blob: 2e2d38ebda4bbf0b323bf0011634f87ac0b67402 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
13#include <linux/slab.h>
14#include <linux/compiler.h>
15#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kernel_stat.h>
20#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036#include <linux/mm_inline.h> /* for page_is_file_cache() */
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 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/*
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
49 * though.
50 *
51 * Shared mappings now work. 15.8.1995 Bruno.
52 *
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 *
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
57 */
58
59/*
60 * Lock ordering:
61 *
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070064 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080067 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * ->i_mmap_lock (truncate->unmap_mapping_range)
69 *
70 * ->mmap_sem
71 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070072 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
74 *
75 * ->mmap_sem
76 * ->lock_page (access_process_vm)
77 *
Nick Piggin82591e62006-10-19 23:29:10 -070078 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080081 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * ->i_alloc_sem (various)
83 *
84 * ->inode_lock
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
87 *
88 * ->i_mmap_lock
89 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080099 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 *
106 * ->task->proc_lock
107 * ->dcache_lock (proc_pid_lookup)
108 */
109
110/*
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700113 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
115void __remove_from_page_cache(struct page *page)
116{
117 struct address_space *mapping = page->mapping;
118
119 radix_tree_delete(&mapping->page_tree, page->index);
120 page->mapping = NULL;
121 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700122 __dec_zone_page_state(page, NR_FILE_PAGES);
Nick Piggin45426812007-07-15 23:38:12 -0700123 BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -0700124 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds3a692792007-12-19 14:05:13 -0800125
126 /*
127 * Some filesystems seem to re-dirty the page even after
128 * the VM has canceled the dirty bit (eg ext3 journaling).
129 *
130 * Fix it up by doing a final dirty accounting check after
131 * having removed the page entirely.
132 */
133 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
134 dec_zone_page_state(page, NR_FILE_DIRTY);
135 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
139void remove_from_page_cache(struct page *page)
140{
141 struct address_space *mapping = page->mapping;
142
Matt Mackallcd7619d2005-05-01 08:59:01 -0700143 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Nick Piggin19fd6232008-07-25 19:45:32 -0700145 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700147 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
150static int sync_page(void *word)
151{
152 struct address_space *mapping;
153 struct page *page;
154
Andi Kleen07808b72005-11-05 17:25:53 +0100155 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700158 * page_mapping() is being called without PG_locked held.
159 * Some knowledge of the state and use of the page is used to
160 * reduce the requirements down to a memory barrier.
161 * The danger here is of a stale page_mapping() return value
162 * indicating a struct address_space different from the one it's
163 * associated with when it is associated with one.
164 * After smp_mb(), it's either the correct page_mapping() for
165 * the page, or an old page_mapping() and the page's own
166 * page_mapping() has gone NULL.
167 * The ->sync_page() address_space operation must tolerate
168 * page_mapping() going NULL. By an amazing coincidence,
169 * this comes about because none of the users of the page
170 * in the ->sync_page() methods make essential use of the
171 * page_mapping(), merely passing the page down to the backing
172 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700173 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700174 * of interest. When page_mapping() does go NULL, the entire
175 * call stack gracefully ignores the page and returns.
176 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 */
178 smp_mb();
179 mapping = page_mapping(page);
180 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
181 mapping->a_ops->sync_page(page);
182 io_schedule();
183 return 0;
184}
185
Matthew Wilcox2687a352007-12-06 11:18:49 -0500186static int sync_page_killable(void *word)
187{
188 sync_page(word);
189 return fatal_signal_pending(current) ? -EINTR : 0;
190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700194 * @mapping: address space structure to write
195 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800196 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700197 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700199 * Start writeback against all of a mapping's dirty pages that lie
200 * within the byte offsets <start, end> inclusive.
201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700203 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * these two operations is that if a dirty page/buffer is encountered, it must
205 * be waited upon, and not just skipped over.
206 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800207int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
208 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 int ret;
211 struct writeback_control wbc = {
212 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800213 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700214 .range_start = start,
215 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 };
217
218 if (!mapping_cap_writeback_dirty(mapping))
219 return 0;
220
221 ret = do_writepages(mapping, &wbc);
222 return ret;
223}
224
225static inline int __filemap_fdatawrite(struct address_space *mapping,
226 int sync_mode)
227{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229}
230
231int filemap_fdatawrite(struct address_space *mapping)
232{
233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
234}
235EXPORT_SYMBOL(filemap_fdatawrite);
236
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400237int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800238 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
241}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400242EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Randy Dunlap485bb992006-06-23 02:03:49 -0700244/**
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
247 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
250 */
251int filemap_flush(struct address_space *mapping)
252{
253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
254}
255EXPORT_SYMBOL(filemap_flush);
256
Randy Dunlap485bb992006-06-23 02:03:49 -0700257/**
258 * wait_on_page_writeback_range - wait for writeback to complete
259 * @mapping: target address_space
260 * @start: beginning page index
261 * @end: ending page index
262 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 * Wait for writeback to complete against pages indexed by start->end
264 * inclusive
265 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800266int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 pgoff_t start, pgoff_t end)
268{
269 struct pagevec pvec;
270 int nr_pages;
271 int ret = 0;
272 pgoff_t index;
273
274 if (end < start)
275 return 0;
276
277 pagevec_init(&pvec, 0);
278 index = start;
279 while ((index <= end) &&
280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
281 PAGECACHE_TAG_WRITEBACK,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
283 unsigned i;
284
285 for (i = 0; i < nr_pages; i++) {
286 struct page *page = pvec.pages[i];
287
288 /* until radix tree lookup accepts end_index */
289 if (page->index > end)
290 continue;
291
292 wait_on_page_writeback(page);
293 if (PageError(page))
294 ret = -EIO;
295 }
296 pagevec_release(&pvec);
297 cond_resched();
298 }
299
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
302 ret = -ENOSPC;
303 if (test_and_clear_bit(AS_EIO, &mapping->flags))
304 ret = -EIO;
305
306 return ret;
307}
308
Randy Dunlap485bb992006-06-23 02:03:49 -0700309/**
310 * sync_page_range - write and wait on all pages in the passed range
311 * @inode: target inode
312 * @mapping: target address_space
313 * @pos: beginning offset in pages to write
314 * @count: number of bytes to write
315 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * Write and wait upon all the pages in the passed range. This is a "data
317 * integrity" operation. It waits upon in-flight writeout before starting and
318 * waiting upon new writeout. If there was an IO error, return it.
319 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800320 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * it is otherwise livelockable.
322 */
323int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800324 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
327 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
328 int ret;
329
330 if (!mapping_cap_writeback_dirty(mapping) || !count)
331 return 0;
332 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
333 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800334 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800336 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
338 if (ret == 0)
339 ret = wait_on_page_writeback_range(mapping, start, end);
340 return ret;
341}
342EXPORT_SYMBOL(sync_page_range);
343
Randy Dunlap485bb992006-06-23 02:03:49 -0700344/**
Randy Dunlap76824862008-03-19 17:00:40 -0700345 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
Randy Dunlap485bb992006-06-23 02:03:49 -0700346 * @inode: target inode
347 * @mapping: target address_space
348 * @pos: beginning offset in pages to write
349 * @count: number of bytes to write
350 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800351 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 * as it forces O_SYNC writers to different parts of the same file
353 * to be serialised right until io completion.
354 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800355int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
356 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
358 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
359 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
360 int ret;
361
362 if (!mapping_cap_writeback_dirty(mapping) || !count)
363 return 0;
364 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
365 if (ret == 0)
366 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
367 if (ret == 0)
368 ret = wait_on_page_writeback_range(mapping, start, end);
369 return ret;
370}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800371EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700374 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700376 *
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
380int filemap_fdatawait(struct address_space *mapping)
381{
382 loff_t i_size = i_size_read(mapping->host);
383
384 if (i_size == 0)
385 return 0;
386
387 return wait_on_page_writeback_range(mapping, 0,
388 (i_size - 1) >> PAGE_CACHE_SHIFT);
389}
390EXPORT_SYMBOL(filemap_fdatawait);
391
392int filemap_write_and_wait(struct address_space *mapping)
393{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800394 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800397 err = filemap_fdatawrite(mapping);
398 /*
399 * Even if the above returned error, the pages may be
400 * written partially (e.g. -ENOSPC), so we wait for it.
