blob: 379ff0bcbf6e88eb98e13550853fa215cbc25789 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
13#include <linux/slab.h>
14#include <linux/compiler.h>
15#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kernel_stat.h>
20#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036#include <linux/mm_inline.h> /* for page_is_file_cache() */
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
42#include <linux/buffer_head.h> /* for generic_osync_inode */
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
Adrian Bunk5ce78522005-09-10 00:26:28 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/*
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
49 * though.
50 *
51 * Shared mappings now work. 15.8.1995 Bruno.
52 *
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 *
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
57 */
58
59/*
60 * Lock ordering:
61 *
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070064 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080067 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * ->i_mmap_lock (truncate->unmap_mapping_range)
69 *
70 * ->mmap_sem
71 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070072 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
74 *
75 * ->mmap_sem
76 * ->lock_page (access_process_vm)
77 *
Nick Piggin82591e62006-10-19 23:29:10 -070078 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080081 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * ->i_alloc_sem (various)
83 *
84 * ->inode_lock
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
87 *
88 * ->i_mmap_lock
89 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080099 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 *
106 * ->task->proc_lock
107 * ->dcache_lock (proc_pid_lookup)
108 */
109
110/*
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700113 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
115void __remove_from_page_cache(struct page *page)
116{
117 struct address_space *mapping = page->mapping;
118
119 radix_tree_delete(&mapping->page_tree, page->index);
120 page->mapping = NULL;
121 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700122 __dec_zone_page_state(page, NR_FILE_PAGES);
Nick Piggin45426812007-07-15 23:38:12 -0700123 BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -0700124 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds3a692792007-12-19 14:05:13 -0800125
126 /*
127 * Some filesystems seem to re-dirty the page even after
128 * the VM has canceled the dirty bit (eg ext3 journaling).
129 *
130 * Fix it up by doing a final dirty accounting check after
131 * having removed the page entirely.
132 */
133 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
134 dec_zone_page_state(page, NR_FILE_DIRTY);
135 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
139void remove_from_page_cache(struct page *page)
140{
141 struct address_space *mapping = page->mapping;
142
Matt Mackallcd7619d2005-05-01 08:59:01 -0700143 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Nick Piggin19fd6232008-07-25 19:45:32 -0700145 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700147 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
150static int sync_page(void *word)
151{
152 struct address_space *mapping;
153 struct page *page;
154
Andi Kleen07808b72005-11-05 17:25:53 +0100155 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700158 * page_mapping() is being called without PG_locked held.
159 * Some knowledge of the state and use of the page is used to
160 * reduce the requirements down to a memory barrier.
161 * The danger here is of a stale page_mapping() return value
162 * indicating a struct address_space different from the one it's
163 * associated with when it is associated with one.
164 * After smp_mb(), it's either the correct page_mapping() for
165 * the page, or an old page_mapping() and the page's own
166 * page_mapping() has gone NULL.
167 * The ->sync_page() address_space operation must tolerate
168 * page_mapping() going NULL. By an amazing coincidence,
169 * this comes about because none of the users of the page
170 * in the ->sync_page() methods make essential use of the
171 * page_mapping(), merely passing the page down to the backing
172 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700173 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700174 * of interest. When page_mapping() does go NULL, the entire
175 * call stack gracefully ignores the page and returns.
176 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 */
178 smp_mb();
179 mapping = page_mapping(page);
180 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
181 mapping->a_ops->sync_page(page);
182 io_schedule();
183 return 0;
184}
185
Matthew Wilcox2687a352007-12-06 11:18:49 -0500186static int sync_page_killable(void *word)
187{
188 sync_page(word);
189 return fatal_signal_pending(current) ? -EINTR : 0;
190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700194 * @mapping: address space structure to write
195 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800196 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700197 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700199 * Start writeback against all of a mapping's dirty pages that lie
200 * within the byte offsets <start, end> inclusive.
201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700203 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * these two operations is that if a dirty page/buffer is encountered, it must
205 * be waited upon, and not just skipped over.
206 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800207int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
208 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 int ret;
211 struct writeback_control wbc = {
212 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800213 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700214 .range_start = start,
215 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 };
217
218 if (!mapping_cap_writeback_dirty(mapping))
219 return 0;
220
221 ret = do_writepages(mapping, &wbc);
222 return ret;
223}
224
225static inline int __filemap_fdatawrite(struct address_space *mapping,
226 int sync_mode)
227{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229}
230
231int filemap_fdatawrite(struct address_space *mapping)
232{
233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
234}
235EXPORT_SYMBOL(filemap_fdatawrite);
236
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400237int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800238 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
241}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400242EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Randy Dunlap485bb992006-06-23 02:03:49 -0700244/**
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
247 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
250 */
251int filemap_flush(struct address_space *mapping)
252{
253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
254}
255EXPORT_SYMBOL(filemap_flush);
256
Randy Dunlap485bb992006-06-23 02:03:49 -0700257/**
258 * wait_on_page_writeback_range - wait for writeback to complete
259 * @mapping: target address_space
260 * @start: beginning page index
261 * @end: ending page index
262 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 * Wait for writeback to complete against pages indexed by start->end
264 * inclusive
265 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800266int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 pgoff_t start, pgoff_t end)
268{
269 struct pagevec pvec;
270 int nr_pages;
271 int ret = 0;
272 pgoff_t index;
273
274 if (end < start)
275 return 0;
276
277 pagevec_init(&pvec, 0);
278 index = start;
279 while ((index <= end) &&
280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
281 PAGECACHE_TAG_WRITEBACK,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
283 unsigned i;
284
285 for (i = 0; i < nr_pages; i++) {
286 struct page *page = pvec.pages[i];
287
288 /* until radix tree lookup accepts end_index */
289 if (page->index > end)
290 continue;
291
292 wait_on_page_writeback(page);
293 if (PageError(page))
294 ret = -EIO;
295 }
296 pagevec_release(&pvec);
297 cond_resched();
298 }
299
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
302 ret = -ENOSPC;
303 if (test_and_clear_bit(AS_EIO, &mapping->flags))
304 ret = -EIO;
305
306 return ret;
307}
308
Randy Dunlap485bb992006-06-23 02:03:49 -0700309/**
310 * sync_page_range - write and wait on all pages in the passed range
311 * @inode: target inode
312 * @mapping: target address_space
313 * @pos: beginning offset in pages to write
314 * @count: number of bytes to write
315 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * Write and wait upon all the pages in the passed range. This is a "data
317 * integrity" operation. It waits upon in-flight writeout before starting and
318 * waiting upon new writeout. If there was an IO error, return it.
