4 * Copyright (C) 1994-1999 Linus Torvalds
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)
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/compiler.h>
16 #include <linux/uaccess.h>
17 #include <linux/aio.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.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>
32 #include <linux/cpuset.h>
37 * FIXME: remove all knowledge of the buffer layer from the core VM
39 #include <linux/buffer_head.h> /* for generic_osync_inode */
44 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
45 loff_t offset
, unsigned long nr_segs
);
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * Shared mappings now work. 15.8.1995 Bruno.
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
64 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
68 * ->i_mmap_lock (truncate->unmap_mapping_range)
72 * ->page_table_lock or pte_lock (various, mainly in memory.c)
73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 * ->lock_page (access_process_vm)
78 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
82 * ->i_alloc_sem (various)
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->dcache_lock (proc_pid_lookup)
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
115 void __remove_from_page_cache(struct page
*page
)
117 struct address_space
*mapping
= page
->mapping
;
119 radix_tree_delete(&mapping
->page_tree
, page
->index
);
120 page
->mapping
= NULL
;
122 __dec_zone_page_state(page
, NR_FILE_PAGES
);
123 BUG_ON(page_mapped(page
));
126 void remove_from_page_cache(struct page
*page
)
128 struct address_space
*mapping
= page
->mapping
;
130 BUG_ON(!PageLocked(page
));
132 write_lock_irq(&mapping
->tree_lock
);
133 __remove_from_page_cache(page
);
134 write_unlock_irq(&mapping
->tree_lock
);
137 static int sync_page(void *word
)
139 struct address_space
*mapping
;
142 page
= container_of((unsigned long *)word
, struct page
, flags
);
145 * 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
160 * ignore it for all cases but swap, where only page_private(page) is
161 * of interest. When page_mapping() does go NULL, the entire
162 * call stack gracefully ignores the page and returns.
166 mapping
= page_mapping(page
);
167 if (mapping
&& mapping
->a_ops
&& mapping
->a_ops
->sync_page
)
168 mapping
->a_ops
->sync_page(page
);
174 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
175 * @mapping: address space structure to write
176 * @start: offset in bytes where the range starts
177 * @end: offset in bytes where the range ends (inclusive)
178 * @sync_mode: enable synchronous operation
180 * Start writeback against all of a mapping's dirty pages that lie
181 * within the byte offsets <start, end> inclusive.
183 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
184 * opposed to a regular memory cleansing writeback. The difference between
185 * these two operations is that if a dirty page/buffer is encountered, it must
186 * be waited upon, and not just skipped over.
188 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
189 loff_t end
, int sync_mode
)
192 struct writeback_control wbc
= {
193 .sync_mode
= sync_mode
,
194 .nr_to_write
= mapping
->nrpages
* 2,
195 .range_start
= start
,
199 if (!mapping_cap_writeback_dirty(mapping
))
202 ret
= do_writepages(mapping
, &wbc
);
206 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
209 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
212 int filemap_fdatawrite(struct address_space
*mapping
)
214 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
216 EXPORT_SYMBOL(filemap_fdatawrite
);
218 static int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
221 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
225 * filemap_flush - mostly a non-blocking flush
226 * @mapping: target address_space
228 * This is a mostly non-blocking flush. Not suitable for data-integrity
229 * purposes - I/O may not be started against all dirty pages.
231 int filemap_flush(struct address_space
*mapping
)
233 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
235 EXPORT_SYMBOL(filemap_flush
);
238 * wait_on_page_writeback_range - wait for writeback to complete
239 * @mapping: target address_space
240 * @start: beginning page index
241 * @end: ending page index
243 * Wait for writeback to complete against pages indexed by start->end
246 int wait_on_page_writeback_range(struct address_space
*mapping
,
247 pgoff_t start
, pgoff_t end
)
257 pagevec_init(&pvec
, 0);
259 while ((index
<= end
) &&
260 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
261 PAGECACHE_TAG_WRITEBACK
,
262 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
265 for (i
= 0; i
< nr_pages
; i
++) {
266 struct page
*page
= pvec
.pages
[i
];
268 /* until radix tree lookup accepts end_index */
269 if (page
->index
> end
)
272 wait_on_page_writeback(page
);
276 pagevec_release(&pvec
);
280 /* Check for outstanding write errors */
281 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
283 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
290 * sync_page_range - write and wait on all pages in the passed range
291 * @inode: target inode
292 * @mapping: target address_space
293 * @pos: beginning offset in pages to write
294 * @count: number of bytes to write
296 * Write and wait upon all the pages in the passed range. This is a "data
297 * integrity" operation. It waits upon in-flight writeout before starting and
298 * waiting upon new writeout. If there was an IO error, return it.
300 * We need to re-take i_mutex during the generic_osync_inode list walk because
301 * it is otherwise livelockable.
303 int sync_page_range(struct inode
*inode
, struct address_space
*mapping
,
304 loff_t pos
, loff_t count
)
306 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
307 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
310 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
312 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
314 mutex_lock(&inode
->i_mutex
);
315 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
316 mutex_unlock(&inode
->i_mutex
);
319 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
322 EXPORT_SYMBOL(sync_page_range
);
325 * sync_page_range_nolock
326 * @inode: target inode
327 * @mapping: target address_space
328 * @pos: beginning offset in pages to write
329 * @count: number of bytes to write
331 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
332 * as it forces O_SYNC writers to different parts of the same file
333 * to be serialised right until io completion.
335 int sync_page_range_nolock(struct inode
*inode
, struct address_space
*mapping
,
336 loff_t pos
, loff_t count
)
338 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
339 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
342 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
344 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
346 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
348 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
351 EXPORT_SYMBOL(sync_page_range_nolock
);
354 * filemap_fdatawait - wait for all under-writeback pages to complete
355 * @mapping: address space structure to wait for
357 * Walk the list of under-writeback pages of the given address space
358 * and wait for all of them.
360 int filemap_fdatawait(struct address_space
*mapping
)
362 loff_t i_size
= i_size_read(mapping
->host
);
367 return wait_on_page_writeback_range(mapping
, 0,
368 (i_size
- 1) >> PAGE_CACHE_SHIFT
);
370 EXPORT_SYMBOL(filemap_fdatawait
);
372 int filemap_write_and_wait(struct address_space
*mapping
)
376 if (mapping
->nrpages
) {
377 err
= filemap_fdatawrite(mapping
);
379 * Even if the above returned error, the pages may be
380 * written partially (e.g. -ENOSPC), so we wait for it.
381 * But the -EIO is special case, it may indicate the worst
382 * thing (e.g. bug) happened, so we avoid waiting for it.
385 int err2
= filemap_fdatawait(mapping
);
392 EXPORT_SYMBOL(filemap_write_and_wait
);
395 * filemap_write_and_wait_range - write out & wait on a file range
396 * @mapping: the address_space for the pages
397 * @lstart: offset in bytes where the range starts
398 * @lend: offset in bytes where the range ends (inclusive)
400 * Write out and wait upon file offsets lstart->lend, inclusive.
