blob: b9a60c43b61a4649f19a418584af345088e74daf [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 *
78 * ->mmap_sem
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080079 * ->i_mutex (msync)
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 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -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
470struct page *page_cache_alloc(struct address_space *x)
471{
472 if (cpuset_do_page_mem_spread()) {
473 int n = cpuset_mem_spread_node();
474 return alloc_pages_node(n, mapping_gfp_mask(x), 0);
475 }
476 return alloc_pages(mapping_gfp_mask(x), 0);
477}
478EXPORT_SYMBOL(page_cache_alloc);
479
480struct page *page_cache_alloc_cold(struct address_space *x)
481{
482 if (cpuset_do_page_mem_spread()) {
483 int n = cpuset_mem_spread_node();
484 return alloc_pages_node(n, mapping_gfp_mask(x)|__GFP_COLD, 0);
485 }
486 return alloc_pages(mapping_gfp_mask(x)|__GFP_COLD, 0);
487}
488EXPORT_SYMBOL(page_cache_alloc_cold);
489#endif
490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491/*
492 * In order to wait for pages to become available there must be
493 * waitqueues associated with pages. By using a hash table of
494 * waitqueues where the bucket discipline is to maintain all
495 * waiters on the same queue and wake all when any of the pages
496 * become available, and for the woken contexts to check to be
497 * sure the appropriate page became available, this saves space
498 * at a cost of "thundering herd" phenomena during rare hash
499 * collisions.
500 */
501static wait_queue_head_t *page_waitqueue(struct page *page)
502{
503 const struct zone *zone = page_zone(page);
504
505 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
506}
507
508static inline void wake_up_page(struct page *page, int bit)
509{
510 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
511}
512
513void fastcall wait_on_page_bit(struct page *page, int bit_nr)
514{
515 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
516
517 if (test_bit(bit_nr, &page->flags))
518 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
519 TASK_UNINTERRUPTIBLE);
520}
521EXPORT_SYMBOL(wait_on_page_bit);
522
523/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700524 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 * @page: the page
526 *
527 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
528 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
529 * mechananism between PageLocked pages and PageWriteback pages is shared.
530 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
531 *
532 * The first mb is necessary to safely close the critical section opened by the
533 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
534 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
535 * parallel wait_on_page_locked()).
536 */
537void fastcall unlock_page(struct page *page)
538{
539 smp_mb__before_clear_bit();
540 if (!TestClearPageLocked(page))
541 BUG();
542 smp_mb__after_clear_bit();
543 wake_up_page(page, PG_locked);
544}
545EXPORT_SYMBOL(unlock_page);
546
Randy Dunlap485bb992006-06-23 02:03:49 -0700547/**
548 * end_page_writeback - end writeback against a page
549 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 */
551void end_page_writeback(struct page *page)
552{
553 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
554 if (!test_clear_page_writeback(page))
555 BUG();
556 }
557 smp_mb__after_clear_bit();
558 wake_up_page(page, PG_writeback);
559}
560EXPORT_SYMBOL(end_page_writeback);
561
Randy Dunlap485bb992006-06-23 02:03:49 -0700562/**
563 * __lock_page - get a lock on the page, assuming we need to sleep to get it
564 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700566 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
568 * chances are that on the second loop, the block layer's plug list is empty,
569 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
570 */
571void fastcall __lock_page(struct page *page)
572{
573 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
574
575 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
576 TASK_UNINTERRUPTIBLE);
577}
578EXPORT_SYMBOL(__lock_page);
579
Randy Dunlap485bb992006-06-23 02:03:49 -0700580/**
581 * find_get_page - find and get a page reference
582 * @mapping: the address_space to search
583 * @offset: the page index
584 *
585 * A rather lightweight function, finding and getting a reference to a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 * hashed page atomically.
587 */
588struct page * find_get_page(struct address_space *mapping, unsigned long offset)
589{
590 struct page *page;
591
592 read_lock_irq(&mapping->tree_lock);
593 page = radix_tree_lookup(&mapping->page_tree, offset);
594 if (page)
595 page_cache_get(page);
596 read_unlock_irq(&mapping->tree_lock);
597 return page;
598}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599EXPORT_SYMBOL(find_get_page);
600
Randy Dunlap485bb992006-06-23 02:03:49 -0700601/**
602 * find_trylock_page - find and lock a page
603 * @mapping: the address_space to search
604 * @offset: the page index
605 *
606 * Same as find_get_page(), but trylock it instead of incrementing the count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
608struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
609{
610 struct page *page;
611
612 read_lock_irq(&mapping->tree_lock);
613 page = radix_tree_lookup(&mapping->page_tree, offset);
614 if (page && TestSetPageLocked(page))
615 page = NULL;
616 read_unlock_irq(&mapping->tree_lock);
617 return page;
618}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619EXPORT_SYMBOL(find_trylock_page);
620
621/**
622 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700623 * @mapping: the address_space to search
624 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 *
626 * Locates the desired pagecache page, locks it, increments its reference
627 * count and returns its address.
628 *
629 * Returns zero if the page was not present. find_lock_page() may sleep.
630 */
631struct page *find_lock_page(struct address_space *mapping,
632 unsigned long offset)
633{
634 struct page *page;
635
636 read_lock_irq(&mapping->tree_lock);
637repeat:
638 page = radix_tree_lookup(&mapping->page_tree, offset);
639 if (page) {
640 page_cache_get(page);
641 if (TestSetPageLocked(page)) {
642 read_unlock_irq(&mapping->tree_lock);
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800643 __lock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 read_lock_irq(&mapping->tree_lock);
645
646 /* Has the page been truncated while we slept? */
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800647 if (unlikely(page->mapping != mapping ||
648 page->index != offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 unlock_page(page);
650 page_cache_release(page);
651 goto repeat;
652 }
653 }
654 }
655 read_unlock_irq(&mapping->tree_lock);
656 return page;
657}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658EXPORT_SYMBOL(find_lock_page);
659
660/**
661 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700662 * @mapping: the page's address_space
663 * @index: the page's index into the mapping
664 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 *
666 * Locates a page in the pagecache. If the page is not present, a new page
667 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
668 * LRU list. The returned page is locked and has its reference count
669 * incremented.
670 *
671 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
672 * allocation!
673 *
674 * find_or_create_page() returns the desired page's address, or zero on
675 * memory exhaustion.
676 */
677struct page *find_or_create_page(struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400678 unsigned long index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
680 struct page *page, *cached_page = NULL;
681 int err;
682repeat:
683 page = find_lock_page(mapping, index);
684 if (!page) {
685 if (!cached_page) {
686 cached_page = alloc_page(gfp_mask);
687 if (!cached_page)
688 return NULL;
689 }
690 err = add_to_page_cache_lru(cached_page, mapping,
691 index, gfp_mask);
692 if (!err) {
693 page = cached_page;
694 cached_page = NULL;
695 } else if (err == -EEXIST)
696 goto repeat;
697 }
698 if (cached_page)
699 page_cache_release(cached_page);
700 return page;
701}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702EXPORT_SYMBOL(find_or_create_page);
703
704/**
705 * find_get_pages - gang pagecache lookup
706 * @mapping: The address_space to search
707 * @start: The starting page index
708 * @nr_pages: The maximum number of pages
709 * @pages: Where the resulting pages are placed
710 *
711 * find_get_pages() will search for and return a group of up to
712 * @nr_pages pages in the mapping. The pages are placed at @pages.