319 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800320 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * it is otherwise livelockable.
322 */
323int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800324 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
327 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
328 int ret;
329
330 if (!mapping_cap_writeback_dirty(mapping) || !count)
331 return 0;
332 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
333 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800334 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800336 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
338 if (ret == 0)
339 ret = wait_on_page_writeback_range(mapping, start, end);
340 return ret;
341}
342EXPORT_SYMBOL(sync_page_range);
343
Randy Dunlap485bb992006-06-23 02:03:49 -0700344/**
Randy Dunlap76824862008-03-19 17:00:40 -0700345 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
Randy Dunlap485bb992006-06-23 02:03:49 -0700346 * @inode: target inode
347 * @mapping: target address_space
348 * @pos: beginning offset in pages to write
349 * @count: number of bytes to write
350 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800351 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 * as it forces O_SYNC writers to different parts of the same file
353 * to be serialised right until io completion.
354 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800355int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
356 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
358 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
359 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
360 int ret;
361
362 if (!mapping_cap_writeback_dirty(mapping) || !count)
363 return 0;
364 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
365 if (ret == 0)
366 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
367 if (ret == 0)
368 ret = wait_on_page_writeback_range(mapping, start, end);
369 return ret;
370}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800371EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700374 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700376 *
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
380int filemap_fdatawait(struct address_space *mapping)
381{
382 loff_t i_size = i_size_read(mapping->host);
383
384 if (i_size == 0)
385 return 0;
386
387 return wait_on_page_writeback_range(mapping, 0,
388 (i_size - 1) >> PAGE_CACHE_SHIFT);
389}
390EXPORT_SYMBOL(filemap_fdatawait);
391
392int filemap_write_and_wait(struct address_space *mapping)
393{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800394 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800397 err = filemap_fdatawrite(mapping);
398 /*
399 * Even if the above returned error, the pages may be
400 * written partially (e.g. -ENOSPC), so we wait for it.
401 * But the -EIO is special case, it may indicate the worst
402 * thing (e.g. bug) happened, so we avoid waiting for it.
403 */
404 if (err != -EIO) {
405 int err2 = filemap_fdatawait(mapping);
406 if (!err)
407 err = err2;
408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800410 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800412EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Randy Dunlap485bb992006-06-23 02:03:49 -0700414/**
415 * filemap_write_and_wait_range - write out & wait on a file range
416 * @mapping: the address_space for the pages
417 * @lstart: offset in bytes where the range starts
418 * @lend: offset in bytes where the range ends (inclusive)
419 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800420 * Write out and wait upon file offsets lstart->lend, inclusive.
421 *
422 * Note that `lend' is inclusive (describes the last byte to be written) so
423 * that this function can be used to write to the very end-of-file (end = -1).
424 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425int filemap_write_and_wait_range(struct address_space *mapping,
426 loff_t lstart, loff_t lend)
427{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800428 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800431 err = __filemap_fdatawrite_range(mapping, lstart, lend,
432 WB_SYNC_ALL);
433 /* See comment of filemap_write_and_wait() */
434 if (err != -EIO) {
435 int err2 = wait_on_page_writeback_range(mapping,
436 lstart >> PAGE_CACHE_SHIFT,
437 lend >> PAGE_CACHE_SHIFT);
438 if (!err)
439 err = err2;
440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800442 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
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);
Nick Piggine2867812008-07-25 19:45:30 -0700479 } else {
480 page->mapping = NULL;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700481 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700482 page_cache_release(page);
483 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800484
Nick Piggin19fd6232008-07-25 19:45:32 -0700485 spin_unlock_irq(&mapping->tree_lock);
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();
Nick Piggin2ae88142006-10-28 10:38:23 -0700524 return alloc_pages_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{
1007 if (!ra->ra_pages)
1008 return;
1009
1010 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001011}
1012
Randy Dunlap485bb992006-06-23 02:03:49 -07001013/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001014 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001015 * @filp: the file to read
1016 * @ppos: current file position
1017 * @desc: read_descriptor
1018 * @actor: read method
1019 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001021 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 *
1023 * This is really ugly. But the goto's actually try to clarify some
1024 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001026static void do_generic_file_read(struct file *filp, loff_t *ppos,
1027 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001029 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001031 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001032 pgoff_t index;
1033 pgoff_t last_index;
1034 pgoff_t prev_index;
1035 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001036 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001040 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1041 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1043 offset = *ppos & ~PAGE_CACHE_MASK;
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 for (;;) {
1046 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001047 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001048 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 unsigned long nr, ret;
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052find_page:
1053 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001054 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001055 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001056 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001057 index, last_index - index);
1058 page = find_get_page(mapping, index);
1059 if (unlikely(page == NULL))
1060 goto no_cached_page;
1061 }
1062 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001063 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001064 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001065 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001067 if (!PageUptodate(page)) {
1068 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1069 !mapping->a_ops->is_partially_uptodate)
1070 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001071 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001072 goto page_not_up_to_date;
1073 if (!mapping->a_ops->is_partially_uptodate(page,
1074 desc, offset))
1075 goto page_not_up_to_date_locked;
1076 unlock_page(page);
1077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001079 /*
1080 * i_size must be checked after we know the page is Uptodate.
1081 *
1082 * Checking i_size after the check allows us to calculate
1083 * the correct value for "nr", which means the zero-filled
1084 * part of the page is not copied back to userspace (unless
1085 * another truncate extends the file - this is desired though).
1086 */
1087
1088 isize = i_size_read(inode);
1089 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1090 if (unlikely(!isize || index > end_index)) {
1091 page_cache_release(page);
1092 goto out;
1093 }
1094
1095 /* nr is the maximum number of bytes to copy from this page */
1096 nr = PAGE_CACHE_SIZE;
1097 if (index == end_index) {
1098 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1099 if (nr <= offset) {
1100 page_cache_release(page);
1101 goto out;
1102 }
1103 }
1104 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106 /* If users can be writing to this page using arbitrary
1107 * virtual addresses, take care about potential aliasing
1108 * before reading the page on the kernel side.
1109 */
1110 if (mapping_writably_mapped(mapping))
1111 flush_dcache_page(page);
1112
1113 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001114 * When a sequential read accesses a page several times,
1115 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 */
Jan Karaec0f1632007-05-06 14:49:25 -07001117 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 mark_page_accessed(page);
1119 prev_index = index;
1120
1121 /*
1122 * Ok, we have the page, and it's up-to-date, so
1123 * now we can copy it to user space...
