blob: c6ebd9f912abbfb7298c99fcd7432a3636857fb1 [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>
28#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Carsten Otteceffc072005-06-23 22:05:25 -070033#include "filemap.h"
Nick Piggin0f8053a2006-03-22 00:08:33 -080034#include "internal.h"
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 * FIXME: remove all knowledge of the buffer layer from the core VM
38 */
39#include <linux/buffer_head.h> /* for generic_osync_inode */
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <asm/mman.h>
42
Adrian Bunk5ce78522005-09-10 00:26:28 -070043static ssize_t
44generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
45 loff_t offset, unsigned long nr_segs);
46
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
113 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
114 */
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123}
124
125void remove_from_page_cache(struct page *page)
126{
127 struct address_space *mapping = page->mapping;
128
Matt Mackallcd7619d2005-05-01 08:59:01 -0700129 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
131 write_lock_irq(&mapping->tree_lock);
132 __remove_from_page_cache(page);
133 write_unlock_irq(&mapping->tree_lock);
134}
135
136static int sync_page(void *word)
137{
138 struct address_space *mapping;
139 struct page *page;
140
Andi Kleen07808b72005-11-05 17:25:53 +0100141 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
143 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700144 * page_mapping() is being called without PG_locked held.
145 * Some knowledge of the state and use of the page is used to
146 * reduce the requirements down to a memory barrier.
147 * The danger here is of a stale page_mapping() return value
148 * indicating a struct address_space different from the one it's
149 * associated with when it is associated with one.
150 * After smp_mb(), it's either the correct page_mapping() for
151 * the page, or an old page_mapping() and the page's own
152 * page_mapping() has gone NULL.
153 * The ->sync_page() address_space operation must tolerate
154 * page_mapping() going NULL. By an amazing coincidence,
155 * this comes about because none of the users of the page
156 * in the ->sync_page() methods make essential use of the
157 * page_mapping(), merely passing the page down to the backing
158 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700159 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700160 * of interest. When page_mapping() does go NULL, the entire
161 * call stack gracefully ignores the page and returns.
162 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 */
164 smp_mb();
165 mapping = page_mapping(page);
166 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
167 mapping->a_ops->sync_page(page);
168 io_schedule();
169 return 0;
170}
171
172/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700173 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700174 * @mapping: address space structure to write
175 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800176 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700177 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700179 * Start writeback against all of a mapping's dirty pages that lie
180 * within the byte offsets <start, end> inclusive.
181 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700183 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 * these two operations is that if a dirty page/buffer is encountered, it must
185 * be waited upon, and not just skipped over.
186 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800187int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
188 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
190 int ret;
191 struct writeback_control wbc = {
192 .sync_mode = sync_mode,
193 .nr_to_write = mapping->nrpages * 2,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700194 .range_start = start,
195 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 };
197
198 if (!mapping_cap_writeback_dirty(mapping))
199 return 0;
200
201 ret = do_writepages(mapping, &wbc);
202 return ret;
203}
204
205static inline int __filemap_fdatawrite(struct address_space *mapping,
206 int sync_mode)
207{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700208 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
210
211int filemap_fdatawrite(struct address_space *mapping)
212{
213 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
214}
215EXPORT_SYMBOL(filemap_fdatawrite);
216
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800217static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
218 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
221}
222
Randy Dunlap485bb992006-06-23 02:03:49 -0700223/**
224 * filemap_flush - mostly a non-blocking flush
225 * @mapping: target address_space
226 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 * This is a mostly non-blocking flush. Not suitable for data-integrity
228 * purposes - I/O may not be started against all dirty pages.
229 */
230int filemap_flush(struct address_space *mapping)
231{
232 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
233}
234EXPORT_SYMBOL(filemap_flush);
235
Randy Dunlap485bb992006-06-23 02:03:49 -0700236/**
237 * wait_on_page_writeback_range - wait for writeback to complete
238 * @mapping: target address_space
239 * @start: beginning page index
240 * @end: ending page index
241 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 * Wait for writeback to complete against pages indexed by start->end
243 * inclusive
244 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800245int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 pgoff_t start, pgoff_t end)
247{
248 struct pagevec pvec;
249 int nr_pages;
250 int ret = 0;
251 pgoff_t index;
252
253 if (end < start)
254 return 0;
255
256 pagevec_init(&pvec, 0);
257 index = start;
258 while ((index <= end) &&
259 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
260 PAGECACHE_TAG_WRITEBACK,
261 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
262 unsigned i;
263
264 for (i = 0; i < nr_pages; i++) {
265 struct page *page = pvec.pages[i];
266
267 /* until radix tree lookup accepts end_index */
268 if (page->index > end)
269 continue;
270
271 wait_on_page_writeback(page);
272 if (PageError(page))
273 ret = -EIO;
274 }
275 pagevec_release(&pvec);
276 cond_resched();
277 }
278
279 /* Check for outstanding write errors */
280 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
281 ret = -ENOSPC;
282 if (test_and_clear_bit(AS_EIO, &mapping->flags))
283 ret = -EIO;
284
285 return ret;
286}
287
Randy Dunlap485bb992006-06-23 02:03:49 -0700288/**
289 * sync_page_range - write and wait on all pages in the passed range
290 * @inode: target inode
291 * @mapping: target address_space
292 * @pos: beginning offset in pages to write
293 * @count: number of bytes to write
294 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * Write and wait upon all the pages in the passed range. This is a "data
296 * integrity" operation. It waits upon in-flight writeout before starting and
297 * waiting upon new writeout. If there was an IO error, return it.
298 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800299 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 * it is otherwise livelockable.
301 */
302int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800303 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
305 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
306 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
307 int ret;
308
309 if (!mapping_cap_writeback_dirty(mapping) || !count)
310 return 0;
311 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
312 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800313 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800315 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 }
317 if (ret == 0)
318 ret = wait_on_page_writeback_range(mapping, start, end);
319 return ret;
320}
321EXPORT_SYMBOL(sync_page_range);
322
Randy Dunlap485bb992006-06-23 02:03:49 -0700323/**
324 * sync_page_range_nolock
325 * @inode: target inode
326 * @mapping: target address_space
327 * @pos: beginning offset in pages to write
328 * @count: number of bytes to write
329 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800330 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 * as it forces O_SYNC writers to different parts of the same file
332 * to be serialised right until io completion.
333 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800334int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
335 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
337 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
338 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
339 int ret;
340
341 if (!mapping_cap_writeback_dirty(mapping) || !count)
342 return 0;
343 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
344 if (ret == 0)
345 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
346 if (ret == 0)
347 ret = wait_on_page_writeback_range(mapping, start, end);
348 return ret;
349}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800350EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700353 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700355 *
356 * Walk the list of under-writeback pages of the given address space
357 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 */
359int filemap_fdatawait(struct address_space *mapping)
360{
361 loff_t i_size = i_size_read(mapping->host);
362
363 if (i_size == 0)
364 return 0;
365
366 return wait_on_page_writeback_range(mapping, 0,
367 (i_size - 1) >> PAGE_CACHE_SHIFT);
368}
369EXPORT_SYMBOL(filemap_fdatawait);
370
371int filemap_write_and_wait(struct address_space *mapping)
372{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800373 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800376 err = filemap_fdatawrite(mapping);
377 /*
378 * Even if the above returned error, the pages may be
379 * written partially (e.g. -ENOSPC), so we wait for it.
380 * But the -EIO is special case, it may indicate the worst
381 * thing (e.g. bug) happened, so we avoid waiting for it.
382 */
383 if (err != -EIO) {
384 int err2 = filemap_fdatawait(mapping);
385 if (!err)
386 err = err2;
387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800389 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800391EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Randy Dunlap485bb992006-06-23 02:03:49 -0700393/**
394 * filemap_write_and_wait_range - write out & wait on a file range
395 * @mapping: the address_space for the pages
396 * @lstart: offset in bytes where the range starts
397 * @lend: offset in bytes where the range ends (inclusive)
398 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800399 * Write out and wait upon file offsets lstart->lend, inclusive.
400 *
401 * Note that `lend' is inclusive (describes the last byte to be written) so
402 * that this function can be used to write to the very end-of-file (end = -1).
