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 */
43 #ifdef CONFIG_DIRECTIO
45 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
46 loff_t offset
, unsigned long nr_segs
);
49 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
50 loff_t offset
, unsigned long nr_segs
)
57 * Shared mappings implemented 30.11.1994. It's not fully working yet,
60 * Shared mappings now work. 15.8.1995 Bruno.
62 * finished 'unifying' the page and buffer cache and SMP-threaded the
63 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
65 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
71 * ->i_mmap_lock (vmtruncate)
72 * ->private_lock (__free_pte->__set_page_dirty_buffers)
73 * ->swap_lock (exclusive_swap_page, others)
74 * ->mapping->tree_lock
77 * ->i_mmap_lock (truncate->unmap_mapping_range)
81 * ->page_table_lock or pte_lock (various, mainly in memory.c)
82 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
85 * ->lock_page (access_process_vm)
87 * ->i_mutex (generic_file_buffered_write)
88 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
91 * ->i_alloc_sem (various)
94 * ->sb_lock (fs/fs-writeback.c)
95 * ->mapping->tree_lock (__sync_single_inode)
98 * ->anon_vma.lock (vma_adjust)
101 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
103 * ->page_table_lock or pte_lock
104 * ->swap_lock (try_to_unmap_one)
105 * ->private_lock (try_to_unmap_one)
106 * ->tree_lock (try_to_unmap_one)
107 * ->zone.lru_lock (follow_page->mark_page_accessed)
108 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
109 * ->private_lock (page_remove_rmap->set_page_dirty)
110 * ->tree_lock (page_remove_rmap->set_page_dirty)
111 * ->inode_lock (page_remove_rmap->set_page_dirty)
112 * ->inode_lock (zap_pte_range->set_page_dirty)
113 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
116 * ->dcache_lock (proc_pid_lookup)
120 * Remove a page from the page cache and free it. Caller has to make
121 * sure the page is locked and that nobody else uses it - or that usage
122 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
124 void __remove_from_page_cache(struct page
*page
)
126 #if 0 // mask by Victor Yu. 02-12-2007
127 struct address_space
*mapping
= page
->mapping
;
129 struct address_space
*mapping
= page
->u
.xx
.mapping
;
132 radix_tree_delete(&mapping
->page_tree
, page
->index
);
133 #if 0 // mask by Victor Yu. 02-12-2007
134 page
->mapping
= NULL
;
136 page
->u
.xx
.mapping
= NULL
;
139 __dec_zone_page_state(page
, NR_FILE_PAGES
);
142 void remove_from_page_cache(struct page
*page
)
144 #if 0 // mask by Victor Yu. 02-12-2007
145 struct address_space
*mapping
= page
->mapping
;
147 struct address_space
*mapping
= page
->u
.xx
.mapping
;
150 BUG_ON(!PageLocked(page
));
152 write_lock_irq(&mapping
->tree_lock
);
153 __remove_from_page_cache(page
);
154 write_unlock_irq(&mapping
->tree_lock
);
157 static int sync_page(void *word
)
159 struct address_space
*mapping
;
162 page
= container_of((unsigned long *)word
, struct page
, flags
);
165 * page_mapping() is being called without PG_locked held.
166 * Some knowledge of the state and use of the page is used to
167 * reduce the requirements down to a memory barrier.
168 * The danger here is of a stale page_mapping() return value
169 * indicating a struct address_space different from the one it's
170 * associated with when it is associated with one.
171 * After smp_mb(), it's either the correct page_mapping() for
172 * the page, or an old page_mapping() and the page's own
173 * page_mapping() has gone NULL.
174 * The ->sync_page() address_space operation must tolerate
175 * page_mapping() going NULL. By an amazing coincidence,
176 * this comes about because none of the users of the page
177 * in the ->sync_page() methods make essential use of the
178 * page_mapping(), merely passing the page down to the backing
179 * device's unplug functions when it's non-NULL, which in turn
180 * ignore it for all cases but swap, where only page_private(page) is
181 * of interest. When page_mapping() does go NULL, the entire
182 * call stack gracefully ignores the page and returns.
186 mapping
= page_mapping(page
);
187 if (mapping
&& mapping
->a_ops
&& mapping
->a_ops
->sync_page
)
188 mapping
->a_ops
->sync_page(page
);
194 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
195 * @mapping: address space structure to write
196 * @start: offset in bytes where the range starts
197 * @end: offset in bytes where the range ends (inclusive)
198 * @sync_mode: enable synchronous operation
200 * Start writeback against all of a mapping's dirty pages that lie
201 * within the byte offsets <start, end> inclusive.
