blob: fd57442186cbec1b55654639415af850ba352a2f [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>
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
41#include <asm/uaccess.h>
42#include <asm/mman.h>
43
Adrian Bunk5ce78522005-09-10 00:26:28 -070044static ssize_t
45generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
46 loff_t offset, unsigned long nr_segs);
47
Linus Torvalds1da177e2005-04-16 15:20:36 -070048/*
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
50 * though.
51 *
52 * Shared mappings now work. 15.8.1995 Bruno.
53 *
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 *
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
58 */
59
60/*
61 * Lock ordering:
62 *
63 * ->i_mmap_lock (vmtruncate)
64 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070065 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080068 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * ->i_mmap_lock (truncate->unmap_mapping_range)
70 *
71 * ->mmap_sem
72 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070073 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
75 *
76 * ->mmap_sem
77 * ->lock_page (access_process_vm)
78 *
79 * ->mmap_sem
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080080 * ->i_mutex (msync)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080082 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 * ->i_alloc_sem (various)
84 *
85 * ->inode_lock
86 * ->sb_lock (fs/fs-writeback.c)
87 * ->mapping->tree_lock (__sync_single_inode)
88 *
89 * ->i_mmap_lock
90 * ->anon_vma.lock (vma_adjust)
91 *
92 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070095 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070096 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * ->private_lock (try_to_unmap_one)
98 * ->tree_lock (try_to_unmap_one)
99 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -0800100 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 * ->private_lock (page_remove_rmap->set_page_dirty)
102 * ->tree_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (page_remove_rmap->set_page_dirty)
104 * ->inode_lock (zap_pte_range->set_page_dirty)
105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
106 *
107 * ->task->proc_lock
108 * ->dcache_lock (proc_pid_lookup)
109 */
110
111/*
112 * Remove a page from the page cache and free it. Caller has to make
113 * sure the page is locked and that nobody else uses it - or that usage
114 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
115 */
116void __remove_from_page_cache(struct page *page)
117{
118 struct address_space *mapping = page->mapping;
119
120 radix_tree_delete(&mapping->page_tree, page->index);
121 page->mapping = NULL;
122 mapping->nrpages--;
123 pagecache_acct(-1);
124}
125
126void remove_from_page_cache(struct page *page)
127{
128 struct address_space *mapping = page->mapping;
129
Matt Mackallcd7619d2005-05-01 08:59:01 -0700130 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 write_lock_irq(&mapping->tree_lock);
133 __remove_from_page_cache(page);
134 write_unlock_irq(&mapping->tree_lock);
135}
136
137static int sync_page(void *word)
138{
139 struct address_space *mapping;
140 struct page *page;
141
Andi Kleen07808b72005-11-05 17:25:53 +0100142 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
144 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700145 * page_mapping() is being called without PG_locked held.
146 * Some knowledge of the state and use of the page is used to
147 * reduce the requirements down to a memory barrier.
148 * The danger here is of a stale page_mapping() return value
149 * indicating a struct address_space different from the one it's
150 * associated with when it is associated with one.
151 * After smp_mb(), it's either the correct page_mapping() for
152 * the page, or an old page_mapping() and the page's own
153 * page_mapping() has gone NULL.
154 * The ->sync_page() address_space operation must tolerate
155 * page_mapping() going NULL. By an amazing coincidence,
156 * this comes about because none of the users of the page
157 * in the ->sync_page() methods make essential use of the
158 * page_mapping(), merely passing the page down to the backing
159 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700160 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700161 * of interest. When page_mapping() does go NULL, the entire
162 * call stack gracefully ignores the page and returns.
163 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 */
165 smp_mb();
166 mapping = page_mapping(page);
167 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
168 mapping->a_ops->sync_page(page);
169 io_schedule();
170 return 0;
171}
172
173/**
174 * filemap_fdatawrite_range - start writeback against all of a mapping's
175 * dirty pages that lie within the byte offsets <start, end>
Martin Waitz67be2dd2005-05-01 08:59:26 -0700176 * @mapping: address space structure to write
177 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800178 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700179 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 *
181 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
182 * opposed to a regular memory * cleansing writeback. The difference between
183 * these two operations is that if a dirty page/buffer is encountered, it must
184 * be waited upon, and not just skipped over.
185 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800186int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
187 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
189 int ret;
190 struct writeback_control wbc = {
191 .sync_mode = sync_mode,
192 .nr_to_write = mapping->nrpages * 2,
193 .start = start,
194 .end = end,
195 };
196
197 if (!mapping_cap_writeback_dirty(mapping))
198 return 0;
199
200 ret = do_writepages(mapping, &wbc);
201 return ret;
202}
203
204static inline int __filemap_fdatawrite(struct address_space *mapping,
205 int sync_mode)
206{
207 return __filemap_fdatawrite_range(mapping, 0, 0, sync_mode);
208}
209
210int filemap_fdatawrite(struct address_space *mapping)
211{
212 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
213}
214EXPORT_SYMBOL(filemap_fdatawrite);
215
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800216static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
217 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
219 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
220}
221
222/*
223 * This is a mostly non-blocking flush. Not suitable for data-integrity
224 * purposes - I/O may not be started against all dirty pages.
225 */
226int filemap_flush(struct address_space *mapping)
227{
228 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
229}
230EXPORT_SYMBOL(filemap_flush);
231
232/*
233 * Wait for writeback to complete against pages indexed by start->end
234 * inclusive
235 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800236int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 pgoff_t start, pgoff_t end)
238{
239 struct pagevec pvec;
240 int nr_pages;
241 int ret = 0;
242 pgoff_t index;
243
244 if (end < start)
245 return 0;
246
247 pagevec_init(&pvec, 0);
248 index = start;
249 while ((index <= end) &&
250 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
251 PAGECACHE_TAG_WRITEBACK,
252 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
253 unsigned i;
254
255 for (i = 0; i < nr_pages; i++) {
256 struct page *page = pvec.pages[i];
257
258 /* until radix tree lookup accepts end_index */
259 if (page->index > end)
260 continue;
261
262 wait_on_page_writeback(page);
263 if (PageError(page))
264 ret = -EIO;
265 }
266 pagevec_release(&pvec);
267 cond_resched();
268 }
269
270 /* Check for outstanding write errors */
271 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
272 ret = -ENOSPC;
273 if (test_and_clear_bit(AS_EIO, &mapping->flags))
274 ret = -EIO;
275
276 return ret;
277}
278
279/*
280 * Write and wait upon all the pages in the passed range. This is a "data
281 * integrity" operation. It waits upon in-flight writeout before starting and
282 * waiting upon new writeout. If there was an IO error, return it.
283 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800284 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 * it is otherwise livelockable.
286 */
287int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800288 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
290 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
291 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
292 int ret;
293
294 if (!mapping_cap_writeback_dirty(mapping) || !count)
295 return 0;
296 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
297 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800298 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800300 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 }
302 if (ret == 0)
303 ret = wait_on_page_writeback_range(mapping, start, end);
304 return ret;
305}
306EXPORT_SYMBOL(sync_page_range);
307
308/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800309 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * as it forces O_SYNC writers to different parts of the same file
311 * to be serialised right until io completion.
312 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800313int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
314 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
316 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
317 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
318 int ret;
319
320 if (!mapping_cap_writeback_dirty(mapping) || !count)
321 return 0;
322 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
323 if (ret == 0)
324 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
325 if (ret == 0)
326 ret = wait_on_page_writeback_range(mapping, start, end);
327 return ret;
328}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800329EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331/**
332 * filemap_fdatawait - walk the list of under-writeback pages of the given
333 * address space and wait for all of them.
334 *
335 * @mapping: address space structure to wait for
336 */
337int filemap_fdatawait(struct address_space *mapping)
338{
339 loff_t i_size = i_size_read(mapping->host);
340
341 if (i_size == 0)
342 return 0;
343
344 return wait_on_page_writeback_range(mapping, 0,
345 (i_size - 1) >> PAGE_CACHE_SHIFT);
346}
347EXPORT_SYMBOL(filemap_fdatawait);
348
349int filemap_write_and_wait(struct address_space *mapping)
350{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800351 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800354 err = filemap_fdatawrite(mapping);
355 /*
356 * Even if the above returned error, the pages may be
357 * written partially (e.g. -ENOSPC), so we wait for it.
358 * But the -EIO is special case, it may indicate the worst
359 * thing (e.g. bug) happened, so we avoid waiting for it.
