blob: 5d5449f3d41c83810e002d7fc41cf0cb55b88536 [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
846 * @_ra: file's readahead state
847 * @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();
894 if (index == next_index)
895 next_index = page_cache_readahead(mapping, &ra, filp,
896 index, last_index - index);
897
898find_page:
899 page = find_get_page(mapping, index);
900 if (unlikely(page == NULL)) {
901 handle_ra_miss(mapping, &ra, index);
902 goto no_cached_page;
903 }
904 if (!PageUptodate(page))
905 goto page_not_up_to_date;
906page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -0700907 /*
908 * i_size must be checked after we know the page is Uptodate.
909 *
910 * Checking i_size after the check allows us to calculate
911 * the correct value for "nr", which means the zero-filled
912 * part of the page is not copied back to userspace (unless
913 * another truncate extends the file - this is desired though).
914 */
915
916 isize = i_size_read(inode);
917 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
918 if (unlikely(!isize || index > end_index)) {
919 page_cache_release(page);
920 goto out;
921 }
922
923 /* nr is the maximum number of bytes to copy from this page */
924 nr = PAGE_CACHE_SIZE;
925 if (index == end_index) {
926 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
927 if (nr <= offset) {
928 page_cache_release(page);
929 goto out;
930 }
931 }
932 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934 /* If users can be writing to this page using arbitrary
935 * virtual addresses, take care about potential aliasing
936 * before reading the page on the kernel side.
937 */
938 if (mapping_writably_mapped(mapping))
939 flush_dcache_page(page);
940
941 /*
Jan Karaec0f1632007-05-06 14:49:25 -0700942 * When a sequential read accesses a page several times,
943 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 */
Jan Karaec0f1632007-05-06 14:49:25 -0700945 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 mark_page_accessed(page);
947 prev_index = index;
948
949 /*
950 * Ok, we have the page, and it's up-to-date, so
951 * now we can copy it to user space...
952 *
953 * The actor routine returns how many bytes were actually used..
954 * NOTE! This may not be the same as how much of a user buffer
955 * we filled up (we may be padding etc), so we can only update
956 * "pos" here (the actor routine has to update the user buffer
957 * pointers and the remaining count).
958 */
959 ret = actor(desc, page, offset, nr);
960 offset += ret;
961 index += offset >> PAGE_CACHE_SHIFT;
962 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -0700963 prev_offset = offset;
964 ra.prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
966 page_cache_release(page);
967 if (ret == nr && desc->count)
968 continue;
969 goto out;
970
971page_not_up_to_date:
972 /* Get exclusive access to the page ... */
973 lock_page(page);
974
Nick Pigginda6052f2006-09-25 23:31:35 -0700975 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (!page->mapping) {
977 unlock_page(page);
978 page_cache_release(page);
979 continue;
980 }
981
982 /* Did somebody else fill it already? */
983 if (PageUptodate(page)) {
984 unlock_page(page);
985 goto page_ok;
986 }
987
988readpage:
989 /* Start the actual read. The read will unlock the page. */
990 error = mapping->a_ops->readpage(filp, page);
991
Zach Brown994fc28c2005-12-15 14:28:17 -0800992 if (unlikely(error)) {
993 if (error == AOP_TRUNCATED_PAGE) {
994 page_cache_release(page);
995 goto find_page;
996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -0800998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
1000 if (!PageUptodate(page)) {
1001 lock_page(page);
1002 if (!PageUptodate(page)) {
1003 if (page->mapping == NULL) {
1004 /*
1005 * invalidate_inode_pages got it
1006 */
1007 unlock_page(page);
1008 page_cache_release(page);
1009 goto find_page;
1010 }
1011 unlock_page(page);
1012 error = -EIO;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001013 shrink_readahead_size_eio(filp, &ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 goto readpage_error;
1015 }
1016 unlock_page(page);
1017 }
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 goto page_ok;
1020
1021readpage_error:
1022 /* UHHUH! A synchronous read error occurred. Report it */
1023 desc->error = error;
1024 page_cache_release(page);
1025 goto out;
1026
1027no_cached_page:
1028 /*
1029 * Ok, it wasn't cached, so we need to create a new
1030 * page..
1031 */
1032 if (!cached_page) {
1033 cached_page = page_cache_alloc_cold(mapping);
1034 if (!cached_page) {
1035 desc->error = -ENOMEM;
1036 goto out;
1037 }
1038 }
1039 error = add_to_page_cache_lru(cached_page, mapping,
1040 index, GFP_KERNEL);
1041 if (error) {
1042 if (error == -EEXIST)
1043 goto find_page;
1044 desc->error = error;
1045 goto out;
1046 }
1047 page = cached_page;
1048 cached_page = NULL;
1049 goto readpage;
1050 }
1051
1052out:
1053 *_ra = ra;
1054
1055 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1056 if (cached_page)
1057 page_cache_release(cached_page);
1058 if (filp)
1059 file_accessed(filp);
1060}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061EXPORT_SYMBOL(do_generic_mapping_read);
1062
1063int file_read_actor(read_descriptor_t *desc, struct page *page,
1064 unsigned long offset, unsigned long size)
1065{
1066 char *kaddr;
1067 unsigned long left, count = desc->count;
1068
1069 if (size > count)
1070 size = count;
1071
1072 /*
1073 * Faults on the destination of a read are common, so do it before
1074 * taking the kmap.