401 * But the -EIO is special case, it may indicate the worst
402 * thing (e.g. bug) happened, so we avoid waiting for it.
403 */
404 if (err != -EIO) {
405 int err2 = filemap_fdatawait(mapping);
406 if (!err)
407 err = err2;
408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800410 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800412EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Randy Dunlap485bb992006-06-23 02:03:49 -0700414/**
415 * filemap_write_and_wait_range - write out & wait on a file range
416 * @mapping: the address_space for the pages
417 * @lstart: offset in bytes where the range starts
418 * @lend: offset in bytes where the range ends (inclusive)
419 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800420 * Write out and wait upon file offsets lstart->lend, inclusive.
421 *
422 * Note that `lend' is inclusive (describes the last byte to be written) so
423 * that this function can be used to write to the very end-of-file (end = -1).
424 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425int filemap_write_and_wait_range(struct address_space *mapping,
426 loff_t lstart, loff_t lend)
427{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800428 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800431 err = __filemap_fdatawrite_range(mapping, lstart, lend,
432 WB_SYNC_ALL);
433 /* See comment of filemap_write_and_wait() */
434 if (err != -EIO) {
435 int err2 = wait_on_page_writeback_range(mapping,
436 lstart >> PAGE_CACHE_SHIFT,
437 lend >> PAGE_CACHE_SHIFT);
438 if (!err)
439 err = err2;
440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800442 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
Randy Dunlap485bb992006-06-23 02:03:49 -0700445/**
Nick Piggine2867812008-07-25 19:45:30 -0700446 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700447 * @page: page to add
448 * @mapping: the page's address_space
449 * @offset: page index
450 * @gfp_mask: page allocation mode
451 *
Nick Piggine2867812008-07-25 19:45:30 -0700452 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * This function does not add the page to the LRU. The caller must do that.
454 */
Nick Piggine2867812008-07-25 19:45:30 -0700455int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400456 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Nick Piggine2867812008-07-25 19:45:30 -0700458 int error;
459
460 VM_BUG_ON(!PageLocked(page));
461
462 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800463 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800464 if (error)
465 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Balbir Singh35c754d2008-02-07 00:14:05 -0800467 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700469 page_cache_get(page);
470 page->mapping = mapping;
471 page->index = offset;
472
Nick Piggin19fd6232008-07-25 19:45:32 -0700473 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700475 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700477 __inc_zone_page_state(page, NR_FILE_PAGES);
Nick Piggine2867812008-07-25 19:45:30 -0700478 } else {
479 page->mapping = NULL;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700480 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700481 page_cache_release(page);
482 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800483
Nick Piggin19fd6232008-07-25 19:45:32 -0700484 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800486 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700487 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800488out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 return error;
490}
Nick Piggine2867812008-07-25 19:45:30 -0700491EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400494 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700496 int ret;
497
498 /*
499 * Splice_read and readahead add shmem/tmpfs pages into the page cache
500 * before shmem_readpage has a chance to mark them as SwapBacked: they
501 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
502 * (called in add_to_page_cache) needs to know where they're going too.
503 */
504 if (mapping_cap_swap_backed(mapping))
505 SetPageSwapBacked(page);
506
507 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
508 if (ret == 0) {
509 if (page_is_file_cache(page))
510 lru_cache_add_file(page);
511 else
512 lru_cache_add_active_anon(page);
513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return ret;
515}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300516EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Paul Jackson44110fe2006-03-24 03:16:04 -0800518#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700519struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800520{
521 if (cpuset_do_page_mem_spread()) {
522 int n = cpuset_mem_spread_node();
Nick Piggin2ae88142006-10-28 10:38:23 -0700523 return alloc_pages_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800524 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700525 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800526}
Nick Piggin2ae88142006-10-28 10:38:23 -0700527EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800528#endif
529
Nick Piggindb376482006-09-25 23:31:24 -0700530static int __sleep_on_page_lock(void *word)
531{
532 io_schedule();
533 return 0;
534}
535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536/*
537 * In order to wait for pages to become available there must be
538 * waitqueues associated with pages. By using a hash table of
539 * waitqueues where the bucket discipline is to maintain all
540 * waiters on the same queue and wake all when any of the pages
541 * become available, and for the woken contexts to check to be
542 * sure the appropriate page became available, this saves space
543 * at a cost of "thundering herd" phenomena during rare hash
544 * collisions.
545 */
546static wait_queue_head_t *page_waitqueue(struct page *page)
547{
548 const struct zone *zone = page_zone(page);
549
550 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
551}
552
553static inline void wake_up_page(struct page *page, int bit)
554{
555 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
556}
557
Harvey Harrison920c7a52008-02-04 22:29:26 -0800558void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
560 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
561
562 if (test_bit(bit_nr, &page->flags))
563 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
564 TASK_UNINTERRUPTIBLE);
565}
566EXPORT_SYMBOL(wait_on_page_bit);
567
568/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100569 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
570 * @page - Page defining the wait queue of interest
571 * @waiter - Waiter to add to the queue
572 *
573 * Add an arbitrary @waiter to the wait queue for the nominated @page.
574 */
575void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
576{
577 wait_queue_head_t *q = page_waitqueue(page);
578 unsigned long flags;
579
580 spin_lock_irqsave(&q->lock, flags);
581 __add_wait_queue(q, waiter);
582 spin_unlock_irqrestore(&q->lock, flags);
583}
584EXPORT_SYMBOL_GPL(add_page_wait_queue);
585
586/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700587 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 * @page: the page
589 *
590 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
591 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
592 * mechananism between PageLocked pages and PageWriteback pages is shared.
593 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
594 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700595 * The mb is necessary to enforce ordering between the clear_bit and the read
596 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800598void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
Nick Piggin8413ac92008-10-18 20:26:59 -0700600 VM_BUG_ON(!PageLocked(page));
601 clear_bit_unlock(PG_locked, &page->flags);
602 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 wake_up_page(page, PG_locked);
604}
605EXPORT_SYMBOL(unlock_page);
606
Randy Dunlap485bb992006-06-23 02:03:49 -0700607/**
608 * end_page_writeback - end writeback against a page
609 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 */
611void end_page_writeback(struct page *page)
612{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700613 if (TestClearPageReclaim(page))
614 rotate_reclaimable_page(page);
615
616 if (!test_clear_page_writeback(page))
617 BUG();
618
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 smp_mb__after_clear_bit();
620 wake_up_page(page, PG_writeback);
621}
622EXPORT_SYMBOL(end_page_writeback);
623
Randy Dunlap485bb992006-06-23 02:03:49 -0700624/**
625 * __lock_page - get a lock on the page, assuming we need to sleep to get it
626 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700628 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
630 * chances are that on the second loop, the block layer's plug list is empty,
631 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
632 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800633void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
635 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
636
637 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
638 TASK_UNINTERRUPTIBLE);
639}
640EXPORT_SYMBOL(__lock_page);
641
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800642int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500643{
644 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
645
646 return __wait_on_bit_lock(page_waitqueue(page), &wait,
647 sync_page_killable, TASK_KILLABLE);
648}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300649EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500650
Randy Dunlap76824862008-03-19 17:00:40 -0700651/**
652 * __lock_page_nosync - get a lock on the page, without calling sync_page()
653 * @page: the page to lock
654 *
Nick Piggindb376482006-09-25 23:31:24 -0700655 * Variant of lock_page that does not require the caller to hold a reference
656 * on the page's mapping.
657 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800658void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700659{
660 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
661 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
662 TASK_UNINTERRUPTIBLE);
663}
664
Randy Dunlap485bb992006-06-23 02:03:49 -0700665/**
666 * find_get_page - find and get a page reference
667 * @mapping: the address_space to search
668 * @offset: the page index
669 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700670 * Is there a pagecache struct page at the given (mapping, offset) tuple?
671 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 */
Nick Piggina60637c2008-07-25 19:45:31 -0700673struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
Nick Piggina60637c2008-07-25 19:45:31 -0700675 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 struct page *page;
677
Nick Piggina60637c2008-07-25 19:45:31 -0700678 rcu_read_lock();
679repeat:
680 page = NULL;
681 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
682 if (pagep) {
683 page = radix_tree_deref_slot(pagep);
684 if (unlikely(!page || page == RADIX_TREE_RETRY))
685 goto repeat;
686
687 if (!page_cache_get_speculative(page))
688 goto repeat;
689
690 /*
691 * Has the page moved?