403 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404int filemap_write_and_wait_range(struct address_space *mapping,
405 loff_t lstart, loff_t lend)
406{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800407 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
409 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800410 err = __filemap_fdatawrite_range(mapping, lstart, lend,
411 WB_SYNC_ALL);
412 /* See comment of filemap_write_and_wait() */
413 if (err != -EIO) {
414 int err2 = wait_on_page_writeback_range(mapping,
415 lstart >> PAGE_CACHE_SHIFT,
416 lend >> PAGE_CACHE_SHIFT);
417 if (!err)
418 err = err2;
419 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800421 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
423
Randy Dunlap485bb992006-06-23 02:03:49 -0700424/**
425 * add_to_page_cache - add newly allocated pagecache pages
426 * @page: page to add
427 * @mapping: the page's address_space
428 * @offset: page index
429 * @gfp_mask: page allocation mode
430 *
431 * This function is used to add newly allocated pagecache pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 * the page is new, so we can just run SetPageLocked() against it.
433 * The other page state flags were set by rmqueue().
434 *
435 * This function does not add the page to the LRU. The caller must do that.
436 */
437int add_to_page_cache(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400438 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
440 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
441
442 if (error == 0) {
443 write_lock_irq(&mapping->tree_lock);
444 error = radix_tree_insert(&mapping->page_tree, offset, page);
445 if (!error) {
446 page_cache_get(page);
447 SetPageLocked(page);
448 page->mapping = mapping;
449 page->index = offset;
450 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700451 __inc_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 }
453 write_unlock_irq(&mapping->tree_lock);
454 radix_tree_preload_end();
455 }
456 return error;
457}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458EXPORT_SYMBOL(add_to_page_cache);
459
460int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400461 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
463 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
464 if (ret == 0)
465 lru_cache_add(page);
466 return ret;
467}
468
Paul Jackson44110fe2006-03-24 03:16:04 -0800469#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700470struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800471{
472 if (cpuset_do_page_mem_spread()) {
473 int n = cpuset_mem_spread_node();
Nick Piggin2ae88142006-10-28 10:38:23 -0700474 return alloc_pages_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800475 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700476 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800477}
Nick Piggin2ae88142006-10-28 10:38:23 -0700478EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800479#endif
480
Nick Piggindb376482006-09-25 23:31:24 -0700481static int __sleep_on_page_lock(void *word)
482{
483 io_schedule();
484 return 0;
485}
486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487/*
488 * In order to wait for pages to become available there must be
489 * waitqueues associated with pages. By using a hash table of
490 * waitqueues where the bucket discipline is to maintain all
491 * waiters on the same queue and wake all when any of the pages
492 * become available, and for the woken contexts to check to be
493 * sure the appropriate page became available, this saves space
494 * at a cost of "thundering herd" phenomena during rare hash
495 * collisions.
496 */
497static wait_queue_head_t *page_waitqueue(struct page *page)
498{
499 const struct zone *zone = page_zone(page);
500
501 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
502}
503
504static inline void wake_up_page(struct page *page, int bit)
505{
506 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
507}
508
509void fastcall wait_on_page_bit(struct page *page, int bit_nr)
510{
511 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
512
513 if (test_bit(bit_nr, &page->flags))
514 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
515 TASK_UNINTERRUPTIBLE);
516}
517EXPORT_SYMBOL(wait_on_page_bit);
518
519/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700520 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 * @page: the page
522 *
523 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
524 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
525 * mechananism between PageLocked pages and PageWriteback pages is shared.
526 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
527 *
528 * The first mb is necessary to safely close the critical section opened by the
529 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
530 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
531 * parallel wait_on_page_locked()).
532 */
533void fastcall unlock_page(struct page *page)
534{
535 smp_mb__before_clear_bit();
536 if (!TestClearPageLocked(page))
537 BUG();
538 smp_mb__after_clear_bit();
539 wake_up_page(page, PG_locked);
540}
541EXPORT_SYMBOL(unlock_page);
542
Randy Dunlap485bb992006-06-23 02:03:49 -0700543/**
544 * end_page_writeback - end writeback against a page
545 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 */
547void end_page_writeback(struct page *page)
548{
549 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
550 if (!test_clear_page_writeback(page))
551 BUG();
552 }
553 smp_mb__after_clear_bit();
554 wake_up_page(page, PG_writeback);
555}
556EXPORT_SYMBOL(end_page_writeback);
557
Randy Dunlap485bb992006-06-23 02:03:49 -0700558/**
559 * __lock_page - get a lock on the page, assuming we need to sleep to get it
560 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700562 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
564 * chances are that on the second loop, the block layer's plug list is empty,
565 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
566 */
567void fastcall __lock_page(struct page *page)
568{
569 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
570
571 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
572 TASK_UNINTERRUPTIBLE);
573}
574EXPORT_SYMBOL(__lock_page);
575
Nick Piggindb376482006-09-25 23:31:24 -0700576/*
577 * Variant of lock_page that does not require the caller to hold a reference
578 * on the page's mapping.
579 */
580void fastcall __lock_page_nosync(struct page *page)
581{
582 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
583 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
584 TASK_UNINTERRUPTIBLE);
585}
586
Randy Dunlap485bb992006-06-23 02:03:49 -0700587/**
588 * find_get_page - find and get a page reference
589 * @mapping: the address_space to search
590 * @offset: the page index
591 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700592 * Is there a pagecache struct page at the given (mapping, offset) tuple?
593 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 */
595struct page * find_get_page(struct address_space *mapping, unsigned long offset)
596{
597 struct page *page;
598
599 read_lock_irq(&mapping->tree_lock);
600 page = radix_tree_lookup(&mapping->page_tree, offset);
601 if (page)
602 page_cache_get(page);
603 read_unlock_irq(&mapping->tree_lock);
604 return page;
605}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606EXPORT_SYMBOL(find_get_page);
607
Randy Dunlap485bb992006-06-23 02:03:49 -0700608/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700610 * @mapping: the address_space to search
611 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 *
613 * Locates the desired pagecache page, locks it, increments its reference
614 * count and returns its address.
615 *
616 * Returns zero if the page was not present. find_lock_page() may sleep.
617 */
618struct page *find_lock_page(struct address_space *mapping,
619 unsigned long offset)
620{
621 struct page *page;
622
623 read_lock_irq(&mapping->tree_lock);
624repeat:
625 page = radix_tree_lookup(&mapping->page_tree, offset);
626 if (page) {
627 page_cache_get(page);
628 if (TestSetPageLocked(page)) {
629 read_unlock_irq(&mapping->tree_lock);
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800630 __lock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 read_lock_irq(&mapping->tree_lock);
632
633 /* Has the page been truncated while we slept? */
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800634 if (unlikely(page->mapping != mapping ||
635 page->index != offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 unlock_page(page);
637 page_cache_release(page);
638 goto repeat;
639 }
640 }
641 }
642 read_unlock_irq(&mapping->tree_lock);
643 return page;
644}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645EXPORT_SYMBOL(find_lock_page);
646
647/**
648 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700649 * @mapping: the page's address_space
650 * @index: the page's index into the mapping
651 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 *
653 * Locates a page in the pagecache. If the page is not present, a new page
654 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
655 * LRU list. The returned page is locked and has its reference count
656 * incremented.
657 *
658 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
659 * allocation!
660 *
661 * find_or_create_page() returns the desired page's address, or zero on
662 * memory exhaustion.
663 */
664struct page *find_or_create_page(struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400665 unsigned long index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 struct page *page, *cached_page = NULL;
668 int err;
669repeat:
670 page = find_lock_page(mapping, index);
671 if (!page) {
672 if (!cached_page) {
Christoph Lameter43c0f3d2007-05-15 23:57:09 -0700673 cached_page =
674 __page_cache_alloc(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 if (!cached_page)
676 return NULL;
677 }
678 err = add_to_page_cache_lru(cached_page, mapping,
679 index, gfp_mask);
680 if (!err) {
681 page = cached_page;
682 cached_page = NULL;
683 } else if (err == -EEXIST)
684 goto repeat;
685 }
686 if (cached_page)
687 page_cache_release(cached_page);
688 return page;
689}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690EXPORT_SYMBOL(find_or_create_page);
691
692/**
693 * find_get_pages - gang pagecache lookup
694 * @mapping: The address_space to search
695 * @start: The starting page index
696 * @nr_pages: The maximum number of pages
697 * @pages: Where the resulting pages are placed
698 *
699 * find_get_pages() will search for and return a group of up to
700 * @nr_pages pages in the mapping. The pages are placed at @pages.
701 * find_get_pages() takes a reference against the returned pages.
702 *
703 * The search returns a group of mapping-contiguous pages with ascending
704 * indexes. There may be holes in the indices due to not-present pages.
705 *
706 * find_get_pages() returns the number of pages which were found.