1075 */
1076 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1077 kaddr = kmap_atomic(page, KM_USER0);
1078 left = __copy_to_user_inatomic(desc->arg.buf,
1079 kaddr + offset, size);
1080 kunmap_atomic(kaddr, KM_USER0);
1081 if (left == 0)
1082 goto success;
1083 }
1084
1085 /* Do it the slow way */
1086 kaddr = kmap(page);
1087 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1088 kunmap(page);
1089
1090 if (left) {
1091 size -= left;
1092 desc->error = -EFAULT;
1093 }
1094success:
1095 desc->count = count - size;
1096 desc->written += size;
1097 desc->arg.buf += size;
1098 return size;
1099}
1100
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001101/*
1102 * Performs necessary checks before doing a write
1103 * @iov: io vector request
1104 * @nr_segs: number of segments in the iovec
1105 * @count: number of bytes to write
1106 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1107 *
1108 * Adjust number of segments and amount of bytes to write (nr_segs should be
1109 * properly initialized first). Returns appropriate error code that caller
1110 * should return or zero in case that write should be allowed.
1111 */
1112int generic_segment_checks(const struct iovec *iov,
1113 unsigned long *nr_segs, size_t *count, int access_flags)
1114{
1115 unsigned long seg;
1116 size_t cnt = 0;
1117 for (seg = 0; seg < *nr_segs; seg++) {
1118 const struct iovec *iv = &iov[seg];
1119
1120 /*
1121 * If any segment has a negative length, or the cumulative
1122 * length ever wraps negative then return -EINVAL.
1123 */
1124 cnt += iv->iov_len;
1125 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1126 return -EINVAL;
1127 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1128 continue;
1129 if (seg == 0)
1130 return -EFAULT;
1131 *nr_segs = seg;
1132 cnt -= iv->iov_len; /* This segment is no good */
1133 break;
1134 }
1135 *count = cnt;
1136 return 0;
1137}
1138EXPORT_SYMBOL(generic_segment_checks);
1139
Randy Dunlap485bb992006-06-23 02:03:49 -07001140/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001141 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001142 * @iocb: kernel I/O control block
1143 * @iov: io vector request
1144 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001145 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001146 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 * This is the "read()" routine for all filesystems
1148 * that can use the page cache directly.
1149 */
1150ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001151generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1152 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153{
1154 struct file *filp = iocb->ki_filp;
1155 ssize_t retval;
1156 unsigned long seg;
1157 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001158 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001161 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1162 if (retval)
1163 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1166 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001167 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 struct address_space *mapping;
1169 struct inode *inode;
1170
1171 mapping = filp->f_mapping;
1172 inode = mapping->host;
1173 retval = 0;
1174 if (!count)
1175 goto out; /* skip atime */
1176 size = i_size_read(inode);
1177 if (pos < size) {
1178 retval = generic_file_direct_IO(READ, iocb,
1179 iov, pos, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 if (retval > 0)
1181 *ppos = pos + retval;
1182 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001183 if (likely(retval != 0)) {
Steven Whitehouse3f1a9aa2006-09-27 14:52:48 -04001184 file_accessed(filp);
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001185 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 }
1188
1189 retval = 0;
1190 if (count) {
1191 for (seg = 0; seg < nr_segs; seg++) {
1192 read_descriptor_t desc;
1193
1194 desc.written = 0;
1195 desc.arg.buf = iov[seg].iov_base;
1196 desc.count = iov[seg].iov_len;
1197 if (desc.count == 0)
1198 continue;
1199 desc.error = 0;
1200 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1201 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001202 if (desc.error) {
1203 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 break;
1205 }
akpm@linux-foundation.orgc44939e2007-07-15 23:38:25 -07001206 if (desc.count > 0)
1207 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
1209 }
1210out:
1211 return retval;
1212}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213EXPORT_SYMBOL(generic_file_aio_read);
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1216{
1217 ssize_t written;
1218 unsigned long count = desc->count;
1219 struct file *file = desc->arg.data;
1220
1221 if (size > count)
1222 size = count;
1223
1224 written = file->f_op->sendpage(file, page, offset,
1225 size, &file->f_pos, size<count);
1226 if (written < 0) {
1227 desc->error = written;
1228 written = 0;
1229 }
1230 desc->count = count - written;
1231 desc->written += written;
1232 return written;
1233}
1234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235static ssize_t
1236do_readahead(struct address_space *mapping, struct file *filp,
1237 unsigned long index, unsigned long nr)
1238{
1239 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1240 return -EINVAL;
1241
1242 force_page_cache_readahead(mapping, filp, index,
1243 max_sane_readahead(nr));
1244 return 0;
1245}
1246
1247asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1248{
1249 ssize_t ret;
1250 struct file *file;
1251
1252 ret = -EBADF;
1253 file = fget(fd);
1254 if (file) {
1255 if (file->f_mode & FMODE_READ) {
1256 struct address_space *mapping = file->f_mapping;
1257 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1258 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1259 unsigned long len = end - start + 1;
1260 ret = do_readahead(mapping, file, start, len);
1261 }
1262 fput(file);
1263 }
1264 return ret;
1265}
1266
1267#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001268static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1269/**
1270 * page_cache_read - adds requested page to the page cache if not already there
1271 * @file: file to read
1272 * @offset: page index
1273 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 * This adds the requested page to the page cache if it isn't already there,
1275 * and schedules an I/O to read in its contents from disk.
1276 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277static int fastcall page_cache_read(struct file * file, unsigned long offset)
1278{
1279 struct address_space *mapping = file->f_mapping;
1280 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001281 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
Zach Brown994fc28c2005-12-15 14:28:17 -08001283 do {
1284 page = page_cache_alloc_cold(mapping);
1285 if (!page)
1286 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
Zach Brown994fc28c2005-12-15 14:28:17 -08001288 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1289 if (ret == 0)
1290 ret = mapping->a_ops->readpage(file, page);
1291 else if (ret == -EEXIST)
1292 ret = 0; /* losing race to add is OK */
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Zach Brown994fc28c2005-12-15 14:28:17 -08001296 } while (ret == AOP_TRUNCATED_PAGE);
1297
1298 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299}
1300
1301#define MMAP_LOTSAMISS (100)
1302
Randy Dunlap485bb992006-06-23 02:03:49 -07001303/**
1304 * filemap_nopage - read in file data for page fault handling
1305 * @area: the applicable vm_area
1306 * @address: target address to read in
1307 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1308 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 * filemap_nopage() is invoked via the vma operations vector for a
1310 * mapped memory region to read in file data during a page fault.
