might have been
2107 * cached by non-direct readahead, or faulted in by get_user_pages()
2108 * if the source of the write was an mmap'ed region of the file
2109 * we're writing. Either one is a pretty crazy thing to do,
2110 * so we don't support it 100%. If this invalidation
2111 * fails, tough, the write still worked...
2112 */
2113 if (mapping->nrpages) {
2114 invalidate_inode_pages2_range(mapping,
2115 pos >> PAGE_CACHE_SHIFT, end);
2116 }
2117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 if (written > 0) {
2119 loff_t end = pos + written;
2120 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2121 i_size_write(inode, end);
2122 mark_inode_dirty(inode);
2123 }
2124 *ppos = end;
2125 }
2126
2127 /*
2128 * Sync the fs metadata but not the minor inode changes and
2129 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002130 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002131 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 */
Christoph Hellwiga969e902008-07-23 21:27:04 -07002133out:
Zach Brown8459d862006-12-10 02:21:05 -08002134 if ((written >= 0 || written == -EIOCBQUEUED) &&
2135 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002136 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2137 if (err < 0)
2138 written = err;
2139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 return written;
2141}
2142EXPORT_SYMBOL(generic_file_direct_write);
2143
Nick Piggineb2be182007-10-16 01:24:57 -07002144/*
2145 * Find or create a page at the given pagecache position. Return the locked
2146 * page. This function is specifically for buffered writes.
2147 */
Nick Piggin54566b22009-01-04 12:00:53 -08002148struct page *grab_cache_page_write_begin(struct address_space *mapping,
2149 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002150{
2151 int status;
2152 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002153 gfp_t gfp_notmask = 0;
2154 if (flags & AOP_FLAG_NOFS)
2155 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002156repeat:
2157 page = find_lock_page(mapping, index);
2158 if (likely(page))
2159 return page;
2160
Nick Piggin54566b22009-01-04 12:00:53 -08002161 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002162 if (!page)
2163 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002164 status = add_to_page_cache_lru(page, mapping, index,
2165 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002166 if (unlikely(status)) {
2167 page_cache_release(page);
2168 if (status == -EEXIST)
2169 goto repeat;
2170 return NULL;
2171 }
2172 return page;
2173}
Nick Piggin54566b22009-01-04 12:00:53 -08002174EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002175
Nick Pigginafddba42007-10-16 01:25:01 -07002176static ssize_t generic_perform_write(struct file *file,
2177 struct iov_iter *i, loff_t pos)
2178{
2179 struct address_space *mapping = file->f_mapping;
2180 const struct address_space_operations *a_ops = mapping->a_ops;
2181 long status = 0;
2182 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002183 unsigned int flags = 0;
2184
2185 /*
2186 * Copies from kernel address space cannot fail (NFSD is a big user).
2187 */
2188 if (segment_eq(get_fs(), KERNEL_DS))
2189 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002190
2191 do {
2192 struct page *page;
2193 pgoff_t index; /* Pagecache index for current page */
2194 unsigned long offset; /* Offset into pagecache page */
2195 unsigned long bytes; /* Bytes to write to page */
2196 size_t copied; /* Bytes copied from user */
2197 void *fsdata;
2198
2199 offset = (pos & (PAGE_CACHE_SIZE - 1));
2200 index = pos >> PAGE_CACHE_SHIFT;
2201 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2202 iov_iter_count(i));
2203
2204again:
2205
2206 /*
2207 * Bring in the user page that we will copy from _first_.
2208 * Otherwise there's a nasty deadlock on copying from the
2209 * same page as we're writing to, without it being marked
2210 * up-to-date.
2211 *
2212 * Not only is this an optimisation, but it is also required
2213 * to check that the address is actually valid, when atomic
2214 * usercopies are used, below.
2215 */
2216 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2217 status = -EFAULT;
2218 break;
2219 }
2220
Nick Piggin674b8922007-10-16 01:25:03 -07002221 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002222 &page, &fsdata);
2223 if (unlikely(status))
2224 break;
2225
2226 pagefault_disable();
2227 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2228 pagefault_enable();
2229 flush_dcache_page(page);
2230
2231 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2232 page, fsdata);
2233 if (unlikely(status < 0))
2234 break;
2235 copied = status;
2236
2237 cond_resched();
2238
Nick Piggin124d3b72008-02-02 15:01:17 +01002239 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002240 if (unlikely(copied == 0)) {
2241 /*
2242 * If we were unable to copy any data at all, we must
2243 * fall back to a single segment length write.
2244 *
2245 * If we didn't fallback here, we could livelock
2246 * because not all segments in the iov can be copied at
2247 * once without a pagefault.
2248 */
2249 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2250 iov_iter_single_seg_count(i));
2251 goto again;
2252 }
Nick Pigginafddba42007-10-16 01:25:01 -07002253 pos += copied;
2254 written += copied;
2255
2256 balance_dirty_pages_ratelimited(mapping);
2257
2258 } while (iov_iter_count(i));
2259
2260 return written ? written : status;
2261}
2262
2263ssize_t
2264generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2265 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2266 size_t count, ssize_t written)
2267{
2268 struct file *file = iocb->ki_filp;
2269 struct address_space *mapping = file->f_mapping;
2270 const struct address_space_operations *a_ops = mapping->a_ops;
2271 struct inode *inode = mapping->host;
2272 ssize_t status;
2273 struct iov_iter i;
2274
2275 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002276 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002279 written += status;
2280 *ppos = pos + status;
2281
2282 /*
2283 * For now, when the user asks for O_SYNC, we'll actually give
2284 * O_DSYNC
2285 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2287 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2288 status = generic_osync_inode(inode, mapping,
2289 OSYNC_METADATA|OSYNC_DATA);
2290 }
2291 }
2292
2293 /*
2294 * If we get here for O_DIRECT writes then we must have fallen through
2295 * to buffered writes (block instantiation inside i_size). So we sync
2296 * the file data here, to try to honour O_DIRECT expectations.