713 * find_get_pages() takes a reference against the returned pages.
714 *
715 * The search returns a group of mapping-contiguous pages with ascending
716 * indexes. There may be holes in the indices due to not-present pages.
717 *
718 * find_get_pages() returns the number of pages which were found.
719 */
720unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
721 unsigned int nr_pages, struct page **pages)
722{
723 unsigned int i;
724 unsigned int ret;
725
726 read_lock_irq(&mapping->tree_lock);
727 ret = radix_tree_gang_lookup(&mapping->page_tree,
728 (void **)pages, start, nr_pages);
729 for (i = 0; i < ret; i++)
730 page_cache_get(pages[i]);
731 read_unlock_irq(&mapping->tree_lock);
732 return ret;
733}
734
Jens Axboeebf43502006-04-27 08:46:01 +0200735/**
736 * find_get_pages_contig - gang contiguous pagecache lookup
737 * @mapping: The address_space to search
738 * @index: The starting page index
739 * @nr_pages: The maximum number of pages
740 * @pages: Where the resulting pages are placed
741 *
742 * find_get_pages_contig() works exactly like find_get_pages(), except
743 * that the returned number of pages are guaranteed to be contiguous.
744 *
745 * find_get_pages_contig() returns the number of pages which were found.
746 */
747unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
748 unsigned int nr_pages, struct page **pages)
749{
750 unsigned int i;
751 unsigned int ret;
752
753 read_lock_irq(&mapping->tree_lock);
754 ret = radix_tree_gang_lookup(&mapping->page_tree,
755 (void **)pages, index, nr_pages);
756 for (i = 0; i < ret; i++) {
757 if (pages[i]->mapping == NULL || pages[i]->index != index)
758 break;
759
760 page_cache_get(pages[i]);
761 index++;
762 }
763 read_unlock_irq(&mapping->tree_lock);
764 return i;
765}
766
Randy Dunlap485bb992006-06-23 02:03:49 -0700767/**
768 * find_get_pages_tag - find and return pages that match @tag
769 * @mapping: the address_space to search
770 * @index: the starting page index
771 * @tag: the tag index
772 * @nr_pages: the maximum number of pages
773 * @pages: where the resulting pages are placed
774 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700776 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 */
778unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
779 int tag, unsigned int nr_pages, struct page **pages)
780{
781 unsigned int i;
782 unsigned int ret;
783
784 read_lock_irq(&mapping->tree_lock);
785 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
786 (void **)pages, *index, nr_pages, tag);
787 for (i = 0; i < ret; i++)
788 page_cache_get(pages[i]);
789 if (ret)
790 *index = pages[ret - 1]->index + 1;
791 read_unlock_irq(&mapping->tree_lock);
792 return ret;
793}
794
Randy Dunlap485bb992006-06-23 02:03:49 -0700795/**
796 * grab_cache_page_nowait - returns locked page at given index in given cache
797 * @mapping: target address_space
798 * @index: the page index
799 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * Same as grab_cache_page, but do not wait if the page is unavailable.
801 * This is intended for speculative data generators, where the data can
802 * be regenerated if the page couldn't be grabbed. This routine should
803 * be safe to call while holding the lock for another page.
804 *
805 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
806 * and deadlock against the caller's locked page.
807 */
808struct page *
809grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
810{
811 struct page *page = find_get_page(mapping, index);
Al Viro6daa0e22005-10-21 03:18:50 -0400812 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
814 if (page) {
815 if (!TestSetPageLocked(page))
816 return page;
817 page_cache_release(page);
818 return NULL;
819 }
820 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
821 page = alloc_pages(gfp_mask, 0);
822 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
823 page_cache_release(page);
824 page = NULL;
825 }
826 return page;
827}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828EXPORT_SYMBOL(grab_cache_page_nowait);
829
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700830/*
831 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
832 * a _large_ part of the i/o request. Imagine the worst scenario:
833 *
834 * ---R__________________________________________B__________
835 * ^ reading here ^ bad block(assume 4k)
836 *
837 * read(R) => miss => readahead(R...B) => media error => frustrating retries
838 * => failing the whole request => read(R) => read(R+1) =>
839 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
840 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
841 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
842 *
843 * It is going insane. Fix it by quickly scaling down the readahead size.
844 */
845static void shrink_readahead_size_eio(struct file *filp,
846 struct file_ra_state *ra)
847{
848 if (!ra->ra_pages)
849 return;
850
851 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700852}
853
Randy Dunlap485bb992006-06-23 02:03:49 -0700854/**
855 * do_generic_mapping_read - generic file read routine
856 * @mapping: address_space to be read
857 * @_ra: file's readahead state
858 * @filp: the file to read
859 * @ppos: current file position
860 * @desc: read_descriptor
861 * @actor: read method
862 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700864 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 *
866 * This is really ugly. But the goto's actually try to clarify some
867 * of the logic when it comes to error handling etc.
868 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700869 * Note the struct file* is only passed for the use of readpage.
870 * It may be NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 */
872void do_generic_mapping_read(struct address_space *mapping,
873 struct file_ra_state *_ra,
874 struct file *filp,
875 loff_t *ppos,
876 read_descriptor_t *desc,
877 read_actor_t actor)
878{
879 struct inode *inode = mapping->host;
880 unsigned long index;
881 unsigned long end_index;
882 unsigned long offset;
883 unsigned long last_index;
884 unsigned long next_index;
885 unsigned long prev_index;
886 loff_t isize;
887 struct page *cached_page;
888 int error;
889 struct file_ra_state ra = *_ra;
890
891 cached_page = NULL;
892 index = *ppos >> PAGE_CACHE_SHIFT;
893 next_index = index;
894 prev_index = ra.prev_page;
895 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
896 offset = *ppos & ~PAGE_CACHE_MASK;
897
898 isize = i_size_read(inode);
899 if (!isize)
900 goto out;
901
902 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
903 for (;;) {
904 struct page *page;
905 unsigned long nr, ret;
906
907 /* nr is the maximum number of bytes to copy from this page */
908 nr = PAGE_CACHE_SIZE;
909 if (index >= end_index) {
910 if (index > end_index)
911 goto out;
912 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
913 if (nr <= offset) {
914 goto out;
915 }
916 }
917 nr = nr - offset;
918
919 cond_resched();
920 if (index == next_index)
921 next_index = page_cache_readahead(mapping, &ra, filp,
922 index, last_index - index);
923
924find_page:
925 page = find_get_page(mapping, index);
926 if (unlikely(page == NULL)) {
927 handle_ra_miss(mapping, &ra, index);
928 goto no_cached_page;
929 }
930 if (!PageUptodate(page))
931 goto page_not_up_to_date;
932page_ok:
933
934 /* If users can be writing to this page using arbitrary
935 * virtual addresses, take care about potential aliasing
936 * before reading the page on the kernel side.
937 */
938 if (mapping_writably_mapped(mapping))
939 flush_dcache_page(page);
940
941 /*
942 * When (part of) the same page is read multiple times
943 * in succession, only mark it as accessed the first time.
944 */
945 if (prev_index != index)
946 mark_page_accessed(page);
947 prev_index = index;
948
949 /*
950 * Ok, we have the page, and it's up-to-date, so
951 * now we can copy it to user space...
952 *
953 * The actor routine returns how many bytes were actually used..
954 * NOTE! This may not be the same as how much of a user buffer
955 * we filled up (we may be padding etc), so we can only update
956 * "pos" here (the actor routine has to update the user buffer
957 * pointers and the remaining count).