360 */
361 if (err != -EIO) {
362 int err2 = filemap_fdatawait(mapping);
363 if (!err)
364 err = err2;
365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800367 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800369EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Andrew Morton469eb4d2006-03-24 03:17:45 -0800371/*
372 * Write out and wait upon file offsets lstart->lend, inclusive.
373 *
374 * Note that `lend' is inclusive (describes the last byte to be written) so
375 * that this function can be used to write to the very end-of-file (end = -1).
376 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377int filemap_write_and_wait_range(struct address_space *mapping,
378 loff_t lstart, loff_t lend)
379{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800380 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
382 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800383 err = __filemap_fdatawrite_range(mapping, lstart, lend,
384 WB_SYNC_ALL);
385 /* See comment of filemap_write_and_wait() */
386 if (err != -EIO) {
387 int err2 = wait_on_page_writeback_range(mapping,
388 lstart >> PAGE_CACHE_SHIFT,
389 lend >> PAGE_CACHE_SHIFT);
390 if (!err)
391 err = err2;
392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800394 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
397/*
398 * This function is used to add newly allocated pagecache pages:
399 * the page is new, so we can just run SetPageLocked() against it.
400 * The other page state flags were set by rmqueue().
401 *
402 * This function does not add the page to the LRU. The caller must do that.
403 */
404int add_to_page_cache(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400405 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
407 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
408
409 if (error == 0) {
410 write_lock_irq(&mapping->tree_lock);
411 error = radix_tree_insert(&mapping->page_tree, offset, page);
412 if (!error) {
413 page_cache_get(page);
414 SetPageLocked(page);
415 page->mapping = mapping;
416 page->index = offset;
417 mapping->nrpages++;
418 pagecache_acct(1);
419 }
420 write_unlock_irq(&mapping->tree_lock);
421 radix_tree_preload_end();
422 }
423 return error;
424}
425
426EXPORT_SYMBOL(add_to_page_cache);
427
428int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400429 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
431 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
432 if (ret == 0)
433 lru_cache_add(page);
434 return ret;
435}
436
Paul Jackson44110fe2006-03-24 03:16:04 -0800437#ifdef CONFIG_NUMA
438struct page *page_cache_alloc(struct address_space *x)
439{
440 if (cpuset_do_page_mem_spread()) {
441 int n = cpuset_mem_spread_node();
442 return alloc_pages_node(n, mapping_gfp_mask(x), 0);
443 }
444 return alloc_pages(mapping_gfp_mask(x), 0);
445}
446EXPORT_SYMBOL(page_cache_alloc);
447
448struct page *page_cache_alloc_cold(struct address_space *x)
449{
450 if (cpuset_do_page_mem_spread()) {
451 int n = cpuset_mem_spread_node();
452 return alloc_pages_node(n, mapping_gfp_mask(x)|__GFP_COLD, 0);
453 }
454 return alloc_pages(mapping_gfp_mask(x)|__GFP_COLD, 0);
455}
456EXPORT_SYMBOL(page_cache_alloc_cold);
457#endif
458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459/*
460 * In order to wait for pages to become available there must be
461 * waitqueues associated with pages. By using a hash table of
462 * waitqueues where the bucket discipline is to maintain all
463 * waiters on the same queue and wake all when any of the pages
464 * become available, and for the woken contexts to check to be
465 * sure the appropriate page became available, this saves space
466 * at a cost of "thundering herd" phenomena during rare hash
467 * collisions.
468 */
469static wait_queue_head_t *page_waitqueue(struct page *page)
470{
471 const struct zone *zone = page_zone(page);
472
473 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
474}
475
476static inline void wake_up_page(struct page *page, int bit)
477{
478 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
479}
480
481void fastcall wait_on_page_bit(struct page *page, int bit_nr)
482{
483 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
484
485 if (test_bit(bit_nr, &page->flags))
486 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
487 TASK_UNINTERRUPTIBLE);
488}
489EXPORT_SYMBOL(wait_on_page_bit);
490
491/**
492 * unlock_page() - unlock a locked page
493 *
494 * @page: the page
495 *
496 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
497 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
498 * mechananism between PageLocked pages and PageWriteback pages is shared.
499 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
500 *
501 * The first mb is necessary to safely close the critical section opened by the
502 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
503 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
504 * parallel wait_on_page_locked()).
505 */
506void fastcall unlock_page(struct page *page)
507{
508 smp_mb__before_clear_bit();
509 if (!TestClearPageLocked(page))
510 BUG();
511 smp_mb__after_clear_bit();
512 wake_up_page(page, PG_locked);
513}
514EXPORT_SYMBOL(unlock_page);
515
516/*
517 * End writeback against a page.
518 */
519void end_page_writeback(struct page *page)
520{
521 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
522 if (!test_clear_page_writeback(page))
523 BUG();
524 }
525 smp_mb__after_clear_bit();
526 wake_up_page(page, PG_writeback);
527}
528EXPORT_SYMBOL(end_page_writeback);
529
530/*
531 * Get a lock on the page, assuming we need to sleep to get it.
532 *
533 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
534 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
535 * chances are that on the second loop, the block layer's plug list is empty,
536 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
537 */
538void fastcall __lock_page(struct page *page)
539{
540 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
541
542 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
543 TASK_UNINTERRUPTIBLE);
544}
545EXPORT_SYMBOL(__lock_page);
546
547/*
548 * a rather lightweight function, finding and getting a reference to a
549 * hashed page atomically.
550 */
551struct page * find_get_page(struct address_space *mapping, unsigned long offset)
552{
553 struct page *page;
554
555 read_lock_irq(&mapping->tree_lock);
556 page = radix_tree_lookup(&mapping->page_tree, offset);
557 if (page)
558 page_cache_get(page);
559 read_unlock_irq(&mapping->tree_lock);
560 return page;
561}
562
563EXPORT_SYMBOL(find_get_page);
564
565/*
566 * Same as above, but trylock it instead of incrementing the count.
567 */
568struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
569{
570 struct page *page;
571
572 read_lock_irq(&mapping->tree_lock);
573 page = radix_tree_lookup(&mapping->page_tree, offset);
574 if (page && TestSetPageLocked(page))
575 page = NULL;
576 read_unlock_irq(&mapping->tree_lock);
577 return page;
578}
579
580EXPORT_SYMBOL(find_trylock_page);
581
582/**
583 * find_lock_page - locate, pin and lock a pagecache page
584 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700585 * @mapping: the address_space to search
586 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 *
588 * Locates the desired pagecache page, locks it, increments its reference
589 * count and returns its address.
590 *
591 * Returns zero if the page was not present. find_lock_page() may sleep.
592 */
593struct page *find_lock_page(struct address_space *mapping,
594 unsigned long offset)
595{
596 struct page *page;
597
598 read_lock_irq(&mapping->tree_lock);
599repeat:
600 page = radix_tree_lookup(&mapping->page_tree, offset);
601 if (page) {
602 page_cache_get(page);
603 if (TestSetPageLocked(page)) {
604 read_unlock_irq(&mapping->tree_lock);
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800605 __lock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 read_lock_irq(&mapping->tree_lock);
607
608 /* Has the page been truncated while we slept? */
Nikita Danilovbbfbb7c2006-01-06 00:11:08 -0800609 if (unlikely(page->mapping != mapping ||
610 page->index != offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 unlock_page(page);
612 page_cache_release(page);
613 goto repeat;
614 }
615 }
616 }
617 read_unlock_irq(&mapping->tree_lock);
618 return page;
619}
620
621EXPORT_SYMBOL(find_lock_page);
622
623/**
624 * find_or_create_page - locate or add a pagecache page
625 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700626 * @mapping: the page's address_space
627 * @index: the page's index into the mapping
628 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 *
630 * Locates a page in the pagecache. If the page is not present, a new page
631 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
632 * LRU list. The returned page is locked and has its reference count
633 * incremented.
634 *
635 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
636 * allocation!
637 *
638 * find_or_create_page() returns the desired page's address, or zero on
639 * memory exhaustion.
