blob: 195339b27e9a2d037e58921d099b1de30fe3461f [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>
28#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Nick Piggin0f8053a2006-03-22 00:08:33 -080034#include "internal.h"
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 * FIXME: remove all knowledge of the buffer layer from the core VM
38 */
39#include <linux/buffer_head.h> /* for generic_osync_inode */
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <asm/mman.h>
42
Adrian Bunk5ce78522005-09-10 00:26:28 -070043static ssize_t
44generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
45 loff_t offset, unsigned long nr_segs);
46
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
113 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
114 */
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));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
125
126void remove_from_page_cache(struct page *page)
127{
128 struct address_space *mapping = page->mapping;
129
Matt Mackallcd7619d2005-05-01 08:59:01 -0700130 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 write_lock_irq(&mapping->tree_lock);
133 __remove_from_page_cache(page);
134 write_unlock_irq(&mapping->tree_lock);
135}
136
137static int sync_page(void *word)
138{
139 struct address_space *mapping;
140 struct page *page;
141
Andi Kleen07808b72005-11-05 17:25:53 +0100142 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
144 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700145 * page_mapping() is being called without PG_locked held.
146 * Some knowledge of the state and use of the page is used to
147 * reduce the requirements down to a memory barrier.
148 * The danger here is of a stale page_mapping() return value
149 * indicating a struct address_space different from the one it's
150 * associated with when it is associated with one.
151 * After smp_mb(), it's either the correct page_mapping() for
152 * the page, or an old page_mapping() and the page's own
153 * page_mapping() has gone NULL.
154 * The ->sync_page() address_space operation must tolerate
155 * page_mapping() going NULL. By an amazing coincidence,
156 * this comes about because none of the users of the page
157 * in the ->sync_page() methods make essential use of the
158 * page_mapping(), merely passing the page down to the backing
159 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700160 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700161 * of interest. When page_mapping() does go NULL, the entire
162 * call stack gracefully ignores the page and returns.
163 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 */
165 smp_mb();
166 mapping = page_mapping(page);
167 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
168 mapping->a_ops->sync_page(page);
169 io_schedule();
170 return 0;
171}
172
173/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700174 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700175 * @mapping: address space structure to write
176 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800177 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700178 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700180 * Start writeback against all of a mapping's dirty pages that lie
181 * within the byte offsets <start, end> inclusive.
182 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700184 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 * these two operations is that if a dirty page/buffer is encountered, it must
186 * be waited upon, and not just skipped over.
187 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800188int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
189 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
191 int ret;
192 struct writeback_control wbc = {
193 .sync_mode = sync_mode,
194 .nr_to_write = mapping->nrpages * 2,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700195 .range_start = start,
196 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 };
198
199 if (!mapping_cap_writeback_dirty(mapping))
200 return 0;
201
202 ret = do_writepages(mapping, &wbc);
203 return ret;
204}
205
206static inline int __filemap_fdatawrite(struct address_space *mapping,
207 int sync_mode)
208{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700209 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
211
212int filemap_fdatawrite(struct address_space *mapping)
213{
214 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
215}
216EXPORT_SYMBOL(filemap_fdatawrite);
217
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800218static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
219 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
221 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
222}
223
Randy Dunlap485bb992006-06-23 02:03:49 -0700224/**
225 * filemap_flush - mostly a non-blocking flush
226 * @mapping: target address_space
227 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 * This is a mostly non-blocking flush. Not suitable for data-integrity
229 * purposes - I/O may not be started against all dirty pages.
230 */
231int filemap_flush(struct address_space *mapping)
232{
233 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
234}
235EXPORT_SYMBOL(filemap_flush);
236
Randy Dunlap485bb992006-06-23 02:03:49 -0700237/**
238 * wait_on_page_writeback_range - wait for writeback to complete
239 * @mapping: target address_space
240 * @start: beginning page index
241 * @end: ending page index
242 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 * Wait for writeback to complete against pages indexed by start->end
244 * inclusive
245 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800246int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 pgoff_t start, pgoff_t end)
248{
249 struct pagevec pvec;
250 int nr_pages;
251 int ret = 0;
252 pgoff_t index;
253
254 if (end < start)
255 return 0;
256
257 pagevec_init(&pvec, 0);
258 index = start;
259 while ((index <= end) &&
260 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
261 PAGECACHE_TAG_WRITEBACK,
262 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
263 unsigned i;
264
265 for (i = 0; i < nr_pages; i++) {
266 struct page *page = pvec.pages[i];
267
268 /* until radix tree lookup accepts end_index */
269 if (page->index > end)
270 continue;
271
272 wait_on_page_writeback(page);
273 if (PageError(page))
274 ret = -EIO;
275 }
276 pagevec_release(&pvec);
277 cond_resched();
278 }
279
280 /* Check for outstanding write errors */
281 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
282 ret = -ENOSPC;
283 if (test_and_clear_bit(AS_EIO, &mapping->flags))
284 ret = -EIO;
285
286 return ret;
287}
288
Randy Dunlap485bb992006-06-23 02:03:49 -0700289/**
290 * sync_page_range - write and wait on all pages in the passed range
291 * @inode: target inode
292 * @mapping: target address_space
293 * @pos: beginning offset in pages to write
294 * @count: number of bytes to write
295 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 * Write and wait upon all the pages in the passed range. This is a "data
297 * integrity" operation. It waits upon in-flight writeout before starting and
298 * waiting upon new writeout. If there was an IO error, return it.
299 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800300 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 * it is otherwise livelockable.
302 */
303int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800304 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
306 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
307 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
308 int ret;
309
310 if (!mapping_cap_writeback_dirty(mapping) || !count)
311 return 0;
312 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
313 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800314 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800316 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 }
318 if (ret == 0)
319 ret = wait_on_page_writeback_range(mapping, start, end);
320 return ret;
321}
322EXPORT_SYMBOL(sync_page_range);
323
Randy Dunlap485bb992006-06-23 02:03:49 -0700324/**
325 * sync_page_range_nolock
326 * @inode: target inode
327 * @mapping: target address_space
328 * @pos: beginning offset in pages to write
329 * @count: number of bytes to write
330 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800331 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 * as it forces O_SYNC writers to different parts of the same file
333 * to be serialised right until io completion.
334 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800335int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
336 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
338 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
339 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
340 int ret;
341
342 if (!mapping_cap_writeback_dirty(mapping) || !count)
343 return 0;
344 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
345 if (ret == 0)
346 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
347 if (ret == 0)
348 ret = wait_on_page_writeback_range(mapping, start, end);
349 return ret;
350}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800351EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700354 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700356 *
357 * Walk the list of under-writeback pages of the given address space
358 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 */
360int filemap_fdatawait(struct address_space *mapping)
361{
362 loff_t i_size = i_size_read(mapping->host);
363
364 if (i_size == 0)
365 return 0;
366
367 return wait_on_page_writeback_range(mapping, 0,
368 (i_size - 1) >> PAGE_CACHE_SHIFT);
369}
370EXPORT_SYMBOL(filemap_fdatawait);
371
372int filemap_write_and_wait(struct address_space *mapping)
373{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800374 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
376 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800377 err = filemap_fdatawrite(mapping);
378 /*
379 * Even if the above returned error, the pages may be
380 * written partially (e.g. -ENOSPC), so we wait for it.
381 * But the -EIO is special case, it may indicate the worst
382 * thing (e.g. bug) happened, so we avoid waiting for it.
383 */
384 if (err != -EIO) {
385 int err2 = filemap_fdatawait(mapping);
386 if (!err)
387 err = err2;
388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800390 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800392EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
Randy Dunlap485bb992006-06-23 02:03:49 -0700394/**
395 * filemap_write_and_wait_range - write out & wait on a file range
396 * @mapping: the address_space for the pages
397 * @lstart: offset in bytes where the range starts
398 * @lend: offset in bytes where the range ends (inclusive)
399 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800400 * Write out and wait upon file offsets lstart->lend, inclusive.
401 *
402 * Note that `lend' is inclusive (describes the last byte to be written) so
403 * that this function can be used to write to the very end-of-file (end = -1).
404 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405int filemap_write_and_wait_range(struct address_space *mapping,
406 loff_t lstart, loff_t lend)
407{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800408 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
410 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800411 err = __filemap_fdatawrite_range(mapping, lstart, lend,
412 WB_SYNC_ALL);
413 /* See comment of filemap_write_and_wait() */
414 if (err != -EIO) {
415 int err2 = wait_on_page_writeback_range(mapping,
416 lstart >> PAGE_CACHE_SHIFT,
417 lend >> PAGE_CACHE_SHIFT);
418 if (!err)
419 err = err2;
420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800422 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
Randy Dunlap485bb992006-06-23 02:03:49 -0700425/**
426 * add_to_page_cache - add newly allocated pagecache pages
427 * @page: page to add
428 * @mapping: the page's address_space
429 * @offset: page index
430 * @gfp_mask: page allocation mode
431 *
432 * This function is used to add newly allocated pagecache pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 * the page is new, so we can just run SetPageLocked() against it.
