blob: ccea3b665c12571ac32d6d936b3862e103f7b995 [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));
Linus Torvalds3a692792007-12-19 14:05:13 -0800124
125 /*
126 * Some filesystems seem to re-dirty the page even after
127 * the VM has canceled the dirty bit (eg ext3 journaling).
128 *
129 * Fix it up by doing a final dirty accounting check after
130 * having removed the page entirely.
131 */
132 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
133 dec_zone_page_state(page, NR_FILE_DIRTY);
134 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
138void remove_from_page_cache(struct page *page)
139{
140 struct address_space *mapping = page->mapping;
141
Matt Mackallcd7619d2005-05-01 08:59:01 -0700142 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Nick Piggin19fd6232008-07-25 19:45:32 -0700144 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700146 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700147 mem_cgroup_uncharge_cache_page(page);
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}
Chris Masonf6995582009-04-15 13:22:37 -0400444EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Randy Dunlap485bb992006-06-23 02:03:49 -0700446/**
Nick Piggine2867812008-07-25 19:45:30 -0700447 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700448 * @page: page to add
449 * @mapping: the page's address_space
450 * @offset: page index
451 * @gfp_mask: page allocation mode
452 *
Nick Piggine2867812008-07-25 19:45:30 -0700453 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 * This function does not add the page to the LRU. The caller must do that.
455 */
Nick Piggine2867812008-07-25 19:45:30 -0700456int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400457 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Nick Piggine2867812008-07-25 19:45:30 -0700459 int error;
460
461 VM_BUG_ON(!PageLocked(page));
462
463 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800464 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800465 if (error)
466 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Balbir Singh35c754d2008-02-07 00:14:05 -0800468 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700470 page_cache_get(page);
471 page->mapping = mapping;
472 page->index = offset;
473
Nick Piggin19fd6232008-07-25 19:45:32 -0700474 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700476 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700478 __inc_zone_page_state(page, NR_FILE_PAGES);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700479 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700480 } else {
481 page->mapping = NULL;
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700482 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700483 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700484 page_cache_release(page);
485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800487 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700488 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800489out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 return error;
491}
Nick Piggine2867812008-07-25 19:45:30 -0700492EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400495 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700497 int ret;
498
499 /*
500 * Splice_read and readahead add shmem/tmpfs pages into the page cache
501 * before shmem_readpage has a chance to mark them as SwapBacked: they
502 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
503 * (called in add_to_page_cache) needs to know where they're going too.
504 */
505 if (mapping_cap_swap_backed(mapping))
506 SetPageSwapBacked(page);
507
508 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
509 if (ret == 0) {
510 if (page_is_file_cache(page))
511 lru_cache_add_file(page);
512 else
513 lru_cache_add_active_anon(page);
514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 return ret;
516}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300517EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Paul Jackson44110fe2006-03-24 03:16:04 -0800519#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700520struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800521{
522 if (cpuset_do_page_mem_spread()) {
523 int n = cpuset_mem_spread_node();
Mel Gorman6484eb32009-06-16 15:31:54 -0700524 return alloc_pages_exact_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800525 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700526 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800527}
Nick Piggin2ae88142006-10-28 10:38:23 -0700528EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800529#endif
530
Nick Piggindb376482006-09-25 23:31:24 -0700531static int __sleep_on_page_lock(void *word)
532{
533 io_schedule();
534 return 0;
535}
536
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537/*
538 * In order to wait for pages to become available there must be
539 * waitqueues associated with pages. By using a hash table of
540 * waitqueues where the bucket discipline is to maintain all
541 * waiters on the same queue and wake all when any of the pages
542 * become available, and for the woken contexts to check to be
543 * sure the appropriate page became available, this saves space
544 * at a cost of "thundering herd" phenomena during rare hash
545 * collisions.
546 */
547static wait_queue_head_t *page_waitqueue(struct page *page)
548{
549 const struct zone *zone = page_zone(page);
550
551 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
552}
553
554static inline void wake_up_page(struct page *page, int bit)
555{
556 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
557}
558
Harvey Harrison920c7a52008-02-04 22:29:26 -0800559void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
561 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
562
563 if (test_bit(bit_nr, &page->flags))
564 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
565 TASK_UNINTERRUPTIBLE);
566}
567EXPORT_SYMBOL(wait_on_page_bit);
568
569/**
David Howells385e1ca2009-04-03 16:42:39 +0100570 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700571 * @page: Page defining the wait queue of interest
572 * @waiter: Waiter to add to the queue
David Howells385e1ca2009-04-03 16:42:39 +0100573 *
574 * Add an arbitrary @waiter to the wait queue for the nominated @page.
575 */
576void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
577{
578 wait_queue_head_t *q = page_waitqueue(page);
579 unsigned long flags;
580
581 spin_lock_irqsave(&q->lock, flags);
582 __add_wait_queue(q, waiter);
583 spin_unlock_irqrestore(&q->lock, flags);
584}
585EXPORT_SYMBOL_GPL(add_page_wait_queue);
586
587/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700588 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 * @page: the page
590 *
591 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
592 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
593 * mechananism between PageLocked pages and PageWriteback pages is shared.
594 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
595 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700596 * The mb is necessary to enforce ordering between the clear_bit and the read
597 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800599void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Nick Piggin8413ac92008-10-18 20:26:59 -0700601 VM_BUG_ON(!PageLocked(page));
602 clear_bit_unlock(PG_locked, &page->flags);
603 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 wake_up_page(page, PG_locked);
605}
606EXPORT_SYMBOL(unlock_page);
607
Randy Dunlap485bb992006-06-23 02:03:49 -0700608/**
609 * end_page_writeback - end writeback against a page
610 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 */
612void end_page_writeback(struct page *page)
613{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700614 if (TestClearPageReclaim(page))
615 rotate_reclaimable_page(page);
616
617 if (!test_clear_page_writeback(page))
618 BUG();
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 smp_mb__after_clear_bit();
621 wake_up_page(page, PG_writeback);
622}
623EXPORT_SYMBOL(end_page_writeback);
624
Randy Dunlap485bb992006-06-23 02:03:49 -0700625/**
626 * __lock_page - get a lock on the page, assuming we need to sleep to get it
627 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700629 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
631 * chances are that on the second loop, the block layer's plug list is empty,
632 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
633 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800634void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
636 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
637
638 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
639 TASK_UNINTERRUPTIBLE);
640}
641EXPORT_SYMBOL(__lock_page);
642
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800643int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500644{
645 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
646
647 return __wait_on_bit_lock(page_waitqueue(page), &wait,
648 sync_page_killable, TASK_KILLABLE);
649}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300650EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500651
Randy Dunlap76824862008-03-19 17:00:40 -0700652/**
653 * __lock_page_nosync - get a lock on the page, without calling sync_page()
654 * @page: the page to lock
655 *
Nick Piggindb376482006-09-25 23:31:24 -0700656 * Variant of lock_page that does not require the caller to hold a reference
657 * on the page's mapping.
658 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800659void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700660{
661 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
662 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
663 TASK_UNINTERRUPTIBLE);
664}
665
Randy Dunlap485bb992006-06-23 02:03:49 -0700666/**
667 * find_get_page - find and get a page reference
668 * @mapping: the address_space to search
669 * @offset: the page index
670 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700671 * Is there a pagecache struct page at the given (mapping, offset) tuple?