1311 *
1312 * The goto's are kind of ugly, but this streamlines the normal case of having
1313 * it in the page cache, and handles the special cases reasonably without
1314 * having a lot of duplicated code.
1315 */
1316struct page *filemap_nopage(struct vm_area_struct *area,
1317 unsigned long address, int *type)
1318{
1319 int error;
1320 struct file *file = area->vm_file;
1321 struct address_space *mapping = file->f_mapping;
1322 struct file_ra_state *ra = &file->f_ra;
1323 struct inode *inode = mapping->host;
1324 struct page *page;
1325 unsigned long size, pgoff;
1326 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1327
1328 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1329
1330retry_all:
1331 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1332 if (pgoff >= size)
1333 goto outside_data_content;
1334
1335 /* If we don't want any read-ahead, don't bother */
1336 if (VM_RandomReadHint(area))
1337 goto no_cached_page;
1338
1339 /*
1340 * The readahead code wants to be told about each and every page
1341 * so it can build and shrink its windows appropriately
1342 *
1343 * For sequential accesses, we use the generic readahead logic.
1344 */
1345 if (VM_SequentialReadHint(area))
1346 page_cache_readahead(mapping, ra, file, pgoff, 1);
1347
1348 /*
1349 * Do we have something in the page cache already?
1350 */
1351retry_find:
1352 page = find_get_page(mapping, pgoff);
1353 if (!page) {
1354 unsigned long ra_pages;
1355
1356 if (VM_SequentialReadHint(area)) {
1357 handle_ra_miss(mapping, ra, pgoff);
1358 goto no_cached_page;
1359 }
1360 ra->mmap_miss++;
1361
1362 /*
1363 * Do we miss much more than hit in this file? If so,
1364 * stop bothering with read-ahead. It will only hurt.
1365 */
1366 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1367 goto no_cached_page;
1368
1369 /*
1370 * To keep the pgmajfault counter straight, we need to
1371 * check did_readaround, as this is an inner loop.
1372 */
1373 if (!did_readaround) {
1374 majmin = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001375 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
1377 did_readaround = 1;
1378 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1379 if (ra_pages) {
1380 pgoff_t start = 0;
1381
1382 if (pgoff > ra_pages / 2)
1383 start = pgoff - ra_pages / 2;
1384 do_page_cache_readahead(mapping, file, start, ra_pages);
1385 }
1386 page = find_get_page(mapping, pgoff);
1387 if (!page)
1388 goto no_cached_page;
1389 }
1390
1391 if (!did_readaround)
1392 ra->mmap_hit++;
1393
1394 /*
1395 * Ok, found a page in the page cache, now we need to check
1396 * that it's up-to-date.
1397 */
1398 if (!PageUptodate(page))
1399 goto page_not_uptodate;
1400
1401success:
1402 /*
1403 * Found the page and have a reference on it.
1404 */
1405 mark_page_accessed(page);
1406 if (type)
1407 *type = majmin;
1408 return page;
1409
1410outside_data_content:
1411 /*
1412 * An external ptracer can access pages that normally aren't
1413 * accessible..
1414 */
1415 if (area->vm_mm == current->mm)
Adam Litke79f5acf2006-09-29 01:58:43 -07001416 return NOPAGE_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 /* Fall through to the non-read-ahead case */
1418no_cached_page:
1419 /*
1420 * We're only likely to ever get here if MADV_RANDOM is in
1421 * effect.
1422 */
1423 error = page_cache_read(file, pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425 /*
1426 * The page we want has now been added to the page cache.
1427 * In the unlikely event that someone removed it in the
1428 * meantime, we'll just come back here and read it again.
1429 */
1430 if (error >= 0)
1431 goto retry_find;
1432
1433 /*
1434 * An error return from page_cache_read can result if the
1435 * system is low on memory, or a problem occurs while trying
1436 * to schedule I/O.
1437 */
1438 if (error == -ENOMEM)
1439 return NOPAGE_OOM;
Adam Litke79f5acf2006-09-29 01:58:43 -07001440 return NOPAGE_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442page_not_uptodate:
1443 if (!did_readaround) {
1444 majmin = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001445 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
1448 /*
1449 * Umm, take care of errors if the page isn't up-to-date.
1450 * Try to re-read it _once_. We do this synchronously,
1451 * because there really aren't any performance issues here
1452 * and we need to check for errors.
1453 */
1454 lock_page(page);
1455
1456 /* Somebody truncated the page on us? */
1457 if (!page->mapping) {
1458 unlock_page(page);
1459 page_cache_release(page);
1460 goto retry_all;
1461 }
1462
1463 /* Somebody else successfully read it in? */
1464 if (PageUptodate(page)) {
1465 unlock_page(page);
1466 goto success;
1467 }
1468 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001469 error = mapping->a_ops->readpage(file, page);
1470 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 wait_on_page_locked(page);
1472 if (PageUptodate(page))
1473 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001474 } else if (error == AOP_TRUNCATED_PAGE) {
1475 page_cache_release(page);
1476 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 }
1478
1479 /*
1480 * Things didn't work out. Return zero to tell the
1481 * mm layer so, possibly freeing the page cache page first.
1482 */
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001483 shrink_readahead_size_eio(file, ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 page_cache_release(page);
Adam Litke79f5acf2006-09-29 01:58:43 -07001485 return NOPAGE_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487EXPORT_SYMBOL(filemap_nopage);
1488
1489static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1490 int nonblock)
1491{
1492 struct address_space *mapping = file->f_mapping;
1493 struct page *page;
1494 int error;
1495
1496 /*
1497 * Do we have something in the page cache already?