434 * The other page state flags were set by rmqueue().
435 *
436 * This function does not add the page to the LRU. The caller must do that.
437 */
438int add_to_page_cache(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400439 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
442
443 if (error == 0) {
444 write_lock_irq(&mapping->tree_lock);
445 error = radix_tree_insert(&mapping->page_tree, offset, page);
446 if (!error) {
447 page_cache_get(page);
448 SetPageLocked(page);
449 page->mapping = mapping;
450 page->index = offset;
451 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700452 __inc_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 }
454 write_unlock_irq(&mapping->tree_lock);
455 radix_tree_preload_end();
456 }
457 return error;
458}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459EXPORT_SYMBOL(add_to_page_cache);
460
461int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400462 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
464 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
465 if (ret == 0)
466 lru_cache_add(page);
467 return ret;
468}
469
Paul Jackson44110fe2006-03-24 03:16:04 -0800470#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700471struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800472{
473 if (cpuset_do_page_mem_spread()) {
474 int n = cpuset_mem_spread_node();
Nick Piggin2ae88142006-10-28 10:38:23 -0700475 return alloc_pages_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800476 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700477 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800478}
Nick Piggin2ae88142006-10-28 10:38:23 -0700479EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800480#endif
481
Nick Piggindb376482006-09-25 23:31:24 -0700482static int __sleep_on_page_lock(void *word)
483{
484 io_schedule();
485 return 0;
486}
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488/*
489 * In order to wait for pages to become available there must be
490 * waitqueues associated with pages. By using a hash table of
491 * waitqueues where the bucket discipline is to maintain all
492 * waiters on the same queue and wake all when any of the pages
493 * become available, and for the woken contexts to check to be
494 * sure the appropriate page became available, this saves space
495 * at a cost of "thundering herd" phenomena during rare hash
496 * collisions.
497 */
498static wait_queue_head_t *page_waitqueue(struct page *page)
499{
500 const struct zone *zone = page_zone(page);
501
502 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
503}
504
505static inline void wake_up_page(struct page *page, int bit)
506{
507 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
508}
509
510void fastcall wait_on_page_bit(struct page *page, int bit_nr)
511{
512 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
513
514 if (test_bit(bit_nr, &page->flags))
515 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
516 TASK_UNINTERRUPTIBLE);
517}
518EXPORT_SYMBOL(wait_on_page_bit);
519
520/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700521 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 * @page: the page
523 *
524 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
525 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
526 * mechananism between PageLocked pages and PageWriteback pages is shared.
527 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
528 *
529 * The first mb is necessary to safely close the critical section opened by the
530 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
531 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
532 * parallel wait_on_page_locked()).
533 */
534void fastcall unlock_page(struct page *page)
535{
536 smp_mb__before_clear_bit();
537 if (!TestClearPageLocked(page))
538 BUG();
539 smp_mb__after_clear_bit();
540 wake_up_page(page, PG_locked);
541}
542EXPORT_SYMBOL(unlock_page);
543
Randy Dunlap485bb992006-06-23 02:03:49 -0700544/**
545 * end_page_writeback - end writeback against a page
546 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
548void end_page_writeback(struct page *page)
549{
550 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
551 if (!test_clear_page_writeback(page))
552 BUG();
553 }
554 smp_mb__after_clear_bit();
555 wake_up_page(page, PG_writeback);
556}
557EXPORT_SYMBOL(end_page_writeback);
558
Randy Dunlap485bb992006-06-23 02:03:49 -0700559/**
560 * __lock_page - get a lock on the page, assuming we need to sleep to get it
561 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700563 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
565 * chances are that on the second loop, the block layer's plug list is empty,
566 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
567 */
568void fastcall __lock_page(struct page *page)
569{
570 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
571
572 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
573 TASK_UNINTERRUPTIBLE);
574}
575EXPORT_SYMBOL(__lock_page);
576
Nick Piggindb376482006-09-25 23:31:24 -0700577/*
578 * Variant of lock_page that does not require the caller to hold a reference
579 * on the page's mapping.
580 */
581void fastcall __lock_page_nosync(struct page *page)
582{
583 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
584 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
585 TASK_UNINTERRUPTIBLE);
586}
587
Randy Dunlap485bb992006-06-23 02:03:49 -0700588/**
589 * find_get_page - find and get a page reference
590 * @mapping: the address_space to search
591 * @offset: the page index
592 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700593 * Is there a pagecache struct page at the given (mapping, offset) tuple?
594 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 */
Fengguang Wu57f6b962007-10-16 01:24:37 -0700596struct page * find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
598 struct page *page;
599
600 read_lock_irq(&mapping->tree_lock);
601 page = radix_tree_lookup(&mapping->page_tree, offset);
602 if (page)
603 page_cache_get(page);
604 read_unlock_irq(&mapping->tree_lock);
605 return page;
606}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607EXPORT_SYMBOL(find_get_page);
608
Randy Dunlap485bb992006-06-23 02:03:49 -0700609/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700611 * @mapping: the address_space to search
612 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 *
614 * Locates the desired pagecache page, locks it, increments its reference
615 * count and returns its address.
616 *
617 * Returns zero if the page was not present. find_lock_page() may sleep.
618 */
619struct page *find_lock_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700620 pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
622 struct page *page;
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624repeat:
Nick Piggin45726cb2007-10-16 01:24:41 -0700625 read_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 page = radix_tree_lookup(&mapping->page_tree, offset);
627 if (page) {
628 page_cache_get(page);
629 if (TestSetPageLocked(page)) {
630 read_unlock_irq(&mapping->tree_lock);
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800631 __lock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 /* Has the page been truncated while we slept? */
Nick Piggin45726cb2007-10-16 01:24:41 -0700634 if (unlikely(page->mapping != mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 unlock_page(page);
636 page_cache_release(page);
637 goto repeat;
638 }
Nick Piggin45726cb2007-10-16 01:24:41 -0700639 VM_BUG_ON(page->index != offset);
640 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
642 }
643 read_unlock_irq(&mapping->tree_lock);
Nick Piggin45726cb2007-10-16 01:24:41 -0700644out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 return page;
646}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647EXPORT_SYMBOL(find_lock_page);
648
649/**
650 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700651 * @mapping: the page's address_space
652 * @index: the page's index into the mapping
653 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 *
655 * Locates a page in the pagecache. If the page is not present, a new page
656 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
657 * LRU list. The returned page is locked and has its reference count
658 * incremented.
659 *
660 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
661 * allocation!
662 *
663 * find_or_create_page() returns the desired page's address, or zero on
664 * memory exhaustion.
665 */
666struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700667 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668{
Nick Piggineb2be182007-10-16 01:24:57 -0700669 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 int err;
671repeat:
672 page = find_lock_page(mapping, index);
673 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700674 page = __page_cache_alloc(gfp_mask);
675 if (!page)
676 return NULL;
677 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
678 if (unlikely(err)) {
679 page_cache_release(page);
680 page = NULL;
681 if (err == -EEXIST)
682 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 return page;
686}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687EXPORT_SYMBOL(find_or_create_page);
688
689/**
690 * find_get_pages - gang pagecache lookup
691 * @mapping: The address_space to search
692 * @start: The starting page index
693 * @nr_pages: The maximum number of pages
694 * @pages: Where the resulting pages are placed
695 *
696 * find_get_pages() will search for and return a group of up to
697 * @nr_pages pages in the mapping. The pages are placed at @pages.
698 * find_get_pages() takes a reference against the returned pages.
699 *
700 * The search returns a group of mapping-contiguous pages with ascending
701 * indexes. There may be holes in the indices due to not-present pages.
702 *
703 * find_get_pages() returns the number of pages which were found.
704 */
705unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
706 unsigned int nr_pages, struct page **pages)
707{
708 unsigned int i;
709 unsigned int ret;
710
711 read_lock_irq(&mapping->tree_lock);
712 ret = radix_tree_gang_lookup(&mapping->page_tree,
713 (void **)pages, start, nr_pages);
714 for (i = 0; i < ret; i++)
715 page_cache_get(pages[i]);
716 read_unlock_irq(&mapping->tree_lock);
717 return ret;
718}
719
Jens Axboeebf43502006-04-27 08:46:01 +0200720/**
721 * find_get_pages_contig - gang contiguous pagecache lookup
722 * @mapping: The address_space to search
723 * @index: The starting page index
724 * @nr_pages: The maximum number of pages
725 * @pages: Where the resulting pages are placed
726 *
727 * find_get_pages_contig() works exactly like find_get_pages(), except
728 * that the returned number of pages are guaranteed to be contiguous.
729 *
730 * find_get_pages_contig() returns the number of pages which were found.