672 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 */
Nick Piggina60637c2008-07-25 19:45:31 -0700674struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
Nick Piggina60637c2008-07-25 19:45:31 -0700676 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 struct page *page;
678
Nick Piggina60637c2008-07-25 19:45:31 -0700679 rcu_read_lock();
680repeat:
681 page = NULL;
682 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
683 if (pagep) {
684 page = radix_tree_deref_slot(pagep);
685 if (unlikely(!page || page == RADIX_TREE_RETRY))
686 goto repeat;
687
688 if (!page_cache_get_speculative(page))
689 goto repeat;
690
691 /*
692 * Has the page moved?
693 * This is part of the lockless pagecache protocol. See
694 * include/linux/pagemap.h for details.
695 */
696 if (unlikely(page != *pagep)) {
697 page_cache_release(page);
698 goto repeat;
699 }
700 }
701 rcu_read_unlock();
702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return page;
704}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705EXPORT_SYMBOL(find_get_page);
706
Randy Dunlap485bb992006-06-23 02:03:49 -0700707/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700709 * @mapping: the address_space to search
710 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 *
712 * Locates the desired pagecache page, locks it, increments its reference
713 * count and returns its address.
714 *
715 * Returns zero if the page was not present. find_lock_page() may sleep.
716 */
Nick Piggina60637c2008-07-25 19:45:31 -0700717struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
719 struct page *page;
720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700722 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700724 lock_page(page);
725 /* Has the page been truncated? */
726 if (unlikely(page->mapping != mapping)) {
727 unlock_page(page);
728 page_cache_release(page);
729 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
Nick Piggina60637c2008-07-25 19:45:31 -0700731 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 return page;
734}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735EXPORT_SYMBOL(find_lock_page);
736
737/**
738 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700739 * @mapping: the page's address_space
740 * @index: the page's index into the mapping
741 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 *
743 * Locates a page in the pagecache. If the page is not present, a new page
744 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
745 * LRU list. The returned page is locked and has its reference count
746 * incremented.
747 *
748 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
749 * allocation!
750 *
751 * find_or_create_page() returns the desired page's address, or zero on
752 * memory exhaustion.
753 */
754struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700755 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
Nick Piggineb2be182007-10-16 01:24:57 -0700757 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 int err;
759repeat:
760 page = find_lock_page(mapping, index);
761 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700762 page = __page_cache_alloc(gfp_mask);
763 if (!page)
764 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800765 /*
766 * We want a regular kernel memory (not highmem or DMA etc)
767 * allocation for the radix tree nodes, but we need to honour
768 * the context-specific requirements the caller has asked for.
769 * GFP_RECLAIM_MASK collects those requirements.
770 */
771 err = add_to_page_cache_lru(page, mapping, index,
772 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700773 if (unlikely(err)) {
774 page_cache_release(page);
775 page = NULL;
776 if (err == -EEXIST)
777 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 return page;
781}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782EXPORT_SYMBOL(find_or_create_page);
783
784/**
785 * find_get_pages - gang pagecache lookup
786 * @mapping: The address_space to search
787 * @start: The starting page index
788 * @nr_pages: The maximum number of pages
789 * @pages: Where the resulting pages are placed
790 *
791 * find_get_pages() will search for and return a group of up to
792 * @nr_pages pages in the mapping. The pages are placed at @pages.
793 * find_get_pages() takes a reference against the returned pages.
794 *
795 * The search returns a group of mapping-contiguous pages with ascending
796 * indexes. There may be holes in the indices due to not-present pages.
797 *
798 * find_get_pages() returns the number of pages which were found.
799 */
800unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
801 unsigned int nr_pages, struct page **pages)
802{
803 unsigned int i;
804 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700805 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Nick Piggina60637c2008-07-25 19:45:31 -0700807 rcu_read_lock();
808restart:
809 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
810 (void ***)pages, start, nr_pages);
811 ret = 0;
812 for (i = 0; i < nr_found; i++) {
813 struct page *page;
814repeat:
815 page = radix_tree_deref_slot((void **)pages[i]);
816 if (unlikely(!page))
817 continue;
818 /*
819 * this can only trigger if nr_found == 1, making livelock
820 * a non issue.
821 */
822 if (unlikely(page == RADIX_TREE_RETRY))
823 goto restart;
824
825 if (!page_cache_get_speculative(page))
826 goto repeat;
827
828 /* Has the page moved? */
829 if (unlikely(page != *((void **)pages[i]))) {
830 page_cache_release(page);
831 goto repeat;
832 }
833
834 pages[ret] = page;
835 ret++;
836 }
837 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 return ret;
839}
840
Jens Axboeebf43502006-04-27 08:46:01 +0200841/**
842 * find_get_pages_contig - gang contiguous pagecache lookup
843 * @mapping: The address_space to search
844 * @index: The starting page index
845 * @nr_pages: The maximum number of pages
846 * @pages: Where the resulting pages are placed
847 *
848 * find_get_pages_contig() works exactly like find_get_pages(), except
849 * that the returned number of pages are guaranteed to be contiguous.
850 *
851 * find_get_pages_contig() returns the number of pages which were found.
852 */
853unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
854 unsigned int nr_pages, struct page **pages)
855{
856 unsigned int i;
857 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700858 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200859
Nick Piggina60637c2008-07-25 19:45:31 -0700860 rcu_read_lock();
861restart:
862 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
863 (void ***)pages, index, nr_pages);
864 ret = 0;
865 for (i = 0; i < nr_found; i++) {
866 struct page *page;
867repeat:
868 page = radix_tree_deref_slot((void **)pages[i]);
869 if (unlikely(!page))
870 continue;
871 /*
872 * this can only trigger if nr_found == 1, making livelock
873 * a non issue.
874 */
875 if (unlikely(page == RADIX_TREE_RETRY))
876 goto restart;
877
878 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200879 break;
880
Nick Piggina60637c2008-07-25 19:45:31 -0700881 if (!page_cache_get_speculative(page))
882 goto repeat;
883
884 /* Has the page moved? */
885 if (unlikely(page != *((void **)pages[i]))) {
886 page_cache_release(page);
887 goto repeat;
888 }
889
890 pages[ret] = page;
891 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200892 index++;
893 }
Nick Piggina60637c2008-07-25 19:45:31 -0700894 rcu_read_unlock();
895 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200896}
David Howellsef71c152007-05-09 02:33:44 -0700897EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200898
Randy Dunlap485bb992006-06-23 02:03:49 -0700899/**
900 * find_get_pages_tag - find and return pages that match @tag
901 * @mapping: the address_space to search
902 * @index: the starting page index
903 * @tag: the tag index
904 * @nr_pages: the maximum number of pages
905 * @pages: where the resulting pages are placed
906 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700908 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 */
910unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
911 int tag, unsigned int nr_pages, struct page **pages)
912{
913 unsigned int i;
914 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700915 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Nick Piggina60637c2008-07-25 19:45:31 -0700917 rcu_read_lock();
918restart:
919 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
920 (void ***)pages, *index, nr_pages, tag);
921 ret = 0;
922 for (i = 0; i < nr_found; i++) {
923 struct page *page;
924repeat:
925 page = radix_tree_deref_slot((void **)pages[i]);
926 if (unlikely(!page))
927 continue;
928 /*
929 * this can only trigger if nr_found == 1, making livelock
930 * a non issue.
