blob: 4eb958c402fe298b011c82c1d1cbe868b3f8e485 [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>
Emil Medve3a424f22007-10-24 14:18:32 +020031#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/security.h>
33#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080034#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070035#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080036#include <linux/memcontrol.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
42#include <linux/buffer_head.h> /* for generic_osync_inode */
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
Adrian Bunk5ce78522005-09-10 00:26:28 -070046static ssize_t
47generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
48 loff_t offset, unsigned long nr_segs);
49
Linus Torvalds1da177e2005-04-16 15:20:36 -070050/*
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * though.
53 *
54 * Shared mappings now work. 15.8.1995 Bruno.
55 *
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 *
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 */
61
62/*
63 * Lock ordering:
64 *
65 * ->i_mmap_lock (vmtruncate)
66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070067 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080070 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * ->i_mmap_lock (truncate->unmap_mapping_range)
72 *
73 * ->mmap_sem
74 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070075 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
Nick Piggin82591e62006-10-19 23:29:10 -070081 * ->i_mutex (generic_file_buffered_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080084 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * ->i_alloc_sem (various)
86 *
87 * ->inode_lock
88 * ->sb_lock (fs/fs-writeback.c)
89 * ->mapping->tree_lock (__sync_single_inode)
90 *
91 * ->i_mmap_lock
92 * ->anon_vma.lock (vma_adjust)
93 *
94 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070095 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070097 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070098 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * ->private_lock (try_to_unmap_one)
100 * ->tree_lock (try_to_unmap_one)
101 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -0800102 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 * ->private_lock (page_remove_rmap->set_page_dirty)
104 * ->tree_lock (page_remove_rmap->set_page_dirty)
105 * ->inode_lock (page_remove_rmap->set_page_dirty)
106 * ->inode_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 *
109 * ->task->proc_lock
110 * ->dcache_lock (proc_pid_lookup)
111 */
112
113/*
114 * Remove a page from the page cache and free it. Caller has to make
115 * sure the page is locked and that nobody else uses it - or that usage
116 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
117 */
118void __remove_from_page_cache(struct page *page)
119{
120 struct address_space *mapping = page->mapping;
121
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800122 mem_cgroup_uncharge_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 radix_tree_delete(&mapping->page_tree, page->index);
124 page->mapping = NULL;
125 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700126 __dec_zone_page_state(page, NR_FILE_PAGES);
Nick Piggin45426812007-07-15 23:38:12 -0700127 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800128
129 /*
130 * Some filesystems seem to re-dirty the page even after
131 * the VM has canceled the dirty bit (eg ext3 journaling).
132 *
133 * Fix it up by doing a final dirty accounting check after
134 * having removed the page entirely.
135 */
136 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
137 dec_zone_page_state(page, NR_FILE_DIRTY);
138 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
142void remove_from_page_cache(struct page *page)
143{
144 struct address_space *mapping = page->mapping;
145
Matt Mackallcd7619d2005-05-01 08:59:01 -0700146 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148 write_lock_irq(&mapping->tree_lock);
149 __remove_from_page_cache(page);
150 write_unlock_irq(&mapping->tree_lock);
151}
152
153static int sync_page(void *word)
154{
155 struct address_space *mapping;
156 struct page *page;
157
Andi Kleen07808b72005-11-05 17:25:53 +0100158 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
160 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700161 * page_mapping() is being called without PG_locked held.
162 * Some knowledge of the state and use of the page is used to
163 * reduce the requirements down to a memory barrier.
164 * The danger here is of a stale page_mapping() return value
165 * indicating a struct address_space different from the one it's
166 * associated with when it is associated with one.
167 * After smp_mb(), it's either the correct page_mapping() for
168 * the page, or an old page_mapping() and the page's own
169 * page_mapping() has gone NULL.
170 * The ->sync_page() address_space operation must tolerate
171 * page_mapping() going NULL. By an amazing coincidence,
172 * this comes about because none of the users of the page
173 * in the ->sync_page() methods make essential use of the
174 * page_mapping(), merely passing the page down to the backing
175 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700176 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700177 * of interest. When page_mapping() does go NULL, the entire
178 * call stack gracefully ignores the page and returns.
179 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
181 smp_mb();
182 mapping = page_mapping(page);
183 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
184 mapping->a_ops->sync_page(page);
185 io_schedule();
186 return 0;
187}
188
Matthew Wilcox2687a352007-12-06 11:18:49 -0500189static int sync_page_killable(void *word)
190{
191 sync_page(word);
192 return fatal_signal_pending(current) ? -EINTR : 0;
193}
194
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700196 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700197 * @mapping: address space structure to write
198 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800199 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700200 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700202 * Start writeback against all of a mapping's dirty pages that lie
203 * within the byte offsets <start, end> inclusive.
204 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700206 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 * these two operations is that if a dirty page/buffer is encountered, it must
208 * be waited upon, and not just skipped over.
209 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800210int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
211 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
213 int ret;
214 struct writeback_control wbc = {
215 .sync_mode = sync_mode,
216 .nr_to_write = mapping->nrpages * 2,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700217 .range_start = start,
218 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 };
220
221 if (!mapping_cap_writeback_dirty(mapping))
222 return 0;
223
224 ret = do_writepages(mapping, &wbc);
225 return ret;
226}
227
228static inline int __filemap_fdatawrite(struct address_space *mapping,
229 int sync_mode)
230{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700231 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
234int filemap_fdatawrite(struct address_space *mapping)
235{
236 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
237}
238EXPORT_SYMBOL(filemap_fdatawrite);
239
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800240static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
241 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
243 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
244}
245
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/**
347 * sync_page_range_nolock
348 * @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{
580 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
581 if (!test_clear_page_writeback(page))
582 BUG();
583 }
584 smp_mb__after_clear_bit();
585 wake_up_page(page, PG_writeback);
586}
587EXPORT_SYMBOL(end_page_writeback);
588
Randy Dunlap485bb992006-06-23 02:03:49 -0700589/**
590 * __lock_page - get a lock on the page, assuming we need to sleep to get it
591 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700593 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
595 * chances are that on the second loop, the block layer's plug list is empty,
596 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
597 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800598void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
600 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
601
602 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
603 TASK_UNINTERRUPTIBLE);
604}
605EXPORT_SYMBOL(__lock_page);
606
Matthew Wilcox2687a352007-12-06 11:18:49 -0500607int fastcall __lock_page_killable(struct page *page)
608{
609 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
610
611 return __wait_on_bit_lock(page_waitqueue(page), &wait,
612 sync_page_killable, TASK_KILLABLE);
613}
614
Nick Piggindb376482006-09-25 23:31:24 -0700615/*
616 * Variant of lock_page that does not require the caller to hold a reference
617 * on the page's mapping.
618 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800619void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700620{
621 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
622 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
623 TASK_UNINTERRUPTIBLE);
624}
625
Randy Dunlap485bb992006-06-23 02:03:49 -0700626/**
627 * find_get_page - find and get a page reference
628 * @mapping: the address_space to search
629 * @offset: the page index
630 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700631 * Is there a pagecache struct page at the given (mapping, offset) tuple?
632 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 */
Fengguang Wu57f6b962007-10-16 01:24:37 -0700634struct page * find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
636 struct page *page;
637
638 read_lock_irq(&mapping->tree_lock);
639 page = radix_tree_lookup(&mapping->page_tree, offset);
640 if (page)
641 page_cache_get(page);
642 read_unlock_irq(&mapping->tree_lock);
643 return page;
644}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645EXPORT_SYMBOL(find_get_page);
646
Randy Dunlap485bb992006-06-23 02:03:49 -0700647/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700649 * @mapping: the address_space to search
650 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 *
652 * Locates the desired pagecache page, locks it, increments its reference
653 * count and returns its address.
654 *
655 * Returns zero if the page was not present. find_lock_page() may sleep.
