blob: a965b6b35f266bce90ad5703ec4c81b7270e9d8e [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 */
12#include <linux/config.h>
13#include <linux/module.h>
14#include <linux/slab.h>
15#include <linux/compiler.h>
16#include <linux/fs.h>
17#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>
Carsten Otteceffc072005-06-23 22:05:25 -070032#include "filemap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * FIXME: remove all knowledge of the buffer layer from the core VM
35 */
36#include <linux/buffer_head.h> /* for generic_osync_inode */
37
38#include <asm/uaccess.h>
39#include <asm/mman.h>
40
Adrian Bunk5ce78522005-09-10 00:26:28 -070041static ssize_t
42generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
43 loff_t offset, unsigned long nr_segs);
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/*
46 * Shared mappings implemented 30.11.1994. It's not fully working yet,
47 * though.
48 *
49 * Shared mappings now work. 15.8.1995 Bruno.
50 *
51 * finished 'unifying' the page and buffer cache and SMP-threaded the
52 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
53 *
54 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
55 */
56
57/*
58 * Lock ordering:
59 *
60 * ->i_mmap_lock (vmtruncate)
61 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070062 * ->swap_lock (exclusive_swap_page, others)
63 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080065 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * ->i_mmap_lock (truncate->unmap_mapping_range)
67 *
68 * ->mmap_sem
69 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070070 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
72 *
73 * ->mmap_sem
74 * ->lock_page (access_process_vm)
75 *
76 * ->mmap_sem
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080077 * ->i_mutex (msync)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080079 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 * ->i_alloc_sem (various)
81 *
82 * ->inode_lock
83 * ->sb_lock (fs/fs-writeback.c)
84 * ->mapping->tree_lock (__sync_single_inode)
85 *
86 * ->i_mmap_lock
87 * ->anon_vma.lock (vma_adjust)
88 *
89 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070090 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070093 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
97 * ->private_lock (page_remove_rmap->set_page_dirty)
98 * ->tree_lock (page_remove_rmap->set_page_dirty)
99 * ->inode_lock (page_remove_rmap->set_page_dirty)
100 * ->inode_lock (zap_pte_range->set_page_dirty)
101 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
102 *
103 * ->task->proc_lock
104 * ->dcache_lock (proc_pid_lookup)
105 */
106
107/*
108 * Remove a page from the page cache and free it. Caller has to make
109 * sure the page is locked and that nobody else uses it - or that usage
110 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
111 */
112void __remove_from_page_cache(struct page *page)
113{
114 struct address_space *mapping = page->mapping;
115
116 radix_tree_delete(&mapping->page_tree, page->index);
117 page->mapping = NULL;
118 mapping->nrpages--;
119 pagecache_acct(-1);
120}
121
122void remove_from_page_cache(struct page *page)
123{
124 struct address_space *mapping = page->mapping;
125
Matt Mackallcd7619d2005-05-01 08:59:01 -0700126 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
128 write_lock_irq(&mapping->tree_lock);
129 __remove_from_page_cache(page);
130 write_unlock_irq(&mapping->tree_lock);
131}
132
133static int sync_page(void *word)
134{
135 struct address_space *mapping;
136 struct page *page;
137
Andi Kleen07808b72005-11-05 17:25:53 +0100138 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
140 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700141 * page_mapping() is being called without PG_locked held.
142 * Some knowledge of the state and use of the page is used to
143 * reduce the requirements down to a memory barrier.
144 * The danger here is of a stale page_mapping() return value
145 * indicating a struct address_space different from the one it's
146 * associated with when it is associated with one.
147 * After smp_mb(), it's either the correct page_mapping() for
148 * the page, or an old page_mapping() and the page's own
149 * page_mapping() has gone NULL.
150 * The ->sync_page() address_space operation must tolerate
151 * page_mapping() going NULL. By an amazing coincidence,
152 * this comes about because none of the users of the page
153 * in the ->sync_page() methods make essential use of the
154 * page_mapping(), merely passing the page down to the backing
155 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700156 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700157 * of interest. When page_mapping() does go NULL, the entire
158 * call stack gracefully ignores the page and returns.
159 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 */
161 smp_mb();
162 mapping = page_mapping(page);
163 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
164 mapping->a_ops->sync_page(page);
165 io_schedule();
166 return 0;
167}
168
169/**
170 * filemap_fdatawrite_range - start writeback against all of a mapping's
171 * dirty pages that lie within the byte offsets <start, end>
Martin Waitz67be2dd2005-05-01 08:59:26 -0700172 * @mapping: address space structure to write
173 * @start: offset in bytes where the range starts
174 * @end: offset in bytes where the range ends
175 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 *
177 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
178 * opposed to a regular memory * cleansing writeback. The difference between
179 * these two operations is that if a dirty page/buffer is encountered, it must
180 * be waited upon, and not just skipped over.
181 */
182static int __filemap_fdatawrite_range(struct address_space *mapping,
183 loff_t start, loff_t end, int sync_mode)
184{
185 int ret;
186 struct writeback_control wbc = {
187 .sync_mode = sync_mode,
188 .nr_to_write = mapping->nrpages * 2,
189 .start = start,
190 .end = end,
191 };
192
193 if (!mapping_cap_writeback_dirty(mapping))
194 return 0;
195
196 ret = do_writepages(mapping, &wbc);
197 return ret;
198}
199
200static inline int __filemap_fdatawrite(struct address_space *mapping,
201 int sync_mode)
202{
203 return __filemap_fdatawrite_range(mapping, 0, 0, sync_mode);
204}
205
206int filemap_fdatawrite(struct address_space *mapping)
207{
208 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
209}
210EXPORT_SYMBOL(filemap_fdatawrite);
211
212static int filemap_fdatawrite_range(struct address_space *mapping,
213 loff_t start, loff_t end)
214{
215 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
216}
217
218/*
219 * This is a mostly non-blocking flush. Not suitable for data-integrity
220 * purposes - I/O may not be started against all dirty pages.
221 */
222int filemap_flush(struct address_space *mapping)
223{
224 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
225}
226EXPORT_SYMBOL(filemap_flush);
227
228/*
229 * Wait for writeback to complete against pages indexed by start->end
230 * inclusive
231 */
232static int wait_on_page_writeback_range(struct address_space *mapping,
233 pgoff_t start, pgoff_t end)
234{
235 struct pagevec pvec;
236 int nr_pages;
237 int ret = 0;
238 pgoff_t index;
239
240 if (end < start)
241 return 0;
242
243 pagevec_init(&pvec, 0);
244 index = start;
245 while ((index <= end) &&
246 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
247 PAGECACHE_TAG_WRITEBACK,
248 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
249 unsigned i;
250
251 for (i = 0; i < nr_pages; i++) {
252 struct page *page = pvec.pages[i];
253
254 /* until radix tree lookup accepts end_index */
255 if (page->index > end)
256 continue;
257
258 wait_on_page_writeback(page);
259 if (PageError(page))
260 ret = -EIO;
261 }
262 pagevec_release(&pvec);
263 cond_resched();
264 }
265
266 /* Check for outstanding write errors */
267 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
268 ret = -ENOSPC;
269 if (test_and_clear_bit(AS_EIO, &mapping->flags))
270 ret = -EIO;
271
272 return ret;
273}
274
275/*
276 * Write and wait upon all the pages in the passed range. This is a "data
277 * integrity" operation. It waits upon in-flight writeout before starting and
278 * waiting upon new writeout. If there was an IO error, return it.
279 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800280 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 * it is otherwise livelockable.
282 */
283int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800284 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
286 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
287 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
288 int ret;
289
290 if (!mapping_cap_writeback_dirty(mapping) || !count)
291 return 0;
292 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
293 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800294 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800296 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 }
298 if (ret == 0)
299 ret = wait_on_page_writeback_range(mapping, start, end);
300 return ret;
301}
302EXPORT_SYMBOL(sync_page_range);
303
304/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800305 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 * as it forces O_SYNC writers to different parts of the same file
307 * to be serialised right until io completion.