958 */
959 ret = actor(desc, page, offset, nr);
960 offset += ret;
961 index += offset >> PAGE_CACHE_SHIFT;
962 offset &= ~PAGE_CACHE_MASK;
963
964 page_cache_release(page);
965 if (ret == nr && desc->count)
966 continue;
967 goto out;
968
969page_not_up_to_date:
970 /* Get exclusive access to the page ... */
971 lock_page(page);
972
973 /* Did it get unhashed before we got the lock? */
974 if (!page->mapping) {
975 unlock_page(page);
976 page_cache_release(page);
977 continue;
978 }
979
980 /* Did somebody else fill it already? */
981 if (PageUptodate(page)) {
982 unlock_page(page);
983 goto page_ok;
984 }
985
986readpage:
987 /* Start the actual read. The read will unlock the page. */
988 error = mapping->a_ops->readpage(filp, page);
989
Zach Brown994fc28c2005-12-15 14:28:17 -0800990 if (unlikely(error)) {
991 if (error == AOP_TRUNCATED_PAGE) {
992 page_cache_release(page);
993 goto find_page;
994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -0800996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 if (!PageUptodate(page)) {
999 lock_page(page);
1000 if (!PageUptodate(page)) {
1001 if (page->mapping == NULL) {
1002 /*
1003 * invalidate_inode_pages got it
1004 */
1005 unlock_page(page);
1006 page_cache_release(page);
1007 goto find_page;
1008 }
1009 unlock_page(page);
1010 error = -EIO;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001011 shrink_readahead_size_eio(filp, &ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 goto readpage_error;
1013 }
1014 unlock_page(page);
1015 }
1016
1017 /*
1018 * i_size must be checked after we have done ->readpage.
1019 *
1020 * Checking i_size after the readpage allows us to calculate
1021 * the correct value for "nr", which means the zero-filled
1022 * part of the page is not copied back to userspace (unless
1023 * another truncate extends the file - this is desired though).
1024 */
1025 isize = i_size_read(inode);
1026 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1027 if (unlikely(!isize || index > end_index)) {
1028 page_cache_release(page);
1029 goto out;
1030 }
1031
1032 /* nr is the maximum number of bytes to copy from this page */
1033 nr = PAGE_CACHE_SIZE;
1034 if (index == end_index) {
1035 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1036 if (nr <= offset) {
1037 page_cache_release(page);
1038 goto out;
1039 }
1040 }
1041 nr = nr - offset;
1042 goto page_ok;
1043
1044readpage_error:
1045 /* UHHUH! A synchronous read error occurred. Report it */
1046 desc->error = error;
1047 page_cache_release(page);
1048 goto out;
1049
1050no_cached_page:
1051 /*
1052 * Ok, it wasn't cached, so we need to create a new
1053 * page..
1054 */
1055 if (!cached_page) {
1056 cached_page = page_cache_alloc_cold(mapping);
1057 if (!cached_page) {
1058 desc->error = -ENOMEM;
1059 goto out;
1060 }
1061 }
1062 error = add_to_page_cache_lru(cached_page, mapping,
1063 index, GFP_KERNEL);
1064 if (error) {
1065 if (error == -EEXIST)
1066 goto find_page;
1067 desc->error = error;
1068 goto out;
1069 }
1070 page = cached_page;
1071 cached_page = NULL;
1072 goto readpage;
1073 }
1074
1075out:
1076 *_ra = ra;
1077
1078 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1079 if (cached_page)
1080 page_cache_release(cached_page);
1081 if (filp)
1082 file_accessed(filp);
1083}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084EXPORT_SYMBOL(do_generic_mapping_read);
1085
1086int file_read_actor(read_descriptor_t *desc, struct page *page,
1087 unsigned long offset, unsigned long size)
1088{
1089 char *kaddr;
1090 unsigned long left, count = desc->count;
1091
1092 if (size > count)
1093 size = count;
1094
1095 /*
1096 * Faults on the destination of a read are common, so do it before
1097 * taking the kmap.
1098 */
1099 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1100 kaddr = kmap_atomic(page, KM_USER0);
1101 left = __copy_to_user_inatomic(desc->arg.buf,
1102 kaddr + offset, size);
1103 kunmap_atomic(kaddr, KM_USER0);
1104 if (left == 0)
1105 goto success;
1106 }
1107
1108 /* Do it the slow way */
1109 kaddr = kmap(page);
1110 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1111 kunmap(page);
1112
1113 if (left) {
1114 size -= left;
1115 desc->error = -EFAULT;
1116 }
1117success:
1118 desc->count = count - size;
1119 desc->written += size;
1120 desc->arg.buf += size;
1121 return size;
1122}
1123
Randy Dunlap485bb992006-06-23 02:03:49 -07001124/**
1125 * __generic_file_aio_read - generic filesystem read routine
1126 * @iocb: kernel I/O control block
1127 * @iov: io vector request
1128 * @nr_segs: number of segments in the iovec
1129 * @ppos: current file position
1130 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 * This is the "read()" routine for all filesystems
1132 * that can use the page cache directly.
1133 */
1134ssize_t
1135__generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1136 unsigned long nr_segs, loff_t *ppos)
1137{
1138 struct file *filp = iocb->ki_filp;
1139 ssize_t retval;
1140 unsigned long seg;
1141 size_t count;
1142
1143 count = 0;
1144 for (seg = 0; seg < nr_segs; seg++) {
1145 const struct iovec *iv = &iov[seg];
1146
1147 /*
1148 * If any segment has a negative length, or the cumulative
1149 * length ever wraps negative then return -EINVAL.