656 */
657struct page *find_lock_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700658 pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
660 struct page *page;
661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662repeat:
Nick Piggin45726cb2007-10-16 01:24:41 -0700663 read_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 page = radix_tree_lookup(&mapping->page_tree, offset);
665 if (page) {
666 page_cache_get(page);
667 if (TestSetPageLocked(page)) {
668 read_unlock_irq(&mapping->tree_lock);
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800669 __lock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
671 /* Has the page been truncated while we slept? */
Nick Piggin45726cb2007-10-16 01:24:41 -0700672 if (unlikely(page->mapping != mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 unlock_page(page);
674 page_cache_release(page);
675 goto repeat;
676 }
Nick Piggin45726cb2007-10-16 01:24:41 -0700677 VM_BUG_ON(page->index != offset);
678 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 }
680 }
681 read_unlock_irq(&mapping->tree_lock);
Nick Piggin45726cb2007-10-16 01:24:41 -0700682out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 return page;
684}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685EXPORT_SYMBOL(find_lock_page);
686
687/**
688 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700689 * @mapping: the page's address_space
690 * @index: the page's index into the mapping
691 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 *
693 * Locates a page in the pagecache. If the page is not present, a new page
694 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
695 * LRU list. The returned page is locked and has its reference count
696 * incremented.
697 *
698 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
699 * allocation!
700 *
701 * find_or_create_page() returns the desired page's address, or zero on
702 * memory exhaustion.
703 */
704struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700705 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
Nick Piggineb2be182007-10-16 01:24:57 -0700707 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 int err;
709repeat:
710 page = find_lock_page(mapping, index);
711 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700712 page = __page_cache_alloc(gfp_mask);
713 if (!page)
714 return NULL;
715 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
716 if (unlikely(err)) {
717 page_cache_release(page);
718 page = NULL;
719 if (err == -EEXIST)
720 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 return page;
724}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725EXPORT_SYMBOL(find_or_create_page);
726
727/**
728 * find_get_pages - gang pagecache lookup
729 * @mapping: The address_space to search
730 * @start: The starting page index
731 * @nr_pages: The maximum number of pages
732 * @pages: Where the resulting pages are placed
733 *
734 * find_get_pages() will search for and return a group of up to
735 * @nr_pages pages in the mapping. The pages are placed at @pages.
736 * find_get_pages() takes a reference against the returned pages.
737 *
738 * The search returns a group of mapping-contiguous pages with ascending
739 * indexes. There may be holes in the indices due to not-present pages.
740 *
741 * find_get_pages() returns the number of pages which were found.
742 */
743unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
744 unsigned int nr_pages, struct page **pages)
745{
746 unsigned int i;
747 unsigned int ret;
748
749 read_lock_irq(&mapping->tree_lock);
750 ret = radix_tree_gang_lookup(&mapping->page_tree,
751 (void **)pages, start, nr_pages);
752 for (i = 0; i < ret; i++)
753 page_cache_get(pages[i]);
754 read_unlock_irq(&mapping->tree_lock);
755 return ret;
756}
757
Jens Axboeebf43502006-04-27 08:46:01 +0200758/**
759 * find_get_pages_contig - gang contiguous pagecache lookup
760 * @mapping: The address_space to search
761 * @index: The starting page index
762 * @nr_pages: The maximum number of pages
763 * @pages: Where the resulting pages are placed
764 *
765 * find_get_pages_contig() works exactly like find_get_pages(), except
766 * that the returned number of pages are guaranteed to be contiguous.
767 *
768 * find_get_pages_contig() returns the number of pages which were found.
769 */
770unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
771 unsigned int nr_pages, struct page **pages)
772{
773 unsigned int i;
774 unsigned int ret;
775
776 read_lock_irq(&mapping->tree_lock);
777 ret = radix_tree_gang_lookup(&mapping->page_tree,
778 (void **)pages, index, nr_pages);
779 for (i = 0; i < ret; i++) {
780 if (pages[i]->mapping == NULL || pages[i]->index != index)
781 break;
782
783 page_cache_get(pages[i]);
784 index++;
785 }
786 read_unlock_irq(&mapping->tree_lock);
787 return i;
788}
David Howellsef71c152007-05-09 02:33:44 -0700789EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200790
Randy Dunlap485bb992006-06-23 02:03:49 -0700791/**
792 * find_get_pages_tag - find and return pages that match @tag
793 * @mapping: the address_space to search
794 * @index: the starting page index
795 * @tag: the tag index
796 * @nr_pages: the maximum number of pages
797 * @pages: where the resulting pages are placed
798 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700800 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 */
802unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
803 int tag, unsigned int nr_pages, struct page **pages)
804{
805 unsigned int i;
806 unsigned int ret;
807
808 read_lock_irq(&mapping->tree_lock);
809 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
810 (void **)pages, *index, nr_pages, tag);
811 for (i = 0; i < ret; i++)
812 page_cache_get(pages[i]);
813 if (ret)
814 *index = pages[ret - 1]->index + 1;
815 read_unlock_irq(&mapping->tree_lock);
816 return ret;
817}
David Howellsef71c152007-05-09 02:33:44 -0700818EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Randy Dunlap485bb992006-06-23 02:03:49 -0700820/**
821 * grab_cache_page_nowait - returns locked page at given index in given cache
822 * @mapping: target address_space
823 * @index: the page index
824 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800825 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 * This is intended for speculative data generators, where the data can
827 * be regenerated if the page couldn't be grabbed. This routine should
828 * be safe to call while holding the lock for another page.
829 *
830 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
831 * and deadlock against the caller's locked page.
832 */
833struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700834grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
836 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
838 if (page) {
839 if (!TestSetPageLocked(page))
840 return page;
841 page_cache_release(page);
842 return NULL;
843 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700844 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
845 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 page_cache_release(page);
847 page = NULL;
848 }
849 return page;
850}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851EXPORT_SYMBOL(grab_cache_page_nowait);
852
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700853/*
854 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
855 * a _large_ part of the i/o request. Imagine the worst scenario:
856 *
857 * ---R__________________________________________B__________
858 * ^ reading here ^ bad block(assume 4k)
859 *
860 * read(R) => miss => readahead(R...B) => media error => frustrating retries
861 * => failing the whole request => read(R) => read(R+1) =>
862 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
863 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
864 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
865 *
866 * It is going insane. Fix it by quickly scaling down the readahead size.
867 */
868static void shrink_readahead_size_eio(struct file *filp,
869 struct file_ra_state *ra)
870{
871 if (!ra->ra_pages)
872 return;
873
874 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700875}
876
Randy Dunlap485bb992006-06-23 02:03:49 -0700877/**
878 * do_generic_mapping_read - generic file read routine
879 * @mapping: address_space to be read
Randy Dunlap8f731f72007-10-18 23:39:28 -0700880 * @ra: file's readahead state
Randy Dunlap485bb992006-06-23 02:03:49 -0700881 * @filp: the file to read
882 * @ppos: current file position
883 * @desc: read_descriptor
884 * @actor: read method
885 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700887 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 *
889 * This is really ugly. But the goto's actually try to clarify some
890 * of the logic when it comes to error handling etc.
891 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700892 * Note the struct file* is only passed for the use of readpage.
893 * It may be NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 */
895void do_generic_mapping_read(struct address_space *mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700896 struct file_ra_state *ra,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 struct file *filp,
898 loff_t *ppos,
899 read_descriptor_t *desc,
900 read_actor_t actor)
901{
902 struct inode *inode = mapping->host;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700903 pgoff_t index;
904 pgoff_t last_index;
905 pgoff_t prev_index;
906 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700907 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700911 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
912 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
914 offset = *ppos & ~PAGE_CACHE_MASK;
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 for (;;) {
917 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700918 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -0700919 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 unsigned long nr, ret;
921
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923find_page:
924 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700925 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700926 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700927 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700928 index, last_index - index);
929 page = find_get_page(mapping, index);
930 if (unlikely(page == NULL))
931 goto no_cached_page;
932 }
933 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700934 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700935 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700936 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 }
938 if (!PageUptodate(page))
939 goto page_not_up_to_date;
940page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -0700941 /*
942 * i_size must be checked after we know the page is Uptodate.
943 *
944 * Checking i_size after the check allows us to calculate
945 * the correct value for "nr", which means the zero-filled
946 * part of the page is not copied back to userspace (unless
947 * another truncate extends the file - this is desired though).
948 */
949
950 isize = i_size_read(inode);
951 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
952 if (unlikely(!isize || index > end_index)) {
953 page_cache_release(page);
954 goto out;
955 }
956
957 /* nr is the maximum number of bytes to copy from this page */
958 nr = PAGE_CACHE_SIZE;
959 if (index == end_index) {
960 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
961 if (nr <= offset) {
962 page_cache_release(page);
963 goto out;
964 }
965 }
966 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
968 /* If users can be writing to this page using arbitrary
969 * virtual addresses, take care about potential aliasing
970 * before reading the page on the kernel side.