707 */
708unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
709 unsigned int nr_pages, struct page **pages)
710{
711 unsigned int i;
712 unsigned int ret;
713
714 read_lock_irq(&mapping->tree_lock);
715 ret = radix_tree_gang_lookup(&mapping->page_tree,
716 (void **)pages, start, nr_pages);
717 for (i = 0; i < ret; i++)
718 page_cache_get(pages[i]);
719 read_unlock_irq(&mapping->tree_lock);
720 return ret;
721}
722
Jens Axboeebf43502006-04-27 08:46:01 +0200723/**
724 * find_get_pages_contig - gang contiguous pagecache lookup
725 * @mapping: The address_space to search
726 * @index: The starting page index
727 * @nr_pages: The maximum number of pages
728 * @pages: Where the resulting pages are placed
729 *
730 * find_get_pages_contig() works exactly like find_get_pages(), except
731 * that the returned number of pages are guaranteed to be contiguous.
732 *
733 * find_get_pages_contig() returns the number of pages which were found.
734 */
735unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
736 unsigned int nr_pages, struct page **pages)
737{
738 unsigned int i;
739 unsigned int ret;
740
741 read_lock_irq(&mapping->tree_lock);
742 ret = radix_tree_gang_lookup(&mapping->page_tree,
743 (void **)pages, index, nr_pages);
744 for (i = 0; i < ret; i++) {
745 if (pages[i]->mapping == NULL || pages[i]->index != index)
746 break;
747
748 page_cache_get(pages[i]);
749 index++;
750 }
751 read_unlock_irq(&mapping->tree_lock);
752 return i;
753}
David Howellsef71c152007-05-09 02:33:44 -0700754EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200755
Randy Dunlap485bb992006-06-23 02:03:49 -0700756/**
757 * find_get_pages_tag - find and return pages that match @tag
758 * @mapping: the address_space to search
759 * @index: the starting page index
760 * @tag: the tag index
761 * @nr_pages: the maximum number of pages
762 * @pages: where the resulting pages are placed
763 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700765 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 */
767unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
768 int tag, unsigned int nr_pages, struct page **pages)
769{
770 unsigned int i;
771 unsigned int ret;
772
773 read_lock_irq(&mapping->tree_lock);
774 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
775 (void **)pages, *index, nr_pages, tag);
776 for (i = 0; i < ret; i++)
777 page_cache_get(pages[i]);
778 if (ret)
779 *index = pages[ret - 1]->index + 1;
780 read_unlock_irq(&mapping->tree_lock);
781 return ret;
782}
David Howellsef71c152007-05-09 02:33:44 -0700783EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Randy Dunlap485bb992006-06-23 02:03:49 -0700785/**
786 * grab_cache_page_nowait - returns locked page at given index in given cache
787 * @mapping: target address_space
788 * @index: the page index
789 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800790 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 * This is intended for speculative data generators, where the data can
792 * be regenerated if the page couldn't be grabbed. This routine should
793 * be safe to call while holding the lock for another page.
794 *
795 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
796 * and deadlock against the caller's locked page.
797 */
798struct page *
799grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
800{
801 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
803 if (page) {
804 if (!TestSetPageLocked(page))
805 return page;
806 page_cache_release(page);
807 return NULL;
808 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700809 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
810 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 page_cache_release(page);
812 page = NULL;
813 }
814 return page;
815}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816EXPORT_SYMBOL(grab_cache_page_nowait);
817
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700818/*
819 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
820 * a _large_ part of the i/o request. Imagine the worst scenario:
821 *
822 * ---R__________________________________________B__________
823 * ^ reading here ^ bad block(assume 4k)
824 *
825 * read(R) => miss => readahead(R...B) => media error => frustrating retries
826 * => failing the whole request => read(R) => read(R+1) =>
827 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
828 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
829 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
830 *
831 * It is going insane. Fix it by quickly scaling down the readahead size.
832 */
833static void shrink_readahead_size_eio(struct file *filp,
834 struct file_ra_state *ra)
835{
836 if (!ra->ra_pages)
837 return;
838
839 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700840}
841
Randy Dunlap485bb992006-06-23 02:03:49 -0700842/**
843 * do_generic_mapping_read - generic file read routine
844 * @mapping: address_space to be read
845 * @_ra: file's readahead state
846 * @filp: the file to read
847 * @ppos: current file position
848 * @desc: read_descriptor
849 * @actor: read method
850 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700852 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 *
854 * This is really ugly. But the goto's actually try to clarify some
855 * of the logic when it comes to error handling etc.
856 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700857 * Note the struct file* is only passed for the use of readpage.
858 * It may be NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 */
860void do_generic_mapping_read(struct address_space *mapping,
861 struct file_ra_state *_ra,
862 struct file *filp,
863 loff_t *ppos,
864 read_descriptor_t *desc,
865 read_actor_t actor)
866{
867 struct inode *inode = mapping->host;
868 unsigned long index;
869 unsigned long end_index;
870 unsigned long offset;
871 unsigned long last_index;
872 unsigned long next_index;
873 unsigned long prev_index;
Jan Karaec0f1632007-05-06 14:49:25 -0700874 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 loff_t isize;
876 struct page *cached_page;
877 int error;
878 struct file_ra_state ra = *_ra;
879
880 cached_page = NULL;
881 index = *ppos >> PAGE_CACHE_SHIFT;
882 next_index = index;
Jan Kara6ce745e2007-05-06 14:49:26 -0700883 prev_index = ra.prev_index;
884 prev_offset = ra.prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
886 offset = *ppos & ~PAGE_CACHE_MASK;
887
888 isize = i_size_read(inode);
889 if (!isize)
890 goto out;
891
892 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
893 for (;;) {
894 struct page *page;
895 unsigned long nr, ret;
896
897 /* nr is the maximum number of bytes to copy from this page */
898 nr = PAGE_CACHE_SIZE;
899 if (index >= end_index) {
900 if (index > end_index)
901 goto out;
902 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
903 if (nr <= offset) {
904 goto out;
905 }
906 }
907 nr = nr - offset;
908
909 cond_resched();
910 if (index == next_index)
911 next_index = page_cache_readahead(mapping, &ra, filp,
912 index, last_index - index);
913
914find_page:
915 page = find_get_page(mapping, index);
916 if (unlikely(page == NULL)) {
917 handle_ra_miss(mapping, &ra, index);
918 goto no_cached_page;
919 }
920 if (!PageUptodate(page))
921 goto page_not_up_to_date;
922page_ok:
923
924 /* If users can be writing to this page using arbitrary
925 * virtual addresses, take care about potential aliasing
926 * before reading the page on the kernel side.
927 */
928 if (mapping_writably_mapped(mapping))
929 flush_dcache_page(page);
930
931 /*
Jan Karaec0f1632007-05-06 14:49:25 -0700932 * When a sequential read accesses a page several times,
933 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 */
Jan Karaec0f1632007-05-06 14:49:25 -0700935 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 mark_page_accessed(page);
937 prev_index = index;
938
939 /*
940 * Ok, we have the page, and it's up-to-date, so
941 * now we can copy it to user space...
942 *
943 * The actor routine returns how many bytes were actually used..
944 * NOTE! This may not be the same as how much of a user buffer
945 * we filled up (we may be padding etc), so we can only update
946 * "pos" here (the actor routine has to update the user buffer
947 * pointers and the remaining count).
948 */
949 ret = actor(desc, page, offset, nr);
950 offset += ret;
951 index += offset >> PAGE_CACHE_SHIFT;
952 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -0700953 prev_offset = offset;
954 ra.prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
956 page_cache_release(page);
957 if (ret == nr && desc->count)
958 continue;
959 goto out;
960
961page_not_up_to_date:
962 /* Get exclusive access to the page ... */
963 lock_page(page);
964
Nick Pigginda6052f2006-09-25 23:31:35 -0700965 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (!page->mapping) {
967 unlock_page(page);
968 page_cache_release(page);
969 continue;
970 }
971
972 /* Did somebody else fill it already? */
973 if (PageUptodate(page)) {
974 unlock_page(page);
975 goto page_ok;
976 }
977
978readpage:
979 /* Start the actual read. The read will unlock the page. */
980 error = mapping->a_ops->readpage(filp, page);
981
Zach Brown994fc28c2005-12-15 14:28:17 -0800982 if (unlikely(error)) {
983 if (error == AOP_TRUNCATED_PAGE) {
984 page_cache_release(page);
985 goto find_page;
986 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -0800988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 if (!PageUptodate(page)) {
991 lock_page(page);
992 if (!PageUptodate(page)) {
993 if (page->mapping == NULL) {
994 /*
995 * invalidate_inode_pages got it
996 */
997 unlock_page(page);
998 page_cache_release(page);
999 goto find_page;
1000 }
1001 unlock_page(page);
1002 error = -EIO;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001003 shrink_readahead_size_eio(filp, &ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 goto readpage_error;
1005 }
1006 unlock_page(page);
1007 }
1008
1009 /*
1010 * i_size must be checked after we have done ->readpage.