308 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800309int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
310 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
312 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
313 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
314 int ret;
315
316 if (!mapping_cap_writeback_dirty(mapping) || !count)
317 return 0;
318 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
319 if (ret == 0)
320 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
321 if (ret == 0)
322 ret = wait_on_page_writeback_range(mapping, start, end);
323 return ret;
324}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800325EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327/**
328 * filemap_fdatawait - walk the list of under-writeback pages of the given
329 * address space and wait for all of them.
330 *
331 * @mapping: address space structure to wait for
332 */
333int filemap_fdatawait(struct address_space *mapping)
334{
335 loff_t i_size = i_size_read(mapping->host);
336
337 if (i_size == 0)
338 return 0;
339
340 return wait_on_page_writeback_range(mapping, 0,
341 (i_size - 1) >> PAGE_CACHE_SHIFT);
342}
343EXPORT_SYMBOL(filemap_fdatawait);
344
345int filemap_write_and_wait(struct address_space *mapping)
346{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800347 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800350 err = filemap_fdatawrite(mapping);
351 /*
352 * Even if the above returned error, the pages may be
353 * written partially (e.g. -ENOSPC), so we wait for it.
354 * But the -EIO is special case, it may indicate the worst
355 * thing (e.g. bug) happened, so we avoid waiting for it.
356 */
357 if (err != -EIO) {
358 int err2 = filemap_fdatawait(mapping);
359 if (!err)
360 err = err2;
361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800363 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800365EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367int filemap_write_and_wait_range(struct address_space *mapping,
368 loff_t lstart, loff_t lend)
369{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800370 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800373 err = __filemap_fdatawrite_range(mapping, lstart, lend,
374 WB_SYNC_ALL);
375 /* See comment of filemap_write_and_wait() */
376 if (err != -EIO) {
377 int err2 = wait_on_page_writeback_range(mapping,
378 lstart >> PAGE_CACHE_SHIFT,
379 lend >> PAGE_CACHE_SHIFT);
380 if (!err)
381 err = err2;
382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800384 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
386
387/*
388 * This function is used to add newly allocated pagecache pages:
389 * the page is new, so we can just run SetPageLocked() against it.
390 * The other page state flags were set by rmqueue().
391 *
392 * This function does not add the page to the LRU. The caller must do that.
393 */
394int add_to_page_cache(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400395 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
397 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
398
399 if (error == 0) {
400 write_lock_irq(&mapping->tree_lock);
401 error = radix_tree_insert(&mapping->page_tree, offset, page);
402 if (!error) {
403 page_cache_get(page);
404 SetPageLocked(page);
405 page->mapping = mapping;
406 page->index = offset;
407 mapping->nrpages++;
408 pagecache_acct(1);
409 }
410 write_unlock_irq(&mapping->tree_lock);
411 radix_tree_preload_end();
412 }
413 return error;
414}
415
416EXPORT_SYMBOL(add_to_page_cache);
417
418int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400419 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
421 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
422 if (ret == 0)
423 lru_cache_add(page);
424 return ret;
425}
426
427/*
428 * In order to wait for pages to become available there must be
429 * waitqueues associated with pages. By using a hash table of
430 * waitqueues where the bucket discipline is to maintain all
431 * waiters on the same queue and wake all when any of the pages
432 * become available, and for the woken contexts to check to be
433 * sure the appropriate page became available, this saves space
434 * at a cost of "thundering herd" phenomena during rare hash
435 * collisions.
436 */
437static wait_queue_head_t *page_waitqueue(struct page *page)
438{
439 const struct zone *zone = page_zone(page);
440
441 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
442}
443
444static inline void wake_up_page(struct page *page, int bit)
445{
446 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
447}
448
449void fastcall wait_on_page_bit(struct page *page, int bit_nr)
450{
451 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
452
453 if (test_bit(bit_nr, &page->flags))
454 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
455 TASK_UNINTERRUPTIBLE);
456}
457EXPORT_SYMBOL(wait_on_page_bit);
458
459/**
460 * unlock_page() - unlock a locked page
461 *
462 * @page: the page
463 *
464 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
465 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
466 * mechananism between PageLocked pages and PageWriteback pages is shared.
467 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
468 *
469 * The first mb is necessary to safely close the critical section opened by the
470 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
471 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
472 * parallel wait_on_page_locked()).
473 */
474void fastcall unlock_page(struct page *page)
475{
476 smp_mb__before_clear_bit();
477 if (!TestClearPageLocked(page))
478 BUG();
479 smp_mb__after_clear_bit();
480 wake_up_page(page, PG_locked);
481}
482EXPORT_SYMBOL(unlock_page);
483
484/*
485 * End writeback against a page.
486 */
487void end_page_writeback(struct page *page)
488{
489 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
490 if (!test_clear_page_writeback(page))
491 BUG();
492 }
493 smp_mb__after_clear_bit();
494 wake_up_page(page, PG_writeback);
495}
496EXPORT_SYMBOL(end_page_writeback);
497
498/*
499 * Get a lock on the page, assuming we need to sleep to get it.
500 *
501 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
502 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
503 * chances are that on the second loop, the block layer's plug list is empty,
504 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
505 */
506void fastcall __lock_page(struct page *page)
507{
508 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
509
510 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
511 TASK_UNINTERRUPTIBLE);
512}
513EXPORT_SYMBOL(__lock_page);
514
515/*
516 * a rather lightweight function, finding and getting a reference to a
517 * hashed page atomically.
518 */
519struct page * find_get_page(struct address_space *mapping, unsigned long offset)
520{
521 struct page *page;
522
523 read_lock_irq(&mapping->tree_lock);
524 page = radix_tree_lookup(&mapping->page_tree, offset);
525 if (page)
526 page_cache_get(page);
527 read_unlock_irq(&mapping->tree_lock);
528 return page;
529}
530
531EXPORT_SYMBOL(find_get_page);
532
533/*
534 * Same as above, but trylock it instead of incrementing the count.
535 */
536struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
537{
538 struct page *page;
539
540 read_lock_irq(&mapping->tree_lock);
541 page = radix_tree_lookup(&mapping->page_tree, offset);
542 if (page && TestSetPageLocked(page))
543 page = NULL;
544 read_unlock_irq(&mapping->tree_lock);
545 return page;
546}
547
548EXPORT_SYMBOL(find_trylock_page);
549
550/**
551 * find_lock_page - locate, pin and lock a pagecache page
552 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700553 * @mapping: the address_space to search
554 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 *
556 * Locates the desired pagecache page, locks it, increments its reference
557 * count and returns its address.
558 *
559 * Returns zero if the page was not present. find_lock_page() may sleep.
560 */
561struct page *find_lock_page(struct address_space *mapping,
562 unsigned long offset)
563{
564 struct page *page;
565
566 read_lock_irq(&mapping->tree_lock);
567repeat:
568 page = radix_tree_lookup(&mapping->page_tree, offset);
569 if (page) {
570 page_cache_get(page);
571 if (TestSetPageLocked(page)) {
572 read_unlock_irq(&mapping->tree_lock);
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800573 __lock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 read_lock_irq(&mapping->tree_lock);
575
576 /* Has the page been truncated while we slept? */
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800577 if (unlikely(page->mapping != mapping ||
578 page->index != offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 unlock_page(page);
580 page_cache_release(page);
581 goto repeat;
582 }
583 }
584 }
585 read_unlock_irq(&mapping->tree_lock);
586 return page;
587}
588
589EXPORT_SYMBOL(find_lock_page);
590
591/**
592 * find_or_create_page - locate or add a pagecache page
593 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700594 * @mapping: the page's address_space
595 * @index: the page's index into the mapping
596 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 *
598 * Locates a page in the pagecache. If the page is not present, a new page
599 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
600 * LRU list. The returned page is locked and has its reference count
601 * incremented.
602 *
603 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
604 * allocation!
605 *
606 * find_or_create_page() returns the desired page's address, or zero on
607 * memory exhaustion.