971 */
972 if (mapping_writably_mapped(mapping))
973 flush_dcache_page(page);
974
975 /*
Jan Karaec0f1632007-05-06 14:49:25 -0700976 * When a sequential read accesses a page several times,
977 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 */
Jan Karaec0f1632007-05-06 14:49:25 -0700979 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 mark_page_accessed(page);
981 prev_index = index;
982
983 /*
984 * Ok, we have the page, and it's up-to-date, so
985 * now we can copy it to user space...
986 *
987 * The actor routine returns how many bytes were actually used..
988 * NOTE! This may not be the same as how much of a user buffer
989 * we filled up (we may be padding etc), so we can only update
990 * "pos" here (the actor routine has to update the user buffer
991 * pointers and the remaining count).
992 */
993 ret = actor(desc, page, offset, nr);
994 offset += ret;
995 index += offset >> PAGE_CACHE_SHIFT;
996 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -0700997 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
999 page_cache_release(page);
1000 if (ret == nr && desc->count)
1001 continue;
1002 goto out;
1003
1004page_not_up_to_date:
1005 /* Get exclusive access to the page ... */
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001006 if (lock_page_killable(page))
1007 goto readpage_eio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
Nick Pigginda6052f2006-09-25 23:31:35 -07001009 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 if (!page->mapping) {
1011 unlock_page(page);
1012 page_cache_release(page);
1013 continue;
1014 }
1015
1016 /* Did somebody else fill it already? */
1017 if (PageUptodate(page)) {
1018 unlock_page(page);
1019 goto page_ok;
1020 }
1021
1022readpage:
1023 /* Start the actual read. The read will unlock the page. */
1024 error = mapping->a_ops->readpage(filp, page);
1025
Zach Brown994fc28c2005-12-15 14:28:17 -08001026 if (unlikely(error)) {
1027 if (error == AOP_TRUNCATED_PAGE) {
1028 page_cache_release(page);
1029 goto find_page;
1030 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 if (!PageUptodate(page)) {
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001035 if (lock_page_killable(page))
1036 goto readpage_eio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 if (!PageUptodate(page)) {
1038 if (page->mapping == NULL) {
1039 /*
1040 * invalidate_inode_pages got it
1041 */
1042 unlock_page(page);
1043 page_cache_release(page);
1044 goto find_page;
1045 }
1046 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001047 shrink_readahead_size_eio(filp, ra);
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001048 goto readpage_eio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 }
1050 unlock_page(page);
1051 }
1052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 goto page_ok;
1054
Matthew Wilcox0b94e972007-12-06 11:19:57 -05001055readpage_eio:
1056 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057readpage_error:
1058 /* UHHUH! A synchronous read error occurred. Report it */
1059 desc->error = error;
1060 page_cache_release(page);
1061 goto out;
1062
1063no_cached_page:
1064 /*
1065 * Ok, it wasn't cached, so we need to create a new
1066 * page..
1067 */
Nick Piggineb2be182007-10-16 01:24:57 -07001068 page = page_cache_alloc_cold(mapping);
1069 if (!page) {
1070 desc->error = -ENOMEM;
1071 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
Nick Piggineb2be182007-10-16 01:24:57 -07001073 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 index, GFP_KERNEL);
1075 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001076 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 if (error == -EEXIST)
1078 goto find_page;
1079 desc->error = error;
1080 goto out;
1081 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 goto readpage;
1083 }
1084
1085out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001086 ra->prev_pos = prev_index;
1087 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1088 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001090 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (filp)
1092 file_accessed(filp);
1093}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094EXPORT_SYMBOL(do_generic_mapping_read);
1095
1096int file_read_actor(read_descriptor_t *desc, struct page *page,
1097 unsigned long offset, unsigned long size)
1098{
1099 char *kaddr;
1100 unsigned long left, count = desc->count;
1101
1102 if (size > count)
1103 size = count;
1104
1105 /*
1106 * Faults on the destination of a read are common, so do it before
1107 * taking the kmap.
1108 */
1109 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1110 kaddr = kmap_atomic(page, KM_USER0);
1111 left = __copy_to_user_inatomic(desc->arg.buf,
1112 kaddr + offset, size);
1113 kunmap_atomic(kaddr, KM_USER0);
1114 if (left == 0)
1115 goto success;
1116 }
1117
1118 /* Do it the slow way */
1119 kaddr = kmap(page);
1120 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1121 kunmap(page);
1122
1123 if (left) {
1124 size -= left;
1125 desc->error = -EFAULT;
1126 }
1127success:
1128 desc->count = count - size;
1129 desc->written += size;
1130 desc->arg.buf += size;
1131 return size;
1132}
1133
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001134/*
1135 * Performs necessary checks before doing a write
1136 * @iov: io vector request
1137 * @nr_segs: number of segments in the iovec
1138 * @count: number of bytes to write
1139 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1140 *
1141 * Adjust number of segments and amount of bytes to write (nr_segs should be
1142 * properly initialized first). Returns appropriate error code that caller
1143 * should return or zero in case that write should be allowed.
1144 */
1145int generic_segment_checks(const struct iovec *iov,
1146 unsigned long *nr_segs, size_t *count, int access_flags)
1147{
1148 unsigned long seg;
1149 size_t cnt = 0;
1150 for (seg = 0; seg < *nr_segs; seg++) {
1151 const struct iovec *iv = &iov[seg];
1152
1153 /*
1154 * If any segment has a negative length, or the cumulative
1155 * length ever wraps negative then return -EINVAL.
1156 */
1157 cnt += iv->iov_len;
1158 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1159 return -EINVAL;
1160 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1161 continue;
1162 if (seg == 0)
1163 return -EFAULT;
1164 *nr_segs = seg;
1165 cnt -= iv->iov_len; /* This segment is no good */
1166 break;
1167 }
1168 *count = cnt;
1169 return 0;
1170}
1171EXPORT_SYMBOL(generic_segment_checks);
1172
Randy Dunlap485bb992006-06-23 02:03:49 -07001173/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001174 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001175 * @iocb: kernel I/O control block
1176 * @iov: io vector request
1177 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001178 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001179 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 * This is the "read()" routine for all filesystems
1181 * that can use the page cache directly.
1182 */
1183ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001184generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1185 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186{
1187 struct file *filp = iocb->ki_filp;
1188 ssize_t retval;
1189 unsigned long seg;
1190 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001191 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
1193 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001194 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1195 if (retval)
1196 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
1198 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1199 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001200 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 struct address_space *mapping;
1202 struct inode *inode;
1203
1204 mapping = filp->f_mapping;
1205 inode = mapping->host;
1206 retval = 0;
1207 if (!count)
1208 goto out; /* skip atime */
1209 size = i_size_read(inode);
1210 if (pos < size) {
1211 retval = generic_file_direct_IO(READ, iocb,
1212 iov, pos, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 if (retval > 0)
1214 *ppos = pos + retval;
1215 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001216 if (likely(retval != 0)) {
Steven Whitehouse3f1a9aa2006-09-27 14:52:48 -04001217 file_accessed(filp);
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001218 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
1221
1222 retval = 0;
1223 if (count) {
1224 for (seg = 0; seg < nr_segs; seg++) {
1225 read_descriptor_t desc;
1226
1227 desc.written = 0;
1228 desc.arg.buf = iov[seg].iov_base;
1229 desc.count = iov[seg].iov_len;
1230 if (desc.count == 0)
1231 continue;
1232 desc.error = 0;
1233 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1234 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001235 if (desc.error) {
1236 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 break;
1238 }
akpm@linux-foundation.orgc44939e2007-07-15 23:38:25 -07001239 if (desc.count > 0)
1240 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 }
1242 }
1243out:
1244 return retval;
1245}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246EXPORT_SYMBOL(generic_file_aio_read);
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248static ssize_t
1249do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001250 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
1252 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1253 return -EINVAL;
1254
1255 force_page_cache_readahead(mapping, filp, index,
1256 max_sane_readahead(nr));
1257 return 0;
1258}
1259
1260asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1261{
1262 ssize_t ret;
1263 struct file *file;
1264
1265 ret = -EBADF;
1266 file = fget(fd);
1267 if (file) {
1268 if (file->f_mode & FMODE_READ) {
1269 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001270 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1271 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 unsigned long len = end - start + 1;
1273 ret = do_readahead(mapping, file, start, len);
1274 }
1275 fput(file);
1276 }
1277 return ret;
1278}
1279
1280#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001281/**
1282 * page_cache_read - adds requested page to the page cache if not already there
1283 * @file: file to read
1284 * @offset: page index
1285 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 * This adds the requested page to the page cache if it isn't already there,
1287 * and schedules an I/O to read in its contents from disk.