1124 *
1125 * The actor routine returns how many bytes were actually used..
1126 * NOTE! This may not be the same as how much of a user buffer
1127 * we filled up (we may be padding etc), so we can only update
1128 * "pos" here (the actor routine has to update the user buffer
1129 * pointers and the remaining count).
1130 */
1131 ret = actor(desc, page, offset, nr);
1132 offset += ret;
1133 index += offset >> PAGE_CACHE_SHIFT;
1134 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001135 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 page_cache_release(page);
1138 if (ret == nr && desc->count)
1139 continue;
1140 goto out;
1141
1142page_not_up_to_date:
1143 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001144 error = lock_page_killable(page);
1145 if (unlikely(error))
1146 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001148page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001149 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (!page->mapping) {
1151 unlock_page(page);
1152 page_cache_release(page);
1153 continue;
1154 }
1155
1156 /* Did somebody else fill it already? */
1157 if (PageUptodate(page)) {
1158 unlock_page(page);
1159 goto page_ok;
1160 }
1161
1162readpage:
1163 /* Start the actual read. The read will unlock the page. */
1164 error = mapping->a_ops->readpage(filp, page);
1165
Zach Brown994fc28c2005-12-15 14:28:17 -08001166 if (unlikely(error)) {
1167 if (error == AOP_TRUNCATED_PAGE) {
1168 page_cache_release(page);
1169 goto find_page;
1170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001172 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
1174 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001175 error = lock_page_killable(page);
1176 if (unlikely(error))
1177 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 if (!PageUptodate(page)) {
1179 if (page->mapping == NULL) {
1180 /*
1181 * invalidate_inode_pages got it
1182 */
1183 unlock_page(page);
1184 page_cache_release(page);
1185 goto find_page;
1186 }
1187 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001188 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001189 error = -EIO;
1190 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 }
1192 unlock_page(page);
1193 }
1194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 goto page_ok;
1196
1197readpage_error:
1198 /* UHHUH! A synchronous read error occurred. Report it */
1199 desc->error = error;
1200 page_cache_release(page);
1201 goto out;
1202
1203no_cached_page:
1204 /*
1205 * Ok, it wasn't cached, so we need to create a new
1206 * page..
1207 */
Nick Piggineb2be182007-10-16 01:24:57 -07001208 page = page_cache_alloc_cold(mapping);
1209 if (!page) {
1210 desc->error = -ENOMEM;
1211 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 }
Nick Piggineb2be182007-10-16 01:24:57 -07001213 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 index, GFP_KERNEL);
1215 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001216 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (error == -EEXIST)
1218 goto find_page;
1219 desc->error = error;
1220 goto out;
1221 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 goto readpage;
1223 }
1224
1225out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001226 ra->prev_pos = prev_index;
1227 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1228 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001230 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001231 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234int file_read_actor(read_descriptor_t *desc, struct page *page,
1235 unsigned long offset, unsigned long size)
1236{
1237 char *kaddr;
1238 unsigned long left, count = desc->count;
1239
1240 if (size > count)
1241 size = count;
1242
1243 /*
1244 * Faults on the destination of a read are common, so do it before
1245 * taking the kmap.
1246 */
1247 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1248 kaddr = kmap_atomic(page, KM_USER0);
1249 left = __copy_to_user_inatomic(desc->arg.buf,
1250 kaddr + offset, size);
1251 kunmap_atomic(kaddr, KM_USER0);
1252 if (left == 0)
1253 goto success;
1254 }
1255
1256 /* Do it the slow way */
1257 kaddr = kmap(page);
1258 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1259 kunmap(page);
1260
1261 if (left) {
1262 size -= left;
1263 desc->error = -EFAULT;
1264 }
1265success:
1266 desc->count = count - size;
1267 desc->written += size;
1268 desc->arg.buf += size;
1269 return size;
1270}
1271
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001272/*
1273 * Performs necessary checks before doing a write
1274 * @iov: io vector request
1275 * @nr_segs: number of segments in the iovec
1276 * @count: number of bytes to write
1277 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1278 *
1279 * Adjust number of segments and amount of bytes to write (nr_segs should be
1280 * properly initialized first). Returns appropriate error code that caller
1281 * should return or zero in case that write should be allowed.
1282 */
1283int generic_segment_checks(const struct iovec *iov,
1284 unsigned long *nr_segs, size_t *count, int access_flags)
1285{
1286 unsigned long seg;
1287 size_t cnt = 0;
1288 for (seg = 0; seg < *nr_segs; seg++) {
1289 const struct iovec *iv = &iov[seg];
1290
1291 /*
1292 * If any segment has a negative length, or the cumulative
1293 * length ever wraps negative then return -EINVAL.
1294 */
1295 cnt += iv->iov_len;
1296 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1297 return -EINVAL;
1298 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1299 continue;
1300 if (seg == 0)
1301 return -EFAULT;
1302 *nr_segs = seg;
1303 cnt -= iv->iov_len; /* This segment is no good */
1304 break;
1305 }
1306 *count = cnt;
1307 return 0;
1308}
1309EXPORT_SYMBOL(generic_segment_checks);
1310
Randy Dunlap485bb992006-06-23 02:03:49 -07001311/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001312 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001313 * @iocb: kernel I/O control block
1314 * @iov: io vector request
1315 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001316 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001317 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 * This is the "read()" routine for all filesystems
1319 * that can use the page cache directly.