931 */
932 if (unlikely(page == RADIX_TREE_RETRY))
933 goto restart;
934
935 if (!page_cache_get_speculative(page))
936 goto repeat;
937
938 /* Has the page moved? */
939 if (unlikely(page != *((void **)pages[i]))) {
940 page_cache_release(page);
941 goto repeat;
942 }
943
944 pages[ret] = page;
945 ret++;
946 }
947 rcu_read_unlock();
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (ret)
950 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 return ret;
953}
David Howellsef71c152007-05-09 02:33:44 -0700954EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Randy Dunlap485bb992006-06-23 02:03:49 -0700956/**
957 * grab_cache_page_nowait - returns locked page at given index in given cache
958 * @mapping: target address_space
959 * @index: the page index
960 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800961 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 * This is intended for speculative data generators, where the data can
963 * be regenerated if the page couldn't be grabbed. This routine should
964 * be safe to call while holding the lock for another page.
965 *
966 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
967 * and deadlock against the caller's locked page.
968 */
969struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700970grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
972 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
974 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200975 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 return page;
977 page_cache_release(page);
978 return NULL;
979 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700980 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800981 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 page_cache_release(page);
983 page = NULL;
984 }
985 return page;
986}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987EXPORT_SYMBOL(grab_cache_page_nowait);
988
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700989/*
990 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
991 * a _large_ part of the i/o request. Imagine the worst scenario:
992 *
993 * ---R__________________________________________B__________
994 * ^ reading here ^ bad block(assume 4k)
995 *
996 * read(R) => miss => readahead(R...B) => media error => frustrating retries
997 * => failing the whole request => read(R) => read(R+1) =>
998 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
999 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1000 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1001 *
1002 * It is going insane. Fix it by quickly scaling down the readahead size.
1003 */
1004static void shrink_readahead_size_eio(struct file *filp,
1005 struct file_ra_state *ra)
1006{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001007 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001008}
1009
Randy Dunlap485bb992006-06-23 02:03:49 -07001010/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001011 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001012 * @filp: the file to read
1013 * @ppos: current file position
1014 * @desc: read_descriptor
1015 * @actor: read method
1016 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001018 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 *
1020 * This is really ugly. But the goto's actually try to clarify some
1021 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001023static void do_generic_file_read(struct file *filp, loff_t *ppos,
1024 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001026 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001028 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001029 pgoff_t index;
1030 pgoff_t last_index;
1031 pgoff_t prev_index;
1032 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001033 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001037 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1038 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1040 offset = *ppos & ~PAGE_CACHE_MASK;
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 for (;;) {
1043 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001044 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001045 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 unsigned long nr, ret;
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049find_page:
1050 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001051 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001052 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001053 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001054 index, last_index - index);
1055 page = find_get_page(mapping, index);
1056 if (unlikely(page == NULL))
1057 goto no_cached_page;
1058 }
1059 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001060 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001061 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001062 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001064 if (!PageUptodate(page)) {
1065 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1066 !mapping->a_ops->is_partially_uptodate)
1067 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001068 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001069 goto page_not_up_to_date;
1070 if (!mapping->a_ops->is_partially_uptodate(page,
1071 desc, offset))
1072 goto page_not_up_to_date_locked;
1073 unlock_page(page);
1074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001076 /*
1077 * i_size must be checked after we know the page is Uptodate.
1078 *
1079 * Checking i_size after the check allows us to calculate
1080 * the correct value for "nr", which means the zero-filled
1081 * part of the page is not copied back to userspace (unless
1082 * another truncate extends the file - this is desired though).
1083 */
1084
1085 isize = i_size_read(inode);
1086 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1087 if (unlikely(!isize || index > end_index)) {
1088 page_cache_release(page);
1089 goto out;
1090 }
1091
1092 /* nr is the maximum number of bytes to copy from this page */
1093 nr = PAGE_CACHE_SIZE;
1094 if (index == end_index) {
1095 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1096 if (nr <= offset) {
1097 page_cache_release(page);
1098 goto out;
1099 }
1100 }
1101 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 /* If users can be writing to this page using arbitrary
1104 * virtual addresses, take care about potential aliasing
1105 * before reading the page on the kernel side.
1106 */
1107 if (mapping_writably_mapped(mapping))
1108 flush_dcache_page(page);
1109
1110 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001111 * When a sequential read accesses a page several times,
1112 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 */
Jan Karaec0f1632007-05-06 14:49:25 -07001114 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 mark_page_accessed(page);
1116 prev_index = index;
1117
1118 /*
1119 * Ok, we have the page, and it's up-to-date, so
1120 * now we can copy it to user space...
1121 *
1122 * The actor routine returns how many bytes were actually used..
1123 * NOTE! This may not be the same as how much of a user buffer
1124 * we filled up (we may be padding etc), so we can only update
1125 * "pos" here (the actor routine has to update the user buffer
1126 * pointers and the remaining count).
1127 */
1128 ret = actor(desc, page, offset, nr);
1129 offset += ret;
1130 index += offset >> PAGE_CACHE_SHIFT;
1131 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001132 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
1134 page_cache_release(page);
1135 if (ret == nr && desc->count)
1136 continue;
1137 goto out;
1138
1139page_not_up_to_date:
1140 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001141 error = lock_page_killable(page);
1142 if (unlikely(error))
1143 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001145page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001146 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 if (!page->mapping) {
1148 unlock_page(page);
1149 page_cache_release(page);
1150 continue;
1151 }
1152
1153 /* Did somebody else fill it already? */
1154 if (PageUptodate(page)) {
1155 unlock_page(page);
1156 goto page_ok;
1157 }
1158
1159readpage:
1160 /* Start the actual read. The read will unlock the page. */
1161 error = mapping->a_ops->readpage(filp, page);
1162
Zach Brown994fc28c2005-12-15 14:28:17 -08001163 if (unlikely(error)) {
1164 if (error == AOP_TRUNCATED_PAGE) {
1165 page_cache_release(page);
1166 goto find_page;
1167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
1171 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001172 error = lock_page_killable(page);
1173 if (unlikely(error))
1174 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (!PageUptodate(page)) {
1176 if (page->mapping == NULL) {
1177 /*
1178 * invalidate_inode_pages got it
1179 */
1180 unlock_page(page);
1181 page_cache_release(page);
1182 goto find_page;
1183 }
1184 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001185 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001186 error = -EIO;
1187 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 }
1189 unlock_page(page);
1190 }
1191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 goto page_ok;
1193
1194readpage_error:
1195 /* UHHUH! A synchronous read error occurred. Report it */
1196 desc->error = error;
1197 page_cache_release(page);
1198 goto out;
1199
1200no_cached_page:
1201 /*
1202 * Ok, it wasn't cached, so we need to create a new
1203 * page..
1204 */
Nick Piggineb2be182007-10-16 01:24:57 -07001205 page = page_cache_alloc_cold(mapping);
1206 if (!page) {
1207 desc->error = -ENOMEM;
1208 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 }
Nick Piggineb2be182007-10-16 01:24:57 -07001210 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 index, GFP_KERNEL);
1212 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001213 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 if (error == -EEXIST)
1215 goto find_page;
1216 desc->error = error;
1217 goto out;
1218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 goto readpage;
1220 }
1221
1222out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001223 ra->prev_pos = prev_index;
1224 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1225 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001227 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001228 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
1231int file_read_actor(read_descriptor_t *desc, struct page *page,
1232 unsigned long offset, unsigned long size)
1233{
1234 char *kaddr;
1235 unsigned long left, count = desc->count;
1236
1237 if (size > count)
1238 size = count;
1239
1240 /*
1241 * Faults on the destination of a read are common, so do it before
1242 * taking the kmap.