402 * Note that `lend' is inclusive (describes the last byte to be written) so
403 * that this function can be used to write to the very end-of-file (end = -1).
405 int filemap_write_and_wait_range(struct address_space
*mapping
,
406 loff_t lstart
, loff_t lend
)
410 if (mapping
->nrpages
) {
411 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
413 /* See comment of filemap_write_and_wait() */
415 int err2
= wait_on_page_writeback_range(mapping
,
416 lstart
>> PAGE_CACHE_SHIFT
,
417 lend
>> PAGE_CACHE_SHIFT
);
426 * add_to_page_cache - add newly allocated pagecache pages
428 * @mapping: the page's address_space
429 * @offset: page index
430 * @gfp_mask: page allocation mode
432 * This function is used to add newly allocated pagecache pages;
433 * the page is new, so we can just run SetPageLocked() against it.
434 * The other page state flags were set by rmqueue().
436 * This function does not add the page to the LRU. The caller must do that.
438 int add_to_page_cache(struct page
*page
, struct address_space
*mapping
,
439 pgoff_t offset
, gfp_t gfp_mask
)
441 int error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
444 write_lock_irq(&mapping
->tree_lock
);
445 error
= radix_tree_insert(&mapping
->page_tree
, offset
, page
);
447 page_cache_get(page
);
449 page
->mapping
= mapping
;
450 page
->index
= offset
;
452 __inc_zone_page_state(page
, NR_FILE_PAGES
);
454 write_unlock_irq(&mapping
->tree_lock
);
455 radix_tree_preload_end();
459 EXPORT_SYMBOL(add_to_page_cache
);
461 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
462 pgoff_t offset
, gfp_t gfp_mask
)
464 int ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
471 struct page
*__page_cache_alloc(gfp_t gfp
)
473 if (cpuset_do_page_mem_spread()) {
474 int n
= cpuset_mem_spread_node();
475 return alloc_pages_node(n
, gfp
, 0);
477 return alloc_pages(gfp
, 0);
479 EXPORT_SYMBOL(__page_cache_alloc
);
482 static int __sleep_on_page_lock(void *word
)
489 * In order to wait for pages to become available there must be
490 * waitqueues associated with pages. By using a hash table of
491 * waitqueues where the bucket discipline is to maintain all
492 * waiters on the same queue and wake all when any of the pages
493 * become available, and for the woken contexts to check to be
494 * sure the appropriate page became available, this saves space
495 * at a cost of "thundering herd" phenomena during rare hash
498 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
500 const struct zone
*zone
= page_zone(page
);
502 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
505 static inline void wake_up_page(struct page
*page
, int bit
)
507 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
510 void fastcall
wait_on_page_bit(struct page
*page
, int bit_nr
)
512 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
514 if (test_bit(bit_nr
, &page
->flags
))
515 __wait_on_bit(page_waitqueue(page
), &wait
, sync_page
,
516 TASK_UNINTERRUPTIBLE
);
518 EXPORT_SYMBOL(wait_on_page_bit
);
521 * unlock_page - unlock a locked page
524 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
525 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
526 * mechananism between PageLocked pages and PageWriteback pages is shared.
527 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
529 * The first mb is necessary to safely close the critical section opened by the
530 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
531 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
532 * parallel wait_on_page_locked()).
534 void fastcall
unlock_page(struct page
*page
)
536 smp_mb__before_clear_bit();
537 if (!TestClearPageLocked(page
))
539 smp_mb__after_clear_bit();
540 wake_up_page(page
, PG_locked
);
542 EXPORT_SYMBOL(unlock_page
);
545 * end_page_writeback - end writeback against a page
548 void end_page_writeback(struct page
*page
)
550 if (!TestClearPageReclaim(page
) || rotate_reclaimable_page(page
)) {
551 if (!test_clear_page_writeback(page
))
554 smp_mb__after_clear_bit();
555 wake_up_page(page
, PG_writeback
);
557 EXPORT_SYMBOL(end_page_writeback
);
560 * __lock_page - get a lock on the page, assuming we need to sleep to get it
561 * @page: the page to lock
563 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
564 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
565 * chances are that on the second loop, the block layer's plug list is empty,
566 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
568 void fastcall
__lock_page(struct page
*page
)
570 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
572 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sync_page
,
573 TASK_UNINTERRUPTIBLE
);
575 EXPORT_SYMBOL(__lock_page
);
578 * Variant of lock_page that does not require the caller to hold a reference
579 * on the page's mapping.
581 void fastcall
__lock_page_nosync(struct page
*page
)
583 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
584 __wait_on_bit_lock(page_waitqueue(page
), &wait
, __sleep_on_page_lock
,
585 TASK_UNINTERRUPTIBLE
);
589 * find_get_page - find and get a page reference
590 * @mapping: the address_space to search
591 * @offset: the page index
593 * Is there a pagecache struct page at the given (mapping, offset) tuple?
594 * If yes, increment its refcount and return it; if no, return NULL.
596 struct page
* find_get_page(struct address_space
*mapping
, unsigned long offset
)
600 read_lock_irq(&mapping
->tree_lock
);
601 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
603 page_cache_get(page
);
604 read_unlock_irq(&mapping
->tree_lock
);
607 EXPORT_SYMBOL(find_get_page
);
610 * find_lock_page - locate, pin and lock a pagecache page
611 * @mapping: the address_space to search
612 * @offset: the page index
614 * Locates the desired pagecache page, locks it, increments its reference
615 * count and returns its address.
617 * Returns zero if the page was not present. find_lock_page() may sleep.
619 struct page
*find_lock_page(struct address_space
*mapping
,
620 unsigned long offset
)
624 read_lock_irq(&mapping
->tree_lock
);
626 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
628 page_cache_get(page
);
629 if (TestSetPageLocked(page
)) {
630 read_unlock_irq(&mapping
->tree_lock
);
632 read_lock_irq(&mapping
->tree_lock
);
634 /* Has the page been truncated while we slept? */
635 if (unlikely(page
->mapping
!= mapping
||
636 page
->index
!= offset
)) {
638 page_cache_release(page
);
643 read_unlock_irq(&mapping
->tree_lock
);
646 EXPORT_SYMBOL(find_lock_page
);
649 * find_or_create_page - locate or add a pagecache page
650 * @mapping: the page's address_space
651 * @index: the page's index into the mapping
652 * @gfp_mask: page allocation mode
654 * Locates a page in the pagecache. If the page is not present, a new page
655 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
656 * LRU list. The returned page is locked and has its reference count
659 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
662 * find_or_create_page() returns the desired page's address, or zero on
665 struct page
*find_or_create_page(struct address_space
*mapping
,
666 unsigned long index
, gfp_t gfp_mask
)
668 struct page
*page
, *cached_page
= NULL
;
671 page
= find_lock_page(mapping
, index
);
675 __page_cache_alloc(gfp_mask
);
679 err
= add_to_page_cache_lru(cached_page
, mapping
,
684 } else if (err
== -EEXIST
)
688 page_cache_release(cached_page
);
691 EXPORT_SYMBOL(find_or_create_page
);
694 * find_get_pages - gang pagecache lookup
695 * @mapping: The address_space to search
696 * @start: The starting page index
697 * @nr_pages: The maximum number of pages
698 * @pages: Where the resulting pages are placed
700 * find_get_pages() will search for and return a group of up to
701 * @nr_pages pages in the mapping. The pages are placed at @pages.