203 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
204 * opposed to a regular memory cleansing writeback. The difference between
205 * these two operations is that if a dirty page/buffer is encountered, it must
206 * be waited upon, and not just skipped over.
208 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
209 loff_t end
, int sync_mode
)
212 struct writeback_control wbc
= {
213 .sync_mode
= sync_mode
,
214 .nr_to_write
= mapping
->nrpages
* 2,
215 .range_start
= start
,
219 if (!mapping_cap_writeback_dirty(mapping
))
222 ret
= do_writepages(mapping
, &wbc
);
226 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
229 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
232 int filemap_fdatawrite(struct address_space
*mapping
)
234 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
236 EXPORT_SYMBOL(filemap_fdatawrite
);
238 static int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
241 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
251 int filemap_flush(struct address_space
*mapping
)
253 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
255 EXPORT_SYMBOL(filemap_flush
);
258 * wait_on_page_writeback_range - wait for writeback to complete
259 * @mapping: target address_space
260 * @start: beginning page index
261 * @end: ending page index
263 * Wait for writeback to complete against pages indexed by start->end
266 int wait_on_page_writeback_range(struct address_space
*mapping
,
267 pgoff_t start
, pgoff_t end
)
277 pagevec_init(&pvec
, 0);
279 while ((index
<= end
) &&
280 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
281 PAGECACHE_TAG_WRITEBACK
,
282 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
285 for (i
= 0; i
< nr_pages
; i
++) {
286 struct page
*page
= pvec
.pages
[i
];
288 /* until radix tree lookup accepts end_index */
289 if (page
->index
> end
)
292 wait_on_page_writeback(page
);
296 pagevec_release(&pvec
);
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
303 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
310 * sync_page_range - write and wait on all pages in the passed range
311 * @inode: target inode
312 * @mapping: target address_space
313 * @pos: beginning offset in pages to write
314 * @count: number of bytes to write
316 * Write and wait upon all the pages in the passed range. This is a "data
317 * integrity" operation. It waits upon in-flight writeout before starting and
318 * waiting upon new writeout. If there was an IO error, return it.
320 * We need to re-take i_mutex during the generic_osync_inode list walk because
321 * it is otherwise livelockable.
323 int sync_page_range(struct inode
*inode
, struct address_space
*mapping
,
324 loff_t pos
, loff_t count
)
326 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
327 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
330 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
332 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
334 mutex_lock(&inode
->i_mutex
);
335 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
336 mutex_unlock(&inode
->i_mutex
);
339 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
342 EXPORT_SYMBOL(sync_page_range
);
345 * sync_page_range_nolock
346 * @inode: target inode
347 * @mapping: target address_space
348 * @pos: beginning offset in pages to write
349 * @count: number of bytes to write
351 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
352 * as it forces O_SYNC writers to different parts of the same file
353 * to be serialised right until io completion.
355 int sync_page_range_nolock(struct inode
*inode
, struct address_space
*mapping
,
356 loff_t pos
, loff_t count
)
358 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
359 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
362 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
364 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
366 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
368 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
371 EXPORT_SYMBOL(sync_page_range_nolock
);
374 * filemap_fdatawait - wait for all under-writeback pages to complete
375 * @mapping: address space structure to wait for
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
380 int filemap_fdatawait(struct address_space
*mapping
)
382 loff_t i_size
= i_size_read(mapping
->host
);
387 return wait_on_page_writeback_range(mapping
, 0,
388 (i_size
- 1) >> PAGE_CACHE_SHIFT
);
390 EXPORT_SYMBOL(filemap_fdatawait
);
392 int filemap_write_and_wait(struct address_space
*mapping
)
396 if (mapping
->nrpages
) {
397 err
= filemap_fdatawrite(mapping
);
399 * Even if the above returned error, the pages may be
400 * written partially (e.g. -ENOSPC), so we wait for it.
401 * But the -EIO is special case, it may indicate the worst
402 * thing (e.g. bug) happened, so we avoid waiting for it.