1243 */
1244 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1245 kaddr = kmap_atomic(page, KM_USER0);
1246 left = __copy_to_user_inatomic(desc->arg.buf,
1247 kaddr + offset, size);
1248 kunmap_atomic(kaddr, KM_USER0);
1249 if (left == 0)
1250 goto success;
1251 }
1252
1253 /* Do it the slow way */
1254 kaddr = kmap(page);
1255 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1256 kunmap(page);
1257
1258 if (left) {
1259 size -= left;
1260 desc->error = -EFAULT;
1261 }
1262success:
1263 desc->count = count - size;
1264 desc->written += size;
1265 desc->arg.buf += size;
1266 return size;
1267}
1268
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001269/*
1270 * Performs necessary checks before doing a write
1271 * @iov: io vector request
1272 * @nr_segs: number of segments in the iovec
1273 * @count: number of bytes to write
1274 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1275 *
1276 * Adjust number of segments and amount of bytes to write (nr_segs should be
1277 * properly initialized first). Returns appropriate error code that caller
1278 * should return or zero in case that write should be allowed.
1279 */
1280int generic_segment_checks(const struct iovec *iov,
1281 unsigned long *nr_segs, size_t *count, int access_flags)
1282{
1283 unsigned long seg;
1284 size_t cnt = 0;
1285 for (seg = 0; seg < *nr_segs; seg++) {
1286 const struct iovec *iv = &iov[seg];
1287
1288 /*
1289 * If any segment has a negative length, or the cumulative
1290 * length ever wraps negative then return -EINVAL.
1291 */
1292 cnt += iv->iov_len;
1293 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1294 return -EINVAL;
1295 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1296 continue;
1297 if (seg == 0)
1298 return -EFAULT;
1299 *nr_segs = seg;
1300 cnt -= iv->iov_len; /* This segment is no good */
1301 break;
1302 }
1303 *count = cnt;
1304 return 0;
1305}
1306EXPORT_SYMBOL(generic_segment_checks);
1307
Randy Dunlap485bb992006-06-23 02:03:49 -07001308/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001309 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001310 * @iocb: kernel I/O control block
1311 * @iov: io vector request
1312 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001313 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001314 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 * This is the "read()" routine for all filesystems
1316 * that can use the page cache directly.
1317 */
1318ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001319generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1320 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
1322 struct file *filp = iocb->ki_filp;
1323 ssize_t retval;
1324 unsigned long seg;
1325 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001326 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
1328 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001329 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1330 if (retval)
1331 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1334 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001335 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 struct address_space *mapping;
1337 struct inode *inode;
1338
1339 mapping = filp->f_mapping;
1340 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 if (!count)
1342 goto out; /* skip atime */
1343 size = i_size_read(inode);
1344 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001345 retval = filemap_write_and_wait_range(mapping, pos,
1346 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001347 if (!retval) {
1348 retval = mapping->a_ops->direct_IO(READ, iocb,
1349 iov, pos, nr_segs);
1350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 if (retval > 0)
1352 *ppos = pos + retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001353 if (retval) {
1354 file_accessed(filp);
1355 goto out;
1356 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001357 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 }
1359
Hugh Dickins11fa9772008-07-23 21:27:34 -07001360 for (seg = 0; seg < nr_segs; seg++) {
1361 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
Hugh Dickins11fa9772008-07-23 21:27:34 -07001363 desc.written = 0;
1364 desc.arg.buf = iov[seg].iov_base;
1365 desc.count = iov[seg].iov_len;
1366 if (desc.count == 0)
1367 continue;
1368 desc.error = 0;
1369 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1370 retval += desc.written;
1371 if (desc.error) {
1372 retval = retval ?: desc.error;
1373 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001375 if (desc.count > 0)
1376 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
1378out:
1379 return retval;
1380}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381EXPORT_SYMBOL(generic_file_aio_read);
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383static ssize_t
1384do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001385 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386{
1387 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1388 return -EINVAL;
1389
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001390 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 return 0;
1392}
1393
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001394SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
1396 ssize_t ret;
1397 struct file *file;
1398
1399 ret = -EBADF;
1400 file = fget(fd);
1401 if (file) {
1402 if (file->f_mode & FMODE_READ) {
1403 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001404 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1405 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 unsigned long len = end - start + 1;
1407 ret = do_readahead(mapping, file, start, len);
1408 }
1409 fput(file);
1410 }
1411 return ret;
1412}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001413#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1414asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1415{
1416 return SYSC_readahead((int) fd, offset, (size_t) count);
1417}
1418SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1419#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
1421#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001422/**
1423 * page_cache_read - adds requested page to the page cache if not already there
1424 * @file: file to read
1425 * @offset: page index
1426 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 * This adds the requested page to the page cache if it isn't already there,
1428 * and schedules an I/O to read in its contents from disk.
1429 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001430static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431{
1432 struct address_space *mapping = file->f_mapping;
1433 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001434 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Zach Brown994fc28c2005-12-15 14:28:17 -08001436 do {
1437 page = page_cache_alloc_cold(mapping);
1438 if (!page)
1439 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Zach Brown994fc28c2005-12-15 14:28:17 -08001441 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1442 if (ret == 0)
1443 ret = mapping->a_ops->readpage(file, page);
1444 else if (ret == -EEXIST)
1445 ret = 0; /* losing race to add is OK */
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Zach Brown994fc28c2005-12-15 14:28:17 -08001449 } while (ret == AOP_TRUNCATED_PAGE);
1450
1451 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
1454#define MMAP_LOTSAMISS (100)
1455
Linus Torvaldsef00e082009-06-16 15:31:25 -07001456/*
1457 * Synchronous readahead happens when we don't even find
1458 * a page in the page cache at all.
1459 */
1460static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1461 struct file_ra_state *ra,
1462 struct file *file,
1463 pgoff_t offset)
1464{
1465 unsigned long ra_pages;
1466 struct address_space *mapping = file->f_mapping;
1467
1468 /* If we don't want any read-ahead, don't bother */
1469 if (VM_RandomReadHint(vma))
1470 return;
1471
Wu Fengguang70ac23c2009-06-16 15:31:28 -07001472 if (VM_SequentialReadHint(vma) ||
1473 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001474 page_cache_sync_readahead(mapping, ra, file, offset,
1475 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001476 return;
1477 }
1478
1479 if (ra->mmap_miss < INT_MAX)
1480 ra->mmap_miss++;
1481
1482 /*
1483 * Do we miss much more than hit in this file? If so,
1484 * stop bothering with read-ahead. It will only hurt.
1485 */
1486 if (ra->mmap_miss > MMAP_LOTSAMISS)
1487 return;
1488
Wu Fengguangd30a1102009-06-16 15:31:30 -07001489 /*
1490 * mmap read-around
1491 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001492 ra_pages = max_sane_readahead(ra->ra_pages);
1493 if (ra_pages) {
Wu Fengguangd30a1102009-06-16 15:31:30 -07001494 ra->start = max_t(long, 0, offset - ra_pages/2);
1495 ra->size = ra_pages;
1496 ra->async_size = 0;
1497 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001498 }
1499}
1500
1501/*
1502 * Asynchronous readahead happens when we find the page and PG_readahead,
1503 * so we want to possibly extend the readahead further..