731 */
732unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
733 unsigned int nr_pages, struct page **pages)
734{
735 unsigned int i;
736 unsigned int ret;
737
738 read_lock_irq(&mapping->tree_lock);
739 ret = radix_tree_gang_lookup(&mapping->page_tree,
740 (void **)pages, index, nr_pages);
741 for (i = 0; i < ret; i++) {
742 if (pages[i]->mapping == NULL || pages[i]->index != index)
743 break;
744
745 page_cache_get(pages[i]);
746 index++;
747 }
748 read_unlock_irq(&mapping->tree_lock);
749 return i;
750}
David Howellsef71c152007-05-09 02:33:44 -0700751EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200752
Randy Dunlap485bb992006-06-23 02:03:49 -0700753/**
754 * find_get_pages_tag - find and return pages that match @tag
755 * @mapping: the address_space to search
756 * @index: the starting page index
757 * @tag: the tag index
758 * @nr_pages: the maximum number of pages
759 * @pages: where the resulting pages are placed
760 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700762 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 */
764unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
765 int tag, unsigned int nr_pages, struct page **pages)
766{
767 unsigned int i;
768 unsigned int ret;
769
770 read_lock_irq(&mapping->tree_lock);
771 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
772 (void **)pages, *index, nr_pages, tag);
773 for (i = 0; i < ret; i++)
774 page_cache_get(pages[i]);
775 if (ret)
776 *index = pages[ret - 1]->index + 1;
777 read_unlock_irq(&mapping->tree_lock);
778 return ret;
779}
David Howellsef71c152007-05-09 02:33:44 -0700780EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Randy Dunlap485bb992006-06-23 02:03:49 -0700782/**
783 * grab_cache_page_nowait - returns locked page at given index in given cache
784 * @mapping: target address_space
785 * @index: the page index
786 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800787 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 * This is intended for speculative data generators, where the data can
789 * be regenerated if the page couldn't be grabbed. This routine should
790 * be safe to call while holding the lock for another page.
791 *
792 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
793 * and deadlock against the caller's locked page.
794 */
795struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700796grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
798 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800 if (page) {
801 if (!TestSetPageLocked(page))
802 return page;
803 page_cache_release(page);
804 return NULL;
805 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700806 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
807 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 page_cache_release(page);
809 page = NULL;
810 }
811 return page;
812}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813EXPORT_SYMBOL(grab_cache_page_nowait);
814
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700815/*
816 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
817 * a _large_ part of the i/o request. Imagine the worst scenario:
818 *
819 * ---R__________________________________________B__________
820 * ^ reading here ^ bad block(assume 4k)
821 *
822 * read(R) => miss => readahead(R...B) => media error => frustrating retries
823 * => failing the whole request => read(R) => read(R+1) =>
824 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
825 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
826 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
827 *
828 * It is going insane. Fix it by quickly scaling down the readahead size.
829 */
830static void shrink_readahead_size_eio(struct file *filp,
831 struct file_ra_state *ra)
832{
833 if (!ra->ra_pages)
834 return;
835
836 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700837}
838
Randy Dunlap485bb992006-06-23 02:03:49 -0700839/**
840 * do_generic_mapping_read - generic file read routine
841 * @mapping: address_space to be read
Stephen Hemmingerf0b85c02007-08-10 13:01:16 -0700842 * @_ra: file's readahead state
Randy Dunlap485bb992006-06-23 02:03:49 -0700843 * @filp: the file to read
844 * @ppos: current file position
845 * @desc: read_descriptor
846 * @actor: read method
847 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700849 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 *
851 * This is really ugly. But the goto's actually try to clarify some
852 * of the logic when it comes to error handling etc.
853 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700854 * Note the struct file* is only passed for the use of readpage.
855 * It may be NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 */
857void do_generic_mapping_read(struct address_space *mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700858 struct file_ra_state *ra,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 struct file *filp,
860 loff_t *ppos,
861 read_descriptor_t *desc,
862 read_actor_t actor)
863{
864 struct inode *inode = mapping->host;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700865 pgoff_t index;
866 pgoff_t last_index;
867 pgoff_t prev_index;
868 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700869 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700873 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
874 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
876 offset = *ppos & ~PAGE_CACHE_MASK;
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 for (;;) {
879 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700880 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -0700881 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 unsigned long nr, ret;
883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885find_page:
886 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700887 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700888 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700889 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700890 index, last_index - index);
891 page = find_get_page(mapping, index);
892 if (unlikely(page == NULL))
893 goto no_cached_page;
894 }
895 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700896 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700897 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700898 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 }
900 if (!PageUptodate(page))
901 goto page_not_up_to_date;
902page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -0700903 /*
904 * i_size must be checked after we know the page is Uptodate.
905 *
906 * Checking i_size after the check allows us to calculate
907 * the correct value for "nr", which means the zero-filled
908 * part of the page is not copied back to userspace (unless
909 * another truncate extends the file - this is desired though).
910 */
911
912 isize = i_size_read(inode);
913 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
914 if (unlikely(!isize || index > end_index)) {
915 page_cache_release(page);
916 goto out;
917 }
918
919 /* nr is the maximum number of bytes to copy from this page */
920 nr = PAGE_CACHE_SIZE;
921 if (index == end_index) {
922 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
923 if (nr <= offset) {
924 page_cache_release(page);
925 goto out;
926 }
927 }
928 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 /* If users can be writing to this page using arbitrary
931 * virtual addresses, take care about potential aliasing
932 * before reading the page on the kernel side.
933 */
934 if (mapping_writably_mapped(mapping))
935 flush_dcache_page(page);
936
937 /*
Jan Karaec0f1632007-05-06 14:49:25 -0700938 * When a sequential read accesses a page several times,
939 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 */
Jan Karaec0f1632007-05-06 14:49:25 -0700941 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 mark_page_accessed(page);
943 prev_index = index;
944
945 /*
946 * Ok, we have the page, and it's up-to-date, so
947 * now we can copy it to user space...
948 *
949 * The actor routine returns how many bytes were actually used..
950 * NOTE! This may not be the same as how much of a user buffer
951 * we filled up (we may be padding etc), so we can only update
952 * "pos" here (the actor routine has to update the user buffer
953 * pointers and the remaining count).
954 */
955 ret = actor(desc, page, offset, nr);
956 offset += ret;
957 index += offset >> PAGE_CACHE_SHIFT;
958 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -0700959 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
961 page_cache_release(page);
962 if (ret == nr && desc->count)
963 continue;
964 goto out;
965
966page_not_up_to_date:
967 /* Get exclusive access to the page ... */
968 lock_page(page);
969
Nick Pigginda6052f2006-09-25 23:31:35 -0700970 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 if (!page->mapping) {
972 unlock_page(page);
973 page_cache_release(page);
974 continue;
975 }
976
977 /* Did somebody else fill it already? */
978 if (PageUptodate(page)) {
979 unlock_page(page);
980 goto page_ok;
981 }
982
983readpage:
984 /* Start the actual read. The read will unlock the page. */
985 error = mapping->a_ops->readpage(filp, page);
986
Zach Brown994fc28c2005-12-15 14:28:17 -0800987 if (unlikely(error)) {
988 if (error == AOP_TRUNCATED_PAGE) {
989 page_cache_release(page);
990 goto find_page;
991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -0800993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 if (!PageUptodate(page)) {
996 lock_page(page);
997 if (!PageUptodate(page)) {
998 if (page->mapping == NULL) {
999 /*
1000 * invalidate_inode_pages got it
1001 */
1002 unlock_page(page);
1003 page_cache_release(page);
1004 goto find_page;
1005 }
1006 unlock_page(page);
1007 error = -EIO;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001008 shrink_readahead_size_eio(filp, ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 goto readpage_error;
1010 }
1011 unlock_page(page);
1012 }
1013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 goto page_ok;
1015
1016readpage_error:
1017 /* UHHUH! A synchronous read error occurred. Report it */
1018 desc->error = error;
1019 page_cache_release(page);
1020 goto out;
1021
1022no_cached_page:
1023 /*
1024 * Ok, it wasn't cached, so we need to create a new
1025 * page..
1026 */
Nick Piggineb2be182007-10-16 01:24:57 -07001027 page = page_cache_alloc_cold(mapping);
1028 if (!page) {
1029 desc->error = -ENOMEM;
1030 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 }
Nick Piggineb2be182007-10-16 01:24:57 -07001032 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 index, GFP_KERNEL);
1034 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001035 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (error == -EEXIST)
1037 goto find_page;
1038 desc->error = error;
1039 goto out;
1040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 goto readpage;
1042 }
1043
1044out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001045 ra->prev_pos = prev_index;
1046 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1047 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001049 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (filp)
1051 file_accessed(filp);
1052}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053EXPORT_SYMBOL(do_generic_mapping_read);
1054
1055int file_read_actor(read_descriptor_t *desc, struct page *page,
1056 unsigned long offset, unsigned long size)
1057{
1058 char *kaddr;
1059 unsigned long left, count = desc->count;
1060
1061 if (size > count)
1062 size = count;
1063
1064 /*
1065 * Faults on the destination of a read are common, so do it before
1066 * taking the kmap.