1011 *
1012 * Checking i_size after the readpage allows us to calculate
1013 * the correct value for "nr", which means the zero-filled
1014 * part of the page is not copied back to userspace (unless
1015 * another truncate extends the file - this is desired though).
1016 */
1017 isize = i_size_read(inode);
1018 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1019 if (unlikely(!isize || index > end_index)) {
1020 page_cache_release(page);
1021 goto out;
1022 }
1023
1024 /* nr is the maximum number of bytes to copy from this page */
1025 nr = PAGE_CACHE_SIZE;
1026 if (index == end_index) {
1027 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1028 if (nr <= offset) {
1029 page_cache_release(page);
1030 goto out;
1031 }
1032 }
1033 nr = nr - offset;
1034 goto page_ok;
1035
1036readpage_error:
1037 /* UHHUH! A synchronous read error occurred. Report it */
1038 desc->error = error;
1039 page_cache_release(page);
1040 goto out;
1041
1042no_cached_page:
1043 /*
1044 * Ok, it wasn't cached, so we need to create a new
1045 * page..
1046 */
1047 if (!cached_page) {
1048 cached_page = page_cache_alloc_cold(mapping);
1049 if (!cached_page) {
1050 desc->error = -ENOMEM;
1051 goto out;
1052 }
1053 }
1054 error = add_to_page_cache_lru(cached_page, mapping,
1055 index, GFP_KERNEL);
1056 if (error) {
1057 if (error == -EEXIST)
1058 goto find_page;
1059 desc->error = error;
1060 goto out;
1061 }
1062 page = cached_page;
1063 cached_page = NULL;
1064 goto readpage;
1065 }
1066
1067out:
1068 *_ra = ra;
1069
1070 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1071 if (cached_page)
1072 page_cache_release(cached_page);
1073 if (filp)
1074 file_accessed(filp);
1075}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076EXPORT_SYMBOL(do_generic_mapping_read);
1077
1078int file_read_actor(read_descriptor_t *desc, struct page *page,
1079 unsigned long offset, unsigned long size)
1080{
1081 char *kaddr;
1082 unsigned long left, count = desc->count;
1083
1084 if (size > count)
1085 size = count;
1086
1087 /*
1088 * Faults on the destination of a read are common, so do it before
1089 * taking the kmap.
1090 */
1091 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1092 kaddr = kmap_atomic(page, KM_USER0);
1093 left = __copy_to_user_inatomic(desc->arg.buf,
1094 kaddr + offset, size);
1095 kunmap_atomic(kaddr, KM_USER0);
1096 if (left == 0)
1097 goto success;
1098 }
1099
1100 /* Do it the slow way */
1101 kaddr = kmap(page);
1102 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1103 kunmap(page);
1104
1105 if (left) {
1106 size -= left;
1107 desc->error = -EFAULT;
1108 }
1109success:
1110 desc->count = count - size;
1111 desc->written += size;
1112 desc->arg.buf += size;
1113 return size;
1114}
1115
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001116/*
1117 * Performs necessary checks before doing a write
1118 * @iov: io vector request
1119 * @nr_segs: number of segments in the iovec
1120 * @count: number of bytes to write
1121 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1122 *
1123 * Adjust number of segments and amount of bytes to write (nr_segs should be
1124 * properly initialized first). Returns appropriate error code that caller
1125 * should return or zero in case that write should be allowed.
1126 */
1127int generic_segment_checks(const struct iovec *iov,
1128 unsigned long *nr_segs, size_t *count, int access_flags)
1129{
1130 unsigned long seg;
1131 size_t cnt = 0;
1132 for (seg = 0; seg < *nr_segs; seg++) {
1133 const struct iovec *iv = &iov[seg];
1134
1135 /*
1136 * If any segment has a negative length, or the cumulative
1137 * length ever wraps negative then return -EINVAL.
1138 */
1139 cnt += iv->iov_len;
1140 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1141 return -EINVAL;
1142 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1143 continue;
1144 if (seg == 0)
1145 return -EFAULT;
1146 *nr_segs = seg;
1147 cnt -= iv->iov_len; /* This segment is no good */
1148 break;
1149 }
1150 *count = cnt;
1151 return 0;
1152}
1153EXPORT_SYMBOL(generic_segment_checks);
1154
Randy Dunlap485bb992006-06-23 02:03:49 -07001155/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001156 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001157 * @iocb: kernel I/O control block
1158 * @iov: io vector request
1159 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001160 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001161 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 * This is the "read()" routine for all filesystems
1163 * that can use the page cache directly.
1164 */
1165ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001166generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1167 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
1169 struct file *filp = iocb->ki_filp;
1170 ssize_t retval;
1171 unsigned long seg;
1172 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001173 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001176 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1177 if (retval)
1178 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1181 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001182 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 struct address_space *mapping;
1184 struct inode *inode;
1185
1186 mapping = filp->f_mapping;
1187 inode = mapping->host;
1188 retval = 0;
1189 if (!count)
1190 goto out; /* skip atime */
1191 size = i_size_read(inode);
1192 if (pos < size) {
1193 retval = generic_file_direct_IO(READ, iocb,
1194 iov, pos, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 if (retval > 0)
1196 *ppos = pos + retval;
1197 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001198 if (likely(retval != 0)) {
Steven Whitehouse3f1a9aa2006-09-27 14:52:48 -04001199 file_accessed(filp);
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001200 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 }
1203
1204 retval = 0;
1205 if (count) {
1206 for (seg = 0; seg < nr_segs; seg++) {
1207 read_descriptor_t desc;
1208
1209 desc.written = 0;
1210 desc.arg.buf = iov[seg].iov_base;
1211 desc.count = iov[seg].iov_len;
1212 if (desc.count == 0)
1213 continue;
1214 desc.error = 0;
1215 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1216 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001217 if (desc.error) {
1218 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 break;
1220 }
1221 }
1222 }
1223out:
1224 return retval;
1225}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226EXPORT_SYMBOL(generic_file_aio_read);
1227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1229{
1230 ssize_t written;
1231 unsigned long count = desc->count;
1232 struct file *file = desc->arg.data;
1233
1234 if (size > count)
1235 size = count;
1236
1237 written = file->f_op->sendpage(file, page, offset,
1238 size, &file->f_pos, size<count);
1239 if (written < 0) {
1240 desc->error = written;
1241 written = 0;
1242 }
1243 desc->count = count - written;
1244 desc->written += written;
1245 return written;
1246}
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248static ssize_t
1249do_readahead(struct address_space *mapping, struct file *filp,
1250 unsigned long index, unsigned long nr)
1251{
1252 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1253 return -EINVAL;
1254
1255 force_page_cache_readahead(mapping, filp, index,
1256 max_sane_readahead(nr));
1257 return 0;
1258}
1259
1260asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1261{
1262 ssize_t ret;
1263 struct file *file;
1264
1265 ret = -EBADF;
1266 file = fget(fd);
1267 if (file) {
1268 if (file->f_mode & FMODE_READ) {
1269 struct address_space *mapping = file->f_mapping;
1270 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1271 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1272 unsigned long len = end - start + 1;
1273 ret = do_readahead(mapping, file, start, len);
1274 }
1275 fput(file);
1276 }
1277 return ret;
1278}
1279
1280#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001281static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1282/**
1283 * page_cache_read - adds requested page to the page cache if not already there
1284 * @file: file to read
1285 * @offset: page index
1286 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 * This adds the requested page to the page cache if it isn't already there,
1288 * and schedules an I/O to read in its contents from disk.
1289 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290static int fastcall page_cache_read(struct file * file, unsigned long offset)
1291{
1292 struct address_space *mapping = file->f_mapping;
1293 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001294 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Zach Brown994fc28c2005-12-15 14:28:17 -08001296 do {
1297 page = page_cache_alloc_cold(mapping);
1298 if (!page)
1299 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Zach Brown994fc28c2005-12-15 14:28:17 -08001301 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1302 if (ret == 0)
1303 ret = mapping->a_ops->readpage(file, page);
1304 else if (ret == -EEXIST)
1305 ret = 0; /* losing race to add is OK */
1306
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Zach Brown994fc28c2005-12-15 14:28:17 -08001309 } while (ret == AOP_TRUNCATED_PAGE);
1310
1311 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312}
1313
1314#define MMAP_LOTSAMISS (100)
1315
Randy Dunlap485bb992006-06-23 02:03:49 -07001316/**
1317 * filemap_nopage - read in file data for page fault handling
1318 * @area: the applicable vm_area
1319 * @address: target address to read in
1320 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1321 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 * filemap_nopage() is invoked via the vma operations vector for a
1323 * mapped memory region to read in file data during a page fault.