1150 */
1151 count += iv->iov_len;
1152 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1153 return -EINVAL;
1154 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1155 continue;
1156 if (seg == 0)
1157 return -EFAULT;
1158 nr_segs = seg;
1159 count -= iv->iov_len; /* This segment is no good */
1160 break;
1161 }
1162
1163 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1164 if (filp->f_flags & O_DIRECT) {
1165 loff_t pos = *ppos, size;
1166 struct address_space *mapping;
1167 struct inode *inode;
1168
1169 mapping = filp->f_mapping;
1170 inode = mapping->host;
1171 retval = 0;
1172 if (!count)
1173 goto out; /* skip atime */
1174 size = i_size_read(inode);
1175 if (pos < size) {
1176 retval = generic_file_direct_IO(READ, iocb,
1177 iov, pos, nr_segs);
Suparna Bhattacharyab5c44c22005-05-21 16:33:36 -07001178 if (retval > 0 && !is_sync_kiocb(iocb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 retval = -EIOCBQUEUED;
1180 if (retval > 0)
1181 *ppos = pos + retval;
1182 }
1183 file_accessed(filp);
1184 goto out;
1185 }
1186
1187 retval = 0;
1188 if (count) {
1189 for (seg = 0; seg < nr_segs; seg++) {
1190 read_descriptor_t desc;
1191
1192 desc.written = 0;
1193 desc.arg.buf = iov[seg].iov_base;
1194 desc.count = iov[seg].iov_len;
1195 if (desc.count == 0)
1196 continue;
1197 desc.error = 0;
1198 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1199 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001200 if (desc.error) {
1201 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 break;
1203 }
1204 }
1205 }
1206out:
1207 return retval;
1208}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209EXPORT_SYMBOL(__generic_file_aio_read);
1210
1211ssize_t
1212generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1213{
1214 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1215
1216 BUG_ON(iocb->ki_pos != pos);
1217 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1218}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219EXPORT_SYMBOL(generic_file_aio_read);
1220
1221ssize_t
1222generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1223{
1224 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1225 struct kiocb kiocb;
1226 ssize_t ret;
1227
1228 init_sync_kiocb(&kiocb, filp);
1229 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1230 if (-EIOCBQUEUED == ret)
1231 ret = wait_on_sync_kiocb(&kiocb);
1232 return ret;
1233}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234EXPORT_SYMBOL(generic_file_read);
1235
1236int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1237{
1238 ssize_t written;
1239 unsigned long count = desc->count;
1240 struct file *file = desc->arg.data;
1241
1242 if (size > count)
1243 size = count;
1244
1245 written = file->f_op->sendpage(file, page, offset,
1246 size, &file->f_pos, size<count);
1247 if (written < 0) {
1248 desc->error = written;
1249 written = 0;
1250 }
1251 desc->count = count - written;
1252 desc->written += written;
1253 return written;
1254}
1255
1256ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1257 size_t count, read_actor_t actor, void *target)
1258{
1259 read_descriptor_t desc;
1260
1261 if (!count)
1262 return 0;
1263
1264 desc.written = 0;
1265 desc.count = count;
1266 desc.arg.data = target;
1267 desc.error = 0;
1268
1269 do_generic_file_read(in_file, ppos, &desc, actor);
1270 if (desc.written)
1271 return desc.written;
1272 return desc.error;
1273}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274EXPORT_SYMBOL(generic_file_sendfile);
1275
1276static ssize_t
1277do_readahead(struct address_space *mapping, struct file *filp,
1278 unsigned long index, unsigned long nr)
1279{
1280 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1281 return -EINVAL;
1282
1283 force_page_cache_readahead(mapping, filp, index,
1284 max_sane_readahead(nr));
1285 return 0;
1286}
1287
1288asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1289{
1290 ssize_t ret;
1291 struct file *file;
1292
1293 ret = -EBADF;
1294 file = fget(fd);
1295 if (file) {
1296 if (file->f_mode & FMODE_READ) {
1297 struct address_space *mapping = file->f_mapping;
1298 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1299 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1300 unsigned long len = end - start + 1;
1301 ret = do_readahead(mapping, file, start, len);
1302 }
1303 fput(file);
1304 }
1305 return ret;
1306}
1307
1308#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001309static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1310/**
1311 * page_cache_read - adds requested page to the page cache if not already there
1312 * @file: file to read
1313 * @offset: page index
1314 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 * This adds the requested page to the page cache if it isn't already there,
1316 * and schedules an I/O to read in its contents from disk.
1317 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318static int fastcall page_cache_read(struct file * file, unsigned long offset)
1319{
1320 struct address_space *mapping = file->f_mapping;
1321 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001322 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Zach Brown994fc28c2005-12-15 14:28:17 -08001324 do {
1325 page = page_cache_alloc_cold(mapping);
1326 if (!page)
1327 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Zach Brown994fc28c2005-12-15 14:28:17 -08001329 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1330 if (ret == 0)
1331 ret = mapping->a_ops->readpage(file, page);
1332 else if (ret == -EEXIST)
1333 ret = 0; /* losing race to add is OK */
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Zach Brown994fc28c2005-12-15 14:28:17 -08001337 } while (ret == AOP_TRUNCATED_PAGE);
1338
1339 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340}
1341
1342#define MMAP_LOTSAMISS (100)
1343
Randy Dunlap485bb992006-06-23 02:03:49 -07001344/**
1345 * filemap_nopage - read in file data for page fault handling
1346 * @area: the applicable vm_area
1347 * @address: target address to read in
1348 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1349 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 * filemap_nopage() is invoked via the vma operations vector for a
1351 * mapped memory region to read in file data during a page fault.
1352 *
1353 * The goto's are kind of ugly, but this streamlines the normal case of having
1354 * it in the page cache, and handles the special cases reasonably without
1355 * having a lot of duplicated code.
1356 */
1357struct page *filemap_nopage(struct vm_area_struct *area,
1358 unsigned long address, int *type)
1359{
1360 int error;
1361 struct file *file = area->vm_file;
1362 struct address_space *mapping = file->f_mapping;
1363 struct file_ra_state *ra = &file->f_ra;
1364 struct inode *inode = mapping->host;
1365 struct page *page;
1366 unsigned long size, pgoff;
1367 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1368
1369 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1370
1371retry_all:
1372 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1373 if (pgoff >= size)
1374 goto outside_data_content;
1375
1376 /* If we don't want any read-ahead, don't bother */
1377 if (VM_RandomReadHint(area))
1378 goto no_cached_page;
1379
1380 /*
1381 * The readahead code wants to be told about each and every page
1382 * so it can build and shrink its windows appropriately
1383 *
1384 * For sequential accesses, we use the generic readahead logic.
1385 */
1386 if (VM_SequentialReadHint(area))
1387 page_cache_readahead(mapping, ra, file, pgoff, 1);
1388
1389 /*
1390 * Do we have something in the page cache already?
1391 */
1392retry_find:
1393 page = find_get_page(mapping, pgoff);
1394 if (!page) {
1395 unsigned long ra_pages;
1396
1397 if (VM_SequentialReadHint(area)) {
1398 handle_ra_miss(mapping, ra, pgoff);
1399 goto no_cached_page;
1400 }
1401 ra->mmap_miss++;
1402
1403 /*
1404 * Do we miss much more than hit in this file? If so,
1405 * stop bothering with read-ahead. It will only hurt.
1406 */
1407 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1408 goto no_cached_page;
1409
1410 /*
1411 * To keep the pgmajfault counter straight, we need to
1412 * check did_readaround, as this is an inner loop.
1413 */
1414 if (!did_readaround) {
1415 majmin = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001416 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 }
1418 did_readaround = 1;
1419 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1420 if (ra_pages) {
1421 pgoff_t start = 0;
1422
1423 if (pgoff > ra_pages / 2)
1424 start = pgoff - ra_pages / 2;
1425 do_page_cache_readahead(mapping, file, start, ra_pages);
1426 }
1427 page = find_get_page(mapping, pgoff);
1428 if (!page)
1429 goto no_cached_page;
1430 }
1431
1432 if (!did_readaround)
1433 ra->mmap_hit++;
1434
1435 /*
1436 * Ok, found a page in the page cache, now we need to check
1437 * that it's up-to-date.
1438 */
1439 if (!PageUptodate(page))
1440 goto page_not_uptodate;
1441
1442success:
1443 /*
1444 * Found the page and have a reference on it.
1445 */
1446 mark_page_accessed(page);
1447 if (type)
1448 *type = majmin;
1449 return page;
1450
1451outside_data_content:
1452 /*
1453 * An external ptracer can access pages that normally aren't
1454 * accessible..
1455 */
1456 if (area->vm_mm == current->mm)
1457 return NULL;
1458 /* Fall through to the non-read-ahead case */
1459no_cached_page:
1460 /*
1461 * We're only likely to ever get here if MADV_RANDOM is in
1462 * effect.
1463 */
1464 error = page_cache_read(file, pgoff);
1465 grab_swap_token();
1466
1467 /*
1468 * The page we want has now been added to the page cache.
1469 * In the unlikely event that someone removed it in the
1470 * meantime, we'll just come back here and read it again.
1471 */
1472 if (error >= 0)
1473 goto retry_find;
1474
1475 /*
1476 * An error return from page_cache_read can result if the
1477 * system is low on memory, or a problem occurs while trying
1478 * to schedule I/O.