702 * find_get_pages() takes a reference against the returned pages.
704 * The search returns a group of mapping-contiguous pages with ascending
705 * indexes. There may be holes in the indices due to not-present pages.
707 * find_get_pages() returns the number of pages which were found.
709 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
710 unsigned int nr_pages
, struct page
**pages
)
715 read_lock_irq(&mapping
->tree_lock
);
716 ret
= radix_tree_gang_lookup(&mapping
->page_tree
,
717 (void **)pages
, start
, nr_pages
);
718 for (i
= 0; i
< ret
; i
++)
719 page_cache_get(pages
[i
]);
720 read_unlock_irq(&mapping
->tree_lock
);
725 * find_get_pages_contig - gang contiguous pagecache lookup
726 * @mapping: The address_space to search
727 * @index: The starting page index
728 * @nr_pages: The maximum number of pages
729 * @pages: Where the resulting pages are placed
731 * find_get_pages_contig() works exactly like find_get_pages(), except
732 * that the returned number of pages are guaranteed to be contiguous.
734 * find_get_pages_contig() returns the number of pages which were found.
736 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
737 unsigned int nr_pages
, struct page
**pages
)
742 read_lock_irq(&mapping
->tree_lock
);
743 ret
= radix_tree_gang_lookup(&mapping
->page_tree
,
744 (void **)pages
, index
, nr_pages
);
745 for (i
= 0; i
< ret
; i
++) {
746 if (pages
[i
]->mapping
== NULL
|| pages
[i
]->index
!= index
)
749 page_cache_get(pages
[i
]);
752 read_unlock_irq(&mapping
->tree_lock
);
755 EXPORT_SYMBOL(find_get_pages_contig
);
758 * find_get_pages_tag - find and return pages that match @tag
759 * @mapping: the address_space to search
760 * @index: the starting page index
761 * @tag: the tag index
762 * @nr_pages: the maximum number of pages
763 * @pages: where the resulting pages are placed
765 * Like find_get_pages, except we only return pages which are tagged with
766 * @tag. We update @index to index the next page for the traversal.
768 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
769 int tag
, unsigned int nr_pages
, struct page
**pages
)
774 read_lock_irq(&mapping
->tree_lock
);
775 ret
= radix_tree_gang_lookup_tag(&mapping
->page_tree
,
776 (void **)pages
, *index
, nr_pages
, tag
);
777 for (i
= 0; i
< ret
; i
++)
778 page_cache_get(pages
[i
]);
780 *index
= pages
[ret
- 1]->index
+ 1;
781 read_unlock_irq(&mapping
->tree_lock
);
784 EXPORT_SYMBOL(find_get_pages_tag
);
787 * grab_cache_page_nowait - returns locked page at given index in given cache
788 * @mapping: target address_space
789 * @index: the page index
791 * Same as grab_cache_page(), but do not wait if the page is unavailable.
792 * This is intended for speculative data generators, where the data can
793 * be regenerated if the page couldn't be grabbed. This routine should
794 * be safe to call while holding the lock for another page.
796 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
797 * and deadlock against the caller's locked page.
800 grab_cache_page_nowait(struct address_space
*mapping
, unsigned long index
)
802 struct page
*page
= find_get_page(mapping
, index
);
805 if (!TestSetPageLocked(page
))
807 page_cache_release(page
);
810 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
811 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_KERNEL
)) {
812 page_cache_release(page
);
817 EXPORT_SYMBOL(grab_cache_page_nowait
);
820 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
821 * a _large_ part of the i/o request. Imagine the worst scenario:
823 * ---R__________________________________________B__________
824 * ^ reading here ^ bad block(assume 4k)
826 * read(R) => miss => readahead(R...B) => media error => frustrating retries
827 * => failing the whole request => read(R) => read(R+1) =>
828 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
829 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
830 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
832 * It is going insane. Fix it by quickly scaling down the readahead size.
834 static void shrink_readahead_size_eio(struct file
*filp
,
835 struct file_ra_state
*ra
)
844 * do_generic_mapping_read - generic file read routine
845 * @mapping: address_space to be read
846 * @_ra: file's readahead state
847 * @filp: the file to read
848 * @ppos: current file position
849 * @desc: read_descriptor
850 * @actor: read method
852 * This is a generic file read routine, and uses the
853 * mapping->a_ops->readpage() function for the actual low-level stuff.
855 * This is really ugly. But the goto's actually try to clarify some
856 * of the logic when it comes to error handling etc.
858 * Note the struct file* is only passed for the use of readpage.
861 void do_generic_mapping_read(struct address_space
*mapping
,
862 struct file_ra_state
*_ra
,
865 read_descriptor_t
*desc
,
868 struct inode
*inode
= mapping
->host
;
870 unsigned long end_index
;
871 unsigned long offset
;
872 unsigned long last_index
;
873 unsigned long next_index
;
874 unsigned long prev_index
;
875 unsigned int prev_offset
;
877 struct page
*cached_page
;
879 struct file_ra_state ra
= *_ra
;
882 index
= *ppos
>> PAGE_CACHE_SHIFT
;
884 prev_index
= ra
.prev_index
;
885 prev_offset
= ra
.prev_offset
;
886 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
887 offset
= *ppos
& ~PAGE_CACHE_MASK
;
889 isize
= i_size_read(inode
);
893 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
896 unsigned long nr
, ret
;
898 /* nr is the maximum number of bytes to copy from this page */
899 nr
= PAGE_CACHE_SIZE
;
900 if (index
>= end_index
) {
901 if (index
> end_index
)
903 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
911 if (index
== next_index
)
912 next_index
= page_cache_readahead(mapping
, &ra
, filp
,
913 index
, last_index
- index
);
916 page
= find_get_page(mapping
, index
);
917 if (unlikely(page
== NULL
)) {
918 handle_ra_miss(mapping
, &ra
, index
);
921 if (!PageUptodate(page
))
922 goto page_not_up_to_date
;
925 /* If users can be writing to this page using arbitrary
926 * virtual addresses, take care about potential aliasing
927 * before reading the page on the kernel side.
929 if (mapping_writably_mapped(mapping
))
930 flush_dcache_page(page
);
933 * When a sequential read accesses a page several times,
934 * only mark it as accessed the first time.
936 if (prev_index
!= index
|| offset
!= prev_offset
)
937 mark_page_accessed(page
);
941 * Ok, we have the page, and it's up-to-date, so
942 * now we can copy it to user space...
944 * The actor routine returns how many bytes were actually used..
945 * NOTE! This may not be the same as how much of a user buffer
946 * we filled up (we may be padding etc), so we can only update
947 * "pos" here (the actor routine has to update the user buffer
948 * pointers and the remaining count).
