blob: 65d9d9e2b755e7dd7e662d4dd5cc45b19514a7ab [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
13#include <linux/slab.h>
14#include <linux/compiler.h>
15#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kernel_stat.h>
20#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080036#include "internal.h"
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 * FIXME: remove all knowledge of the buffer layer from the core VM
40 */
41#include <linux/buffer_head.h> /* for generic_osync_inode */
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <asm/mman.h>
44
Adrian Bunk5ce78522005-09-10 00:26:28 -070045static ssize_t
46generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
47 loff_t offset, unsigned long nr_segs);
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049/*
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * though.
52 *
53 * Shared mappings now work. 15.8.1995 Bruno.
54 *
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 *
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
59 */
60
61/*
62 * Lock ordering:
63 *
64 * ->i_mmap_lock (vmtruncate)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070066 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080069 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 * ->i_mmap_lock (truncate->unmap_mapping_range)
71 *
72 * ->mmap_sem
73 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070074 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 *
77 * ->mmap_sem
78 * ->lock_page (access_process_vm)
79 *
Nick Piggin82591e62006-10-19 23:29:10 -070080 * ->i_mutex (generic_file_buffered_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080083 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * ->i_alloc_sem (various)
85 *
86 * ->inode_lock
87 * ->sb_lock (fs/fs-writeback.c)
88 * ->mapping->tree_lock (__sync_single_inode)
89 *
90 * ->i_mmap_lock
91 * ->anon_vma.lock (vma_adjust)
92 *
93 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070096 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070097 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 * ->private_lock (try_to_unmap_one)
99 * ->tree_lock (try_to_unmap_one)
100 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -0800101 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 * ->private_lock (page_remove_rmap->set_page_dirty)
103 * ->tree_lock (page_remove_rmap->set_page_dirty)
104 * ->inode_lock (page_remove_rmap->set_page_dirty)
105 * ->inode_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 *
108 * ->task->proc_lock
109 * ->dcache_lock (proc_pid_lookup)
110 */
111
112/*
113 * Remove a page from the page cache and free it. Caller has to make
114 * sure the page is locked and that nobody else uses it - or that usage
115 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
116 */
117void __remove_from_page_cache(struct page *page)
118{
119 struct address_space *mapping = page->mapping;
120
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800121 mem_cgroup_uncharge_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 radix_tree_delete(&mapping->page_tree, page->index);
123 page->mapping = NULL;
124 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700125 __dec_zone_page_state(page, NR_FILE_PAGES);
Nick Piggin45426812007-07-15 23:38:12 -0700126 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800127
128 /*
129 * Some filesystems seem to re-dirty the page even after
130 * the VM has canceled the dirty bit (eg ext3 journaling).
131 *
132 * Fix it up by doing a final dirty accounting check after
133 * having removed the page entirely.
134 */
135 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
136 dec_zone_page_state(page, NR_FILE_DIRTY);
137 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139}
140
141void remove_from_page_cache(struct page *page)
142{
143 struct address_space *mapping = page->mapping;
144
Matt Mackallcd7619d2005-05-01 08:59:01 -0700145 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147 write_lock_irq(&mapping->tree_lock);
148 __remove_from_page_cache(page);
149 write_unlock_irq(&mapping->tree_lock);
150}
151
152static int sync_page(void *word)
153{
154 struct address_space *mapping;
155 struct page *page;
156
Andi Kleen07808b72005-11-05 17:25:53 +0100157 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700160 * page_mapping() is being called without PG_locked held.
161 * Some knowledge of the state and use of the page is used to
162 * reduce the requirements down to a memory barrier.
163 * The danger here is of a stale page_mapping() return value
164 * indicating a struct address_space different from the one it's
165 * associated with when it is associated with one.
166 * After smp_mb(), it's either the correct page_mapping() for
167 * the page, or an old page_mapping() and the page's own
168 * page_mapping() has gone NULL.
169 * The ->sync_page() address_space operation must tolerate
170 * page_mapping() going NULL. By an amazing coincidence,
171 * this comes about because none of the users of the page
172 * in the ->sync_page() methods make essential use of the
173 * page_mapping(), merely passing the page down to the backing
174 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700175 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700176 * of interest. When page_mapping() does go NULL, the entire
177 * call stack gracefully ignores the page and returns.
178 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 */
180 smp_mb();
181 mapping = page_mapping(page);
182 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
183 mapping->a_ops->sync_page(page);
184 io_schedule();
185 return 0;
186}
187
Matthew Wilcox2687a352007-12-06 11:18:49 -0500188static int sync_page_killable(void *word)
189{
190 sync_page(word);
191 return fatal_signal_pending(current) ? -EINTR : 0;
192}
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700195 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700196 * @mapping: address space structure to write
197 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800198 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700199 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700201 * Start writeback against all of a mapping's dirty pages that lie
202 * within the byte offsets <start, end> inclusive.
203 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700205 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 * these two operations is that if a dirty page/buffer is encountered, it must
207 * be waited upon, and not just skipped over.
208 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800209int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
210 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
212 int ret;
213 struct writeback_control wbc = {
214 .sync_mode = sync_mode,
215 .nr_to_write = mapping->nrpages * 2,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700216 .range_start = start,
217 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 };
219
220 if (!mapping_cap_writeback_dirty(mapping))
221 return 0;
222
223 ret = do_writepages(mapping, &wbc);
224 return ret;
225}
226
227static inline int __filemap_fdatawrite(struct address_space *mapping,
228 int sync_mode)
229{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700230 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
232
233int filemap_fdatawrite(struct address_space *mapping)
234{
235 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
236}
237EXPORT_SYMBOL(filemap_fdatawrite);
238
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400239int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800240 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
242 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
243}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400244EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Randy Dunlap485bb992006-06-23 02:03:49 -0700246/**
247 * filemap_flush - mostly a non-blocking flush
248 * @mapping: target address_space
249 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 * This is a mostly non-blocking flush. Not suitable for data-integrity
251 * purposes - I/O may not be started against all dirty pages.
252 */
253int filemap_flush(struct address_space *mapping)
254{
255 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
256}
257EXPORT_SYMBOL(filemap_flush);
258
Randy Dunlap485bb992006-06-23 02:03:49 -0700259/**
260 * wait_on_page_writeback_range - wait for writeback to complete
261 * @mapping: target address_space
262 * @start: beginning page index
263 * @end: ending page index
264 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 * Wait for writeback to complete against pages indexed by start->end
266 * inclusive
267 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800268int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 pgoff_t start, pgoff_t end)
270{
271 struct pagevec pvec;
272 int nr_pages;
273 int ret = 0;
274 pgoff_t index;
275
276 if (end < start)
277 return 0;
278
279 pagevec_init(&pvec, 0);
280 index = start;
281 while ((index <= end) &&
282 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
283 PAGECACHE_TAG_WRITEBACK,
284 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
285 unsigned i;
286
287 for (i = 0; i < nr_pages; i++) {
288 struct page *page = pvec.pages[i];
289
290 /* until radix tree lookup accepts end_index */
291 if (page->index > end)
292 continue;
293
294 wait_on_page_writeback(page);
295 if (PageError(page))
296 ret = -EIO;
297 }
298 pagevec_release(&pvec);
299 cond_resched();
300 }
301
302 /* Check for outstanding write errors */
303 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
304 ret = -ENOSPC;
305 if (test_and_clear_bit(AS_EIO, &mapping->flags))
306 ret = -EIO;
307
308 return ret;
309}
310
Randy Dunlap485bb992006-06-23 02:03:49 -0700311/**
312 * sync_page_range - write and wait on all pages in the passed range
313 * @inode: target inode
314 * @mapping: target address_space
315 * @pos: beginning offset in pages to write
316 * @count: number of bytes to write
317 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * Write and wait upon all the pages in the passed range. This is a "data
319 * integrity" operation. It waits upon in-flight writeout before starting and
320 * waiting upon new writeout. If there was an IO error, return it.
321 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800322 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 * it is otherwise livelockable.
324 */
325int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800326 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
328 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
329 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
330 int ret;
331
332 if (!mapping_cap_writeback_dirty(mapping) || !count)
333 return 0;
334 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
335 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800336 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800338 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
340 if (ret == 0)
341 ret = wait_on_page_writeback_range(mapping, start, end);
342 return ret;
343}
344EXPORT_SYMBOL(sync_page_range);
345
Randy Dunlap485bb992006-06-23 02:03:49 -0700346/**
Randy Dunlap76824862008-03-19 17:00:40 -0700347 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
Randy Dunlap485bb992006-06-23 02:03:49 -0700348 * @inode: target inode
349 * @mapping: target address_space
350 * @pos: beginning offset in pages to write
351 * @count: number of bytes to write
352 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800353 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * as it forces O_SYNC writers to different parts of the same file
355 * to be serialised right until io completion.
356 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800357int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
358 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
360 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
361 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
362 int ret;
363
364 if (!mapping_cap_writeback_dirty(mapping) || !count)
365 return 0;
366 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
367 if (ret == 0)
368 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
369 if (ret == 0)
370 ret = wait_on_page_writeback_range(mapping, start, end);
371 return ret;
372}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800373EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700376 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700378 *
379 * Walk the list of under-writeback pages of the given address space
380 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 */
382int filemap_fdatawait(struct address_space *mapping)
383{
384 loff_t i_size = i_size_read(mapping->host);
385
386 if (i_size == 0)
387 return 0;
388
389 return wait_on_page_writeback_range(mapping, 0,
390 (i_size - 1) >> PAGE_CACHE_SHIFT);
391}
392EXPORT_SYMBOL(filemap_fdatawait);
393
394int filemap_write_and_wait(struct address_space *mapping)
395{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800396 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800399 err = filemap_fdatawrite(mapping);
400 /*
401 * Even if the above returned error, the pages may be
402 * written partially (e.g. -ENOSPC), so we wait for it.