640 */
641struct page *find_or_create_page(struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400642 unsigned long index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
644 struct page *page, *cached_page = NULL;
645 int err;
646repeat:
647 page = find_lock_page(mapping, index);
648 if (!page) {
649 if (!cached_page) {
650 cached_page = alloc_page(gfp_mask);
651 if (!cached_page)
652 return NULL;
653 }
654 err = add_to_page_cache_lru(cached_page, mapping,
655 index, gfp_mask);
656 if (!err) {
657 page = cached_page;
658 cached_page = NULL;
659 } else if (err == -EEXIST)
660 goto repeat;
661 }
662 if (cached_page)
663 page_cache_release(cached_page);
664 return page;
665}
666
667EXPORT_SYMBOL(find_or_create_page);
668
669/**
670 * find_get_pages - gang pagecache lookup
671 * @mapping: The address_space to search
672 * @start: The starting page index
673 * @nr_pages: The maximum number of pages
674 * @pages: Where the resulting pages are placed
675 *
676 * find_get_pages() will search for and return a group of up to
677 * @nr_pages pages in the mapping. The pages are placed at @pages.
678 * find_get_pages() takes a reference against the returned pages.
679 *
680 * The search returns a group of mapping-contiguous pages with ascending
681 * indexes. There may be holes in the indices due to not-present pages.
682 *
683 * find_get_pages() returns the number of pages which were found.
684 */
685unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
686 unsigned int nr_pages, struct page **pages)
687{
688 unsigned int i;
689 unsigned int ret;
690
691 read_lock_irq(&mapping->tree_lock);
692 ret = radix_tree_gang_lookup(&mapping->page_tree,
693 (void **)pages, start, nr_pages);
694 for (i = 0; i < ret; i++)
695 page_cache_get(pages[i]);
696 read_unlock_irq(&mapping->tree_lock);
697 return ret;
698}
699
Jens Axboeebf43502006-04-27 08:46:01 +0200700/**
701 * find_get_pages_contig - gang contiguous pagecache lookup
702 * @mapping: The address_space to search
703 * @index: The starting page index
704 * @nr_pages: The maximum number of pages
705 * @pages: Where the resulting pages are placed
706 *
707 * find_get_pages_contig() works exactly like find_get_pages(), except
708 * that the returned number of pages are guaranteed to be contiguous.
709 *
710 * find_get_pages_contig() returns the number of pages which were found.
711 */
712unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
713 unsigned int nr_pages, struct page **pages)
714{
715 unsigned int i;
716 unsigned int ret;
717
718 read_lock_irq(&mapping->tree_lock);
719 ret = radix_tree_gang_lookup(&mapping->page_tree,
720 (void **)pages, index, nr_pages);
721 for (i = 0; i < ret; i++) {
722 if (pages[i]->mapping == NULL || pages[i]->index != index)
723 break;
724
725 page_cache_get(pages[i]);
726 index++;
727 }
728 read_unlock_irq(&mapping->tree_lock);
729 return i;
730}
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/*
733 * Like find_get_pages, except we only return pages which are tagged with
734 * `tag'. We update *index to index the next page for the traversal.
735 */
736unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
737 int tag, unsigned int nr_pages, struct page **pages)
738{
739 unsigned int i;
740 unsigned int ret;
741
742 read_lock_irq(&mapping->tree_lock);
743 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
744 (void **)pages, *index, nr_pages, tag);
745 for (i = 0; i < ret; i++)
746 page_cache_get(pages[i]);
747 if (ret)
748 *index = pages[ret - 1]->index + 1;
749 read_unlock_irq(&mapping->tree_lock);
750 return ret;
751}
752
753/*
754 * Same as grab_cache_page, but do not wait if the page is unavailable.
755 * This is intended for speculative data generators, where the data can
756 * be regenerated if the page couldn't be grabbed. This routine should
757 * be safe to call while holding the lock for another page.
758 *
759 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
760 * and deadlock against the caller's locked page.
761 */
762struct page *
763grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
764{
765 struct page *page = find_get_page(mapping, index);
Al Viro6daa0e22005-10-21 03:18:50 -0400766 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 if (page) {
769 if (!TestSetPageLocked(page))
770 return page;
771 page_cache_release(page);
772 return NULL;
773 }
774 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
775 page = alloc_pages(gfp_mask, 0);
776 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
777 page_cache_release(page);
778 page = NULL;
779 }
780 return page;
781}
782
783EXPORT_SYMBOL(grab_cache_page_nowait);
784
785/*
786 * This is a generic file read routine, and uses the
787 * mapping->a_ops->readpage() function for the actual low-level
788 * stuff.
789 *
790 * This is really ugly. But the goto's actually try to clarify some
791 * of the logic when it comes to error handling etc.
792 *
793 * Note the struct file* is only passed for the use of readpage. It may be
794 * NULL.
795 */
796void do_generic_mapping_read(struct address_space *mapping,
797 struct file_ra_state *_ra,
798 struct file *filp,
799 loff_t *ppos,
800 read_descriptor_t *desc,
801 read_actor_t actor)
802{
803 struct inode *inode = mapping->host;
804 unsigned long index;
805 unsigned long end_index;
806 unsigned long offset;
807 unsigned long last_index;
808 unsigned long next_index;
809 unsigned long prev_index;
810 loff_t isize;
811 struct page *cached_page;
812 int error;
813 struct file_ra_state ra = *_ra;
814
815 cached_page = NULL;
816 index = *ppos >> PAGE_CACHE_SHIFT;
817 next_index = index;
818 prev_index = ra.prev_page;
819 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
820 offset = *ppos & ~PAGE_CACHE_MASK;
821
822 isize = i_size_read(inode);
823 if (!isize)
824 goto out;
825
826 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
827 for (;;) {
828 struct page *page;
829 unsigned long nr, ret;
830
831 /* nr is the maximum number of bytes to copy from this page */
832 nr = PAGE_CACHE_SIZE;
833 if (index >= end_index) {
834 if (index > end_index)
835 goto out;
836 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
837 if (nr <= offset) {
838 goto out;
839 }
840 }
841 nr = nr - offset;
842
843 cond_resched();
844 if (index == next_index)
845 next_index = page_cache_readahead(mapping, &ra, filp,
846 index, last_index - index);
847
848find_page:
849 page = find_get_page(mapping, index);
850 if (unlikely(page == NULL)) {
851 handle_ra_miss(mapping, &ra, index);
852 goto no_cached_page;
853 }
854 if (!PageUptodate(page))
855 goto page_not_up_to_date;
856page_ok:
857
858 /* If users can be writing to this page using arbitrary
859 * virtual addresses, take care about potential aliasing
860 * before reading the page on the kernel side.
861 */
862 if (mapping_writably_mapped(mapping))
863 flush_dcache_page(page);
864
865 /*
866 * When (part of) the same page is read multiple times
867 * in succession, only mark it as accessed the first time.
868 */
869 if (prev_index != index)
870 mark_page_accessed(page);
871 prev_index = index;
872
873 /*
874 * Ok, we have the page, and it's up-to-date, so
875 * now we can copy it to user space...
876 *
877 * The actor routine returns how many bytes were actually used..
878 * NOTE! This may not be the same as how much of a user buffer
879 * we filled up (we may be padding etc), so we can only update
880 * "pos" here (the actor routine has to update the user buffer
881 * pointers and the remaining count).
882 */
883 ret = actor(desc, page, offset, nr);
884 offset += ret;
885 index += offset >> PAGE_CACHE_SHIFT;
886 offset &= ~PAGE_CACHE_MASK;
887
888 page_cache_release(page);
889 if (ret == nr && desc->count)
890 continue;
891 goto out;
892
893page_not_up_to_date:
894 /* Get exclusive access to the page ... */
895 lock_page(page);
896
897 /* Did it get unhashed before we got the lock? */
898 if (!page->mapping) {
899 unlock_page(page);
900 page_cache_release(page);
901 continue;
902 }
903
904 /* Did somebody else fill it already? */
905 if (PageUptodate(page)) {
906 unlock_page(page);
907 goto page_ok;
908 }
909
910readpage:
911 /* Start the actual read. The read will unlock the page. */
912 error = mapping->a_ops->readpage(filp, page);
913
Zach Brown994fc28c2005-12-15 14:28:17 -0800914 if (unlikely(error)) {
915 if (error == AOP_TRUNCATED_PAGE) {
916 page_cache_release(page);
917 goto find_page;
918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -0800920 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922 if (!PageUptodate(page)) {
923 lock_page(page);
924 if (!PageUptodate(page)) {
925 if (page->mapping == NULL) {
926 /*
927 * invalidate_inode_pages got it
928 */
929 unlock_page(page);
930 page_cache_release(page);
931 goto find_page;
932 }
933 unlock_page(page);
934 error = -EIO;
935 goto readpage_error;
936 }
937 unlock_page(page);
938 }
939
940 /*
941 * i_size must be checked after we have done ->readpage.
