blob: 15c8413ee929200006e4eb5e8a96bf5b04f0f213 [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>
Carsten Otteceffc072005-06-23 22:05:25 -070033#include "filemap.h"
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 */
596struct page * find_get_page(struct address_space *mapping, unsigned long offset)
597{
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,
620 unsigned long offset)
621{
622 struct page *page;
623
624 read_lock_irq(&mapping->tree_lock);
625repeat:
626 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 read_lock_irq(&mapping->tree_lock);
633
634 /* Has the page been truncated while we slept? */
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800635 if (unlikely(page->mapping != mapping ||
636 page->index != offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 unlock_page(page);
638 page_cache_release(page);
639 goto repeat;
640 }
641 }
642 }
643 read_unlock_irq(&mapping->tree_lock);
644 return page;
645}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646EXPORT_SYMBOL(find_lock_page);
647
648/**
649 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700650 * @mapping: the page's address_space
651 * @index: the page's index into the mapping
652 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 *
654 * Locates a page in the pagecache. If the page is not present, a new page
655 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
656 * LRU list. The returned page is locked and has its reference count
657 * incremented.
658 *
659 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
660 * allocation!
661 *
662 * find_or_create_page() returns the desired page's address, or zero on
663 * memory exhaustion.
664 */
665struct page *find_or_create_page(struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400666 unsigned long index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
668 struct page *page, *cached_page = NULL;
669 int err;
670repeat:
671 page = find_lock_page(mapping, index);
672 if (!page) {
673 if (!cached_page) {
Christoph Lameter43c0f3d2007-05-15 23:57:09 -0700674 cached_page =
675 __page_cache_alloc(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (!cached_page)
677 return NULL;
678 }
679 err = add_to_page_cache_lru(cached_page, mapping,
680 index, gfp_mask);
681 if (!err) {
682 page = cached_page;
683 cached_page = NULL;
684 } else if (err == -EEXIST)
685 goto repeat;
686 }
687 if (cached_page)
688 page_cache_release(cached_page);
689 return page;
690}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691EXPORT_SYMBOL(find_or_create_page);
692
693/**
694 * find_get_pages - gang pagecache lookup
695 * @mapping: The address_space to search
696 * @start: The starting page index
697 * @nr_pages: The maximum number of pages
698 * @pages: Where the resulting pages are placed
699 *
700 * find_get_pages() will search for and return a group of up to
701 * @nr_pages pages in the mapping. The pages are placed at @pages.
702 * find_get_pages() takes a reference against the returned pages.
703 *
704 * The search returns a group of mapping-contiguous pages with ascending
705 * indexes. There may be holes in the indices due to not-present pages.
706 *
707 * find_get_pages() returns the number of pages which were found.
708 */
709unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
710 unsigned int nr_pages, struct page **pages)
711{
712 unsigned int i;
713 unsigned int ret;
714
715 read_lock_irq(&mapping->tree_lock);
716 ret = radix_tree_gang_lookup(&mapping->page_tree,
717 (void **)pages, start, nr_pages);
718 for (i = 0; i < ret; i++)
719 page_cache_get(pages[i]);
720 read_unlock_irq(&mapping->tree_lock);
721 return ret;
722}
723
Jens Axboeebf43502006-04-27 08:46:01 +0200724/**
725 * find_get_pages_contig - gang contiguous pagecache lookup
726 * @mapping: The address_space to search
727 * @index: The starting page index
728 * @nr_pages: The maximum number of pages
729 * @pages: Where the resulting pages are placed
730 *
731 * find_get_pages_contig() works exactly like find_get_pages(), except
732 * that the returned number of pages are guaranteed to be contiguous.
733 *
734 * find_get_pages_contig() returns the number of pages which were found.
735 */
736unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
737 unsigned int nr_pages, struct page **pages)
738{
739 unsigned int i;
740 unsigned int ret;
741
742 read_lock_irq(&mapping->tree_lock);
743 ret = radix_tree_gang_lookup(&mapping->page_tree,
744 (void **)pages, index, nr_pages);
745 for (i = 0; i < ret; i++) {
746 if (pages[i]->mapping == NULL || pages[i]->index != index)
747 break;
748
749 page_cache_get(pages[i]);
750 index++;
751 }
752 read_unlock_irq(&mapping->tree_lock);
753 return i;
754}
David Howellsef71c152007-05-09 02:33:44 -0700755EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200756
Randy Dunlap485bb992006-06-23 02:03:49 -0700757/**
758 * find_get_pages_tag - find and return pages that match @tag
759 * @mapping: the address_space to search
760 * @index: the starting page index
761 * @tag: the tag index
762 * @nr_pages: the maximum number of pages
763 * @pages: where the resulting pages are placed
764 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700766 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 */
768unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
769 int tag, unsigned int nr_pages, struct page **pages)
770{
771 unsigned int i;
772 unsigned int ret;
773
774 read_lock_irq(&mapping->tree_lock);
775 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
776 (void **)pages, *index, nr_pages, tag);
777 for (i = 0; i < ret; i++)
778 page_cache_get(pages[i]);
779 if (ret)
780 *index = pages[ret - 1]->index + 1;
781 read_unlock_irq(&mapping->tree_lock);
782 return ret;
783}
David Howellsef71c152007-05-09 02:33:44 -0700784EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Randy Dunlap485bb992006-06-23 02:03:49 -0700786/**
787 * grab_cache_page_nowait - returns locked page at given index in given cache
788 * @mapping: target address_space
789 * @index: the page index
790 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800791 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 * This is intended for speculative data generators, where the data can
793 * be regenerated if the page couldn't be grabbed. This routine should
794 * be safe to call while holding the lock for another page.
795 *
796 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
797 * and deadlock against the caller's locked page.