692 * This is part of the lockless pagecache protocol. See
693 * include/linux/pagemap.h for details.
694 */
695 if (unlikely(page != *pagep)) {
696 page_cache_release(page);
697 goto repeat;
698 }
699 }
700 rcu_read_unlock();
701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 return page;
703}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704EXPORT_SYMBOL(find_get_page);
705
Randy Dunlap485bb992006-06-23 02:03:49 -0700706/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700708 * @mapping: the address_space to search
709 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 *
711 * Locates the desired pagecache page, locks it, increments its reference
712 * count and returns its address.
713 *
714 * Returns zero if the page was not present. find_lock_page() may sleep.
715 */
Nick Piggina60637c2008-07-25 19:45:31 -0700716struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
718 struct page *page;
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700721 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700723 lock_page(page);
724 /* Has the page been truncated? */
725 if (unlikely(page->mapping != mapping)) {
726 unlock_page(page);
727 page_cache_release(page);
728 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 }
Nick Piggina60637c2008-07-25 19:45:31 -0700730 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 return page;
733}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734EXPORT_SYMBOL(find_lock_page);
735
736/**
737 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700738 * @mapping: the page's address_space
739 * @index: the page's index into the mapping
740 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 *
742 * Locates a page in the pagecache. If the page is not present, a new page
743 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
744 * LRU list. The returned page is locked and has its reference count
745 * incremented.
746 *
747 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
748 * allocation!
749 *
750 * find_or_create_page() returns the desired page's address, or zero on
751 * memory exhaustion.
752 */
753struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700754 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
Nick Piggineb2be182007-10-16 01:24:57 -0700756 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 int err;
758repeat:
759 page = find_lock_page(mapping, index);
760 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700761 page = __page_cache_alloc(gfp_mask);
762 if (!page)
763 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800764 /*
765 * We want a regular kernel memory (not highmem or DMA etc)
766 * allocation for the radix tree nodes, but we need to honour
767 * the context-specific requirements the caller has asked for.
768 * GFP_RECLAIM_MASK collects those requirements.
769 */
770 err = add_to_page_cache_lru(page, mapping, index,
771 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700772 if (unlikely(err)) {
773 page_cache_release(page);
774 page = NULL;
775 if (err == -EEXIST)
776 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return page;
780}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781EXPORT_SYMBOL(find_or_create_page);
782
783/**
784 * find_get_pages - gang pagecache lookup
785 * @mapping: The address_space to search
786 * @start: The starting page index
787 * @nr_pages: The maximum number of pages
788 * @pages: Where the resulting pages are placed
789 *
790 * find_get_pages() will search for and return a group of up to
791 * @nr_pages pages in the mapping. The pages are placed at @pages.
792 * find_get_pages() takes a reference against the returned pages.
793 *
794 * The search returns a group of mapping-contiguous pages with ascending
795 * indexes. There may be holes in the indices due to not-present pages.
796 *
797 * find_get_pages() returns the number of pages which were found.
798 */
799unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
800 unsigned int nr_pages, struct page **pages)
801{
802 unsigned int i;
803 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700804 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Nick Piggina60637c2008-07-25 19:45:31 -0700806 rcu_read_lock();
807restart:
808 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
809 (void ***)pages, start, nr_pages);
810 ret = 0;
811 for (i = 0; i < nr_found; i++) {
812 struct page *page;
813repeat:
814 page = radix_tree_deref_slot((void **)pages[i]);
815 if (unlikely(!page))
816 continue;
817 /*
818 * this can only trigger if nr_found == 1, making livelock
819 * a non issue.
820 */
821 if (unlikely(page == RADIX_TREE_RETRY))
822 goto restart;
823
824 if (!page_cache_get_speculative(page))
825 goto repeat;
826
827 /* Has the page moved? */
828 if (unlikely(page != *((void **)pages[i]))) {
829 page_cache_release(page);
830 goto repeat;
831 }
832
833 pages[ret] = page;
834 ret++;
835 }
836 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return ret;
838}
839
Jens Axboeebf43502006-04-27 08:46:01 +0200840/**
841 * find_get_pages_contig - gang contiguous pagecache lookup
842 * @mapping: The address_space to search
843 * @index: The starting page index
844 * @nr_pages: The maximum number of pages
845 * @pages: Where the resulting pages are placed
846 *
847 * find_get_pages_contig() works exactly like find_get_pages(), except
848 * that the returned number of pages are guaranteed to be contiguous.
849 *
850 * find_get_pages_contig() returns the number of pages which were found.
851 */
852unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
853 unsigned int nr_pages, struct page **pages)
854{
855 unsigned int i;
856 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700857 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200858
Nick Piggina60637c2008-07-25 19:45:31 -0700859 rcu_read_lock();
860restart:
861 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
862 (void ***)pages, index, nr_pages);
863 ret = 0;
864 for (i = 0; i < nr_found; i++) {
865 struct page *page;
866repeat:
867 page = radix_tree_deref_slot((void **)pages[i]);
868 if (unlikely(!page))
869 continue;
870 /*
871 * this can only trigger if nr_found == 1, making livelock
872 * a non issue.
873 */
874 if (unlikely(page == RADIX_TREE_RETRY))
875 goto restart;
876
877 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200878 break;
879
Nick Piggina60637c2008-07-25 19:45:31 -0700880 if (!page_cache_get_speculative(page))
881 goto repeat;
882
883 /* Has the page moved? */
884 if (unlikely(page != *((void **)pages[i]))) {
885 page_cache_release(page);
886 goto repeat;
887 }
888
889 pages[ret] = page;
890 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200891 index++;
892 }
Nick Piggina60637c2008-07-25 19:45:31 -0700893 rcu_read_unlock();
894 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200895}
David Howellsef71c152007-05-09 02:33:44 -0700896EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200897
Randy Dunlap485bb992006-06-23 02:03:49 -0700898/**
899 * find_get_pages_tag - find and return pages that match @tag
900 * @mapping: the address_space to search
901 * @index: the starting page index
902 * @tag: the tag index
903 * @nr_pages: the maximum number of pages
904 * @pages: where the resulting pages are placed
905 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700907 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 */
909unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
910 int tag, unsigned int nr_pages, struct page **pages)
911{
912 unsigned int i;
913 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700914 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Nick Piggina60637c2008-07-25 19:45:31 -0700916 rcu_read_lock();
917restart:
918 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
919 (void ***)pages, *index, nr_pages, tag);
920 ret = 0;
921 for (i = 0; i < nr_found; i++) {
922 struct page *page;
923repeat:
924 page = radix_tree_deref_slot((void **)pages[i]);
925 if (unlikely(!page))
926 continue;
927 /*
928 * this can only trigger if nr_found == 1, making livelock
929 * a non issue.
930 */
931 if (unlikely(page == RADIX_TREE_RETRY))
932 goto restart;
933
934 if (!page_cache_get_speculative(page))
935 goto repeat;
936
937 /* Has the page moved? */
938 if (unlikely(page != *((void **)pages[i]))) {
939 page_cache_release(page);
940 goto repeat;
941 }
942
943 pages[ret] = page;
944 ret++;
945 }
946 rcu_read_unlock();
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (ret)
949 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 return ret;
952}
David Howellsef71c152007-05-09 02:33:44 -0700953EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Randy Dunlap485bb992006-06-23 02:03:49 -0700955/**
956 * grab_cache_page_nowait - returns locked page at given index in given cache
957 * @mapping: target address_space
958 * @index: the page index
959 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800960 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 * This is intended for speculative data generators, where the data can
962 * be regenerated if the page couldn't be grabbed. This routine should
963 * be safe to call while holding the lock for another page.
964 *
965 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
966 * and deadlock against the caller's locked page.
967 */
968struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700969grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970{
971 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200974 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 return page;
976 page_cache_release(page);
977 return NULL;
978 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700979 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800980 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 page_cache_release(page);
982 page = NULL;
983 }
984 return page;
985}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986EXPORT_SYMBOL(grab_cache_page_nowait);
987
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700988/*
989 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
990 * a _large_ part of the i/o request. Imagine the worst scenario:
991 *
992 * ---R__________________________________________B__________
993 * ^ reading here ^ bad block(assume 4k)
994 *
995 * read(R) => miss => readahead(R...B) => media error => frustrating retries
996 * => failing the whole request => read(R) => read(R+1) =>
997 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
998 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
999 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1000 *
1001 * It is going insane. Fix it by quickly scaling down the readahead size.