1479 */
1480 if (error == -ENOMEM)
1481 return NOPAGE_OOM;
1482 return NULL;
1483
1484page_not_uptodate:
1485 if (!did_readaround) {
1486 majmin = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001487 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 }
1489 lock_page(page);
1490
1491 /* Did it get unhashed while we waited for it? */
1492 if (!page->mapping) {
1493 unlock_page(page);
1494 page_cache_release(page);
1495 goto retry_all;
1496 }
1497
1498 /* Did somebody else get it up-to-date? */
1499 if (PageUptodate(page)) {
1500 unlock_page(page);
1501 goto success;
1502 }
1503
Zach Brown994fc28c2005-12-15 14:28:17 -08001504 error = mapping->a_ops->readpage(file, page);
1505 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 wait_on_page_locked(page);
1507 if (PageUptodate(page))
1508 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001509 } else if (error == AOP_TRUNCATED_PAGE) {
1510 page_cache_release(page);
1511 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
1513
1514 /*
1515 * Umm, take care of errors if the page isn't up-to-date.
1516 * Try to re-read it _once_. We do this synchronously,
1517 * because there really aren't any performance issues here
1518 * and we need to check for errors.
1519 */
1520 lock_page(page);
1521
1522 /* Somebody truncated the page on us? */
1523 if (!page->mapping) {
1524 unlock_page(page);
1525 page_cache_release(page);
1526 goto retry_all;
1527 }
1528
1529 /* Somebody else successfully read it in? */
1530 if (PageUptodate(page)) {
1531 unlock_page(page);
1532 goto success;
1533 }
1534 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001535 error = mapping->a_ops->readpage(file, page);
1536 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 wait_on_page_locked(page);
1538 if (PageUptodate(page))
1539 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001540 } else if (error == AOP_TRUNCATED_PAGE) {
1541 page_cache_release(page);
1542 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 }
1544
1545 /*
1546 * Things didn't work out. Return zero to tell the
1547 * mm layer so, possibly freeing the page cache page first.
1548 */
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001549 shrink_readahead_size_eio(file, ra);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 page_cache_release(page);
1551 return NULL;
1552}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553EXPORT_SYMBOL(filemap_nopage);
1554
1555static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1556 int nonblock)
1557{
1558 struct address_space *mapping = file->f_mapping;
1559 struct page *page;
1560 int error;
1561
1562 /*
1563 * Do we have something in the page cache already?
1564 */
1565retry_find:
1566 page = find_get_page(mapping, pgoff);
1567 if (!page) {
1568 if (nonblock)
1569 return NULL;
1570 goto no_cached_page;
1571 }
1572
1573 /*
1574 * Ok, found a page in the page cache, now we need to check
1575 * that it's up-to-date.
1576 */
Jeff Moyerd3457342005-04-16 15:24:05 -07001577 if (!PageUptodate(page)) {
1578 if (nonblock) {
1579 page_cache_release(page);
1580 return NULL;
1581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 goto page_not_uptodate;
Jeff Moyerd3457342005-04-16 15:24:05 -07001583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585success:
1586 /*
1587 * Found the page and have a reference on it.
1588 */
1589 mark_page_accessed(page);
1590 return page;
1591
1592no_cached_page:
1593 error = page_cache_read(file, pgoff);
1594
1595 /*
1596 * The page we want has now been added to the page cache.
1597 * In the unlikely event that someone removed it in the
1598 * meantime, we'll just come back here and read it again.
1599 */
1600 if (error >= 0)
1601 goto retry_find;
1602
1603 /*
1604 * An error return from page_cache_read can result if the
1605 * system is low on memory, or a problem occurs while trying
1606 * to schedule I/O.
1607 */
1608 return NULL;
1609
1610page_not_uptodate:
1611 lock_page(page);
1612
1613 /* Did it get unhashed while we waited for it? */
1614 if (!page->mapping) {
1615 unlock_page(page);
1616 goto err;
1617 }
1618
1619 /* Did somebody else get it up-to-date? */
1620 if (PageUptodate(page)) {
1621 unlock_page(page);
1622 goto success;
1623 }
1624
Zach Brown994fc28c2005-12-15 14:28:17 -08001625 error = mapping->a_ops->readpage(file, page);
1626 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 wait_on_page_locked(page);
1628 if (PageUptodate(page))
1629 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001630 } else if (error == AOP_TRUNCATED_PAGE) {
1631 page_cache_release(page);
1632 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 }
1634
1635 /*
1636 * Umm, take care of errors if the page isn't up-to-date.
1637 * Try to re-read it _once_. We do this synchronously,
1638 * because there really aren't any performance issues here
1639 * and we need to check for errors.
1640 */
1641 lock_page(page);
1642
1643 /* Somebody truncated the page on us? */
1644 if (!page->mapping) {
1645 unlock_page(page);
1646 goto err;
1647 }
1648 /* Somebody else successfully read it in? */
1649 if (PageUptodate(page)) {
1650 unlock_page(page);
1651 goto success;
1652 }
1653
1654 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001655 error = mapping->a_ops->readpage(file, page);
1656 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 wait_on_page_locked(page);
1658 if (PageUptodate(page))
1659 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001660 } else if (error == AOP_TRUNCATED_PAGE) {
1661 page_cache_release(page);
1662 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
1664
1665 /*
1666 * Things didn't work out. Return zero to tell the
1667 * mm layer so, possibly freeing the page cache page first.
1668 */
1669err:
1670 page_cache_release(page);
1671
1672 return NULL;
1673}
1674
1675int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1676 unsigned long len, pgprot_t prot, unsigned long pgoff,
1677 int nonblock)
1678{
1679 struct file *file = vma->vm_file;
1680 struct address_space *mapping = file->f_mapping;
1681 struct inode *inode = mapping->host;
1682 unsigned long size;
1683 struct mm_struct *mm = vma->vm_mm;
1684 struct page *page;
1685 int err;
1686
1687 if (!nonblock)
1688 force_page_cache_readahead(mapping, vma->vm_file,
1689 pgoff, len >> PAGE_CACHE_SHIFT);
1690
1691repeat:
1692 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1693 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1694 return -EINVAL;
1695
1696 page = filemap_getpage(file, pgoff, nonblock);
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001697
1698 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1699 * done in shmem_populate calling shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 if (!page && !nonblock)
1701 return -ENOMEM;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 if (page) {
1704 err = install_page(mm, vma, addr, page, prot);
1705 if (err) {
1706 page_cache_release(page);
1707 return err;
1708 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001709 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001710 /* No page was found just because we can't read it in now (being
1711 * here implies nonblock != 0), but the page may exist, so set
1712 * the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 err = install_file_pte(mm, vma, addr, pgoff, prot);
1714 if (err)
1715 return err;
1716 }
1717
1718 len -= PAGE_SIZE;
1719 addr += PAGE_SIZE;
1720 pgoff++;
1721 if (len)
1722 goto repeat;
1723
1724 return 0;
1725}
Nikita Danilovb1459462005-10-29 18:17:02 -07001726EXPORT_SYMBOL(filemap_populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
1728struct vm_operations_struct generic_file_vm_ops = {
1729 .nopage = filemap_nopage,
1730 .populate = filemap_populate,
1731};
1732
1733/* This is used for a general mmap of a disk file */
1734
1735int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1736{
1737 struct address_space *mapping = file->f_mapping;
1738
1739 if (!mapping->a_ops->readpage)
1740 return -ENOEXEC;
1741 file_accessed(file);
1742 vma->vm_ops = &generic_file_vm_ops;
1743 return 0;
1744}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
1746/*
1747 * This is for filesystems which do not implement ->writepage.