1067 */
1068 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1069 kaddr = kmap_atomic(page, KM_USER0);
1070 left = __copy_to_user_inatomic(desc->arg.buf,
1071 kaddr + offset, size);
1072 kunmap_atomic(kaddr, KM_USER0);
1073 if (left == 0)
1074 goto success;
1075 }
1076
1077 /* Do it the slow way */
1078 kaddr = kmap(page);
1079 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1080 kunmap(page);
1081
1082 if (left) {
1083 size -= left;
1084 desc->error = -EFAULT;
1085 }
1086success:
1087 desc->count = count - size;
1088 desc->written += size;
1089 desc->arg.buf += size;
1090 return size;
1091}
1092
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001093/*
1094 * Performs necessary checks before doing a write
1095 * @iov: io vector request
1096 * @nr_segs: number of segments in the iovec
1097 * @count: number of bytes to write
1098 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1099 *
1100 * Adjust number of segments and amount of bytes to write (nr_segs should be
1101 * properly initialized first). Returns appropriate error code that caller
1102 * should return or zero in case that write should be allowed.
1103 */
1104int generic_segment_checks(const struct iovec *iov,
1105 unsigned long *nr_segs, size_t *count, int access_flags)
1106{
1107 unsigned long seg;
1108 size_t cnt = 0;
1109 for (seg = 0; seg < *nr_segs; seg++) {
1110 const struct iovec *iv = &iov[seg];
1111
1112 /*
1113 * If any segment has a negative length, or the cumulative
1114 * length ever wraps negative then return -EINVAL.
1115 */
1116 cnt += iv->iov_len;
1117 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1118 return -EINVAL;
1119 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1120 continue;
1121 if (seg == 0)
1122 return -EFAULT;
1123 *nr_segs = seg;
1124 cnt -= iv->iov_len; /* This segment is no good */
1125 break;
1126 }
1127 *count = cnt;
1128 return 0;
1129}
1130EXPORT_SYMBOL(generic_segment_checks);
1131
Randy Dunlap485bb992006-06-23 02:03:49 -07001132/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001133 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001134 * @iocb: kernel I/O control block
1135 * @iov: io vector request
1136 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001137 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001138 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 * This is the "read()" routine for all filesystems
1140 * that can use the page cache directly.
1141 */
1142ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001143generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1144 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145{
1146 struct file *filp = iocb->ki_filp;
1147 ssize_t retval;
1148 unsigned long seg;
1149 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001150 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
1152 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001153 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1154 if (retval)
1155 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
1157 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1158 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001159 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 struct address_space *mapping;
1161 struct inode *inode;
1162
1163 mapping = filp->f_mapping;
1164 inode = mapping->host;
1165 retval = 0;
1166 if (!count)
1167 goto out; /* skip atime */
1168 size = i_size_read(inode);
1169 if (pos < size) {
1170 retval = generic_file_direct_IO(READ, iocb,
1171 iov, pos, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (retval > 0)
1173 *ppos = pos + retval;
1174 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001175 if (likely(retval != 0)) {
Steven Whitehouse3f1a9aa2006-09-27 14:52:48 -04001176 file_accessed(filp);
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001177 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 }
1180
1181 retval = 0;
1182 if (count) {
1183 for (seg = 0; seg < nr_segs; seg++) {
1184 read_descriptor_t desc;
1185
1186 desc.written = 0;
1187 desc.arg.buf = iov[seg].iov_base;
1188 desc.count = iov[seg].iov_len;
1189 if (desc.count == 0)
1190 continue;
1191 desc.error = 0;
1192 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1193 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001194 if (desc.error) {
1195 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 break;
1197 }
akpm@linux-foundation.orgc44939e2007-07-15 23:38:25 -07001198 if (desc.count > 0)
1199 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
1201 }
1202out:
1203 return retval;
1204}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205EXPORT_SYMBOL(generic_file_aio_read);
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207static ssize_t
1208do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001209 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
1211 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1212 return -EINVAL;
1213
1214 force_page_cache_readahead(mapping, filp, index,
1215 max_sane_readahead(nr));
1216 return 0;
1217}
1218
1219asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1220{
1221 ssize_t ret;
1222 struct file *file;
1223
1224 ret = -EBADF;
1225 file = fget(fd);
1226 if (file) {
1227 if (file->f_mode & FMODE_READ) {
1228 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001229 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1230 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 unsigned long len = end - start + 1;
1232 ret = do_readahead(mapping, file, start, len);
1233 }
1234 fput(file);
1235 }
1236 return ret;
1237}
1238
1239#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001240/**
1241 * page_cache_read - adds requested page to the page cache if not already there
1242 * @file: file to read
1243 * @offset: page index
1244 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 * This adds the requested page to the page cache if it isn't already there,
1246 * and schedules an I/O to read in its contents from disk.
1247 */
Fengguang Wu57f6b962007-10-16 01:24:37 -07001248static int fastcall page_cache_read(struct file * file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249{
1250 struct address_space *mapping = file->f_mapping;
1251 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001252 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Zach Brown994fc28c2005-12-15 14:28:17 -08001254 do {
1255 page = page_cache_alloc_cold(mapping);
1256 if (!page)
1257 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Zach Brown994fc28c2005-12-15 14:28:17 -08001259 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1260 if (ret == 0)
1261 ret = mapping->a_ops->readpage(file, page);
1262 else if (ret == -EEXIST)
1263 ret = 0; /* losing race to add is OK */
1264
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Zach Brown994fc28c2005-12-15 14:28:17 -08001267 } while (ret == AOP_TRUNCATED_PAGE);
1268
1269 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
1271
1272#define MMAP_LOTSAMISS (100)
1273
Randy Dunlap485bb992006-06-23 02:03:49 -07001274/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001275 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001276 * @vma: vma in which the fault was taken
1277 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001278 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001279 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 * mapped memory region to read in file data during a page fault.
1281 *
1282 * The goto's are kind of ugly, but this streamlines the normal case of having
1283 * it in the page cache, and handles the special cases reasonably without
1284 * having a lot of duplicated code.
1285 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001286int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287{
1288 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001289 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 struct address_space *mapping = file->f_mapping;
1291 struct file_ra_state *ra = &file->f_ra;
1292 struct inode *inode = mapping->host;
1293 struct page *page;
Nick Piggin54cb8822007-07-19 01:46:59 -07001294 unsigned long size;
1295 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001296 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001299 if (vmf->pgoff >= size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 goto outside_data_content;
1301
1302 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001303 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 goto no_cached_page;
1305
1306 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 * Do we have something in the page cache already?
1308 */
1309retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001310 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001311 /*
1312 * For sequential accesses, we use the generic readahead logic.
1313 */
1314 if (VM_SequentialReadHint(vma)) {
1315 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001316 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001317 vmf->pgoff, 1);
1318 page = find_lock_page(mapping, vmf->pgoff);
1319 if (!page)
1320 goto no_cached_page;
1321 }
1322 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001323 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001324 vmf->pgoff, 1);
1325 }
1326 }
1327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 if (!page) {
1329 unsigned long ra_pages;
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 ra->mmap_miss++;
1332
1333 /*
1334 * Do we miss much more than hit in this file? If so,
1335 * stop bothering with read-ahead. It will only hurt.
1336 */
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001337 if (ra->mmap_miss > MMAP_LOTSAMISS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 goto no_cached_page;
1339
1340 /*
1341 * To keep the pgmajfault counter straight, we need to
1342 * check did_readaround, as this is an inner loop.
1343 */
1344 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001345 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001346 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 }
1348 did_readaround = 1;
1349 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1350 if (ra_pages) {
1351 pgoff_t start = 0;
1352
Nick Piggind0217ac2007-07-19 01:47:03 -07001353 if (vmf->pgoff > ra_pages / 2)
1354 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 do_page_cache_readahead(mapping, file, start, ra_pages);
1356 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001357 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 if (!page)
1359 goto no_cached_page;
1360 }
1361
1362 if (!did_readaround)
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001363 ra->mmap_miss--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001366 * We have a locked page in the page cache, now we need to check
1367 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 */
Nick Piggind00806b2007-07-19 01:46:57 -07001369 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 goto page_not_uptodate;
1371
Nick Piggind00806b2007-07-19 01:46:57 -07001372 /* Must recheck i_size under page lock */
1373 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001374 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001375 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001376 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001377 goto outside_data_content;
1378 }
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 /*
1381 * Found the page and have a reference on it.
1382 */
1383 mark_page_accessed(page);
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001384 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001385 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001386 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
1388outside_data_content:
1389 /*
1390 * An external ptracer can access pages that normally aren't
1391 * accessible..
1392 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001393 if (vma->vm_mm == current->mm)
1394 return VM_FAULT_SIGBUS;
1395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 /* Fall through to the non-read-ahead case */
1397no_cached_page:
1398 /*
1399 * We're only likely to ever get here if MADV_RANDOM is in
1400 * effect.
1401 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001402 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
1404 /*
1405 * The page we want has now been added to the page cache.
1406 * In the unlikely event that someone removed it in the
1407 * meantime, we'll just come back here and read it again.
1408 */
1409 if (error >= 0)
1410 goto retry_find;
1411
1412 /*
1413 * An error return from page_cache_read can result if the
1414 * system is low on memory, or a problem occurs while trying
1415 * to schedule I/O.