2297 */
2298 if (unlikely(file->f_flags & O_DIRECT) && written)
Nick Piggin48b47c52009-01-06 14:40:22 -08002299 status = filemap_write_and_wait_range(mapping,
2300 pos, pos + written - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 return written ? written : status;
2303}
2304EXPORT_SYMBOL(generic_file_buffered_write);
2305
Adrian Bunk5ce78522005-09-10 00:26:28 -07002306static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2308 unsigned long nr_segs, loff_t *ppos)
2309{
2310 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002311 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 size_t ocount; /* original count */
2313 size_t count; /* after file limit checks */
2314 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 loff_t pos;
2316 ssize_t written;
2317 ssize_t err;
2318
2319 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002320 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2321 if (err)
2322 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
2324 count = ocount;
2325 pos = *ppos;
2326
2327 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2328
2329 /* We can write back this queue in page reclaim */
2330 current->backing_dev_info = mapping->backing_dev_info;
2331 written = 0;
2332
2333 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2334 if (err)
2335 goto out;
2336
2337 if (count == 0)
2338 goto out;
2339
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002340 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 if (err)
2342 goto out;
2343
Christoph Hellwig870f4812006-01-09 20:52:01 -08002344 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2347 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002348 loff_t endbyte;
2349 ssize_t written_buffered;
2350
2351 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2352 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 if (written < 0 || written == count)
2354 goto out;
2355 /*
2356 * direct-io write to a hole: fall through to buffered I/O
2357 * for completing the rest of the request.
2358 */
2359 pos += written;
2360 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002361 written_buffered = generic_file_buffered_write(iocb, iov,
2362 nr_segs, pos, ppos, count,
2363 written);
2364 /*
2365 * If generic_file_buffered_write() retuned a synchronous error
2366 * then we want to return the number of bytes which were
2367 * direct-written, or the error code if that was zero. Note
2368 * that this differs from normal direct-io semantics, which
2369 * will return -EFOO even if some bytes were written.
2370 */
2371 if (written_buffered < 0) {
2372 err = written_buffered;
2373 goto out;
2374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002376 /*
2377 * We need to ensure that the page cache pages are written to
2378 * disk and invalidated to preserve the expected O_DIRECT
2379 * semantics.
2380 */
2381 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002382 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2383 SYNC_FILE_RANGE_WAIT_BEFORE|
2384 SYNC_FILE_RANGE_WRITE|
2385 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002386 if (err == 0) {
2387 written = written_buffered;
2388 invalidate_mapping_pages(mapping,
2389 pos >> PAGE_CACHE_SHIFT,
2390 endbyte >> PAGE_CACHE_SHIFT);
2391 } else {
2392 /*
2393 * We don't know how much we wrote, so just return
2394 * the number of bytes which were direct-written
2395 */
2396 }
2397 } else {
2398 written = generic_file_buffered_write(iocb, iov, nr_segs,
2399 pos, ppos, count, written);
2400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401out:
2402 current->backing_dev_info = NULL;
2403 return written ? written : err;
2404}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Badari Pulavarty027445c2006-09-30 23:28:46 -07002406ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2407 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
2409 struct file *file = iocb->ki_filp;
2410 struct address_space *mapping = file->f_mapping;
2411 struct inode *inode = mapping->host;
2412 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413
Badari Pulavarty027445c2006-09-30 23:28:46 -07002414 BUG_ON(iocb->ki_pos != pos);
2415
2416 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2417 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
2419 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002420 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
2422 err = sync_page_range_nolock(inode, mapping, pos, ret);
2423 if (err < 0)
2424 ret = err;
2425 }
2426 return ret;
2427}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002428EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
Badari Pulavarty027445c2006-09-30 23:28:46 -07002430ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2431 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432{
2433 struct file *file = iocb->ki_filp;
2434 struct address_space *mapping = file->f_mapping;
2435 struct inode *inode = mapping->host;
2436 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
2438 BUG_ON(iocb->ki_pos != pos);
2439
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002440 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002441 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2442 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002443 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
2445 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2446 ssize_t err;
2447
2448 err = sync_page_range(inode, mapping, pos, ret);
2449 if (err < 0)
2450 ret = err;
2451 }
2452 return ret;
2453}
2454EXPORT_SYMBOL(generic_file_aio_write);
2455
David Howellscf9a2ae2006-08-29 19:05:54 +01002456/**
2457 * try_to_release_page() - release old fs-specific metadata on a page
2458 *
2459 * @page: the page which the kernel is trying to free
2460 * @gfp_mask: memory allocation flags (and I/O mode)
2461 *
2462 * The address_space is to try to release any data against the page
2463 * (presumably at page->private). If the release was successful, return `1'.
2464 * Otherwise return zero.
2465 *
2466 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002467 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002468 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002469 */
2470int try_to_release_page(struct page *page, gfp_t gfp_mask)
2471{
2472 struct address_space * const mapping = page->mapping;
2473
2474 BUG_ON(!PageLocked(page));
2475 if (PageWriteback(page))
2476 return 0;
2477
2478 if (mapping && mapping->a_ops->releasepage)
2479 return mapping->a_ops->releasepage(page, gfp_mask);
2480 return try_to_free_buffers(page);
2481}
2482
2483EXPORT_SYMBOL(try_to_release_page);