403 * But the -EIO is special case, it may indicate the worst
404 * thing (e.g. bug) happened, so we avoid waiting for it.
405 */
406 if (err != -EIO) {
407 int err2 = filemap_fdatawait(mapping);
408 if (!err)
409 err = err2;
410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800412 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800414EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
Randy Dunlap485bb992006-06-23 02:03:49 -0700416/**
417 * filemap_write_and_wait_range - write out & wait on a file range
418 * @mapping: the address_space for the pages
419 * @lstart: offset in bytes where the range starts
420 * @lend: offset in bytes where the range ends (inclusive)
421 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800422 * Write out and wait upon file offsets lstart->lend, inclusive.
423 *
424 * Note that `lend' is inclusive (describes the last byte to be written) so
425 * that this function can be used to write to the very end-of-file (end = -1).
426 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427int filemap_write_and_wait_range(struct address_space *mapping,
428 loff_t lstart, loff_t lend)
429{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800430 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
432 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800433 err = __filemap_fdatawrite_range(mapping, lstart, lend,
434 WB_SYNC_ALL);
435 /* See comment of filemap_write_and_wait() */
436 if (err != -EIO) {
437 int err2 = wait_on_page_writeback_range(mapping,
438 lstart >> PAGE_CACHE_SHIFT,
439 lend >> PAGE_CACHE_SHIFT);
440 if (!err)
441 err = err2;
442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800444 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
446
Randy Dunlap485bb992006-06-23 02:03:49 -0700447/**
448 * add_to_page_cache - add newly allocated pagecache pages
449 * @page: page to add
450 * @mapping: the page's address_space
451 * @offset: page index
452 * @gfp_mask: page allocation mode
453 *
454 * This function is used to add newly allocated pagecache pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * the page is new, so we can just run SetPageLocked() against it.
456 * The other page state flags were set by rmqueue().
457 *
458 * This function does not add the page to the LRU. The caller must do that.
459 */
460int add_to_page_cache(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400461 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
Badari Pulavarty4c6bc8d2008-02-07 00:14:05 -0800463 int error = mem_cgroup_cache_charge(page, current->mm,
464 gfp_mask & ~__GFP_HIGHMEM);
Balbir Singh35c754d2008-02-07 00:14:05 -0800465 if (error)
466 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Balbir Singh35c754d2008-02-07 00:14:05 -0800468 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (error == 0) {
470 write_lock_irq(&mapping->tree_lock);
471 error = radix_tree_insert(&mapping->page_tree, offset, page);
472 if (!error) {
473 page_cache_get(page);
474 SetPageLocked(page);
475 page->mapping = mapping;
476 page->index = offset;
477 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700478 __inc_zone_page_state(page, NR_FILE_PAGES);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800479 } else
480 mem_cgroup_uncharge_page(page);
481
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 write_unlock_irq(&mapping->tree_lock);
483 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800484 } else
485 mem_cgroup_uncharge_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800486out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return error;
488}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489EXPORT_SYMBOL(add_to_page_cache);
490
491int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400492 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
494 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
495 if (ret == 0)
496 lru_cache_add(page);
497 return ret;
498}
499
Paul Jackson44110fe2006-03-24 03:16:04 -0800500#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700501struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800502{
503 if (cpuset_do_page_mem_spread()) {
504 int n = cpuset_mem_spread_node();
Nick Piggin2ae88142006-10-28 10:38:23 -0700505 return alloc_pages_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800506 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700507 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800508}
Nick Piggin2ae88142006-10-28 10:38:23 -0700509EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800510#endif
511
Nick Piggindb376482006-09-25 23:31:24 -0700512static int __sleep_on_page_lock(void *word)
513{
514 io_schedule();
515 return 0;
516}
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518/*
519 * In order to wait for pages to become available there must be
520 * waitqueues associated with pages. By using a hash table of
521 * waitqueues where the bucket discipline is to maintain all
522 * waiters on the same queue and wake all when any of the pages
523 * become available, and for the woken contexts to check to be
524 * sure the appropriate page became available, this saves space
525 * at a cost of "thundering herd" phenomena during rare hash
526 * collisions.
527 */
528static wait_queue_head_t *page_waitqueue(struct page *page)
529{
530 const struct zone *zone = page_zone(page);
531
532 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
533}
534
535static inline void wake_up_page(struct page *page, int bit)
536{
537 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
538}
539
Harvey Harrison920c7a52008-02-04 22:29:26 -0800540void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
542 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
543
544 if (test_bit(bit_nr, &page->flags))
545 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
546 TASK_UNINTERRUPTIBLE);
547}
548EXPORT_SYMBOL(wait_on_page_bit);
549
550/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700551 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 * @page: the page
553 *
554 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
555 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
556 * mechananism between PageLocked pages and PageWriteback pages is shared.
557 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
558 *
559 * The first mb is necessary to safely close the critical section opened by the
560 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
561 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
562 * parallel wait_on_page_locked()).
563 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800564void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
566 smp_mb__before_clear_bit();
567 if (!TestClearPageLocked(page))
568 BUG();
569 smp_mb__after_clear_bit();
570 wake_up_page(page, PG_locked);
571}
572EXPORT_SYMBOL(unlock_page);
573
Randy Dunlap485bb992006-06-23 02:03:49 -0700574/**
575 * end_page_writeback - end writeback against a page
576 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 */
578void end_page_writeback(struct page *page)
579{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700580 if (TestClearPageReclaim(page))
581 rotate_reclaimable_page(page);
582
583 if (!test_clear_page_writeback(page))
584 BUG();
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 smp_mb__after_clear_bit();
587 wake_up_page(page, PG_writeback);
588}
589EXPORT_SYMBOL(end_page_writeback);
590
Randy Dunlap485bb992006-06-23 02:03:49 -0700591/**
592 * __lock_page - get a lock on the page, assuming we need to sleep to get it
593 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700595 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
597 * chances are that on the second loop, the block layer's plug list is empty,
598 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
599 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800600void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
602 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
603
604 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
605 TASK_UNINTERRUPTIBLE);
606}
607EXPORT_SYMBOL(__lock_page);
608
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800609int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500610{
611 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
612
613 return __wait_on_bit_lock(page_waitqueue(page), &wait,
614 sync_page_killable, TASK_KILLABLE);
615}
616
Randy Dunlap76824862008-03-19 17:00:40 -0700617/**
618 * __lock_page_nosync - get a lock on the page, without calling sync_page()
619 * @page: the page to lock
620 *
Nick Piggindb376482006-09-25 23:31:24 -0700621 * Variant of lock_page that does not require the caller to hold a reference
622 * on the page's mapping.
623 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800624void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700625{
626 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
627 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
628 TASK_UNINTERRUPTIBLE);
629}
630
Randy Dunlap485bb992006-06-23 02:03:49 -0700631/**
632 * find_get_page - find and get a page reference
633 * @mapping: the address_space to search
634 * @offset: the page index
635 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700636 * Is there a pagecache struct page at the given (mapping, offset) tuple?
637 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 */
Fengguang Wu57f6b962007-10-16 01:24:37 -0700639struct page * find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
641 struct page *page;
642
643 read_lock_irq(&mapping->tree_lock);
644 page = radix_tree_lookup(&mapping->page_tree, offset);
645 if (page)
646 page_cache_get(page);
647 read_unlock_irq(&mapping->tree_lock);
648 return page;
649}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650EXPORT_SYMBOL(find_get_page);
651
Randy Dunlap485bb992006-06-23 02:03:49 -0700652/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700654 * @mapping: the address_space to search
655 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 *
657 * Locates the desired pagecache page, locks it, increments its reference
658 * count and returns its address.
659 *
660 * Returns zero if the page was not present. find_lock_page() may sleep.
661 */
662struct page *find_lock_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700663 pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
665 struct page *page;
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667repeat:
Nick Piggin45726cb2007-10-16 01:24:41 -0700668 read_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 page = radix_tree_lookup(&mapping->page_tree, offset);
670 if (page) {
671 page_cache_get(page);
672 if (TestSetPageLocked(page)) {
673 read_unlock_irq(&mapping->tree_lock);
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800674 __lock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
676 /* Has the page been truncated while we slept? */
Nick Piggin45726cb2007-10-16 01:24:41 -0700677 if (unlikely(page->mapping != mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 unlock_page(page);
679 page_cache_release(page);
680 goto repeat;
681 }
Nick Piggin45726cb2007-10-16 01:24:41 -0700682 VM_BUG_ON(page->index != offset);
683 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
685 }
686 read_unlock_irq(&mapping->tree_lock);
Nick Piggin45726cb2007-10-16 01:24:41 -0700687out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 return page;
689}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690EXPORT_SYMBOL(find_lock_page);
691
692/**
693 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700694 * @mapping: the page's address_space
695 * @index: the page's index into the mapping
696 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 *
698 * Locates a page in the pagecache. If the page is not present, a new page
699 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
700 * LRU list. The returned page is locked and has its reference count
701 * incremented.
702 *
703 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
704 * allocation!
705 *
706 * find_or_create_page() returns the desired page's address, or zero on
707 * memory exhaustion.