1288 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001289static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290{
1291 struct address_space *mapping = file->f_mapping;
1292 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001293 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Zach Brown994fc28c2005-12-15 14:28:17 -08001295 do {
1296 page = page_cache_alloc_cold(mapping);
1297 if (!page)
1298 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Zach Brown994fc28c2005-12-15 14:28:17 -08001300 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1301 if (ret == 0)
1302 ret = mapping->a_ops->readpage(file, page);
1303 else if (ret == -EEXIST)
1304 ret = 0; /* losing race to add is OK */
1305
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Zach Brown994fc28c2005-12-15 14:28:17 -08001308 } while (ret == AOP_TRUNCATED_PAGE);
1309
1310 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311}
1312
1313#define MMAP_LOTSAMISS (100)
1314
Randy Dunlap485bb992006-06-23 02:03:49 -07001315/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001316 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001317 * @vma: vma in which the fault was taken
1318 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001319 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001320 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 * mapped memory region to read in file data during a page fault.
1322 *
1323 * The goto's are kind of ugly, but this streamlines the normal case of having
1324 * it in the page cache, and handles the special cases reasonably without
1325 * having a lot of duplicated code.
1326 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001327int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
1329 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001330 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 struct address_space *mapping = file->f_mapping;
1332 struct file_ra_state *ra = &file->f_ra;
1333 struct inode *inode = mapping->host;
1334 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001335 pgoff_t size;
Nick Piggin54cb8822007-07-19 01:46:59 -07001336 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001337 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001340 if (vmf->pgoff >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001341 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001344 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 goto no_cached_page;
1346
1347 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 * Do we have something in the page cache already?
1349 */
1350retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001351 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001352 /*
1353 * For sequential accesses, we use the generic readahead logic.
1354 */
1355 if (VM_SequentialReadHint(vma)) {
1356 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001357 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001358 vmf->pgoff, 1);
1359 page = find_lock_page(mapping, vmf->pgoff);
1360 if (!page)
1361 goto no_cached_page;
1362 }
1363 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001364 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001365 vmf->pgoff, 1);
1366 }
1367 }
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 if (!page) {
1370 unsigned long ra_pages;
1371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 ra->mmap_miss++;
1373
1374 /*
1375 * Do we miss much more than hit in this file? If so,
1376 * stop bothering with read-ahead. It will only hurt.
1377 */
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001378 if (ra->mmap_miss > MMAP_LOTSAMISS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 goto no_cached_page;
1380
1381 /*
1382 * To keep the pgmajfault counter straight, we need to
1383 * check did_readaround, as this is an inner loop.
1384 */
1385 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001386 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001387 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 }
1389 did_readaround = 1;
1390 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1391 if (ra_pages) {
1392 pgoff_t start = 0;
1393
Nick Piggind0217ac2007-07-19 01:47:03 -07001394 if (vmf->pgoff > ra_pages / 2)
1395 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 do_page_cache_readahead(mapping, file, start, ra_pages);
1397 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001398 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 if (!page)
1400 goto no_cached_page;
1401 }
1402
1403 if (!did_readaround)
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001404 ra->mmap_miss--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001407 * We have a locked page in the page cache, now we need to check
1408 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 */
Nick Piggind00806b2007-07-19 01:46:57 -07001410 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 goto page_not_uptodate;
1412
Nick Piggind00806b2007-07-19 01:46:57 -07001413 /* Must recheck i_size under page lock */
1414 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001415 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001416 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001417 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001418 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001419 }
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 /*
1422 * Found the page and have a reference on it.
1423 */
1424 mark_page_accessed(page);
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001425 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001426 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001427 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429no_cached_page:
1430 /*
1431 * We're only likely to ever get here if MADV_RANDOM is in
1432 * effect.
1433 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001434 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 /*
1437 * The page we want has now been added to the page cache.
1438 * In the unlikely event that someone removed it in the
1439 * meantime, we'll just come back here and read it again.
1440 */
1441 if (error >= 0)
1442 goto retry_find;
1443
1444 /*
1445 * An error return from page_cache_read can result if the
1446 * system is low on memory, or a problem occurs while trying
1447 * to schedule I/O.
1448 */
1449 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001450 return VM_FAULT_OOM;
1451 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001454 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001456 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001457 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
1460 /*
1461 * Umm, take care of errors if the page isn't up-to-date.
1462 * Try to re-read it _once_. We do this synchronously,
1463 * because there really aren't any performance issues here
1464 * and we need to check for errors.
1465 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001467 error = mapping->a_ops->readpage(file, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001469
1470 if (!error || error == AOP_TRUNCATED_PAGE)
1471 goto retry_find;
1472
1473 /* Things didn't work out. Return zero to tell the mm layer so. */
1474 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001475 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001476}
1477EXPORT_SYMBOL(filemap_fault);
1478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001480 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481};
1482
1483/* This is used for a general mmap of a disk file */
1484
1485int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1486{
1487 struct address_space *mapping = file->f_mapping;
1488
1489 if (!mapping->a_ops->readpage)
1490 return -ENOEXEC;
1491 file_accessed(file);
1492 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001493 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 return 0;
1495}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497/*
1498 * This is for filesystems which do not implement ->writepage.
1499 */
1500int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1501{
1502 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1503 return -EINVAL;
1504 return generic_file_mmap(file, vma);
1505}
1506#else
1507int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1508{
1509 return -ENOSYS;
1510}
1511int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1512{
1513 return -ENOSYS;
1514}
1515#endif /* CONFIG_MMU */
1516
1517EXPORT_SYMBOL(generic_file_mmap);
1518EXPORT_SYMBOL(generic_file_readonly_mmap);
1519
Nick Piggin6fe69002007-05-06 14:49:04 -07001520static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001521 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 int (*filler)(void *,struct page*),
1523 void *data)
1524{
Nick Piggineb2be182007-10-16 01:24:57 -07001525 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 int err;
1527repeat:
1528 page = find_get_page(mapping, index);
1529 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001530 page = page_cache_alloc_cold(mapping);
1531 if (!page)
1532 return ERR_PTR(-ENOMEM);
1533 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1534 if (unlikely(err)) {
1535 page_cache_release(page);
1536 if (err == -EEXIST)
1537 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 return ERR_PTR(err);
1540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 err = filler(data, page);
1542 if (err < 0) {
1543 page_cache_release(page);
1544 page = ERR_PTR(err);
1545 }
1546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 return page;
1548}
1549
Nick Piggin6fe69002007-05-06 14:49:04 -07001550/*
1551 * Same as read_cache_page, but don't wait for page to become unlocked
1552 * after submitting it to the filler.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001554struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001555 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 int (*filler)(void *,struct page*),
1557 void *data)
1558{
1559 struct page *page;
1560 int err;
1561
1562retry:
1563 page = __read_cache_page(mapping, index, filler, data);
1564 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001565 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 if (PageUptodate(page))
1567 goto out;
1568
1569 lock_page(page);
1570 if (!page->mapping) {
1571 unlock_page(page);
1572 page_cache_release(page);
1573 goto retry;
1574 }
1575 if (PageUptodate(page)) {
1576 unlock_page(page);
1577 goto out;
1578 }
1579 err = filler(data, page);
1580 if (err < 0) {
1581 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001582 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 }
David Howellsc855ff32007-05-09 13:42:20 +01001584out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001585 mark_page_accessed(page);
1586 return page;
1587}
1588EXPORT_SYMBOL(read_cache_page_async);
1589
1590/**
1591 * read_cache_page - read into page cache, fill it if needed
1592 * @mapping: the page's address_space
1593 * @index: the page index
1594 * @filler: function to perform the read
1595 * @data: destination for read data
1596 *
1597 * Read into the page cache. If a page already exists, and PageUptodate() is
1598 * not set, try to fill the page then wait for it to become unlocked.