1320 */
1321ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001322generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1323 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324{
1325 struct file *filp = iocb->ki_filp;
1326 ssize_t retval;
1327 unsigned long seg;
1328 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001329 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
1331 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001332 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1333 if (retval)
1334 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
1336 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1337 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001338 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 struct address_space *mapping;
1340 struct inode *inode;
1341
1342 mapping = filp->f_mapping;
1343 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 if (!count)
1345 goto out; /* skip atime */
1346 size = i_size_read(inode);
1347 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001348 retval = filemap_write_and_wait_range(mapping, pos,
1349 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001350 if (!retval) {
1351 retval = mapping->a_ops->direct_IO(READ, iocb,
1352 iov, pos, nr_segs);
1353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 if (retval > 0)
1355 *ppos = pos + retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001356 if (retval) {
1357 file_accessed(filp);
1358 goto out;
1359 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001360 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
1362
Hugh Dickins11fa9772008-07-23 21:27:34 -07001363 for (seg = 0; seg < nr_segs; seg++) {
1364 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Hugh Dickins11fa9772008-07-23 21:27:34 -07001366 desc.written = 0;
1367 desc.arg.buf = iov[seg].iov_base;
1368 desc.count = iov[seg].iov_len;
1369 if (desc.count == 0)
1370 continue;
1371 desc.error = 0;
1372 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1373 retval += desc.written;
1374 if (desc.error) {
1375 retval = retval ?: desc.error;
1376 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001378 if (desc.count > 0)
1379 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
1381out:
1382 return retval;
1383}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384EXPORT_SYMBOL(generic_file_aio_read);
1385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386static ssize_t
1387do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001388 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
1390 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1391 return -EINVAL;
1392
1393 force_page_cache_readahead(mapping, filp, index,
1394 max_sane_readahead(nr));
1395 return 0;
1396}
1397
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001398SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
1400 ssize_t ret;
1401 struct file *file;
1402
1403 ret = -EBADF;
1404 file = fget(fd);
1405 if (file) {
1406 if (file->f_mode & FMODE_READ) {
1407 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001408 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1409 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 unsigned long len = end - start + 1;
1411 ret = do_readahead(mapping, file, start, len);
1412 }
1413 fput(file);
1414 }
1415 return ret;
1416}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001417#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1418asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1419{
1420 return SYSC_readahead((int) fd, offset, (size_t) count);
1421}
1422SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1423#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001426/**
1427 * page_cache_read - adds requested page to the page cache if not already there
1428 * @file: file to read
1429 * @offset: page index
1430 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 * This adds the requested page to the page cache if it isn't already there,
1432 * and schedules an I/O to read in its contents from disk.
1433 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001434static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
1436 struct address_space *mapping = file->f_mapping;
1437 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001438 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Zach Brown994fc28c2005-12-15 14:28:17 -08001440 do {
1441 page = page_cache_alloc_cold(mapping);
1442 if (!page)
1443 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
Zach Brown994fc28c2005-12-15 14:28:17 -08001445 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1446 if (ret == 0)
1447 ret = mapping->a_ops->readpage(file, page);
1448 else if (ret == -EEXIST)
1449 ret = 0; /* losing race to add is OK */
1450
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
Zach Brown994fc28c2005-12-15 14:28:17 -08001453 } while (ret == AOP_TRUNCATED_PAGE);
1454
1455 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
1458#define MMAP_LOTSAMISS (100)
1459
Randy Dunlap485bb992006-06-23 02:03:49 -07001460/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001461 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001462 * @vma: vma in which the fault was taken
1463 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001464 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001465 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 * mapped memory region to read in file data during a page fault.
1467 *
1468 * The goto's are kind of ugly, but this streamlines the normal case of having
1469 * it in the page cache, and handles the special cases reasonably without
1470 * having a lot of duplicated code.
1471 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001472int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
1474 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001475 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 struct address_space *mapping = file->f_mapping;
1477 struct file_ra_state *ra = &file->f_ra;
1478 struct inode *inode = mapping->host;
1479 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001480 pgoff_t size;
Nick Piggin54cb8822007-07-19 01:46:59 -07001481 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001482 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001485 if (vmf->pgoff >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001486 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
1488 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001489 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 goto no_cached_page;
1491
1492 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 * Do we have something in the page cache already?
1494 */
1495retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001496 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001497 /*
1498 * For sequential accesses, we use the generic readahead logic.
1499 */
1500 if (VM_SequentialReadHint(vma)) {
1501 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001502 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001503 vmf->pgoff, 1);
1504 page = find_lock_page(mapping, vmf->pgoff);
1505 if (!page)
1506 goto no_cached_page;
1507 }
1508 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001509 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001510 vmf->pgoff, 1);
1511 }
1512 }
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 if (!page) {
1515 unsigned long ra_pages;
1516
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 ra->mmap_miss++;
1518
1519 /*
1520 * Do we miss much more than hit in this file? If so,
1521 * stop bothering with read-ahead. It will only hurt.
1522 */
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001523 if (ra->mmap_miss > MMAP_LOTSAMISS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 goto no_cached_page;
1525
1526 /*
1527 * To keep the pgmajfault counter straight, we need to
1528 * check did_readaround, as this is an inner loop.
1529 */
1530 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001531 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001532 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 }
1534 did_readaround = 1;
1535 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1536 if (ra_pages) {
1537 pgoff_t start = 0;
1538
Nick Piggind0217ac2007-07-19 01:47:03 -07001539 if (vmf->pgoff > ra_pages / 2)
1540 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 do_page_cache_readahead(mapping, file, start, ra_pages);
1542 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001543 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 if (!page)
1545 goto no_cached_page;
1546 }
1547
1548 if (!did_readaround)
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001549 ra->mmap_miss--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
1551 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001552 * We have a locked page in the page cache, now we need to check
1553 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 */
Nick Piggind00806b2007-07-19 01:46:57 -07001555 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 goto page_not_uptodate;
1557
Nick Piggind00806b2007-07-19 01:46:57 -07001558 /* Must recheck i_size under page lock */
1559 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001560 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001561 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001562 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001563 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001564 }
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 /*
1567 * Found the page and have a reference on it.
1568 */
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001569 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001570 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001571 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573no_cached_page:
1574 /*
1575 * We're only likely to ever get here if MADV_RANDOM is in
1576 * effect.
1577 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001578 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 /*
1581 * The page we want has now been added to the page cache.
1582 * In the unlikely event that someone removed it in the
1583 * meantime, we'll just come back here and read it again.
1584 */
1585 if (error >= 0)
1586 goto retry_find;
1587
1588 /*
1589 * An error return from page_cache_read can result if the
1590 * system is low on memory, or a problem occurs while trying
1591 * to schedule I/O.
1592 */
1593 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001594 return VM_FAULT_OOM;
1595 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001598 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001600 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001601 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
1604 /*
1605 * Umm, take care of errors if the page isn't up-to-date.
1606 * Try to re-read it _once_. We do this synchronously,
1607 * because there really aren't any performance issues here
1608 * and we need to check for errors.