708 */
709struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700710 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
Nick Piggineb2be182007-10-16 01:24:57 -0700712 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 int err;
714repeat:
715 page = find_lock_page(mapping, index);
716 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700717 page = __page_cache_alloc(gfp_mask);
718 if (!page)
719 return NULL;
720 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
721 if (unlikely(err)) {
722 page_cache_release(page);
723 page = NULL;
724 if (err == -EEXIST)
725 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 return page;
729}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730EXPORT_SYMBOL(find_or_create_page);
731
732/**
733 * find_get_pages - gang pagecache lookup
734 * @mapping: The address_space to search
735 * @start: The starting page index
736 * @nr_pages: The maximum number of pages
737 * @pages: Where the resulting pages are placed
738 *
739 * find_get_pages() will search for and return a group of up to
740 * @nr_pages pages in the mapping. The pages are placed at @pages.
741 * find_get_pages() takes a reference against the returned pages.
742 *
743 * The search returns a group of mapping-contiguous pages with ascending
744 * indexes. There may be holes in the indices due to not-present pages.
745 *
746 * find_get_pages() returns the number of pages which were found.
747 */
748unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
749 unsigned int nr_pages, struct page **pages)
750{
751 unsigned int i;
752 unsigned int ret;
753
754 read_lock_irq(&mapping->tree_lock);
755 ret = radix_tree_gang_lookup(&mapping->page_tree,
756 (void **)pages, start, nr_pages);
757 for (i = 0; i < ret; i++)
758 page_cache_get(pages[i]);
759 read_unlock_irq(&mapping->tree_lock);
760 return ret;
761}
762
Jens Axboeebf43502006-04-27 08:46:01 +0200763/**
764 * find_get_pages_contig - gang contiguous pagecache lookup
765 * @mapping: The address_space to search
766 * @index: The starting page index
767 * @nr_pages: The maximum number of pages
768 * @pages: Where the resulting pages are placed
769 *
770 * find_get_pages_contig() works exactly like find_get_pages(), except
771 * that the returned number of pages are guaranteed to be contiguous.
772 *
773 * find_get_pages_contig() returns the number of pages which were found.
774 */
775unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
776 unsigned int nr_pages, struct page **pages)
777{
778 unsigned int i;
779 unsigned int ret;
780
781 read_lock_irq(&mapping->tree_lock);
782 ret = radix_tree_gang_lookup(&mapping->page_tree,
783 (void **)pages, index, nr_pages);
784 for (i = 0; i < ret; i++) {
785 if (pages[i]->mapping == NULL || pages[i]->index != index)
786 break;
787
788 page_cache_get(pages[i]);
789 index++;
790 }
791 read_unlock_irq(&mapping->tree_lock);
792 return i;
793}
David Howellsef71c152007-05-09 02:33:44 -0700794EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200795
Randy Dunlap485bb992006-06-23 02:03:49 -0700796/**
797 * find_get_pages_tag - find and return pages that match @tag
798 * @mapping: the address_space to search
799 * @index: the starting page index
800 * @tag: the tag index
801 * @nr_pages: the maximum number of pages
802 * @pages: where the resulting pages are placed
803 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700805 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 */
807unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
808 int tag, unsigned int nr_pages, struct page **pages)
809{
810 unsigned int i;
811 unsigned int ret;
812
813 read_lock_irq(&mapping->tree_lock);
814 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
815 (void **)pages, *index, nr_pages, tag);
816 for (i = 0; i < ret; i++)
817 page_cache_get(pages[i]);
818 if (ret)
819 *index = pages[ret - 1]->index + 1;
820 read_unlock_irq(&mapping->tree_lock);
821 return ret;
822}
David Howellsef71c152007-05-09 02:33:44 -0700823EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Randy Dunlap485bb992006-06-23 02:03:49 -0700825/**
826 * grab_cache_page_nowait - returns locked page at given index in given cache
827 * @mapping: target address_space
828 * @index: the page index
829 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800830 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 * This is intended for speculative data generators, where the data can
832 * be regenerated if the page couldn't be grabbed. This routine should
833 * be safe to call while holding the lock for another page.
834 *
835 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
836 * and deadlock against the caller's locked page.
837 */
838struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700839grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840{
841 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 if (page) {
844 if (!TestSetPageLocked(page))
845 return page;
846 page_cache_release(page);
847 return NULL;
848 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700849 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
850 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 page_cache_release(page);
852 page = NULL;
853 }
854 return page;
855}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856EXPORT_SYMBOL(grab_cache_page_nowait);
857
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700858/*
859 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
860 * a _large_ part of the i/o request. Imagine the worst scenario:
861 *
862 * ---R__________________________________________B__________
863 * ^ reading here ^ bad block(assume 4k)
864 *
865 * read(R) => miss => readahead(R...B) => media error => frustrating retries
866 * => failing the whole request => read(R) => read(R+1) =>
867 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
868 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
869 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
870 *
871 * It is going insane. Fix it by quickly scaling down the readahead size.
872 */
873static void shrink_readahead_size_eio(struct file *filp,
874 struct file_ra_state *ra)
875{
876 if (!ra->ra_pages)
877 return;
878
879 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700880}
881
Randy Dunlap485bb992006-06-23 02:03:49 -0700882/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800883 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700884 * @filp: the file to read
885 * @ppos: current file position
886 * @desc: read_descriptor
887 * @actor: read method
888 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700890 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 *
892 * This is really ugly. But the goto's actually try to clarify some
893 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800895static void do_generic_file_read(struct file *filp, loff_t *ppos,
896 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
Christoph Hellwig36e78912008-02-08 04:21:24 -0800898 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -0800900 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700901 pgoff_t index;
902 pgoff_t last_index;
903 pgoff_t prev_index;
904 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700905 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700909 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
910 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
912 offset = *ppos & ~PAGE_CACHE_MASK;
913
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 for (;;) {
915 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700916 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -0700917 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 unsigned long nr, ret;
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921find_page:
922 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700923 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700924 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700925 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700926 index, last_index - index);
927 page = find_get_page(mapping, index);
928 if (unlikely(page == NULL))
929 goto no_cached_page;
930 }
931 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700932 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700933 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700934 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 }
936 if (!PageUptodate(page))
937 goto page_not_up_to_date;
938page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -0700939 /*
940 * i_size must be checked after we know the page is Uptodate.
941 *
942 * Checking i_size after the check allows us to calculate
943 * the correct value for "nr", which means the zero-filled
944 * part of the page is not copied back to userspace (unless
945 * another truncate extends the file - this is desired though).
946 */
947
948 isize = i_size_read(inode);
949 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
950 if (unlikely(!isize || index > end_index)) {
951 page_cache_release(page);
952 goto out;
953 }
954
955 /* nr is the maximum number of bytes to copy from this page */
956 nr = PAGE_CACHE_SIZE;
957 if (index == end_index) {
958 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
959 if (nr <= offset) {
960 page_cache_release(page);
961 goto out;
962 }
963 }
964 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
966 /* If users can be writing to this page using arbitrary
967 * virtual addresses, take care about potential aliasing
968 * before reading the page on the kernel side.
969 */
970 if (mapping_writably_mapped(mapping))
971 flush_dcache_page(page);
972
973 /*
Jan Karaec0f1632007-05-06 14:49:25 -0700974 * When a sequential read accesses a page several times,
975 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 */
Jan Karaec0f1632007-05-06 14:49:25 -0700977 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 mark_page_accessed(page);
979 prev_index = index;
980
981 /*
982 * Ok, we have the page, and it's up-to-date, so
983 * now we can copy it to user space...
984 *
985 * The actor routine returns how many bytes were actually used..
986 * NOTE! This may not be the same as how much of a user buffer
987 * we filled up (we may be padding etc), so we can only update
988 * "pos" here (the actor routine has to update the user buffer
989 * pointers and the remaining count).
990 */
991 ret = actor(desc, page, offset, nr);
992 offset += ret;
993 index += offset >> PAGE_CACHE_SHIFT;
994 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -0700995 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
997 page_cache_release(page);
998 if (ret == nr && desc->count)
999 continue;
1000 goto out;
1001
1002page_not_up_to_date:
1003 /* Get exclusive access to the page ... */
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001004 if (lock_page_killable(page))
1005 goto readpage_eio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
Nick Pigginda6052f2006-09-25 23:31:35 -07001007 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (!page->mapping) {
1009 unlock_page(page);
1010 page_cache_release(page);
1011 continue;
1012 }
1013
1014 /* Did somebody else fill it already? */
1015 if (PageUptodate(page)) {
1016 unlock_page(page);
1017 goto page_ok;
1018 }
1019
1020readpage:
1021 /* Start the actual read. The read will unlock the page. */
1022 error = mapping->a_ops->readpage(filp, page);
1023
Zach Brown994fc28c2005-12-15 14:28:17 -08001024 if (unlikely(error)) {
1025 if (error == AOP_TRUNCATED_PAGE) {
1026 page_cache_release(page);
1027 goto find_page;
1028 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001030 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
1032 if (!PageUptodate(page)) {
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001033 if (lock_page_killable(page))
1034 goto readpage_eio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (!PageUptodate(page)) {
1036 if (page->mapping == NULL) {
1037 /*
1038 * invalidate_inode_pages got it
1039 */
1040 unlock_page(page);
1041 page_cache_release(page);
1042 goto find_page;
1043 }
1044 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001045 shrink_readahead_size_eio(filp, ra);
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001046 goto readpage_eio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
1048 unlock_page(page);
1049 }
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 goto page_ok;
1052
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001053readpage_eio:
1054 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055readpage_error:
1056 /* UHHUH! A synchronous read error occurred. Report it */
1057 desc->error = error;
1058 page_cache_release(page);
1059 goto out;
1060
1061no_cached_page:
1062 /*
1063 * Ok, it wasn't cached, so we need to create a new
1064 * page..