1748 */
1749int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1750{
1751 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1752 return -EINVAL;
1753 return generic_file_mmap(file, vma);
1754}
1755#else
1756int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1757{
1758 return -ENOSYS;
1759}
1760int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1761{
1762 return -ENOSYS;
1763}
1764#endif /* CONFIG_MMU */
1765
1766EXPORT_SYMBOL(generic_file_mmap);
1767EXPORT_SYMBOL(generic_file_readonly_mmap);
1768
1769static inline struct page *__read_cache_page(struct address_space *mapping,
1770 unsigned long index,
1771 int (*filler)(void *,struct page*),
1772 void *data)
1773{
1774 struct page *page, *cached_page = NULL;
1775 int err;
1776repeat:
1777 page = find_get_page(mapping, index);
1778 if (!page) {
1779 if (!cached_page) {
1780 cached_page = page_cache_alloc_cold(mapping);
1781 if (!cached_page)
1782 return ERR_PTR(-ENOMEM);
1783 }
1784 err = add_to_page_cache_lru(cached_page, mapping,
1785 index, GFP_KERNEL);
1786 if (err == -EEXIST)
1787 goto repeat;
1788 if (err < 0) {
1789 /* Presumably ENOMEM for radix tree node */
1790 page_cache_release(cached_page);
1791 return ERR_PTR(err);
1792 }
1793 page = cached_page;
1794 cached_page = NULL;
1795 err = filler(data, page);
1796 if (err < 0) {
1797 page_cache_release(page);
1798 page = ERR_PTR(err);
1799 }
1800 }
1801 if (cached_page)
1802 page_cache_release(cached_page);
1803 return page;
1804}
1805
Randy Dunlap485bb992006-06-23 02:03:49 -07001806/**
1807 * read_cache_page - read into page cache, fill it if needed
1808 * @mapping: the page's address_space
1809 * @index: the page index
1810 * @filler: function to perform the read
1811 * @data: destination for read data
1812 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 * Read into the page cache. If a page already exists,
1814 * and PageUptodate() is not set, try to fill the page.
1815 */
1816struct page *read_cache_page(struct address_space *mapping,
1817 unsigned long index,
1818 int (*filler)(void *,struct page*),
1819 void *data)
1820{
1821 struct page *page;
1822 int err;
1823
1824retry:
1825 page = __read_cache_page(mapping, index, filler, data);
1826 if (IS_ERR(page))
1827 goto out;
1828 mark_page_accessed(page);
1829 if (PageUptodate(page))
1830 goto out;
1831
1832 lock_page(page);
1833 if (!page->mapping) {
1834 unlock_page(page);
1835 page_cache_release(page);
1836 goto retry;
1837 }
1838 if (PageUptodate(page)) {
1839 unlock_page(page);
1840 goto out;
1841 }
1842 err = filler(data, page);
1843 if (err < 0) {
1844 page_cache_release(page);
1845 page = ERR_PTR(err);
1846 }
1847 out:
1848 return page;
1849}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850EXPORT_SYMBOL(read_cache_page);
1851
1852/*
1853 * If the page was newly created, increment its refcount and add it to the
1854 * caller's lru-buffering pagevec. This function is specifically for
1855 * generic_file_write().
1856 */
1857static inline struct page *
1858__grab_cache_page(struct address_space *mapping, unsigned long index,
1859 struct page **cached_page, struct pagevec *lru_pvec)
1860{
1861 int err;
1862 struct page *page;
1863repeat:
1864 page = find_lock_page(mapping, index);
1865 if (!page) {
1866 if (!*cached_page) {
1867 *cached_page = page_cache_alloc(mapping);
1868 if (!*cached_page)
1869 return NULL;
1870 }
1871 err = add_to_page_cache(*cached_page, mapping,
1872 index, GFP_KERNEL);
1873 if (err == -EEXIST)
1874 goto repeat;
1875 if (err == 0) {
1876 page = *cached_page;
1877 page_cache_get(page);
1878 if (!pagevec_add(lru_pvec, page))
1879 __pagevec_lru_add(lru_pvec);
1880 *cached_page = NULL;
1881 }
1882 }
1883 return page;
1884}
1885
1886/*
1887 * The logic we want is
1888 *
1889 * if suid or (sgid and xgrp)
1890 * remove privs
1891 */
1892int remove_suid(struct dentry *dentry)
1893{
1894 mode_t mode = dentry->d_inode->i_mode;
1895 int kill = 0;
1896 int result = 0;
1897
1898 /* suid always must be killed */
1899 if (unlikely(mode & S_ISUID))
1900 kill = ATTR_KILL_SUID;
1901
1902 /*
1903 * sgid without any exec bits is just a mandatory locking mark; leave
1904 * it alone. If some exec bits are set, it's a real sgid; kill it.
1905 */
1906 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1907 kill |= ATTR_KILL_SGID;
1908
1909 if (unlikely(kill && !capable(CAP_FSETID))) {
1910 struct iattr newattrs;
1911
1912 newattrs.ia_valid = ATTR_FORCE | kill;
1913 result = notify_change(dentry, &newattrs);
1914 }
1915 return result;
1916}
1917EXPORT_SYMBOL(remove_suid);
1918
Carsten Otteceffc072005-06-23 22:05:25 -07001919size_t
NeilBrown01408c42006-06-25 05:47:58 -07001920__filemap_copy_from_user_iovec_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 const struct iovec *iov, size_t base, size_t bytes)
1922{
1923 size_t copied = 0, left = 0;
1924
1925 while (bytes) {
1926 char __user *buf = iov->iov_base + base;
1927 int copy = min(bytes, iov->iov_len - base);
1928
1929 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001930 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 copied += copy;
1932 bytes -= copy;
1933 vaddr += copy;
1934 iov++;
1935
NeilBrown01408c42006-06-25 05:47:58 -07001936 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 }
1939 return copied - left;
1940}
1941
1942/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 * Performs necessary checks before doing a write
1944 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001945 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 * Returns appropriate error code that caller should return or
1947 * zero in case that write should be allowed.
1948 */
1949inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1950{
1951 struct inode *inode = file->f_mapping->host;
1952 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1953
1954 if (unlikely(*pos < 0))
1955 return -EINVAL;
1956
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (!isblk) {
1958 /* FIXME: this is for backwards compatibility with 2.4 */
1959 if (file->f_flags & O_APPEND)
1960 *pos = i_size_read(inode);
1961
1962 if (limit != RLIM_INFINITY) {
1963 if (*pos >= limit) {
1964 send_sig(SIGXFSZ, current, 0);
1965 return -EFBIG;
1966 }
1967 if (*count > limit - (typeof(limit))*pos) {
1968 *count = limit - (typeof(limit))*pos;
1969 }
1970 }
1971 }
1972
1973 /*
1974 * LFS rule
1975 */
1976 if (unlikely(*pos + *count > MAX_NON_LFS &&
1977 !(file->f_flags & O_LARGEFILE))) {
1978 if (*pos >= MAX_NON_LFS) {
1979 send_sig(SIGXFSZ, current, 0);
1980 return -EFBIG;
1981 }
1982 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1983 *count = MAX_NON_LFS - (unsigned long)*pos;
1984 }
1985 }
1986
1987 /*
1988 * Are we about to exceed the fs block limit ?