1504 */
1505static void do_async_mmap_readahead(struct vm_area_struct *vma,
1506 struct file_ra_state *ra,
1507 struct file *file,
1508 struct page *page,
1509 pgoff_t offset)
1510{
1511 struct address_space *mapping = file->f_mapping;
1512
1513 /* If we don't want any read-ahead, don't bother */
1514 if (VM_RandomReadHint(vma))
1515 return;
1516 if (ra->mmap_miss > 0)
1517 ra->mmap_miss--;
1518 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001519 page_cache_async_readahead(mapping, ra, file,
1520 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001521}
1522
Randy Dunlap485bb992006-06-23 02:03:49 -07001523/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001524 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001525 * @vma: vma in which the fault was taken
1526 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001527 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001528 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 * mapped memory region to read in file data during a page fault.
1530 *
1531 * The goto's are kind of ugly, but this streamlines the normal case of having
1532 * it in the page cache, and handles the special cases reasonably without
1533 * having a lot of duplicated code.
1534 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001535int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
1537 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001538 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 struct address_space *mapping = file->f_mapping;
1540 struct file_ra_state *ra = &file->f_ra;
1541 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001542 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001544 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001545 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001548 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001549 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 * Do we have something in the page cache already?
1553 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001554 page = find_get_page(mapping, offset);
1555 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001557 * We found the page, so try async readahead before
1558 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001560 do_async_mmap_readahead(vma, ra, file, page, offset);
1561 lock_page(page);
1562
1563 /* Did it get truncated? */
1564 if (unlikely(page->mapping != mapping)) {
1565 unlock_page(page);
1566 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 goto no_cached_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 }
Linus Torvaldsef00e082009-06-16 15:31:25 -07001569 } else {
1570 /* No page in the page cache at all */
1571 do_sync_mmap_readahead(vma, ra, file, offset);
1572 count_vm_event(PGMAJFAULT);
1573 ret = VM_FAULT_MAJOR;
1574retry_find:
1575 page = find_lock_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 if (!page)
1577 goto no_cached_page;
1578 }
1579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001581 * We have a locked page in the page cache, now we need to check
1582 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 */
Nick Piggind00806b2007-07-19 01:46:57 -07001584 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 goto page_not_uptodate;
1586
Linus Torvaldsef00e082009-06-16 15:31:25 -07001587 /*
1588 * Found the page and have a reference on it.
1589 * We must recheck i_size under page lock.
1590 */
Nick Piggind00806b2007-07-19 01:46:57 -07001591 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001592 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001593 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001594 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001595 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001596 }
1597
Linus Torvaldsef00e082009-06-16 15:31:25 -07001598 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001599 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001600 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602no_cached_page:
1603 /*
1604 * We're only likely to ever get here if MADV_RANDOM is in
1605 * effect.
1606 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001607 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
1609 /*
1610 * The page we want has now been added to the page cache.
1611 * In the unlikely event that someone removed it in the
1612 * meantime, we'll just come back here and read it again.
1613 */
1614 if (error >= 0)
1615 goto retry_find;
1616
1617 /*
1618 * An error return from page_cache_read can result if the
1619 * system is low on memory, or a problem occurs while trying
1620 * to schedule I/O.
1621 */
1622 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001623 return VM_FAULT_OOM;
1624 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
1626page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 /*
1628 * Umm, take care of errors if the page isn't up-to-date.
1629 * Try to re-read it _once_. We do this synchronously,
1630 * because there really aren't any performance issues here
1631 * and we need to check for errors.
1632 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001634 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001635 if (!error) {
1636 wait_on_page_locked(page);
1637 if (!PageUptodate(page))
1638 error = -EIO;
1639 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001641
1642 if (!error || error == AOP_TRUNCATED_PAGE)
1643 goto retry_find;
1644
1645 /* Things didn't work out. Return zero to tell the mm layer so. */
1646 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001647 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001648}
1649EXPORT_SYMBOL(filemap_fault);
1650
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001652 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653};
1654
1655/* This is used for a general mmap of a disk file */
1656
1657int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1658{
1659 struct address_space *mapping = file->f_mapping;
1660
1661 if (!mapping->a_ops->readpage)
1662 return -ENOEXEC;
1663 file_accessed(file);
1664 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001665 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 return 0;
1667}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669/*
1670 * This is for filesystems which do not implement ->writepage.
1671 */
1672int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1673{
1674 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1675 return -EINVAL;
1676 return generic_file_mmap(file, vma);
1677}
1678#else
1679int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1680{
1681 return -ENOSYS;
1682}
1683int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1684{
1685 return -ENOSYS;
1686}
1687#endif /* CONFIG_MMU */
1688
1689EXPORT_SYMBOL(generic_file_mmap);
1690EXPORT_SYMBOL(generic_file_readonly_mmap);
1691
Nick Piggin6fe69002007-05-06 14:49:04 -07001692static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001693 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 int (*filler)(void *,struct page*),
1695 void *data)
1696{
Nick Piggineb2be182007-10-16 01:24:57 -07001697 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 int err;
1699repeat:
1700 page = find_get_page(mapping, index);
1701 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001702 page = page_cache_alloc_cold(mapping);
1703 if (!page)
1704 return ERR_PTR(-ENOMEM);
1705 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1706 if (unlikely(err)) {
1707 page_cache_release(page);
1708 if (err == -EEXIST)
1709 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return ERR_PTR(err);
1712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 err = filler(data, page);
1714 if (err < 0) {
1715 page_cache_release(page);
1716 page = ERR_PTR(err);
1717 }
1718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 return page;
1720}
1721
Randy Dunlap76824862008-03-19 17:00:40 -07001722/**
1723 * read_cache_page_async - read into page cache, fill it if needed
1724 * @mapping: the page's address_space
1725 * @index: the page index
1726 * @filler: function to perform the read
1727 * @data: destination for read data
1728 *
Nick Piggin6fe69002007-05-06 14:49:04 -07001729 * Same as read_cache_page, but don't wait for page to become unlocked
1730 * after submitting it to the filler.
Randy Dunlap76824862008-03-19 17:00:40 -07001731 *
1732 * Read into the page cache. If a page already exists, and PageUptodate() is
1733 * not set, try to fill the page but don't wait for it to become unlocked.
1734 *
1735 * If the page does not get brought uptodate, return -EIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001737struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001738 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 int (*filler)(void *,struct page*),
1740 void *data)
1741{
1742 struct page *page;
1743 int err;
1744
1745retry:
1746 page = __read_cache_page(mapping, index, filler, data);
1747 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001748 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 if (PageUptodate(page))
1750 goto out;
1751
1752 lock_page(page);
1753 if (!page->mapping) {
1754 unlock_page(page);
1755 page_cache_release(page);
1756 goto retry;
1757 }
1758 if (PageUptodate(page)) {
1759 unlock_page(page);
1760 goto out;
1761 }
1762 err = filler(data, page);
1763 if (err < 0) {
1764 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001765 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
David Howellsc855ff32007-05-09 13:42:20 +01001767out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001768 mark_page_accessed(page);
1769 return page;
1770}
1771EXPORT_SYMBOL(read_cache_page_async);
1772
1773/**
1774 * read_cache_page - read into page cache, fill it if needed
1775 * @mapping: the page's address_space
1776 * @index: the page index
1777 * @filler: function to perform the read
1778 * @data: destination for read data
1779 *
1780 * Read into the page cache. If a page already exists, and PageUptodate() is
1781 * not set, try to fill the page then wait for it to become unlocked.
1782 *
1783 * If the page does not get brought uptodate, return -EIO.