798 */
799struct page *
800grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
801{
802 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804 if (page) {
805 if (!TestSetPageLocked(page))
806 return page;
807 page_cache_release(page);
808 return NULL;
809 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700810 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
811 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 page_cache_release(page);
813 page = NULL;
814 }
815 return page;
816}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817EXPORT_SYMBOL(grab_cache_page_nowait);
818
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700819/*
820 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
821 * a _large_ part of the i/o request. Imagine the worst scenario:
822 *
823 * ---R__________________________________________B__________
824 * ^ reading here ^ bad block(assume 4k)
825 *
826 * read(R) => miss => readahead(R...B) => media error => frustrating retries
827 * => failing the whole request => read(R) => read(R+1) =>
828 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
829 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
830 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
831 *
832 * It is going insane. Fix it by quickly scaling down the readahead size.
833 */
834static void shrink_readahead_size_eio(struct file *filp,
835 struct file_ra_state *ra)
836{
837 if (!ra->ra_pages)
838 return;
839
840 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700841}
842
Randy Dunlap485bb992006-06-23 02:03:49 -0700843/**
844 * do_generic_mapping_read - generic file read routine
845 * @mapping: address_space to be read
Stephen Hemmingerf0b85c02007-08-10 13:01:16 -0700846 * @_ra: file's readahead state
Randy Dunlap485bb992006-06-23 02:03:49 -0700847 * @filp: the file to read
848 * @ppos: current file position
849 * @desc: read_descriptor
850 * @actor: read method
851 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700853 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 *
855 * This is really ugly. But the goto's actually try to clarify some
856 * of the logic when it comes to error handling etc.
857 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700858 * Note the struct file* is only passed for the use of readpage.
859 * It may be NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 */
861void do_generic_mapping_read(struct address_space *mapping,
862 struct file_ra_state *_ra,
863 struct file *filp,
864 loff_t *ppos,
865 read_descriptor_t *desc,
866 read_actor_t actor)
867{
868 struct inode *inode = mapping->host;
869 unsigned long index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 unsigned long offset;
871 unsigned long last_index;
872 unsigned long next_index;
873 unsigned long prev_index;
Jan Karaec0f1632007-05-06 14:49:25 -0700874 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 struct page *cached_page;
876 int error;
877 struct file_ra_state ra = *_ra;
878
879 cached_page = NULL;
880 index = *ppos >> PAGE_CACHE_SHIFT;
881 next_index = index;
Jan Kara6ce745e2007-05-06 14:49:26 -0700882 prev_index = ra.prev_index;
883 prev_offset = ra.prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
885 offset = *ppos & ~PAGE_CACHE_MASK;
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 for (;;) {
888 struct page *page;
NeilBrowna32ea1e2007-07-17 04:03:04 -0700889 unsigned long end_index;
890 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 unsigned long nr, ret;
892
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894find_page:
895 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700896 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700897 page_cache_sync_readahead(mapping,
898 &ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700899 index, last_index - index);
900 page = find_get_page(mapping, index);
901 if (unlikely(page == NULL))
902 goto no_cached_page;
903 }
904 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700905 page_cache_async_readahead(mapping,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700906 &ra, filp, page,
907 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 }
909 if (!PageUptodate(page))
910 goto page_not_up_to_date;
911page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -0700912 /*
913 * i_size must be checked after we know the page is Uptodate.
914 *
915 * Checking i_size after the check allows us to calculate
916 * the correct value for "nr", which means the zero-filled
917 * part of the page is not copied back to userspace (unless
918 * another truncate extends the file - this is desired though).
919 */
920
921 isize = i_size_read(inode);
922 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
923 if (unlikely(!isize || index > end_index)) {
924 page_cache_release(page);
925 goto out;
926 }
927
928 /* nr is the maximum number of bytes to copy from this page */
929 nr = PAGE_CACHE_SIZE;
930 if (index == end_index) {
931 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
932 if (nr <= offset) {
933 page_cache_release(page);
934 goto out;
935 }
936 }
937 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
939 /* If users can be writing to this page using arbitrary
940 * virtual addresses, take care about potential aliasing
941 * before reading the page on the kernel side.
942 */
943 if (mapping_writably_mapped(mapping))
944 flush_dcache_page(page);
945
946 /*
Jan Karaec0f1632007-05-06 14:49:25 -0700947 * When a sequential read accesses a page several times,
948 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 */
Jan Karaec0f1632007-05-06 14:49:25 -0700950 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 mark_page_accessed(page);
952 prev_index = index;
953
954 /*
955 * Ok, we have the page, and it's up-to-date, so
956 * now we can copy it to user space...
957 *
958 * The actor routine returns how many bytes were actually used..
959 * NOTE! This may not be the same as how much of a user buffer
960 * we filled up (we may be padding etc), so we can only update
961 * "pos" here (the actor routine has to update the user buffer
962 * pointers and the remaining count).
963 */
964 ret = actor(desc, page, offset, nr);
965 offset += ret;
966 index += offset >> PAGE_CACHE_SHIFT;
967 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -0700968 prev_offset = offset;
969 ra.prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 page_cache_release(page);
972 if (ret == nr && desc->count)
973 continue;
974 goto out;
975
976page_not_up_to_date:
977 /* Get exclusive access to the page ... */
978 lock_page(page);
979
Nick Pigginda6052f2006-09-25 23:31:35 -0700980 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 if (!page->mapping) {
982 unlock_page(page);
983 page_cache_release(page);
984 continue;
985 }
986
987 /* Did somebody else fill it already? */
988 if (PageUptodate(page)) {
989 unlock_page(page);
990 goto page_ok;
991 }
992
993readpage:
994 /* Start the actual read. The read will unlock the page. */
995 error = mapping->a_ops->readpage(filp, page);
996
Zach Brown994fc28c2005-12-15 14:28:17 -0800997 if (unlikely(error)) {
998 if (error == AOP_TRUNCATED_PAGE) {
999 page_cache_release(page);
1000 goto find_page;
1001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001003 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
1005 if (!PageUptodate(page)) {
1006 lock_page(page);
1007 if (!PageUptodate(page)) {
1008 if (page->mapping == NULL) {
1009 /*
1010 * invalidate_inode_pages got it
1011 */
1012 unlock_page(page);
1013 page_cache_release(page);
1014 goto find_page;
1015 }
1016 unlock_page(page);
1017 error = -EIO;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001018 shrink_readahead_size_eio(filp, &ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 goto readpage_error;
1020 }
1021 unlock_page(page);
1022 }
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 goto page_ok;
1025
1026readpage_error:
1027 /* UHHUH! A synchronous read error occurred. Report it */
1028 desc->error = error;
1029 page_cache_release(page);
1030 goto out;
1031
1032no_cached_page:
1033 /*
1034 * Ok, it wasn't cached, so we need to create a new
1035 * page..