1989 *
1990 * If we have written data it becomes a short write. If we have
1991 * exceeded without writing data we send a signal and return EFBIG.
1992 * Linus frestrict idea will clean these up nicely..
1993 */
1994 if (likely(!isblk)) {
1995 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1996 if (*count || *pos > inode->i_sb->s_maxbytes) {
1997 send_sig(SIGXFSZ, current, 0);
1998 return -EFBIG;
1999 }
2000 /* zero-length writes at ->s_maxbytes are OK */
2001 }
2002
2003 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2004 *count = inode->i_sb->s_maxbytes - *pos;
2005 } else {
2006 loff_t isize;
2007 if (bdev_read_only(I_BDEV(inode)))
2008 return -EPERM;
2009 isize = i_size_read(inode);
2010 if (*pos >= isize) {
2011 if (*count || *pos > isize)
2012 return -ENOSPC;
2013 }
2014
2015 if (*pos + *count > isize)
2016 *count = isize - *pos;
2017 }
2018 return 0;
2019}
2020EXPORT_SYMBOL(generic_write_checks);
2021
2022ssize_t
2023generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2024 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2025 size_t count, size_t ocount)
2026{
2027 struct file *file = iocb->ki_filp;
2028 struct address_space *mapping = file->f_mapping;
2029 struct inode *inode = mapping->host;
2030 ssize_t written;
2031
2032 if (count != ocount)
2033 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2034
2035 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2036 if (written > 0) {
2037 loff_t end = pos + written;
2038 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2039 i_size_write(inode, end);
2040 mark_inode_dirty(inode);
2041 }
2042 *ppos = end;
2043 }
2044
2045 /*
2046 * Sync the fs metadata but not the minor inode changes and
2047 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002048 * i_mutex is held, which protects generic_osync_inode() from
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 * livelocking.
2050 */
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002051 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2052 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2053 if (err < 0)
2054 written = err;
2055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (written == count && !is_sync_kiocb(iocb))
2057 written = -EIOCBQUEUED;
2058 return written;
2059}
2060EXPORT_SYMBOL(generic_file_direct_write);
2061
2062ssize_t
2063generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2064 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2065 size_t count, ssize_t written)
2066{
2067 struct file *file = iocb->ki_filp;
2068 struct address_space * mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002069 const struct address_space_operations *a_ops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 struct inode *inode = mapping->host;
2071 long status = 0;
2072 struct page *page;
2073 struct page *cached_page = NULL;
2074 size_t bytes;
2075 struct pagevec lru_pvec;
2076 const struct iovec *cur_iov = iov; /* current iovec */
2077 size_t iov_base = 0; /* offset in the current iovec */
2078 char __user *buf;
2079
2080 pagevec_init(&lru_pvec, 0);
2081
2082 /*
2083 * handle partial DIO write. Adjust cur_iov if needed.
2084 */
2085 if (likely(nr_segs == 1))
2086 buf = iov->iov_base + written;
2087 else {
2088 filemap_set_next_iovec(&cur_iov, &iov_base, written);
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002089 buf = cur_iov->iov_base + iov_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 }
2091
2092 do {
2093 unsigned long index;
2094 unsigned long offset;
2095 size_t copied;
2096
2097 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2098 index = pos >> PAGE_CACHE_SHIFT;
2099 bytes = PAGE_CACHE_SIZE - offset;
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07002100
2101 /* Limit the size of the copy to the caller's write size */
2102 bytes = min(bytes, count);
2103
2104 /*
2105 * Limit the size of the copy to that of the current segment,
2106 * because fault_in_pages_readable() doesn't know how to walk
2107 * segments.
2108 */
2109 bytes = min(bytes, cur_iov->iov_len - iov_base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
2111 /*
2112 * Bring in the user page that we will copy from _first_.
2113 * Otherwise there's a nasty deadlock on copying from the
2114 * same page as we're writing to, without it being marked
2115 * up-to-date.
2116 */
Vladimir V. Saveliev6527c2b2006-06-27 02:53:57 -07002117 fault_in_pages_readable(buf, bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2120 if (!page) {
2121 status = -ENOMEM;
2122 break;
2123 }
2124
Andrew Morton81b0c872006-06-29 02:24:26 -07002125 if (unlikely(bytes == 0)) {
2126 status = 0;
2127 copied = 0;
2128 goto zero_length_segment;
2129 }
2130
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2132 if (unlikely(status)) {
2133 loff_t isize = i_size_read(inode);
Zach Brown994fc28c2005-12-15 14:28:17 -08002134
2135 if (status != AOP_TRUNCATED_PAGE)
2136 unlock_page(page);
2137 page_cache_release(page);
2138 if (status == AOP_TRUNCATED_PAGE)
2139 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 /*
2141 * prepare_write() may have instantiated a few blocks
2142 * outside i_size. Trim these off again.
2143 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (pos + bytes > isize)
2145 vmtruncate(inode, isize);
2146 break;
2147 }
2148 if (likely(nr_segs == 1))
2149 copied = filemap_copy_from_user(page, offset,
2150 buf, bytes);
2151 else
2152 copied = filemap_copy_from_user_iovec(page, offset,
2153 cur_iov, iov_base, bytes);
2154 flush_dcache_page(page);
2155 status = a_ops->commit_write(file, page, offset, offset+bytes);
Zach Brown994fc28c2005-12-15 14:28:17 -08002156 if (status == AOP_TRUNCATED_PAGE) {
2157 page_cache_release(page);
2158 continue;
2159 }
Andrew Morton81b0c872006-06-29 02:24:26 -07002160zero_length_segment:
2161 if (likely(copied >= 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 if (!status)
2163 status = copied;
2164
2165 if (status >= 0) {
2166 written += status;
2167 count -= status;
2168 pos += status;
2169 buf += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002170 if (unlikely(nr_segs > 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 filemap_set_next_iovec(&cur_iov,
2172 &iov_base, status);
Badari Pulavartyb0cfbd92005-06-25 14:55:42 -07002173 if (count)
2174 buf = cur_iov->iov_base +
2175 iov_base;
Martin Schwidefskya5117182005-06-06 13:35:54 -07002176 } else {
2177 iov_base += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 }
2180 }
2181 if (unlikely(copied != bytes))
2182 if (status >= 0)
2183 status = -EFAULT;
2184 unlock_page(page);
2185 mark_page_accessed(page);
2186 page_cache_release(page);
2187 if (status < 0)
2188 break;
2189 balance_dirty_pages_ratelimited(mapping);
2190 cond_resched();
2191 } while (count);
2192 *ppos = pos;
2193
2194 if (cached_page)
2195 page_cache_release(cached_page);
2196
2197 /*
2198 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2199 */
2200 if (likely(status >= 0)) {
2201 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2202 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2203 status = generic_osync_inode(inode, mapping,
2204 OSYNC_METADATA|OSYNC_DATA);
2205 }
2206 }
2207
2208 /*
2209 * If we get here for O_DIRECT writes then we must have fallen through
2210 * to buffered writes (block instantiation inside i_size). So we sync
2211 * the file data here, to try to honour O_DIRECT expectations.