1784 */
1785struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001786 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001787 int (*filler)(void *,struct page*),
1788 void *data)
1789{
1790 struct page *page;
1791
1792 page = read_cache_page_async(mapping, index, filler, data);
1793 if (IS_ERR(page))
1794 goto out;
1795 wait_on_page_locked(page);
1796 if (!PageUptodate(page)) {
1797 page_cache_release(page);
1798 page = ERR_PTR(-EIO);
1799 }
1800 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 return page;
1802}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803EXPORT_SYMBOL(read_cache_page);
1804
1805/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 * The logic we want is
1807 *
1808 * if suid or (sgid and xgrp)
1809 * remove privs
1810 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001811int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812{
1813 mode_t mode = dentry->d_inode->i_mode;
1814 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
1816 /* suid always must be killed */
1817 if (unlikely(mode & S_ISUID))
1818 kill = ATTR_KILL_SUID;
1819
1820 /*
1821 * sgid without any exec bits is just a mandatory locking mark; leave
1822 * it alone. If some exec bits are set, it's a real sgid; kill it.
1823 */
1824 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1825 kill |= ATTR_KILL_SGID;
1826
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001827 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001828 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Jens Axboe01de85e2006-10-17 19:50:36 +02001830 return 0;
1831}
Mark Fashehd23a1472006-10-17 17:05:18 -07001832EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001833
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001834static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001835{
1836 struct iattr newattrs;
1837
1838 newattrs.ia_valid = ATTR_FORCE | kill;
1839 return notify_change(dentry, &newattrs);
1840}
1841
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001842int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001843{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001844 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001845 int killsuid = should_remove_suid(dentry);
1846 int killpriv = security_inode_need_killpriv(dentry);
1847 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001848
Serge E. Hallynb5376772007-10-16 23:31:36 -07001849 if (killpriv < 0)
1850 return killpriv;
1851 if (killpriv)
1852 error = security_inode_killpriv(dentry);
1853 if (!error && killsuid)
1854 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001855
Serge E. Hallynb5376772007-10-16 23:31:36 -07001856 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001858EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
Nick Piggin2f718ff2007-10-16 01:24:59 -07001860static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 const struct iovec *iov, size_t base, size_t bytes)
1862{
Ingo Molnarf1800532009-03-02 11:00:57 +01001863 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
1865 while (bytes) {
1866 char __user *buf = iov->iov_base + base;
1867 int copy = min(bytes, iov->iov_len - base);
1868
1869 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001870 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 copied += copy;
1872 bytes -= copy;
1873 vaddr += copy;
1874 iov++;
1875
NeilBrown01408c42006-06-25 05:47:58 -07001876 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 }
1879 return copied - left;
1880}
1881
1882/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001883 * Copy as much as we can into the page and return the number of bytes which
1884 * were sucessfully copied. If a fault is encountered then return the number of
1885 * bytes which were copied.
1886 */
1887size_t iov_iter_copy_from_user_atomic(struct page *page,
1888 struct iov_iter *i, unsigned long offset, size_t bytes)
1889{
1890 char *kaddr;
1891 size_t copied;
1892
1893 BUG_ON(!in_atomic());
1894 kaddr = kmap_atomic(page, KM_USER0);
1895 if (likely(i->nr_segs == 1)) {
1896 int left;
1897 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001898 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001899 copied = bytes - left;
1900 } else {
1901 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1902 i->iov, i->iov_offset, bytes);
1903 }
1904 kunmap_atomic(kaddr, KM_USER0);
1905
1906 return copied;
1907}
Nick Piggin89e10782007-10-16 01:25:07 -07001908EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001909
1910/*
1911 * This has the same sideeffects and return value as
1912 * iov_iter_copy_from_user_atomic().
1913 * The difference is that it attempts to resolve faults.
1914 * Page must not be locked.
1915 */
1916size_t iov_iter_copy_from_user(struct page *page,
1917 struct iov_iter *i, unsigned long offset, size_t bytes)
1918{
1919 char *kaddr;
1920 size_t copied;
1921
1922 kaddr = kmap(page);
1923 if (likely(i->nr_segs == 1)) {
1924 int left;
1925 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001926 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001927 copied = bytes - left;
1928 } else {
1929 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1930 i->iov, i->iov_offset, bytes);
1931 }
1932 kunmap(page);
1933 return copied;
1934}
Nick Piggin89e10782007-10-16 01:25:07 -07001935EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001936
Nick Pigginf7009262008-03-10 11:43:59 -07001937void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001938{
Nick Pigginf7009262008-03-10 11:43:59 -07001939 BUG_ON(i->count < bytes);
1940
Nick Piggin2f718ff2007-10-16 01:24:59 -07001941 if (likely(i->nr_segs == 1)) {
1942 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001943 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001944 } else {
1945 const struct iovec *iov = i->iov;
1946 size_t base = i->iov_offset;
1947
Nick Piggin124d3b72008-02-02 15:01:17 +01001948 /*
1949 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001950 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001951 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001952 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001953 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001954
Nick Pigginf7009262008-03-10 11:43:59 -07001955 copy = min(bytes, iov->iov_len - base);
1956 BUG_ON(!i->count || i->count < copy);
1957 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001958 bytes -= copy;
1959 base += copy;
1960 if (iov->iov_len == base) {
1961 iov++;
1962 base = 0;
1963 }
1964 }
1965 i->iov = iov;
1966 i->iov_offset = base;
1967 }
1968}
Nick Piggin89e10782007-10-16 01:25:07 -07001969EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001970
Nick Pigginafddba42007-10-16 01:25:01 -07001971/*
1972 * Fault in the first iovec of the given iov_iter, to a maximum length
1973 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1974 * accessed (ie. because it is an invalid address).
1975 *
1976 * writev-intensive code may want this to prefault several iovecs -- that
1977 * would be possible (callers must not rely on the fact that _only_ the
1978 * first iovec will be faulted with the current implementation).
1979 */
1980int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001981{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001982 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001983 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1984 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001985}
Nick Piggin89e10782007-10-16 01:25:07 -07001986EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001987
1988/*
1989 * Return the count of just the current iov_iter segment.
1990 */
1991size_t iov_iter_single_seg_count(struct iov_iter *i)
1992{
1993 const struct iovec *iov = i->iov;
1994 if (i->nr_segs == 1)
1995 return i->count;
1996 else
1997 return min(i->count, iov->iov_len - i->iov_offset);
1998}
Nick Piggin89e10782007-10-16 01:25:07 -07001999EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002000
2001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 * Performs necessary checks before doing a write
2003 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002004 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 * Returns appropriate error code that caller should return or
2006 * zero in case that write should be allowed.
2007 */
2008inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2009{
2010 struct inode *inode = file->f_mapping->host;
2011 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2012
2013 if (unlikely(*pos < 0))
2014 return -EINVAL;
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 if (!isblk) {
2017 /* FIXME: this is for backwards compatibility with 2.4 */
2018 if (file->f_flags & O_APPEND)
2019 *pos = i_size_read(inode);
2020
2021 if (limit != RLIM_INFINITY) {
2022 if (*pos >= limit) {
2023 send_sig(SIGXFSZ, current, 0);
2024 return -EFBIG;
2025 }
2026 if (*count > limit - (typeof(limit))*pos) {
2027 *count = limit - (typeof(limit))*pos;
2028 }
2029 }
2030 }
2031
2032 /*
2033 * LFS rule
2034 */
2035 if (unlikely(*pos + *count > MAX_NON_LFS &&
2036 !(file->f_flags & O_LARGEFILE))) {
2037 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 return -EFBIG;
2039 }
2040 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2041 *count = MAX_NON_LFS - (unsigned long)*pos;
2042 }
2043 }
2044
2045 /*
2046 * Are we about to exceed the fs block limit ?