1609 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001611 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001612 if (!error) {
1613 wait_on_page_locked(page);
1614 if (!PageUptodate(page))
1615 error = -EIO;
1616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001618
1619 if (!error || error == AOP_TRUNCATED_PAGE)
1620 goto retry_find;
1621
1622 /* Things didn't work out. Return zero to tell the mm layer so. */
1623 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001624 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001625}
1626EXPORT_SYMBOL(filemap_fault);
1627
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001629 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630};
1631
1632/* This is used for a general mmap of a disk file */
1633
1634int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1635{
1636 struct address_space *mapping = file->f_mapping;
1637
1638 if (!mapping->a_ops->readpage)
1639 return -ENOEXEC;
1640 file_accessed(file);
1641 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001642 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 return 0;
1644}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646/*
1647 * This is for filesystems which do not implement ->writepage.
1648 */
1649int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1650{
1651 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1652 return -EINVAL;
1653 return generic_file_mmap(file, vma);
1654}
1655#else
1656int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1657{
1658 return -ENOSYS;
1659}
1660int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1661{
1662 return -ENOSYS;
1663}
1664#endif /* CONFIG_MMU */
1665
1666EXPORT_SYMBOL(generic_file_mmap);
1667EXPORT_SYMBOL(generic_file_readonly_mmap);
1668
Nick Piggin6fe69002007-05-06 14:49:04 -07001669static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001670 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 int (*filler)(void *,struct page*),
1672 void *data)
1673{
Nick Piggineb2be182007-10-16 01:24:57 -07001674 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 int err;
1676repeat:
1677 page = find_get_page(mapping, index);
1678 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001679 page = page_cache_alloc_cold(mapping);
1680 if (!page)
1681 return ERR_PTR(-ENOMEM);
1682 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1683 if (unlikely(err)) {
1684 page_cache_release(page);
1685 if (err == -EEXIST)
1686 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 return ERR_PTR(err);
1689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 err = filler(data, page);
1691 if (err < 0) {
1692 page_cache_release(page);
1693 page = ERR_PTR(err);
1694 }
1695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 return page;
1697}
1698
Randy Dunlap76824862008-03-19 17:00:40 -07001699/**
1700 * read_cache_page_async - read into page cache, fill it if needed
1701 * @mapping: the page's address_space
1702 * @index: the page index
1703 * @filler: function to perform the read
1704 * @data: destination for read data
1705 *
Nick Piggin6fe69002007-05-06 14:49:04 -07001706 * Same as read_cache_page, but don't wait for page to become unlocked
1707 * after submitting it to the filler.
Randy Dunlap76824862008-03-19 17:00:40 -07001708 *
1709 * Read into the page cache. If a page already exists, and PageUptodate() is
1710 * not set, try to fill the page but don't wait for it to become unlocked.
1711 *
1712 * If the page does not get brought uptodate, return -EIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001714struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001715 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 int (*filler)(void *,struct page*),
1717 void *data)
1718{
1719 struct page *page;
1720 int err;
1721
1722retry:
1723 page = __read_cache_page(mapping, index, filler, data);
1724 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001725 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 if (PageUptodate(page))
1727 goto out;
1728
1729 lock_page(page);
1730 if (!page->mapping) {
1731 unlock_page(page);
1732 page_cache_release(page);
1733 goto retry;
1734 }
1735 if (PageUptodate(page)) {
1736 unlock_page(page);
1737 goto out;
1738 }
1739 err = filler(data, page);
1740 if (err < 0) {
1741 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001742 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 }
David Howellsc855ff32007-05-09 13:42:20 +01001744out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001745 mark_page_accessed(page);
1746 return page;
1747}
1748EXPORT_SYMBOL(read_cache_page_async);
1749
1750/**
1751 * read_cache_page - read into page cache, fill it if needed
1752 * @mapping: the page's address_space
1753 * @index: the page index
1754 * @filler: function to perform the read
1755 * @data: destination for read data
1756 *
1757 * Read into the page cache. If a page already exists, and PageUptodate() is
1758 * not set, try to fill the page then wait for it to become unlocked.
1759 *
1760 * If the page does not get brought uptodate, return -EIO.
1761 */
1762struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001763 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001764 int (*filler)(void *,struct page*),
1765 void *data)
1766{
1767 struct page *page;
1768
1769 page = read_cache_page_async(mapping, index, filler, data);
1770 if (IS_ERR(page))
1771 goto out;
1772 wait_on_page_locked(page);
1773 if (!PageUptodate(page)) {
1774 page_cache_release(page);
1775 page = ERR_PTR(-EIO);
1776 }
1777 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 return page;
1779}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780EXPORT_SYMBOL(read_cache_page);
1781
1782/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 * The logic we want is
1784 *
1785 * if suid or (sgid and xgrp)
1786 * remove privs
1787 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001788int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789{
1790 mode_t mode = dentry->d_inode->i_mode;
1791 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
1793 /* suid always must be killed */
1794 if (unlikely(mode & S_ISUID))
1795 kill = ATTR_KILL_SUID;
1796
1797 /*
1798 * sgid without any exec bits is just a mandatory locking mark; leave
1799 * it alone. If some exec bits are set, it's a real sgid; kill it.
1800 */
1801 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1802 kill |= ATTR_KILL_SGID;
1803
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001804 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001805 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
Jens Axboe01de85e2006-10-17 19:50:36 +02001807 return 0;
1808}
Mark Fashehd23a1472006-10-17 17:05:18 -07001809EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001810
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001811static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001812{
1813 struct iattr newattrs;
1814
1815 newattrs.ia_valid = ATTR_FORCE | kill;
1816 return notify_change(dentry, &newattrs);
1817}
1818
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001819int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001820{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001821 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001822 int killsuid = should_remove_suid(dentry);
1823 int killpriv = security_inode_need_killpriv(dentry);
1824 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001825
Serge E. Hallynb5376772007-10-16 23:31:36 -07001826 if (killpriv < 0)
1827 return killpriv;
1828 if (killpriv)
1829 error = security_inode_killpriv(dentry);
1830 if (!error && killsuid)
1831 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001832
Serge E. Hallynb5376772007-10-16 23:31:36 -07001833 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001835EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Nick Piggin2f718ff2007-10-16 01:24:59 -07001837static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 const struct iovec *iov, size_t base, size_t bytes)
1839{
Ingo Molnarf1800532009-03-02 11:00:57 +01001840 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
1842 while (bytes) {
1843 char __user *buf = iov->iov_base + base;
1844 int copy = min(bytes, iov->iov_len - base);
1845
1846 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001847 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 copied += copy;
1849 bytes -= copy;
1850 vaddr += copy;
1851 iov++;
1852
NeilBrown01408c42006-06-25 05:47:58 -07001853 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 }
1856 return copied - left;
1857}
1858
1859/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001860 * Copy as much as we can into the page and return the number of bytes which
1861 * were sucessfully copied. If a fault is encountered then return the number of
1862 * bytes which were copied.