1002 */
1003static void shrink_readahead_size_eio(struct file *filp,
1004 struct file_ra_state *ra)
1005{
1006 if (!ra->ra_pages)
1007 return;
1008
1009 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001010}
1011
Randy Dunlap485bb992006-06-23 02:03:49 -07001012/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001013 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001014 * @filp: the file to read
1015 * @ppos: current file position
1016 * @desc: read_descriptor
1017 * @actor: read method
1018 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001020 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 *
1022 * This is really ugly. But the goto's actually try to clarify some
1023 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001025static void do_generic_file_read(struct file *filp, loff_t *ppos,
1026 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001028 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001030 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001031 pgoff_t index;
1032 pgoff_t last_index;
1033 pgoff_t prev_index;
1034 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001035 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001039 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1040 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1042 offset = *ppos & ~PAGE_CACHE_MASK;
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 for (;;) {
1045 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001046 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001047 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 unsigned long nr, ret;
1049
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051find_page:
1052 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001053 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001054 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001055 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001056 index, last_index - index);
1057 page = find_get_page(mapping, index);
1058 if (unlikely(page == NULL))
1059 goto no_cached_page;
1060 }
1061 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001062 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001063 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001064 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001066 if (!PageUptodate(page)) {
1067 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1068 !mapping->a_ops->is_partially_uptodate)
1069 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001070 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001071 goto page_not_up_to_date;
1072 if (!mapping->a_ops->is_partially_uptodate(page,
1073 desc, offset))
1074 goto page_not_up_to_date_locked;
1075 unlock_page(page);
1076 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001078 /*
1079 * i_size must be checked after we know the page is Uptodate.
1080 *
1081 * Checking i_size after the check allows us to calculate
1082 * the correct value for "nr", which means the zero-filled
1083 * part of the page is not copied back to userspace (unless
1084 * another truncate extends the file - this is desired though).
1085 */
1086
1087 isize = i_size_read(inode);
1088 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1089 if (unlikely(!isize || index > end_index)) {
1090 page_cache_release(page);
1091 goto out;
1092 }
1093
1094 /* nr is the maximum number of bytes to copy from this page */
1095 nr = PAGE_CACHE_SIZE;
1096 if (index == end_index) {
1097 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1098 if (nr <= offset) {
1099 page_cache_release(page);
1100 goto out;
1101 }
1102 }
1103 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
1105 /* If users can be writing to this page using arbitrary
1106 * virtual addresses, take care about potential aliasing
1107 * before reading the page on the kernel side.
1108 */
1109 if (mapping_writably_mapped(mapping))
1110 flush_dcache_page(page);
1111
1112 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001113 * When a sequential read accesses a page several times,
1114 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 */
Jan Karaec0f1632007-05-06 14:49:25 -07001116 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 mark_page_accessed(page);
1118 prev_index = index;
1119
1120 /*
1121 * Ok, we have the page, and it's up-to-date, so
1122 * now we can copy it to user space...
1123 *
1124 * The actor routine returns how many bytes were actually used..
1125 * NOTE! This may not be the same as how much of a user buffer
1126 * we filled up (we may be padding etc), so we can only update
1127 * "pos" here (the actor routine has to update the user buffer
1128 * pointers and the remaining count).
1129 */
1130 ret = actor(desc, page, offset, nr);
1131 offset += ret;
1132 index += offset >> PAGE_CACHE_SHIFT;
1133 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001134 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 page_cache_release(page);
1137 if (ret == nr && desc->count)
1138 continue;
1139 goto out;
1140
1141page_not_up_to_date:
1142 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001143 error = lock_page_killable(page);
1144 if (unlikely(error))
1145 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001147page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001148 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 if (!page->mapping) {
1150 unlock_page(page);
1151 page_cache_release(page);
1152 continue;
1153 }
1154
1155 /* Did somebody else fill it already? */
1156 if (PageUptodate(page)) {
1157 unlock_page(page);
1158 goto page_ok;
1159 }
1160
1161readpage:
1162 /* Start the actual read. The read will unlock the page. */
1163 error = mapping->a_ops->readpage(filp, page);
1164
Zach Brown994fc28c2005-12-15 14:28:17 -08001165 if (unlikely(error)) {
1166 if (error == AOP_TRUNCATED_PAGE) {
1167 page_cache_release(page);
1168 goto find_page;
1169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001174 error = lock_page_killable(page);
1175 if (unlikely(error))
1176 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (!PageUptodate(page)) {
1178 if (page->mapping == NULL) {
1179 /*
1180 * invalidate_inode_pages got it
1181 */
1182 unlock_page(page);
1183 page_cache_release(page);
1184 goto find_page;
1185 }
1186 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001187 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001188 error = -EIO;
1189 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 }
1191 unlock_page(page);
1192 }
1193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 goto page_ok;
1195
1196readpage_error:
1197 /* UHHUH! A synchronous read error occurred. Report it */
1198 desc->error = error;
1199 page_cache_release(page);
1200 goto out;
1201
1202no_cached_page:
1203 /*
1204 * Ok, it wasn't cached, so we need to create a new
1205 * page..
1206 */
Nick Piggineb2be182007-10-16 01:24:57 -07001207 page = page_cache_alloc_cold(mapping);
1208 if (!page) {
1209 desc->error = -ENOMEM;
1210 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 }
Nick Piggineb2be182007-10-16 01:24:57 -07001212 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 index, GFP_KERNEL);
1214 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001215 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (error == -EEXIST)
1217 goto find_page;
1218 desc->error = error;
1219 goto out;
1220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 goto readpage;
1222 }
1223
1224out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001225 ra->prev_pos = prev_index;
1226 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1227 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001229 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001230 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233int file_read_actor(read_descriptor_t *desc, struct page *page,
1234 unsigned long offset, unsigned long size)
1235{
1236 char *kaddr;
1237 unsigned long left, count = desc->count;
1238
1239 if (size > count)
1240 size = count;
1241
1242 /*
1243 * Faults on the destination of a read are common, so do it before
1244 * taking the kmap.
1245 */
1246 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1247 kaddr = kmap_atomic(page, KM_USER0);
1248 left = __copy_to_user_inatomic(desc->arg.buf,
1249 kaddr + offset, size);
1250 kunmap_atomic(kaddr, KM_USER0);
1251 if (left == 0)
1252 goto success;
1253 }
1254
1255 /* Do it the slow way */
1256 kaddr = kmap(page);
1257 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1258 kunmap(page);
1259
1260 if (left) {
1261 size -= left;
1262 desc->error = -EFAULT;
1263 }
1264success:
1265 desc->count = count - size;
1266 desc->written += size;
1267 desc->arg.buf += size;
1268 return size;
1269}
1270
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001271/*
1272 * Performs necessary checks before doing a write
1273 * @iov: io vector request
1274 * @nr_segs: number of segments in the iovec
1275 * @count: number of bytes to write
1276 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1277 *
1278 * Adjust number of segments and amount of bytes to write (nr_segs should be
1279 * properly initialized first). Returns appropriate error code that caller
1280 * should return or zero in case that write should be allowed.
1281 */
1282int generic_segment_checks(const struct iovec *iov,
1283 unsigned long *nr_segs, size_t *count, int access_flags)
1284{
1285 unsigned long seg;
1286 size_t cnt = 0;
1287 for (seg = 0; seg < *nr_segs; seg++) {
1288 const struct iovec *iv = &iov[seg];
1289
1290 /*
1291 * If any segment has a negative length, or the cumulative
1292 * length ever wraps negative then return -EINVAL.
1293 */
1294 cnt += iv->iov_len;
1295 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1296 return -EINVAL;
1297 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1298 continue;
1299 if (seg == 0)
1300 return -EFAULT;
1301 *nr_segs = seg;
1302 cnt -= iv->iov_len; /* This segment is no good */
1303 break;
1304 }
1305 *count = cnt;
1306 return 0;
1307}
1308EXPORT_SYMBOL(generic_segment_checks);
1309
Randy Dunlap485bb992006-06-23 02:03:49 -07001310/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001311 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001312 * @iocb: kernel I/O control block
1313 * @iov: io vector request
1314 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001315 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001316 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 * This is the "read()" routine for all filesystems
1318 * that can use the page cache directly.