1036 */
1037 if (!cached_page) {
1038 cached_page = page_cache_alloc_cold(mapping);
1039 if (!cached_page) {
1040 desc->error = -ENOMEM;
1041 goto out;
1042 }
1043 }
1044 error = add_to_page_cache_lru(cached_page, mapping,
1045 index, GFP_KERNEL);
1046 if (error) {
1047 if (error == -EEXIST)
1048 goto find_page;
1049 desc->error = error;
1050 goto out;
1051 }
1052 page = cached_page;
1053 cached_page = NULL;
1054 goto readpage;
1055 }
1056
1057out:
1058 *_ra = ra;
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001059 _ra->prev_index = prev_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
1061 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1062 if (cached_page)
1063 page_cache_release(cached_page);
1064 if (filp)
1065 file_accessed(filp);
1066}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067EXPORT_SYMBOL(do_generic_mapping_read);
1068
1069int file_read_actor(read_descriptor_t *desc, struct page *page,
1070 unsigned long offset, unsigned long size)
1071{
1072 char *kaddr;
1073 unsigned long left, count = desc->count;
1074
1075 if (size > count)
1076 size = count;
1077
1078 /*
1079 * Faults on the destination of a read are common, so do it before
1080 * taking the kmap.
1081 */
1082 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1083 kaddr = kmap_atomic(page, KM_USER0);
1084 left = __copy_to_user_inatomic(desc->arg.buf,
1085 kaddr + offset, size);
1086 kunmap_atomic(kaddr, KM_USER0);
1087 if (left == 0)
1088 goto success;
1089 }
1090
1091 /* Do it the slow way */
1092 kaddr = kmap(page);
1093 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1094 kunmap(page);
1095
1096 if (left) {
1097 size -= left;
1098 desc->error = -EFAULT;
1099 }
1100success:
1101 desc->count = count - size;
1102 desc->written += size;
1103 desc->arg.buf += size;
1104 return size;
1105}
1106
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001107/*
1108 * Performs necessary checks before doing a write
1109 * @iov: io vector request
1110 * @nr_segs: number of segments in the iovec
1111 * @count: number of bytes to write
1112 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1113 *
1114 * Adjust number of segments and amount of bytes to write (nr_segs should be
1115 * properly initialized first). Returns appropriate error code that caller
1116 * should return or zero in case that write should be allowed.
1117 */
1118int generic_segment_checks(const struct iovec *iov,
1119 unsigned long *nr_segs, size_t *count, int access_flags)
1120{
1121 unsigned long seg;
1122 size_t cnt = 0;
1123 for (seg = 0; seg < *nr_segs; seg++) {
1124 const struct iovec *iv = &iov[seg];
1125
1126 /*
1127 * If any segment has a negative length, or the cumulative
1128 * length ever wraps negative then return -EINVAL.
1129 */
1130 cnt += iv->iov_len;
1131 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1132 return -EINVAL;
1133 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1134 continue;
1135 if (seg == 0)
1136 return -EFAULT;
1137 *nr_segs = seg;
1138 cnt -= iv->iov_len; /* This segment is no good */
1139 break;
1140 }
1141 *count = cnt;
1142 return 0;
1143}
1144EXPORT_SYMBOL(generic_segment_checks);
1145
Randy Dunlap485bb992006-06-23 02:03:49 -07001146/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001147 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001148 * @iocb: kernel I/O control block
1149 * @iov: io vector request
1150 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001151 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001152 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 * This is the "read()" routine for all filesystems
1154 * that can use the page cache directly.
1155 */
1156ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001157generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1158 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159{
1160 struct file *filp = iocb->ki_filp;
1161 ssize_t retval;
1162 unsigned long seg;
1163 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001164 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
1166 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001167 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1168 if (retval)
1169 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
1171 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1172 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001173 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 struct address_space *mapping;
1175 struct inode *inode;
1176
1177 mapping = filp->f_mapping;
1178 inode = mapping->host;
1179 retval = 0;
1180 if (!count)
1181 goto out; /* skip atime */
1182 size = i_size_read(inode);
1183 if (pos < size) {
1184 retval = generic_file_direct_IO(READ, iocb,
1185 iov, pos, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 if (retval > 0)
1187 *ppos = pos + retval;
1188 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001189 if (likely(retval != 0)) {
Steven Whitehouse3f1a9aa2006-09-27 14:52:48 -04001190 file_accessed(filp);
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001191 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 }
1194
1195 retval = 0;
1196 if (count) {
1197 for (seg = 0; seg < nr_segs; seg++) {
1198 read_descriptor_t desc;
1199
1200 desc.written = 0;
1201 desc.arg.buf = iov[seg].iov_base;
1202 desc.count = iov[seg].iov_len;
1203 if (desc.count == 0)
1204 continue;
1205 desc.error = 0;
1206 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1207 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001208 if (desc.error) {
1209 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 break;
1211 }
akpm@linux-foundation.orgc44939e2007-07-15 23:38:25 -07001212 if (desc.count > 0)
1213 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215 }
1216out:
1217 return retval;
1218}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219EXPORT_SYMBOL(generic_file_aio_read);
1220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221static ssize_t
1222do_readahead(struct address_space *mapping, struct file *filp,
1223 unsigned long index, unsigned long nr)
1224{
1225 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1226 return -EINVAL;
1227
1228 force_page_cache_readahead(mapping, filp, index,
1229 max_sane_readahead(nr));
1230 return 0;
1231}
1232
1233asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1234{
1235 ssize_t ret;
1236 struct file *file;
1237
1238 ret = -EBADF;
1239 file = fget(fd);
1240 if (file) {
1241 if (file->f_mode & FMODE_READ) {
1242 struct address_space *mapping = file->f_mapping;
1243 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1244 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1245 unsigned long len = end - start + 1;
1246 ret = do_readahead(mapping, file, start, len);
1247 }
1248 fput(file);
1249 }
1250 return ret;
1251}
1252
1253#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001254static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1255/**
1256 * page_cache_read - adds requested page to the page cache if not already there
1257 * @file: file to read
1258 * @offset: page index
1259 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 * This adds the requested page to the page cache if it isn't already there,
1261 * and schedules an I/O to read in its contents from disk.