950 ret
= actor(desc
, page
, offset
, nr
);
952 index
+= offset
>> PAGE_CACHE_SHIFT
;
953 offset
&= ~PAGE_CACHE_MASK
;
954 prev_offset
= offset
;
955 ra
.prev_offset
= offset
;
957 page_cache_release(page
);
958 if (ret
== nr
&& desc
->count
)
963 /* Get exclusive access to the page ... */
966 /* Did it get truncated before we got the lock? */
967 if (!page
->mapping
) {
969 page_cache_release(page
);
973 /* Did somebody else fill it already? */
974 if (PageUptodate(page
)) {
980 /* Start the actual read. The read will unlock the page. */
981 error
= mapping
->a_ops
->readpage(filp
, page
);
983 if (unlikely(error
)) {
984 if (error
== AOP_TRUNCATED_PAGE
) {
985 page_cache_release(page
);
991 if (!PageUptodate(page
)) {
993 if (!PageUptodate(page
)) {
994 if (page
->mapping
== NULL
) {
996 * invalidate_inode_pages got it
999 page_cache_release(page
);
1004 shrink_readahead_size_eio(filp
, &ra
);
1005 goto readpage_error
;
1011 * i_size must be checked after we have done ->readpage.
1013 * Checking i_size after the readpage allows us to calculate
1014 * the correct value for "nr", which means the zero-filled
1015 * part of the page is not copied back to userspace (unless
1016 * another truncate extends the file - this is desired though).
1018 isize
= i_size_read(inode
);
1019 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1020 if (unlikely(!isize
|| index
> end_index
)) {
1021 page_cache_release(page
);
1025 /* nr is the maximum number of bytes to copy from this page */
1026 nr
= PAGE_CACHE_SIZE
;
1027 if (index
== end_index
) {
1028 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1030 page_cache_release(page
);
1038 /* UHHUH! A synchronous read error occurred. Report it */
1039 desc
->error
= error
;
1040 page_cache_release(page
);
1045 * Ok, it wasn't cached, so we need to create a new
1049 cached_page
= page_cache_alloc_cold(mapping
);
1051 desc
->error
= -ENOMEM
;
1055 error
= add_to_page_cache_lru(cached_page
, mapping
,
1058 if (error
== -EEXIST
)
1060 desc
->error
= error
;
1071 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1073 page_cache_release(cached_page
);
1075 file_accessed(filp
);
1077 EXPORT_SYMBOL(do_generic_mapping_read
);
1079 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
1080 unsigned long offset
, unsigned long size
)
1083 unsigned long left
, count
= desc
->count
;
1089 * Faults on the destination of a read are common, so do it before
1092 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
1093 kaddr
= kmap_atomic(page
, KM_USER0
);
1094 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
1095 kaddr
+ offset
, size
);
1096 kunmap_atomic(kaddr
, KM_USER0
);
1101 /* Do it the slow way */
1103 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
1108 desc
->error
= -EFAULT
;
1111 desc
->count
= count
- size
;
1112 desc
->written
+= size
;
1113 desc
->arg
.buf
+= size
;
1118 * Performs necessary checks before doing a write
1119 * @iov: io vector request
1120 * @nr_segs: number of segments in the iovec
1121 * @count: number of bytes to write
1122 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1124 * Adjust number of segments and amount of bytes to write (nr_segs should be
1125 * properly initialized first). Returns appropriate error code that caller
1126 * should return or zero in case that write should be allowed.
1128 int generic_segment_checks(const struct iovec
*iov
,
1129 unsigned long *nr_segs
, size_t *count
, int access_flags
)
1133 for (seg
= 0; seg
< *nr_segs
; seg
++) {
1134 const struct iovec
*iv
= &iov
[seg
];
1137 * If any segment has a negative length, or the cumulative
1138 * length ever wraps negative then return -EINVAL.
1141 if (unlikely((ssize_t
)(cnt
|iv
->iov_len
) < 0))
1143 if (access_ok(access_flags
, iv
->iov_base
, iv
->iov_len
))
1148 cnt
-= iv
->iov_len
; /* This segment is no good */
1154 EXPORT_SYMBOL(generic_segment_checks
);
1157 * generic_file_aio_read - generic filesystem read routine
1158 * @iocb: kernel I/O control block
1159 * @iov: io vector request
1160 * @nr_segs: number of segments in the iovec
1161 * @pos: current file position
1163 * This is the "read()" routine for all filesystems
1164 * that can use the page cache directly.
1167 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1168 unsigned long nr_segs
, loff_t pos
)
1170 struct file
*filp
= iocb
->ki_filp
;
1174 loff_t
*ppos
= &iocb
->ki_pos
;
1177 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1181 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1182 if (filp
->f_flags
& O_DIRECT
) {
1184 struct address_space
*mapping
;
1185 struct inode
*inode
;
1187 mapping
= filp
->f_mapping
;
1188 inode
= mapping
->host
;
1191 goto out
; /* skip atime */
1192 size
= i_size_read(inode
);
1194 retval
= generic_file_direct_IO(READ
, iocb
,
1197 *ppos
= pos
+ retval
;
1199 if (likely(retval
!= 0)) {
1200 file_accessed(filp
);
1207 for (seg
= 0; seg
< nr_segs
; seg
++) {
1208 read_descriptor_t desc
;
1211 desc
.arg
.buf
= iov
[seg
].iov_base
;
1212 desc
.count
= iov
[seg
].iov_len
;
1213 if (desc
.count
== 0)
1216 do_generic_file_read(filp
,ppos
,&desc
,file_read_actor
);
1217 retval
+= desc
.written
;
1219 retval
= retval
?: desc
.error
;
1229 EXPORT_SYMBOL(generic_file_aio_read
);
1231 int file_send_actor(read_descriptor_t
* desc
, struct page
*page
, unsigned long offset
, unsigned long size
)
1234 unsigned long count
= desc
->count
;
1235 struct file
*file
= desc
->arg
.data
;
1240 written
= file
->f_op
->sendpage(file
, page
, offset
,
1241 size
, &file
->f_pos
, size
<count
);
1243 desc
->error
= written
;
1246 desc
->count
= count
- written
;
1247 desc
->written
+= written
;
1252 do_readahead(struct address_space
*mapping
, struct file
*filp
,
1253 unsigned long index
, unsigned long nr
)
1255 if (!mapping
|| !mapping
->a_ops
|| !mapping
->a_ops
->readpage
)
1258 force_page_cache_readahead(mapping
, filp
, index
,
1259 max_sane_readahead(nr
));
1263 asmlinkage ssize_t
sys_readahead(int fd
, loff_t offset
, size_t count
)
1271 if (file
->f_mode
& FMODE_READ
) {
1272 struct address_space
*mapping
= file
->f_mapping
;
1273 unsigned long start
= offset
>> PAGE_CACHE_SHIFT
;
1274 unsigned long end
= (offset
+ count
- 1) >> PAGE_CACHE_SHIFT
;
1275 unsigned long len
= end
- start
+ 1;
1276 ret
= do_readahead(mapping
, file
, start
, len
);
1284 static int FASTCALL(page_cache_read(struct file
* file
, unsigned long offset
));
1286 * page_cache_read - adds requested page to the page cache if not already there
1287 * @file: file to read
1288 * @offset: page index
1290 * This adds the requested page to the page cache if it isn't already there,
1291 * and schedules an I/O to read in its contents from disk.