2212 */
2213 if (unlikely(file->f_flags & O_DIRECT) && written)
2214 status = filemap_write_and_wait(mapping);
2215
2216 pagevec_lru_add(&lru_pvec);
2217 return written ? written : status;
2218}
2219EXPORT_SYMBOL(generic_file_buffered_write);
2220
Adrian Bunk5ce78522005-09-10 00:26:28 -07002221static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2223 unsigned long nr_segs, loff_t *ppos)
2224{
2225 struct file *file = iocb->ki_filp;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002226 const struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 size_t ocount; /* original count */
2228 size_t count; /* after file limit checks */
2229 struct inode *inode = mapping->host;
2230 unsigned long seg;
2231 loff_t pos;
2232 ssize_t written;
2233 ssize_t err;
2234
2235 ocount = 0;
2236 for (seg = 0; seg < nr_segs; seg++) {
2237 const struct iovec *iv = &iov[seg];
2238
2239 /*
2240 * If any segment has a negative length, or the cumulative
2241 * length ever wraps negative then return -EINVAL.
2242 */
2243 ocount += iv->iov_len;
2244 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2245 return -EINVAL;
2246 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2247 continue;
2248 if (seg == 0)
2249 return -EFAULT;
2250 nr_segs = seg;
2251 ocount -= iv->iov_len; /* This segment is no good */
2252 break;
2253 }
2254
2255 count = ocount;
2256 pos = *ppos;
2257
2258 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2259
2260 /* We can write back this queue in page reclaim */
2261 current->backing_dev_info = mapping->backing_dev_info;
2262 written = 0;
2263
2264 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2265 if (err)
2266 goto out;
2267
2268 if (count == 0)
2269 goto out;
2270
2271 err = remove_suid(file->f_dentry);
2272 if (err)
2273 goto out;
2274
Christoph Hellwig870f4812006-01-09 20:52:01 -08002275 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
2277 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2278 if (unlikely(file->f_flags & O_DIRECT)) {
2279 written = generic_file_direct_write(iocb, iov,
2280 &nr_segs, pos, ppos, count, ocount);
2281 if (written < 0 || written == count)
2282 goto out;
2283 /*
2284 * direct-io write to a hole: fall through to buffered I/O
2285 * for completing the rest of the request.
2286 */
2287 pos += written;
2288 count -= written;
2289 }
2290
2291 written = generic_file_buffered_write(iocb, iov, nr_segs,
2292 pos, ppos, count, written);
2293out:
2294 current->backing_dev_info = NULL;
2295 return written ? written : err;
2296}
2297EXPORT_SYMBOL(generic_file_aio_write_nolock);
2298
2299ssize_t
2300generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2301 unsigned long nr_segs, loff_t *ppos)
2302{
2303 struct file *file = iocb->ki_filp;
2304 struct address_space *mapping = file->f_mapping;
2305 struct inode *inode = mapping->host;
2306 ssize_t ret;
2307 loff_t pos = *ppos;
2308
2309 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2310
2311 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2312 int err;
2313
2314 err = sync_page_range_nolock(inode, mapping, pos, ret);
2315 if (err < 0)
2316 ret = err;
2317 }
2318 return ret;
2319}
2320
Adrian Bunk5ce78522005-09-10 00:26:28 -07002321static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322__generic_file_write_nolock(struct file *file, const struct iovec *iov,
2323 unsigned long nr_segs, loff_t *ppos)
2324{
2325 struct kiocb kiocb;
2326 ssize_t ret;
2327
2328 init_sync_kiocb(&kiocb, file);
2329 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2330 if (ret == -EIOCBQUEUED)
2331 ret = wait_on_sync_kiocb(&kiocb);
2332 return ret;
2333}
2334
2335ssize_t
2336generic_file_write_nolock(struct file *file, const struct iovec *iov,
2337 unsigned long nr_segs, loff_t *ppos)
2338{
2339 struct kiocb kiocb;
2340 ssize_t ret;
2341
2342 init_sync_kiocb(&kiocb, file);
2343 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2344 if (-EIOCBQUEUED == ret)
2345 ret = wait_on_sync_kiocb(&kiocb);
2346 return ret;
2347}
2348EXPORT_SYMBOL(generic_file_write_nolock);
2349
2350ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2351 size_t count, loff_t pos)
2352{
2353 struct file *file = iocb->ki_filp;
2354 struct address_space *mapping = file->f_mapping;
2355 struct inode *inode = mapping->host;
2356 ssize_t ret;
2357 struct iovec local_iov = { .iov_base = (void __user *)buf,
2358 .iov_len = count };
2359
2360 BUG_ON(iocb->ki_pos != pos);
2361
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002362 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2364 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002365 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
2367 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2368 ssize_t err;
2369
2370 err = sync_page_range(inode, mapping, pos, ret);
2371 if (err < 0)
2372 ret = err;
2373 }
2374 return ret;
2375}
2376EXPORT_SYMBOL(generic_file_aio_write);
2377
2378ssize_t generic_file_write(struct file *file, const char __user *buf,
2379 size_t count, loff_t *ppos)
2380{
2381 struct address_space *mapping = file->f_mapping;
2382 struct inode *inode = mapping->host;
2383 ssize_t ret;
2384 struct iovec local_iov = { .iov_base = (void __user *)buf,
2385 .iov_len = count };
2386
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002387 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002389 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
2391 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2392 ssize_t err;
2393
2394 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2395 if (err < 0)
2396 ret = err;
2397 }
2398 return ret;
2399}
2400EXPORT_SYMBOL(generic_file_write);
2401
2402ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2403 unsigned long nr_segs, loff_t *ppos)
2404{
2405 struct kiocb kiocb;
2406 ssize_t ret;
2407
2408 init_sync_kiocb(&kiocb, filp);
2409 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2410 if (-EIOCBQUEUED == ret)
2411 ret = wait_on_sync_kiocb(&kiocb);
2412 return ret;
2413}
2414EXPORT_SYMBOL(generic_file_readv);
2415
2416ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2417 unsigned long nr_segs, loff_t *ppos)
2418{
2419 struct address_space *mapping = file->f_mapping;
2420 struct inode *inode = mapping->host;
2421 ssize_t ret;
2422
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002423 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002425 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
2427 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2428 int err;
2429
2430 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2431 if (err < 0)
2432 ret = err;
2433 }
2434 return ret;
2435}
2436EXPORT_SYMBOL(generic_file_writev);
2437
2438/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002439 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 * went wrong during pagecache shootdown.
2441 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002442static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2444 loff_t offset, unsigned long nr_segs)
2445{
2446 struct file *file = iocb->ki_filp;
2447 struct address_space *mapping = file->f_mapping;
2448 ssize_t retval;
2449 size_t write_len = 0;
2450
2451 /*
2452 * If it's a write, unmap all mmappings of the file up-front. This
2453 * will cause any pte dirty bits to be propagated into the pageframes
2454 * for the subsequent filemap_write_and_wait().
2455 */
2456 if (rw == WRITE) {
2457 write_len = iov_length(iov, nr_segs);
2458 if (mapping_mapped(mapping))
2459 unmap_mapping_range(mapping, offset, write_len, 0);
2460 }
2461
2462 retval = filemap_write_and_wait(mapping);
2463 if (retval == 0) {
2464 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2465 offset, nr_segs);
2466 if (rw == WRITE && mapping->nrpages) {
2467 pgoff_t end = (offset + write_len - 1)
2468 >> PAGE_CACHE_SHIFT;
2469 int err = invalidate_inode_pages2_range(mapping,
2470 offset >> PAGE_CACHE_SHIFT, end);
2471 if (err)
2472 retval = err;
2473 }
2474 }
2475 return retval;
2476}