1416 */
1417 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001418 return VM_FAULT_OOM;
1419 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
1421page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001422 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001424 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001425 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 /*
1429 * Umm, take care of errors if the page isn't up-to-date.
1430 * Try to re-read it _once_. We do this synchronously,
1431 * because there really aren't any performance issues here
1432 * and we need to check for errors.
1433 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001435 error = mapping->a_ops->readpage(file, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001437
1438 if (!error || error == AOP_TRUNCATED_PAGE)
1439 goto retry_find;
1440
1441 /* Things didn't work out. Return zero to tell the mm layer so. */
1442 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001443 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001444}
1445EXPORT_SYMBOL(filemap_fault);
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001448 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449};
1450
1451/* This is used for a general mmap of a disk file */
1452
1453int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1454{
1455 struct address_space *mapping = file->f_mapping;
1456
1457 if (!mapping->a_ops->readpage)
1458 return -ENOEXEC;
1459 file_accessed(file);
1460 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001461 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 return 0;
1463}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465/*
1466 * This is for filesystems which do not implement ->writepage.
1467 */
1468int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1469{
1470 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1471 return -EINVAL;
1472 return generic_file_mmap(file, vma);
1473}
1474#else
1475int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1476{
1477 return -ENOSYS;
1478}
1479int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1480{
1481 return -ENOSYS;
1482}
1483#endif /* CONFIG_MMU */
1484
1485EXPORT_SYMBOL(generic_file_mmap);
1486EXPORT_SYMBOL(generic_file_readonly_mmap);
1487
Nick Piggin6fe69002007-05-06 14:49:04 -07001488static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001489 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 int (*filler)(void *,struct page*),
1491 void *data)
1492{
Nick Piggineb2be182007-10-16 01:24:57 -07001493 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 int err;
1495repeat:
1496 page = find_get_page(mapping, index);
1497 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001498 page = page_cache_alloc_cold(mapping);
1499 if (!page)
1500 return ERR_PTR(-ENOMEM);
1501 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1502 if (unlikely(err)) {
1503 page_cache_release(page);
1504 if (err == -EEXIST)
1505 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 return ERR_PTR(err);
1508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 err = filler(data, page);
1510 if (err < 0) {
1511 page_cache_release(page);
1512 page = ERR_PTR(err);
1513 }
1514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 return page;
1516}
1517
Nick Piggin6fe69002007-05-06 14:49:04 -07001518/*
1519 * Same as read_cache_page, but don't wait for page to become unlocked
1520 * after submitting it to the filler.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001522struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001523 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 int (*filler)(void *,struct page*),
1525 void *data)
1526{
1527 struct page *page;
1528 int err;
1529
1530retry:
1531 page = __read_cache_page(mapping, index, filler, data);
1532 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001533 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 if (PageUptodate(page))
1535 goto out;
1536
1537 lock_page(page);
1538 if (!page->mapping) {
1539 unlock_page(page);
1540 page_cache_release(page);
1541 goto retry;
1542 }
1543 if (PageUptodate(page)) {
1544 unlock_page(page);
1545 goto out;
1546 }
1547 err = filler(data, page);
1548 if (err < 0) {
1549 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001550 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 }
David Howellsc855ff32007-05-09 13:42:20 +01001552out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001553 mark_page_accessed(page);
1554 return page;
1555}
1556EXPORT_SYMBOL(read_cache_page_async);
1557
1558/**
1559 * read_cache_page - read into page cache, fill it if needed
1560 * @mapping: the page's address_space
1561 * @index: the page index
1562 * @filler: function to perform the read
1563 * @data: destination for read data
1564 *
1565 * Read into the page cache. If a page already exists, and PageUptodate() is
1566 * not set, try to fill the page then wait for it to become unlocked.
1567 *
1568 * If the page does not get brought uptodate, return -EIO.
1569 */
1570struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001571 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001572 int (*filler)(void *,struct page*),
1573 void *data)
1574{
1575 struct page *page;
1576
1577 page = read_cache_page_async(mapping, index, filler, data);
1578 if (IS_ERR(page))
1579 goto out;
1580 wait_on_page_locked(page);
1581 if (!PageUptodate(page)) {
1582 page_cache_release(page);
1583 page = ERR_PTR(-EIO);
1584 }
1585 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 return page;
1587}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588EXPORT_SYMBOL(read_cache_page);
1589
1590/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 * The logic we want is
1592 *
1593 * if suid or (sgid and xgrp)
1594 * remove privs
1595 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001596int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
1598 mode_t mode = dentry->d_inode->i_mode;
1599 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 /* suid always must be killed */
1602 if (unlikely(mode & S_ISUID))
1603 kill = ATTR_KILL_SUID;
1604
1605 /*
1606 * sgid without any exec bits is just a mandatory locking mark; leave
1607 * it alone. If some exec bits are set, it's a real sgid; kill it.
1608 */
1609 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1610 kill |= ATTR_KILL_SGID;
1611
Jens Axboe01de85e2006-10-17 19:50:36 +02001612 if (unlikely(kill && !capable(CAP_FSETID)))
1613 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
Jens Axboe01de85e2006-10-17 19:50:36 +02001615 return 0;
1616}
Mark Fashehd23a1472006-10-17 17:05:18 -07001617EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001618
1619int __remove_suid(struct dentry *dentry, int kill)
1620{
1621 struct iattr newattrs;
1622
1623 newattrs.ia_valid = ATTR_FORCE | kill;
1624 return notify_change(dentry, &newattrs);
1625}
1626
1627int remove_suid(struct dentry *dentry)
1628{
1629 int kill = should_remove_suid(dentry);
1630
1631 if (unlikely(kill))
1632 return __remove_suid(dentry, kill);
1633
1634 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635}
1636EXPORT_SYMBOL(remove_suid);
1637
Nick Piggin2f718ff2007-10-16 01:24:59 -07001638static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 const struct iovec *iov, size_t base, size_t bytes)
1640{
1641 size_t copied = 0, left = 0;
1642
1643 while (bytes) {
1644 char __user *buf = iov->iov_base + base;
1645 int copy = min(bytes, iov->iov_len - base);
1646
1647 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001648 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 copied += copy;
1650 bytes -= copy;
1651 vaddr += copy;
1652 iov++;
1653
NeilBrown01408c42006-06-25 05:47:58 -07001654 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 }
1657 return copied - left;
1658}
1659
1660/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001661 * Copy as much as we can into the page and return the number of bytes which
1662 * were sucessfully copied. If a fault is encountered then return the number of
1663 * bytes which were copied.
1664 */
1665size_t iov_iter_copy_from_user_atomic(struct page *page,
1666 struct iov_iter *i, unsigned long offset, size_t bytes)
1667{
1668 char *kaddr;
1669 size_t copied;
1670
1671 BUG_ON(!in_atomic());
1672 kaddr = kmap_atomic(page, KM_USER0);
1673 if (likely(i->nr_segs == 1)) {
1674 int left;
1675 char __user *buf = i->iov->iov_base + i->iov_offset;
1676 left = __copy_from_user_inatomic_nocache(kaddr + offset,
1677 buf, bytes);
1678 copied = bytes - left;
1679 } else {
1680 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1681 i->iov, i->iov_offset, bytes);
1682 }
1683 kunmap_atomic(kaddr, KM_USER0);
1684
1685 return copied;
1686}
Nick Piggin89e10782007-10-16 01:25:07 -07001687EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001688
1689/*
1690 * This has the same sideeffects and return value as
1691 * iov_iter_copy_from_user_atomic().
1692 * The difference is that it attempts to resolve faults.
1693 * Page must not be locked.
1694 */
1695size_t iov_iter_copy_from_user(struct page *page,
1696 struct iov_iter *i, unsigned long offset, size_t bytes)
1697{
1698 char *kaddr;
1699 size_t copied;
1700
1701 kaddr = kmap(page);
1702 if (likely(i->nr_segs == 1)) {
1703 int left;
1704 char __user *buf = i->iov->iov_base + i->iov_offset;
1705 left = __copy_from_user_nocache(kaddr + offset, buf, bytes);
1706 copied = bytes - left;
1707 } else {
1708 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1709 i->iov, i->iov_offset, bytes);
1710 }
1711 kunmap(page);
1712 return copied;
1713}
Nick Piggin89e10782007-10-16 01:25:07 -07001714EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001715
1716static void __iov_iter_advance_iov(struct iov_iter *i, size_t bytes)
1717{
1718 if (likely(i->nr_segs == 1)) {
1719 i->iov_offset += bytes;
1720 } else {
1721 const struct iovec *iov = i->iov;
1722 size_t base = i->iov_offset;
1723
1724 while (bytes) {
1725 int copy = min(bytes, iov->iov_len - base);
1726
1727 bytes -= copy;
1728 base += copy;
1729 if (iov->iov_len == base) {
1730 iov++;
1731 base = 0;
1732 }
1733 }
1734 i->iov = iov;
1735 i->iov_offset = base;
1736 }
1737}
1738
1739void iov_iter_advance(struct iov_iter *i, size_t bytes)
1740{
1741 BUG_ON(i->count < bytes);
1742
1743 __iov_iter_advance_iov(i, bytes);
1744 i->count -= bytes;
1745}
Nick Piggin89e10782007-10-16 01:25:07 -07001746EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001747
Nick Pigginafddba42007-10-16 01:25:01 -07001748/*
1749 * Fault in the first iovec of the given iov_iter, to a maximum length
1750 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1751 * accessed (ie. because it is an invalid address).