2047 *
2048 * If we have written data it becomes a short write. If we have
2049 * exceeded without writing data we send a signal and return EFBIG.
2050 * Linus frestrict idea will clean these up nicely..
2051 */
2052 if (likely(!isblk)) {
2053 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2054 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return -EFBIG;
2056 }
2057 /* zero-length writes at ->s_maxbytes are OK */
2058 }
2059
2060 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2061 *count = inode->i_sb->s_maxbytes - *pos;
2062 } else {
David Howells93614012006-09-30 20:45:40 +02002063#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 loff_t isize;
2065 if (bdev_read_only(I_BDEV(inode)))
2066 return -EPERM;
2067 isize = i_size_read(inode);
2068 if (*pos >= isize) {
2069 if (*count || *pos > isize)
2070 return -ENOSPC;
2071 }
2072
2073 if (*pos + *count > isize)
2074 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002075#else
2076 return -EPERM;
2077#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 }
2079 return 0;
2080}
2081EXPORT_SYMBOL(generic_write_checks);
2082
Nick Pigginafddba42007-10-16 01:25:01 -07002083int pagecache_write_begin(struct file *file, struct address_space *mapping,
2084 loff_t pos, unsigned len, unsigned flags,
2085 struct page **pagep, void **fsdata)
2086{
2087 const struct address_space_operations *aops = mapping->a_ops;
2088
Nick Piggin4e02ed42008-10-29 14:00:55 -07002089 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002090 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002091}
2092EXPORT_SYMBOL(pagecache_write_begin);
2093
2094int pagecache_write_end(struct file *file, struct address_space *mapping,
2095 loff_t pos, unsigned len, unsigned copied,
2096 struct page *page, void *fsdata)
2097{
2098 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002099
Nick Piggin4e02ed42008-10-29 14:00:55 -07002100 mark_page_accessed(page);
2101 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002102}
2103EXPORT_SYMBOL(pagecache_write_end);
2104
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105ssize_t
2106generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2107 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2108 size_t count, size_t ocount)
2109{
2110 struct file *file = iocb->ki_filp;
2111 struct address_space *mapping = file->f_mapping;
2112 struct inode *inode = mapping->host;
2113 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002114 size_t write_len;
2115 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117 if (count != ocount)
2118 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2119
Christoph Hellwiga969e902008-07-23 21:27:04 -07002120 write_len = iov_length(iov, *nr_segs);
2121 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002122
Nick Piggin48b47c52009-01-06 14:40:22 -08002123 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002124 if (written)
2125 goto out;
2126
2127 /*
2128 * After a write we want buffered reads to be sure to go to disk to get
2129 * the new data. We invalidate clean cached page from the region we're
2130 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002131 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002132 */
2133 if (mapping->nrpages) {
2134 written = invalidate_inode_pages2_range(mapping,
2135 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002136 /*
2137 * If a page can not be invalidated, return 0 to fall back
2138 * to buffered write.
2139 */
2140 if (written) {
2141 if (written == -EBUSY)
2142 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002143 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002144 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002145 }
2146
2147 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2148
2149 /*
2150 * Finally, try again to invalidate clean pages which might have been
2151 * cached by non-direct readahead, or faulted in by get_user_pages()
2152 * if the source of the write was an mmap'ed region of the file
2153 * we're writing. Either one is a pretty crazy thing to do,
2154 * so we don't support it 100%. If this invalidation
2155 * fails, tough, the write still worked...
2156 */
2157 if (mapping->nrpages) {
2158 invalidate_inode_pages2_range(mapping,
2159 pos >> PAGE_CACHE_SHIFT, end);
2160 }
2161
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 if (written > 0) {
2163 loff_t end = pos + written;
2164 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2165 i_size_write(inode, end);
2166 mark_inode_dirty(inode);
2167 }
2168 *ppos = end;
2169 }
2170
2171 /*
2172 * Sync the fs metadata but not the minor inode changes and
2173 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002174 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002175 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 */
Christoph Hellwiga969e902008-07-23 21:27:04 -07002177out:
Zach Brown8459d862006-12-10 02:21:05 -08002178 if ((written >= 0 || written == -EIOCBQUEUED) &&
2179 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002180 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2181 if (err < 0)
2182 written = err;
2183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 return written;
2185}
2186EXPORT_SYMBOL(generic_file_direct_write);
2187
Nick Piggineb2be182007-10-16 01:24:57 -07002188/*
2189 * Find or create a page at the given pagecache position. Return the locked
2190 * page. This function is specifically for buffered writes.
2191 */
Nick Piggin54566b22009-01-04 12:00:53 -08002192struct page *grab_cache_page_write_begin(struct address_space *mapping,
2193 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002194{
2195 int status;
2196 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002197 gfp_t gfp_notmask = 0;
2198 if (flags & AOP_FLAG_NOFS)
2199 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002200repeat:
2201 page = find_lock_page(mapping, index);
2202 if (likely(page))
2203 return page;
2204
Nick Piggin54566b22009-01-04 12:00:53 -08002205 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002206 if (!page)
2207 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002208 status = add_to_page_cache_lru(page, mapping, index,
2209 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002210 if (unlikely(status)) {
2211 page_cache_release(page);
2212 if (status == -EEXIST)
2213 goto repeat;
2214 return NULL;
2215 }
2216 return page;
2217}
Nick Piggin54566b22009-01-04 12:00:53 -08002218EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002219
Nick Pigginafddba42007-10-16 01:25:01 -07002220static ssize_t generic_perform_write(struct file *file,
2221 struct iov_iter *i, loff_t pos)
2222{
2223 struct address_space *mapping = file->f_mapping;
2224 const struct address_space_operations *a_ops = mapping->a_ops;
2225 long status = 0;
2226 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002227 unsigned int flags = 0;
2228
2229 /*
2230 * Copies from kernel address space cannot fail (NFSD is a big user).
2231 */
2232 if (segment_eq(get_fs(), KERNEL_DS))
2233 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002234
2235 do {
2236 struct page *page;
2237 pgoff_t index; /* Pagecache index for current page */
2238 unsigned long offset; /* Offset into pagecache page */
2239 unsigned long bytes; /* Bytes to write to page */
2240 size_t copied; /* Bytes copied from user */
2241 void *fsdata;
2242
2243 offset = (pos & (PAGE_CACHE_SIZE - 1));
2244 index = pos >> PAGE_CACHE_SHIFT;
2245 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2246 iov_iter_count(i));
2247
2248again:
2249
2250 /*
2251 * Bring in the user page that we will copy from _first_.
2252 * Otherwise there's a nasty deadlock on copying from the
2253 * same page as we're writing to, without it being marked
2254 * up-to-date.
2255 *
2256 * Not only is this an optimisation, but it is also required
2257 * to check that the address is actually valid, when atomic
2258 * usercopies are used, below.