608 */
609struct page *find_or_create_page(struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400610 unsigned long index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
612 struct page *page, *cached_page = NULL;
613 int err;
614repeat:
615 page = find_lock_page(mapping, index);
616 if (!page) {
617 if (!cached_page) {
618 cached_page = alloc_page(gfp_mask);
619 if (!cached_page)
620 return NULL;
621 }
622 err = add_to_page_cache_lru(cached_page, mapping,
623 index, gfp_mask);
624 if (!err) {
625 page = cached_page;
626 cached_page = NULL;
627 } else if (err == -EEXIST)
628 goto repeat;
629 }
630 if (cached_page)
631 page_cache_release(cached_page);
632 return page;
633}
634
635EXPORT_SYMBOL(find_or_create_page);
636
637/**
638 * find_get_pages - gang pagecache lookup
639 * @mapping: The address_space to search
640 * @start: The starting page index
641 * @nr_pages: The maximum number of pages
642 * @pages: Where the resulting pages are placed
643 *
644 * find_get_pages() will search for and return a group of up to
645 * @nr_pages pages in the mapping. The pages are placed at @pages.
646 * find_get_pages() takes a reference against the returned pages.
647 *
648 * The search returns a group of mapping-contiguous pages with ascending
649 * indexes. There may be holes in the indices due to not-present pages.
650 *
651 * find_get_pages() returns the number of pages which were found.
652 */
653unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
654 unsigned int nr_pages, struct page **pages)
655{
656 unsigned int i;
657 unsigned int ret;
658
659 read_lock_irq(&mapping->tree_lock);
660 ret = radix_tree_gang_lookup(&mapping->page_tree,
661 (void **)pages, start, nr_pages);
662 for (i = 0; i < ret; i++)
663 page_cache_get(pages[i]);
664 read_unlock_irq(&mapping->tree_lock);
665 return ret;
666}
667
668/*
669 * Like find_get_pages, except we only return pages which are tagged with
670 * `tag'. We update *index to index the next page for the traversal.
671 */
672unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
673 int tag, unsigned int nr_pages, struct page **pages)
674{
675 unsigned int i;
676 unsigned int ret;
677
678 read_lock_irq(&mapping->tree_lock);
679 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
680 (void **)pages, *index, nr_pages, tag);
681 for (i = 0; i < ret; i++)
682 page_cache_get(pages[i]);
683 if (ret)
684 *index = pages[ret - 1]->index + 1;
685 read_unlock_irq(&mapping->tree_lock);
686 return ret;
687}
688
689/*
690 * Same as grab_cache_page, but do not wait if the page is unavailable.
691 * This is intended for speculative data generators, where the data can
692 * be regenerated if the page couldn't be grabbed. This routine should
693 * be safe to call while holding the lock for another page.
694 *
695 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
696 * and deadlock against the caller's locked page.
697 */
698struct page *
699grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
700{
701 struct page *page = find_get_page(mapping, index);
Al Viro6daa0e22005-10-21 03:18:50 -0400702 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
704 if (page) {
705 if (!TestSetPageLocked(page))
706 return page;
707 page_cache_release(page);
708 return NULL;
709 }
710 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
711 page = alloc_pages(gfp_mask, 0);
712 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
713 page_cache_release(page);
714 page = NULL;
715 }
716 return page;
717}
718
719EXPORT_SYMBOL(grab_cache_page_nowait);
720
721/*
722 * This is a generic file read routine, and uses the
723 * mapping->a_ops->readpage() function for the actual low-level
724 * stuff.
725 *
726 * This is really ugly. But the goto's actually try to clarify some
727 * of the logic when it comes to error handling etc.
728 *
729 * Note the struct file* is only passed for the use of readpage. It may be
730 * NULL.
731 */
732void do_generic_mapping_read(struct address_space *mapping,
733 struct file_ra_state *_ra,
734 struct file *filp,
735 loff_t *ppos,
736 read_descriptor_t *desc,
737 read_actor_t actor)
738{
739 struct inode *inode = mapping->host;
740 unsigned long index;
741 unsigned long end_index;
742 unsigned long offset;
743 unsigned long last_index;
744 unsigned long next_index;
745 unsigned long prev_index;
746 loff_t isize;
747 struct page *cached_page;
748 int error;
749 struct file_ra_state ra = *_ra;
750
751 cached_page = NULL;
752 index = *ppos >> PAGE_CACHE_SHIFT;
753 next_index = index;
754 prev_index = ra.prev_page;
755 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
756 offset = *ppos & ~PAGE_CACHE_MASK;
757
758 isize = i_size_read(inode);
759 if (!isize)
760 goto out;
761
762 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
763 for (;;) {
764 struct page *page;
765 unsigned long nr, ret;
766
767 /* nr is the maximum number of bytes to copy from this page */
768 nr = PAGE_CACHE_SIZE;
769 if (index >= end_index) {
770 if (index > end_index)
771 goto out;
772 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
773 if (nr <= offset) {
774 goto out;
775 }
776 }
777 nr = nr - offset;
778
779 cond_resched();
780 if (index == next_index)
781 next_index = page_cache_readahead(mapping, &ra, filp,
782 index, last_index - index);
783
784find_page:
785 page = find_get_page(mapping, index);
786 if (unlikely(page == NULL)) {
787 handle_ra_miss(mapping, &ra, index);
788 goto no_cached_page;
789 }
790 if (!PageUptodate(page))
791 goto page_not_up_to_date;
792page_ok:
793
794 /* If users can be writing to this page using arbitrary
795 * virtual addresses, take care about potential aliasing
796 * before reading the page on the kernel side.
797 */
798 if (mapping_writably_mapped(mapping))
799 flush_dcache_page(page);
800
801 /*
802 * When (part of) the same page is read multiple times
803 * in succession, only mark it as accessed the first time.
804 */
805 if (prev_index != index)
806 mark_page_accessed(page);
807 prev_index = index;
808
809 /*
810 * Ok, we have the page, and it's up-to-date, so
811 * now we can copy it to user space...
812 *
813 * The actor routine returns how many bytes were actually used..
814 * NOTE! This may not be the same as how much of a user buffer
815 * we filled up (we may be padding etc), so we can only update
816 * "pos" here (the actor routine has to update the user buffer
817 * pointers and the remaining count).
818 */
819 ret = actor(desc, page, offset, nr);
820 offset += ret;
821 index += offset >> PAGE_CACHE_SHIFT;
822 offset &= ~PAGE_CACHE_MASK;
823
824 page_cache_release(page);
825 if (ret == nr && desc->count)
826 continue;
827 goto out;
828
829page_not_up_to_date:
830 /* Get exclusive access to the page ... */
831 lock_page(page);
832
833 /* Did it get unhashed before we got the lock? */
834 if (!page->mapping) {
835 unlock_page(page);
836 page_cache_release(page);
837 continue;
838 }
839
840 /* Did somebody else fill it already? */
841 if (PageUptodate(page)) {
842 unlock_page(page);
843 goto page_ok;
844 }
845
846readpage:
847 /* Start the actual read. The read will unlock the page. */
848 error = mapping->a_ops->readpage(filp, page);
849
Zach Brown994fc28c2005-12-15 14:28:17 -0800850 if (unlikely(error)) {
851 if (error == AOP_TRUNCATED_PAGE) {
852 page_cache_release(page);
853 goto find_page;
854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -0800856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 if (!PageUptodate(page)) {
859 lock_page(page);
860 if (!PageUptodate(page)) {
861 if (page->mapping == NULL) {
862 /*
863 * invalidate_inode_pages got it
864 */
865 unlock_page(page);
866 page_cache_release(page);
867 goto find_page;
868 }
869 unlock_page(page);
870 error = -EIO;
871 goto readpage_error;
872 }
873 unlock_page(page);
874 }
875
876 /*
877 * i_size must be checked after we have done ->readpage.
878 *
879 * Checking i_size after the readpage allows us to calculate
880 * the correct value for "nr", which means the zero-filled
881 * part of the page is not copied back to userspace (unless
882 * another truncate extends the file - this is desired though).
883 */
884 isize = i_size_read(inode);
885 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
886 if (unlikely(!isize || index > end_index)) {
887 page_cache_release(page);
888 goto out;
889 }
890
891 /* nr is the maximum number of bytes to copy from this page */
892 nr = PAGE_CACHE_SIZE;
893 if (index == end_index) {
894 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
895 if (nr <= offset) {
896 page_cache_release(page);
897 goto out;
898 }
899 }
900 nr = nr - offset;
901 goto page_ok;
902
903readpage_error:
904 /* UHHUH! A synchronous read error occurred. Report it */
905 desc->error = error;
906 page_cache_release(page);
907 goto out;
908
909no_cached_page:
910 /*
911 * Ok, it wasn't cached, so we need to create a new
912 * page..