1498 */
1499retry_find:
1500 page = find_get_page(mapping, pgoff);
1501 if (!page) {
1502 if (nonblock)
1503 return NULL;
1504 goto no_cached_page;
1505 }
1506
1507 /*
1508 * Ok, found a page in the page cache, now we need to check
1509 * that it's up-to-date.
1510 */
Jeff Moyerd3457342005-04-16 15:24:05 -07001511 if (!PageUptodate(page)) {
1512 if (nonblock) {
1513 page_cache_release(page);
1514 return NULL;
1515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 goto page_not_uptodate;
Jeff Moyerd3457342005-04-16 15:24:05 -07001517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
1519success:
1520 /*
1521 * Found the page and have a reference on it.
1522 */
1523 mark_page_accessed(page);
1524 return page;
1525
1526no_cached_page:
1527 error = page_cache_read(file, pgoff);
1528
1529 /*
1530 * The page we want has now been added to the page cache.
1531 * In the unlikely event that someone removed it in the
1532 * meantime, we'll just come back here and read it again.
1533 */
1534 if (error >= 0)
1535 goto retry_find;
1536
1537 /*
1538 * An error return from page_cache_read can result if the
1539 * system is low on memory, or a problem occurs while trying
1540 * to schedule I/O.
1541 */
1542 return NULL;
1543
1544page_not_uptodate:
1545 lock_page(page);
1546
Nick Pigginda6052f2006-09-25 23:31:35 -07001547 /* Did it get truncated while we waited for it? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 if (!page->mapping) {
1549 unlock_page(page);
1550 goto err;
1551 }
1552
1553 /* Did somebody else get it up-to-date? */
1554 if (PageUptodate(page)) {
1555 unlock_page(page);
1556 goto success;
1557 }
1558
Zach Brown994fc28c2005-12-15 14:28:17 -08001559 error = mapping->a_ops->readpage(file, page);
1560 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 wait_on_page_locked(page);
1562 if (PageUptodate(page))
1563 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001564 } else if (error == AOP_TRUNCATED_PAGE) {
1565 page_cache_release(page);
1566 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 }
1568
1569 /*
1570 * Umm, take care of errors if the page isn't up-to-date.
1571 * Try to re-read it _once_. We do this synchronously,
1572 * because there really aren't any performance issues here
1573 * and we need to check for errors.
1574 */
1575 lock_page(page);
1576
1577 /* Somebody truncated the page on us? */
1578 if (!page->mapping) {
1579 unlock_page(page);
1580 goto err;
1581 }
1582 /* Somebody else successfully read it in? */
1583 if (PageUptodate(page)) {
1584 unlock_page(page);
1585 goto success;
1586 }
1587
1588 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001589 error = mapping->a_ops->readpage(file, page);
1590 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 wait_on_page_locked(page);
1592 if (PageUptodate(page))
1593 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001594 } else if (error == AOP_TRUNCATED_PAGE) {
1595 page_cache_release(page);
1596 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 }
1598
1599 /*
1600 * Things didn't work out. Return zero to tell the
1601 * mm layer so, possibly freeing the page cache page first.
1602 */
1603err:
1604 page_cache_release(page);
1605
1606 return NULL;
1607}
1608
1609int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1610 unsigned long len, pgprot_t prot, unsigned long pgoff,
1611 int nonblock)
1612{
1613 struct file *file = vma->vm_file;
1614 struct address_space *mapping = file->f_mapping;
1615 struct inode *inode = mapping->host;
1616 unsigned long size;
1617 struct mm_struct *mm = vma->vm_mm;
1618 struct page *page;
1619 int err;
1620
1621 if (!nonblock)
1622 force_page_cache_readahead(mapping, vma->vm_file,
1623 pgoff, len >> PAGE_CACHE_SHIFT);
1624
1625repeat:
1626 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1627 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1628 return -EINVAL;
1629
1630 page = filemap_getpage(file, pgoff, nonblock);
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001631
1632 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1633 * done in shmem_populate calling shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 if (!page && !nonblock)
1635 return -ENOMEM;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001636
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (page) {
1638 err = install_page(mm, vma, addr, page, prot);
1639 if (err) {
1640 page_cache_release(page);
1641 return err;
1642 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001643 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001644 /* No page was found just because we can't read it in now (being
1645 * here implies nonblock != 0), but the page may exist, so set
1646 * the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 err = install_file_pte(mm, vma, addr, pgoff, prot);
1648 if (err)
1649 return err;
1650 }
1651
1652 len -= PAGE_SIZE;
1653 addr += PAGE_SIZE;
1654 pgoff++;
1655 if (len)
1656 goto repeat;
1657
1658 return 0;
1659}
Nikita Danilovb1459462005-10-29 18:17:02 -07001660EXPORT_SYMBOL(filemap_populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
1662struct vm_operations_struct generic_file_vm_ops = {
1663 .nopage = filemap_nopage,
1664 .populate = filemap_populate,
1665};
1666
1667/* This is used for a general mmap of a disk file */
1668
1669int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1670{
1671 struct address_space *mapping = file->f_mapping;
1672
1673 if (!mapping->a_ops->readpage)
1674 return -ENOEXEC;
1675 file_accessed(file);
1676 vma->vm_ops = &generic_file_vm_ops;
1677 return 0;
1678}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
1680/*
1681 * This is for filesystems which do not implement ->writepage.