1863 */
1864size_t iov_iter_copy_from_user_atomic(struct page *page,
1865 struct iov_iter *i, unsigned long offset, size_t bytes)
1866{
1867 char *kaddr;
1868 size_t copied;
1869
1870 BUG_ON(!in_atomic());
1871 kaddr = kmap_atomic(page, KM_USER0);
1872 if (likely(i->nr_segs == 1)) {
1873 int left;
1874 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001875 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001876 copied = bytes - left;
1877 } else {
1878 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1879 i->iov, i->iov_offset, bytes);
1880 }
1881 kunmap_atomic(kaddr, KM_USER0);
1882
1883 return copied;
1884}
Nick Piggin89e10782007-10-16 01:25:07 -07001885EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001886
1887/*
1888 * This has the same sideeffects and return value as
1889 * iov_iter_copy_from_user_atomic().
1890 * The difference is that it attempts to resolve faults.
1891 * Page must not be locked.
1892 */
1893size_t iov_iter_copy_from_user(struct page *page,
1894 struct iov_iter *i, unsigned long offset, size_t bytes)
1895{
1896 char *kaddr;
1897 size_t copied;
1898
1899 kaddr = kmap(page);
1900 if (likely(i->nr_segs == 1)) {
1901 int left;
1902 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001903 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001904 copied = bytes - left;
1905 } else {
1906 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1907 i->iov, i->iov_offset, bytes);
1908 }
1909 kunmap(page);
1910 return copied;
1911}
Nick Piggin89e10782007-10-16 01:25:07 -07001912EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001913
Nick Pigginf7009262008-03-10 11:43:59 -07001914void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001915{
Nick Pigginf7009262008-03-10 11:43:59 -07001916 BUG_ON(i->count < bytes);
1917
Nick Piggin2f718ff2007-10-16 01:24:59 -07001918 if (likely(i->nr_segs == 1)) {
1919 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001920 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001921 } else {
1922 const struct iovec *iov = i->iov;
1923 size_t base = i->iov_offset;
1924
Nick Piggin124d3b72008-02-02 15:01:17 +01001925 /*
1926 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001927 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001928 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001929 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001930 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001931
Nick Pigginf7009262008-03-10 11:43:59 -07001932 copy = min(bytes, iov->iov_len - base);
1933 BUG_ON(!i->count || i->count < copy);
1934 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001935 bytes -= copy;
1936 base += copy;
1937 if (iov->iov_len == base) {
1938 iov++;
1939 base = 0;
1940 }
1941 }
1942 i->iov = iov;
1943 i->iov_offset = base;
1944 }
1945}
Nick Piggin89e10782007-10-16 01:25:07 -07001946EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001947
Nick Pigginafddba42007-10-16 01:25:01 -07001948/*
1949 * Fault in the first iovec of the given iov_iter, to a maximum length
1950 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1951 * accessed (ie. because it is an invalid address).
1952 *
1953 * writev-intensive code may want this to prefault several iovecs -- that
1954 * would be possible (callers must not rely on the fact that _only_ the
1955 * first iovec will be faulted with the current implementation).
1956 */
1957int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001958{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001959 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001960 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1961 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001962}
Nick Piggin89e10782007-10-16 01:25:07 -07001963EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001964
1965/*
1966 * Return the count of just the current iov_iter segment.
1967 */
1968size_t iov_iter_single_seg_count(struct iov_iter *i)
1969{
1970 const struct iovec *iov = i->iov;
1971 if (i->nr_segs == 1)
1972 return i->count;
1973 else
1974 return min(i->count, iov->iov_len - i->iov_offset);
1975}
Nick Piggin89e10782007-10-16 01:25:07 -07001976EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001977
1978/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 * Performs necessary checks before doing a write
1980 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001981 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 * Returns appropriate error code that caller should return or
1983 * zero in case that write should be allowed.
1984 */
1985inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1986{
1987 struct inode *inode = file->f_mapping->host;
1988 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1989
1990 if (unlikely(*pos < 0))
1991 return -EINVAL;
1992
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 if (!isblk) {
1994 /* FIXME: this is for backwards compatibility with 2.4 */
1995 if (file->f_flags & O_APPEND)
1996 *pos = i_size_read(inode);
1997
1998 if (limit != RLIM_INFINITY) {
1999 if (*pos >= limit) {
2000 send_sig(SIGXFSZ, current, 0);
2001 return -EFBIG;
2002 }
2003 if (*count > limit - (typeof(limit))*pos) {
2004 *count = limit - (typeof(limit))*pos;
2005 }
2006 }
2007 }
2008
2009 /*
2010 * LFS rule
2011 */
2012 if (unlikely(*pos + *count > MAX_NON_LFS &&
2013 !(file->f_flags & O_LARGEFILE))) {
2014 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 return -EFBIG;
2016 }
2017 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2018 *count = MAX_NON_LFS - (unsigned long)*pos;
2019 }
2020 }
2021
2022 /*
2023 * Are we about to exceed the fs block limit ?
2024 *
2025 * If we have written data it becomes a short write. If we have
2026 * exceeded without writing data we send a signal and return EFBIG.
2027 * Linus frestrict idea will clean these up nicely..