1293 static int fastcall
page_cache_read(struct file
* file
, unsigned long offset
)
1295 struct address_space
*mapping
= file
->f_mapping
;
1300 page
= page_cache_alloc_cold(mapping
);
1304 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1306 ret
= mapping
->a_ops
->readpage(file
, page
);
1307 else if (ret
== -EEXIST
)
1308 ret
= 0; /* losing race to add is OK */
1310 page_cache_release(page
);
1312 } while (ret
== AOP_TRUNCATED_PAGE
);
1317 #define MMAP_LOTSAMISS (100)
1320 * filemap_nopage - read in file data for page fault handling
1321 * @area: the applicable vm_area
1322 * @address: target address to read in
1323 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1325 * filemap_nopage() is invoked via the vma operations vector for a
1326 * mapped memory region to read in file data during a page fault.
1328 * The goto's are kind of ugly, but this streamlines the normal case of having
1329 * it in the page cache, and handles the special cases reasonably without
1330 * having a lot of duplicated code.
1332 struct page
*filemap_nopage(struct vm_area_struct
*area
,
1333 unsigned long address
, int *type
)
1336 struct file
*file
= area
->vm_file
;
1337 struct address_space
*mapping
= file
->f_mapping
;
1338 struct file_ra_state
*ra
= &file
->f_ra
;
1339 struct inode
*inode
= mapping
->host
;
1341 unsigned long size
, pgoff
;
1342 int did_readaround
= 0, majmin
= VM_FAULT_MINOR
;
1344 pgoff
= ((address
-area
->vm_start
) >> PAGE_CACHE_SHIFT
) + area
->vm_pgoff
;
1347 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1349 goto outside_data_content
;
1351 /* If we don't want any read-ahead, don't bother */
1352 if (VM_RandomReadHint(area
))
1353 goto no_cached_page
;
1356 * The readahead code wants to be told about each and every page
1357 * so it can build and shrink its windows appropriately
1359 * For sequential accesses, we use the generic readahead logic.
1361 if (VM_SequentialReadHint(area
))
1362 page_cache_readahead(mapping
, ra
, file
, pgoff
, 1);
1365 * Do we have something in the page cache already?
1368 page
= find_get_page(mapping
, pgoff
);
1370 unsigned long ra_pages
;
1372 if (VM_SequentialReadHint(area
)) {
1373 handle_ra_miss(mapping
, ra
, pgoff
);
1374 goto no_cached_page
;
1379 * Do we miss much more than hit in this file? If so,
1380 * stop bothering with read-ahead. It will only hurt.
1382 if (ra
->mmap_miss
> ra
->mmap_hit
+ MMAP_LOTSAMISS
)
1383 goto no_cached_page
;
1386 * To keep the pgmajfault counter straight, we need to
1387 * check did_readaround, as this is an inner loop.
1389 if (!did_readaround
) {
1390 majmin
= VM_FAULT_MAJOR
;
1391 count_vm_event(PGMAJFAULT
);
1394 ra_pages
= max_sane_readahead(file
->f_ra
.ra_pages
);
1398 if (pgoff
> ra_pages
/ 2)
1399 start
= pgoff
- ra_pages
/ 2;
1400 do_page_cache_readahead(mapping
, file
, start
, ra_pages
);
1402 page
= find_get_page(mapping
, pgoff
);
1404 goto no_cached_page
;
1407 if (!did_readaround
)
1411 * Ok, found a page in the page cache, now we need to check
1412 * that it's up-to-date.
1414 if (!PageUptodate(page
))
1415 goto page_not_uptodate
;
1419 * Found the page and have a reference on it.
1421 mark_page_accessed(page
);
1426 outside_data_content
:
1428 * An external ptracer can access pages that normally aren't
1431 if (area
->vm_mm
== current
->mm
)
1432 return NOPAGE_SIGBUS
;
1433 /* Fall through to the non-read-ahead case */
1436 * We're only likely to ever get here if MADV_RANDOM is in
1439 error
= page_cache_read(file
, pgoff
);
1442 * The page we want has now been added to the page cache.
1443 * In the unlikely event that someone removed it in the
1444 * meantime, we'll just come back here and read it again.
1450 * An error return from page_cache_read can result if the
1451 * system is low on memory, or a problem occurs while trying
1454 if (error
== -ENOMEM
)
1456 return NOPAGE_SIGBUS
;
1459 if (!did_readaround
) {
1460 majmin
= VM_FAULT_MAJOR
;
1461 count_vm_event(PGMAJFAULT
);
1465 * Umm, take care of errors if the page isn't up-to-date.
1466 * Try to re-read it _once_. We do this synchronously,
1467 * because there really aren't any performance issues here
1468 * and we need to check for errors.
1472 /* Somebody truncated the page on us? */
1473 if (!page
->mapping
) {
1475 page_cache_release(page
);
1479 /* Somebody else successfully read it in? */
1480 if (PageUptodate(page
)) {
1484 ClearPageError(page
);
1485 error
= mapping
->a_ops
->readpage(file
, page
);
1487 wait_on_page_locked(page
);
1488 if (PageUptodate(page
))
1490 } else if (error
== AOP_TRUNCATED_PAGE
) {
1491 page_cache_release(page
);
1496 * Things didn't work out. Return zero to tell the
1497 * mm layer so, possibly freeing the page cache page first.
1499 shrink_readahead_size_eio(file
, ra
);
1500 page_cache_release(page
);
1501 return NOPAGE_SIGBUS
;
1503 EXPORT_SYMBOL(filemap_nopage
);
1505 static struct page
* filemap_getpage(struct file
*file
, unsigned long pgoff
,
1508 struct address_space
*mapping
= file
->f_mapping
;
1513 * Do we have something in the page cache already?
1516 page
= find_get_page(mapping
, pgoff
);
1520 goto no_cached_page
;
1524 * Ok, found a page in the page cache, now we need to check
1525 * that it's up-to-date.
1527 if (!PageUptodate(page
)) {
1529 page_cache_release(page
);
1532 goto page_not_uptodate
;
1537 * Found the page and have a reference on it.
1539 mark_page_accessed(page
);
1543 error
= page_cache_read(file
, pgoff
);
1546 * The page we want has now been added to the page cache.
1547 * In the unlikely event that someone removed it in the
1548 * meantime, we'll just come back here and read it again.
1554 * An error return from page_cache_read can result if the
1555 * system is low on memory, or a problem occurs while trying
1563 /* Did it get truncated while we waited for it? */
1564 if (!page
->mapping
) {
1569 /* Did somebody else get it up-to-date? */
1570 if (PageUptodate(page
)) {
1575 error
= mapping
->a_ops
->readpage(file
, page
);
1577 wait_on_page_locked(page
);
1578 if (PageUptodate(page
))
1580 } else if (error
== AOP_TRUNCATED_PAGE
) {
1581 page_cache_release(page
);
1586 * Umm, take care of errors if the page isn't up-to-date.