1324 *
1325 * The goto's are kind of ugly, but this streamlines the normal case of having
1326 * it in the page cache, and handles the special cases reasonably without
1327 * having a lot of duplicated code.
1328 */
1329struct page *filemap_nopage(struct vm_area_struct *area,
1330 unsigned long address, int *type)
1331{
1332 int error;
1333 struct file *file = area->vm_file;
1334 struct address_space *mapping = file->f_mapping;
1335 struct file_ra_state *ra = &file->f_ra;
1336 struct inode *inode = mapping->host;
1337 struct page *page;
1338 unsigned long size, pgoff;
1339 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1340
1341 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1342
1343retry_all:
1344 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1345 if (pgoff >= size)
1346 goto outside_data_content;
1347
1348 /* If we don't want any read-ahead, don't bother */
1349 if (VM_RandomReadHint(area))
1350 goto no_cached_page;
1351
1352 /*
1353 * The readahead code wants to be told about each and every page
1354 * so it can build and shrink its windows appropriately
1355 *
1356 * For sequential accesses, we use the generic readahead logic.
1357 */
1358 if (VM_SequentialReadHint(area))
1359 page_cache_readahead(mapping, ra, file, pgoff, 1);
1360
1361 /*
1362 * Do we have something in the page cache already?
1363 */
1364retry_find:
1365 page = find_get_page(mapping, pgoff);
1366 if (!page) {
1367 unsigned long ra_pages;
1368
1369 if (VM_SequentialReadHint(area)) {
1370 handle_ra_miss(mapping, ra, pgoff);
1371 goto no_cached_page;
1372 }
1373 ra->mmap_miss++;
1374
1375 /*
1376 * Do we miss much more than hit in this file? If so,
1377 * stop bothering with read-ahead. It will only hurt.
1378 */
1379 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1380 goto no_cached_page;
1381
1382 /*
1383 * To keep the pgmajfault counter straight, we need to
1384 * check did_readaround, as this is an inner loop.
1385 */
1386 if (!did_readaround) {
1387 majmin = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001388 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 }
1390 did_readaround = 1;
1391 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1392 if (ra_pages) {
1393 pgoff_t start = 0;
1394
1395 if (pgoff > ra_pages / 2)
1396 start = pgoff - ra_pages / 2;
1397 do_page_cache_readahead(mapping, file, start, ra_pages);
1398 }
1399 page = find_get_page(mapping, pgoff);
1400 if (!page)
1401 goto no_cached_page;
1402 }
1403
1404 if (!did_readaround)
1405 ra->mmap_hit++;
1406
1407 /*
1408 * Ok, found a page in the page cache, now we need to check
1409 * that it's up-to-date.
1410 */
1411 if (!PageUptodate(page))
1412 goto page_not_uptodate;
1413
1414success:
1415 /*
1416 * Found the page and have a reference on it.
1417 */
1418 mark_page_accessed(page);
1419 if (type)
1420 *type = majmin;
1421 return page;
1422
1423outside_data_content:
1424 /*
1425 * An external ptracer can access pages that normally aren't
1426 * accessible..
1427 */
1428 if (area->vm_mm == current->mm)
Adam Litke79f5acf2006-09-29 01:58:43 -07001429 return NOPAGE_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 /* Fall through to the non-read-ahead case */
1431no_cached_page:
1432 /*
1433 * We're only likely to ever get here if MADV_RANDOM is in
1434 * effect.
1435 */
1436 error = page_cache_read(file, pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
1438 /*
1439 * The page we want has now been added to the page cache.
1440 * In the unlikely event that someone removed it in the
1441 * meantime, we'll just come back here and read it again.
1442 */
1443 if (error >= 0)
1444 goto retry_find;
1445
1446 /*
1447 * An error return from page_cache_read can result if the
1448 * system is low on memory, or a problem occurs while trying
1449 * to schedule I/O.
1450 */
1451 if (error == -ENOMEM)
1452 return NOPAGE_OOM;
Adam Litke79f5acf2006-09-29 01:58:43 -07001453 return NOPAGE_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
1455page_not_uptodate:
1456 if (!did_readaround) {
1457 majmin = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001458 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
1461 /*
1462 * Umm, take care of errors if the page isn't up-to-date.
1463 * Try to re-read it _once_. We do this synchronously,
1464 * because there really aren't any performance issues here
1465 * and we need to check for errors.
1466 */
1467 lock_page(page);
1468
1469 /* Somebody truncated the page on us? */
1470 if (!page->mapping) {
1471 unlock_page(page);
1472 page_cache_release(page);
1473 goto retry_all;
1474 }
1475
1476 /* Somebody else successfully read it in? */
1477 if (PageUptodate(page)) {
1478 unlock_page(page);
1479 goto success;
1480 }
1481 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001482 error = mapping->a_ops->readpage(file, page);
1483 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 wait_on_page_locked(page);
1485 if (PageUptodate(page))
1486 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001487 } else if (error == AOP_TRUNCATED_PAGE) {
1488 page_cache_release(page);
1489 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 }
1491
1492 /*
1493 * Things didn't work out. Return zero to tell the
1494 * mm layer so, possibly freeing the page cache page first.
1495 */
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001496 shrink_readahead_size_eio(file, ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 page_cache_release(page);
Adam Litke79f5acf2006-09-29 01:58:43 -07001498 return NOPAGE_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500EXPORT_SYMBOL(filemap_nopage);
1501
1502static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1503 int nonblock)
1504{
1505 struct address_space *mapping = file->f_mapping;
1506 struct page *page;
1507 int error;
1508
1509 /*
1510 * Do we have something in the page cache already?
1511 */
1512retry_find:
1513 page = find_get_page(mapping, pgoff);
1514 if (!page) {
1515 if (nonblock)
1516 return NULL;
1517 goto no_cached_page;
1518 }
1519
1520 /*
1521 * Ok, found a page in the page cache, now we need to check
1522 * that it's up-to-date.
1523 */
Jeff Moyerd3457342005-04-16 15:24:05 -07001524 if (!PageUptodate(page)) {
1525 if (nonblock) {
1526 page_cache_release(page);
1527 return NULL;
1528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 goto page_not_uptodate;
Jeff Moyerd3457342005-04-16 15:24:05 -07001530 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
1532success:
1533 /*
1534 * Found the page and have a reference on it.
1535 */
1536 mark_page_accessed(page);
1537 return page;
1538
1539no_cached_page:
1540 error = page_cache_read(file, pgoff);
1541
1542 /*
1543 * The page we want has now been added to the page cache.
1544 * In the unlikely event that someone removed it in the
1545 * meantime, we'll just come back here and read it again.
1546 */
1547 if (error >= 0)
1548 goto retry_find;
1549
1550 /*
1551 * An error return from page_cache_read can result if the
1552 * system is low on memory, or a problem occurs while trying
1553 * to schedule I/O.
1554 */
1555 return NULL;
1556
1557page_not_uptodate:
1558 lock_page(page);
1559
Nick Pigginda6052f2006-09-25 23:31:35 -07001560 /* Did it get truncated while we waited for it? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (!page->mapping) {
1562 unlock_page(page);
1563 goto err;
1564 }
1565
1566 /* Did somebody else get it up-to-date? */
1567 if (PageUptodate(page)) {
1568 unlock_page(page);
1569 goto success;
1570 }
1571
Zach Brown994fc28c2005-12-15 14:28:17 -08001572 error = mapping->a_ops->readpage(file, page);
1573 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 wait_on_page_locked(page);
1575 if (PageUptodate(page))
1576 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001577 } else if (error == AOP_TRUNCATED_PAGE) {
1578 page_cache_release(page);
1579 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581
1582 /*
1583 * Umm, take care of errors if the page isn't up-to-date.
1584 * Try to re-read it _once_. We do this synchronously,
1585 * because there really aren't any performance issues here
1586 * and we need to check for errors.
1587 */
1588 lock_page(page);
1589
1590 /* Somebody truncated the page on us? */
1591 if (!page->mapping) {
1592 unlock_page(page);
1593 goto err;
1594 }
1595 /* Somebody else successfully read it in? */
1596 if (PageUptodate(page)) {
1597 unlock_page(page);
1598 goto success;
1599 }
1600
1601 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001602 error = mapping->a_ops->readpage(file, page);
1603 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 wait_on_page_locked(page);
1605 if (PageUptodate(page))
1606 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001607 } else if (error == AOP_TRUNCATED_PAGE) {
1608 page_cache_release(page);
1609 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 }
1611
1612 /*
1613 * Things didn't work out. Return zero to tell the
1614 * mm layer so, possibly freeing the page cache page first.