1065 */
Nick Piggineb2be182007-10-16 01:24:57 -07001066 page = page_cache_alloc_cold(mapping);
1067 if (!page) {
1068 desc->error = -ENOMEM;
1069 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Nick Piggineb2be182007-10-16 01:24:57 -07001071 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 index, GFP_KERNEL);
1073 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001074 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 if (error == -EEXIST)
1076 goto find_page;
1077 desc->error = error;
1078 goto out;
1079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 goto readpage;
1081 }
1082
1083out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001084 ra->prev_pos = prev_index;
1085 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1086 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001088 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (filp)
1090 file_accessed(filp);
1091}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093int file_read_actor(read_descriptor_t *desc, struct page *page,
1094 unsigned long offset, unsigned long size)
1095{
1096 char *kaddr;
1097 unsigned long left, count = desc->count;
1098
1099 if (size > count)
1100 size = count;
1101
1102 /*
1103 * Faults on the destination of a read are common, so do it before
1104 * taking the kmap.
1105 */
1106 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1107 kaddr = kmap_atomic(page, KM_USER0);
1108 left = __copy_to_user_inatomic(desc->arg.buf,
1109 kaddr + offset, size);
1110 kunmap_atomic(kaddr, KM_USER0);
1111 if (left == 0)
1112 goto success;
1113 }
1114
1115 /* Do it the slow way */
1116 kaddr = kmap(page);
1117 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1118 kunmap(page);
1119
1120 if (left) {
1121 size -= left;
1122 desc->error = -EFAULT;
1123 }
1124success:
1125 desc->count = count - size;
1126 desc->written += size;
1127 desc->arg.buf += size;
1128 return size;
1129}
1130
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001131/*
1132 * Performs necessary checks before doing a write
1133 * @iov: io vector request
1134 * @nr_segs: number of segments in the iovec
1135 * @count: number of bytes to write
1136 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1137 *
1138 * Adjust number of segments and amount of bytes to write (nr_segs should be
1139 * properly initialized first). Returns appropriate error code that caller
1140 * should return or zero in case that write should be allowed.
1141 */
1142int generic_segment_checks(const struct iovec *iov,
1143 unsigned long *nr_segs, size_t *count, int access_flags)
1144{
1145 unsigned long seg;
1146 size_t cnt = 0;
1147 for (seg = 0; seg < *nr_segs; seg++) {
1148 const struct iovec *iv = &iov[seg];
1149
1150 /*
1151 * If any segment has a negative length, or the cumulative
1152 * length ever wraps negative then return -EINVAL.
1153 */
1154 cnt += iv->iov_len;
1155 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1156 return -EINVAL;
1157 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1158 continue;
1159 if (seg == 0)
1160 return -EFAULT;
1161 *nr_segs = seg;
1162 cnt -= iv->iov_len; /* This segment is no good */
1163 break;
1164 }
1165 *count = cnt;
1166 return 0;
1167}
1168EXPORT_SYMBOL(generic_segment_checks);
1169
Randy Dunlap485bb992006-06-23 02:03:49 -07001170/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001171 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001172 * @iocb: kernel I/O control block
1173 * @iov: io vector request
1174 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001175 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001176 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 * This is the "read()" routine for all filesystems
1178 * that can use the page cache directly.
1179 */
1180ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001181generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1182 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183{
1184 struct file *filp = iocb->ki_filp;
1185 ssize_t retval;
1186 unsigned long seg;
1187 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001188 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
1190 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001191 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1192 if (retval)
1193 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1196 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001197 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 struct address_space *mapping;
1199 struct inode *inode;
1200
1201 mapping = filp->f_mapping;
1202 inode = mapping->host;
1203 retval = 0;
1204 if (!count)
1205 goto out; /* skip atime */
1206 size = i_size_read(inode);
1207 if (pos < size) {
1208 retval = generic_file_direct_IO(READ, iocb,
1209 iov, pos, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 if (retval > 0)
1211 *ppos = pos + retval;
1212 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001213 if (likely(retval != 0)) {
Steven Whitehouse3f1a9aa2006-09-27 14:52:48 -04001214 file_accessed(filp);
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001215 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 }
1218
1219 retval = 0;
1220 if (count) {
1221 for (seg = 0; seg < nr_segs; seg++) {
1222 read_descriptor_t desc;
1223
1224 desc.written = 0;
1225 desc.arg.buf = iov[seg].iov_base;
1226 desc.count = iov[seg].iov_len;
1227 if (desc.count == 0)
1228 continue;
1229 desc.error = 0;
1230 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1231 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001232 if (desc.error) {
1233 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 break;
1235 }
akpm@linux-foundation.orgc44939e2007-07-15 23:38:25 -07001236 if (desc.count > 0)
1237 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 }
1239 }
1240out:
1241 return retval;
1242}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243EXPORT_SYMBOL(generic_file_aio_read);
1244
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245static ssize_t
1246do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001247 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248{
1249 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1250 return -EINVAL;
1251
1252 force_page_cache_readahead(mapping, filp, index,
1253 max_sane_readahead(nr));
1254 return 0;
1255}
1256
1257asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1258{
1259 ssize_t ret;
1260 struct file *file;
1261
1262 ret = -EBADF;
1263 file = fget(fd);
1264 if (file) {
1265 if (file->f_mode & FMODE_READ) {
1266 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001267 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1268 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 unsigned long len = end - start + 1;
1270 ret = do_readahead(mapping, file, start, len);
1271 }
1272 fput(file);
1273 }
1274 return ret;
1275}
1276
1277#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001278/**
1279 * page_cache_read - adds requested page to the page cache if not already there
1280 * @file: file to read
1281 * @offset: page index
1282 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 * This adds the requested page to the page cache if it isn't already there,
1284 * and schedules an I/O to read in its contents from disk.
1285 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001286static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287{
1288 struct address_space *mapping = file->f_mapping;
1289 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001290 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
Zach Brown994fc28c2005-12-15 14:28:17 -08001292 do {
1293 page = page_cache_alloc_cold(mapping);
1294 if (!page)
1295 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
Zach Brown994fc28c2005-12-15 14:28:17 -08001297 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1298 if (ret == 0)
1299 ret = mapping->a_ops->readpage(file, page);
1300 else if (ret == -EEXIST)
1301 ret = 0; /* losing race to add is OK */
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Zach Brown994fc28c2005-12-15 14:28:17 -08001305 } while (ret == AOP_TRUNCATED_PAGE);
1306
1307 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308}
1309
1310#define MMAP_LOTSAMISS (100)
1311
Randy Dunlap485bb992006-06-23 02:03:49 -07001312/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001313 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001314 * @vma: vma in which the fault was taken
1315 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001316 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001317 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 * mapped memory region to read in file data during a page fault.
1319 *
1320 * The goto's are kind of ugly, but this streamlines the normal case of having
1321 * it in the page cache, and handles the special cases reasonably without
1322 * having a lot of duplicated code.
1323 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001324int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
1326 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001327 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 struct address_space *mapping = file->f_mapping;
1329 struct file_ra_state *ra = &file->f_ra;
1330 struct inode *inode = mapping->host;
1331 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001332 pgoff_t size;
Nick Piggin54cb8822007-07-19 01:46:59 -07001333 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001334 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001337 if (vmf->pgoff >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001338 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
1340 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001341 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 goto no_cached_page;
1343
1344 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 * Do we have something in the page cache already?
1346 */
1347retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001348 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001349 /*
1350 * For sequential accesses, we use the generic readahead logic.
1351 */
1352 if (VM_SequentialReadHint(vma)) {
1353 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001354 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001355 vmf->pgoff, 1);
1356 page = find_lock_page(mapping, vmf->pgoff);
1357 if (!page)
1358 goto no_cached_page;
1359 }
1360 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001361 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001362 vmf->pgoff, 1);
1363 }
1364 }
1365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (!page) {
1367 unsigned long ra_pages;
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 ra->mmap_miss++;
1370
1371 /*
1372 * Do we miss much more than hit in this file? If so,
1373 * stop bothering with read-ahead. It will only hurt.
1374 */
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001375 if (ra->mmap_miss > MMAP_LOTSAMISS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 goto no_cached_page;
1377
1378 /*
1379 * To keep the pgmajfault counter straight, we need to
1380 * check did_readaround, as this is an inner loop.
1381 */
1382 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001383 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001384 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 }
1386 did_readaround = 1;
1387 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1388 if (ra_pages) {
1389 pgoff_t start = 0;
1390
Nick Piggind0217ac2007-07-19 01:47:03 -07001391 if (vmf->pgoff > ra_pages / 2)
1392 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 do_page_cache_readahead(mapping, file, start, ra_pages);
1394 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001395 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 if (!page)
1397 goto no_cached_page;
1398 }
1399
1400 if (!did_readaround)
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001401 ra->mmap_miss--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001404 * We have a locked page in the page cache, now we need to check
1405 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 */
Nick Piggind00806b2007-07-19 01:46:57 -07001407 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 goto page_not_uptodate;
1409
Nick Piggind00806b2007-07-19 01:46:57 -07001410 /* Must recheck i_size under page lock */
1411 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001412 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001413 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001414 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001415 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001416 }
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 /*
1419 * Found the page and have a reference on it.