1587 * Try to re-read it _once_. We do this synchronously,
1588 * because there really aren't any performance issues here
1589 * and we need to check for errors.
1593 /* Somebody truncated the page on us? */
1594 if (!page
->mapping
) {
1598 /* Somebody else successfully read it in? */
1599 if (PageUptodate(page
)) {
1604 ClearPageError(page
);
1605 error
= mapping
->a_ops
->readpage(file
, page
);
1607 wait_on_page_locked(page
);
1608 if (PageUptodate(page
))
1610 } else if (error
== AOP_TRUNCATED_PAGE
) {
1611 page_cache_release(page
);
1616 * Things didn't work out. Return zero to tell the
1617 * mm layer so, possibly freeing the page cache page first.
1620 page_cache_release(page
);
1625 int filemap_populate(struct vm_area_struct
*vma
, unsigned long addr
,
1626 unsigned long len
, pgprot_t prot
, unsigned long pgoff
,
1629 struct file
*file
= vma
->vm_file
;
1630 struct address_space
*mapping
= file
->f_mapping
;
1631 struct inode
*inode
= mapping
->host
;
1633 struct mm_struct
*mm
= vma
->vm_mm
;
1638 force_page_cache_readahead(mapping
, vma
->vm_file
,
1639 pgoff
, len
>> PAGE_CACHE_SHIFT
);
1642 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1643 if (pgoff
+ (len
>> PAGE_CACHE_SHIFT
) > size
)
1646 page
= filemap_getpage(file
, pgoff
, nonblock
);
1648 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1649 * done in shmem_populate calling shmem_getpage */
1650 if (!page
&& !nonblock
)
1654 err
= install_page(mm
, vma
, addr
, page
, prot
);
1656 page_cache_release(page
);
1659 } else if (vma
->vm_flags
& VM_NONLINEAR
) {
1660 /* No page was found just because we can't read it in now (being
1661 * here implies nonblock != 0), but the page may exist, so set
1662 * the PTE to fault it in later. */
1663 err
= install_file_pte(mm
, vma
, addr
, pgoff
, prot
);
1676 EXPORT_SYMBOL(filemap_populate
);
1678 struct vm_operations_struct generic_file_vm_ops
= {
1679 .nopage
= filemap_nopage
,
1680 .populate
= filemap_populate
,
1683 /* This is used for a general mmap of a disk file */
1685 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1687 struct address_space
*mapping
= file
->f_mapping
;
1689 if (!mapping
->a_ops
->readpage
)
1691 file_accessed(file
);
1692 vma
->vm_ops
= &generic_file_vm_ops
;
1697 * This is for filesystems which do not implement ->writepage.
1699 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1701 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
1703 return generic_file_mmap(file
, vma
);
1706 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1710 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1714 #endif /* CONFIG_MMU */
1716 EXPORT_SYMBOL(generic_file_mmap
);
1717 EXPORT_SYMBOL(generic_file_readonly_mmap
);
1719 static struct page
*__read_cache_page(struct address_space
*mapping
,
1720 unsigned long index
,
1721 int (*filler
)(void *,struct page
*),
1724 struct page
*page
, *cached_page
= NULL
;
1727 page
= find_get_page(mapping
, index
);
1730 cached_page
= page_cache_alloc_cold(mapping
);
1732 return ERR_PTR(-ENOMEM
);
1734 err
= add_to_page_cache_lru(cached_page
, mapping
,
1739 /* Presumably ENOMEM for radix tree node */
1740 page_cache_release(cached_page
);
1741 return ERR_PTR(err
);
1745 err
= filler(data
, page
);
1747 page_cache_release(page
);
1748 page
= ERR_PTR(err
);
1752 page_cache_release(cached_page
);
1757 * Same as read_cache_page, but don't wait for page to become unlocked
1758 * after submitting it to the filler.
1760 struct page
*read_cache_page_async(struct address_space
*mapping
,
1761 unsigned long index
,
1762 int (*filler
)(void *,struct page
*),
1769 page
= __read_cache_page(mapping
, index
, filler
, data
);
1772 if (PageUptodate(page
))
1776 if (!page
->mapping
) {
1778 page_cache_release(page
);
1781 if (PageUptodate(page
)) {
1785 err
= filler(data
, page
);
1787 page_cache_release(page
);
1788 return ERR_PTR(err
);
1791 mark_page_accessed(page
);
1794 EXPORT_SYMBOL(read_cache_page_async
);
1797 * read_cache_page - read into page cache, fill it if needed
1798 * @mapping: the page's address_space
1799 * @index: the page index
1800 * @filler: function to perform the read
1801 * @data: destination for read data
1803 * Read into the page cache. If a page already exists, and PageUptodate() is
1804 * not set, try to fill the page then wait for it to become unlocked.
1806 * If the page does not get brought uptodate, return -EIO.
1808 struct page
*read_cache_page(struct address_space
*mapping
,
1809 unsigned long index
,
1810 int (*filler
)(void *,struct page
*),
1815 page
= read_cache_page_async(mapping
, index
, filler
, data
);
1818 wait_on_page_locked(page
);
1819 if (!PageUptodate(page
)) {
1820 page_cache_release(page
);
1821 page
= ERR_PTR(-EIO
);
1826 EXPORT_SYMBOL(read_cache_page
);
1829 * If the page was newly created, increment its refcount and add it to the
1830 * caller's lru-buffering pagevec. This function is specifically for
1831 * generic_file_write().
1833 static inline struct page
*
1834 __grab_cache_page(struct address_space
*mapping
, unsigned long index
,
1835 struct page
**cached_page
, struct pagevec
*lru_pvec
)
1840 page
= find_lock_page(mapping
, index
);
1842 if (!*cached_page
) {
1843 *cached_page
= page_cache_alloc(mapping
);
1847 err
= add_to_page_cache(*cached_page
, mapping
,
1852 page
= *cached_page
;
1853 page_cache_get(page
);
1854 if (!pagevec_add(lru_pvec
, page
))
1855 __pagevec_lru_add(lru_pvec
);
1856 *cached_page
= NULL
;
1863 * The logic we want is
1865 * if suid or (sgid and xgrp)
1868 int should_remove_suid(struct dentry
*dentry
)
1870 mode_t mode
= dentry
->d_inode
->i_mode
;
1873 /* suid always must be killed */
1874 if (unlikely(mode
& S_ISUID
))
1875 kill
= ATTR_KILL_SUID
;
1878 * sgid without any exec bits is just a mandatory locking mark; leave
1879 * it alone. If some exec bits are set, it's a real sgid; kill it.