942 *
943 * Checking i_size after the readpage allows us to calculate
944 * the correct value for "nr", which means the zero-filled
945 * part of the page is not copied back to userspace (unless
946 * another truncate extends the file - this is desired though).
947 */
948 isize = i_size_read(inode);
949 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
950 if (unlikely(!isize || index > end_index)) {
951 page_cache_release(page);
952 goto out;
953 }
954
955 /* nr is the maximum number of bytes to copy from this page */
956 nr = PAGE_CACHE_SIZE;
957 if (index == end_index) {
958 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
959 if (nr <= offset) {
960 page_cache_release(page);
961 goto out;
962 }
963 }
964 nr = nr - offset;
965 goto page_ok;
966
967readpage_error:
968 /* UHHUH! A synchronous read error occurred. Report it */
969 desc->error = error;
970 page_cache_release(page);
971 goto out;
972
973no_cached_page:
974 /*
975 * Ok, it wasn't cached, so we need to create a new
976 * page..
977 */
978 if (!cached_page) {
979 cached_page = page_cache_alloc_cold(mapping);
980 if (!cached_page) {
981 desc->error = -ENOMEM;
982 goto out;
983 }
984 }
985 error = add_to_page_cache_lru(cached_page, mapping,
986 index, GFP_KERNEL);
987 if (error) {
988 if (error == -EEXIST)
989 goto find_page;
990 desc->error = error;
991 goto out;
992 }
993 page = cached_page;
994 cached_page = NULL;
995 goto readpage;
996 }
997
998out:
999 *_ra = ra;
1000
1001 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1002 if (cached_page)
1003 page_cache_release(cached_page);
1004 if (filp)
1005 file_accessed(filp);
1006}
1007
1008EXPORT_SYMBOL(do_generic_mapping_read);
1009
1010int file_read_actor(read_descriptor_t *desc, struct page *page,
1011 unsigned long offset, unsigned long size)
1012{
1013 char *kaddr;
1014 unsigned long left, count = desc->count;
1015
1016 if (size > count)
1017 size = count;
1018
1019 /*
1020 * Faults on the destination of a read are common, so do it before
1021 * taking the kmap.
1022 */
1023 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1024 kaddr = kmap_atomic(page, KM_USER0);
1025 left = __copy_to_user_inatomic(desc->arg.buf,
1026 kaddr + offset, size);
1027 kunmap_atomic(kaddr, KM_USER0);
1028 if (left == 0)
1029 goto success;
1030 }
1031
1032 /* Do it the slow way */
1033 kaddr = kmap(page);
1034 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1035 kunmap(page);
1036
1037 if (left) {
1038 size -= left;
1039 desc->error = -EFAULT;
1040 }
1041success:
1042 desc->count = count - size;
1043 desc->written += size;
1044 desc->arg.buf += size;
1045 return size;
1046}
1047
1048/*
1049 * This is the "read()" routine for all filesystems
1050 * that can use the page cache directly.
1051 */
1052ssize_t
1053__generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1054 unsigned long nr_segs, loff_t *ppos)
1055{
1056 struct file *filp = iocb->ki_filp;
1057 ssize_t retval;
1058 unsigned long seg;
1059 size_t count;
1060
1061 count = 0;
1062 for (seg = 0; seg < nr_segs; seg++) {
1063 const struct iovec *iv = &iov[seg];
1064
1065 /*
1066 * If any segment has a negative length, or the cumulative
1067 * length ever wraps negative then return -EINVAL.
1068 */
1069 count += iv->iov_len;
1070 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1071 return -EINVAL;
1072 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1073 continue;
1074 if (seg == 0)
1075 return -EFAULT;
1076 nr_segs = seg;
1077 count -= iv->iov_len; /* This segment is no good */
1078 break;
1079 }
1080
1081 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1082 if (filp->f_flags & O_DIRECT) {
1083 loff_t pos = *ppos, size;
1084 struct address_space *mapping;
1085 struct inode *inode;
1086
1087 mapping = filp->f_mapping;
1088 inode = mapping->host;
1089 retval = 0;
1090 if (!count)
1091 goto out; /* skip atime */
1092 size = i_size_read(inode);
1093 if (pos < size) {
1094 retval = generic_file_direct_IO(READ, iocb,
1095 iov, pos, nr_segs);
Suparna Bhattacharyab5c44c22005-05-21 16:33:36 -07001096 if (retval > 0 && !is_sync_kiocb(iocb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 retval = -EIOCBQUEUED;
1098 if (retval > 0)
1099 *ppos = pos + retval;
1100 }
1101 file_accessed(filp);
1102 goto out;
1103 }
1104
1105 retval = 0;
1106 if (count) {
1107 for (seg = 0; seg < nr_segs; seg++) {
1108 read_descriptor_t desc;
1109
1110 desc.written = 0;
1111 desc.arg.buf = iov[seg].iov_base;
1112 desc.count = iov[seg].iov_len;
1113 if (desc.count == 0)
1114 continue;
1115 desc.error = 0;
1116 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1117 retval += desc.written;
Tejun Heo39e88ca2005-10-30 15:02:40 -08001118 if (desc.error) {
1119 retval = retval ?: desc.error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 break;
1121 }
1122 }
1123 }
1124out:
1125 return retval;
1126}
1127
1128EXPORT_SYMBOL(__generic_file_aio_read);
1129
1130ssize_t
1131generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1132{
1133 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1134
1135 BUG_ON(iocb->ki_pos != pos);
1136 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1137}
1138
1139EXPORT_SYMBOL(generic_file_aio_read);
1140
1141ssize_t
1142generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1143{
1144 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1145 struct kiocb kiocb;
1146 ssize_t ret;
1147
1148 init_sync_kiocb(&kiocb, filp);
1149 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1150 if (-EIOCBQUEUED == ret)
1151 ret = wait_on_sync_kiocb(&kiocb);
1152 return ret;
1153}
1154
1155EXPORT_SYMBOL(generic_file_read);
1156
1157int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1158{
1159 ssize_t written;
1160 unsigned long count = desc->count;
1161 struct file *file = desc->arg.data;
1162
1163 if (size > count)
1164 size = count;
1165
1166 written = file->f_op->sendpage(file, page, offset,
1167 size, &file->f_pos, size<count);
1168 if (written < 0) {
1169 desc->error = written;
1170 written = 0;
1171 }
1172 desc->count = count - written;
1173 desc->written += written;
1174 return written;
1175}
1176
1177ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1178 size_t count, read_actor_t actor, void *target)
1179{
1180 read_descriptor_t desc;
1181
1182 if (!count)
1183 return 0;
1184
1185 desc.written = 0;
1186 desc.count = count;
1187 desc.arg.data = target;
1188 desc.error = 0;
1189
1190 do_generic_file_read(in_file, ppos, &desc, actor);
1191 if (desc.written)
1192 return desc.written;
1193 return desc.error;
1194}
1195
1196EXPORT_SYMBOL(generic_file_sendfile);
1197
1198static ssize_t
1199do_readahead(struct address_space *mapping, struct file *filp,
1200 unsigned long index, unsigned long nr)
1201{
1202 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1203 return -EINVAL;
1204
1205 force_page_cache_readahead(mapping, filp, index,
1206 max_sane_readahead(nr));
1207 return 0;
1208}
1209
1210asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1211{
1212 ssize_t ret;
1213 struct file *file;
1214
1215 ret = -EBADF;
1216 file = fget(fd);
1217 if (file) {
1218 if (file->f_mode & FMODE_READ) {
1219 struct address_space *mapping = file->f_mapping;
1220 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1221 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1222 unsigned long len = end - start + 1;
1223 ret = do_readahead(mapping, file, start, len);
1224 }
1225 fput(file);
1226 }
1227 return ret;
1228}
1229
1230#ifdef CONFIG_MMU
1231/*
1232 * This adds the requested page to the page cache if it isn't already there,
1233 * and schedules an I/O to read in its contents from disk.
1234 */
1235static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1236static int fastcall page_cache_read(struct file * file, unsigned long offset)
1237{
1238 struct address_space *mapping = file->f_mapping;
1239 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001240 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Zach Brown994fc28c2005-12-15 14:28:17 -08001242 do {
1243 page = page_cache_alloc_cold(mapping);
1244 if (!page)
1245 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Zach Brown994fc28c2005-12-15 14:28:17 -08001247 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1248 if (ret == 0)
1249 ret = mapping->a_ops->readpage(file, page);
1250 else if (ret == -EEXIST)
1251 ret = 0; /* losing race to add is OK */
1252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Zach Brown994fc28c2005-12-15 14:28:17 -08001255 } while (ret == AOP_TRUNCATED_PAGE);
1256
1257 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258}
1259
1260#define MMAP_LOTSAMISS (100)
1261
1262/*
1263 * filemap_nopage() is invoked via the vma operations vector for a
1264 * mapped memory region to read in file data during a page fault.