1615 */
1616err:
1617 page_cache_release(page);
1618
1619 return NULL;
1620}
1621
1622int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1623 unsigned long len, pgprot_t prot, unsigned long pgoff,
1624 int nonblock)
1625{
1626 struct file *file = vma->vm_file;
1627 struct address_space *mapping = file->f_mapping;
1628 struct inode *inode = mapping->host;
1629 unsigned long size;
1630 struct mm_struct *mm = vma->vm_mm;
1631 struct page *page;
1632 int err;
1633
1634 if (!nonblock)
1635 force_page_cache_readahead(mapping, vma->vm_file,
1636 pgoff, len >> PAGE_CACHE_SHIFT);
1637
1638repeat:
1639 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1640 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1641 return -EINVAL;
1642
1643 page = filemap_getpage(file, pgoff, nonblock);
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001644
1645 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1646 * done in shmem_populate calling shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 if (!page && !nonblock)
1648 return -ENOMEM;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 if (page) {
1651 err = install_page(mm, vma, addr, page, prot);
1652 if (err) {
1653 page_cache_release(page);
1654 return err;
1655 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001656 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001657 /* No page was found just because we can't read it in now (being
1658 * here implies nonblock != 0), but the page may exist, so set
1659 * the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 err = install_file_pte(mm, vma, addr, pgoff, prot);
1661 if (err)
1662 return err;
1663 }
1664
1665 len -= PAGE_SIZE;
1666 addr += PAGE_SIZE;
1667 pgoff++;
1668 if (len)
1669 goto repeat;
1670
1671 return 0;
1672}
Nikita Danilovb1459462005-10-29 18:17:02 -07001673EXPORT_SYMBOL(filemap_populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
1675struct vm_operations_struct generic_file_vm_ops = {
1676 .nopage = filemap_nopage,
1677 .populate = filemap_populate,
1678};
1679
1680/* This is used for a general mmap of a disk file */
1681
1682int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1683{
1684 struct address_space *mapping = file->f_mapping;
1685
1686 if (!mapping->a_ops->readpage)
1687 return -ENOEXEC;
1688 file_accessed(file);
1689 vma->vm_ops = &generic_file_vm_ops;
1690 return 0;
1691}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693/*
1694 * This is for filesystems which do not implement ->writepage.
1695 */
1696int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1697{
1698 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1699 return -EINVAL;
1700 return generic_file_mmap(file, vma);
1701}
1702#else
1703int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1704{
1705 return -ENOSYS;
1706}
1707int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1708{
1709 return -ENOSYS;
1710}
1711#endif /* CONFIG_MMU */
1712
1713EXPORT_SYMBOL(generic_file_mmap);
1714EXPORT_SYMBOL(generic_file_readonly_mmap);
1715
Nick Piggin6fe69002007-05-06 14:49:04 -07001716static struct page *__read_cache_page(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 unsigned long index,
1718 int (*filler)(void *,struct page*),
1719 void *data)
1720{
1721 struct page *page, *cached_page = NULL;
1722 int err;
1723repeat:
1724 page = find_get_page(mapping, index);
1725 if (!page) {
1726 if (!cached_page) {
1727 cached_page = page_cache_alloc_cold(mapping);
1728 if (!cached_page)
1729 return ERR_PTR(-ENOMEM);
1730 }
1731 err = add_to_page_cache_lru(cached_page, mapping,
1732 index, GFP_KERNEL);
1733 if (err == -EEXIST)
1734 goto repeat;
1735 if (err < 0) {
1736 /* Presumably ENOMEM for radix tree node */
1737 page_cache_release(cached_page);
1738 return ERR_PTR(err);
1739 }
1740 page = cached_page;
1741 cached_page = NULL;
1742 err = filler(data, page);
1743 if (err < 0) {
1744 page_cache_release(page);
1745 page = ERR_PTR(err);
1746 }
1747 }
1748 if (cached_page)
1749 page_cache_release(cached_page);
1750 return page;
1751}
1752
Nick Piggin6fe69002007-05-06 14:49:04 -07001753/*
1754 * Same as read_cache_page, but don't wait for page to become unlocked
1755 * after submitting it to the filler.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001757struct page *read_cache_page_async(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 unsigned long index,
1759 int (*filler)(void *,struct page*),
1760 void *data)
1761{
1762 struct page *page;
1763 int err;
1764
1765retry:
1766 page = __read_cache_page(mapping, index, filler, data);
1767 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001768 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 if (PageUptodate(page))
1770 goto out;
1771
1772 lock_page(page);
1773 if (!page->mapping) {
1774 unlock_page(page);
1775 page_cache_release(page);
1776 goto retry;
1777 }
1778 if (PageUptodate(page)) {
1779 unlock_page(page);
1780 goto out;
1781 }
1782 err = filler(data, page);
1783 if (err < 0) {
1784 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001785 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 }
David Howellsc855ff32007-05-09 13:42:20 +01001787out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001788 mark_page_accessed(page);
1789 return page;
1790}
1791EXPORT_SYMBOL(read_cache_page_async);
1792
1793/**
1794 * read_cache_page - read into page cache, fill it if needed
1795 * @mapping: the page's address_space
1796 * @index: the page index
1797 * @filler: function to perform the read
1798 * @data: destination for read data
1799 *
1800 * Read into the page cache. If a page already exists, and PageUptodate() is
1801 * not set, try to fill the page then wait for it to become unlocked.
1802 *
1803 * If the page does not get brought uptodate, return -EIO.
1804 */
1805struct page *read_cache_page(struct address_space *mapping,
1806 unsigned long index,
1807 int (*filler)(void *,struct page*),
1808 void *data)
1809{
1810 struct page *page;
1811
1812 page = read_cache_page_async(mapping, index, filler, data);
1813 if (IS_ERR(page))
1814 goto out;
1815 wait_on_page_locked(page);
1816 if (!PageUptodate(page)) {
1817 page_cache_release(page);
1818 page = ERR_PTR(-EIO);
1819 }
1820 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 return page;
1822}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823EXPORT_SYMBOL(read_cache_page);
1824
1825/*
1826 * If the page was newly created, increment its refcount and add it to the
1827 * caller's lru-buffering pagevec. This function is specifically for
1828 * generic_file_write().
1829 */
1830static inline struct page *
1831__grab_cache_page(struct address_space *mapping, unsigned long index,
1832 struct page **cached_page, struct pagevec *lru_pvec)
1833{
1834 int err;
1835 struct page *page;
1836repeat:
1837 page = find_lock_page(mapping, index);
1838 if (!page) {
1839 if (!*cached_page) {
1840 *cached_page = page_cache_alloc(mapping);
1841 if (!*cached_page)
1842 return NULL;
1843 }
1844 err = add_to_page_cache(*cached_page, mapping,
1845 index, GFP_KERNEL);
1846 if (err == -EEXIST)
1847 goto repeat;
1848 if (err == 0) {
1849 page = *cached_page;
1850 page_cache_get(page);
1851 if (!pagevec_add(lru_pvec, page))
1852 __pagevec_lru_add(lru_pvec);
1853 *cached_page = NULL;
1854 }
1855 }
1856 return page;
1857}
1858
1859/*
1860 * The logic we want is
1861 *
1862 * if suid or (sgid and xgrp)
1863 * remove privs
1864 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001865int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866{
1867 mode_t mode = dentry->d_inode->i_mode;
1868 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
1870 /* suid always must be killed */
1871 if (unlikely(mode & S_ISUID))
1872 kill = ATTR_KILL_SUID;
1873
1874 /*
1875 * sgid without any exec bits is just a mandatory locking mark; leave
1876 * it alone. If some exec bits are set, it's a real sgid; kill it.
1877 */
1878 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1879 kill |= ATTR_KILL_SGID;
1880
Jens Axboe01de85e2006-10-17 19:50:36 +02001881 if (unlikely(kill && !capable(CAP_FSETID)))
1882 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
Jens Axboe01de85e2006-10-17 19:50:36 +02001884 return 0;
1885}
Mark Fashehd23a1472006-10-17 17:05:18 -07001886EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001887
1888int __remove_suid(struct dentry *dentry, int kill)
1889{
1890 struct iattr newattrs;
1891
1892 newattrs.ia_valid = ATTR_FORCE | kill;
1893 return notify_change(dentry, &newattrs);
1894}
1895
1896int remove_suid(struct dentry *dentry)
1897{
1898 int kill = should_remove_suid(dentry);
1899
1900 if (unlikely(kill))
1901 return __remove_suid(dentry, kill);
1902
1903 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904}
1905EXPORT_SYMBOL(remove_suid);
1906
Carsten Otteceffc072005-06-23 22:05:25 -07001907size_t
NeilBrown01408c42006-06-25 05:47:58 -07001908__filemap_copy_from_user_iovec_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 const struct iovec *iov, size_t base, size_t bytes)
1910{
1911 size_t copied = 0, left = 0;
1912
1913 while (bytes) {
1914 char __user *buf = iov->iov_base + base;
1915 int copy = min(bytes, iov->iov_len - base);
1916
1917 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001918 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 copied += copy;
1920 bytes -= copy;
1921 vaddr += copy;
1922 iov++;
1923
NeilBrown01408c42006-06-25 05:47:58 -07001924 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 }
1927 return copied - left;
1928}
1929
1930/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 * Performs necessary checks before doing a write
1932 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001933 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 * Returns appropriate error code that caller should return or
1935 * zero in case that write should be allowed.