2028 */
2029 if (likely(!isblk)) {
2030 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2031 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 return -EFBIG;
2033 }
2034 /* zero-length writes at ->s_maxbytes are OK */
2035 }
2036
2037 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2038 *count = inode->i_sb->s_maxbytes - *pos;
2039 } else {
David Howells93614012006-09-30 20:45:40 +02002040#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 loff_t isize;
2042 if (bdev_read_only(I_BDEV(inode)))
2043 return -EPERM;
2044 isize = i_size_read(inode);
2045 if (*pos >= isize) {
2046 if (*count || *pos > isize)
2047 return -ENOSPC;
2048 }
2049
2050 if (*pos + *count > isize)
2051 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002052#else
2053 return -EPERM;
2054#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 }
2056 return 0;
2057}
2058EXPORT_SYMBOL(generic_write_checks);
2059
Nick Pigginafddba42007-10-16 01:25:01 -07002060int pagecache_write_begin(struct file *file, struct address_space *mapping,
2061 loff_t pos, unsigned len, unsigned flags,
2062 struct page **pagep, void **fsdata)
2063{
2064 const struct address_space_operations *aops = mapping->a_ops;
2065
Nick Piggin4e02ed42008-10-29 14:00:55 -07002066 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002067 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002068}
2069EXPORT_SYMBOL(pagecache_write_begin);
2070
2071int pagecache_write_end(struct file *file, struct address_space *mapping,
2072 loff_t pos, unsigned len, unsigned copied,
2073 struct page *page, void *fsdata)
2074{
2075 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002076
Nick Piggin4e02ed42008-10-29 14:00:55 -07002077 mark_page_accessed(page);
2078 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002079}
2080EXPORT_SYMBOL(pagecache_write_end);
2081
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082ssize_t
2083generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2084 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2085 size_t count, size_t ocount)
2086{
2087 struct file *file = iocb->ki_filp;
2088 struct address_space *mapping = file->f_mapping;
2089 struct inode *inode = mapping->host;
2090 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002091 size_t write_len;
2092 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
2094 if (count != ocount)
2095 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2096
Christoph Hellwiga969e902008-07-23 21:27:04 -07002097 write_len = iov_length(iov, *nr_segs);
2098 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002099
Nick Piggin48b47c52009-01-06 14:40:22 -08002100 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002101 if (written)
2102 goto out;
2103
2104 /*
2105 * After a write we want buffered reads to be sure to go to disk to get
2106 * the new data. We invalidate clean cached page from the region we're
2107 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002108 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002109 */
2110 if (mapping->nrpages) {
2111 written = invalidate_inode_pages2_range(mapping,
2112 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002113 /*
2114 * If a page can not be invalidated, return 0 to fall back
2115 * to buffered write.
2116 */
2117 if (written) {
2118 if (written == -EBUSY)
2119 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002120 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002121 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002122 }
2123
2124 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2125
2126 /*
2127 * Finally, try again to invalidate clean pages which might have been
2128 * cached by non-direct readahead, or faulted in by get_user_pages()
2129 * if the source of the write was an mmap'ed region of the file
2130 * we're writing. Either one is a pretty crazy thing to do,
2131 * so we don't support it 100%. If this invalidation
2132 * fails, tough, the write still worked...
2133 */
2134 if (mapping->nrpages) {
2135 invalidate_inode_pages2_range(mapping,
2136 pos >> PAGE_CACHE_SHIFT, end);
2137 }
2138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 if (written > 0) {
2140 loff_t end = pos + written;
2141 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2142 i_size_write(inode, end);
2143 mark_inode_dirty(inode);
2144 }
2145 *ppos = end;
2146 }
2147
2148 /*
2149 * Sync the fs metadata but not the minor inode changes and
2150 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002151 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002152 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 */
Christoph Hellwiga969e902008-07-23 21:27:04 -07002154out:
Zach Brown8459d862006-12-10 02:21:05 -08002155 if ((written >= 0 || written == -EIOCBQUEUED) &&
2156 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002157 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2158 if (err < 0)
2159 written = err;
2160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 return written;
2162}
2163EXPORT_SYMBOL(generic_file_direct_write);
2164
Nick Piggineb2be182007-10-16 01:24:57 -07002165/*
2166 * Find or create a page at the given pagecache position. Return the locked
2167 * page. This function is specifically for buffered writes.
2168 */
Nick Piggin54566b22009-01-04 12:00:53 -08002169struct page *grab_cache_page_write_begin(struct address_space *mapping,
2170 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002171{
2172 int status;
2173 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002174 gfp_t gfp_notmask = 0;
2175 if (flags & AOP_FLAG_NOFS)
2176 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002177repeat:
2178 page = find_lock_page(mapping, index);
2179 if (likely(page))
2180 return page;
2181
Nick Piggin54566b22009-01-04 12:00:53 -08002182 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002183 if (!page)
2184 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002185 status = add_to_page_cache_lru(page, mapping, index,
2186 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002187 if (unlikely(status)) {
2188 page_cache_release(page);
2189 if (status == -EEXIST)
2190 goto repeat;
2191 return NULL;
2192 }
2193 return page;
2194}
Nick Piggin54566b22009-01-04 12:00:53 -08002195EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002196
Nick Pigginafddba42007-10-16 01:25:01 -07002197static ssize_t generic_perform_write(struct file *file,
2198 struct iov_iter *i, loff_t pos)
2199{
2200 struct address_space *mapping = file->f_mapping;
2201 const struct address_space_operations *a_ops = mapping->a_ops;
2202 long status = 0;
2203 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002204 unsigned int flags = 0;
2205
2206 /*
2207 * Copies from kernel address space cannot fail (NFSD is a big user).
2208 */
2209 if (segment_eq(get_fs(), KERNEL_DS))
2210 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002211
2212 do {
2213 struct page *page;
2214 pgoff_t index; /* Pagecache index for current page */
2215 unsigned long offset; /* Offset into pagecache page */
2216 unsigned long bytes; /* Bytes to write to page */
2217 size_t copied; /* Bytes copied from user */
2218 void *fsdata;
2219
2220 offset = (pos & (PAGE_CACHE_SIZE - 1));
2221 index = pos >> PAGE_CACHE_SHIFT;
2222 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2223 iov_iter_count(i));
2224
2225again:
2226
2227 /*
2228 * Bring in the user page that we will copy from _first_.
2229 * Otherwise there's a nasty deadlock on copying from the
2230 * same page as we're writing to, without it being marked
2231 * up-to-date.
2232 *
2233 * Not only is this an optimisation, but it is also required
2234 * to check that the address is actually valid, when atomic
2235 * usercopies are used, below.
2236 */
2237 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2238 status = -EFAULT;
2239 break;
2240 }
2241
Nick Piggin674b8922007-10-16 01:25:03 -07002242 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002243 &page, &fsdata);
2244 if (unlikely(status))
2245 break;
2246
2247 pagefault_disable();
2248 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2249 pagefault_enable();
2250 flush_dcache_page(page);
2251
2252 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2253 page, fsdata);
2254 if (unlikely(status < 0))
2255 break;
2256 copied = status;
2257
2258 cond_resched();
2259
Nick Piggin124d3b72008-02-02 15:01:17 +01002260 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002261 if (unlikely(copied == 0)) {
2262 /*
2263 * If we were unable to copy any data at all, we must
2264 * fall back to a single segment length write.
2265 *
2266 * If we didn't fallback here, we could livelock
2267 * because not all segments in the iov can be copied at
2268 * once without a pagefault.