1319 */
1320ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001321generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1322 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323{
1324 struct file *filp = iocb->ki_filp;
1325 ssize_t retval;
1326 unsigned long seg;
1327 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001328 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
1330 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001331 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1332 if (retval)
1333 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1336 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001337 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 struct address_space *mapping;
1339 struct inode *inode;
1340
1341 mapping = filp->f_mapping;
1342 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (!count)
1344 goto out; /* skip atime */
1345 size = i_size_read(inode);
1346 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001347 retval = filemap_write_and_wait_range(mapping, pos,
1348 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001349 if (!retval) {
1350 retval = mapping->a_ops->direct_IO(READ, iocb,
1351 iov, pos, nr_segs);
1352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 if (retval > 0)
1354 *ppos = pos + retval;
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001355 if (retval) {
1356 file_accessed(filp);
1357 goto out;
1358 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 }
1361
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001362 for (seg = 0; seg < nr_segs; seg++) {
1363 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001365 desc.written = 0;
1366 desc.arg.buf = iov[seg].iov_base;
1367 desc.count = iov[seg].iov_len;
1368 if (desc.count == 0)
1369 continue;
1370 desc.error = 0;
1371 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1372 retval += desc.written;
1373 if (desc.error) {
1374 retval = retval ?: desc.error;
1375 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
Hugh Dickins11fa977e2008-07-23 21:27:34 -07001377 if (desc.count > 0)
1378 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 }
1380out:
1381 return retval;
1382}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383EXPORT_SYMBOL(generic_file_aio_read);
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385static ssize_t
1386do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001387 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
1389 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1390 return -EINVAL;
1391
1392 force_page_cache_readahead(mapping, filp, index,
1393 max_sane_readahead(nr));
1394 return 0;
1395}
1396
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001397SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398{
1399 ssize_t ret;
1400 struct file *file;
1401
1402 ret = -EBADF;
1403 file = fget(fd);
1404 if (file) {
1405 if (file->f_mode & FMODE_READ) {
1406 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001407 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1408 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 unsigned long len = end - start + 1;
1410 ret = do_readahead(mapping, file, start, len);
1411 }
1412 fput(file);
1413 }
1414 return ret;
1415}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001416#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1417asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1418{
1419 return SYSC_readahead((int) fd, offset, (size_t) count);
1420}
1421SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1422#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001425/**
1426 * page_cache_read - adds requested page to the page cache if not already there
1427 * @file: file to read
1428 * @offset: page index
1429 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 * This adds the requested page to the page cache if it isn't already there,
1431 * and schedules an I/O to read in its contents from disk.
1432 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001433static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434{
1435 struct address_space *mapping = file->f_mapping;
1436 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001437 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
Zach Brown994fc28c2005-12-15 14:28:17 -08001439 do {
1440 page = page_cache_alloc_cold(mapping);
1441 if (!page)
1442 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Zach Brown994fc28c2005-12-15 14:28:17 -08001444 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1445 if (ret == 0)
1446 ret = mapping->a_ops->readpage(file, page);
1447 else if (ret == -EEXIST)
1448 ret = 0; /* losing race to add is OK */
1449
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Zach Brown994fc28c2005-12-15 14:28:17 -08001452 } while (ret == AOP_TRUNCATED_PAGE);
1453
1454 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455}
1456
1457#define MMAP_LOTSAMISS (100)
1458
Randy Dunlap485bb992006-06-23 02:03:49 -07001459/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001460 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001461 * @vma: vma in which the fault was taken
1462 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001463 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001464 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 * mapped memory region to read in file data during a page fault.
1466 *
1467 * The goto's are kind of ugly, but this streamlines the normal case of having
1468 * it in the page cache, and handles the special cases reasonably without
1469 * having a lot of duplicated code.
1470 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001471int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
1473 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001474 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 struct address_space *mapping = file->f_mapping;
1476 struct file_ra_state *ra = &file->f_ra;
1477 struct inode *inode = mapping->host;
1478 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001479 pgoff_t size;
Nick Piggin54cb8822007-07-19 01:46:59 -07001480 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001481 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001484 if (vmf->pgoff >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001485 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
1487 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001488 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 goto no_cached_page;
1490
1491 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 * Do we have something in the page cache already?
1493 */
1494retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001495 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001496 /*
1497 * For sequential accesses, we use the generic readahead logic.
1498 */
1499 if (VM_SequentialReadHint(vma)) {
1500 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001501 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001502 vmf->pgoff, 1);
1503 page = find_lock_page(mapping, vmf->pgoff);
1504 if (!page)
1505 goto no_cached_page;
1506 }
1507 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001508 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001509 vmf->pgoff, 1);
1510 }
1511 }
1512
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 if (!page) {
1514 unsigned long ra_pages;
1515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 ra->mmap_miss++;
1517
1518 /*
1519 * Do we miss much more than hit in this file? If so,
1520 * stop bothering with read-ahead. It will only hurt.
1521 */
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001522 if (ra->mmap_miss > MMAP_LOTSAMISS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 goto no_cached_page;
1524
1525 /*
1526 * To keep the pgmajfault counter straight, we need to
1527 * check did_readaround, as this is an inner loop.
1528 */
1529 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001530 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001531 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 }
1533 did_readaround = 1;
1534 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1535 if (ra_pages) {
1536 pgoff_t start = 0;
1537
Nick Piggind0217ac2007-07-19 01:47:03 -07001538 if (vmf->pgoff > ra_pages / 2)
1539 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 do_page_cache_readahead(mapping, file, start, ra_pages);
1541 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001542 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 if (!page)
1544 goto no_cached_page;
1545 }
1546
1547 if (!did_readaround)
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001548 ra->mmap_miss--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001551 * We have a locked page in the page cache, now we need to check
1552 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
Nick Piggind00806b2007-07-19 01:46:57 -07001554 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 goto page_not_uptodate;
1556
Nick Piggind00806b2007-07-19 01:46:57 -07001557 /* Must recheck i_size under page lock */
1558 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001559 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001560 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001561 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001562 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001563 }
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 /*
1566 * Found the page and have a reference on it.
1567 */
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001568 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001569 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001570 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572no_cached_page:
1573 /*
1574 * We're only likely to ever get here if MADV_RANDOM is in
1575 * effect.
1576 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001577 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
1579 /*
1580 * The page we want has now been added to the page cache.
1581 * In the unlikely event that someone removed it in the
1582 * meantime, we'll just come back here and read it again.
1583 */
1584 if (error >= 0)
1585 goto retry_find;
1586
1587 /*
1588 * An error return from page_cache_read can result if the
1589 * system is low on memory, or a problem occurs while trying
1590 * to schedule I/O.
1591 */
1592 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001593 return VM_FAULT_OOM;
1594 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
1596page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001597 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001599 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001600 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602
1603 /*
1604 * Umm, take care of errors if the page isn't up-to-date.
1605 * Try to re-read it _once_. We do this synchronously,
1606 * because there really aren't any performance issues here
1607 * and we need to check for errors.
1608 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001610 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001611 if (!error) {
1612 wait_on_page_locked(page);
1613 if (!PageUptodate(page))
1614 error = -EIO;
1615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001617
1618 if (!error || error == AOP_TRUNCATED_PAGE)
1619 goto retry_find;
1620
1621 /* Things didn't work out. Return zero to tell the mm layer so. */
1622 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001623 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001624}
1625EXPORT_SYMBOL(filemap_fault);
1626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001628 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629};
1630
1631/* This is used for a general mmap of a disk file */
1632
1633int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1634{
1635 struct address_space *mapping = file->f_mapping;
1636
1637 if (!mapping->a_ops->readpage)
1638 return -ENOEXEC;
1639 file_accessed(file);
1640 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001641 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 return 0;
1643}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
1645/*
1646 * This is for filesystems which do not implement ->writepage.