1599 *
1600 * If the page does not get brought uptodate, return -EIO.
1601 */
1602struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001603 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001604 int (*filler)(void *,struct page*),
1605 void *data)
1606{
1607 struct page *page;
1608
1609 page = read_cache_page_async(mapping, index, filler, data);
1610 if (IS_ERR(page))
1611 goto out;
1612 wait_on_page_locked(page);
1613 if (!PageUptodate(page)) {
1614 page_cache_release(page);
1615 page = ERR_PTR(-EIO);
1616 }
1617 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 return page;
1619}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620EXPORT_SYMBOL(read_cache_page);
1621
1622/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 * The logic we want is
1624 *
1625 * if suid or (sgid and xgrp)
1626 * remove privs
1627 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001628int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629{
1630 mode_t mode = dentry->d_inode->i_mode;
1631 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
1633 /* suid always must be killed */
1634 if (unlikely(mode & S_ISUID))
1635 kill = ATTR_KILL_SUID;
1636
1637 /*
1638 * sgid without any exec bits is just a mandatory locking mark; leave
1639 * it alone. If some exec bits are set, it's a real sgid; kill it.
1640 */
1641 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1642 kill |= ATTR_KILL_SGID;
1643
Jens Axboe01de85e2006-10-17 19:50:36 +02001644 if (unlikely(kill && !capable(CAP_FSETID)))
1645 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Jens Axboe01de85e2006-10-17 19:50:36 +02001647 return 0;
1648}
Mark Fashehd23a1472006-10-17 17:05:18 -07001649EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001650
1651int __remove_suid(struct dentry *dentry, int kill)
1652{
1653 struct iattr newattrs;
1654
1655 newattrs.ia_valid = ATTR_FORCE | kill;
1656 return notify_change(dentry, &newattrs);
1657}
1658
1659int remove_suid(struct dentry *dentry)
1660{
Serge E. Hallynb5376772007-10-16 23:31:36 -07001661 int killsuid = should_remove_suid(dentry);
1662 int killpriv = security_inode_need_killpriv(dentry);
1663 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001664
Serge E. Hallynb5376772007-10-16 23:31:36 -07001665 if (killpriv < 0)
1666 return killpriv;
1667 if (killpriv)
1668 error = security_inode_killpriv(dentry);
1669 if (!error && killsuid)
1670 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001671
Serge E. Hallynb5376772007-10-16 23:31:36 -07001672 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
1674EXPORT_SYMBOL(remove_suid);
1675
Nick Piggin2f718ff2007-10-16 01:24:59 -07001676static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 const struct iovec *iov, size_t base, size_t bytes)
1678{
1679 size_t copied = 0, left = 0;
1680
1681 while (bytes) {
1682 char __user *buf = iov->iov_base + base;
1683 int copy = min(bytes, iov->iov_len - base);
1684
1685 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001686 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 copied += copy;
1688 bytes -= copy;
1689 vaddr += copy;
1690 iov++;
1691
NeilBrown01408c42006-06-25 05:47:58 -07001692 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 }
1695 return copied - left;
1696}
1697
1698/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001699 * Copy as much as we can into the page and return the number of bytes which
1700 * were sucessfully copied. If a fault is encountered then return the number of
1701 * bytes which were copied.
1702 */
1703size_t iov_iter_copy_from_user_atomic(struct page *page,
1704 struct iov_iter *i, unsigned long offset, size_t bytes)
1705{
1706 char *kaddr;
1707 size_t copied;
1708
1709 BUG_ON(!in_atomic());
1710 kaddr = kmap_atomic(page, KM_USER0);
1711 if (likely(i->nr_segs == 1)) {
1712 int left;
1713 char __user *buf = i->iov->iov_base + i->iov_offset;
1714 left = __copy_from_user_inatomic_nocache(kaddr + offset,
1715 buf, bytes);
1716 copied = bytes - left;
1717 } else {
1718 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1719 i->iov, i->iov_offset, bytes);
1720 }
1721 kunmap_atomic(kaddr, KM_USER0);
1722
1723 return copied;
1724}
Nick Piggin89e10782007-10-16 01:25:07 -07001725EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001726
1727/*
1728 * This has the same sideeffects and return value as
1729 * iov_iter_copy_from_user_atomic().
1730 * The difference is that it attempts to resolve faults.
1731 * Page must not be locked.
1732 */
1733size_t iov_iter_copy_from_user(struct page *page,
1734 struct iov_iter *i, unsigned long offset, size_t bytes)
1735{
1736 char *kaddr;
1737 size_t copied;
1738
1739 kaddr = kmap(page);
1740 if (likely(i->nr_segs == 1)) {
1741 int left;
1742 char __user *buf = i->iov->iov_base + i->iov_offset;
1743 left = __copy_from_user_nocache(kaddr + offset, buf, bytes);
1744 copied = bytes - left;
1745 } else {
1746 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1747 i->iov, i->iov_offset, bytes);
1748 }
1749 kunmap(page);
1750 return copied;
1751}
Nick Piggin89e10782007-10-16 01:25:07 -07001752EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001753
1754static void __iov_iter_advance_iov(struct iov_iter *i, size_t bytes)
1755{
1756 if (likely(i->nr_segs == 1)) {
1757 i->iov_offset += bytes;
1758 } else {
1759 const struct iovec *iov = i->iov;
1760 size_t base = i->iov_offset;
1761
Nick Piggin124d3b72008-02-02 15:01:17 +01001762 /*
1763 * The !iov->iov_len check ensures we skip over unlikely
1764 * zero-length segments.
1765 */
1766 while (bytes || !iov->iov_len) {
Nick Piggin2f718ff2007-10-16 01:24:59 -07001767 int copy = min(bytes, iov->iov_len - base);
1768
1769 bytes -= copy;
1770 base += copy;
1771 if (iov->iov_len == base) {
1772 iov++;
1773 base = 0;
1774 }
1775 }
1776 i->iov = iov;
1777 i->iov_offset = base;
1778 }
1779}
1780
1781void iov_iter_advance(struct iov_iter *i, size_t bytes)
1782{
1783 BUG_ON(i->count < bytes);
1784
1785 __iov_iter_advance_iov(i, bytes);
1786 i->count -= bytes;
1787}
Nick Piggin89e10782007-10-16 01:25:07 -07001788EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001789
Nick Pigginafddba42007-10-16 01:25:01 -07001790/*
1791 * Fault in the first iovec of the given iov_iter, to a maximum length
1792 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1793 * accessed (ie. because it is an invalid address).
1794 *
1795 * writev-intensive code may want this to prefault several iovecs -- that
1796 * would be possible (callers must not rely on the fact that _only_ the
1797 * first iovec will be faulted with the current implementation).
1798 */
1799int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001800{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001801 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001802 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1803 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001804}
Nick Piggin89e10782007-10-16 01:25:07 -07001805EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001806
1807/*
1808 * Return the count of just the current iov_iter segment.
1809 */
1810size_t iov_iter_single_seg_count(struct iov_iter *i)
1811{
1812 const struct iovec *iov = i->iov;
1813 if (i->nr_segs == 1)
1814 return i->count;
1815 else
1816 return min(i->count, iov->iov_len - i->iov_offset);
1817}
Nick Piggin89e10782007-10-16 01:25:07 -07001818EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001819
1820/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 * Performs necessary checks before doing a write
1822 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001823 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 * Returns appropriate error code that caller should return or
1825 * zero in case that write should be allowed.
1826 */
1827inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1828{
1829 struct inode *inode = file->f_mapping->host;
1830 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1831
1832 if (unlikely(*pos < 0))
1833 return -EINVAL;
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 if (!isblk) {
1836 /* FIXME: this is for backwards compatibility with 2.4 */
1837 if (file->f_flags & O_APPEND)
1838 *pos = i_size_read(inode);
1839
1840 if (limit != RLIM_INFINITY) {
1841 if (*pos >= limit) {
1842 send_sig(SIGXFSZ, current, 0);
1843 return -EFBIG;
1844 }
1845 if (*count > limit - (typeof(limit))*pos) {
1846 *count = limit - (typeof(limit))*pos;
1847 }
1848 }
1849 }
1850
1851 /*
1852 * LFS rule
1853 */
1854 if (unlikely(*pos + *count > MAX_NON_LFS &&
1855 !(file->f_flags & O_LARGEFILE))) {
1856 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 return -EFBIG;
1858 }
1859 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1860 *count = MAX_NON_LFS - (unsigned long)*pos;
1861 }
1862 }
1863
1864 /*
1865 * Are we about to exceed the fs block limit ?