1262 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263static int fastcall page_cache_read(struct file * file, unsigned long offset)
1264{
1265 struct address_space *mapping = file->f_mapping;
1266 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001267 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
Zach Brown994fc28c2005-12-15 14:28:17 -08001269 do {
1270 page = page_cache_alloc_cold(mapping);
1271 if (!page)
1272 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Zach Brown994fc28c2005-12-15 14:28:17 -08001274 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1275 if (ret == 0)
1276 ret = mapping->a_ops->readpage(file, page);
1277 else if (ret == -EEXIST)
1278 ret = 0; /* losing race to add is OK */
1279
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
Zach Brown994fc28c2005-12-15 14:28:17 -08001282 } while (ret == AOP_TRUNCATED_PAGE);
1283
1284 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285}
1286
1287#define MMAP_LOTSAMISS (100)
1288
Randy Dunlap485bb992006-06-23 02:03:49 -07001289/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001290 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001291 * @vma: vma in which the fault was taken
1292 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001293 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001294 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 * mapped memory region to read in file data during a page fault.
1296 *
1297 * The goto's are kind of ugly, but this streamlines the normal case of having
1298 * it in the page cache, and handles the special cases reasonably without
1299 * having a lot of duplicated code.
1300 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001301int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
1303 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001304 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 struct address_space *mapping = file->f_mapping;
1306 struct file_ra_state *ra = &file->f_ra;
1307 struct inode *inode = mapping->host;
1308 struct page *page;
Nick Piggin54cb8822007-07-19 01:46:59 -07001309 unsigned long size;
1310 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001311 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001314 if (vmf->pgoff >= size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 goto outside_data_content;
1316
1317 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001318 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 goto no_cached_page;
1320
1321 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 * Do we have something in the page cache already?
1323 */
1324retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001325 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001326 /*
1327 * For sequential accesses, we use the generic readahead logic.
1328 */
1329 if (VM_SequentialReadHint(vma)) {
1330 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001331 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001332 vmf->pgoff, 1);
1333 page = find_lock_page(mapping, vmf->pgoff);
1334 if (!page)
1335 goto no_cached_page;
1336 }
1337 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001338 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001339 vmf->pgoff, 1);
1340 }
1341 }
1342
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (!page) {
1344 unsigned long ra_pages;
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 ra->mmap_miss++;
1347
1348 /*
1349 * Do we miss much more than hit in this file? If so,
1350 * stop bothering with read-ahead. It will only hurt.
1351 */
1352 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1353 goto no_cached_page;
1354
1355 /*
1356 * To keep the pgmajfault counter straight, we need to
1357 * check did_readaround, as this is an inner loop.
1358 */
1359 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001360 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001361 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 }
1363 did_readaround = 1;
1364 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1365 if (ra_pages) {
1366 pgoff_t start = 0;
1367
Nick Piggind0217ac2007-07-19 01:47:03 -07001368 if (vmf->pgoff > ra_pages / 2)
1369 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 do_page_cache_readahead(mapping, file, start, ra_pages);
1371 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001372 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if (!page)
1374 goto no_cached_page;
1375 }
1376
1377 if (!did_readaround)
1378 ra->mmap_hit++;
1379
1380 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001381 * We have a locked page in the page cache, now we need to check
1382 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 */
Nick Piggind00806b2007-07-19 01:46:57 -07001384 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 goto page_not_uptodate;
1386
Nick Piggind00806b2007-07-19 01:46:57 -07001387 /* Must recheck i_size under page lock */
1388 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001389 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001390 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001391 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001392 goto outside_data_content;
1393 }
1394
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 /*
1396 * Found the page and have a reference on it.
1397 */
1398 mark_page_accessed(page);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001399 ra->prev_index = page->index;
Nick Piggind0217ac2007-07-19 01:47:03 -07001400 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001401 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403outside_data_content:
1404 /*
1405 * An external ptracer can access pages that normally aren't
1406 * accessible..
1407 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001408 if (vma->vm_mm == current->mm)
1409 return VM_FAULT_SIGBUS;
1410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 /* Fall through to the non-read-ahead case */
1412no_cached_page:
1413 /*
1414 * We're only likely to ever get here if MADV_RANDOM is in
1415 * effect.
1416 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001417 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
1419 /*
1420 * The page we want has now been added to the page cache.
1421 * In the unlikely event that someone removed it in the
1422 * meantime, we'll just come back here and read it again.
1423 */
1424 if (error >= 0)
1425 goto retry_find;
1426
1427 /*
1428 * An error return from page_cache_read can result if the
1429 * system is low on memory, or a problem occurs while trying
1430 * to schedule I/O.
1431 */
1432 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001433 return VM_FAULT_OOM;
1434 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001437 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001439 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001440 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
1443 /*
1444 * Umm, take care of errors if the page isn't up-to-date.
1445 * Try to re-read it _once_. We do this synchronously,
1446 * because there really aren't any performance issues here
1447 * and we need to check for errors.
1448 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001450 error = mapping->a_ops->readpage(file, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001452
1453 if (!error || error == AOP_TRUNCATED_PAGE)
1454 goto retry_find;
1455
1456 /* Things didn't work out. Return zero to tell the mm layer so. */
1457 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001458 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001459}
1460EXPORT_SYMBOL(filemap_fault);
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001463 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464};
1465
1466/* This is used for a general mmap of a disk file */
1467
1468int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1469{
1470 struct address_space *mapping = file->f_mapping;
1471
1472 if (!mapping->a_ops->readpage)
1473 return -ENOEXEC;
1474 file_accessed(file);
1475 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001476 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return 0;
1478}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
1480/*
1481 * This is for filesystems which do not implement ->writepage.