2269 */
2270 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2271 iov_iter_single_seg_count(i));
2272 goto again;
2273 }
Nick Pigginafddba42007-10-16 01:25:01 -07002274 pos += copied;
2275 written += copied;
2276
2277 balance_dirty_pages_ratelimited(mapping);
2278
2279 } while (iov_iter_count(i));
2280
2281 return written ? written : status;
2282}
2283
2284ssize_t
2285generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2286 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2287 size_t count, ssize_t written)
2288{
2289 struct file *file = iocb->ki_filp;
2290 struct address_space *mapping = file->f_mapping;
2291 const struct address_space_operations *a_ops = mapping->a_ops;
2292 struct inode *inode = mapping->host;
2293 ssize_t status;
2294 struct iov_iter i;
2295
2296 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002297 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002298
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002300 written += status;
2301 *ppos = pos + status;
2302
2303 /*
2304 * For now, when the user asks for O_SYNC, we'll actually give
2305 * O_DSYNC
2306 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2308 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2309 status = generic_osync_inode(inode, mapping,
2310 OSYNC_METADATA|OSYNC_DATA);
2311 }
2312 }
2313
2314 /*
2315 * If we get here for O_DIRECT writes then we must have fallen through
2316 * to buffered writes (block instantiation inside i_size). So we sync
2317 * the file data here, to try to honour O_DIRECT expectations.
2318 */
2319 if (unlikely(file->f_flags & O_DIRECT) && written)
Nick Piggin48b47c52009-01-06 14:40:22 -08002320 status = filemap_write_and_wait_range(mapping,
2321 pos, pos + written - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 return written ? written : status;
2324}
2325EXPORT_SYMBOL(generic_file_buffered_write);
2326
Adrian Bunk5ce78522005-09-10 00:26:28 -07002327static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2329 unsigned long nr_segs, loff_t *ppos)
2330{
2331 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002332 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 size_t ocount; /* original count */
2334 size_t count; /* after file limit checks */
2335 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 loff_t pos;
2337 ssize_t written;
2338 ssize_t err;
2339
2340 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002341 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2342 if (err)
2343 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
2345 count = ocount;
2346 pos = *ppos;
2347
2348 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2349
2350 /* We can write back this queue in page reclaim */
2351 current->backing_dev_info = mapping->backing_dev_info;
2352 written = 0;
2353
2354 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2355 if (err)
2356 goto out;
2357
2358 if (count == 0)
2359 goto out;
2360
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002361 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 if (err)
2363 goto out;
2364
Christoph Hellwig870f4812006-01-09 20:52:01 -08002365 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
2367 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2368 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002369 loff_t endbyte;
2370 ssize_t written_buffered;
2371
2372 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2373 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 if (written < 0 || written == count)
2375 goto out;
2376 /*
2377 * direct-io write to a hole: fall through to buffered I/O
2378 * for completing the rest of the request.
2379 */
2380 pos += written;
2381 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002382 written_buffered = generic_file_buffered_write(iocb, iov,
2383 nr_segs, pos, ppos, count,
2384 written);
2385 /*
2386 * If generic_file_buffered_write() retuned a synchronous error
2387 * then we want to return the number of bytes which were
2388 * direct-written, or the error code if that was zero. Note
2389 * that this differs from normal direct-io semantics, which
2390 * will return -EFOO even if some bytes were written.
2391 */
2392 if (written_buffered < 0) {
2393 err = written_buffered;
2394 goto out;
2395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002397 /*
2398 * We need to ensure that the page cache pages are written to
2399 * disk and invalidated to preserve the expected O_DIRECT
2400 * semantics.
2401 */
2402 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002403 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2404 SYNC_FILE_RANGE_WAIT_BEFORE|
2405 SYNC_FILE_RANGE_WRITE|
2406 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002407 if (err == 0) {
2408 written = written_buffered;
2409 invalidate_mapping_pages(mapping,
2410 pos >> PAGE_CACHE_SHIFT,
2411 endbyte >> PAGE_CACHE_SHIFT);
2412 } else {
2413 /*
2414 * We don't know how much we wrote, so just return
2415 * the number of bytes which were direct-written
2416 */
2417 }
2418 } else {
2419 written = generic_file_buffered_write(iocb, iov, nr_segs,
2420 pos, ppos, count, written);
2421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422out:
2423 current->backing_dev_info = NULL;
2424 return written ? written : err;
2425}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
Badari Pulavarty027445c2006-09-30 23:28:46 -07002427ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2428 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429{
2430 struct file *file = iocb->ki_filp;
2431 struct address_space *mapping = file->f_mapping;
2432 struct inode *inode = mapping->host;
2433 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
Badari Pulavarty027445c2006-09-30 23:28:46 -07002435 BUG_ON(iocb->ki_pos != pos);
2436
2437 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2438 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
2440 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002441 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
2443 err = sync_page_range_nolock(inode, mapping, pos, ret);
2444 if (err < 0)
2445 ret = err;
2446 }
2447 return ret;
2448}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002449EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450
Badari Pulavarty027445c2006-09-30 23:28:46 -07002451ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2452 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
2459 BUG_ON(iocb->ki_pos != pos);
2460
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002461 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002462 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2463 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002464 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
2466 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2467 ssize_t err;
2468
2469 err = sync_page_range(inode, mapping, pos, ret);
2470 if (err < 0)
2471 ret = err;
2472 }
2473 return ret;
2474}
2475EXPORT_SYMBOL(generic_file_aio_write);
2476
David Howellscf9a2ae2006-08-29 19:05:54 +01002477/**
2478 * try_to_release_page() - release old fs-specific metadata on a page
2479 *
2480 * @page: the page which the kernel is trying to free
2481 * @gfp_mask: memory allocation flags (and I/O mode)
2482 *
2483 * The address_space is to try to release any data against the page
2484 * (presumably at page->private). If the release was successful, return `1'.
2485 * Otherwise return zero.
2486 *
David Howells266cf652009-04-03 16:42:36 +01002487 * This may also be called if PG_fscache is set on a page, indicating that the
2488 * page is known to the local caching routines.
2489 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002490 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002491 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002492 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002493 */
2494int try_to_release_page(struct page *page, gfp_t gfp_mask)
2495{
2496 struct address_space * const mapping = page->mapping;
2497
2498 BUG_ON(!PageLocked(page));
2499 if (PageWriteback(page))
2500 return 0;
2501
2502 if (mapping && mapping->a_ops->releasepage)
2503 return mapping->a_ops->releasepage(page, gfp_mask);
2504 return try_to_free_buffers(page);
2505}
2506
2507EXPORT_SYMBOL(try_to_release_page);