1647 */
1648int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1649{
1650 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1651 return -EINVAL;
1652 return generic_file_mmap(file, vma);
1653}
1654#else
1655int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1656{
1657 return -ENOSYS;
1658}
1659int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1660{
1661 return -ENOSYS;
1662}
1663#endif /* CONFIG_MMU */
1664
1665EXPORT_SYMBOL(generic_file_mmap);
1666EXPORT_SYMBOL(generic_file_readonly_mmap);
1667
Nick Piggin6fe69002007-05-06 14:49:04 -07001668static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001669 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 int (*filler)(void *,struct page*),
1671 void *data)
1672{
Nick Piggineb2be182007-10-16 01:24:57 -07001673 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 int err;
1675repeat:
1676 page = find_get_page(mapping, index);
1677 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001678 page = page_cache_alloc_cold(mapping);
1679 if (!page)
1680 return ERR_PTR(-ENOMEM);
1681 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1682 if (unlikely(err)) {
1683 page_cache_release(page);
1684 if (err == -EEXIST)
1685 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 return ERR_PTR(err);
1688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 err = filler(data, page);
1690 if (err < 0) {
1691 page_cache_release(page);
1692 page = ERR_PTR(err);
1693 }
1694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 return page;
1696}
1697
Randy Dunlap76824862008-03-19 17:00:40 -07001698/**
1699 * read_cache_page_async - read into page cache, fill it if needed
1700 * @mapping: the page's address_space
1701 * @index: the page index
1702 * @filler: function to perform the read
1703 * @data: destination for read data
1704 *
Nick Piggin6fe69002007-05-06 14:49:04 -07001705 * Same as read_cache_page, but don't wait for page to become unlocked
1706 * after submitting it to the filler.
Randy Dunlap76824862008-03-19 17:00:40 -07001707 *
1708 * Read into the page cache. If a page already exists, and PageUptodate() is
1709 * not set, try to fill the page but don't wait for it to become unlocked.
1710 *
1711 * If the page does not get brought uptodate, return -EIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001713struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001714 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 int (*filler)(void *,struct page*),
1716 void *data)
1717{
1718 struct page *page;
1719 int err;
1720
1721retry:
1722 page = __read_cache_page(mapping, index, filler, data);
1723 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001724 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 if (PageUptodate(page))
1726 goto out;
1727
1728 lock_page(page);
1729 if (!page->mapping) {
1730 unlock_page(page);
1731 page_cache_release(page);
1732 goto retry;
1733 }
1734 if (PageUptodate(page)) {
1735 unlock_page(page);
1736 goto out;
1737 }
1738 err = filler(data, page);
1739 if (err < 0) {
1740 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001741 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 }
David Howellsc855ff32007-05-09 13:42:20 +01001743out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001744 mark_page_accessed(page);
1745 return page;
1746}
1747EXPORT_SYMBOL(read_cache_page_async);
1748
1749/**
1750 * read_cache_page - read into page cache, fill it if needed
1751 * @mapping: the page's address_space
1752 * @index: the page index
1753 * @filler: function to perform the read
1754 * @data: destination for read data
1755 *
1756 * Read into the page cache. If a page already exists, and PageUptodate() is
1757 * not set, try to fill the page then wait for it to become unlocked.
1758 *
1759 * If the page does not get brought uptodate, return -EIO.
1760 */
1761struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001762 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001763 int (*filler)(void *,struct page*),
1764 void *data)
1765{
1766 struct page *page;
1767
1768 page = read_cache_page_async(mapping, index, filler, data);
1769 if (IS_ERR(page))
1770 goto out;
1771 wait_on_page_locked(page);
1772 if (!PageUptodate(page)) {
1773 page_cache_release(page);
1774 page = ERR_PTR(-EIO);
1775 }
1776 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 return page;
1778}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779EXPORT_SYMBOL(read_cache_page);
1780
1781/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 * The logic we want is
1783 *
1784 * if suid or (sgid and xgrp)
1785 * remove privs
1786 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001787int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788{
1789 mode_t mode = dentry->d_inode->i_mode;
1790 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
1792 /* suid always must be killed */
1793 if (unlikely(mode & S_ISUID))
1794 kill = ATTR_KILL_SUID;
1795
1796 /*
1797 * sgid without any exec bits is just a mandatory locking mark; leave
1798 * it alone. If some exec bits are set, it's a real sgid; kill it.
1799 */
1800 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1801 kill |= ATTR_KILL_SGID;
1802
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001803 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001804 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
Jens Axboe01de85e2006-10-17 19:50:36 +02001806 return 0;
1807}
Mark Fashehd23a1472006-10-17 17:05:18 -07001808EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001809
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001810static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001811{
1812 struct iattr newattrs;
1813
1814 newattrs.ia_valid = ATTR_FORCE | kill;
1815 return notify_change(dentry, &newattrs);
1816}
1817
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001818int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001819{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001820 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001821 int killsuid = should_remove_suid(dentry);
1822 int killpriv = security_inode_need_killpriv(dentry);
1823 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001824
Serge E. Hallynb5376772007-10-16 23:31:36 -07001825 if (killpriv < 0)
1826 return killpriv;
1827 if (killpriv)
1828 error = security_inode_killpriv(dentry);
1829 if (!error && killsuid)
1830 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001831
Serge E. Hallynb5376772007-10-16 23:31:36 -07001832 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001834EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Nick Piggin2f718ff2007-10-16 01:24:59 -07001836static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 const struct iovec *iov, size_t base, size_t bytes)
1838{
Ingo Molnarf1800532009-03-02 11:00:57 +01001839 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
1841 while (bytes) {
1842 char __user *buf = iov->iov_base + base;
1843 int copy = min(bytes, iov->iov_len - base);
1844
1845 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001846 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 copied += copy;
1848 bytes -= copy;
1849 vaddr += copy;
1850 iov++;
1851
NeilBrown01408c42006-06-25 05:47:58 -07001852 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 }
1855 return copied - left;
1856}
1857
1858/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001859 * Copy as much as we can into the page and return the number of bytes which
1860 * were sucessfully copied. If a fault is encountered then return the number of
1861 * bytes which were copied.
1862 */
1863size_t iov_iter_copy_from_user_atomic(struct page *page,
1864 struct iov_iter *i, unsigned long offset, size_t bytes)
1865{
1866 char *kaddr;
1867 size_t copied;
1868
1869 BUG_ON(!in_atomic());
1870 kaddr = kmap_atomic(page, KM_USER0);
1871 if (likely(i->nr_segs == 1)) {
1872 int left;
1873 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001874 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001875 copied = bytes - left;
1876 } else {
1877 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1878 i->iov, i->iov_offset, bytes);
1879 }
1880 kunmap_atomic(kaddr, KM_USER0);
1881
1882 return copied;
1883}
Nick Piggin89e10782007-10-16 01:25:07 -07001884EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001885
1886/*
1887 * This has the same sideeffects and return value as
1888 * iov_iter_copy_from_user_atomic().
1889 * The difference is that it attempts to resolve faults.
1890 * Page must not be locked.
1891 */
1892size_t iov_iter_copy_from_user(struct page *page,
1893 struct iov_iter *i, unsigned long offset, size_t bytes)
1894{
1895 char *kaddr;
1896 size_t copied;
1897
1898 kaddr = kmap(page);
1899 if (likely(i->nr_segs == 1)) {
1900 int left;
1901 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001902 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001903 copied = bytes - left;
1904 } else {
1905 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1906 i->iov, i->iov_offset, bytes);
1907 }
1908 kunmap(page);
1909 return copied;
1910}
Nick Piggin89e10782007-10-16 01:25:07 -07001911EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001912
Nick Pigginf7009262008-03-10 11:43:59 -07001913void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001914{
Nick Pigginf7009262008-03-10 11:43:59 -07001915 BUG_ON(i->count < bytes);
1916
Nick Piggin2f718ff2007-10-16 01:24:59 -07001917 if (likely(i->nr_segs == 1)) {
1918 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001919 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001920 } else {
1921 const struct iovec *iov = i->iov;
1922 size_t base = i->iov_offset;
1923
Nick Piggin124d3b72008-02-02 15:01:17 +01001924 /*
1925 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001926 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001927 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001928 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001929 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001930
Nick Pigginf7009262008-03-10 11:43:59 -07001931 copy = min(bytes, iov->iov_len - base);
1932 BUG_ON(!i->count || i->count < copy);
1933 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001934 bytes -= copy;
1935 base += copy;
1936 if (iov->iov_len == base) {
1937 iov++;
1938 base = 0;
1939 }
1940 }
1941 i->iov = iov;
1942 i->iov_offset = base;
1943 }
1944}
Nick Piggin89e10782007-10-16 01:25:07 -07001945EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001946
Nick Pigginafddba42007-10-16 01:25:01 -07001947/*
1948 * Fault in the first iovec of the given iov_iter, to a maximum length
1949 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1950 * accessed (ie. because it is an invalid address).