1420 */
1421 mark_page_accessed(page);
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001422 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001423 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001424 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426no_cached_page:
1427 /*
1428 * We're only likely to ever get here if MADV_RANDOM is in
1429 * effect.
1430 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001431 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
1433 /*
1434 * The page we want has now been added to the page cache.
1435 * In the unlikely event that someone removed it in the
1436 * meantime, we'll just come back here and read it again.
1437 */
1438 if (error >= 0)
1439 goto retry_find;
1440
1441 /*
1442 * An error return from page_cache_read can result if the
1443 * system is low on memory, or a problem occurs while trying
1444 * to schedule I/O.
1445 */
1446 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001447 return VM_FAULT_OOM;
1448 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001451 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001453 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001454 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 /*
1458 * Umm, take care of errors if the page isn't up-to-date.
1459 * Try to re-read it _once_. We do this synchronously,
1460 * because there really aren't any performance issues here
1461 * and we need to check for errors.
1462 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001464 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001465 if (!error) {
1466 wait_on_page_locked(page);
1467 if (!PageUptodate(page))
1468 error = -EIO;
1469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001471
1472 if (!error || error == AOP_TRUNCATED_PAGE)
1473 goto retry_find;
1474
1475 /* Things didn't work out. Return zero to tell the mm layer so. */
1476 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001477 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001478}
1479EXPORT_SYMBOL(filemap_fault);
1480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001482 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483};
1484
1485/* This is used for a general mmap of a disk file */
1486
1487int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1488{
1489 struct address_space *mapping = file->f_mapping;
1490
1491 if (!mapping->a_ops->readpage)
1492 return -ENOEXEC;
1493 file_accessed(file);
1494 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001495 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 return 0;
1497}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
1499/*
1500 * This is for filesystems which do not implement ->writepage.
1501 */
1502int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1503{
1504 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1505 return -EINVAL;
1506 return generic_file_mmap(file, vma);
1507}
1508#else
1509int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1510{
1511 return -ENOSYS;
1512}
1513int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1514{
1515 return -ENOSYS;
1516}
1517#endif /* CONFIG_MMU */
1518
1519EXPORT_SYMBOL(generic_file_mmap);
1520EXPORT_SYMBOL(generic_file_readonly_mmap);
1521
Nick Piggin6fe69002007-05-06 14:49:04 -07001522static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001523 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 int (*filler)(void *,struct page*),
1525 void *data)
1526{
Nick Piggineb2be182007-10-16 01:24:57 -07001527 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 int err;
1529repeat:
1530 page = find_get_page(mapping, index);
1531 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001532 page = page_cache_alloc_cold(mapping);
1533 if (!page)
1534 return ERR_PTR(-ENOMEM);
1535 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1536 if (unlikely(err)) {
1537 page_cache_release(page);
1538 if (err == -EEXIST)
1539 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 return ERR_PTR(err);
1542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 err = filler(data, page);
1544 if (err < 0) {
1545 page_cache_release(page);
1546 page = ERR_PTR(err);
1547 }
1548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 return page;
1550}
1551
Randy Dunlap76824862008-03-19 17:00:40 -07001552/**
1553 * read_cache_page_async - read into page cache, fill it if needed
1554 * @mapping: the page's address_space
1555 * @index: the page index
1556 * @filler: function to perform the read
1557 * @data: destination for read data
1558 *
Nick Piggin6fe69002007-05-06 14:49:04 -07001559 * Same as read_cache_page, but don't wait for page to become unlocked
1560 * after submitting it to the filler.
Randy Dunlap76824862008-03-19 17:00:40 -07001561 *
1562 * Read into the page cache. If a page already exists, and PageUptodate() is
1563 * not set, try to fill the page but don't wait for it to become unlocked.
1564 *
1565 * If the page does not get brought uptodate, return -EIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001567struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001568 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 int (*filler)(void *,struct page*),
1570 void *data)
1571{
1572 struct page *page;
1573 int err;
1574
1575retry:
1576 page = __read_cache_page(mapping, index, filler, data);
1577 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001578 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 if (PageUptodate(page))
1580 goto out;
1581
1582 lock_page(page);
1583 if (!page->mapping) {
1584 unlock_page(page);
1585 page_cache_release(page);
1586 goto retry;
1587 }
1588 if (PageUptodate(page)) {
1589 unlock_page(page);
1590 goto out;
1591 }
1592 err = filler(data, page);
1593 if (err < 0) {
1594 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001595 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
David Howellsc855ff32007-05-09 13:42:20 +01001597out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001598 mark_page_accessed(page);
1599 return page;
1600}
1601EXPORT_SYMBOL(read_cache_page_async);
1602
1603/**
1604 * read_cache_page - read into page cache, fill it if needed
1605 * @mapping: the page's address_space
1606 * @index: the page index
1607 * @filler: function to perform the read
1608 * @data: destination for read data
1609 *
1610 * Read into the page cache. If a page already exists, and PageUptodate() is
1611 * not set, try to fill the page then wait for it to become unlocked.
1612 *
1613 * If the page does not get brought uptodate, return -EIO.
1614 */
1615struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001616 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001617 int (*filler)(void *,struct page*),
1618 void *data)
1619{
1620 struct page *page;
1621
1622 page = read_cache_page_async(mapping, index, filler, data);
1623 if (IS_ERR(page))
1624 goto out;
1625 wait_on_page_locked(page);
1626 if (!PageUptodate(page)) {
1627 page_cache_release(page);
1628 page = ERR_PTR(-EIO);
1629 }
1630 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 return page;
1632}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633EXPORT_SYMBOL(read_cache_page);
1634
1635/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 * The logic we want is
1637 *
1638 * if suid or (sgid and xgrp)
1639 * remove privs
1640 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001641int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642{
1643 mode_t mode = dentry->d_inode->i_mode;
1644 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646 /* suid always must be killed */
1647 if (unlikely(mode & S_ISUID))
1648 kill = ATTR_KILL_SUID;
1649
1650 /*
1651 * sgid without any exec bits is just a mandatory locking mark; leave
1652 * it alone. If some exec bits are set, it's a real sgid; kill it.
1653 */
1654 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1655 kill |= ATTR_KILL_SGID;
1656
Jens Axboe01de85e2006-10-17 19:50:36 +02001657 if (unlikely(kill && !capable(CAP_FSETID)))
1658 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Jens Axboe01de85e2006-10-17 19:50:36 +02001660 return 0;
1661}
Mark Fashehd23a1472006-10-17 17:05:18 -07001662EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001663
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001664static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001665{
1666 struct iattr newattrs;
1667
1668 newattrs.ia_valid = ATTR_FORCE | kill;
1669 return notify_change(dentry, &newattrs);
1670}
1671
1672int remove_suid(struct dentry *dentry)
1673{
Serge E. Hallynb5376772007-10-16 23:31:36 -07001674 int killsuid = should_remove_suid(dentry);
1675 int killpriv = security_inode_need_killpriv(dentry);
1676 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001677
Serge E. Hallynb5376772007-10-16 23:31:36 -07001678 if (killpriv < 0)
1679 return killpriv;
1680 if (killpriv)
1681 error = security_inode_killpriv(dentry);
1682 if (!error && killsuid)
1683 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001684
Serge E. Hallynb5376772007-10-16 23:31:36 -07001685 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686}
1687EXPORT_SYMBOL(remove_suid);
1688
Nick Piggin2f718ff2007-10-16 01:24:59 -07001689static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 const struct iovec *iov, size_t base, size_t bytes)
1691{
1692 size_t copied = 0, left = 0;
1693
1694 while (bytes) {
1695 char __user *buf = iov->iov_base + base;
1696 int copy = min(bytes, iov->iov_len - base);
1697
1698 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001699 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 copied += copy;
1701 bytes -= copy;
1702 vaddr += copy;
1703 iov++;
1704
NeilBrown01408c42006-06-25 05:47:58 -07001705 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 }
1708 return copied - left;
1709}
1710
1711/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001712 * Copy as much as we can into the page and return the number of bytes which
1713 * were sucessfully copied. If a fault is encountered then return the number of
1714 * bytes which were copied.
1715 */
1716size_t iov_iter_copy_from_user_atomic(struct page *page,
1717 struct iov_iter *i, unsigned long offset, size_t bytes)
1718{
1719 char *kaddr;
1720 size_t copied;
1721
1722 BUG_ON(!in_atomic());
1723 kaddr = kmap_atomic(page, KM_USER0);
1724 if (likely(i->nr_segs == 1)) {
1725 int left;
1726 char __user *buf = i->iov->iov_base + i->iov_offset;
1727 left = __copy_from_user_inatomic_nocache(kaddr + offset,
1728 buf, bytes);
1729 copied = bytes - left;
1730 } else {
1731 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1732 i->iov, i->iov_offset, bytes);
1733 }
1734 kunmap_atomic(kaddr, KM_USER0);
1735
1736 return copied;
1737}
Nick Piggin89e10782007-10-16 01:25:07 -07001738EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001739
1740/*
1741 * This has the same sideeffects and return value as
1742 * iov_iter_copy_from_user_atomic().
1743 * The difference is that it attempts to resolve faults.
1744 * Page must not be locked.