1265 *
1266 * The goto's are kind of ugly, but this streamlines the normal case of having
1267 * it in the page cache, and handles the special cases reasonably without
1268 * having a lot of duplicated code.
1269 */
1270struct page *filemap_nopage(struct vm_area_struct *area,
1271 unsigned long address, int *type)
1272{
1273 int error;
1274 struct file *file = area->vm_file;
1275 struct address_space *mapping = file->f_mapping;
1276 struct file_ra_state *ra = &file->f_ra;
1277 struct inode *inode = mapping->host;
1278 struct page *page;
1279 unsigned long size, pgoff;
1280 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1281
1282 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1283
1284retry_all:
1285 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1286 if (pgoff >= size)
1287 goto outside_data_content;
1288
1289 /* If we don't want any read-ahead, don't bother */
1290 if (VM_RandomReadHint(area))
1291 goto no_cached_page;
1292
1293 /*
1294 * The readahead code wants to be told about each and every page
1295 * so it can build and shrink its windows appropriately
1296 *
1297 * For sequential accesses, we use the generic readahead logic.
1298 */
1299 if (VM_SequentialReadHint(area))
1300 page_cache_readahead(mapping, ra, file, pgoff, 1);
1301
1302 /*
1303 * Do we have something in the page cache already?
1304 */
1305retry_find:
1306 page = find_get_page(mapping, pgoff);
1307 if (!page) {
1308 unsigned long ra_pages;
1309
1310 if (VM_SequentialReadHint(area)) {
1311 handle_ra_miss(mapping, ra, pgoff);
1312 goto no_cached_page;
1313 }
1314 ra->mmap_miss++;
1315
1316 /*
1317 * Do we miss much more than hit in this file? If so,
1318 * stop bothering with read-ahead. It will only hurt.
1319 */
1320 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1321 goto no_cached_page;
1322
1323 /*
1324 * To keep the pgmajfault counter straight, we need to
1325 * check did_readaround, as this is an inner loop.
1326 */
1327 if (!did_readaround) {
1328 majmin = VM_FAULT_MAJOR;
1329 inc_page_state(pgmajfault);
1330 }
1331 did_readaround = 1;
1332 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1333 if (ra_pages) {
1334 pgoff_t start = 0;
1335
1336 if (pgoff > ra_pages / 2)
1337 start = pgoff - ra_pages / 2;
1338 do_page_cache_readahead(mapping, file, start, ra_pages);
1339 }
1340 page = find_get_page(mapping, pgoff);
1341 if (!page)
1342 goto no_cached_page;
1343 }
1344
1345 if (!did_readaround)
1346 ra->mmap_hit++;
1347
1348 /*
1349 * Ok, found a page in the page cache, now we need to check
1350 * that it's up-to-date.
1351 */
1352 if (!PageUptodate(page))
1353 goto page_not_uptodate;
1354
1355success:
1356 /*
1357 * Found the page and have a reference on it.
1358 */
1359 mark_page_accessed(page);
1360 if (type)
1361 *type = majmin;
1362 return page;
1363
1364outside_data_content:
1365 /*
1366 * An external ptracer can access pages that normally aren't
1367 * accessible..
1368 */
1369 if (area->vm_mm == current->mm)
1370 return NULL;
1371 /* Fall through to the non-read-ahead case */
1372no_cached_page:
1373 /*
1374 * We're only likely to ever get here if MADV_RANDOM is in
1375 * effect.
1376 */
1377 error = page_cache_read(file, pgoff);
1378 grab_swap_token();
1379
1380 /*
1381 * The page we want has now been added to the page cache.
1382 * In the unlikely event that someone removed it in the
1383 * meantime, we'll just come back here and read it again.
1384 */
1385 if (error >= 0)
1386 goto retry_find;
1387
1388 /*
1389 * An error return from page_cache_read can result if the
1390 * system is low on memory, or a problem occurs while trying
1391 * to schedule I/O.
1392 */
1393 if (error == -ENOMEM)
1394 return NOPAGE_OOM;
1395 return NULL;
1396
1397page_not_uptodate:
1398 if (!did_readaround) {
1399 majmin = VM_FAULT_MAJOR;
1400 inc_page_state(pgmajfault);
1401 }
1402 lock_page(page);
1403
1404 /* Did it get unhashed while we waited for it? */
1405 if (!page->mapping) {
1406 unlock_page(page);
1407 page_cache_release(page);
1408 goto retry_all;
1409 }
1410
1411 /* Did somebody else get it up-to-date? */
1412 if (PageUptodate(page)) {
1413 unlock_page(page);
1414 goto success;
1415 }
1416
Zach Brown994fc28c2005-12-15 14:28:17 -08001417 error = mapping->a_ops->readpage(file, page);
1418 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 wait_on_page_locked(page);
1420 if (PageUptodate(page))
1421 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001422 } else if (error == AOP_TRUNCATED_PAGE) {
1423 page_cache_release(page);
1424 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 }
1426
1427 /*
1428 * Umm, take care of errors if the page isn't up-to-date.
1429 * Try to re-read it _once_. We do this synchronously,
1430 * because there really aren't any performance issues here
1431 * and we need to check for errors.
1432 */
1433 lock_page(page);
1434
1435 /* Somebody truncated the page on us? */
1436 if (!page->mapping) {
1437 unlock_page(page);
1438 page_cache_release(page);
1439 goto retry_all;
1440 }
1441
1442 /* Somebody else successfully read it in? */
1443 if (PageUptodate(page)) {
1444 unlock_page(page);
1445 goto success;
1446 }
1447 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001448 error = mapping->a_ops->readpage(file, page);
1449 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 wait_on_page_locked(page);
1451 if (PageUptodate(page))
1452 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001453 } else if (error == AOP_TRUNCATED_PAGE) {
1454 page_cache_release(page);
1455 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 }
1457
1458 /*
1459 * Things didn't work out. Return zero to tell the
1460 * mm layer so, possibly freeing the page cache page first.
1461 */
1462 page_cache_release(page);
1463 return NULL;
1464}
1465
1466EXPORT_SYMBOL(filemap_nopage);
1467
1468static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1469 int nonblock)
1470{
1471 struct address_space *mapping = file->f_mapping;
1472 struct page *page;
1473 int error;
1474
1475 /*
1476 * Do we have something in the page cache already?
1477 */
1478retry_find:
1479 page = find_get_page(mapping, pgoff);
1480 if (!page) {
1481 if (nonblock)
1482 return NULL;
1483 goto no_cached_page;
1484 }
1485
1486 /*
1487 * Ok, found a page in the page cache, now we need to check
1488 * that it's up-to-date.
1489 */
Jeff Moyerd3457342005-04-16 15:24:05 -07001490 if (!PageUptodate(page)) {
1491 if (nonblock) {
1492 page_cache_release(page);
1493 return NULL;
1494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 goto page_not_uptodate;
Jeff Moyerd3457342005-04-16 15:24:05 -07001496 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498success:
1499 /*
1500 * Found the page and have a reference on it.
1501 */
1502 mark_page_accessed(page);
1503 return page;
1504
1505no_cached_page:
1506 error = page_cache_read(file, pgoff);
1507
1508 /*
1509 * The page we want has now been added to the page cache.
1510 * In the unlikely event that someone removed it in the
1511 * meantime, we'll just come back here and read it again.
1512 */
1513 if (error >= 0)
1514 goto retry_find;
1515
1516 /*
1517 * An error return from page_cache_read can result if the
1518 * system is low on memory, or a problem occurs while trying
1519 * to schedule I/O.
1520 */
1521 return NULL;
1522
1523page_not_uptodate:
1524 lock_page(page);
1525
1526 /* Did it get unhashed while we waited for it? */
1527 if (!page->mapping) {
1528 unlock_page(page);
1529 goto err;
1530 }
1531
1532 /* Did somebody else get it up-to-date? */
1533 if (PageUptodate(page)) {
1534 unlock_page(page);
1535 goto success;
1536 }
1537
Zach Brown994fc28c2005-12-15 14:28:17 -08001538 error = mapping->a_ops->readpage(file, page);
1539 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 wait_on_page_locked(page);
1541 if (PageUptodate(page))
1542 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001543 } else if (error == AOP_TRUNCATED_PAGE) {
1544 page_cache_release(page);
1545 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 }
1547
1548 /*
1549 * Umm, take care of errors if the page isn't up-to-date.