913 */
914 if (!cached_page) {
915 cached_page = page_cache_alloc_cold(mapping);
916 if (!cached_page) {
917 desc->error = -ENOMEM;
918 goto out;
919 }
920 }
921 error = add_to_page_cache_lru(cached_page, mapping,
922 index, GFP_KERNEL);
923 if (error) {
924 if (error == -EEXIST)
925 goto find_page;
926 desc->error = error;
927 goto out;
928 }
929 page = cached_page;
930 cached_page = NULL;
931 goto readpage;
932 }
933
934out:
935 *_ra = ra;
936
937 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
938 if (cached_page)
939 page_cache_release(cached_page);
940 if (filp)
941 file_accessed(filp);
942}
943
944EXPORT_SYMBOL(do_generic_mapping_read);
945
946int file_read_actor(read_descriptor_t *desc, struct page *page,
947 unsigned long offset, unsigned long size)
948{
949 char *kaddr;
950 unsigned long left, count = desc->count;
951
952 if (size > count)
953 size = count;
954
955 /*
956 * Faults on the destination of a read are common, so do it before
957 * taking the kmap.
958 */
959 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
960 kaddr = kmap_atomic(page, KM_USER0);
961 left = __copy_to_user_inatomic(desc->arg.buf,
962 kaddr + offset, size);
963 kunmap_atomic(kaddr, KM_USER0);
964 if (left == 0)
965 goto success;
966 }
967
968 /* Do it the slow way */
969 kaddr = kmap(page);
970 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
971 kunmap(page);
972
973 if (left) {
974 size -= left;
975 desc->error = -EFAULT;
976 }
977success:
978 desc->count = count - size;
979 desc->written += size;
980 desc->arg.buf += size;
981 return size;
982}
983
984/*
985 * This is the "read()" routine for all filesystems
986 * that can use the page cache directly.
987 */
988ssize_t
989__generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
990 unsigned long nr_segs, loff_t *ppos)
991{
992 struct file *filp = iocb->ki_filp;
993 ssize_t retval;
994 unsigned long seg;
995 size_t count;
996
997 count = 0;
998 for (seg = 0; seg < nr_segs; seg++) {
999 const struct iovec *iv = &iov[seg];
1000
1001 /*
1002 * If any segment has a negative length, or the cumulative
1003 * length ever wraps negative then return -EINVAL.
1004 */
1005 count += iv->iov_len;
1006 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1007 return -EINVAL;
1008 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1009 continue;
1010 if (seg == 0)
1011 return -EFAULT;
1012 nr_segs = seg;
1013 count -= iv->iov_len; /* This segment is no good */
1014 break;
1015 }
1016
1017 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1018 if (filp->f_flags & O_DIRECT) {
1019 loff_t pos = *ppos, size;
1020 struct address_space *mapping;
1021 struct inode *inode;
1022
1023 mapping = filp->f_mapping;
1024 inode = mapping->host;
1025 retval = 0;
1026 if (!count)
1027 goto out; /* skip atime */
1028 size = i_size_read(inode);
1029 if (pos < size) {
1030 retval = generic_file_direct_IO(READ, iocb,
1031 iov, pos, nr_segs);
Suparna Bhattacharyab5c44c22005-05-21 16:33:36 -07001032 if (retval > 0 && !is_sync_kiocb(iocb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 retval = -EIOCBQUEUED;
1034 if (retval > 0)
1035 *ppos = pos + retval;
1036 }
1037 file_accessed(filp);
1038 goto out;
1039 }
1040
1041 retval = 0;
1042 if (count) {
1043 for (seg = 0; seg < nr_segs; seg++) {
1044 read_descriptor_t desc;
1045
1046 desc.written = 0;
1047 desc.arg.buf = iov[seg].iov_base;
1048 desc.count = iov[seg].iov_len;
1049 if (desc.count == 0)
1050 continue;
1051 desc.error = 0;
1052 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1053 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001054 if (desc.error) {
1055 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 break;
1057 }
1058 }
1059 }
1060out:
1061 return retval;
1062}
1063
1064EXPORT_SYMBOL(__generic_file_aio_read);
1065
1066ssize_t
1067generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1068{
1069 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1070
1071 BUG_ON(iocb->ki_pos != pos);
1072 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1073}
1074
1075EXPORT_SYMBOL(generic_file_aio_read);
1076
1077ssize_t
1078generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1079{
1080 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1081 struct kiocb kiocb;
1082 ssize_t ret;
1083
1084 init_sync_kiocb(&kiocb, filp);
1085 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1086 if (-EIOCBQUEUED == ret)
1087 ret = wait_on_sync_kiocb(&kiocb);
1088 return ret;
1089}
1090
1091EXPORT_SYMBOL(generic_file_read);
1092
1093int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1094{
1095 ssize_t written;
1096 unsigned long count = desc->count;
1097 struct file *file = desc->arg.data;
1098
1099 if (size > count)
1100 size = count;
1101
1102 written = file->f_op->sendpage(file, page, offset,
1103 size, &file->f_pos, size<count);
1104 if (written < 0) {
1105 desc->error = written;
1106 written = 0;
1107 }
1108 desc->count = count - written;
1109 desc->written += written;
1110 return written;
1111}
1112
1113ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1114 size_t count, read_actor_t actor, void *target)
1115{
1116 read_descriptor_t desc;
1117
1118 if (!count)
1119 return 0;
1120
1121 desc.written = 0;
1122 desc.count = count;
1123 desc.arg.data = target;
1124 desc.error = 0;
1125
1126 do_generic_file_read(in_file, ppos, &desc, actor);
1127 if (desc.written)
1128 return desc.written;
1129 return desc.error;
1130}
1131
1132EXPORT_SYMBOL(generic_file_sendfile);
1133
1134static ssize_t
1135do_readahead(struct address_space *mapping, struct file *filp,
1136 unsigned long index, unsigned long nr)
1137{
1138 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1139 return -EINVAL;
1140
1141 force_page_cache_readahead(mapping, filp, index,
1142 max_sane_readahead(nr));
1143 return 0;
1144}
1145
1146asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1147{
1148 ssize_t ret;
1149 struct file *file;
1150
1151 ret = -EBADF;
1152 file = fget(fd);
1153 if (file) {
1154 if (file->f_mode & FMODE_READ) {
1155 struct address_space *mapping = file->f_mapping;
1156 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1157 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1158 unsigned long len = end - start + 1;
1159 ret = do_readahead(mapping, file, start, len);
1160 }
1161 fput(file);
1162 }
1163 return ret;
1164}
1165
1166#ifdef CONFIG_MMU
1167/*
1168 * This adds the requested page to the page cache if it isn't already there,
1169 * and schedules an I/O to read in its contents from disk.
1170 */
1171static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1172static int fastcall page_cache_read(struct file * file, unsigned long offset)
1173{
1174 struct address_space *mapping = file->f_mapping;
1175 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001176 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Zach Brown994fc28c2005-12-15 14:28:17 -08001178 do {
1179 page = page_cache_alloc_cold(mapping);
1180 if (!page)
1181 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
Zach Brown994fc28c2005-12-15 14:28:17 -08001183 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1184 if (ret == 0)
1185 ret = mapping->a_ops->readpage(file, page);
1186 else if (ret == -EEXIST)
1187 ret = 0; /* losing race to add is OK */
1188
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Zach Brown994fc28c2005-12-15 14:28:17 -08001191 } while (ret == AOP_TRUNCATED_PAGE);
1192
1193 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194}
1195
1196#define MMAP_LOTSAMISS (100)
1197
1198/*
1199 * filemap_nopage() is invoked via the vma operations vector for a
1200 * mapped memory region to read in file data during a page fault.
1201 *
1202 * The goto's are kind of ugly, but this streamlines the normal case of having
1203 * it in the page cache, and handles the special cases reasonably without
1204 * having a lot of duplicated code.