1682 */
1683int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1684{
1685 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1686 return -EINVAL;
1687 return generic_file_mmap(file, vma);
1688}
1689#else
1690int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1691{
1692 return -ENOSYS;
1693}
1694int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1695{
1696 return -ENOSYS;
1697}
1698#endif /* CONFIG_MMU */
1699
1700EXPORT_SYMBOL(generic_file_mmap);
1701EXPORT_SYMBOL(generic_file_readonly_mmap);
1702
Nick Piggin6fe69002007-05-06 14:49:04 -07001703static struct page *__read_cache_page(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 unsigned long index,
1705 int (*filler)(void *,struct page*),
1706 void *data)
1707{
1708 struct page *page, *cached_page = NULL;
1709 int err;
1710repeat:
1711 page = find_get_page(mapping, index);
1712 if (!page) {
1713 if (!cached_page) {
1714 cached_page = page_cache_alloc_cold(mapping);
1715 if (!cached_page)
1716 return ERR_PTR(-ENOMEM);
1717 }
1718 err = add_to_page_cache_lru(cached_page, mapping,
1719 index, GFP_KERNEL);
1720 if (err == -EEXIST)
1721 goto repeat;
1722 if (err < 0) {
1723 /* Presumably ENOMEM for radix tree node */
1724 page_cache_release(cached_page);
1725 return ERR_PTR(err);
1726 }
1727 page = cached_page;
1728 cached_page = NULL;
1729 err = filler(data, page);
1730 if (err < 0) {
1731 page_cache_release(page);
1732 page = ERR_PTR(err);
1733 }
1734 }
1735 if (cached_page)
1736 page_cache_release(cached_page);
1737 return page;
1738}
1739
Nick Piggin6fe69002007-05-06 14:49:04 -07001740/*
1741 * Same as read_cache_page, but don't wait for page to become unlocked
1742 * after submitting it to the filler.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001744struct page *read_cache_page_async(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 unsigned long index,
1746 int (*filler)(void *,struct page*),
1747 void *data)
1748{
1749 struct page *page;
1750 int err;
1751
1752retry:
1753 page = __read_cache_page(mapping, index, filler, data);
1754 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001755 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 if (PageUptodate(page))
1757 goto out;
1758
1759 lock_page(page);
1760 if (!page->mapping) {
1761 unlock_page(page);
1762 page_cache_release(page);
1763 goto retry;
1764 }
1765 if (PageUptodate(page)) {
1766 unlock_page(page);
1767 goto out;
1768 }
1769 err = filler(data, page);
1770 if (err < 0) {
1771 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001772 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 }
David Howellsc855ff32007-05-09 13:42:20 +01001774out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001775 mark_page_accessed(page);
1776 return page;
1777}
1778EXPORT_SYMBOL(read_cache_page_async);
1779
1780/**
1781 * read_cache_page - read into page cache, fill it if needed
1782 * @mapping: the page's address_space
1783 * @index: the page index
1784 * @filler: function to perform the read
1785 * @data: destination for read data
1786 *
1787 * Read into the page cache. If a page already exists, and PageUptodate() is
1788 * not set, try to fill the page then wait for it to become unlocked.
1789 *
1790 * If the page does not get brought uptodate, return -EIO.
1791 */
1792struct page *read_cache_page(struct address_space *mapping,
1793 unsigned long index,
1794 int (*filler)(void *,struct page*),
1795 void *data)
1796{
1797 struct page *page;
1798
1799 page = read_cache_page_async(mapping, index, filler, data);
1800 if (IS_ERR(page))
1801 goto out;
1802 wait_on_page_locked(page);
1803 if (!PageUptodate(page)) {
1804 page_cache_release(page);
1805 page = ERR_PTR(-EIO);
1806 }
1807 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 return page;
1809}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810EXPORT_SYMBOL(read_cache_page);
1811
1812/*
1813 * If the page was newly created, increment its refcount and add it to the
1814 * caller's lru-buffering pagevec. This function is specifically for
1815 * generic_file_write().
1816 */
1817static inline struct page *
1818__grab_cache_page(struct address_space *mapping, unsigned long index,
1819 struct page **cached_page, struct pagevec *lru_pvec)
1820{
1821 int err;
1822 struct page *page;
1823repeat:
1824 page = find_lock_page(mapping, index);
1825 if (!page) {
1826 if (!*cached_page) {
1827 *cached_page = page_cache_alloc(mapping);
1828 if (!*cached_page)
1829 return NULL;
1830 }
1831 err = add_to_page_cache(*cached_page, mapping,
1832 index, GFP_KERNEL);
1833 if (err == -EEXIST)
1834 goto repeat;
1835 if (err == 0) {
1836 page = *cached_page;
1837 page_cache_get(page);
1838 if (!pagevec_add(lru_pvec, page))
1839 __pagevec_lru_add(lru_pvec);
1840 *cached_page = NULL;
1841 }
1842 }
1843 return page;
1844}
1845
1846/*
1847 * The logic we want is
1848 *
1849 * if suid or (sgid and xgrp)
1850 * remove privs
1851 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001852int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853{
1854 mode_t mode = dentry->d_inode->i_mode;
1855 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
1857 /* suid always must be killed */
1858 if (unlikely(mode & S_ISUID))
1859 kill = ATTR_KILL_SUID;
1860
1861 /*
1862 * sgid without any exec bits is just a mandatory locking mark; leave
1863 * it alone. If some exec bits are set, it's a real sgid; kill it.