1550 * Try to re-read it _once_. We do this synchronously,
1551 * because there really aren't any performance issues here
1552 * and we need to check for errors.
1553 */
1554 lock_page(page);
1555
1556 /* Somebody truncated the page on us? */
1557 if (!page->mapping) {
1558 unlock_page(page);
1559 goto err;
1560 }
1561 /* Somebody else successfully read it in? */
1562 if (PageUptodate(page)) {
1563 unlock_page(page);
1564 goto success;
1565 }
1566
1567 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001568 error = mapping->a_ops->readpage(file, page);
1569 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 wait_on_page_locked(page);
1571 if (PageUptodate(page))
1572 goto success;
Zach Brown994fc28c2005-12-15 14:28:17 -08001573 } else if (error == AOP_TRUNCATED_PAGE) {
1574 page_cache_release(page);
1575 goto retry_find;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 }
1577
1578 /*
1579 * Things didn't work out. Return zero to tell the
1580 * mm layer so, possibly freeing the page cache page first.
1581 */
1582err:
1583 page_cache_release(page);
1584
1585 return NULL;
1586}
1587
1588int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1589 unsigned long len, pgprot_t prot, unsigned long pgoff,
1590 int nonblock)
1591{
1592 struct file *file = vma->vm_file;
1593 struct address_space *mapping = file->f_mapping;
1594 struct inode *inode = mapping->host;
1595 unsigned long size;
1596 struct mm_struct *mm = vma->vm_mm;
1597 struct page *page;
1598 int err;
1599
1600 if (!nonblock)
1601 force_page_cache_readahead(mapping, vma->vm_file,
1602 pgoff, len >> PAGE_CACHE_SHIFT);
1603
1604repeat:
1605 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1606 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1607 return -EINVAL;
1608
1609 page = filemap_getpage(file, pgoff, nonblock);
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001610
1611 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1612 * done in shmem_populate calling shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 if (!page && !nonblock)
1614 return -ENOMEM;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001615
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 if (page) {
1617 err = install_page(mm, vma, addr, page, prot);
1618 if (err) {
1619 page_cache_release(page);
1620 return err;
1621 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001622 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001623 /* No page was found just because we can't read it in now (being
1624 * here implies nonblock != 0), but the page may exist, so set
1625 * the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 err = install_file_pte(mm, vma, addr, pgoff, prot);
1627 if (err)
1628 return err;
1629 }
1630
1631 len -= PAGE_SIZE;
1632 addr += PAGE_SIZE;
1633 pgoff++;
1634 if (len)
1635 goto repeat;
1636
1637 return 0;
1638}
Nikita Danilovb1459462005-10-29 18:17:02 -07001639EXPORT_SYMBOL(filemap_populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
1641struct vm_operations_struct generic_file_vm_ops = {
1642 .nopage = filemap_nopage,
1643 .populate = filemap_populate,
1644};
1645
1646/* This is used for a general mmap of a disk file */
1647
1648int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1649{
1650 struct address_space *mapping = file->f_mapping;
1651
1652 if (!mapping->a_ops->readpage)
1653 return -ENOEXEC;
1654 file_accessed(file);
1655 vma->vm_ops = &generic_file_vm_ops;
1656 return 0;
1657}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659/*
1660 * This is for filesystems which do not implement ->writepage.
1661 */
1662int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1663{
1664 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1665 return -EINVAL;
1666 return generic_file_mmap(file, vma);
1667}
1668#else
1669int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1670{
1671 return -ENOSYS;
1672}
1673int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1674{
1675 return -ENOSYS;
1676}
1677#endif /* CONFIG_MMU */
1678
1679EXPORT_SYMBOL(generic_file_mmap);
1680EXPORT_SYMBOL(generic_file_readonly_mmap);
1681
1682static inline struct page *__read_cache_page(struct address_space *mapping,
1683 unsigned long index,
1684 int (*filler)(void *,struct page*),
1685 void *data)
1686{
1687 struct page *page, *cached_page = NULL;
1688 int err;
1689repeat:
1690 page = find_get_page(mapping, index);
1691 if (!page) {
1692 if (!cached_page) {
1693 cached_page = page_cache_alloc_cold(mapping);
1694 if (!cached_page)
1695 return ERR_PTR(-ENOMEM);
1696 }
1697 err = add_to_page_cache_lru(cached_page, mapping,
1698 index, GFP_KERNEL);
1699 if (err == -EEXIST)
1700 goto repeat;
1701 if (err < 0) {
1702 /* Presumably ENOMEM for radix tree node */
1703 page_cache_release(cached_page);
1704 return ERR_PTR(err);
1705 }
1706 page = cached_page;
1707 cached_page = NULL;
1708 err = filler(data, page);
1709 if (err < 0) {
1710 page_cache_release(page);
1711 page = ERR_PTR(err);
1712 }
1713 }
1714 if (cached_page)
1715 page_cache_release(cached_page);
1716 return page;
1717}
1718
1719/*
1720 * Read into the page cache. If a page already exists,
1721 * and PageUptodate() is not set, try to fill the page.
1722 */
1723struct page *read_cache_page(struct address_space *mapping,
1724 unsigned long index,
1725 int (*filler)(void *,struct page*),
1726 void *data)
1727{
1728 struct page *page;
1729 int err;
1730
1731retry:
1732 page = __read_cache_page(mapping, index, filler, data);
1733 if (IS_ERR(page))
1734 goto out;
1735 mark_page_accessed(page);
1736 if (PageUptodate(page))
1737 goto out;
1738
1739 lock_page(page);
1740 if (!page->mapping) {
1741 unlock_page(page);
1742 page_cache_release(page);
1743 goto retry;
1744 }
1745 if (PageUptodate(page)) {
1746 unlock_page(page);
1747 goto out;
1748 }
1749 err = filler(data, page);
1750 if (err < 0) {
1751 page_cache_release(page);
1752 page = ERR_PTR(err);
1753 }
1754 out:
1755 return page;
1756}
1757
1758EXPORT_SYMBOL(read_cache_page);
1759
1760/*
1761 * If the page was newly created, increment its refcount and add it to the
1762 * caller's lru-buffering pagevec. This function is specifically for
1763 * generic_file_write().
1764 */
1765static inline struct page *
1766__grab_cache_page(struct address_space *mapping, unsigned long index,
1767 struct page **cached_page, struct pagevec *lru_pvec)
1768{
1769 int err;
1770 struct page *page;
1771repeat:
1772 page = find_lock_page(mapping, index);
1773 if (!page) {
1774 if (!*cached_page) {
1775 *cached_page = page_cache_alloc(mapping);
1776 if (!*cached_page)
1777 return NULL;
1778 }
1779 err = add_to_page_cache(*cached_page, mapping,
1780 index, GFP_KERNEL);
1781 if (err == -EEXIST)
1782 goto repeat;
1783 if (err == 0) {
1784 page = *cached_page;
1785 page_cache_get(page);
1786 if (!pagevec_add(lru_pvec, page))
1787 __pagevec_lru_add(lru_pvec);
1788 *cached_page = NULL;
1789 }
1790 }
1791 return page;
1792}
1793
1794/*
1795 * The logic we want is
1796 *
1797 * if suid or (sgid and xgrp)
1798 * remove privs
1799 */
1800int remove_suid(struct dentry *dentry)
1801{
1802 mode_t mode = dentry->d_inode->i_mode;
1803 int kill = 0;
1804 int result = 0;
1805
1806 /* suid always must be killed */
1807 if (unlikely(mode & S_ISUID))
1808 kill = ATTR_KILL_SUID;
1809
1810 /*
1811 * sgid without any exec bits is just a mandatory locking mark; leave
1812 * it alone. If some exec bits are set, it's a real sgid; kill it.