1951 *
1952 * writev-intensive code may want this to prefault several iovecs -- that
1953 * would be possible (callers must not rely on the fact that _only_ the
1954 * first iovec will be faulted with the current implementation).
1955 */
1956int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001957{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001958 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001959 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1960 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001961}
Nick Piggin89e10782007-10-16 01:25:07 -07001962EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001963
1964/*
1965 * Return the count of just the current iov_iter segment.
1966 */
1967size_t iov_iter_single_seg_count(struct iov_iter *i)
1968{
1969 const struct iovec *iov = i->iov;
1970 if (i->nr_segs == 1)
1971 return i->count;
1972 else
1973 return min(i->count, iov->iov_len - i->iov_offset);
1974}
Nick Piggin89e10782007-10-16 01:25:07 -07001975EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001976
1977/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 * Performs necessary checks before doing a write
1979 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001980 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 * Returns appropriate error code that caller should return or
1982 * zero in case that write should be allowed.
1983 */
1984inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1985{
1986 struct inode *inode = file->f_mapping->host;
1987 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1988
1989 if (unlikely(*pos < 0))
1990 return -EINVAL;
1991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 if (!isblk) {
1993 /* FIXME: this is for backwards compatibility with 2.4 */
1994 if (file->f_flags & O_APPEND)
1995 *pos = i_size_read(inode);
1996
1997 if (limit != RLIM_INFINITY) {
1998 if (*pos >= limit) {
1999 send_sig(SIGXFSZ, current, 0);
2000 return -EFBIG;
2001 }
2002 if (*count > limit - (typeof(limit))*pos) {
2003 *count = limit - (typeof(limit))*pos;
2004 }
2005 }
2006 }
2007
2008 /*
2009 * LFS rule
2010 */
2011 if (unlikely(*pos + *count > MAX_NON_LFS &&
2012 !(file->f_flags & O_LARGEFILE))) {
2013 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 return -EFBIG;
2015 }
2016 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2017 *count = MAX_NON_LFS - (unsigned long)*pos;
2018 }
2019 }
2020
2021 /*
2022 * Are we about to exceed the fs block limit ?
2023 *
2024 * If we have written data it becomes a short write. If we have
2025 * exceeded without writing data we send a signal and return EFBIG.
2026 * Linus frestrict idea will clean these up nicely..
2027 */
2028 if (likely(!isblk)) {
2029 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2030 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 return -EFBIG;
2032 }
2033 /* zero-length writes at ->s_maxbytes are OK */
2034 }
2035
2036 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2037 *count = inode->i_sb->s_maxbytes - *pos;
2038 } else {
David Howells93614012006-09-30 20:45:40 +02002039#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 loff_t isize;
2041 if (bdev_read_only(I_BDEV(inode)))
2042 return -EPERM;
2043 isize = i_size_read(inode);
2044 if (*pos >= isize) {
2045 if (*count || *pos > isize)
2046 return -ENOSPC;
2047 }
2048
2049 if (*pos + *count > isize)
2050 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002051#else
2052 return -EPERM;
2053#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 }
2055 return 0;
2056}
2057EXPORT_SYMBOL(generic_write_checks);
2058
Nick Pigginafddba42007-10-16 01:25:01 -07002059int pagecache_write_begin(struct file *file, struct address_space *mapping,
2060 loff_t pos, unsigned len, unsigned flags,
2061 struct page **pagep, void **fsdata)
2062{
2063 const struct address_space_operations *aops = mapping->a_ops;
2064
Nick Piggin4e02ed42008-10-29 14:00:55 -07002065 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002066 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002067}
2068EXPORT_SYMBOL(pagecache_write_begin);
2069
2070int pagecache_write_end(struct file *file, struct address_space *mapping,
2071 loff_t pos, unsigned len, unsigned copied,
2072 struct page *page, void *fsdata)
2073{
2074 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002075
Nick Piggin4e02ed42008-10-29 14:00:55 -07002076 mark_page_accessed(page);
2077 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002078}
2079EXPORT_SYMBOL(pagecache_write_end);
2080
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081ssize_t
2082generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2083 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2084 size_t count, size_t ocount)
2085{
2086 struct file *file = iocb->ki_filp;
2087 struct address_space *mapping = file->f_mapping;
2088 struct inode *inode = mapping->host;
2089 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002090 size_t write_len;
2091 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
2093 if (count != ocount)
2094 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2095
Christoph Hellwiga969e902008-07-23 21:27:04 -07002096 write_len = iov_length(iov, *nr_segs);
2097 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002098
Nick Piggin48b47c52009-01-06 14:40:22 -08002099 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002100 if (written)
2101 goto out;
2102
2103 /*
2104 * After a write we want buffered reads to be sure to go to disk to get
2105 * the new data. We invalidate clean cached page from the region we're
2106 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002107 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002108 */
2109 if (mapping->nrpages) {
2110 written = invalidate_inode_pages2_range(mapping,
2111 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002112 /*
2113 * If a page can not be invalidated, return 0 to fall back
2114 * to buffered write.
2115 */
2116 if (written) {
2117 if (written == -EBUSY)
2118 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002119 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002120 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002121 }
2122
2123 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2124
2125 /*
2126 * Finally, try again to invalidate clean pages which might have been
2127 * cached by non-direct readahead, or faulted in by get_user_pages()
2128 * if the source of the write was an mmap'ed region of the file
2129 * we're writing. Either one is a pretty crazy thing to do,
2130 * so we don't support it 100%. If this invalidation
2131 * fails, tough, the write still worked...
2132 */
2133 if (mapping->nrpages) {
2134 invalidate_inode_pages2_range(mapping,
2135 pos >> PAGE_CACHE_SHIFT, end);
2136 }
2137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 if (written > 0) {
2139 loff_t end = pos + written;
2140 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2141 i_size_write(inode, end);
2142 mark_inode_dirty(inode);
2143 }
2144 *ppos = end;
2145 }
2146
2147 /*
2148 * Sync the fs metadata but not the minor inode changes and
2149 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002150 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002151 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 */
Christoph Hellwiga969e902008-07-23 21:27:04 -07002153out:
Zach Brown8459d862006-12-10 02:21:05 -08002154 if ((written >= 0 || written == -EIOCBQUEUED) &&
2155 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002156 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2157 if (err < 0)
2158 written = err;
2159 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 return written;
2161}
2162EXPORT_SYMBOL(generic_file_direct_write);
2163
Nick Piggineb2be182007-10-16 01:24:57 -07002164/*
2165 * Find or create a page at the given pagecache position. Return the locked
2166 * page. This function is specifically for buffered writes.
2167 */
Nick Piggin54566b22009-01-04 12:00:53 -08002168struct page *grab_cache_page_write_begin(struct address_space *mapping,
2169 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002170{
2171 int status;
2172 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002173 gfp_t gfp_notmask = 0;
2174 if (flags & AOP_FLAG_NOFS)
2175 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002176repeat:
2177 page = find_lock_page(mapping, index);
2178 if (likely(page))
2179 return page;
2180
Nick Piggin54566b22009-01-04 12:00:53 -08002181 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002182 if (!page)
2183 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002184 status = add_to_page_cache_lru(page, mapping, index,
2185 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002186 if (unlikely(status)) {
2187 page_cache_release(page);
2188 if (status == -EEXIST)
2189 goto repeat;
2190 return NULL;
2191 }
2192 return page;
2193}
Nick Piggin54566b22009-01-04 12:00:53 -08002194EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002195
Nick Pigginafddba42007-10-16 01:25:01 -07002196static ssize_t generic_perform_write(struct file *file,
2197 struct iov_iter *i, loff_t pos)
2198{
2199 struct address_space *mapping = file->f_mapping;
2200 const struct address_space_operations *a_ops = mapping->a_ops;
2201 long status = 0;
2202 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002203 unsigned int flags = 0;
2204
2205 /*
2206 * Copies from kernel address space cannot fail (NFSD is a big user).