1864 */
1865 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1866 kill |= ATTR_KILL_SGID;
1867
Jens Axboe01de85e2006-10-17 19:50:36 +02001868 if (unlikely(kill && !capable(CAP_FSETID)))
1869 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Jens Axboe01de85e2006-10-17 19:50:36 +02001871 return 0;
1872}
Mark Fashehd23a1472006-10-17 17:05:18 -07001873EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001874
1875int __remove_suid(struct dentry *dentry, int kill)
1876{
1877 struct iattr newattrs;
1878
1879 newattrs.ia_valid = ATTR_FORCE | kill;
1880 return notify_change(dentry, &newattrs);
1881}
1882
1883int remove_suid(struct dentry *dentry)
1884{
1885 int kill = should_remove_suid(dentry);
1886
1887 if (unlikely(kill))
1888 return __remove_suid(dentry, kill);
1889
1890 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891}
1892EXPORT_SYMBOL(remove_suid);
1893
Carsten Otteceffc072005-06-23 22:05:25 -07001894size_t
NeilBrown01408c42006-06-25 05:47:58 -07001895__filemap_copy_from_user_iovec_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 const struct iovec *iov, size_t base, size_t bytes)
1897{
1898 size_t copied = 0, left = 0;
1899
1900 while (bytes) {
1901 char __user *buf = iov->iov_base + base;
1902 int copy = min(bytes, iov->iov_len - base);
1903
1904 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001905 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 copied += copy;
1907 bytes -= copy;
1908 vaddr += copy;
1909 iov++;
1910
NeilBrown01408c42006-06-25 05:47:58 -07001911 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 }
1914 return copied - left;
1915}
1916
1917/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 * Performs necessary checks before doing a write
1919 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001920 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 * Returns appropriate error code that caller should return or
1922 * zero in case that write should be allowed.
1923 */
1924inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1925{
1926 struct inode *inode = file->f_mapping->host;
1927 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1928
1929 if (unlikely(*pos < 0))
1930 return -EINVAL;
1931
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 if (!isblk) {
1933 /* FIXME: this is for backwards compatibility with 2.4 */
1934 if (file->f_flags & O_APPEND)
1935 *pos = i_size_read(inode);
1936
1937 if (limit != RLIM_INFINITY) {
1938 if (*pos >= limit) {
1939 send_sig(SIGXFSZ, current, 0);
1940 return -EFBIG;
1941 }
1942 if (*count > limit - (typeof(limit))*pos) {
1943 *count = limit - (typeof(limit))*pos;
1944 }
1945 }
1946 }
1947
1948 /*
1949 * LFS rule
1950 */
1951 if (unlikely(*pos + *count > MAX_NON_LFS &&
1952 !(file->f_flags & O_LARGEFILE))) {
1953 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return -EFBIG;
1955 }
1956 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1957 *count = MAX_NON_LFS - (unsigned long)*pos;
1958 }
1959 }
1960
1961 /*
1962 * Are we about to exceed the fs block limit ?
1963 *
1964 * If we have written data it becomes a short write. If we have
1965 * exceeded without writing data we send a signal and return EFBIG.
1966 * Linus frestrict idea will clean these up nicely..
1967 */
1968 if (likely(!isblk)) {
1969 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1970 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 return -EFBIG;
1972 }
1973 /* zero-length writes at ->s_maxbytes are OK */
1974 }
1975
1976 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1977 *count = inode->i_sb->s_maxbytes - *pos;
1978 } else {
David Howells93614012006-09-30 20:45:40 +02001979#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 loff_t isize;
1981 if (bdev_read_only(I_BDEV(inode)))
1982 return -EPERM;
1983 isize = i_size_read(inode);
1984 if (*pos >= isize) {
1985 if (*count || *pos > isize)
1986 return -ENOSPC;
1987 }
1988
1989 if (*pos + *count > isize)
1990 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02001991#else
1992 return -EPERM;
1993#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
1995 return 0;
1996}
1997EXPORT_SYMBOL(generic_write_checks);
1998
1999ssize_t
2000generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2001 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2002 size_t count, size_t ocount)
2003{
2004 struct file *file = iocb->ki_filp;
2005 struct address_space *mapping = file->f_mapping;
2006 struct inode *inode = mapping->host;
2007 ssize_t written;
2008
2009 if (count != ocount)
2010 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2011
2012 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2013 if (written > 0) {
2014 loff_t end = pos + written;
2015 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2016 i_size_write(inode, end);
2017 mark_inode_dirty(inode);
2018 }
2019 *ppos = end;
2020 }
2021
2022 /*
2023 * Sync the fs metadata but not the minor inode changes and
2024 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002025 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002026 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 */
Zach Brown8459d862006-12-10 02:21:05 -08002028 if ((written >= 0 || written == -EIOCBQUEUED) &&
2029 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002030 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2031 if (err < 0)
2032 written = err;
2033 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 return written;
2035}
2036EXPORT_SYMBOL(generic_file_direct_write);
2037
2038ssize_t
2039generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2040 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2041 size_t count, ssize_t written)
2042{
2043 struct file *file = iocb->ki_filp;
2044 struct address_space * mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002045 const struct address_space_operations *a_ops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 struct inode *inode = mapping->host;
2047 long status = 0;
2048 struct page *page;
2049 struct page *cached_page = NULL;
2050 size_t bytes;
2051 struct pagevec lru_pvec;
2052 const struct iovec *cur_iov = iov; /* current iovec */
2053 size_t iov_base = 0; /* offset in the current iovec */
2054 char __user *buf;
2055
2056 pagevec_init(&lru_pvec, 0);
2057
2058 /*
2059 * handle partial DIO write. Adjust cur_iov if needed.
2060 */
2061 if (likely(nr_segs == 1))
2062 buf = iov->iov_base + written;
2063 else {
2064 filemap_set_next_iovec(&cur_iov, &iov_base, written);
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002065 buf = cur_iov->iov_base + iov_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 }
2067
2068 do {
2069 unsigned long index;
2070 unsigned long offset;
2071 size_t copied;
2072
2073 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2074 index = pos >> PAGE_CACHE_SHIFT;
2075 bytes = PAGE_CACHE_SIZE - offset;
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07002076
2077 /* Limit the size of the copy to the caller's write size */
2078 bytes = min(bytes, count);
2079
NeilBrown29dbb3f2007-02-16 01:28:38 -08002080 /* We only need to worry about prefaulting when writes are from
2081 * user-space. NFSd uses vfs_writev with several non-aligned
2082 * segments in the vector, and limiting to one segment a time is
2083 * a noticeable performance for re-write
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07002084 */
NeilBrown29dbb3f2007-02-16 01:28:38 -08002085 if (!segment_eq(get_fs(), KERNEL_DS)) {
2086 /*
2087 * Limit the size of the copy to that of the current
2088 * segment, because fault_in_pages_readable() doesn't
2089 * know how to walk segments.