1482 */
1483int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1484{
1485 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1486 return -EINVAL;
1487 return generic_file_mmap(file, vma);
1488}
1489#else
1490int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1491{
1492 return -ENOSYS;
1493}
1494int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1495{
1496 return -ENOSYS;
1497}
1498#endif /* CONFIG_MMU */
1499
1500EXPORT_SYMBOL(generic_file_mmap);
1501EXPORT_SYMBOL(generic_file_readonly_mmap);
1502
Nick Piggin6fe69002007-05-06 14:49:04 -07001503static struct page *__read_cache_page(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 unsigned long index,
1505 int (*filler)(void *,struct page*),
1506 void *data)
1507{
1508 struct page *page, *cached_page = NULL;
1509 int err;
1510repeat:
1511 page = find_get_page(mapping, index);
1512 if (!page) {
1513 if (!cached_page) {
1514 cached_page = page_cache_alloc_cold(mapping);
1515 if (!cached_page)
1516 return ERR_PTR(-ENOMEM);
1517 }
1518 err = add_to_page_cache_lru(cached_page, mapping,
1519 index, GFP_KERNEL);
1520 if (err == -EEXIST)
1521 goto repeat;
1522 if (err < 0) {
1523 /* Presumably ENOMEM for radix tree node */
1524 page_cache_release(cached_page);
1525 return ERR_PTR(err);
1526 }
1527 page = cached_page;
1528 cached_page = NULL;
1529 err = filler(data, page);
1530 if (err < 0) {
1531 page_cache_release(page);
1532 page = ERR_PTR(err);
1533 }
1534 }
1535 if (cached_page)
1536 page_cache_release(cached_page);
1537 return page;
1538}
1539
Nick Piggin6fe69002007-05-06 14:49:04 -07001540/*
1541 * Same as read_cache_page, but don't wait for page to become unlocked
1542 * after submitting it to the filler.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001544struct page *read_cache_page_async(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 unsigned long index,
1546 int (*filler)(void *,struct page*),
1547 void *data)
1548{
1549 struct page *page;
1550 int err;
1551
1552retry:
1553 page = __read_cache_page(mapping, index, filler, data);
1554 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001555 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if (PageUptodate(page))
1557 goto out;
1558
1559 lock_page(page);
1560 if (!page->mapping) {
1561 unlock_page(page);
1562 page_cache_release(page);
1563 goto retry;
1564 }
1565 if (PageUptodate(page)) {
1566 unlock_page(page);
1567 goto out;
1568 }
1569 err = filler(data, page);
1570 if (err < 0) {
1571 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001572 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 }
David Howellsc855ff32007-05-09 13:42:20 +01001574out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001575 mark_page_accessed(page);
1576 return page;
1577}
1578EXPORT_SYMBOL(read_cache_page_async);
1579
1580/**
1581 * read_cache_page - read into page cache, fill it if needed
1582 * @mapping: the page's address_space
1583 * @index: the page index
1584 * @filler: function to perform the read
1585 * @data: destination for read data
1586 *
1587 * Read into the page cache. If a page already exists, and PageUptodate() is
1588 * not set, try to fill the page then wait for it to become unlocked.
1589 *
1590 * If the page does not get brought uptodate, return -EIO.
1591 */
1592struct page *read_cache_page(struct address_space *mapping,
1593 unsigned long index,
1594 int (*filler)(void *,struct page*),
1595 void *data)
1596{
1597 struct page *page;
1598
1599 page = read_cache_page_async(mapping, index, filler, data);
1600 if (IS_ERR(page))
1601 goto out;
1602 wait_on_page_locked(page);
1603 if (!PageUptodate(page)) {
1604 page_cache_release(page);
1605 page = ERR_PTR(-EIO);
1606 }
1607 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 return page;
1609}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610EXPORT_SYMBOL(read_cache_page);
1611
1612/*
1613 * If the page was newly created, increment its refcount and add it to the
1614 * caller's lru-buffering pagevec. This function is specifically for
1615 * generic_file_write().
1616 */
1617static inline struct page *
1618__grab_cache_page(struct address_space *mapping, unsigned long index,
1619 struct page **cached_page, struct pagevec *lru_pvec)
1620{
1621 int err;
1622 struct page *page;
1623repeat:
1624 page = find_lock_page(mapping, index);
1625 if (!page) {
1626 if (!*cached_page) {
1627 *cached_page = page_cache_alloc(mapping);
1628 if (!*cached_page)
1629 return NULL;
1630 }
1631 err = add_to_page_cache(*cached_page, mapping,
1632 index, GFP_KERNEL);
1633 if (err == -EEXIST)
1634 goto repeat;
1635 if (err == 0) {
1636 page = *cached_page;
1637 page_cache_get(page);
1638 if (!pagevec_add(lru_pvec, page))
1639 __pagevec_lru_add(lru_pvec);
1640 *cached_page = NULL;
1641 }
1642 }
1643 return page;
1644}
1645
1646/*
1647 * The logic we want is
1648 *
1649 * if suid or (sgid and xgrp)
1650 * remove privs
1651 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001652int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653{
1654 mode_t mode = dentry->d_inode->i_mode;
1655 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657 /* suid always must be killed */
1658 if (unlikely(mode & S_ISUID))
1659 kill = ATTR_KILL_SUID;
1660
1661 /*
1662 * sgid without any exec bits is just a mandatory locking mark; leave
1663 * it alone. If some exec bits are set, it's a real sgid; kill it.