2207 */
2208 if (segment_eq(get_fs(), KERNEL_DS))
2209 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002210
2211 do {
2212 struct page *page;
2213 pgoff_t index; /* Pagecache index for current page */
2214 unsigned long offset; /* Offset into pagecache page */
2215 unsigned long bytes; /* Bytes to write to page */
2216 size_t copied; /* Bytes copied from user */
2217 void *fsdata;
2218
2219 offset = (pos & (PAGE_CACHE_SIZE - 1));
2220 index = pos >> PAGE_CACHE_SHIFT;
2221 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2222 iov_iter_count(i));
2223
2224again:
2225
2226 /*
2227 * Bring in the user page that we will copy from _first_.
2228 * Otherwise there's a nasty deadlock on copying from the
2229 * same page as we're writing to, without it being marked
2230 * up-to-date.
2231 *
2232 * Not only is this an optimisation, but it is also required
2233 * to check that the address is actually valid, when atomic
2234 * usercopies are used, below.
2235 */
2236 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2237 status = -EFAULT;
2238 break;
2239 }
2240
Nick Piggin674b8922007-10-16 01:25:03 -07002241 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002242 &page, &fsdata);
2243 if (unlikely(status))
2244 break;
2245
2246 pagefault_disable();
2247 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2248 pagefault_enable();
2249 flush_dcache_page(page);
2250
2251 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2252 page, fsdata);
2253 if (unlikely(status < 0))
2254 break;
2255 copied = status;
2256
2257 cond_resched();
2258
Nick Piggin124d3b72008-02-02 15:01:17 +01002259 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002260 if (unlikely(copied == 0)) {
2261 /*
2262 * If we were unable to copy any data at all, we must
2263 * fall back to a single segment length write.
2264 *
2265 * If we didn't fallback here, we could livelock
2266 * because not all segments in the iov can be copied at
2267 * once without a pagefault.
2268 */
2269 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2270 iov_iter_single_seg_count(i));
2271 goto again;
2272 }
Nick Pigginafddba42007-10-16 01:25:01 -07002273 pos += copied;
2274 written += copied;
2275
2276 balance_dirty_pages_ratelimited(mapping);
2277
2278 } while (iov_iter_count(i));
2279
2280 return written ? written : status;
2281}
2282
2283ssize_t
2284generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2285 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2286 size_t count, ssize_t written)
2287{
2288 struct file *file = iocb->ki_filp;
2289 struct address_space *mapping = file->f_mapping;
2290 const struct address_space_operations *a_ops = mapping->a_ops;
2291 struct inode *inode = mapping->host;
2292 ssize_t status;
2293 struct iov_iter i;
2294
2295 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002296 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002299 written += status;
2300 *ppos = pos + status;
2301
2302 /*
2303 * For now, when the user asks for O_SYNC, we'll actually give
2304 * O_DSYNC
2305 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2307 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2308 status = generic_osync_inode(inode, mapping,
2309 OSYNC_METADATA|OSYNC_DATA);
2310 }
2311 }
2312
2313 /*
2314 * If we get here for O_DIRECT writes then we must have fallen through
2315 * to buffered writes (block instantiation inside i_size). So we sync
2316 * the file data here, to try to honour O_DIRECT expectations.
2317 */
2318 if (unlikely(file->f_flags & O_DIRECT) && written)
Nick Piggin48b47c52009-01-06 14:40:22 -08002319 status = filemap_write_and_wait_range(mapping,
2320 pos, pos + written - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 return written ? written : status;
2323}
2324EXPORT_SYMBOL(generic_file_buffered_write);
2325
Adrian Bunk5ce78522005-09-10 00:26:28 -07002326static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2328 unsigned long nr_segs, loff_t *ppos)
2329{
2330 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002331 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 size_t ocount; /* original count */
2333 size_t count; /* after file limit checks */
2334 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 loff_t pos;
2336 ssize_t written;
2337 ssize_t err;
2338
2339 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002340 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2341 if (err)
2342 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
2344 count = ocount;
2345 pos = *ppos;
2346
2347 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2348
2349 /* We can write back this queue in page reclaim */
2350 current->backing_dev_info = mapping->backing_dev_info;
2351 written = 0;
2352
2353 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2354 if (err)
2355 goto out;
2356
2357 if (count == 0)
2358 goto out;
2359
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002360 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 if (err)
2362 goto out;
2363
Christoph Hellwig870f4812006-01-09 20:52:01 -08002364 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
2366 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2367 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002368 loff_t endbyte;
2369 ssize_t written_buffered;
2370
2371 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2372 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 if (written < 0 || written == count)
2374 goto out;
2375 /*
2376 * direct-io write to a hole: fall through to buffered I/O
2377 * for completing the rest of the request.
2378 */
2379 pos += written;
2380 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002381 written_buffered = generic_file_buffered_write(iocb, iov,
2382 nr_segs, pos, ppos, count,
2383 written);
2384 /*
2385 * If generic_file_buffered_write() retuned a synchronous error
2386 * then we want to return the number of bytes which were
2387 * direct-written, or the error code if that was zero. Note
2388 * that this differs from normal direct-io semantics, which
2389 * will return -EFOO even if some bytes were written.
2390 */
2391 if (written_buffered < 0) {
2392 err = written_buffered;
2393 goto out;
2394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002396 /*
2397 * We need to ensure that the page cache pages are written to
2398 * disk and invalidated to preserve the expected O_DIRECT
2399 * semantics.
2400 */
2401 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002402 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2403 SYNC_FILE_RANGE_WAIT_BEFORE|
2404 SYNC_FILE_RANGE_WRITE|
2405 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002406 if (err == 0) {
2407 written = written_buffered;
2408 invalidate_mapping_pages(mapping,
2409 pos >> PAGE_CACHE_SHIFT,
2410 endbyte >> PAGE_CACHE_SHIFT);
2411 } else {
2412 /*
2413 * We don't know how much we wrote, so just return
2414 * the number of bytes which were direct-written
2415 */
2416 }
2417 } else {
2418 written = generic_file_buffered_write(iocb, iov, nr_segs,
2419 pos, ppos, count, written);
2420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421out:
2422 current->backing_dev_info = NULL;
2423 return written ? written : err;
2424}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Badari Pulavarty027445c2006-09-30 23:28:46 -07002426ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2427 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428{
2429 struct file *file = iocb->ki_filp;
2430 struct address_space *mapping = file->f_mapping;
2431 struct inode *inode = mapping->host;
2432 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Badari Pulavarty027445c2006-09-30 23:28:46 -07002434 BUG_ON(iocb->ki_pos != pos);
2435
2436 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2437 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
2439 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002440 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
2442 err = sync_page_range_nolock(inode, mapping, pos, ret);
2443 if (err < 0)
2444 ret = err;
2445 }
2446 return ret;
2447}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002448EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Badari Pulavarty027445c2006-09-30 23:28:46 -07002450ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2451 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452{
2453 struct file *file = iocb->ki_filp;
2454 struct address_space *mapping = file->f_mapping;
2455 struct inode *inode = mapping->host;
2456 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
2458 BUG_ON(iocb->ki_pos != pos);
2459
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002460 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002461 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2462 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002463 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464
2465 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2466 ssize_t err;
2467
2468 err = sync_page_range(inode, mapping, pos, ret);
2469 if (err < 0)
2470 ret = err;
2471 }
2472 return ret;
2473}
2474EXPORT_SYMBOL(generic_file_aio_write);
2475
David Howellscf9a2ae2006-08-29 19:05:54 +01002476/**
2477 * try_to_release_page() - release old fs-specific metadata on a page
2478 *
2479 * @page: the page which the kernel is trying to free
2480 * @gfp_mask: memory allocation flags (and I/O mode)
2481 *
2482 * The address_space is to try to release any data against the page
2483 * (presumably at page->private). If the release was successful, return `1'.
2484 * Otherwise return zero.
2485 *
David Howells266cf652009-04-03 16:42:36 +01002486 * This may also be called if PG_fscache is set on a page, indicating that the
2487 * page is known to the local caching routines.
2488 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002489 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002490 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002491 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002492 */
2493int try_to_release_page(struct page *page, gfp_t gfp_mask)
2494{
2495 struct address_space * const mapping = page->mapping;
2496
2497 BUG_ON(!PageLocked(page));
2498 if (PageWriteback(page))
2499 return 0;
2500
2501 if (mapping && mapping->a_ops->releasepage)
2502 return mapping->a_ops->releasepage(page, gfp_mask);
2503 return try_to_free_buffers(page);
2504}
2505
2506EXPORT_SYMBOL(try_to_release_page);