1813 */
1814 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1815 kill |= ATTR_KILL_SGID;
1816
1817 if (unlikely(kill && !capable(CAP_FSETID))) {
1818 struct iattr newattrs;
1819
1820 newattrs.ia_valid = ATTR_FORCE | kill;
1821 result = notify_change(dentry, &newattrs);
1822 }
1823 return result;
1824}
1825EXPORT_SYMBOL(remove_suid);
1826
Carsten Otteceffc072005-06-23 22:05:25 -07001827size_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828__filemap_copy_from_user_iovec(char *vaddr,
1829 const struct iovec *iov, size_t base, size_t bytes)
1830{
1831 size_t copied = 0, left = 0;
1832
1833 while (bytes) {
1834 char __user *buf = iov->iov_base + base;
1835 int copy = min(bytes, iov->iov_len - base);
1836
1837 base = 0;
1838 left = __copy_from_user_inatomic(vaddr, buf, copy);
1839 copied += copy;
1840 bytes -= copy;
1841 vaddr += copy;
1842 iov++;
1843
1844 if (unlikely(left)) {
1845 /* zero the rest of the target like __copy_from_user */
1846 if (bytes)
1847 memset(vaddr, 0, bytes);
1848 break;
1849 }
1850 }
1851 return copied - left;
1852}
1853
1854/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 * Performs necessary checks before doing a write
1856 *
1857 * Can adjust writing position aor amount of bytes to write.
1858 * Returns appropriate error code that caller should return or
1859 * zero in case that write should be allowed.
1860 */
1861inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1862{
1863 struct inode *inode = file->f_mapping->host;
1864 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1865
1866 if (unlikely(*pos < 0))
1867 return -EINVAL;
1868
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 if (!isblk) {
1870 /* FIXME: this is for backwards compatibility with 2.4 */
1871 if (file->f_flags & O_APPEND)
1872 *pos = i_size_read(inode);
1873
1874 if (limit != RLIM_INFINITY) {
1875 if (*pos >= limit) {
1876 send_sig(SIGXFSZ, current, 0);
1877 return -EFBIG;
1878 }
1879 if (*count > limit - (typeof(limit))*pos) {
1880 *count = limit - (typeof(limit))*pos;
1881 }
1882 }
1883 }
1884
1885 /*
1886 * LFS rule
1887 */
1888 if (unlikely(*pos + *count > MAX_NON_LFS &&
1889 !(file->f_flags & O_LARGEFILE))) {
1890 if (*pos >= MAX_NON_LFS) {
1891 send_sig(SIGXFSZ, current, 0);
1892 return -EFBIG;
1893 }
1894 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1895 *count = MAX_NON_LFS - (unsigned long)*pos;
1896 }
1897 }
1898
1899 /*
1900 * Are we about to exceed the fs block limit ?
1901 *
1902 * If we have written data it becomes a short write. If we have
1903 * exceeded without writing data we send a signal and return EFBIG.
1904 * Linus frestrict idea will clean these up nicely..
1905 */
1906 if (likely(!isblk)) {
1907 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1908 if (*count || *pos > inode->i_sb->s_maxbytes) {
1909 send_sig(SIGXFSZ, current, 0);
1910 return -EFBIG;
1911 }
1912 /* zero-length writes at ->s_maxbytes are OK */
1913 }
1914
1915 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1916 *count = inode->i_sb->s_maxbytes - *pos;
1917 } else {
1918 loff_t isize;
1919 if (bdev_read_only(I_BDEV(inode)))
1920 return -EPERM;
1921 isize = i_size_read(inode);
1922 if (*pos >= isize) {
1923 if (*count || *pos > isize)
1924 return -ENOSPC;
1925 }
1926
1927 if (*pos + *count > isize)
1928 *count = isize - *pos;
1929 }
1930 return 0;
1931}
1932EXPORT_SYMBOL(generic_write_checks);
1933
1934ssize_t
1935generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1936 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1937 size_t count, size_t ocount)
1938{
1939 struct file *file = iocb->ki_filp;
1940 struct address_space *mapping = file->f_mapping;
1941 struct inode *inode = mapping->host;
1942 ssize_t written;
1943
1944 if (count != ocount)
1945 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1946
1947 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1948 if (written > 0) {
1949 loff_t end = pos + written;
1950 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1951 i_size_write(inode, end);
1952 mark_inode_dirty(inode);
1953 }
1954 *ppos = end;
1955 }
1956
1957 /*
1958 * Sync the fs metadata but not the minor inode changes and
1959 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001960 * i_mutex is held, which protects generic_osync_inode() from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 * livelocking.
1962 */
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07001963 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
1964 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1965 if (err < 0)
1966 written = err;
1967 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 if (written == count && !is_sync_kiocb(iocb))
1969 written = -EIOCBQUEUED;
1970 return written;
1971}
1972EXPORT_SYMBOL(generic_file_direct_write);
1973
1974ssize_t
1975generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
1976 unsigned long nr_segs, loff_t pos, loff_t *ppos,
1977 size_t count, ssize_t written)
1978{
1979 struct file *file = iocb->ki_filp;
1980 struct address_space * mapping = file->f_mapping;
1981 struct address_space_operations *a_ops = mapping->a_ops;
1982 struct inode *inode = mapping->host;
1983 long status = 0;
1984 struct page *page;
1985 struct page *cached_page = NULL;
1986 size_t bytes;
1987 struct pagevec lru_pvec;
1988 const struct iovec *cur_iov = iov; /* current iovec */
1989 size_t iov_base = 0; /* offset in the current iovec */
1990 char __user *buf;
1991
1992 pagevec_init(&lru_pvec, 0);
1993
1994 /*
1995 * handle partial DIO write. Adjust cur_iov if needed.
1996 */
1997 if (likely(nr_segs == 1))
1998 buf = iov->iov_base + written;
1999 else {
2000 filemap_set_next_iovec(&cur_iov, &iov_base, written);
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002001 buf = cur_iov->iov_base + iov_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 }
2003
2004 do {
2005 unsigned long index;
2006 unsigned long offset;
Martin Schwidefskya5117182005-06-06 13:35:54 -07002007 unsigned long maxlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 size_t copied;
2009
2010 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2011 index = pos >> PAGE_CACHE_SHIFT;
2012 bytes = PAGE_CACHE_SIZE - offset;
2013 if (bytes > count)
2014 bytes = count;
2015
2016 /*
2017 * Bring in the user page that we will copy from _first_.
2018 * Otherwise there's a nasty deadlock on copying from the
2019 * same page as we're writing to, without it being marked
2020 * up-to-date.
2021 */
Martin Schwidefskya5117182005-06-06 13:35:54 -07002022 maxlen = cur_iov->iov_len - iov_base;
2023 if (maxlen > bytes)
2024 maxlen = bytes;
2025 fault_in_pages_readable(buf, maxlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
2027 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2028 if (!page) {
2029 status = -ENOMEM;
2030 break;
2031 }
2032
2033 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2034 if (unlikely(status)) {
2035 loff_t isize = i_size_read(inode);
Zach Brown994fc28c2005-12-15 14:28:17 -08002036
2037 if (status != AOP_TRUNCATED_PAGE)
2038 unlock_page(page);
2039 page_cache_release(page);
2040 if (status == AOP_TRUNCATED_PAGE)
2041 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 /*
2043 * prepare_write() may have instantiated a few blocks
2044 * outside i_size. Trim these off again.
2045 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 if (pos + bytes > isize)
2047 vmtruncate(inode, isize);
2048 break;
2049 }
2050 if (likely(nr_segs == 1))
2051 copied = filemap_copy_from_user(page, offset,
2052 buf, bytes);
2053 else
2054 copied = filemap_copy_from_user_iovec(page, offset,
2055 cur_iov, iov_base, bytes);
2056 flush_dcache_page(page);
2057 status = a_ops->commit_write(file, page, offset, offset+bytes);
Zach Brown994fc28c2005-12-15 14:28:17 -08002058 if (status == AOP_TRUNCATED_PAGE) {
2059 page_cache_release(page);
2060 continue;
2061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 if (likely(copied > 0)) {
2063 if (!status)
2064 status = copied;
2065
2066 if (status >= 0) {
2067 written += status;
2068 count -= status;
2069 pos += status;
2070 buf += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002071 if (unlikely(nr_segs > 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 filemap_set_next_iovec(&cur_iov,
2073 &iov_base, status);
Badari Pulavartyb0cfbd92005-06-25 14:55:42 -07002074 if (count)
2075 buf = cur_iov->iov_base +
2076 iov_base;
Martin Schwidefskya5117182005-06-06 13:35:54 -07002077 } else {
2078 iov_base += status;
akpm@osdl.orgf021e922005-05-01 08:58:35 -07002079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 }
2081 }
2082 if (unlikely(copied != bytes))
2083 if (status >= 0)
2084 status = -EFAULT;
2085 unlock_page(page);
2086 mark_page_accessed(page);
2087 page_cache_release(page);
2088 if (status < 0)
2089 break;
2090 balance_dirty_pages_ratelimited(mapping);
2091 cond_resched();
2092 } while (count);
2093 *ppos = pos;
2094
2095 if (cached_page)
2096 page_cache_release(cached_page);
2097
2098 /*
2099 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2100 */
2101 if (likely(status >= 0)) {
2102 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2103 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2104 status = generic_osync_inode(inode, mapping,
2105 OSYNC_METADATA|OSYNC_DATA);
2106 }
2107 }
2108
2109 /*
2110 * If we get here for O_DIRECT writes then we must have fallen through
2111 * to buffered writes (block instantiation inside i_size). So we sync
2112 * the file data here, to try to honour O_DIRECT expectations.