1664 */
1665 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1666 kill |= ATTR_KILL_SGID;
1667
Jens Axboe01de85e2006-10-17 19:50:36 +02001668 if (unlikely(kill && !capable(CAP_FSETID)))
1669 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
Jens Axboe01de85e2006-10-17 19:50:36 +02001671 return 0;
1672}
Mark Fashehd23a1472006-10-17 17:05:18 -07001673EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001674
1675int __remove_suid(struct dentry *dentry, int kill)
1676{
1677 struct iattr newattrs;
1678
1679 newattrs.ia_valid = ATTR_FORCE | kill;
1680 return notify_change(dentry, &newattrs);
1681}
1682
1683int remove_suid(struct dentry *dentry)
1684{
1685 int kill = should_remove_suid(dentry);
1686
1687 if (unlikely(kill))
1688 return __remove_suid(dentry, kill);
1689
1690 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691}
1692EXPORT_SYMBOL(remove_suid);
1693
Carsten Otteceffc072005-06-23 22:05:25 -07001694size_t
NeilBrown01408c42006-06-25 05:47:58 -07001695__filemap_copy_from_user_iovec_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 const struct iovec *iov, size_t base, size_t bytes)
1697{
1698 size_t copied = 0, left = 0;
1699
1700 while (bytes) {
1701 char __user *buf = iov->iov_base + base;
1702 int copy = min(bytes, iov->iov_len - base);
1703
1704 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001705 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 copied += copy;
1707 bytes -= copy;
1708 vaddr += copy;
1709 iov++;
1710
NeilBrown01408c42006-06-25 05:47:58 -07001711 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 }
1714 return copied - left;
1715}
1716
1717/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 * Performs necessary checks before doing a write
1719 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001720 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 * Returns appropriate error code that caller should return or
1722 * zero in case that write should be allowed.
1723 */
1724inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1725{
1726 struct inode *inode = file->f_mapping->host;
1727 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1728
1729 if (unlikely(*pos < 0))
1730 return -EINVAL;
1731
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 if (!isblk) {
1733 /* FIXME: this is for backwards compatibility with 2.4 */
1734 if (file->f_flags & O_APPEND)
1735 *pos = i_size_read(inode);
1736
1737 if (limit != RLIM_INFINITY) {
1738 if (*pos >= limit) {
1739 send_sig(SIGXFSZ, current, 0);
1740 return -EFBIG;
1741 }
1742 if (*count > limit - (typeof(limit))*pos) {
1743 *count = limit - (typeof(limit))*pos;
1744 }
1745 }
1746 }
1747
1748 /*
1749 * LFS rule
1750 */
1751 if (unlikely(*pos + *count > MAX_NON_LFS &&
1752 !(file->f_flags & O_LARGEFILE))) {
1753 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 return -EFBIG;
1755 }
1756 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1757 *count = MAX_NON_LFS - (unsigned long)*pos;
1758 }
1759 }
1760
1761 /*
1762 * Are we about to exceed the fs block limit ?
1763 *
1764 * If we have written data it becomes a short write. If we have
1765 * exceeded without writing data we send a signal and return EFBIG.
1766 * Linus frestrict idea will clean these up nicely..
1767 */
1768 if (likely(!isblk)) {
1769 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1770 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 return -EFBIG;
1772 }
1773 /* zero-length writes at ->s_maxbytes are OK */
1774 }
1775
1776 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1777 *count = inode->i_sb->s_maxbytes - *pos;
1778 } else {
David Howells93614012006-09-30 20:45:40 +02001779#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 loff_t isize;
1781 if (bdev_read_only(I_BDEV(inode)))
1782 return -EPERM;
1783 isize = i_size_read(inode);
1784 if (*pos >= isize) {
1785 if (*count || *pos > isize)
1786 return -ENOSPC;
1787 }
1788
1789 if (*pos + *count > isize)
1790 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02001791#else
1792 return -EPERM;
1793#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 }
1795 return 0;
1796}
1797EXPORT_SYMBOL(generic_write_checks);
1798
1799ssize_t
1800generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1801 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1802 size_t count, size_t ocount)
1803{
1804 struct file *file = iocb->ki_filp;
1805 struct address_space *mapping = file->f_mapping;
1806 struct inode *inode = mapping->host;
1807 ssize_t written;
1808
1809 if (count != ocount)
1810 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1811
1812 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1813 if (written > 0) {
1814 loff_t end = pos + written;
1815 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1816 i_size_write(inode, end);
1817 mark_inode_dirty(inode);
1818 }
1819 *ppos = end;
1820 }
1821
1822 /*
1823 * Sync the fs metadata but not the minor inode changes and
1824 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001825 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08001826 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 */
Zach Brown8459d862006-12-10 02:21:05 -08001828 if ((written >= 0 || written == -EIOCBQUEUED) &&
1829 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07001830 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1831 if (err < 0)
1832 written = err;
1833 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 return written;
1835}
1836EXPORT_SYMBOL(generic_file_direct_write);
1837
1838ssize_t
1839generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
1840 unsigned long nr_segs, loff_t pos, loff_t *ppos,
1841 size_t count, ssize_t written)
1842{
1843 struct file *file = iocb->ki_filp;
1844 struct address_space * mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001845 const struct address_space_operations *a_ops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 struct inode *inode = mapping->host;
1847 long status = 0;
1848 struct page *page;
1849 struct page *cached_page = NULL;
1850 size_t bytes;
1851 struct pagevec lru_pvec;
1852 const struct iovec *cur_iov = iov; /* current iovec */
1853 size_t iov_base = 0; /* offset in the current iovec */
1854 char __user *buf;
1855
1856 pagevec_init(&lru_pvec, 0);
1857
1858 /*
1859 * handle partial DIO write. Adjust cur_iov if needed.