2090 */
2091 bytes = min(bytes, cur_iov->iov_len - iov_base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
NeilBrown29dbb3f2007-02-16 01:28:38 -08002093 /*
2094 * Bring in the user page that we will copy from
2095 * _first_. Otherwise there's a nasty deadlock on
2096 * copying from the same page as we're writing to,
2097 * without it being marked up-to-date.
2098 */
2099 fault_in_pages_readable(buf, bytes);
2100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2102 if (!page) {
2103 status = -ENOMEM;
2104 break;
2105 }
2106
Andrew Morton81b0c872006-06-29 02:24:26 -07002107 if (unlikely(bytes == 0)) {
2108 status = 0;
2109 copied = 0;
2110 goto zero_length_segment;
2111 }
2112
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2114 if (unlikely(status)) {
2115 loff_t isize = i_size_read(inode);
Zach Brown994fc28c2005-12-15 14:28:17 -08002116
2117 if (status != AOP_TRUNCATED_PAGE)
2118 unlock_page(page);
2119 page_cache_release(page);
2120 if (status == AOP_TRUNCATED_PAGE)
2121 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 /*
2123 * prepare_write() may have instantiated a few blocks
2124 * outside i_size. Trim these off again.
2125 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 if (pos + bytes > isize)
2127 vmtruncate(inode, isize);
2128 break;
2129 }
2130 if (likely(nr_segs == 1))
2131 copied = filemap_copy_from_user(page, offset,
2132 buf, bytes);
2133 else
2134 copied = filemap_copy_from_user_iovec(page, offset,
2135 cur_iov, iov_base, bytes);
2136 flush_dcache_page(page);
2137 status = a_ops->commit_write(file, page, offset, offset+bytes);
Zach Brown994fc28c2005-12-15 14:28:17 -08002138 if (status == AOP_TRUNCATED_PAGE) {
2139 page_cache_release(page);
2140 continue;
2141 }
Andrew Morton81b0c872006-06-29 02:24:26 -07002142zero_length_segment:
2143 if (likely(copied >= 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (!status)
2145 status = copied;
2146
2147 if (status >= 0) {
2148 written += status;
2149 count -= status;
2150 pos += status;
2151 buf += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002152 if (unlikely(nr_segs > 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 filemap_set_next_iovec(&cur_iov,
2154 &iov_base, status);
Badari Pulavartyb0cfbd92005-06-25 14:55:42 -07002155 if (count)
2156 buf = cur_iov->iov_base +
2157 iov_base;
Martin Schwidefskya5117182005-06-06 13:35:54 -07002158 } else {
2159 iov_base += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 }
2162 }
2163 if (unlikely(copied != bytes))
2164 if (status >= 0)
2165 status = -EFAULT;
2166 unlock_page(page);
2167 mark_page_accessed(page);
2168 page_cache_release(page);
2169 if (status < 0)
2170 break;
2171 balance_dirty_pages_ratelimited(mapping);
2172 cond_resched();
2173 } while (count);
2174 *ppos = pos;
2175
2176 if (cached_page)
2177 page_cache_release(cached_page);
2178
2179 /*
2180 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2181 */
2182 if (likely(status >= 0)) {
2183 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2184 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2185 status = generic_osync_inode(inode, mapping,
2186 OSYNC_METADATA|OSYNC_DATA);
2187 }
2188 }
2189
2190 /*
2191 * If we get here for O_DIRECT writes then we must have fallen through
2192 * to buffered writes (block instantiation inside i_size). So we sync
2193 * the file data here, to try to honour O_DIRECT expectations.
2194 */
2195 if (unlikely(file->f_flags & O_DIRECT) && written)
2196 status = filemap_write_and_wait(mapping);
2197
2198 pagevec_lru_add(&lru_pvec);
2199 return written ? written : status;
2200}
2201EXPORT_SYMBOL(generic_file_buffered_write);
2202
Adrian Bunk5ce78522005-09-10 00:26:28 -07002203static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2205 unsigned long nr_segs, loff_t *ppos)
2206{
2207 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002208 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 size_t ocount; /* original count */
2210 size_t count; /* after file limit checks */
2211 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 loff_t pos;
2213 ssize_t written;
2214 ssize_t err;
2215
2216 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002217 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2218 if (err)
2219 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
2221 count = ocount;
2222 pos = *ppos;
2223
2224 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2225
2226 /* We can write back this queue in page reclaim */
2227 current->backing_dev_info = mapping->backing_dev_info;
2228 written = 0;
2229
2230 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2231 if (err)
2232 goto out;
2233
2234 if (count == 0)
2235 goto out;
2236
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002237 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 if (err)
2239 goto out;
2240
Christoph Hellwig870f4812006-01-09 20:52:01 -08002241 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
2243 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2244 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002245 loff_t endbyte;
2246 ssize_t written_buffered;
2247
2248 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2249 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 if (written < 0 || written == count)
2251 goto out;
2252 /*
2253 * direct-io write to a hole: fall through to buffered I/O
2254 * for completing the rest of the request.
2255 */
2256 pos += written;
2257 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002258 written_buffered = generic_file_buffered_write(iocb, iov,
2259 nr_segs, pos, ppos, count,
2260 written);
2261 /*
2262 * If generic_file_buffered_write() retuned a synchronous error
2263 * then we want to return the number of bytes which were
2264 * direct-written, or the error code if that was zero. Note
2265 * that this differs from normal direct-io semantics, which
2266 * will return -EFOO even if some bytes were written.