1881 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1882 kill
|= ATTR_KILL_SGID
;
1884 if (unlikely(kill
&& !capable(CAP_FSETID
)))
1889 EXPORT_SYMBOL(should_remove_suid
);
1891 int __remove_suid(struct dentry
*dentry
, int kill
)
1893 struct iattr newattrs
;
1895 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1896 return notify_change(dentry
, &newattrs
);
1899 int remove_suid(struct dentry
*dentry
)
1901 int kill
= should_remove_suid(dentry
);
1904 return __remove_suid(dentry
, kill
);
1908 EXPORT_SYMBOL(remove_suid
);
1911 __filemap_copy_from_user_iovec_inatomic(char *vaddr
,
1912 const struct iovec
*iov
, size_t base
, size_t bytes
)
1914 size_t copied
= 0, left
= 0;
1917 char __user
*buf
= iov
->iov_base
+ base
;
1918 int copy
= min(bytes
, iov
->iov_len
- base
);
1921 left
= __copy_from_user_inatomic_nocache(vaddr
, buf
, copy
);
1930 return copied
- left
;
1934 * Performs necessary checks before doing a write
1936 * Can adjust writing position or amount of bytes to write.
1937 * Returns appropriate error code that caller should return or
1938 * zero in case that write should be allowed.
1940 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
1942 struct inode
*inode
= file
->f_mapping
->host
;
1943 unsigned long limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1945 if (unlikely(*pos
< 0))
1949 /* FIXME: this is for backwards compatibility with 2.4 */
1950 if (file
->f_flags
& O_APPEND
)
1951 *pos
= i_size_read(inode
);
1953 if (limit
!= RLIM_INFINITY
) {
1954 if (*pos
>= limit
) {
1955 send_sig(SIGXFSZ
, current
, 0);
1958 if (*count
> limit
- (typeof(limit
))*pos
) {
1959 *count
= limit
- (typeof(limit
))*pos
;
1967 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
1968 !(file
->f_flags
& O_LARGEFILE
))) {
1969 if (*pos
>= MAX_NON_LFS
) {
1972 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
1973 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
1978 * Are we about to exceed the fs block limit ?
1980 * If we have written data it becomes a short write. If we have
1981 * exceeded without writing data we send a signal and return EFBIG.
1982 * Linus frestrict idea will clean these up nicely..
1984 if (likely(!isblk
)) {
1985 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
1986 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
1989 /* zero-length writes at ->s_maxbytes are OK */
1992 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
1993 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
1997 if (bdev_read_only(I_BDEV(inode
)))
1999 isize
= i_size_read(inode
);
2000 if (*pos
>= isize
) {
2001 if (*count
|| *pos
> isize
)
2005 if (*pos
+ *count
> isize
)
2006 *count
= isize
- *pos
;
2013 EXPORT_SYMBOL(generic_write_checks
);
2016 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2017 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2018 size_t count
, size_t ocount
)
2020 struct file
*file
= iocb
->ki_filp
;
2021 struct address_space
*mapping
= file
->f_mapping
;
2022 struct inode
*inode
= mapping
->host
;
2025 if (count
!= ocount
)
2026 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2028 written
= generic_file_direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2030 loff_t end
= pos
+ written
;
2031 if (end
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2032 i_size_write(inode
, end
);
2033 mark_inode_dirty(inode
);
2039 * Sync the fs metadata but not the minor inode changes and
2040 * of course not the data as we did direct DMA for the IO.
2041 * i_mutex is held, which protects generic_osync_inode() from
2042 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
2044 if ((written
>= 0 || written
== -EIOCBQUEUED
) &&
2045 ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2046 int err
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
2052 EXPORT_SYMBOL(generic_file_direct_write
);
2055 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2056 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2057 size_t count
, ssize_t written
)
2059 struct file
*file
= iocb
->ki_filp
;
2060 struct address_space
* mapping
= file
->f_mapping
;
2061 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2062 struct inode
*inode
= mapping
->host
;
2065 struct page
*cached_page
= NULL
;
2067 struct pagevec lru_pvec
;
2068 const struct iovec
*cur_iov
= iov
; /* current iovec */
2069 size_t iov_base
= 0; /* offset in the current iovec */
2072 pagevec_init(&lru_pvec
, 0);
2075 * handle partial DIO write. Adjust cur_iov if needed.
2077 if (likely(nr_segs
== 1))
2078 buf
= iov
->iov_base
+ written
;
2080 filemap_set_next_iovec(&cur_iov
, &iov_base
, written
);
2081 buf
= cur_iov
->iov_base
+ iov_base
;
2085 unsigned long index
;
2086 unsigned long offset
;
2089 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
2090 index
= pos
>> PAGE_CACHE_SHIFT
;
2091 bytes
= PAGE_CACHE_SIZE
- offset
;
2093 /* Limit the size of the copy to the caller's write size */
2094 bytes
= min(bytes
, count
);
2096 /* We only need to worry about prefaulting when writes are from
2097 * user-space. NFSd uses vfs_writev with several non-aligned
2098 * segments in the vector, and limiting to one segment a time is
2099 * a noticeable performance for re-write
2101 if (!segment_eq(get_fs(), KERNEL_DS
)) {
2103 * Limit the size of the copy to that of the current
2104 * segment, because fault_in_pages_readable() doesn't
2105 * know how to walk segments.
2107 bytes
= min(bytes
, cur_iov
->iov_len
- iov_base
);
2110 * Bring in the user page that we will copy from
2111 * _first_. Otherwise there's a nasty deadlock on
2112 * copying from the same page as we're writing to,
2113 * without it being marked up-to-date.
2115 fault_in_pages_readable(buf
, bytes
);
2117 page
= __grab_cache_page(mapping
,index
,&cached_page
,&lru_pvec
);
2123 if (unlikely(bytes
== 0)) {
2126 goto zero_length_segment
;
2129 status
= a_ops
->prepare_write(file
, page
, offset
, offset
+bytes
);
2130 if (unlikely(status
)) {
2131 loff_t isize
= i_size_read(inode
);
2133 if (status
!= AOP_TRUNCATED_PAGE
)
2135 page_cache_release(page
);
2136 if (status
== AOP_TRUNCATED_PAGE
)
2139 * prepare_write() may have instantiated a few blocks
2140 * outside i_size. Trim these off again.
2142 if (pos
+ bytes
> isize
)
2143 vmtruncate(inode
, isize
);
2146 if (likely(nr_segs
== 1))
2147 copied
= filemap_copy_from_user(page
, offset
,
2150 copied
= filemap_copy_from_user_iovec(page
, offset
,
2151 cur_iov
, iov_base
, bytes
);
2152 flush_dcache_page(page
);
2153 status
= a_ops
->commit_write(file
, page
, offset
, offset
+bytes
);
2154 if (status
== AOP_TRUNCATED_PAGE
) {
2155 page_cache_release(page
);
2158 zero_length_segment
:
2159 if (likely(copied
>= 0)) {
2168 if (unlikely(nr_segs
> 1)) {
2169 filemap_set_next_iovec(&cur_iov
,
2172 buf
= cur_iov
->iov_base
+
2179 if (unlikely(copied
!= bytes
))
2183 mark_page_accessed(page
);
2184 page_cache_release(page
);
2187 balance_dirty_pages_ratelimited(mapping
);
2193 page_cache_release(cached_page
);
2196 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2198 if (likely(status
>= 0)) {
2199 if (unlikely((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2200 if (!a_ops
->writepage
|| !is_sync_kiocb(iocb
))
2201 status
= generic_osync_inode(inode
, mapping
,
2202 OSYNC_METADATA
|OSYNC_DATA
);
2207 * If we get here for O_DIRECT writes then we must have fallen through
2208 * to buffered writes (block instantiation inside i_size). So we sync
2209 * the file data here, to try to honour O_DIRECT expectations.