1745 */
1746size_t iov_iter_copy_from_user(struct page *page,
1747 struct iov_iter *i, unsigned long offset, size_t bytes)
1748{
1749 char *kaddr;
1750 size_t copied;
1751
1752 kaddr = kmap(page);
1753 if (likely(i->nr_segs == 1)) {
1754 int left;
1755 char __user *buf = i->iov->iov_base + i->iov_offset;
1756 left = __copy_from_user_nocache(kaddr + offset, buf, bytes);
1757 copied = bytes - left;
1758 } else {
1759 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1760 i->iov, i->iov_offset, bytes);
1761 }
1762 kunmap(page);
1763 return copied;
1764}
Nick Piggin89e10782007-10-16 01:25:07 -07001765EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001766
Nick Pigginf7009262008-03-10 11:43:59 -07001767void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001768{
Nick Pigginf7009262008-03-10 11:43:59 -07001769 BUG_ON(i->count < bytes);
1770
Nick Piggin2f718ff2007-10-16 01:24:59 -07001771 if (likely(i->nr_segs == 1)) {
1772 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001773 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001774 } else {
1775 const struct iovec *iov = i->iov;
1776 size_t base = i->iov_offset;
1777
Nick Piggin124d3b72008-02-02 15:01:17 +01001778 /*
1779 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001780 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001781 */
Nick Pigginf7009262008-03-10 11:43:59 -07001782 while (bytes || unlikely(!iov->iov_len && i->count)) {
1783 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001784
Nick Pigginf7009262008-03-10 11:43:59 -07001785 copy = min(bytes, iov->iov_len - base);
1786 BUG_ON(!i->count || i->count < copy);
1787 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001788 bytes -= copy;
1789 base += copy;
1790 if (iov->iov_len == base) {
1791 iov++;
1792 base = 0;
1793 }
1794 }
1795 i->iov = iov;
1796 i->iov_offset = base;
1797 }
1798}
Nick Piggin89e10782007-10-16 01:25:07 -07001799EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001800
Nick Pigginafddba42007-10-16 01:25:01 -07001801/*
1802 * Fault in the first iovec of the given iov_iter, to a maximum length
1803 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1804 * accessed (ie. because it is an invalid address).
1805 *
1806 * writev-intensive code may want this to prefault several iovecs -- that
1807 * would be possible (callers must not rely on the fact that _only_ the
1808 * first iovec will be faulted with the current implementation).
1809 */
1810int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001811{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001812 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001813 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1814 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001815}
Nick Piggin89e10782007-10-16 01:25:07 -07001816EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001817
1818/*
1819 * Return the count of just the current iov_iter segment.
1820 */
1821size_t iov_iter_single_seg_count(struct iov_iter *i)
1822{
1823 const struct iovec *iov = i->iov;
1824 if (i->nr_segs == 1)
1825 return i->count;
1826 else
1827 return min(i->count, iov->iov_len - i->iov_offset);
1828}
Nick Piggin89e10782007-10-16 01:25:07 -07001829EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001830
1831/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 * Performs necessary checks before doing a write
1833 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001834 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 * Returns appropriate error code that caller should return or
1836 * zero in case that write should be allowed.
1837 */
1838inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1839{
1840 struct inode *inode = file->f_mapping->host;
1841 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1842
1843 if (unlikely(*pos < 0))
1844 return -EINVAL;
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 if (!isblk) {
1847 /* FIXME: this is for backwards compatibility with 2.4 */
1848 if (file->f_flags & O_APPEND)
1849 *pos = i_size_read(inode);
1850
1851 if (limit != RLIM_INFINITY) {
1852 if (*pos >= limit) {
1853 send_sig(SIGXFSZ, current, 0);
1854 return -EFBIG;
1855 }
1856 if (*count > limit - (typeof(limit))*pos) {
1857 *count = limit - (typeof(limit))*pos;
1858 }
1859 }
1860 }
1861
1862 /*
1863 * LFS rule
1864 */
1865 if (unlikely(*pos + *count > MAX_NON_LFS &&
1866 !(file->f_flags & O_LARGEFILE))) {
1867 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 return -EFBIG;
1869 }
1870 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1871 *count = MAX_NON_LFS - (unsigned long)*pos;
1872 }
1873 }
1874
1875 /*
1876 * Are we about to exceed the fs block limit ?
1877 *
1878 * If we have written data it becomes a short write. If we have
1879 * exceeded without writing data we send a signal and return EFBIG.
1880 * Linus frestrict idea will clean these up nicely..
1881 */
1882 if (likely(!isblk)) {
1883 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1884 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 return -EFBIG;
1886 }
1887 /* zero-length writes at ->s_maxbytes are OK */
1888 }
1889
1890 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1891 *count = inode->i_sb->s_maxbytes - *pos;
1892 } else {
David Howells93614012006-09-30 20:45:40 +02001893#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 loff_t isize;
1895 if (bdev_read_only(I_BDEV(inode)))
1896 return -EPERM;
1897 isize = i_size_read(inode);
1898 if (*pos >= isize) {
1899 if (*count || *pos > isize)
1900 return -ENOSPC;
1901 }
1902
1903 if (*pos + *count > isize)
1904 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02001905#else
1906 return -EPERM;
1907#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 }
1909 return 0;
1910}
1911EXPORT_SYMBOL(generic_write_checks);
1912
Nick Pigginafddba42007-10-16 01:25:01 -07001913int pagecache_write_begin(struct file *file, struct address_space *mapping,
1914 loff_t pos, unsigned len, unsigned flags,
1915 struct page **pagep, void **fsdata)
1916{
1917 const struct address_space_operations *aops = mapping->a_ops;
1918
1919 if (aops->write_begin) {
1920 return aops->write_begin(file, mapping, pos, len, flags,
1921 pagep, fsdata);
1922 } else {
1923 int ret;
1924 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1925 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1926 struct inode *inode = mapping->host;
1927 struct page *page;
1928again:
1929 page = __grab_cache_page(mapping, index);
1930 *pagep = page;
1931 if (!page)
1932 return -ENOMEM;
1933
1934 if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) {
1935 /*
1936 * There is no way to resolve a short write situation
1937 * for a !Uptodate page (except by double copying in
1938 * the caller done by generic_perform_write_2copy).
1939 *
1940 * Instead, we have to bring it uptodate here.
1941 */
1942 ret = aops->readpage(file, page);
1943 page_cache_release(page);
1944 if (ret) {
1945 if (ret == AOP_TRUNCATED_PAGE)
1946 goto again;
1947 return ret;
1948 }
1949 goto again;
1950 }
1951
1952 ret = aops->prepare_write(file, page, offset, offset+len);
1953 if (ret) {
Nick Piggin55144762007-10-16 01:25:26 -07001954 unlock_page(page);
Nick Pigginafddba42007-10-16 01:25:01 -07001955 page_cache_release(page);
1956 if (pos + len > inode->i_size)
1957 vmtruncate(inode, inode->i_size);
Nick Pigginafddba42007-10-16 01:25:01 -07001958 }
1959 return ret;
1960 }
1961}
1962EXPORT_SYMBOL(pagecache_write_begin);
1963
1964int pagecache_write_end(struct file *file, struct address_space *mapping,
1965 loff_t pos, unsigned len, unsigned copied,
1966 struct page *page, void *fsdata)
1967{
1968 const struct address_space_operations *aops = mapping->a_ops;
1969 int ret;
1970
1971 if (aops->write_end) {
1972 mark_page_accessed(page);
1973 ret = aops->write_end(file, mapping, pos, len, copied,
1974 page, fsdata);
1975 } else {
1976 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1977 struct inode *inode = mapping->host;
1978
1979 flush_dcache_page(page);
1980 ret = aops->commit_write(file, page, offset, offset+len);
1981 unlock_page(page);
1982 mark_page_accessed(page);
1983 page_cache_release(page);
Nick Pigginafddba42007-10-16 01:25:01 -07001984
1985 if (ret < 0) {
1986 if (pos + len > inode->i_size)
1987 vmtruncate(inode, inode->i_size);
1988 } else if (ret > 0)
1989 ret = min_t(size_t, copied, ret);
1990 else
1991 ret = copied;
1992 }
1993
1994 return ret;
1995}
1996EXPORT_SYMBOL(pagecache_write_end);
1997
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998ssize_t
1999generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2000 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2001 size_t count, size_t ocount)
2002{
2003 struct file *file = iocb->ki_filp;
2004 struct address_space *mapping = file->f_mapping;
2005 struct inode *inode = mapping->host;
2006 ssize_t written;
2007
2008 if (count != ocount)
2009 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2010
2011 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2012 if (written > 0) {
2013 loff_t end = pos + written;
2014 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2015 i_size_write(inode, end);
2016 mark_inode_dirty(inode);
2017 }
2018 *ppos = end;
2019 }
2020
2021 /*
2022 * Sync the fs metadata but not the minor inode changes and
2023 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002024 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002025 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 */
Zach Brown8459d862006-12-10 02:21:05 -08002027 if ((written >= 0 || written == -EIOCBQUEUED) &&
2028 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002029 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2030 if (err < 0)
2031 written = err;
2032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 return written;
2034}
2035EXPORT_SYMBOL(generic_file_direct_write);
2036
Nick Piggineb2be182007-10-16 01:24:57 -07002037/*
2038 * Find or create a page at the given pagecache position. Return the locked
2039 * page. This function is specifically for buffered writes.