1866 *
1867 * If we have written data it becomes a short write. If we have
1868 * exceeded without writing data we send a signal and return EFBIG.
1869 * Linus frestrict idea will clean these up nicely..
1870 */
1871 if (likely(!isblk)) {
1872 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1873 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 return -EFBIG;
1875 }
1876 /* zero-length writes at ->s_maxbytes are OK */
1877 }
1878
1879 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1880 *count = inode->i_sb->s_maxbytes - *pos;
1881 } else {
David Howells93614012006-09-30 20:45:40 +02001882#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 loff_t isize;
1884 if (bdev_read_only(I_BDEV(inode)))
1885 return -EPERM;
1886 isize = i_size_read(inode);
1887 if (*pos >= isize) {
1888 if (*count || *pos > isize)
1889 return -ENOSPC;
1890 }
1891
1892 if (*pos + *count > isize)
1893 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02001894#else
1895 return -EPERM;
1896#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 }
1898 return 0;
1899}
1900EXPORT_SYMBOL(generic_write_checks);
1901
Nick Pigginafddba42007-10-16 01:25:01 -07001902int pagecache_write_begin(struct file *file, struct address_space *mapping,
1903 loff_t pos, unsigned len, unsigned flags,
1904 struct page **pagep, void **fsdata)
1905{
1906 const struct address_space_operations *aops = mapping->a_ops;
1907
1908 if (aops->write_begin) {
1909 return aops->write_begin(file, mapping, pos, len, flags,
1910 pagep, fsdata);
1911 } else {
1912 int ret;
1913 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1914 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1915 struct inode *inode = mapping->host;
1916 struct page *page;
1917again:
1918 page = __grab_cache_page(mapping, index);
1919 *pagep = page;
1920 if (!page)
1921 return -ENOMEM;
1922
1923 if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) {
1924 /*
1925 * There is no way to resolve a short write situation
1926 * for a !Uptodate page (except by double copying in
1927 * the caller done by generic_perform_write_2copy).
1928 *
1929 * Instead, we have to bring it uptodate here.
1930 */
1931 ret = aops->readpage(file, page);
1932 page_cache_release(page);
1933 if (ret) {
1934 if (ret == AOP_TRUNCATED_PAGE)
1935 goto again;
1936 return ret;
1937 }
1938 goto again;
1939 }
1940
1941 ret = aops->prepare_write(file, page, offset, offset+len);
1942 if (ret) {
Nick Piggin55144762007-10-16 01:25:26 -07001943 unlock_page(page);
Nick Pigginafddba42007-10-16 01:25:01 -07001944 page_cache_release(page);
1945 if (pos + len > inode->i_size)
1946 vmtruncate(inode, inode->i_size);
Nick Pigginafddba42007-10-16 01:25:01 -07001947 }
1948 return ret;
1949 }
1950}
1951EXPORT_SYMBOL(pagecache_write_begin);
1952
1953int pagecache_write_end(struct file *file, struct address_space *mapping,
1954 loff_t pos, unsigned len, unsigned copied,
1955 struct page *page, void *fsdata)
1956{
1957 const struct address_space_operations *aops = mapping->a_ops;
1958 int ret;
1959
1960 if (aops->write_end) {
1961 mark_page_accessed(page);
1962 ret = aops->write_end(file, mapping, pos, len, copied,
1963 page, fsdata);
1964 } else {
1965 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1966 struct inode *inode = mapping->host;
1967
1968 flush_dcache_page(page);
1969 ret = aops->commit_write(file, page, offset, offset+len);
1970 unlock_page(page);
1971 mark_page_accessed(page);
1972 page_cache_release(page);
Nick Pigginafddba42007-10-16 01:25:01 -07001973
1974 if (ret < 0) {
1975 if (pos + len > inode->i_size)
1976 vmtruncate(inode, inode->i_size);
1977 } else if (ret > 0)
1978 ret = min_t(size_t, copied, ret);
1979 else
1980 ret = copied;
1981 }
1982
1983 return ret;
1984}
1985EXPORT_SYMBOL(pagecache_write_end);
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987ssize_t
1988generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1989 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1990 size_t count, size_t ocount)
1991{
1992 struct file *file = iocb->ki_filp;
1993 struct address_space *mapping = file->f_mapping;
1994 struct inode *inode = mapping->host;
1995 ssize_t written;
1996
1997 if (count != ocount)
1998 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1999
2000 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2001 if (written > 0) {
2002 loff_t end = pos + written;
2003 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2004 i_size_write(inode, end);
2005 mark_inode_dirty(inode);
2006 }
2007 *ppos = end;
2008 }
2009
2010 /*
2011 * Sync the fs metadata but not the minor inode changes and
2012 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002013 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002014 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 */
Zach Brown8459d862006-12-10 02:21:05 -08002016 if ((written >= 0 || written == -EIOCBQUEUED) &&
2017 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002018 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2019 if (err < 0)
2020 written = err;
2021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 return written;
2023}
2024EXPORT_SYMBOL(generic_file_direct_write);
2025
Nick Piggineb2be182007-10-16 01:24:57 -07002026/*
2027 * Find or create a page at the given pagecache position. Return the locked
2028 * page. This function is specifically for buffered writes.
2029 */
Nick Pigginafddba42007-10-16 01:25:01 -07002030struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index)
Nick Piggineb2be182007-10-16 01:24:57 -07002031{
2032 int status;
2033 struct page *page;
2034repeat:
2035 page = find_lock_page(mapping, index);
2036 if (likely(page))
2037 return page;
2038
2039 page = page_cache_alloc(mapping);
2040 if (!page)
2041 return NULL;
2042 status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
2043 if (unlikely(status)) {
2044 page_cache_release(page);
2045 if (status == -EEXIST)
2046 goto repeat;
2047 return NULL;
2048 }
2049 return page;
2050}
Nick Pigginafddba42007-10-16 01:25:01 -07002051EXPORT_SYMBOL(__grab_cache_page);
Nick Piggineb2be182007-10-16 01:24:57 -07002052
Nick Pigginafddba42007-10-16 01:25:01 -07002053static ssize_t generic_perform_write_2copy(struct file *file,
2054 struct iov_iter *i, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055{
Andrew Mortonae374612007-10-16 01:24:55 -07002056 struct address_space *mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002057 const struct address_space_operations *a_ops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002058 struct inode *inode = mapping->host;
2059 long status = 0;
2060 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
2062 do {
Nick Piggin08291422007-10-16 01:24:59 -07002063 struct page *src_page;
Nick Piggineb2be182007-10-16 01:24:57 -07002064 struct page *page;
Andrew Mortonae374612007-10-16 01:24:55 -07002065 pgoff_t index; /* Pagecache index for current page */
2066 unsigned long offset; /* Offset into pagecache page */
Nick Piggin08291422007-10-16 01:24:59 -07002067 unsigned long bytes; /* Bytes to write to page */
Andrew Mortonae374612007-10-16 01:24:55 -07002068 size_t copied; /* Bytes copied from user */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Andrew Mortonae374612007-10-16 01:24:55 -07002070 offset = (pos & (PAGE_CACHE_SIZE - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002072 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002073 iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Nick Piggin08291422007-10-16 01:24:59 -07002075 /*
2076 * a non-NULL src_page indicates that we're doing the
2077 * copy via get_user_pages and kmap.
2078 */
2079 src_page = NULL;
Andrew Mortonae374612007-10-16 01:24:55 -07002080
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002081 /*
2082 * Bring in the user page that we will copy from _first_.
2083 * Otherwise there's a nasty deadlock on copying from the
2084 * same page as we're writing to, without it being marked
2085 * up-to-date.
Nick Piggin08291422007-10-16 01:24:59 -07002086 *
2087 * Not only is this an optimisation, but it is also required
2088 * to check that the address is actually valid, when atomic
2089 * usercopies are used, below.