2211 if (unlikely(file
->f_flags
& O_DIRECT
) && written
)
2212 status
= filemap_write_and_wait(mapping
);
2214 pagevec_lru_add(&lru_pvec
);
2215 return written
? written
: status
;
2217 EXPORT_SYMBOL(generic_file_buffered_write
);
2220 __generic_file_aio_write_nolock(struct kiocb
*iocb
, const struct iovec
*iov
,
2221 unsigned long nr_segs
, loff_t
*ppos
)
2223 struct file
*file
= iocb
->ki_filp
;
2224 struct address_space
* mapping
= file
->f_mapping
;
2225 size_t ocount
; /* original count */
2226 size_t count
; /* after file limit checks */
2227 struct inode
*inode
= mapping
->host
;
2233 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
2240 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2242 /* We can write back this queue in page reclaim */
2243 current
->backing_dev_info
= mapping
->backing_dev_info
;
2246 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2253 err
= remove_suid(file
->f_path
.dentry
);
2257 file_update_time(file
);
2259 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2260 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2262 ssize_t written_buffered
;
2264 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2265 ppos
, count
, ocount
);
2266 if (written
< 0 || written
== count
)
2269 * direct-io write to a hole: fall through to buffered I/O
2270 * for completing the rest of the request.
2274 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2275 nr_segs
, pos
, ppos
, count
,
2278 * If generic_file_buffered_write() retuned a synchronous error
2279 * then we want to return the number of bytes which were
2280 * direct-written, or the error code if that was zero. Note
2281 * that this differs from normal direct-io semantics, which
2282 * will return -EFOO even if some bytes were written.
2284 if (written_buffered
< 0) {
2285 err
= written_buffered
;
2290 * We need to ensure that the page cache pages are written to
2291 * disk and invalidated to preserve the expected O_DIRECT
2294 endbyte
= pos
+ written_buffered
- written
- 1;
2295 err
= do_sync_mapping_range(file
->f_mapping
, pos
, endbyte
,
2296 SYNC_FILE_RANGE_WAIT_BEFORE
|
2297 SYNC_FILE_RANGE_WRITE
|
2298 SYNC_FILE_RANGE_WAIT_AFTER
);
2300 written
= written_buffered
;
2301 invalidate_mapping_pages(mapping
,
2302 pos
>> PAGE_CACHE_SHIFT
,
2303 endbyte
>> PAGE_CACHE_SHIFT
);
2306 * We don't know how much we wrote, so just return
2307 * the number of bytes which were direct-written
2311 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2312 pos
, ppos
, count
, written
);
2315 current
->backing_dev_info
= NULL
;
2316 return written
? written
: err
;
2319 ssize_t
generic_file_aio_write_nolock(struct kiocb
*iocb
,
2320 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
2322 struct file
*file
= iocb
->ki_filp
;
2323 struct address_space
*mapping
= file
->f_mapping
;
2324 struct inode
*inode
= mapping
->host
;
2327 BUG_ON(iocb
->ki_pos
!= pos
);
2329 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
,
2332 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2335 err
= sync_page_range_nolock(inode
, mapping
, pos
, ret
);
2341 EXPORT_SYMBOL(generic_file_aio_write_nolock
);
2343 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2344 unsigned long nr_segs
, loff_t pos
)
2346 struct file
*file
= iocb
->ki_filp
;
2347 struct address_space
*mapping
= file
->f_mapping
;
2348 struct inode
*inode
= mapping
->host
;
2351 BUG_ON(iocb
->ki_pos
!= pos
);
2353 mutex_lock(&inode
->i_mutex
);
2354 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
,
2356 mutex_unlock(&inode
->i_mutex
);
2358 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2361 err
= sync_page_range(inode
, mapping
, pos
, ret
);
2367 EXPORT_SYMBOL(generic_file_aio_write
);
2370 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2371 * went wrong during pagecache shootdown.
2374 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
2375 loff_t offset
, unsigned long nr_segs
)
2377 struct file
*file
= iocb
->ki_filp
;
2378 struct address_space
*mapping
= file
->f_mapping
;
2381 pgoff_t end
= 0; /* silence gcc */
2384 * If it's a write, unmap all mmappings of the file up-front. This
2385 * will cause any pte dirty bits to be propagated into the pageframes
2386 * for the subsequent filemap_write_and_wait().
2389 write_len
= iov_length(iov
, nr_segs
);
2390 end
= (offset
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2391 if (mapping_mapped(mapping
))
2392 unmap_mapping_range(mapping
, offset
, write_len
, 0);
2395 retval
= filemap_write_and_wait(mapping
);
2400 * After a write we want buffered reads to be sure to go to disk to get
2401 * the new data. We invalidate clean cached page from the region we're
2402 * about to write. We do this *before* the write so that we can return
2403 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2405 if (rw
== WRITE
&& mapping
->nrpages
) {
2406 retval
= invalidate_inode_pages2_range(mapping
,
2407 offset
>> PAGE_CACHE_SHIFT
, end
);
2412 retval
= mapping
->a_ops
->direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2417 * Finally, try again to invalidate clean pages which might have been
2418 * faulted in by get_user_pages() if the source of the write was an
2419 * mmap()ed region of the file we're writing. That's a pretty crazy
2420 * thing to do, so we don't support it 100%. If this invalidation
2421 * fails and we have -EIOCBQUEUED we ignore the failure.
2423 if (rw
== WRITE
&& mapping
->nrpages
) {
2424 int err
= invalidate_inode_pages2_range(mapping
,
2425 offset
>> PAGE_CACHE_SHIFT
, end
);
2426 if (err
&& retval
>= 0)
2434 * try_to_release_page() - release old fs-specific metadata on a page
2436 * @page: the page which the kernel is trying to free
2437 * @gfp_mask: memory allocation flags (and I/O mode)
2439 * The address_space is to try to release any data against the page
2440 * (presumably at page->private). If the release was successful, return `1'.
2441 * Otherwise return zero.
2443 * The @gfp_mask argument specifies whether I/O may be performed to release
2444 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2446 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2448 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2450 struct address_space
* const mapping
= page
->mapping
;
2452 BUG_ON(!PageLocked(page
));
2453 if (PageWriteback(page
))
2456 if (mapping
&& mapping
->a_ops
->releasepage
)
2457 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2458 return try_to_free_buffers(page
);
2461 EXPORT_SYMBOL(try_to_release_page
);