1205 */
1206struct page *filemap_nopage(struct vm_area_struct *area,
1207 unsigned long address, int *type)
1208{
1209 int error;
1210 struct file *file = area->vm_file;
1211 struct address_space *mapping = file->f_mapping;
1212 struct file_ra_state *ra = &file->f_ra;
1213 struct inode *inode = mapping->host;
1214 struct page *page;
1215 unsigned long size, pgoff;
1216 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1217
1218 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1219
1220retry_all:
1221 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1222 if (pgoff >= size)
1223 goto outside_data_content;
1224
1225 /* If we don't want any read-ahead, don't bother */
1226 if (VM_RandomReadHint(area))
1227 goto no_cached_page;
1228
1229 /*
1230 * The readahead code wants to be told about each and every page
1231 * so it can build and shrink its windows appropriately
1232 *
1233 * For sequential accesses, we use the generic readahead logic.
1234 */
1235 if (VM_SequentialReadHint(area))
1236 page_cache_readahead(mapping, ra, file, pgoff, 1);
1237
1238 /*
1239 * Do we have something in the page cache already?
1240 */
1241retry_find:
1242 page = find_get_page(mapping, pgoff);
1243 if (!page) {
1244 unsigned long ra_pages;
1245
1246 if (VM_SequentialReadHint(area)) {
1247 handle_ra_miss(mapping, ra, pgoff);
1248 goto no_cached_page;
1249 }
1250 ra->mmap_miss++;
1251
1252 /*
1253 * Do we miss much more than hit in this file? If so,
1254 * stop bothering with read-ahead. It will only hurt.
1255 */
1256 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1257 goto no_cached_page;
1258
1259 /*
1260 * To keep the pgmajfault counter straight, we need to
1261 * check did_readaround, as this is an inner loop.
1262 */
1263 if (!did_readaround) {
1264 majmin = VM_FAULT_MAJOR;
1265 inc_page_state(pgmajfault);
1266 }
1267 did_readaround = 1;
1268 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1269 if (ra_pages) {
1270 pgoff_t start = 0;
1271
1272 if (pgoff > ra_pages / 2)
1273 start = pgoff - ra_pages / 2;
1274 do_page_cache_readahead(mapping, file, start, ra_pages);
1275 }
1276 page = find_get_page(mapping, pgoff);
1277 if (!page)
1278 goto no_cached_page;
1279 }
1280
1281 if (!did_readaround)
1282 ra->mmap_hit++;
1283
1284 /*
1285 * Ok, found a page in the page cache, now we need to check
1286 * that it's up-to-date.
1287 */
1288 if (!PageUptodate(page))
1289 goto page_not_uptodate;
1290
1291success:
1292 /*
1293 * Found the page and have a reference on it.
1294 */
1295 mark_page_accessed(page);
1296 if (type)
1297 *type = majmin;
1298 return page;
1299
1300outside_data_content:
1301 /*
1302 * An external ptracer can access pages that normally aren't
1303 * accessible..
1304 */
1305 if (area->vm_mm == current->mm)
1306 return NULL;
1307 /* Fall through to the non-read-ahead case */
1308no_cached_page:
1309 /*
1310 * We're only likely to ever get here if MADV_RANDOM is in
1311 * effect.
1312 */
1313 error = page_cache_read(file, pgoff);
1314 grab_swap_token();
1315
1316 /*
1317 * The page we want has now been added to the page cache.
1318 * In the unlikely event that someone removed it in the
1319 * meantime, we'll just come back here and read it again.
1320 */
1321 if (error >= 0)
1322 goto retry_find;
1323
1324 /*
1325 * An error return from page_cache_read can result if the
1326 * system is low on memory, or a problem occurs while trying
1327 * to schedule I/O.
1328 */
1329 if (error == -ENOMEM)
1330 return NOPAGE_OOM;
1331 return NULL;
1332
1333page_not_uptodate:
1334 if (!did_readaround) {
1335 majmin = VM_FAULT_MAJOR;
1336 inc_page_state(pgmajfault);
1337 }
1338 lock_page(page);
1339
1340 /* Did it get unhashed while we waited for it? */
1341 if (!page->mapping) {
1342 unlock_page(page);
1343 page_cache_release(page);
1344 goto retry_all;
1345 }
1346
1347 /* Did somebody else get it up-to-date? */
1348 if (PageUptodate(page)) {
1349 unlock_page(page);
1350 goto success;
1351 }
1352
Zach Brown994fc28c2005-12-15 14:28:17 -08001353 error = mapping->a_ops->readpage(file, page);
1354 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 wait_on_page_locked(page);
1356 if (PageUptodate(page))
1357 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001358 } else if (error == AOP_TRUNCATED_PAGE) {
1359 page_cache_release(page);
1360 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
1362
1363 /*
1364 * Umm, take care of errors if the page isn't up-to-date.
1365 * Try to re-read it _once_. We do this synchronously,
1366 * because there really aren't any performance issues here
1367 * and we need to check for errors.
1368 */
1369 lock_page(page);
1370
1371 /* Somebody truncated the page on us? */
1372 if (!page->mapping) {
1373 unlock_page(page);
1374 page_cache_release(page);
1375 goto retry_all;
1376 }
1377
1378 /* Somebody else successfully read it in? */
1379 if (PageUptodate(page)) {
1380 unlock_page(page);
1381 goto success;
1382 }
1383 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001384 error = mapping->a_ops->readpage(file, page);
1385 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 wait_on_page_locked(page);
1387 if (PageUptodate(page))
1388 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001389 } else if (error == AOP_TRUNCATED_PAGE) {
1390 page_cache_release(page);
1391 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 }
1393
1394 /*
1395 * Things didn't work out. Return zero to tell the
1396 * mm layer so, possibly freeing the page cache page first.
1397 */
1398 page_cache_release(page);
1399 return NULL;
1400}
1401
1402EXPORT_SYMBOL(filemap_nopage);
1403
1404static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1405 int nonblock)
1406{
1407 struct address_space *mapping = file->f_mapping;
1408 struct page *page;
1409 int error;
1410
1411 /*
1412 * Do we have something in the page cache already?
1413 */
1414retry_find:
1415 page = find_get_page(mapping, pgoff);
1416 if (!page) {
1417 if (nonblock)
1418 return NULL;
1419 goto no_cached_page;
1420 }
1421
1422 /*
1423 * Ok, found a page in the page cache, now we need to check
1424 * that it's up-to-date.
1425 */
Jeff Moyerd3457342005-04-16 15:24:05 -07001426 if (!PageUptodate(page)) {
1427 if (nonblock) {
1428 page_cache_release(page);
1429 return NULL;
1430 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 goto page_not_uptodate;
Jeff Moyerd3457342005-04-16 15:24:05 -07001432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
1434success:
1435 /*
1436 * Found the page and have a reference on it.
1437 */
1438 mark_page_accessed(page);
1439 return page;
1440
1441no_cached_page:
1442 error = page_cache_read(file, pgoff);
1443
1444 /*
1445 * The page we want has now been added to the page cache.
1446 * In the unlikely event that someone removed it in the
1447 * meantime, we'll just come back here and read it again.
1448 */
1449 if (error >= 0)
1450 goto retry_find;
1451
1452 /*
1453 * An error return from page_cache_read can result if the
1454 * system is low on memory, or a problem occurs while trying
1455 * to schedule I/O.
1456 */
1457 return NULL;
1458
1459page_not_uptodate:
1460 lock_page(page);
1461
1462 /* Did it get unhashed while we waited for it? */
1463 if (!page->mapping) {
1464 unlock_page(page);
1465 goto err;
1466 }
1467
1468 /* Did somebody else get it up-to-date? */
1469 if (PageUptodate(page)) {
1470 unlock_page(page);
1471 goto success;
1472 }
1473
Zach Brown994fc28c2005-12-15 14:28:17 -08001474 error = mapping->a_ops->readpage(file, page);
1475 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 wait_on_page_locked(page);
1477 if (PageUptodate(page))
1478 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001479 } else if (error == AOP_TRUNCATED_PAGE) {
1480 page_cache_release(page);
1481 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483
1484 /*
1485 * Umm, take care of errors if the page isn't up-to-date.
1486 * Try to re-read it _once_. We do this synchronously,
1487 * because there really aren't any performance issues here
1488 * and we need to check for errors.