2267 */
2268 if (written_buffered < 0) {
2269 err = written_buffered;
2270 goto out;
2271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002273 /*
2274 * We need to ensure that the page cache pages are written to
2275 * disk and invalidated to preserve the expected O_DIRECT
2276 * semantics.
2277 */
2278 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002279 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2280 SYNC_FILE_RANGE_WAIT_BEFORE|
2281 SYNC_FILE_RANGE_WRITE|
2282 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002283 if (err == 0) {
2284 written = written_buffered;
2285 invalidate_mapping_pages(mapping,
2286 pos >> PAGE_CACHE_SHIFT,
2287 endbyte >> PAGE_CACHE_SHIFT);
2288 } else {
2289 /*
2290 * We don't know how much we wrote, so just return
2291 * the number of bytes which were direct-written
2292 */
2293 }
2294 } else {
2295 written = generic_file_buffered_write(iocb, iov, nr_segs,
2296 pos, ppos, count, written);
2297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298out:
2299 current->backing_dev_info = NULL;
2300 return written ? written : err;
2301}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
Badari Pulavarty027445c2006-09-30 23:28:46 -07002303ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2304 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305{
2306 struct file *file = iocb->ki_filp;
2307 struct address_space *mapping = file->f_mapping;
2308 struct inode *inode = mapping->host;
2309 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
Badari Pulavarty027445c2006-09-30 23:28:46 -07002311 BUG_ON(iocb->ki_pos != pos);
2312
2313 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2314 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315
2316 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002317 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
2319 err = sync_page_range_nolock(inode, mapping, pos, ret);
2320 if (err < 0)
2321 ret = err;
2322 }
2323 return ret;
2324}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002325EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
Badari Pulavarty027445c2006-09-30 23:28:46 -07002327ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2328 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329{
2330 struct file *file = iocb->ki_filp;
2331 struct address_space *mapping = file->f_mapping;
2332 struct inode *inode = mapping->host;
2333 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334
2335 BUG_ON(iocb->ki_pos != pos);
2336
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002337 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002338 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2339 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002340 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
2342 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2343 ssize_t err;
2344
2345 err = sync_page_range(inode, mapping, pos, ret);
2346 if (err < 0)
2347 ret = err;
2348 }
2349 return ret;
2350}
2351EXPORT_SYMBOL(generic_file_aio_write);
2352
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002354 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 * went wrong during pagecache shootdown.
2356 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002357static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2359 loff_t offset, unsigned long nr_segs)
2360{
2361 struct file *file = iocb->ki_filp;
2362 struct address_space *mapping = file->f_mapping;
2363 ssize_t retval;
Zach Brown65b82912007-03-16 13:38:11 -08002364 size_t write_len;
2365 pgoff_t end = 0; /* silence gcc */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
2367 /*
2368 * If it's a write, unmap all mmappings of the file up-front. This
2369 * will cause any pte dirty bits to be propagated into the pageframes
2370 * for the subsequent filemap_write_and_wait().
2371 */
2372 if (rw == WRITE) {
2373 write_len = iov_length(iov, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002374 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 if (mapping_mapped(mapping))
2376 unmap_mapping_range(mapping, offset, write_len, 0);
2377 }
2378
2379 retval = filemap_write_and_wait(mapping);
Zach Brown65b82912007-03-16 13:38:11 -08002380 if (retval)
2381 goto out;
2382
2383 /*
2384 * After a write we want buffered reads to be sure to go to disk to get
2385 * the new data. We invalidate clean cached page from the region we're
2386 * about to write. We do this *before* the write so that we can return
2387 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2388 */
2389 if (rw == WRITE && mapping->nrpages) {
2390 retval = invalidate_inode_pages2_range(mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002392 if (retval)
2393 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 }
Zach Brown65b82912007-03-16 13:38:11 -08002395
2396 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2397 if (retval)
2398 goto out;
2399
2400 /*
2401 * Finally, try again to invalidate clean pages which might have been
2402 * faulted in by get_user_pages() if the source of the write was an
2403 * mmap()ed region of the file we're writing. That's a pretty crazy
2404 * thing to do, so we don't support it 100%. If this invalidation
2405 * fails and we have -EIOCBQUEUED we ignore the failure.
2406 */
2407 if (rw == WRITE && mapping->nrpages) {
2408 int err = invalidate_inode_pages2_range(mapping,
2409 offset >> PAGE_CACHE_SHIFT, end);
2410 if (err && retval >= 0)
2411 retval = err;
2412 }
2413out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 return retval;
2415}
David Howellscf9a2ae2006-08-29 19:05:54 +01002416
2417/**
2418 * try_to_release_page() - release old fs-specific metadata on a page
2419 *
2420 * @page: the page which the kernel is trying to free
2421 * @gfp_mask: memory allocation flags (and I/O mode)
2422 *
2423 * The address_space is to try to release any data against the page
2424 * (presumably at page->private). If the release was successful, return `1'.
2425 * Otherwise return zero.
2426 *
2427 * The @gfp_mask argument specifies whether I/O may be performed to release
2428 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2429 *
2430 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2431 */
2432int try_to_release_page(struct page *page, gfp_t gfp_mask)
2433{
2434 struct address_space * const mapping = page->mapping;
2435
2436 BUG_ON(!PageLocked(page));
2437 if (PageWriteback(page))
2438 return 0;
2439
2440 if (mapping && mapping->a_ops->releasepage)
2441 return mapping->a_ops->releasepage(page, gfp_mask);
2442 return try_to_free_buffers(page);
2443}
2444
2445EXPORT_SYMBOL(try_to_release_page);