2113 */
2114 if (unlikely(file->f_flags & O_DIRECT) && written)
2115 status = filemap_write_and_wait(mapping);
2116
2117 pagevec_lru_add(&lru_pvec);
2118 return written ? written : status;
2119}
2120EXPORT_SYMBOL(generic_file_buffered_write);
2121
Adrian Bunk5ce78522005-09-10 00:26:28 -07002122static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2124 unsigned long nr_segs, loff_t *ppos)
2125{
2126 struct file *file = iocb->ki_filp;
2127 struct address_space * mapping = file->f_mapping;
2128 size_t ocount; /* original count */
2129 size_t count; /* after file limit checks */
2130 struct inode *inode = mapping->host;
2131 unsigned long seg;
2132 loff_t pos;
2133 ssize_t written;
2134 ssize_t err;
2135
2136 ocount = 0;
2137 for (seg = 0; seg < nr_segs; seg++) {
2138 const struct iovec *iv = &iov[seg];
2139
2140 /*
2141 * If any segment has a negative length, or the cumulative
2142 * length ever wraps negative then return -EINVAL.
2143 */
2144 ocount += iv->iov_len;
2145 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2146 return -EINVAL;
2147 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2148 continue;
2149 if (seg == 0)
2150 return -EFAULT;
2151 nr_segs = seg;
2152 ocount -= iv->iov_len; /* This segment is no good */
2153 break;
2154 }
2155
2156 count = ocount;
2157 pos = *ppos;
2158
2159 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2160
2161 /* We can write back this queue in page reclaim */
2162 current->backing_dev_info = mapping->backing_dev_info;
2163 written = 0;
2164
2165 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2166 if (err)
2167 goto out;
2168
2169 if (count == 0)
2170 goto out;
2171
2172 err = remove_suid(file->f_dentry);
2173 if (err)
2174 goto out;
2175
Christoph Hellwig870f4812006-01-09 20:52:01 -08002176 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
2178 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2179 if (unlikely(file->f_flags & O_DIRECT)) {
2180 written = generic_file_direct_write(iocb, iov,
2181 &nr_segs, pos, ppos, count, ocount);
2182 if (written < 0 || written == count)
2183 goto out;
2184 /*
2185 * direct-io write to a hole: fall through to buffered I/O
2186 * for completing the rest of the request.
2187 */
2188 pos += written;
2189 count -= written;
2190 }
2191
2192 written = generic_file_buffered_write(iocb, iov, nr_segs,
2193 pos, ppos, count, written);
2194out:
2195 current->backing_dev_info = NULL;
2196 return written ? written : err;
2197}
2198EXPORT_SYMBOL(generic_file_aio_write_nolock);
2199
2200ssize_t
2201generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2202 unsigned long nr_segs, loff_t *ppos)
2203{
2204 struct file *file = iocb->ki_filp;
2205 struct address_space *mapping = file->f_mapping;
2206 struct inode *inode = mapping->host;
2207 ssize_t ret;
2208 loff_t pos = *ppos;
2209
2210 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2211
2212 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2213 int err;
2214
2215 err = sync_page_range_nolock(inode, mapping, pos, ret);
2216 if (err < 0)
2217 ret = err;
2218 }
2219 return ret;
2220}
2221
Adrian Bunk5ce78522005-09-10 00:26:28 -07002222static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223__generic_file_write_nolock(struct file *file, const struct iovec *iov,
2224 unsigned long nr_segs, loff_t *ppos)
2225{
2226 struct kiocb kiocb;
2227 ssize_t ret;
2228
2229 init_sync_kiocb(&kiocb, file);
2230 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2231 if (ret == -EIOCBQUEUED)
2232 ret = wait_on_sync_kiocb(&kiocb);
2233 return ret;
2234}
2235
2236ssize_t
2237generic_file_write_nolock(struct file *file, const struct iovec *iov,
2238 unsigned long nr_segs, loff_t *ppos)
2239{
2240 struct kiocb kiocb;
2241 ssize_t ret;
2242
2243 init_sync_kiocb(&kiocb, file);
2244 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2245 if (-EIOCBQUEUED == ret)
2246 ret = wait_on_sync_kiocb(&kiocb);
2247 return ret;
2248}
2249EXPORT_SYMBOL(generic_file_write_nolock);
2250
2251ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2252 size_t count, loff_t pos)
2253{
2254 struct file *file = iocb->ki_filp;
2255 struct address_space *mapping = file->f_mapping;
2256 struct inode *inode = mapping->host;
2257 ssize_t ret;
2258 struct iovec local_iov = { .iov_base = (void __user *)buf,
2259 .iov_len = count };
2260
2261 BUG_ON(iocb->ki_pos != pos);
2262
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002263 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2265 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002266 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
2268 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2269 ssize_t err;
2270
2271 err = sync_page_range(inode, mapping, pos, ret);
2272 if (err < 0)
2273 ret = err;
2274 }
2275 return ret;
2276}
2277EXPORT_SYMBOL(generic_file_aio_write);
2278
2279ssize_t generic_file_write(struct file *file, const char __user *buf,
2280 size_t count, loff_t *ppos)
2281{
2282 struct address_space *mapping = file->f_mapping;
2283 struct inode *inode = mapping->host;
2284 ssize_t ret;
2285 struct iovec local_iov = { .iov_base = (void __user *)buf,
2286 .iov_len = count };
2287
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002288 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002290 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
2292 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2293 ssize_t err;
2294
2295 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2296 if (err < 0)
2297 ret = err;
2298 }
2299 return ret;
2300}
2301EXPORT_SYMBOL(generic_file_write);
2302
2303ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2304 unsigned long nr_segs, loff_t *ppos)
2305{
2306 struct kiocb kiocb;
2307 ssize_t ret;
2308
2309 init_sync_kiocb(&kiocb, filp);
2310 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2311 if (-EIOCBQUEUED == ret)
2312 ret = wait_on_sync_kiocb(&kiocb);
2313 return ret;
2314}
2315EXPORT_SYMBOL(generic_file_readv);
2316
2317ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2318 unsigned long nr_segs, loff_t *ppos)
2319{
2320 struct address_space *mapping = file->f_mapping;
2321 struct inode *inode = mapping->host;
2322 ssize_t ret;
2323
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002324 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002326 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
2328 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2329 int err;
2330
2331 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2332 if (err < 0)
2333 ret = err;
2334 }
2335 return ret;
2336}
2337EXPORT_SYMBOL(generic_file_writev);
2338
2339/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002340 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 * went wrong during pagecache shootdown.
2342 */
Adrian Bunk5ce78522005-09-10 00:26:28 -07002343static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2345 loff_t offset, unsigned long nr_segs)
2346{
2347 struct file *file = iocb->ki_filp;
2348 struct address_space *mapping = file->f_mapping;
2349 ssize_t retval;
2350 size_t write_len = 0;
2351
2352 /*
2353 * If it's a write, unmap all mmappings of the file up-front. This
2354 * will cause any pte dirty bits to be propagated into the pageframes
2355 * for the subsequent filemap_write_and_wait().
2356 */
2357 if (rw == WRITE) {
2358 write_len = iov_length(iov, nr_segs);
2359 if (mapping_mapped(mapping))
2360 unmap_mapping_range(mapping, offset, write_len, 0);
2361 }
2362
2363 retval = filemap_write_and_wait(mapping);
2364 if (retval == 0) {
2365 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2366 offset, nr_segs);
2367 if (rw == WRITE && mapping->nrpages) {
2368 pgoff_t end = (offset + write_len - 1)
2369 >> PAGE_CACHE_SHIFT;
2370 int err = invalidate_inode_pages2_range(mapping,
2371 offset >> PAGE_CACHE_SHIFT, end);
2372 if (err)
2373 retval = err;
2374 }
2375 }
2376 return retval;
2377}