2259 */
2260 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2261 status = -EFAULT;
2262 break;
2263 }
2264
Nick Piggin674b8922007-10-16 01:25:03 -07002265 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002266 &page, &fsdata);
2267 if (unlikely(status))
2268 break;
2269
2270 pagefault_disable();
2271 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2272 pagefault_enable();
2273 flush_dcache_page(page);
2274
Josef Bacikc8236db2009-07-05 12:08:18 -07002275 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002276 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2277 page, fsdata);
2278 if (unlikely(status < 0))
2279 break;
2280 copied = status;
2281
2282 cond_resched();
2283
Nick Piggin124d3b72008-02-02 15:01:17 +01002284 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002285 if (unlikely(copied == 0)) {
2286 /*
2287 * If we were unable to copy any data at all, we must
2288 * fall back to a single segment length write.
2289 *
2290 * If we didn't fallback here, we could livelock
2291 * because not all segments in the iov can be copied at
2292 * once without a pagefault.
2293 */
2294 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2295 iov_iter_single_seg_count(i));
2296 goto again;
2297 }
Nick Pigginafddba42007-10-16 01:25:01 -07002298 pos += copied;
2299 written += copied;
2300
2301 balance_dirty_pages_ratelimited(mapping);
2302
2303 } while (iov_iter_count(i));
2304
2305 return written ? written : status;
2306}
2307
2308ssize_t
2309generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2310 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2311 size_t count, ssize_t written)
2312{
2313 struct file *file = iocb->ki_filp;
2314 struct address_space *mapping = file->f_mapping;
2315 const struct address_space_operations *a_ops = mapping->a_ops;
2316 struct inode *inode = mapping->host;
2317 ssize_t status;
2318 struct iov_iter i;
2319
2320 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002321 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002324 written += status;
2325 *ppos = pos + status;
2326
2327 /*
2328 * For now, when the user asks for O_SYNC, we'll actually give
2329 * O_DSYNC
2330 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2332 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2333 status = generic_osync_inode(inode, mapping,
2334 OSYNC_METADATA|OSYNC_DATA);
2335 }
2336 }
2337
2338 /*
2339 * If we get here for O_DIRECT writes then we must have fallen through
2340 * to buffered writes (block instantiation inside i_size). So we sync
2341 * the file data here, to try to honour O_DIRECT expectations.
2342 */
2343 if (unlikely(file->f_flags & O_DIRECT) && written)
Nick Piggin48b47c52009-01-06 14:40:22 -08002344 status = filemap_write_and_wait_range(mapping,
2345 pos, pos + written - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 return written ? written : status;
2348}
2349EXPORT_SYMBOL(generic_file_buffered_write);
2350
Adrian Bunk5ce78522005-09-10 00:26:28 -07002351static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2353 unsigned long nr_segs, loff_t *ppos)
2354{
2355 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002356 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 size_t ocount; /* original count */
2358 size_t count; /* after file limit checks */
2359 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 loff_t pos;
2361 ssize_t written;
2362 ssize_t err;
2363
2364 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002365 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2366 if (err)
2367 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
2369 count = ocount;
2370 pos = *ppos;
2371
2372 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2373
2374 /* We can write back this queue in page reclaim */
2375 current->backing_dev_info = mapping->backing_dev_info;
2376 written = 0;
2377
2378 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2379 if (err)
2380 goto out;
2381
2382 if (count == 0)
2383 goto out;
2384
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002385 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 if (err)
2387 goto out;
2388
Christoph Hellwig870f4812006-01-09 20:52:01 -08002389 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
2391 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2392 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002393 loff_t endbyte;
2394 ssize_t written_buffered;
2395
2396 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2397 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 if (written < 0 || written == count)
2399 goto out;
2400 /*
2401 * direct-io write to a hole: fall through to buffered I/O
2402 * for completing the rest of the request.
2403 */
2404 pos += written;
2405 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002406 written_buffered = generic_file_buffered_write(iocb, iov,
2407 nr_segs, pos, ppos, count,
2408 written);
2409 /*
2410 * If generic_file_buffered_write() retuned a synchronous error
2411 * then we want to return the number of bytes which were
2412 * direct-written, or the error code if that was zero. Note
2413 * that this differs from normal direct-io semantics, which
2414 * will return -EFOO even if some bytes were written.
2415 */
2416 if (written_buffered < 0) {
2417 err = written_buffered;
2418 goto out;
2419 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002421 /*
2422 * We need to ensure that the page cache pages are written to
2423 * disk and invalidated to preserve the expected O_DIRECT
2424 * semantics.
2425 */
2426 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002427 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2428 SYNC_FILE_RANGE_WAIT_BEFORE|
2429 SYNC_FILE_RANGE_WRITE|
2430 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002431 if (err == 0) {
2432 written = written_buffered;
2433 invalidate_mapping_pages(mapping,
2434 pos >> PAGE_CACHE_SHIFT,
2435 endbyte >> PAGE_CACHE_SHIFT);
2436 } else {
2437 /*
2438 * We don't know how much we wrote, so just return
2439 * the number of bytes which were direct-written
2440 */
2441 }
2442 } else {
2443 written = generic_file_buffered_write(iocb, iov, nr_segs,
2444 pos, ppos, count, written);
2445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446out:
2447 current->backing_dev_info = NULL;
2448 return written ? written : err;
2449}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450
Badari Pulavarty027445c2006-09-30 23:28:46 -07002451ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2452 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453{
2454 struct file *file = iocb->ki_filp;
2455 struct address_space *mapping = file->f_mapping;
2456 struct inode *inode = mapping->host;
2457 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
Badari Pulavarty027445c2006-09-30 23:28:46 -07002459 BUG_ON(iocb->ki_pos != pos);
2460
2461 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2462 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
2464 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002465 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
2467 err = sync_page_range_nolock(inode, mapping, pos, ret);
2468 if (err < 0)
2469 ret = err;
2470 }
2471 return ret;
2472}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002473EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
Badari Pulavarty027445c2006-09-30 23:28:46 -07002475ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2476 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477{
2478 struct file *file = iocb->ki_filp;
2479 struct address_space *mapping = file->f_mapping;
2480 struct inode *inode = mapping->host;
2481 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
2483 BUG_ON(iocb->ki_pos != pos);
2484
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002485 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002486 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2487 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002488 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
2490 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2491 ssize_t err;
2492
2493 err = sync_page_range(inode, mapping, pos, ret);
2494 if (err < 0)
2495 ret = err;
2496 }
2497 return ret;
2498}
2499EXPORT_SYMBOL(generic_file_aio_write);
2500
David Howellscf9a2ae2006-08-29 19:05:54 +01002501/**
2502 * try_to_release_page() - release old fs-specific metadata on a page
2503 *
2504 * @page: the page which the kernel is trying to free
2505 * @gfp_mask: memory allocation flags (and I/O mode)
2506 *
2507 * The address_space is to try to release any data against the page
2508 * (presumably at page->private). If the release was successful, return `1'.
2509 * Otherwise return zero.
2510 *
David Howells266cf652009-04-03 16:42:36 +01002511 * This may also be called if PG_fscache is set on a page, indicating that the
2512 * page is known to the local caching routines.
2513 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002514 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002515 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002516 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002517 */
2518int try_to_release_page(struct page *page, gfp_t gfp_mask)
2519{
2520 struct address_space * const mapping = page->mapping;
2521
2522 BUG_ON(!PageLocked(page));
2523 if (PageWriteback(page))
2524 return 0;
2525
2526 if (mapping && mapping->a_ops->releasepage)
2527 return mapping->a_ops->releasepage(page, gfp_mask);
2528 return try_to_free_buffers(page);
2529}
2530
2531EXPORT_SYMBOL(try_to_release_page);