1860 */
1861 if (likely(nr_segs == 1))
1862 buf = iov->iov_base + written;
1863 else {
1864 filemap_set_next_iovec(&cur_iov, &iov_base, written);
akpm@osdl.orgf021e922005-05-01 08:58:35 -07001865 buf = cur_iov->iov_base + iov_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 }
1867
1868 do {
1869 unsigned long index;
1870 unsigned long offset;
1871 size_t copied;
1872
1873 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1874 index = pos >> PAGE_CACHE_SHIFT;
1875 bytes = PAGE_CACHE_SIZE - offset;
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07001876
1877 /* Limit the size of the copy to the caller's write size */
1878 bytes = min(bytes, count);
1879
NeilBrown29dbb3f2007-02-16 01:28:38 -08001880 /* We only need to worry about prefaulting when writes are from
1881 * user-space. NFSd uses vfs_writev with several non-aligned
1882 * segments in the vector, and limiting to one segment a time is
1883 * a noticeable performance for re-write
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07001884 */
NeilBrown29dbb3f2007-02-16 01:28:38 -08001885 if (!segment_eq(get_fs(), KERNEL_DS)) {
1886 /*
1887 * Limit the size of the copy to that of the current
1888 * segment, because fault_in_pages_readable() doesn't
1889 * know how to walk segments.
1890 */
1891 bytes = min(bytes, cur_iov->iov_len - iov_base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
NeilBrown29dbb3f2007-02-16 01:28:38 -08001893 /*
1894 * Bring in the user page that we will copy from
1895 * _first_. Otherwise there's a nasty deadlock on
1896 * copying from the same page as we're writing to,
1897 * without it being marked up-to-date.
1898 */
1899 fault_in_pages_readable(buf, bytes);
1900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
1902 if (!page) {
1903 status = -ENOMEM;
1904 break;
1905 }
1906
Andrew Morton81b0c872006-06-29 02:24:26 -07001907 if (unlikely(bytes == 0)) {
1908 status = 0;
1909 copied = 0;
1910 goto zero_length_segment;
1911 }
1912
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 status = a_ops->prepare_write(file, page, offset, offset+bytes);
1914 if (unlikely(status)) {
1915 loff_t isize = i_size_read(inode);
Zach Brown994fc28c2005-12-15 14:28:17 -08001916
1917 if (status != AOP_TRUNCATED_PAGE)
1918 unlock_page(page);
1919 page_cache_release(page);
1920 if (status == AOP_TRUNCATED_PAGE)
1921 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 /*
1923 * prepare_write() may have instantiated a few blocks
1924 * outside i_size. Trim these off again.
1925 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 if (pos + bytes > isize)
1927 vmtruncate(inode, isize);
1928 break;
1929 }
1930 if (likely(nr_segs == 1))
1931 copied = filemap_copy_from_user(page, offset,
1932 buf, bytes);
1933 else
1934 copied = filemap_copy_from_user_iovec(page, offset,
1935 cur_iov, iov_base, bytes);
1936 flush_dcache_page(page);
1937 status = a_ops->commit_write(file, page, offset, offset+bytes);
Zach Brown994fc28c2005-12-15 14:28:17 -08001938 if (status == AOP_TRUNCATED_PAGE) {
1939 page_cache_release(page);
1940 continue;
1941 }
Andrew Morton81b0c872006-06-29 02:24:26 -07001942zero_length_segment:
1943 if (likely(copied >= 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 if (!status)
1945 status = copied;
1946
1947 if (status >= 0) {
1948 written += status;
1949 count -= status;
1950 pos += status;
1951 buf += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07001952 if (unlikely(nr_segs > 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 filemap_set_next_iovec(&cur_iov,
1954 &iov_base, status);
Badari Pulavartyb0cfbd92005-06-25 14:55:42 -07001955 if (count)
1956 buf = cur_iov->iov_base +
1957 iov_base;
Martin Schwidefskya5117182005-06-06 13:35:54 -07001958 } else {
1959 iov_base += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07001960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 }
1962 }
1963 if (unlikely(copied != bytes))
1964 if (status >= 0)
1965 status = -EFAULT;
1966 unlock_page(page);
1967 mark_page_accessed(page);
1968 page_cache_release(page);
1969 if (status < 0)
1970 break;
1971 balance_dirty_pages_ratelimited(mapping);
1972 cond_resched();
1973 } while (count);
1974 *ppos = pos;
1975
1976 if (cached_page)
1977 page_cache_release(cached_page);
1978
1979 /*
1980 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
1981 */
1982 if (likely(status >= 0)) {
1983 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
1984 if (!a_ops->writepage || !is_sync_kiocb(iocb))
1985 status = generic_osync_inode(inode, mapping,
1986 OSYNC_METADATA|OSYNC_DATA);
1987 }
1988 }
1989
1990 /*
1991 * If we get here for O_DIRECT writes then we must have fallen through
1992 * to buffered writes (block instantiation inside i_size). So we sync
1993 * the file data here, to try to honour O_DIRECT expectations.
1994 */
1995 if (unlikely(file->f_flags & O_DIRECT) && written)
1996 status = filemap_write_and_wait(mapping);
1997
1998 pagevec_lru_add(&lru_pvec);
1999 return written ? written : status;
2000}
2001EXPORT_SYMBOL(generic_file_buffered_write);
2002
Adrian Bunk5ce78522005-09-10 00:26:28 -07002003static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2005 unsigned long nr_segs, loff_t *ppos)
2006{
2007 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002008 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 size_t ocount; /* original count */
2010 size_t count; /* after file limit checks */
2011 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 loff_t pos;
2013 ssize_t written;
2014 ssize_t err;
2015
2016 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002017 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2018 if (err)
2019 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
2021 count = ocount;
2022 pos = *ppos;
2023
2024 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2025
2026 /* We can write back this queue in page reclaim */
2027 current->backing_dev_info = mapping->backing_dev_info;
2028 written = 0;
2029
2030 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2031 if (err)
2032 goto out;
2033
2034 if (count == 0)
2035 goto out;
2036
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002037 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 if (err)
2039 goto out;
2040
Christoph Hellwig870f4812006-01-09 20:52:01 -08002041 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
2043 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2044 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002045 loff_t endbyte;
2046 ssize_t written_buffered;
2047
2048 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2049 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (written < 0 || written == count)
2051 goto out;
2052 /*
2053 * direct-io write to a hole: fall through to buffered I/O
2054 * for completing the rest of the request.