1936 */
1937inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1938{
1939 struct inode *inode = file->f_mapping->host;
1940 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1941
1942 if (unlikely(*pos < 0))
1943 return -EINVAL;
1944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 if (!isblk) {
1946 /* FIXME: this is for backwards compatibility with 2.4 */
1947 if (file->f_flags & O_APPEND)
1948 *pos = i_size_read(inode);
1949
1950 if (limit != RLIM_INFINITY) {
1951 if (*pos >= limit) {
1952 send_sig(SIGXFSZ, current, 0);
1953 return -EFBIG;
1954 }
1955 if (*count > limit - (typeof(limit))*pos) {
1956 *count = limit - (typeof(limit))*pos;
1957 }
1958 }
1959 }
1960
1961 /*
1962 * LFS rule
1963 */
1964 if (unlikely(*pos + *count > MAX_NON_LFS &&
1965 !(file->f_flags & O_LARGEFILE))) {
1966 if (*pos >= MAX_NON_LFS) {
1967 send_sig(SIGXFSZ, current, 0);
1968 return -EFBIG;
1969 }
1970 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1971 *count = MAX_NON_LFS - (unsigned long)*pos;
1972 }
1973 }
1974
1975 /*
1976 * Are we about to exceed the fs block limit ?
1977 *
1978 * If we have written data it becomes a short write. If we have
1979 * exceeded without writing data we send a signal and return EFBIG.
1980 * Linus frestrict idea will clean these up nicely..
1981 */
1982 if (likely(!isblk)) {
1983 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1984 if (*count || *pos > inode->i_sb->s_maxbytes) {
1985 send_sig(SIGXFSZ, current, 0);
1986 return -EFBIG;
1987 }
1988 /* zero-length writes at ->s_maxbytes are OK */
1989 }
1990
1991 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1992 *count = inode->i_sb->s_maxbytes - *pos;
1993 } else {
David Howells93614012006-09-30 20:45:40 +02001994#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 loff_t isize;
1996 if (bdev_read_only(I_BDEV(inode)))
1997 return -EPERM;
1998 isize = i_size_read(inode);
1999 if (*pos >= isize) {
2000 if (*count || *pos > isize)
2001 return -ENOSPC;
2002 }
2003
2004 if (*pos + *count > isize)
2005 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002006#else
2007 return -EPERM;
2008#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 }
2010 return 0;
2011}
2012EXPORT_SYMBOL(generic_write_checks);
2013
2014ssize_t
2015generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2016 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2017 size_t count, size_t ocount)
2018{
2019 struct file *file = iocb->ki_filp;
2020 struct address_space *mapping = file->f_mapping;
2021 struct inode *inode = mapping->host;
2022 ssize_t written;
2023
2024 if (count != ocount)
2025 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2026
2027 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2028 if (written > 0) {
2029 loff_t end = pos + written;
2030 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2031 i_size_write(inode, end);
2032 mark_inode_dirty(inode);
2033 }
2034 *ppos = end;
2035 }
2036
2037 /*
2038 * Sync the fs metadata but not the minor inode changes and
2039 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002040 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002041 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 */
Zach Brown8459d862006-12-10 02:21:05 -08002043 if ((written >= 0 || written == -EIOCBQUEUED) &&
2044 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002045 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2046 if (err < 0)
2047 written = err;
2048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 return written;
2050}
2051EXPORT_SYMBOL(generic_file_direct_write);
2052
2053ssize_t
2054generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2055 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2056 size_t count, ssize_t written)
2057{
2058 struct file *file = iocb->ki_filp;
2059 struct address_space * mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002060 const struct address_space_operations *a_ops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 struct inode *inode = mapping->host;
2062 long status = 0;
2063 struct page *page;
2064 struct page *cached_page = NULL;
2065 size_t bytes;
2066 struct pagevec lru_pvec;
2067 const struct iovec *cur_iov = iov; /* current iovec */
2068 size_t iov_base = 0; /* offset in the current iovec */
2069 char __user *buf;
2070
2071 pagevec_init(&lru_pvec, 0);
2072
2073 /*
2074 * handle partial DIO write. Adjust cur_iov if needed.
2075 */
2076 if (likely(nr_segs == 1))
2077 buf = iov->iov_base + written;
2078 else {
2079 filemap_set_next_iovec(&cur_iov, &iov_base, written);
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002080 buf = cur_iov->iov_base + iov_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 }
2082
2083 do {
2084 unsigned long index;
2085 unsigned long offset;
2086 size_t copied;
2087
2088 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2089 index = pos >> PAGE_CACHE_SHIFT;
2090 bytes = PAGE_CACHE_SIZE - offset;
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07002091
2092 /* Limit the size of the copy to the caller's write size */
2093 bytes = min(bytes, count);
2094
NeilBrown29dbb3f2007-02-16 01:28:38 -08002095 /* We only need to worry about prefaulting when writes are from
2096 * user-space. NFSd uses vfs_writev with several non-aligned
2097 * segments in the vector, and limiting to one segment a time is
2098 * a noticeable performance for re-write
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07002099 */
NeilBrown29dbb3f2007-02-16 01:28:38 -08002100 if (!segment_eq(get_fs(), KERNEL_DS)) {
2101 /*
2102 * Limit the size of the copy to that of the current
2103 * segment, because fault_in_pages_readable() doesn't
2104 * know how to walk segments.
2105 */
2106 bytes = min(bytes, cur_iov->iov_len - iov_base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
NeilBrown29dbb3f2007-02-16 01:28:38 -08002108 /*
2109 * Bring in the user page that we will copy from
2110 * _first_. Otherwise there's a nasty deadlock on
2111 * copying from the same page as we're writing to,
2112 * without it being marked up-to-date.
2113 */
2114 fault_in_pages_readable(buf, bytes);
2115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2117 if (!page) {
2118 status = -ENOMEM;
2119 break;
2120 }
2121
Andrew Morton81b0c872006-06-29 02:24:26 -07002122 if (unlikely(bytes == 0)) {
2123 status = 0;
2124 copied = 0;
2125 goto zero_length_segment;
2126 }
2127
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2129 if (unlikely(status)) {
2130 loff_t isize = i_size_read(inode);
Zach Brown994fc28c2005-12-15 14:28:17 -08002131
2132 if (status != AOP_TRUNCATED_PAGE)
2133 unlock_page(page);
2134 page_cache_release(page);
2135 if (status == AOP_TRUNCATED_PAGE)
2136 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 /*
2138 * prepare_write() may have instantiated a few blocks
2139 * outside i_size. Trim these off again.
2140 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 if (pos + bytes > isize)
2142 vmtruncate(inode, isize);
2143 break;
2144 }
2145 if (likely(nr_segs == 1))
2146 copied = filemap_copy_from_user(page, offset,
2147 buf, bytes);
2148 else
2149 copied = filemap_copy_from_user_iovec(page, offset,
2150 cur_iov, iov_base, bytes);
2151 flush_dcache_page(page);
2152 status = a_ops->commit_write(file, page, offset, offset+bytes);
Zach Brown994fc28c2005-12-15 14:28:17 -08002153 if (status == AOP_TRUNCATED_PAGE) {
2154 page_cache_release(page);
2155 continue;
2156 }
Andrew Morton81b0c872006-06-29 02:24:26 -07002157zero_length_segment:
2158 if (likely(copied >= 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 if (!status)
2160 status = copied;
2161
2162 if (status >= 0) {
2163 written += status;
2164 count -= status;
2165 pos += status;
2166 buf += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002167 if (unlikely(nr_segs > 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 filemap_set_next_iovec(&cur_iov,
2169 &iov_base, status);
Badari Pulavartyb0cfbd92005-06-25 14:55:42 -07002170 if (count)
2171 buf = cur_iov->iov_base +
2172 iov_base;
Martin Schwidefskya5117182005-06-06 13:35:54 -07002173 } else {
2174 iov_base += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002175 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 }
2177 }
2178 if (unlikely(copied != bytes))
2179 if (status >= 0)
2180 status = -EFAULT;
2181 unlock_page(page);
2182 mark_page_accessed(page);
2183 page_cache_release(page);
2184 if (status < 0)
2185 break;
2186 balance_dirty_pages_ratelimited(mapping);
2187 cond_resched();
2188 } while (count);
2189 *ppos = pos;
2190
2191 if (cached_page)
2192 page_cache_release(cached_page);
2193
2194 /*
2195 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2196 */
2197 if (likely(status >= 0)) {
2198 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2199 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2200 status = generic_osync_inode(inode, mapping,
2201 OSYNC_METADATA|OSYNC_DATA);
2202 }
2203 }
2204
2205 /*
2206 * If we get here for O_DIRECT writes then we must have fallen through
2207 * to buffered writes (block instantiation inside i_size). So we sync
2208 * the file data here, to try to honour O_DIRECT expectations.