2040 */
Nick Pigginafddba42007-10-16 01:25:01 -07002041struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index)
Nick Piggineb2be182007-10-16 01:24:57 -07002042{
2043 int status;
2044 struct page *page;
2045repeat:
2046 page = find_lock_page(mapping, index);
2047 if (likely(page))
2048 return page;
2049
2050 page = page_cache_alloc(mapping);
2051 if (!page)
2052 return NULL;
2053 status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
2054 if (unlikely(status)) {
2055 page_cache_release(page);
2056 if (status == -EEXIST)
2057 goto repeat;
2058 return NULL;
2059 }
2060 return page;
2061}
Nick Pigginafddba42007-10-16 01:25:01 -07002062EXPORT_SYMBOL(__grab_cache_page);
Nick Piggineb2be182007-10-16 01:24:57 -07002063
Nick Pigginafddba42007-10-16 01:25:01 -07002064static ssize_t generic_perform_write_2copy(struct file *file,
2065 struct iov_iter *i, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066{
Andrew Mortonae374612007-10-16 01:24:55 -07002067 struct address_space *mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002068 const struct address_space_operations *a_ops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002069 struct inode *inode = mapping->host;
2070 long status = 0;
2071 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
2073 do {
Nick Piggin08291422007-10-16 01:24:59 -07002074 struct page *src_page;
Nick Piggineb2be182007-10-16 01:24:57 -07002075 struct page *page;
Andrew Mortonae374612007-10-16 01:24:55 -07002076 pgoff_t index; /* Pagecache index for current page */
2077 unsigned long offset; /* Offset into pagecache page */
Nick Piggin08291422007-10-16 01:24:59 -07002078 unsigned long bytes; /* Bytes to write to page */
Andrew Mortonae374612007-10-16 01:24:55 -07002079 size_t copied; /* Bytes copied from user */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Andrew Mortonae374612007-10-16 01:24:55 -07002081 offset = (pos & (PAGE_CACHE_SIZE - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002083 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002084 iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
Nick Piggin08291422007-10-16 01:24:59 -07002086 /*
2087 * a non-NULL src_page indicates that we're doing the
2088 * copy via get_user_pages and kmap.
2089 */
2090 src_page = NULL;
Andrew Mortonae374612007-10-16 01:24:55 -07002091
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002092 /*
2093 * Bring in the user page that we will copy from _first_.
2094 * Otherwise there's a nasty deadlock on copying from the
2095 * same page as we're writing to, without it being marked
2096 * up-to-date.
Nick Piggin08291422007-10-16 01:24:59 -07002097 *
2098 * Not only is this an optimisation, but it is also required
2099 * to check that the address is actually valid, when atomic
2100 * usercopies are used, below.
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002101 */
Nick Pigginafddba42007-10-16 01:25:01 -07002102 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
Nick Piggin08291422007-10-16 01:24:59 -07002103 status = -EFAULT;
2104 break;
2105 }
Nick Piggineb2be182007-10-16 01:24:57 -07002106
2107 page = __grab_cache_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 if (!page) {
2109 status = -ENOMEM;
2110 break;
2111 }
2112
Nick Piggin08291422007-10-16 01:24:59 -07002113 /*
2114 * non-uptodate pages cannot cope with short copies, and we
2115 * cannot take a pagefault with the destination page locked.
2116 * So pin the source page to copy it.
2117 */
Nick Piggin674b8922007-10-16 01:25:03 -07002118 if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) {
Nick Piggin08291422007-10-16 01:24:59 -07002119 unlock_page(page);
2120
2121 src_page = alloc_page(GFP_KERNEL);
2122 if (!src_page) {
2123 page_cache_release(page);
2124 status = -ENOMEM;
2125 break;
2126 }
2127
2128 /*
2129 * Cannot get_user_pages with a page locked for the
2130 * same reason as we can't take a page fault with a
2131 * page locked (as explained below).
2132 */
Nick Pigginafddba42007-10-16 01:25:01 -07002133 copied = iov_iter_copy_from_user(src_page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002134 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002135 if (unlikely(copied == 0)) {
2136 status = -EFAULT;
2137 page_cache_release(page);
2138 page_cache_release(src_page);
2139 break;
2140 }
2141 bytes = copied;
2142
2143 lock_page(page);
2144 /*
2145 * Can't handle the page going uptodate here, because
2146 * that means we would use non-atomic usercopies, which
2147 * zero out the tail of the page, which can cause
2148 * zeroes to become transiently visible. We could just
2149 * use a non-zeroing copy, but the APIs aren't too
2150 * consistent.
2151 */
2152 if (unlikely(!page->mapping || PageUptodate(page))) {
2153 unlock_page(page);
2154 page_cache_release(page);
2155 page_cache_release(src_page);
2156 continue;
2157 }
Nick Piggin08291422007-10-16 01:24:59 -07002158 }
2159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 status = a_ops->prepare_write(file, page, offset, offset+bytes);
Nick Piggin64649a52007-10-16 01:24:56 -07002161 if (unlikely(status))
2162 goto fs_write_aop_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08002163
Nick Piggin08291422007-10-16 01:24:59 -07002164 if (!src_page) {
2165 /*
2166 * Must not enter the pagefault handler here, because
2167 * we hold the page lock, so we might recursively
2168 * deadlock on the same lock, or get an ABBA deadlock
2169 * against a different lock, or against the mmap_sem
2170 * (which nests outside the page lock). So increment
2171 * preempt count, and use _atomic usercopies.
2172 *
2173 * The page is uptodate so we are OK to encounter a
2174 * short copy: if unmodified parts of the page are
2175 * marked dirty and written out to disk, it doesn't
2176 * really matter.
2177 */
2178 pagefault_disable();
Nick Pigginafddba42007-10-16 01:25:01 -07002179 copied = iov_iter_copy_from_user_atomic(page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002180 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002181 pagefault_enable();
2182 } else {
2183 void *src, *dst;
2184 src = kmap_atomic(src_page, KM_USER0);
2185 dst = kmap_atomic(page, KM_USER1);
2186 memcpy(dst + offset, src + offset, bytes);
2187 kunmap_atomic(dst, KM_USER1);
2188 kunmap_atomic(src, KM_USER0);
2189 copied = bytes;
2190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 flush_dcache_page(page);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002192
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 status = a_ops->commit_write(file, page, offset, offset+bytes);
Nick Piggin55144762007-10-16 01:25:26 -07002194 if (unlikely(status < 0))
Nick Piggin64649a52007-10-16 01:24:56 -07002195 goto fs_write_aop_error;
Nick Piggin64649a52007-10-16 01:24:56 -07002196 if (unlikely(status > 0)) /* filesystem did partial write */
Nick Piggin08291422007-10-16 01:24:59 -07002197 copied = min_t(size_t, copied, status);
2198
2199 unlock_page(page);
2200 mark_page_accessed(page);
2201 page_cache_release(page);
2202 if (src_page)
2203 page_cache_release(src_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Nick Pigginafddba42007-10-16 01:25:01 -07002205 iov_iter_advance(i, copied);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002206 pos += copied;
Nick Pigginafddba42007-10-16 01:25:01 -07002207 written += copied;
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002208
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 balance_dirty_pages_ratelimited(mapping);
2210 cond_resched();
Nick Piggin64649a52007-10-16 01:24:56 -07002211 continue;
2212
2213fs_write_aop_error:
Nick Piggin55144762007-10-16 01:25:26 -07002214 unlock_page(page);
Nick Piggin64649a52007-10-16 01:24:56 -07002215 page_cache_release(page);
Nick Piggin08291422007-10-16 01:24:59 -07002216 if (src_page)
2217 page_cache_release(src_page);
Nick Piggin64649a52007-10-16 01:24:56 -07002218
2219 /*
2220 * prepare_write() may have instantiated a few blocks
2221 * outside i_size. Trim these off again. Don't need
2222 * i_size_read because we hold i_mutex.
2223 */
2224 if (pos + bytes > inode->i_size)
2225 vmtruncate(inode, inode->i_size);
Nick Piggin55144762007-10-16 01:25:26 -07002226 break;
Nick Pigginafddba42007-10-16 01:25:01 -07002227 } while (iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
Nick Pigginafddba42007-10-16 01:25:01 -07002229 return written ? written : status;
2230}
2231
2232static ssize_t generic_perform_write(struct file *file,
2233 struct iov_iter *i, loff_t pos)
2234{
2235 struct address_space *mapping = file->f_mapping;
2236 const struct address_space_operations *a_ops = mapping->a_ops;
2237 long status = 0;
2238 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002239 unsigned int flags = 0;
2240
2241 /*
2242 * Copies from kernel address space cannot fail (NFSD is a big user).
2243 */
2244 if (segment_eq(get_fs(), KERNEL_DS))
2245 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002246
2247 do {
2248 struct page *page;
2249 pgoff_t index; /* Pagecache index for current page */
2250 unsigned long offset; /* Offset into pagecache page */
2251 unsigned long bytes; /* Bytes to write to page */
2252 size_t copied; /* Bytes copied from user */
2253 void *fsdata;
2254
2255 offset = (pos & (PAGE_CACHE_SIZE - 1));
2256 index = pos >> PAGE_CACHE_SHIFT;
2257 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2258 iov_iter_count(i));
2259
2260again:
2261
2262 /*
2263 * Bring in the user page that we will copy from _first_.
2264 * Otherwise there's a nasty deadlock on copying from the
2265 * same page as we're writing to, without it being marked
2266 * up-to-date.
2267 *
2268 * Not only is this an optimisation, but it is also required
2269 * to check that the address is actually valid, when atomic
2270 * usercopies are used, below.