1752 *
1753 * writev-intensive code may want this to prefault several iovecs -- that
1754 * would be possible (callers must not rely on the fact that _only_ the
1755 * first iovec will be faulted with the current implementation).
1756 */
1757int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001758{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001759 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001760 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1761 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001762}
Nick Piggin89e10782007-10-16 01:25:07 -07001763EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001764
1765/*
1766 * Return the count of just the current iov_iter segment.
1767 */
1768size_t iov_iter_single_seg_count(struct iov_iter *i)
1769{
1770 const struct iovec *iov = i->iov;
1771 if (i->nr_segs == 1)
1772 return i->count;
1773 else
1774 return min(i->count, iov->iov_len - i->iov_offset);
1775}
Nick Piggin89e10782007-10-16 01:25:07 -07001776EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001777
1778/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 * Performs necessary checks before doing a write
1780 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001781 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 * Returns appropriate error code that caller should return or
1783 * zero in case that write should be allowed.
1784 */
1785inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1786{
1787 struct inode *inode = file->f_mapping->host;
1788 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1789
1790 if (unlikely(*pos < 0))
1791 return -EINVAL;
1792
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 if (!isblk) {
1794 /* FIXME: this is for backwards compatibility with 2.4 */
1795 if (file->f_flags & O_APPEND)
1796 *pos = i_size_read(inode);
1797
1798 if (limit != RLIM_INFINITY) {
1799 if (*pos >= limit) {
1800 send_sig(SIGXFSZ, current, 0);
1801 return -EFBIG;
1802 }
1803 if (*count > limit - (typeof(limit))*pos) {
1804 *count = limit - (typeof(limit))*pos;
1805 }
1806 }
1807 }
1808
1809 /*
1810 * LFS rule
1811 */
1812 if (unlikely(*pos + *count > MAX_NON_LFS &&
1813 !(file->f_flags & O_LARGEFILE))) {
1814 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 return -EFBIG;
1816 }
1817 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1818 *count = MAX_NON_LFS - (unsigned long)*pos;
1819 }
1820 }
1821
1822 /*
1823 * Are we about to exceed the fs block limit ?
1824 *
1825 * If we have written data it becomes a short write. If we have
1826 * exceeded without writing data we send a signal and return EFBIG.
1827 * Linus frestrict idea will clean these up nicely..
1828 */
1829 if (likely(!isblk)) {
1830 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1831 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 return -EFBIG;
1833 }
1834 /* zero-length writes at ->s_maxbytes are OK */
1835 }
1836
1837 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1838 *count = inode->i_sb->s_maxbytes - *pos;
1839 } else {
David Howells93614012006-09-30 20:45:40 +02001840#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 loff_t isize;
1842 if (bdev_read_only(I_BDEV(inode)))
1843 return -EPERM;
1844 isize = i_size_read(inode);
1845 if (*pos >= isize) {
1846 if (*count || *pos > isize)
1847 return -ENOSPC;
1848 }
1849
1850 if (*pos + *count > isize)
1851 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02001852#else
1853 return -EPERM;
1854#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 }
1856 return 0;
1857}
1858EXPORT_SYMBOL(generic_write_checks);
1859
Nick Pigginafddba42007-10-16 01:25:01 -07001860int pagecache_write_begin(struct file *file, struct address_space *mapping,
1861 loff_t pos, unsigned len, unsigned flags,
1862 struct page **pagep, void **fsdata)
1863{
1864 const struct address_space_operations *aops = mapping->a_ops;
1865
1866 if (aops->write_begin) {
1867 return aops->write_begin(file, mapping, pos, len, flags,
1868 pagep, fsdata);
1869 } else {
1870 int ret;
1871 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1872 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1873 struct inode *inode = mapping->host;
1874 struct page *page;
1875again:
1876 page = __grab_cache_page(mapping, index);
1877 *pagep = page;
1878 if (!page)
1879 return -ENOMEM;
1880
1881 if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) {
1882 /*
1883 * There is no way to resolve a short write situation
1884 * for a !Uptodate page (except by double copying in
1885 * the caller done by generic_perform_write_2copy).
1886 *
1887 * Instead, we have to bring it uptodate here.
1888 */
1889 ret = aops->readpage(file, page);
1890 page_cache_release(page);
1891 if (ret) {
1892 if (ret == AOP_TRUNCATED_PAGE)
1893 goto again;
1894 return ret;
1895 }
1896 goto again;
1897 }
1898
1899 ret = aops->prepare_write(file, page, offset, offset+len);
1900 if (ret) {
1901 if (ret != AOP_TRUNCATED_PAGE)
1902 unlock_page(page);
1903 page_cache_release(page);
1904 if (pos + len > inode->i_size)
1905 vmtruncate(inode, inode->i_size);
1906 if (ret == AOP_TRUNCATED_PAGE)
1907 goto again;
1908 }
1909 return ret;
1910 }
1911}
1912EXPORT_SYMBOL(pagecache_write_begin);
1913
1914int pagecache_write_end(struct file *file, struct address_space *mapping,
1915 loff_t pos, unsigned len, unsigned copied,
1916 struct page *page, void *fsdata)
1917{
1918 const struct address_space_operations *aops = mapping->a_ops;
1919 int ret;
1920
1921 if (aops->write_end) {
1922 mark_page_accessed(page);
1923 ret = aops->write_end(file, mapping, pos, len, copied,
1924 page, fsdata);
1925 } else {
1926 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1927 struct inode *inode = mapping->host;
1928
1929 flush_dcache_page(page);
1930 ret = aops->commit_write(file, page, offset, offset+len);
1931 unlock_page(page);
1932 mark_page_accessed(page);
1933 page_cache_release(page);
1934 BUG_ON(ret == AOP_TRUNCATED_PAGE); /* can't deal with */
1935
1936 if (ret < 0) {
1937 if (pos + len > inode->i_size)
1938 vmtruncate(inode, inode->i_size);
1939 } else if (ret > 0)
1940 ret = min_t(size_t, copied, ret);
1941 else
1942 ret = copied;
1943 }
1944
1945 return ret;
1946}
1947EXPORT_SYMBOL(pagecache_write_end);
1948
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949ssize_t
1950generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1951 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1952 size_t count, size_t ocount)
1953{
1954 struct file *file = iocb->ki_filp;
1955 struct address_space *mapping = file->f_mapping;
1956 struct inode *inode = mapping->host;
1957 ssize_t written;
1958
1959 if (count != ocount)
1960 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1961
1962 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1963 if (written > 0) {
1964 loff_t end = pos + written;
1965 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1966 i_size_write(inode, end);
1967 mark_inode_dirty(inode);
1968 }
1969 *ppos = end;
1970 }
1971
1972 /*
1973 * Sync the fs metadata but not the minor inode changes and
1974 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001975 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08001976 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 */
Zach Brown8459d862006-12-10 02:21:05 -08001978 if ((written >= 0 || written == -EIOCBQUEUED) &&
1979 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07001980 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1981 if (err < 0)
1982 written = err;
1983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 return written;
1985}
1986EXPORT_SYMBOL(generic_file_direct_write);
1987
Nick Piggineb2be182007-10-16 01:24:57 -07001988/*
1989 * Find or create a page at the given pagecache position. Return the locked
1990 * page. This function is specifically for buffered writes.
1991 */
Nick Pigginafddba42007-10-16 01:25:01 -07001992struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index)
Nick Piggineb2be182007-10-16 01:24:57 -07001993{
1994 int status;
1995 struct page *page;
1996repeat:
1997 page = find_lock_page(mapping, index);
1998 if (likely(page))
1999 return page;
2000
2001 page = page_cache_alloc(mapping);
2002 if (!page)
2003 return NULL;
2004 status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
2005 if (unlikely(status)) {
2006 page_cache_release(page);
2007 if (status == -EEXIST)
2008 goto repeat;
2009 return NULL;
2010 }
2011 return page;
2012}
Nick Pigginafddba42007-10-16 01:25:01 -07002013EXPORT_SYMBOL(__grab_cache_page);
Nick Piggineb2be182007-10-16 01:24:57 -07002014
Nick Pigginafddba42007-10-16 01:25:01 -07002015static ssize_t generic_perform_write_2copy(struct file *file,
2016 struct iov_iter *i, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
Andrew Mortonae374612007-10-16 01:24:55 -07002018 struct address_space *mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002019 const struct address_space_operations *a_ops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002020 struct inode *inode = mapping->host;
2021 long status = 0;
2022 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
2024 do {
Nick Piggin08291422007-10-16 01:24:59 -07002025 struct page *src_page;
Nick Piggineb2be182007-10-16 01:24:57 -07002026 struct page *page;
Andrew Mortonae374612007-10-16 01:24:55 -07002027 pgoff_t index; /* Pagecache index for current page */
2028 unsigned long offset; /* Offset into pagecache page */
Nick Piggin08291422007-10-16 01:24:59 -07002029 unsigned long bytes; /* Bytes to write to page */
Andrew Mortonae374612007-10-16 01:24:55 -07002030 size_t copied; /* Bytes copied from user */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
Andrew Mortonae374612007-10-16 01:24:55 -07002032 offset = (pos & (PAGE_CACHE_SIZE - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002034 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002035 iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
Nick Piggin08291422007-10-16 01:24:59 -07002037 /*
2038 * a non-NULL src_page indicates that we're doing the
2039 * copy via get_user_pages and kmap.