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002090 */
Nick Pigginafddba42007-10-16 01:25:01 -07002091 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
Nick Piggin08291422007-10-16 01:24:59 -07002092 status = -EFAULT;
2093 break;
2094 }
Nick Piggineb2be182007-10-16 01:24:57 -07002095
2096 page = __grab_cache_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 if (!page) {
2098 status = -ENOMEM;
2099 break;
2100 }
2101
Nick Piggin08291422007-10-16 01:24:59 -07002102 /*
2103 * non-uptodate pages cannot cope with short copies, and we
2104 * cannot take a pagefault with the destination page locked.
2105 * So pin the source page to copy it.
2106 */
Nick Piggin674b8922007-10-16 01:25:03 -07002107 if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) {
Nick Piggin08291422007-10-16 01:24:59 -07002108 unlock_page(page);
2109
2110 src_page = alloc_page(GFP_KERNEL);
2111 if (!src_page) {
2112 page_cache_release(page);
2113 status = -ENOMEM;
2114 break;
2115 }
2116
2117 /*
2118 * Cannot get_user_pages with a page locked for the
2119 * same reason as we can't take a page fault with a
2120 * page locked (as explained below).
2121 */
Nick Pigginafddba42007-10-16 01:25:01 -07002122 copied = iov_iter_copy_from_user(src_page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002123 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002124 if (unlikely(copied == 0)) {
2125 status = -EFAULT;
2126 page_cache_release(page);
2127 page_cache_release(src_page);
2128 break;
2129 }
2130 bytes = copied;
2131
2132 lock_page(page);
2133 /*
2134 * Can't handle the page going uptodate here, because
2135 * that means we would use non-atomic usercopies, which
2136 * zero out the tail of the page, which can cause
2137 * zeroes to become transiently visible. We could just
2138 * use a non-zeroing copy, but the APIs aren't too
2139 * consistent.
2140 */
2141 if (unlikely(!page->mapping || PageUptodate(page))) {
2142 unlock_page(page);
2143 page_cache_release(page);
2144 page_cache_release(src_page);
2145 continue;
2146 }
Nick Piggin08291422007-10-16 01:24:59 -07002147 }
2148
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 status = a_ops->prepare_write(file, page, offset, offset+bytes);
Nick Piggin64649a52007-10-16 01:24:56 -07002150 if (unlikely(status))
2151 goto fs_write_aop_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08002152
Nick Piggin08291422007-10-16 01:24:59 -07002153 if (!src_page) {
2154 /*
2155 * Must not enter the pagefault handler here, because
2156 * we hold the page lock, so we might recursively
2157 * deadlock on the same lock, or get an ABBA deadlock
2158 * against a different lock, or against the mmap_sem
2159 * (which nests outside the page lock). So increment
2160 * preempt count, and use _atomic usercopies.
2161 *
2162 * The page is uptodate so we are OK to encounter a
2163 * short copy: if unmodified parts of the page are
2164 * marked dirty and written out to disk, it doesn't
2165 * really matter.
2166 */
2167 pagefault_disable();
Nick Pigginafddba42007-10-16 01:25:01 -07002168 copied = iov_iter_copy_from_user_atomic(page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002169 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002170 pagefault_enable();
2171 } else {
2172 void *src, *dst;
2173 src = kmap_atomic(src_page, KM_USER0);
2174 dst = kmap_atomic(page, KM_USER1);
2175 memcpy(dst + offset, src + offset, bytes);
2176 kunmap_atomic(dst, KM_USER1);
2177 kunmap_atomic(src, KM_USER0);
2178 copied = bytes;
2179 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 flush_dcache_page(page);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002181
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 status = a_ops->commit_write(file, page, offset, offset+bytes);
Nick Piggin55144762007-10-16 01:25:26 -07002183 if (unlikely(status < 0))
Nick Piggin64649a52007-10-16 01:24:56 -07002184 goto fs_write_aop_error;
Nick Piggin64649a52007-10-16 01:24:56 -07002185 if (unlikely(status > 0)) /* filesystem did partial write */
Nick Piggin08291422007-10-16 01:24:59 -07002186 copied = min_t(size_t, copied, status);
2187
2188 unlock_page(page);
2189 mark_page_accessed(page);
2190 page_cache_release(page);
2191 if (src_page)
2192 page_cache_release(src_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Nick Pigginafddba42007-10-16 01:25:01 -07002194 iov_iter_advance(i, copied);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002195 pos += copied;
Nick Pigginafddba42007-10-16 01:25:01 -07002196 written += copied;
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002197
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 balance_dirty_pages_ratelimited(mapping);
2199 cond_resched();
Nick Piggin64649a52007-10-16 01:24:56 -07002200 continue;
2201
2202fs_write_aop_error:
Nick Piggin55144762007-10-16 01:25:26 -07002203 unlock_page(page);
Nick Piggin64649a52007-10-16 01:24:56 -07002204 page_cache_release(page);
Nick Piggin08291422007-10-16 01:24:59 -07002205 if (src_page)
2206 page_cache_release(src_page);
Nick Piggin64649a52007-10-16 01:24:56 -07002207
2208 /*
2209 * prepare_write() may have instantiated a few blocks
2210 * outside i_size. Trim these off again. Don't need
2211 * i_size_read because we hold i_mutex.
2212 */
2213 if (pos + bytes > inode->i_size)
2214 vmtruncate(inode, inode->i_size);
Nick Piggin55144762007-10-16 01:25:26 -07002215 break;
Nick Pigginafddba42007-10-16 01:25:01 -07002216 } while (iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
Nick Pigginafddba42007-10-16 01:25:01 -07002218 return written ? written : status;
2219}
2220
2221static ssize_t generic_perform_write(struct file *file,
2222 struct iov_iter *i, loff_t pos)
2223{
2224 struct address_space *mapping = file->f_mapping;
2225 const struct address_space_operations *a_ops = mapping->a_ops;
2226 long status = 0;
2227 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002228 unsigned int flags = 0;
2229
2230 /*
2231 * Copies from kernel address space cannot fail (NFSD is a big user).
2232 */
2233 if (segment_eq(get_fs(), KERNEL_DS))
2234 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002235
2236 do {
2237 struct page *page;
2238 pgoff_t index; /* Pagecache index for current page */
2239 unsigned long offset; /* Offset into pagecache page */
2240 unsigned long bytes; /* Bytes to write to page */
2241 size_t copied; /* Bytes copied from user */
2242 void *fsdata;
2243
2244 offset = (pos & (PAGE_CACHE_SIZE - 1));
2245 index = pos >> PAGE_CACHE_SHIFT;
2246 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2247 iov_iter_count(i));
2248
2249again:
2250
2251 /*
2252 * Bring in the user page that we will copy from _first_.
2253 * Otherwise there's a nasty deadlock on copying from the
2254 * same page as we're writing to, without it being marked
2255 * up-to-date.
2256 *
2257 * Not only is this an optimisation, but it is also required
2258 * to check that the address is actually valid, when atomic
2259 * usercopies are used, below.
2260 */
2261 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2262 status = -EFAULT;
2263 break;
2264 }
2265
Nick Piggin674b8922007-10-16 01:25:03 -07002266 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002267 &page, &fsdata);
2268 if (unlikely(status))
2269 break;
2270
2271 pagefault_disable();
2272 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2273 pagefault_enable();
2274 flush_dcache_page(page);
2275
2276 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2277 page, fsdata);
2278 if (unlikely(status < 0))
2279 break;
2280 copied = status;
2281
2282 cond_resched();
2283
Nick Piggin124d3b72008-02-02 15:01:17 +01002284 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002285 if (unlikely(copied == 0)) {
2286 /*
2287 * If we were unable to copy any data at all, we must
2288 * fall back to a single segment length write.
2289 *
2290 * If we didn't fallback here, we could livelock
2291 * because not all segments in the iov can be copied at
2292 * once without a pagefault.