2209 */
2210 if (unlikely(file->f_flags & O_DIRECT) && written)
2211 status = filemap_write_and_wait(mapping);
2212
2213 pagevec_lru_add(&lru_pvec);
2214 return written ? written : status;
2215}
2216EXPORT_SYMBOL(generic_file_buffered_write);
2217
Adrian Bunk5ce78522005-09-10 00:26:28 -07002218static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2220 unsigned long nr_segs, loff_t *ppos)
2221{
2222 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002223 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 size_t ocount; /* original count */
2225 size_t count; /* after file limit checks */
2226 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 loff_t pos;
2228 ssize_t written;
2229 ssize_t err;
2230
2231 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002232 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2233 if (err)
2234 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235
2236 count = ocount;
2237 pos = *ppos;
2238
2239 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2240
2241 /* We can write back this queue in page reclaim */
2242 current->backing_dev_info = mapping->backing_dev_info;
2243 written = 0;
2244
2245 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2246 if (err)
2247 goto out;
2248
2249 if (count == 0)
2250 goto out;
2251
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002252 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (err)
2254 goto out;
2255
Christoph Hellwig870f4812006-01-09 20:52:01 -08002256 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257
2258 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2259 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002260 loff_t endbyte;
2261 ssize_t written_buffered;
2262
2263 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2264 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 if (written < 0 || written == count)
2266 goto out;
2267 /*
2268 * direct-io write to a hole: fall through to buffered I/O
2269 * for completing the rest of the request.
2270 */
2271 pos += written;
2272 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002273 written_buffered = generic_file_buffered_write(iocb, iov,
2274 nr_segs, pos, ppos, count,
2275 written);
2276 /*
2277 * If generic_file_buffered_write() retuned a synchronous error
2278 * then we want to return the number of bytes which were
2279 * direct-written, or the error code if that was zero. Note
2280 * that this differs from normal direct-io semantics, which
2281 * will return -EFOO even if some bytes were written.
2282 */
2283 if (written_buffered < 0) {
2284 err = written_buffered;
2285 goto out;
2286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002288 /*
2289 * We need to ensure that the page cache pages are written to
2290 * disk and invalidated to preserve the expected O_DIRECT
2291 * semantics.
2292 */
2293 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002294 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2295 SYNC_FILE_RANGE_WAIT_BEFORE|
2296 SYNC_FILE_RANGE_WRITE|
2297 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002298 if (err == 0) {
2299 written = written_buffered;
2300 invalidate_mapping_pages(mapping,
2301 pos >> PAGE_CACHE_SHIFT,
2302 endbyte >> PAGE_CACHE_SHIFT);
2303 } else {
2304 /*
2305 * We don't know how much we wrote, so just return
2306 * the number of bytes which were direct-written
2307 */
2308 }
2309 } else {
2310 written = generic_file_buffered_write(iocb, iov, nr_segs,
2311 pos, ppos, count, written);
2312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313out:
2314 current->backing_dev_info = NULL;
2315 return written ? written : err;
2316}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Badari Pulavarty027445c2006-09-30 23:28:46 -07002318ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2319 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320{
2321 struct file *file = iocb->ki_filp;
2322 struct address_space *mapping = file->f_mapping;
2323 struct inode *inode = mapping->host;
2324 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
Badari Pulavarty027445c2006-09-30 23:28:46 -07002326 BUG_ON(iocb->ki_pos != pos);
2327
2328 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2329 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
2331 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002332 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
2334 err = sync_page_range_nolock(inode, mapping, pos, ret);
2335 if (err < 0)
2336 ret = err;
2337 }
2338 return ret;
2339}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002340EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Badari Pulavarty027445c2006-09-30 23:28:46 -07002342ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2343 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344{
2345 struct file *file = iocb->ki_filp;
2346 struct address_space *mapping = file->f_mapping;
2347 struct inode *inode = mapping->host;
2348 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349
2350 BUG_ON(iocb->ki_pos != pos);
2351
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002352 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002353 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2354 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002355 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
2357 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2358 ssize_t err;
2359
2360 err = sync_page_range(inode, mapping, pos, ret);
2361 if (err < 0)
2362 ret = err;
2363 }
2364 return ret;
2365}
2366EXPORT_SYMBOL(generic_file_aio_write);
2367
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002369 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 * went wrong during pagecache shootdown.
2371 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002372static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2374 loff_t offset, unsigned long nr_segs)
2375{
2376 struct file *file = iocb->ki_filp;
2377 struct address_space *mapping = file->f_mapping;
2378 ssize_t retval;
Zach Brown65b82912007-03-16 13:38:11 -08002379 size_t write_len;
2380 pgoff_t end = 0; /* silence gcc */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
2382 /*
2383 * If it's a write, unmap all mmappings of the file up-front. This
2384 * will cause any pte dirty bits to be propagated into the pageframes
2385 * for the subsequent filemap_write_and_wait().
2386 */
2387 if (rw == WRITE) {
2388 write_len = iov_length(iov, nr_segs);
Zach Brown65b82912007-03-16 13:38:11 -08002389 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 if (mapping_mapped(mapping))
2391 unmap_mapping_range(mapping, offset, write_len, 0);
2392 }
2393
2394 retval = filemap_write_and_wait(mapping);
Zach Brown65b82912007-03-16 13:38:11 -08002395 if (retval)
2396 goto out;
2397
2398 /*
2399 * After a write we want buffered reads to be sure to go to disk to get
2400 * the new data. We invalidate clean cached page from the region we're
2401 * about to write. We do this *before* the write so that we can return
2402 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2403 */
2404 if (rw == WRITE && mapping->nrpages) {
2405 retval = invalidate_inode_pages2_range(mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 offset >> PAGE_CACHE_SHIFT, end);
Zach Brown65b82912007-03-16 13:38:11 -08002407 if (retval)
2408 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 }
Zach Brown65b82912007-03-16 13:38:11 -08002410
2411 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2412 if (retval)
2413 goto out;
2414
2415 /*
2416 * Finally, try again to invalidate clean pages which might have been
2417 * faulted in by get_user_pages() if the source of the write was an
2418 * mmap()ed region of the file we're writing. That's a pretty crazy
2419 * thing to do, so we don't support it 100%. If this invalidation
2420 * fails and we have -EIOCBQUEUED we ignore the failure.
2421 */
2422 if (rw == WRITE && mapping->nrpages) {
2423 int err = invalidate_inode_pages2_range(mapping,
2424 offset >> PAGE_CACHE_SHIFT, end);
2425 if (err && retval >= 0)
2426 retval = err;
2427 }
2428out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 return retval;
2430}
David Howellscf9a2ae2006-08-29 19:05:54 +01002431
2432/**
2433 * try_to_release_page() - release old fs-specific metadata on a page
2434 *
2435 * @page: the page which the kernel is trying to free
2436 * @gfp_mask: memory allocation flags (and I/O mode)
2437 *
2438 * The address_space is to try to release any data against the page
2439 * (presumably at page->private). If the release was successful, return `1'.
2440 * Otherwise return zero.
2441 *
2442 * The @gfp_mask argument specifies whether I/O may be performed to release
2443 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2444 *
2445 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2446 */
2447int try_to_release_page(struct page *page, gfp_t gfp_mask)
2448{
2449 struct address_space * const mapping = page->mapping;
2450
2451 BUG_ON(!PageLocked(page));
2452 if (PageWriteback(page))
2453 return 0;
2454
2455 if (mapping && mapping->a_ops->releasepage)
2456 return mapping->a_ops->releasepage(page, gfp_mask);
2457 return try_to_free_buffers(page);
2458}
2459
2460EXPORT_SYMBOL(try_to_release_page);