1489 */
1490 lock_page(page);
1491
1492 /* Somebody truncated the page on us? */
1493 if (!page->mapping) {
1494 unlock_page(page);
1495 goto err;
1496 }
1497 /* Somebody else successfully read it in? */
1498 if (PageUptodate(page)) {
1499 unlock_page(page);
1500 goto success;
1501 }
1502
1503 ClearPageError(page);
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 * Things didn't work out. Return zero to tell the
1516 * mm layer so, possibly freeing the page cache page first.
1517 */
1518err:
1519 page_cache_release(page);
1520
1521 return NULL;
1522}
1523
1524int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1525 unsigned long len, pgprot_t prot, unsigned long pgoff,
1526 int nonblock)
1527{
1528 struct file *file = vma->vm_file;
1529 struct address_space *mapping = file->f_mapping;
1530 struct inode *inode = mapping->host;
1531 unsigned long size;
1532 struct mm_struct *mm = vma->vm_mm;
1533 struct page *page;
1534 int err;
1535
1536 if (!nonblock)
1537 force_page_cache_readahead(mapping, vma->vm_file,
1538 pgoff, len >> PAGE_CACHE_SHIFT);
1539
1540repeat:
1541 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1542 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1543 return -EINVAL;
1544
1545 page = filemap_getpage(file, pgoff, nonblock);
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001546
1547 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1548 * done in shmem_populate calling shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 if (!page && !nonblock)
1550 return -ENOMEM;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001551
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (page) {
1553 err = install_page(mm, vma, addr, page, prot);
1554 if (err) {
1555 page_cache_release(page);
1556 return err;
1557 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001558 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001559 /* No page was found just because we can't read it in now (being
1560 * here implies nonblock != 0), but the page may exist, so set
1561 * the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 err = install_file_pte(mm, vma, addr, pgoff, prot);
1563 if (err)
1564 return err;
1565 }
1566
1567 len -= PAGE_SIZE;
1568 addr += PAGE_SIZE;
1569 pgoff++;
1570 if (len)
1571 goto repeat;
1572
1573 return 0;
1574}
Nikita Danilovb1459462005-10-29 18:17:02 -07001575EXPORT_SYMBOL(filemap_populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
1577struct vm_operations_struct generic_file_vm_ops = {
1578 .nopage = filemap_nopage,
1579 .populate = filemap_populate,
1580};
1581
1582/* This is used for a general mmap of a disk file */
1583
1584int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1585{
1586 struct address_space *mapping = file->f_mapping;
1587
1588 if (!mapping->a_ops->readpage)
1589 return -ENOEXEC;
1590 file_accessed(file);
1591 vma->vm_ops = &generic_file_vm_ops;
1592 return 0;
1593}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595/*
1596 * This is for filesystems which do not implement ->writepage.
1597 */
1598int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1599{
1600 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1601 return -EINVAL;
1602 return generic_file_mmap(file, vma);
1603}
1604#else
1605int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1606{
1607 return -ENOSYS;
1608}
1609int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1610{
1611 return -ENOSYS;
1612}
1613#endif /* CONFIG_MMU */
1614
1615EXPORT_SYMBOL(generic_file_mmap);
1616EXPORT_SYMBOL(generic_file_readonly_mmap);
1617
1618static inline struct page *__read_cache_page(struct address_space *mapping,
1619 unsigned long index,
1620 int (*filler)(void *,struct page*),
1621 void *data)
1622{
1623 struct page *page, *cached_page = NULL;
1624 int err;
1625repeat:
1626 page = find_get_page(mapping, index);
1627 if (!page) {
1628 if (!cached_page) {
1629 cached_page = page_cache_alloc_cold(mapping);
1630 if (!cached_page)
1631 return ERR_PTR(-ENOMEM);
1632 }
1633 err = add_to_page_cache_lru(cached_page, mapping,
1634 index, GFP_KERNEL);
1635 if (err == -EEXIST)
1636 goto repeat;
1637 if (err < 0) {
1638 /* Presumably ENOMEM for radix tree node */
1639 page_cache_release(cached_page);
1640 return ERR_PTR(err);
1641 }
1642 page = cached_page;
1643 cached_page = NULL;
1644 err = filler(data, page);
1645 if (err < 0) {
1646 page_cache_release(page);
1647 page = ERR_PTR(err);
1648 }
1649 }
1650 if (cached_page)
1651 page_cache_release(cached_page);
1652 return page;
1653}
1654
1655/*
1656 * Read into the page cache. If a page already exists,
1657 * and PageUptodate() is not set, try to fill the page.
1658 */
1659struct page *read_cache_page(struct address_space *mapping,
1660 unsigned long index,
1661 int (*filler)(void *,struct page*),
1662 void *data)
1663{
1664 struct page *page;
1665 int err;
1666
1667retry:
1668 page = __read_cache_page(mapping, index, filler, data);
1669 if (IS_ERR(page))
1670 goto out;
1671 mark_page_accessed(page);
1672 if (PageUptodate(page))
1673 goto out;
1674
1675 lock_page(page);
1676 if (!page->mapping) {
1677 unlock_page(page);
1678 page_cache_release(page);
1679 goto retry;
1680 }
1681 if (PageUptodate(page)) {
1682 unlock_page(page);
1683 goto out;
1684 }
1685 err = filler(data, page);
1686 if (err < 0) {
1687 page_cache_release(page);
1688 page = ERR_PTR(err);
1689 }
1690 out:
1691 return page;
1692}
1693
1694EXPORT_SYMBOL(read_cache_page);
1695
1696/*
1697 * If the page was newly created, increment its refcount and add it to the
1698 * caller's lru-buffering pagevec. This function is specifically for
1699 * generic_file_write().
1700 */
1701static inline struct page *
1702__grab_cache_page(struct address_space *mapping, unsigned long index,
1703 struct page **cached_page, struct pagevec *lru_pvec)
1704{
1705 int err;
1706 struct page *page;
1707repeat:
1708 page = find_lock_page(mapping, index);
1709 if (!page) {
1710 if (!*cached_page) {
1711 *cached_page = page_cache_alloc(mapping);
1712 if (!*cached_page)
1713 return NULL;
1714 }
1715 err = add_to_page_cache(*cached_page, mapping,
1716 index, GFP_KERNEL);
1717 if (err == -EEXIST)
1718 goto repeat;
1719 if (err == 0) {
1720 page = *cached_page;
1721 page_cache_get(page);
1722 if (!pagevec_add(lru_pvec, page))
1723 __pagevec_lru_add(lru_pvec);
1724 *cached_page = NULL;
1725 }
1726 }
1727 return page;
1728}
1729
1730/*
1731 * The logic we want is
1732 *
1733 * if suid or (sgid and xgrp)
1734 * remove privs
1735 */
1736int remove_suid(struct dentry *dentry)
1737{
1738 mode_t mode = dentry->d_inode->i_mode;
1739 int kill = 0;
1740 int result = 0;
1741
1742 /* suid always must be killed */
1743 if (unlikely(mode & S_ISUID))
1744 kill = ATTR_KILL_SUID;
1745
1746 /*
1747 * sgid without any exec bits is just a mandatory locking mark; leave
1748 * it alone. If some exec bits are set, it's a real sgid; kill it.
1749 */
1750 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1751 kill |= ATTR_KILL_SGID;
1752
1753 if (unlikely(kill && !capable(CAP_FSETID))) {
1754 struct iattr newattrs;
1755
1756 newattrs.ia_valid = ATTR_FORCE | kill;
1757 result = notify_change(dentry, &newattrs);
1758 }
1759 return result;
1760}
1761EXPORT_SYMBOL(remove_suid);
1762
Carsten Otteceffc072005-06-23 22:05:25 -07001763size_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764__filemap_copy_from_user_iovec(char *vaddr,
1765 const struct iovec *iov, size_t base, size_t bytes)
1766{
1767 size_t copied = 0, left = 0;
1768
1769 while (bytes) {
1770 char __user *buf = iov->iov_base + base;
1771 int copy = min(bytes, iov->iov_len - base);
1772
1773 base = 0;
1774 left = __copy_from_user_inatomic(vaddr, buf, copy);
1775 copied += copy;
1776 bytes -= copy;
1777 vaddr += copy;
1778 iov++;
1779
1780 if (unlikely(left)) {
1781 /* zero the rest of the target like __copy_from_user */
1782 if (bytes)
1783 memset(vaddr, 0, bytes);
1784 break;
1785 }
1786 }
1787 return copied - left;
1788}
1789
1790/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 * Performs necessary checks before doing a write
1792 *
1793 * Can adjust writing position aor amount of bytes to write.