2271 */
2272 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2273 status = -EFAULT;
2274 break;
2275 }
2276
Nick Piggin674b8922007-10-16 01:25:03 -07002277 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002278 &page, &fsdata);
2279 if (unlikely(status))
2280 break;
2281
2282 pagefault_disable();
2283 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2284 pagefault_enable();
2285 flush_dcache_page(page);
2286
2287 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2288 page, fsdata);
2289 if (unlikely(status < 0))
2290 break;
2291 copied = status;
2292
2293 cond_resched();
2294
Nick Piggin124d3b72008-02-02 15:01:17 +01002295 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002296 if (unlikely(copied == 0)) {
2297 /*
2298 * If we were unable to copy any data at all, we must
2299 * fall back to a single segment length write.
2300 *
2301 * If we didn't fallback here, we could livelock
2302 * because not all segments in the iov can be copied at
2303 * once without a pagefault.
2304 */
2305 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2306 iov_iter_single_seg_count(i));
2307 goto again;
2308 }
Nick Pigginafddba42007-10-16 01:25:01 -07002309 pos += copied;
2310 written += copied;
2311
2312 balance_dirty_pages_ratelimited(mapping);
2313
2314 } while (iov_iter_count(i));
2315
2316 return written ? written : status;
2317}
2318
2319ssize_t
2320generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2321 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2322 size_t count, ssize_t written)
2323{
2324 struct file *file = iocb->ki_filp;
2325 struct address_space *mapping = file->f_mapping;
2326 const struct address_space_operations *a_ops = mapping->a_ops;
2327 struct inode *inode = mapping->host;
2328 ssize_t status;
2329 struct iov_iter i;
2330
2331 iov_iter_init(&i, iov, nr_segs, count, written);
2332 if (a_ops->write_begin)
2333 status = generic_perform_write(file, &i, pos);
2334 else
2335 status = generic_perform_write_2copy(file, &i, pos);
2336
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002338 written += status;
2339 *ppos = pos + status;
2340
2341 /*
2342 * For now, when the user asks for O_SYNC, we'll actually give
2343 * O_DSYNC
2344 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2346 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2347 status = generic_osync_inode(inode, mapping,
2348 OSYNC_METADATA|OSYNC_DATA);
2349 }
2350 }
2351
2352 /*
2353 * If we get here for O_DIRECT writes then we must have fallen through
2354 * to buffered writes (block instantiation inside i_size). So we sync
2355 * the file data here, to try to honour O_DIRECT expectations.
2356 */
2357 if (unlikely(file->f_flags & O_DIRECT) && written)
2358 status = filemap_write_and_wait(mapping);
2359
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 return written ? written : status;
2361}
2362EXPORT_SYMBOL(generic_file_buffered_write);
2363
Adrian Bunk5ce78522005-09-10 00:26:28 -07002364static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2366 unsigned long nr_segs, loff_t *ppos)
2367{
2368 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002369 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 size_t ocount; /* original count */
2371 size_t count; /* after file limit checks */
2372 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 loff_t pos;
2374 ssize_t written;
2375 ssize_t err;
2376
2377 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002378 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2379 if (err)
2380 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
2382 count = ocount;
2383 pos = *ppos;
2384
2385 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2386
2387 /* We can write back this queue in page reclaim */
2388 current->backing_dev_info = mapping->backing_dev_info;
2389 written = 0;
2390
2391 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2392 if (err)
2393 goto out;
2394
2395 if (count == 0)
2396 goto out;
2397
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002398 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (err)
2400 goto out;
2401
Christoph Hellwig870f4812006-01-09 20:52:01 -08002402 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
2404 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2405 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002406 loff_t endbyte;
2407 ssize_t written_buffered;
2408
2409 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2410 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 if (written < 0 || written == count)
2412 goto out;
2413 /*
2414 * direct-io write to a hole: fall through to buffered I/O
2415 * for completing the rest of the request.
2416 */
2417 pos += written;
2418 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002419 written_buffered = generic_file_buffered_write(iocb, iov,
2420 nr_segs, pos, ppos, count,
2421 written);
2422 /*
2423 * If generic_file_buffered_write() retuned a synchronous error
2424 * then we want to return the number of bytes which were
2425 * direct-written, or the error code if that was zero. Note
2426 * that this differs from normal direct-io semantics, which
2427 * will return -EFOO even if some bytes were written.
2428 */
2429 if (written_buffered < 0) {
2430 err = written_buffered;
2431 goto out;
2432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002434 /*
2435 * We need to ensure that the page cache pages are written to
2436 * disk and invalidated to preserve the expected O_DIRECT
2437 * semantics.
2438 */
2439 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002440 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2441 SYNC_FILE_RANGE_WAIT_BEFORE|
2442 SYNC_FILE_RANGE_WRITE|
2443 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002444 if (err == 0) {
2445 written = written_buffered;
2446 invalidate_mapping_pages(mapping,
2447 pos >> PAGE_CACHE_SHIFT,
2448 endbyte >> PAGE_CACHE_SHIFT);
2449 } else {
2450 /*
2451 * We don't know how much we wrote, so just return
2452 * the number of bytes which were direct-written
2453 */
2454 }
2455 } else {
2456 written = generic_file_buffered_write(iocb, iov, nr_segs,
2457 pos, ppos, count, written);
2458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459out:
2460 current->backing_dev_info = NULL;
2461 return written ? written : err;
2462}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Badari Pulavarty027445c2006-09-30 23:28:46 -07002464ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2465 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
2467 struct file *file = iocb->ki_filp;
2468 struct address_space *mapping = file->f_mapping;
2469 struct inode *inode = mapping->host;
2470 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Badari Pulavarty027445c2006-09-30 23:28:46 -07002472 BUG_ON(iocb->ki_pos != pos);
2473
2474 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2475 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
2477 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002478 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
2480 err = sync_page_range_nolock(inode, mapping, pos, ret);
2481 if (err < 0)
2482 ret = err;
2483 }
2484 return ret;
2485}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002486EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
Badari Pulavarty027445c2006-09-30 23:28:46 -07002488ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2489 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490{
2491 struct file *file = iocb->ki_filp;
2492 struct address_space *mapping = file->f_mapping;
2493 struct inode *inode = mapping->host;
2494 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
2496 BUG_ON(iocb->ki_pos != pos);
2497
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002498 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002499 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2500 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002501 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
2503 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2504 ssize_t err;
2505
2506 err = sync_page_range(inode, mapping, pos, ret);
2507 if (err < 0)
2508 ret = err;
2509 }
2510 return ret;
2511}
2512EXPORT_SYMBOL(generic_file_aio_write);
2513
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002515 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 * went wrong during pagecache shootdown.
2517 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002518static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2520 loff_t offset, unsigned long nr_segs)
2521{
2522 struct file *file = iocb->ki_filp;
2523 struct address_space *mapping = file->f_mapping;
2524 ssize_t retval;
Zach Brown65b82912007-03-16 13:38:11 -08002525 size_t write_len;
2526 pgoff_t end = 0; /* silence gcc */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
2528 /*
2529 * If it's a write, unmap all mmappings of the file up-front. This
2530 * will cause any pte dirty bits to be propagated into the pageframes
2531 * for the subsequent filemap_write_and_wait().
2532 */
2533 if (rw == WRITE) {
2534 write_len = iov_length(iov, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002535 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 if (mapping_mapped(mapping))
2537 unmap_mapping_range(mapping, offset, write_len, 0);
2538 }
2539
2540 retval = filemap_write_and_wait(mapping);
Zach Brown65b82912007-03-16 13:38:11 -08002541 if (retval)
2542 goto out;
2543
2544 /*
2545 * After a write we want buffered reads to be sure to go to disk to get
2546 * the new data. We invalidate clean cached page from the region we're
2547 * about to write. We do this *before* the write so that we can return
2548 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2549 */
2550 if (rw == WRITE && mapping->nrpages) {
2551 retval = invalidate_inode_pages2_range(mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002553 if (retval)
2554 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 }
Zach Brown65b82912007-03-16 13:38:11 -08002556
2557 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002558
2559 /*
2560 * Finally, try again to invalidate clean pages which might have been
Zach Brownbdb76ef2007-10-30 11:45:46 -07002561 * cached by non-direct readahead, or faulted in by get_user_pages()
2562 * if the source of the write was an mmap'ed region of the file
2563 * we're writing. Either one is a pretty crazy thing to do,
2564 * so we don't support it 100%. If this invalidation
2565 * fails, tough, the write still worked...
Zach Brown65b82912007-03-16 13:38:11 -08002566 */
2567 if (rw == WRITE && mapping->nrpages) {
Zach Brownbdb76ef2007-10-30 11:45:46 -07002568 invalidate_inode_pages2_range(mapping, offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002569 }
2570out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 return retval;
2572}
David Howellscf9a2ae2006-08-29 19:05:54 +01002573
2574/**
2575 * try_to_release_page() - release old fs-specific metadata on a page
2576 *
2577 * @page: the page which the kernel is trying to free
2578 * @gfp_mask: memory allocation flags (and I/O mode)
2579 *
2580 * The address_space is to try to release any data against the page
2581 * (presumably at page->private). If the release was successful, return `1'.
2582 * Otherwise return zero.
2583 *
2584 * The @gfp_mask argument specifies whether I/O may be performed to release
2585 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2586 *
2587 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2588 */
2589int try_to_release_page(struct page *page, gfp_t gfp_mask)
2590{
2591 struct address_space * const mapping = page->mapping;
2592
2593 BUG_ON(!PageLocked(page));
2594 if (PageWriteback(page))
2595 return 0;
2596
2597 if (mapping && mapping->a_ops->releasepage)
2598 return mapping->a_ops->releasepage(page, gfp_mask);
2599 return try_to_free_buffers(page);
2600}
2601
2602EXPORT_SYMBOL(try_to_release_page);