2293 */
2294 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2295 iov_iter_single_seg_count(i));
2296 goto again;
2297 }
Nick Pigginafddba42007-10-16 01:25:01 -07002298 pos += copied;
2299 written += copied;
2300
2301 balance_dirty_pages_ratelimited(mapping);
2302
2303 } while (iov_iter_count(i));
2304
2305 return written ? written : status;
2306}
2307
2308ssize_t
2309generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2310 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2311 size_t count, ssize_t written)
2312{
2313 struct file *file = iocb->ki_filp;
2314 struct address_space *mapping = file->f_mapping;
2315 const struct address_space_operations *a_ops = mapping->a_ops;
2316 struct inode *inode = mapping->host;
2317 ssize_t status;
2318 struct iov_iter i;
2319
2320 iov_iter_init(&i, iov, nr_segs, count, written);
2321 if (a_ops->write_begin)
2322 status = generic_perform_write(file, &i, pos);
2323 else
2324 status = generic_perform_write_2copy(file, &i, pos);
2325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002327 written += status;
2328 *ppos = pos + status;
2329
2330 /*
2331 * For now, when the user asks for O_SYNC, we'll actually give
2332 * O_DSYNC
2333 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2335 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2336 status = generic_osync_inode(inode, mapping,
2337 OSYNC_METADATA|OSYNC_DATA);
2338 }
2339 }
2340
2341 /*
2342 * If we get here for O_DIRECT writes then we must have fallen through
2343 * to buffered writes (block instantiation inside i_size). So we sync
2344 * the file data here, to try to honour O_DIRECT expectations.
2345 */
2346 if (unlikely(file->f_flags & O_DIRECT) && written)
2347 status = filemap_write_and_wait(mapping);
2348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 return written ? written : status;
2350}
2351EXPORT_SYMBOL(generic_file_buffered_write);
2352
Adrian Bunk5ce78522005-09-10 00:26:28 -07002353static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2355 unsigned long nr_segs, loff_t *ppos)
2356{
2357 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002358 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 size_t ocount; /* original count */
2360 size_t count; /* after file limit checks */
2361 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 loff_t pos;
2363 ssize_t written;
2364 ssize_t err;
2365
2366 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002367 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2368 if (err)
2369 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370
2371 count = ocount;
2372 pos = *ppos;
2373
2374 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2375
2376 /* We can write back this queue in page reclaim */
2377 current->backing_dev_info = mapping->backing_dev_info;
2378 written = 0;
2379
2380 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2381 if (err)
2382 goto out;
2383
2384 if (count == 0)
2385 goto out;
2386
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002387 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 if (err)
2389 goto out;
2390
Christoph Hellwig870f4812006-01-09 20:52:01 -08002391 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
2393 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2394 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002395 loff_t endbyte;
2396 ssize_t written_buffered;
2397
2398 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2399 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 if (written < 0 || written == count)
2401 goto out;
2402 /*
2403 * direct-io write to a hole: fall through to buffered I/O
2404 * for completing the rest of the request.
2405 */
2406 pos += written;
2407 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002408 written_buffered = generic_file_buffered_write(iocb, iov,
2409 nr_segs, pos, ppos, count,
2410 written);
2411 /*
2412 * If generic_file_buffered_write() retuned a synchronous error
2413 * then we want to return the number of bytes which were
2414 * direct-written, or the error code if that was zero. Note
2415 * that this differs from normal direct-io semantics, which
2416 * will return -EFOO even if some bytes were written.
2417 */
2418 if (written_buffered < 0) {
2419 err = written_buffered;
2420 goto out;
2421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002423 /*
2424 * We need to ensure that the page cache pages are written to
2425 * disk and invalidated to preserve the expected O_DIRECT
2426 * semantics.
2427 */
2428 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002429 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2430 SYNC_FILE_RANGE_WAIT_BEFORE|
2431 SYNC_FILE_RANGE_WRITE|
2432 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002433 if (err == 0) {
2434 written = written_buffered;
2435 invalidate_mapping_pages(mapping,
2436 pos >> PAGE_CACHE_SHIFT,
2437 endbyte >> PAGE_CACHE_SHIFT);
2438 } else {
2439 /*
2440 * We don't know how much we wrote, so just return
2441 * the number of bytes which were direct-written
2442 */
2443 }
2444 } else {
2445 written = generic_file_buffered_write(iocb, iov, nr_segs,
2446 pos, ppos, count, written);
2447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448out:
2449 current->backing_dev_info = NULL;
2450 return written ? written : err;
2451}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
Badari Pulavarty027445c2006-09-30 23:28:46 -07002453ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2454 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455{
2456 struct file *file = iocb->ki_filp;
2457 struct address_space *mapping = file->f_mapping;
2458 struct inode *inode = mapping->host;
2459 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
Badari Pulavarty027445c2006-09-30 23:28:46 -07002461 BUG_ON(iocb->ki_pos != pos);
2462
2463 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2464 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
2466 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002467 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
2469 err = sync_page_range_nolock(inode, mapping, pos, ret);
2470 if (err < 0)
2471 ret = err;
2472 }
2473 return ret;
2474}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002475EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Badari Pulavarty027445c2006-09-30 23:28:46 -07002477ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2478 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
2480 struct file *file = iocb->ki_filp;
2481 struct address_space *mapping = file->f_mapping;
2482 struct inode *inode = mapping->host;
2483 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
2485 BUG_ON(iocb->ki_pos != pos);
2486
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002487 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002488 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2489 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002490 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
2492 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2493 ssize_t err;
2494
2495 err = sync_page_range(inode, mapping, pos, ret);
2496 if (err < 0)
2497 ret = err;
2498 }
2499 return ret;
2500}
2501EXPORT_SYMBOL(generic_file_aio_write);
2502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002504 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 * went wrong during pagecache shootdown.
2506 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002507static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2509 loff_t offset, unsigned long nr_segs)
2510{
2511 struct file *file = iocb->ki_filp;
2512 struct address_space *mapping = file->f_mapping;
2513 ssize_t retval;
Zach Brown65b82912007-03-16 13:38:11 -08002514 size_t write_len;
2515 pgoff_t end = 0; /* silence gcc */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
2517 /*
2518 * If it's a write, unmap all mmappings of the file up-front. This
2519 * will cause any pte dirty bits to be propagated into the pageframes
2520 * for the subsequent filemap_write_and_wait().
2521 */
2522 if (rw == WRITE) {
2523 write_len = iov_length(iov, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002524 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 if (mapping_mapped(mapping))
2526 unmap_mapping_range(mapping, offset, write_len, 0);
2527 }
2528
2529 retval = filemap_write_and_wait(mapping);
Zach Brown65b82912007-03-16 13:38:11 -08002530 if (retval)
2531 goto out;
2532
2533 /*
2534 * After a write we want buffered reads to be sure to go to disk to get
2535 * the new data. We invalidate clean cached page from the region we're
2536 * about to write. We do this *before* the write so that we can return
2537 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2538 */
2539 if (rw == WRITE && mapping->nrpages) {
2540 retval = invalidate_inode_pages2_range(mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002542 if (retval)
2543 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 }
Zach Brown65b82912007-03-16 13:38:11 -08002545
2546 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002547
2548 /*
2549 * Finally, try again to invalidate clean pages which might have been
Zach Brownbdb76ef2007-10-30 11:45:46 -07002550 * cached by non-direct readahead, or faulted in by get_user_pages()
2551 * if the source of the write was an mmap'ed region of the file
2552 * we're writing. Either one is a pretty crazy thing to do,
2553 * so we don't support it 100%. If this invalidation
2554 * fails, tough, the write still worked...
Zach Brown65b82912007-03-16 13:38:11 -08002555 */
2556 if (rw == WRITE && mapping->nrpages) {
Zach Brownbdb76ef2007-10-30 11:45:46 -07002557 invalidate_inode_pages2_range(mapping, offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002558 }
2559out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 return retval;
2561}
David Howellscf9a2ae2006-08-29 19:05:54 +01002562
2563/**
2564 * try_to_release_page() - release old fs-specific metadata on a page
2565 *
2566 * @page: the page which the kernel is trying to free
2567 * @gfp_mask: memory allocation flags (and I/O mode)
2568 *
2569 * The address_space is to try to release any data against the page
2570 * (presumably at page->private). If the release was successful, return `1'.
2571 * Otherwise return zero.
2572 *
2573 * The @gfp_mask argument specifies whether I/O may be performed to release
2574 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2575 *
2576 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2577 */
2578int try_to_release_page(struct page *page, gfp_t gfp_mask)
2579{
2580 struct address_space * const mapping = page->mapping;
2581
2582 BUG_ON(!PageLocked(page));
2583 if (PageWriteback(page))
2584 return 0;
2585
2586 if (mapping && mapping->a_ops->releasepage)
2587 return mapping->a_ops->releasepage(page, gfp_mask);
2588 return try_to_free_buffers(page);
2589}
2590
2591EXPORT_SYMBOL(try_to_release_page);