1794 * Returns appropriate error code that caller should return or
1795 * zero in case that write should be allowed.
1796 */
1797inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1798{
1799 struct inode *inode = file->f_mapping->host;
1800 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1801
1802 if (unlikely(*pos < 0))
1803 return -EINVAL;
1804
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 if (!isblk) {
1806 /* FIXME: this is for backwards compatibility with 2.4 */
1807 if (file->f_flags & O_APPEND)
1808 *pos = i_size_read(inode);
1809
1810 if (limit != RLIM_INFINITY) {
1811 if (*pos >= limit) {
1812 send_sig(SIGXFSZ, current, 0);
1813 return -EFBIG;
1814 }
1815 if (*count > limit - (typeof(limit))*pos) {
1816 *count = limit - (typeof(limit))*pos;
1817 }
1818 }
1819 }
1820
1821 /*
1822 * LFS rule
1823 */
1824 if (unlikely(*pos + *count > MAX_NON_LFS &&
1825 !(file->f_flags & O_LARGEFILE))) {
1826 if (*pos >= MAX_NON_LFS) {
1827 send_sig(SIGXFSZ, current, 0);
1828 return -EFBIG;
1829 }
1830 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1831 *count = MAX_NON_LFS - (unsigned long)*pos;
1832 }
1833 }
1834
1835 /*
1836 * Are we about to exceed the fs block limit ?
1837 *
1838 * If we have written data it becomes a short write. If we have
1839 * exceeded without writing data we send a signal and return EFBIG.
1840 * Linus frestrict idea will clean these up nicely..
1841 */
1842 if (likely(!isblk)) {
1843 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1844 if (*count || *pos > inode->i_sb->s_maxbytes) {
1845 send_sig(SIGXFSZ, current, 0);
1846 return -EFBIG;
1847 }
1848 /* zero-length writes at ->s_maxbytes are OK */
1849 }
1850
1851 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1852 *count = inode->i_sb->s_maxbytes - *pos;
1853 } else {
1854 loff_t isize;
1855 if (bdev_read_only(I_BDEV(inode)))
1856 return -EPERM;
1857 isize = i_size_read(inode);
1858 if (*pos >= isize) {
1859 if (*count || *pos > isize)
1860 return -ENOSPC;
1861 }
1862
1863 if (*pos + *count > isize)
1864 *count = isize - *pos;
1865 }
1866 return 0;
1867}
1868EXPORT_SYMBOL(generic_write_checks);
1869
1870ssize_t
1871generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1872 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1873 size_t count, size_t ocount)
1874{
1875 struct file *file = iocb->ki_filp;
1876 struct address_space *mapping = file->f_mapping;
1877 struct inode *inode = mapping->host;
1878 ssize_t written;
1879
1880 if (count != ocount)
1881 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1882
1883 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1884 if (written > 0) {
1885 loff_t end = pos + written;
1886 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1887 i_size_write(inode, end);
1888 mark_inode_dirty(inode);
1889 }
1890 *ppos = end;
1891 }
1892
1893 /*
1894 * Sync the fs metadata but not the minor inode changes and
1895 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001896 * i_mutex is held, which protects generic_osync_inode() from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 * livelocking.
1898 */
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07001899 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
1900 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1901 if (err < 0)
1902 written = err;
1903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 if (written == count && !is_sync_kiocb(iocb))
1905 written = -EIOCBQUEUED;
1906 return written;
1907}
1908EXPORT_SYMBOL(generic_file_direct_write);
1909
1910ssize_t
1911generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
1912 unsigned long nr_segs, loff_t pos, loff_t *ppos,
1913 size_t count, ssize_t written)
1914{
1915 struct file *file = iocb->ki_filp;
1916 struct address_space * mapping = file->f_mapping;
1917 struct address_space_operations *a_ops = mapping->a_ops;
1918 struct inode *inode = mapping->host;
1919 long status = 0;
1920 struct page *page;
1921 struct page *cached_page = NULL;
1922 size_t bytes;
1923 struct pagevec lru_pvec;
1924 const struct iovec *cur_iov = iov; /* current iovec */
1925 size_t iov_base = 0; /* offset in the current iovec */
1926 char __user *buf;
1927
1928 pagevec_init(&lru_pvec, 0);
1929
1930 /*
1931 * handle partial DIO write. Adjust cur_iov if needed.
1932 */
1933 if (likely(nr_segs == 1))
1934 buf = iov->iov_base + written;
1935 else {
1936 filemap_set_next_iovec(&cur_iov, &iov_base, written);
akpm@osdl.orgf021e922005-05-01 08:58:35 -07001937 buf = cur_iov->iov_base + iov_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 }
1939
1940 do {
1941 unsigned long index;
1942 unsigned long offset;
Martin Schwidefskya5117182005-06-06 13:35:54 -07001943 unsigned long maxlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 size_t copied;
1945
1946 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1947 index = pos >> PAGE_CACHE_SHIFT;
1948 bytes = PAGE_CACHE_SIZE - offset;
1949 if (bytes > count)
1950 bytes = count;
1951
1952 /*
1953 * Bring in the user page that we will copy from _first_.
1954 * Otherwise there's a nasty deadlock on copying from the
1955 * same page as we're writing to, without it being marked
1956 * up-to-date.
1957 */
Martin Schwidefskya5117182005-06-06 13:35:54 -07001958 maxlen = cur_iov->iov_len - iov_base;
1959 if (maxlen > bytes)
1960 maxlen = bytes;
1961 fault_in_pages_readable(buf, maxlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
1964 if (!page) {
1965 status = -ENOMEM;
1966 break;
1967 }
1968
1969 status = a_ops->prepare_write(file, page, offset, offset+bytes);
1970 if (unlikely(status)) {
1971 loff_t isize = i_size_read(inode);
Zach Brown994fc28c2005-12-15 14:28:17 -08001972
1973 if (status != AOP_TRUNCATED_PAGE)
1974 unlock_page(page);
1975 page_cache_release(page);
1976 if (status == AOP_TRUNCATED_PAGE)
1977 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 /*
1979 * prepare_write() may have instantiated a few blocks
1980 * outside i_size. Trim these off again.
1981 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 if (pos + bytes > isize)
1983 vmtruncate(inode, isize);
1984 break;
1985 }
1986 if (likely(nr_segs == 1))
1987 copied = filemap_copy_from_user(page, offset,
1988 buf, bytes);
1989 else
1990 copied = filemap_copy_from_user_iovec(page, offset,
1991 cur_iov, iov_base, bytes);
1992 flush_dcache_page(page);
1993 status = a_ops->commit_write(file, page, offset, offset+bytes);
Zach Brown994fc28c2005-12-15 14:28:17 -08001994 if (status == AOP_TRUNCATED_PAGE) {
1995 page_cache_release(page);
1996 continue;
1997 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 if (likely(copied > 0)) {
1999 if (!status)
2000 status = copied;
2001
2002 if (status >= 0) {
2003 written += status;
2004 count -= status;
2005 pos += status;
2006 buf += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002007 if (unlikely(nr_segs > 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 filemap_set_next_iovec(&cur_iov,
2009 &iov_base, status);
Badari Pulavartyb0cfbd92005-06-25 14:55:42 -07002010 if (count)
2011 buf = cur_iov->iov_base +
2012 iov_base;
Martin Schwidefskya5117182005-06-06 13:35:54 -07002013 } else {
2014 iov_base += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002015 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 }
2017 }
2018 if (unlikely(copied != bytes))
2019 if (status >= 0)
2020 status = -EFAULT;
2021 unlock_page(page);
2022 mark_page_accessed(page);
2023 page_cache_release(page);
2024 if (status < 0)
2025 break;
2026 balance_dirty_pages_ratelimited(mapping);
2027 cond_resched();
2028 } while (count);
2029 *ppos = pos;
2030
2031 if (cached_page)
2032 page_cache_release(cached_page);
2033
2034 /*
2035 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2036 */
2037 if (likely(status >= 0)) {
2038 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2039 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2040 status = generic_osync_inode(inode, mapping,
2041 OSYNC_METADATA|OSYNC_DATA);
2042 }
2043 }
2044
2045 /*
2046 * If we get here for O_DIRECT writes then we must have fallen through
2047 * to buffered writes (block instantiation inside i_size). So we sync
2048 * the file data here, to try to honour O_DIRECT expectations.