2040 */
2041 src_page = NULL;
Andrew Mortonae374612007-10-16 01:24:55 -07002042
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002043 /*
2044 * Bring in the user page that we will copy from _first_.
2045 * Otherwise there's a nasty deadlock on copying from the
2046 * same page as we're writing to, without it being marked
2047 * up-to-date.
Nick Piggin08291422007-10-16 01:24:59 -07002048 *
2049 * Not only is this an optimisation, but it is also required
2050 * to check that the address is actually valid, when atomic
2051 * usercopies are used, below.
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002052 */
Nick Pigginafddba42007-10-16 01:25:01 -07002053 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
Nick Piggin08291422007-10-16 01:24:59 -07002054 status = -EFAULT;
2055 break;
2056 }
Nick Piggineb2be182007-10-16 01:24:57 -07002057
2058 page = __grab_cache_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 if (!page) {
2060 status = -ENOMEM;
2061 break;
2062 }
2063
Nick Piggin08291422007-10-16 01:24:59 -07002064 /*
2065 * non-uptodate pages cannot cope with short copies, and we
2066 * cannot take a pagefault with the destination page locked.
2067 * So pin the source page to copy it.
2068 */
Nick Piggin674b8922007-10-16 01:25:03 -07002069 if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) {
Nick Piggin08291422007-10-16 01:24:59 -07002070 unlock_page(page);
2071
2072 src_page = alloc_page(GFP_KERNEL);
2073 if (!src_page) {
2074 page_cache_release(page);
2075 status = -ENOMEM;
2076 break;
2077 }
2078
2079 /*
2080 * Cannot get_user_pages with a page locked for the
2081 * same reason as we can't take a page fault with a
2082 * page locked (as explained below).
2083 */
Nick Pigginafddba42007-10-16 01:25:01 -07002084 copied = iov_iter_copy_from_user(src_page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002085 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002086 if (unlikely(copied == 0)) {
2087 status = -EFAULT;
2088 page_cache_release(page);
2089 page_cache_release(src_page);
2090 break;
2091 }
2092 bytes = copied;
2093
2094 lock_page(page);
2095 /*
2096 * Can't handle the page going uptodate here, because
2097 * that means we would use non-atomic usercopies, which
2098 * zero out the tail of the page, which can cause
2099 * zeroes to become transiently visible. We could just
2100 * use a non-zeroing copy, but the APIs aren't too
2101 * consistent.
2102 */
2103 if (unlikely(!page->mapping || PageUptodate(page))) {
2104 unlock_page(page);
2105 page_cache_release(page);
2106 page_cache_release(src_page);
2107 continue;
2108 }
Nick Piggin08291422007-10-16 01:24:59 -07002109 }
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 status = a_ops->prepare_write(file, page, offset, offset+bytes);
Nick Piggin64649a52007-10-16 01:24:56 -07002112 if (unlikely(status))
2113 goto fs_write_aop_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08002114
Nick Piggin08291422007-10-16 01:24:59 -07002115 if (!src_page) {
2116 /*
2117 * Must not enter the pagefault handler here, because
2118 * we hold the page lock, so we might recursively
2119 * deadlock on the same lock, or get an ABBA deadlock
2120 * against a different lock, or against the mmap_sem
2121 * (which nests outside the page lock). So increment
2122 * preempt count, and use _atomic usercopies.
2123 *
2124 * The page is uptodate so we are OK to encounter a
2125 * short copy: if unmodified parts of the page are
2126 * marked dirty and written out to disk, it doesn't
2127 * really matter.
2128 */
2129 pagefault_disable();
Nick Pigginafddba42007-10-16 01:25:01 -07002130 copied = iov_iter_copy_from_user_atomic(page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002131 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002132 pagefault_enable();
2133 } else {
2134 void *src, *dst;
2135 src = kmap_atomic(src_page, KM_USER0);
2136 dst = kmap_atomic(page, KM_USER1);
2137 memcpy(dst + offset, src + offset, bytes);
2138 kunmap_atomic(dst, KM_USER1);
2139 kunmap_atomic(src, KM_USER0);
2140 copied = bytes;
2141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 flush_dcache_page(page);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002143
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 status = a_ops->commit_write(file, page, offset, offset+bytes);
Nick Piggin64649a52007-10-16 01:24:56 -07002145 if (unlikely(status < 0 || status == AOP_TRUNCATED_PAGE))
2146 goto fs_write_aop_error;
Nick Piggin64649a52007-10-16 01:24:56 -07002147 if (unlikely(status > 0)) /* filesystem did partial write */
Nick Piggin08291422007-10-16 01:24:59 -07002148 copied = min_t(size_t, copied, status);
2149
2150 unlock_page(page);
2151 mark_page_accessed(page);
2152 page_cache_release(page);
2153 if (src_page)
2154 page_cache_release(src_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
Nick Pigginafddba42007-10-16 01:25:01 -07002156 iov_iter_advance(i, copied);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002157 pos += copied;
Nick Pigginafddba42007-10-16 01:25:01 -07002158 written += copied;
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 balance_dirty_pages_ratelimited(mapping);
2161 cond_resched();
Nick Piggin64649a52007-10-16 01:24:56 -07002162 continue;
2163
2164fs_write_aop_error:
2165 if (status != AOP_TRUNCATED_PAGE)
2166 unlock_page(page);
2167 page_cache_release(page);
Nick Piggin08291422007-10-16 01:24:59 -07002168 if (src_page)
2169 page_cache_release(src_page);
Nick Piggin64649a52007-10-16 01:24:56 -07002170
2171 /*
2172 * prepare_write() may have instantiated a few blocks
2173 * outside i_size. Trim these off again. Don't need
2174 * i_size_read because we hold i_mutex.
2175 */
2176 if (pos + bytes > inode->i_size)
2177 vmtruncate(inode, inode->i_size);
2178 if (status == AOP_TRUNCATED_PAGE)
2179 continue;
2180 else
2181 break;
Nick Pigginafddba42007-10-16 01:25:01 -07002182 } while (iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
Nick Pigginafddba42007-10-16 01:25:01 -07002184 return written ? written : status;
2185}
2186
2187static ssize_t generic_perform_write(struct file *file,
2188 struct iov_iter *i, loff_t pos)
2189{
2190 struct address_space *mapping = file->f_mapping;
2191 const struct address_space_operations *a_ops = mapping->a_ops;
2192 long status = 0;
2193 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002194 unsigned int flags = 0;
2195
2196 /*
2197 * Copies from kernel address space cannot fail (NFSD is a big user).
2198 */
2199 if (segment_eq(get_fs(), KERNEL_DS))
2200 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002201
2202 do {
2203 struct page *page;
2204 pgoff_t index; /* Pagecache index for current page */
2205 unsigned long offset; /* Offset into pagecache page */
2206 unsigned long bytes; /* Bytes to write to page */
2207 size_t copied; /* Bytes copied from user */
2208 void *fsdata;
2209
2210 offset = (pos & (PAGE_CACHE_SIZE - 1));
2211 index = pos >> PAGE_CACHE_SHIFT;
2212 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2213 iov_iter_count(i));
2214
2215again:
2216
2217 /*
2218 * Bring in the user page that we will copy from _first_.
2219 * Otherwise there's a nasty deadlock on copying from the
2220 * same page as we're writing to, without it being marked
2221 * up-to-date.
2222 *
2223 * Not only is this an optimisation, but it is also required
2224 * to check that the address is actually valid, when atomic
2225 * usercopies are used, below.
2226 */
2227 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2228 status = -EFAULT;
2229 break;
2230 }
2231
Nick Piggin674b8922007-10-16 01:25:03 -07002232 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002233 &page, &fsdata);
2234 if (unlikely(status))
2235 break;
2236
2237 pagefault_disable();
2238 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2239 pagefault_enable();
2240 flush_dcache_page(page);
2241
2242 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2243 page, fsdata);
2244 if (unlikely(status < 0))
2245 break;
2246 copied = status;
2247
2248 cond_resched();
2249
2250 if (unlikely(copied == 0)) {
2251 /*
2252 * If we were unable to copy any data at all, we must
2253 * fall back to a single segment length write.