2055 */
2056 pos += written;
2057 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002058 written_buffered = generic_file_buffered_write(iocb, iov,
2059 nr_segs, pos, ppos, count,
2060 written);
2061 /*
2062 * If generic_file_buffered_write() retuned a synchronous error
2063 * then we want to return the number of bytes which were
2064 * direct-written, or the error code if that was zero. Note
2065 * that this differs from normal direct-io semantics, which
2066 * will return -EFOO even if some bytes were written.
2067 */
2068 if (written_buffered < 0) {
2069 err = written_buffered;
2070 goto out;
2071 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002073 /*
2074 * We need to ensure that the page cache pages are written to
2075 * disk and invalidated to preserve the expected O_DIRECT
2076 * semantics.
2077 */
2078 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002079 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2080 SYNC_FILE_RANGE_WAIT_BEFORE|
2081 SYNC_FILE_RANGE_WRITE|
2082 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002083 if (err == 0) {
2084 written = written_buffered;
2085 invalidate_mapping_pages(mapping,
2086 pos >> PAGE_CACHE_SHIFT,
2087 endbyte >> PAGE_CACHE_SHIFT);
2088 } else {
2089 /*
2090 * We don't know how much we wrote, so just return
2091 * the number of bytes which were direct-written
2092 */
2093 }
2094 } else {
2095 written = generic_file_buffered_write(iocb, iov, nr_segs,
2096 pos, ppos, count, written);
2097 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098out:
2099 current->backing_dev_info = NULL;
2100 return written ? written : err;
2101}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
Badari Pulavarty027445c2006-09-30 23:28:46 -07002103ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2104 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105{
2106 struct file *file = iocb->ki_filp;
2107 struct address_space *mapping = file->f_mapping;
2108 struct inode *inode = mapping->host;
2109 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
Badari Pulavarty027445c2006-09-30 23:28:46 -07002111 BUG_ON(iocb->ki_pos != pos);
2112
2113 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2114 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
2116 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002117 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 err = sync_page_range_nolock(inode, mapping, pos, ret);
2120 if (err < 0)
2121 ret = err;
2122 }
2123 return ret;
2124}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002125EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
Badari Pulavarty027445c2006-09-30 23:28:46 -07002127ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2128 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129{
2130 struct file *file = iocb->ki_filp;
2131 struct address_space *mapping = file->f_mapping;
2132 struct inode *inode = mapping->host;
2133 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
2135 BUG_ON(iocb->ki_pos != pos);
2136
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002137 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002138 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2139 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002140 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
2142 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2143 ssize_t err;
2144
2145 err = sync_page_range(inode, mapping, pos, ret);
2146 if (err < 0)
2147 ret = err;
2148 }
2149 return ret;
2150}
2151EXPORT_SYMBOL(generic_file_aio_write);
2152
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002154 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 * went wrong during pagecache shootdown.
2156 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002157static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2159 loff_t offset, unsigned long nr_segs)
2160{
2161 struct file *file = iocb->ki_filp;
2162 struct address_space *mapping = file->f_mapping;
2163 ssize_t retval;
Zach Brown65b82912007-03-16 13:38:11 -08002164 size_t write_len;
2165 pgoff_t end = 0; /* silence gcc */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
2167 /*
2168 * If it's a write, unmap all mmappings of the file up-front. This
2169 * will cause any pte dirty bits to be propagated into the pageframes
2170 * for the subsequent filemap_write_and_wait().
2171 */
2172 if (rw == WRITE) {
2173 write_len = iov_length(iov, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002174 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 if (mapping_mapped(mapping))
2176 unmap_mapping_range(mapping, offset, write_len, 0);
2177 }
2178
2179 retval = filemap_write_and_wait(mapping);
Zach Brown65b82912007-03-16 13:38:11 -08002180 if (retval)
2181 goto out;
2182
2183 /*
2184 * After a write we want buffered reads to be sure to go to disk to get
2185 * the new data. We invalidate clean cached page from the region we're
2186 * about to write. We do this *before* the write so that we can return
2187 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2188 */
2189 if (rw == WRITE && mapping->nrpages) {
2190 retval = invalidate_inode_pages2_range(mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002192 if (retval)
2193 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 }
Zach Brown65b82912007-03-16 13:38:11 -08002195
2196 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2197 if (retval)
2198 goto out;
2199
2200 /*
2201 * Finally, try again to invalidate clean pages which might have been
2202 * faulted in by get_user_pages() if the source of the write was an
2203 * mmap()ed region of the file we're writing. That's a pretty crazy
2204 * thing to do, so we don't support it 100%. If this invalidation
2205 * fails and we have -EIOCBQUEUED we ignore the failure.
2206 */
2207 if (rw == WRITE && mapping->nrpages) {
2208 int err = invalidate_inode_pages2_range(mapping,
2209 offset >> PAGE_CACHE_SHIFT, end);
2210 if (err && retval >= 0)
2211 retval = err;
2212 }
2213out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 return retval;
2215}
David Howellscf9a2ae2006-08-29 19:05:54 +01002216
2217/**
2218 * try_to_release_page() - release old fs-specific metadata on a page
2219 *
2220 * @page: the page which the kernel is trying to free
2221 * @gfp_mask: memory allocation flags (and I/O mode)
2222 *
2223 * The address_space is to try to release any data against the page
2224 * (presumably at page->private). If the release was successful, return `1'.
2225 * Otherwise return zero.
2226 *
2227 * The @gfp_mask argument specifies whether I/O may be performed to release
2228 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2229 *
2230 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2231 */
2232int try_to_release_page(struct page *page, gfp_t gfp_mask)
2233{
2234 struct address_space * const mapping = page->mapping;
2235
2236 BUG_ON(!PageLocked(page));
2237 if (PageWriteback(page))
2238 return 0;
2239
2240 if (mapping && mapping->a_ops->releasepage)
2241 return mapping->a_ops->releasepage(page, gfp_mask);
2242 return try_to_free_buffers(page);
2243}
2244
2245EXPORT_SYMBOL(try_to_release_page);