2049 */
2050 if (unlikely(file->f_flags & O_DIRECT) && written)
2051 status = filemap_write_and_wait(mapping);
2052
2053 pagevec_lru_add(&lru_pvec);
2054 return written ? written : status;
2055}
2056EXPORT_SYMBOL(generic_file_buffered_write);
2057
Adrian Bunk5ce78522005-09-10 00:26:28 -07002058static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2060 unsigned long nr_segs, loff_t *ppos)
2061{
2062 struct file *file = iocb->ki_filp;
2063 struct address_space * mapping = file->f_mapping;
2064 size_t ocount; /* original count */
2065 size_t count; /* after file limit checks */
2066 struct inode *inode = mapping->host;
2067 unsigned long seg;
2068 loff_t pos;
2069 ssize_t written;
2070 ssize_t err;
2071
2072 ocount = 0;
2073 for (seg = 0; seg < nr_segs; seg++) {
2074 const struct iovec *iv = &iov[seg];
2075
2076 /*
2077 * If any segment has a negative length, or the cumulative
2078 * length ever wraps negative then return -EINVAL.
2079 */
2080 ocount += iv->iov_len;
2081 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2082 return -EINVAL;
2083 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2084 continue;
2085 if (seg == 0)
2086 return -EFAULT;
2087 nr_segs = seg;
2088 ocount -= iv->iov_len; /* This segment is no good */
2089 break;
2090 }
2091
2092 count = ocount;
2093 pos = *ppos;
2094
2095 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2096
2097 /* We can write back this queue in page reclaim */
2098 current->backing_dev_info = mapping->backing_dev_info;
2099 written = 0;
2100
2101 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2102 if (err)
2103 goto out;
2104
2105 if (count == 0)
2106 goto out;
2107
2108 err = remove_suid(file->f_dentry);
2109 if (err)
2110 goto out;
2111
Christoph Hellwig870f4812006-01-09 20:52:01 -08002112 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2115 if (unlikely(file->f_flags & O_DIRECT)) {
2116 written = generic_file_direct_write(iocb, iov,
2117 &nr_segs, pos, ppos, count, ocount);
2118 if (written < 0 || written == count)
2119 goto out;
2120 /*
2121 * direct-io write to a hole: fall through to buffered I/O
2122 * for completing the rest of the request.
2123 */
2124 pos += written;
2125 count -= written;
2126 }
2127
2128 written = generic_file_buffered_write(iocb, iov, nr_segs,
2129 pos, ppos, count, written);
2130out:
2131 current->backing_dev_info = NULL;
2132 return written ? written : err;
2133}
2134EXPORT_SYMBOL(generic_file_aio_write_nolock);
2135
2136ssize_t
2137generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2138 unsigned long nr_segs, loff_t *ppos)
2139{
2140 struct file *file = iocb->ki_filp;
2141 struct address_space *mapping = file->f_mapping;
2142 struct inode *inode = mapping->host;
2143 ssize_t ret;
2144 loff_t pos = *ppos;
2145
2146 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2147
2148 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2149 int err;
2150
2151 err = sync_page_range_nolock(inode, mapping, pos, ret);
2152 if (err < 0)
2153 ret = err;
2154 }
2155 return ret;
2156}
2157
Adrian Bunk5ce78522005-09-10 00:26:28 -07002158static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159__generic_file_write_nolock(struct file *file, const struct iovec *iov,
2160 unsigned long nr_segs, loff_t *ppos)
2161{
2162 struct kiocb kiocb;
2163 ssize_t ret;
2164
2165 init_sync_kiocb(&kiocb, file);
2166 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2167 if (ret == -EIOCBQUEUED)
2168 ret = wait_on_sync_kiocb(&kiocb);
2169 return ret;
2170}
2171
2172ssize_t
2173generic_file_write_nolock(struct file *file, const struct iovec *iov,
2174 unsigned long nr_segs, loff_t *ppos)
2175{
2176 struct kiocb kiocb;
2177 ssize_t ret;
2178
2179 init_sync_kiocb(&kiocb, file);
2180 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2181 if (-EIOCBQUEUED == ret)
2182 ret = wait_on_sync_kiocb(&kiocb);
2183 return ret;
2184}
2185EXPORT_SYMBOL(generic_file_write_nolock);
2186
2187ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2188 size_t count, loff_t pos)
2189{
2190 struct file *file = iocb->ki_filp;
2191 struct address_space *mapping = file->f_mapping;
2192 struct inode *inode = mapping->host;
2193 ssize_t ret;
2194 struct iovec local_iov = { .iov_base = (void __user *)buf,
2195 .iov_len = count };
2196
2197 BUG_ON(iocb->ki_pos != pos);
2198
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002199 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2201 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002202 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
2204 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2205 ssize_t err;
2206
2207 err = sync_page_range(inode, mapping, pos, ret);
2208 if (err < 0)
2209 ret = err;
2210 }
2211 return ret;
2212}
2213EXPORT_SYMBOL(generic_file_aio_write);
2214
2215ssize_t generic_file_write(struct file *file, const char __user *buf,
2216 size_t count, loff_t *ppos)
2217{
2218 struct address_space *mapping = file->f_mapping;
2219 struct inode *inode = mapping->host;
2220 ssize_t ret;
2221 struct iovec local_iov = { .iov_base = (void __user *)buf,
2222 .iov_len = count };
2223
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002224 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002226 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
2228 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2229 ssize_t err;
2230
2231 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2232 if (err < 0)
2233 ret = err;
2234 }
2235 return ret;
2236}
2237EXPORT_SYMBOL(generic_file_write);
2238
2239ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2240 unsigned long nr_segs, loff_t *ppos)
2241{
2242 struct kiocb kiocb;
2243 ssize_t ret;
2244
2245 init_sync_kiocb(&kiocb, filp);
2246 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2247 if (-EIOCBQUEUED == ret)
2248 ret = wait_on_sync_kiocb(&kiocb);
2249 return ret;
2250}
2251EXPORT_SYMBOL(generic_file_readv);
2252
2253ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2254 unsigned long nr_segs, loff_t *ppos)
2255{
2256 struct address_space *mapping = file->f_mapping;
2257 struct inode *inode = mapping->host;
2258 ssize_t ret;
2259
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002260 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002262 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263
2264 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2265 int err;
2266
2267 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2268 if (err < 0)
2269 ret = err;
2270 }
2271 return ret;
2272}
2273EXPORT_SYMBOL(generic_file_writev);
2274
2275/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002276 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 * went wrong during pagecache shootdown.
2278 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002279static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2281 loff_t offset, unsigned long nr_segs)
2282{
2283 struct file *file = iocb->ki_filp;
2284 struct address_space *mapping = file->f_mapping;
2285 ssize_t retval;
2286 size_t write_len = 0;
2287
2288 /*
2289 * If it's a write, unmap all mmappings of the file up-front. This
2290 * will cause any pte dirty bits to be propagated into the pageframes
2291 * for the subsequent filemap_write_and_wait().
2292 */
2293 if (rw == WRITE) {
2294 write_len = iov_length(iov, nr_segs);
2295 if (mapping_mapped(mapping))
2296 unmap_mapping_range(mapping, offset, write_len, 0);
2297 }
2298
2299 retval = filemap_write_and_wait(mapping);
2300 if (retval == 0) {
2301 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2302 offset, nr_segs);
2303 if (rw == WRITE && mapping->nrpages) {
2304 pgoff_t end = (offset + write_len - 1)
2305 >> PAGE_CACHE_SHIFT;
2306 int err = invalidate_inode_pages2_range(mapping,
2307 offset >> PAGE_CACHE_SHIFT, end);
2308 if (err)
2309 retval = err;
2310 }
2311 }
2312 return retval;
2313}