2254 *
2255 * If we didn't fallback here, we could livelock
2256 * because not all segments in the iov can be copied at
2257 * once without a pagefault.
2258 */
2259 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2260 iov_iter_single_seg_count(i));
2261 goto again;
2262 }
2263 iov_iter_advance(i, copied);
2264 pos += copied;
2265 written += copied;
2266
2267 balance_dirty_pages_ratelimited(mapping);
2268
2269 } while (iov_iter_count(i));
2270
2271 return written ? written : status;
2272}
2273
2274ssize_t
2275generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2276 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2277 size_t count, ssize_t written)
2278{
2279 struct file *file = iocb->ki_filp;
2280 struct address_space *mapping = file->f_mapping;
2281 const struct address_space_operations *a_ops = mapping->a_ops;
2282 struct inode *inode = mapping->host;
2283 ssize_t status;
2284 struct iov_iter i;
2285
2286 iov_iter_init(&i, iov, nr_segs, count, written);
2287 if (a_ops->write_begin)
2288 status = generic_perform_write(file, &i, pos);
2289 else
2290 status = generic_perform_write_2copy(file, &i, pos);
2291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002293 written += status;
2294 *ppos = pos + status;
2295
2296 /*
2297 * For now, when the user asks for O_SYNC, we'll actually give
2298 * O_DSYNC
2299 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2301 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2302 status = generic_osync_inode(inode, mapping,
2303 OSYNC_METADATA|OSYNC_DATA);
2304 }
2305 }
2306
2307 /*
2308 * If we get here for O_DIRECT writes then we must have fallen through
2309 * to buffered writes (block instantiation inside i_size). So we sync
2310 * the file data here, to try to honour O_DIRECT expectations.
2311 */
2312 if (unlikely(file->f_flags & O_DIRECT) && written)
2313 status = filemap_write_and_wait(mapping);
2314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 return written ? written : status;
2316}
2317EXPORT_SYMBOL(generic_file_buffered_write);
2318
Adrian Bunk5ce78522005-09-10 00:26:28 -07002319static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2321 unsigned long nr_segs, loff_t *ppos)
2322{
2323 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002324 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 size_t ocount; /* original count */
2326 size_t count; /* after file limit checks */
2327 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 loff_t pos;
2329 ssize_t written;
2330 ssize_t err;
2331
2332 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002333 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2334 if (err)
2335 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
2337 count = ocount;
2338 pos = *ppos;
2339
2340 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2341
2342 /* We can write back this queue in page reclaim */
2343 current->backing_dev_info = mapping->backing_dev_info;
2344 written = 0;
2345
2346 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2347 if (err)
2348 goto out;
2349
2350 if (count == 0)
2351 goto out;
2352
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002353 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 if (err)
2355 goto out;
2356
Christoph Hellwig870f4812006-01-09 20:52:01 -08002357 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358
2359 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2360 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002361 loff_t endbyte;
2362 ssize_t written_buffered;
2363
2364 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2365 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 if (written < 0 || written == count)
2367 goto out;
2368 /*
2369 * direct-io write to a hole: fall through to buffered I/O
2370 * for completing the rest of the request.
2371 */
2372 pos += written;
2373 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002374 written_buffered = generic_file_buffered_write(iocb, iov,
2375 nr_segs, pos, ppos, count,
2376 written);
2377 /*
2378 * If generic_file_buffered_write() retuned a synchronous error
2379 * then we want to return the number of bytes which were
2380 * direct-written, or the error code if that was zero. Note
2381 * that this differs from normal direct-io semantics, which
2382 * will return -EFOO even if some bytes were written.
2383 */
2384 if (written_buffered < 0) {
2385 err = written_buffered;
2386 goto out;
2387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002389 /*
2390 * We need to ensure that the page cache pages are written to
2391 * disk and invalidated to preserve the expected O_DIRECT
2392 * semantics.
2393 */
2394 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002395 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2396 SYNC_FILE_RANGE_WAIT_BEFORE|
2397 SYNC_FILE_RANGE_WRITE|
2398 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002399 if (err == 0) {
2400 written = written_buffered;
2401 invalidate_mapping_pages(mapping,
2402 pos >> PAGE_CACHE_SHIFT,
2403 endbyte >> PAGE_CACHE_SHIFT);
2404 } else {
2405 /*
2406 * We don't know how much we wrote, so just return
2407 * the number of bytes which were direct-written
2408 */
2409 }
2410 } else {
2411 written = generic_file_buffered_write(iocb, iov, nr_segs,
2412 pos, ppos, count, written);
2413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414out:
2415 current->backing_dev_info = NULL;
2416 return written ? written : err;
2417}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
Badari Pulavarty027445c2006-09-30 23:28:46 -07002419ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2420 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421{
2422 struct file *file = iocb->ki_filp;
2423 struct address_space *mapping = file->f_mapping;
2424 struct inode *inode = mapping->host;
2425 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
Badari Pulavarty027445c2006-09-30 23:28:46 -07002427 BUG_ON(iocb->ki_pos != pos);
2428
2429 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2430 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
2432 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002433 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
2435 err = sync_page_range_nolock(inode, mapping, pos, ret);
2436 if (err < 0)
2437 ret = err;
2438 }
2439 return ret;
2440}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002441EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
Badari Pulavarty027445c2006-09-30 23:28:46 -07002443ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2444 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445{
2446 struct file *file = iocb->ki_filp;
2447 struct address_space *mapping = file->f_mapping;
2448 struct inode *inode = mapping->host;
2449 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450
2451 BUG_ON(iocb->ki_pos != pos);
2452
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002453 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002454 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2455 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002456 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
2458 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2459 ssize_t err;
2460
2461 err = sync_page_range(inode, mapping, pos, ret);
2462 if (err < 0)
2463 ret = err;
2464 }
2465 return ret;
2466}
2467EXPORT_SYMBOL(generic_file_aio_write);
2468
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002470 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 * went wrong during pagecache shootdown.
2472 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002473static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2475 loff_t offset, unsigned long nr_segs)
2476{
2477 struct file *file = iocb->ki_filp;
2478 struct address_space *mapping = file->f_mapping;
2479 ssize_t retval;
Zach Brown65b82912007-03-16 13:38:11 -08002480 size_t write_len;
2481 pgoff_t end = 0; /* silence gcc */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
2483 /*
2484 * If it's a write, unmap all mmappings of the file up-front. This
2485 * will cause any pte dirty bits to be propagated into the pageframes
2486 * for the subsequent filemap_write_and_wait().
2487 */
2488 if (rw == WRITE) {
2489 write_len = iov_length(iov, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002490 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 if (mapping_mapped(mapping))
2492 unmap_mapping_range(mapping, offset, write_len, 0);
2493 }
2494
2495 retval = filemap_write_and_wait(mapping);
Zach Brown65b82912007-03-16 13:38:11 -08002496 if (retval)
2497 goto out;
2498
2499 /*
2500 * After a write we want buffered reads to be sure to go to disk to get
2501 * the new data. We invalidate clean cached page from the region we're
2502 * about to write. We do this *before* the write so that we can return
2503 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2504 */
2505 if (rw == WRITE && mapping->nrpages) {
2506 retval = invalidate_inode_pages2_range(mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002508 if (retval)
2509 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 }
Zach Brown65b82912007-03-16 13:38:11 -08002511
2512 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2513 if (retval)
2514 goto out;
2515
2516 /*
2517 * Finally, try again to invalidate clean pages which might have been
2518 * faulted in by get_user_pages() if the source of the write was an
2519 * mmap()ed region of the file we're writing. That's a pretty crazy
2520 * thing to do, so we don't support it 100%. If this invalidation
2521 * fails and we have -EIOCBQUEUED we ignore the failure.
2522 */
2523 if (rw == WRITE && mapping->nrpages) {
2524 int err = invalidate_inode_pages2_range(mapping,
2525 offset >> PAGE_CACHE_SHIFT, end);
2526 if (err && retval >= 0)
2527 retval = err;
2528 }
2529out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 return retval;
2531}
David Howellscf9a2ae2006-08-29 19:05:54 +01002532
2533/**
2534 * try_to_release_page() - release old fs-specific metadata on a page
2535 *
2536 * @page: the page which the kernel is trying to free
2537 * @gfp_mask: memory allocation flags (and I/O mode)
2538 *
2539 * The address_space is to try to release any data against the page
2540 * (presumably at page->private). If the release was successful, return `1'.
2541 * Otherwise return zero.
2542 *
2543 * The @gfp_mask argument specifies whether I/O may be performed to release
2544 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2545 *
2546 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2547 */
2548int try_to_release_page(struct page *page, gfp_t gfp_mask)
2549{
2550 struct address_space * const mapping = page->mapping;
2551
2552 BUG_ON(!PageLocked(page));
2553 if (PageWriteback(page))
2554 return 0;
2555
2556 if (mapping && mapping->a_ops->releasepage)
2557 return mapping->a_ops->releasepage(page, gfp_mask);
2558 return try_to_free_buffers(page);
2559}
2560
2561EXPORT_SYMBOL(try_to_release_page);