405 int err2
= filemap_fdatawait(mapping
);
412 EXPORT_SYMBOL(filemap_write_and_wait
);
415 * filemap_write_and_wait_range - write out & wait on a file range
416 * @mapping: the address_space for the pages
417 * @lstart: offset in bytes where the range starts
418 * @lend: offset in bytes where the range ends (inclusive)
420 * Write out and wait upon file offsets lstart->lend, inclusive.
422 * Note that `lend' is inclusive (describes the last byte to be written) so
423 * that this function can be used to write to the very end-of-file (end = -1).
425 int filemap_write_and_wait_range(struct address_space
*mapping
,
426 loff_t lstart
, loff_t lend
)
430 if (mapping
->nrpages
) {
431 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
433 /* See comment of filemap_write_and_wait() */
435 int err2
= wait_on_page_writeback_range(mapping
,
436 lstart
>> PAGE_CACHE_SHIFT
,
437 lend
>> PAGE_CACHE_SHIFT
);
446 * add_to_page_cache - add newly allocated pagecache pages
448 * @mapping: the page's address_space
449 * @offset: page index
450 * @gfp_mask: page allocation mode
452 * This function is used to add newly allocated pagecache pages;
453 * the page is new, so we can just run SetPageLocked() against it.
454 * The other page state flags were set by rmqueue().
456 * This function does not add the page to the LRU. The caller must do that.
458 int add_to_page_cache(struct page
*page
, struct address_space
*mapping
,
459 pgoff_t offset
, gfp_t gfp_mask
)
461 int error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
464 write_lock_irq(&mapping
->tree_lock
);
465 error
= radix_tree_insert(&mapping
->page_tree
, offset
, page
);
467 page_cache_get(page
);
469 #if 0 // mask by Victor Yu. 02-12-2007
470 page
->mapping
= mapping
;
472 page
->u
.xx
.mapping
= mapping
;
474 page
->index
= offset
;
476 __inc_zone_page_state(page
, NR_FILE_PAGES
);
478 write_unlock_irq(&mapping
->tree_lock
);
479 radix_tree_preload_end();
483 EXPORT_SYMBOL(add_to_page_cache
);
485 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
486 pgoff_t offset
, gfp_t gfp_mask
)
488 int ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
495 struct page
*__page_cache_alloc(gfp_t gfp
)
497 if (cpuset_do_page_mem_spread()) {
498 int n
= cpuset_mem_spread_node();
499 return alloc_pages_node(n
, gfp
, 0);
501 return alloc_pages(gfp
, 0);
503 EXPORT_SYMBOL(__page_cache_alloc
);
506 static int __sleep_on_page_lock(void *word
)
513 * In order to wait for pages to become available there must be
514 * waitqueues associated with pages. By using a hash table of
515 * waitqueues where the bucket discipline is to maintain all
516 * waiters on the same queue and wake all when any of the pages
517 * become available, and for the woken contexts to check to be
518 * sure the appropriate page became available, this saves space
519 * at a cost of "thundering herd" phenomena during rare hash
522 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
524 const struct zone
*zone
= page_zone(page
);
526 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
529 static inline void wake_up_page(struct page
*page
, int bit
)
531 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
534 void fastcall
wait_on_page_bit(struct page
*page
, int bit_nr
)
536 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
538 if (test_bit(bit_nr
, &page
->flags
))
539 __wait_on_bit(page_waitqueue(page
), &wait
, sync_page
,
540 TASK_UNINTERRUPTIBLE
);
542 EXPORT_SYMBOL(wait_on_page_bit
);
545 * unlock_page - unlock a locked page
548 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
549 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
550 * mechananism between PageLocked pages and PageWriteback pages is shared.
551 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
553 * The first mb is necessary to safely close the critical section opened by the
554 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
555 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
556 * parallel wait_on_page_locked()).
558 void fastcall
unlock_page(struct page
*page
)
560 smp_mb__before_clear_bit();
561 if (!TestClearPageLocked(page
))
563 smp_mb__after_clear_bit();
564 wake_up_page(page
, PG_locked
);
566 EXPORT_SYMBOL(unlock_page
);
569 * end_page_writeback - end writeback against a page
572 void end_page_writeback(struct page
*page
)
574 if (!TestClearPageReclaim(page
) || rotate_reclaimable_page(page
)) {
575 if (!test_clear_page_writeback(page
))
578 smp_mb__after_clear_bit();
579 wake_up_page(page
, PG_writeback
);
581 EXPORT_SYMBOL(end_page_writeback
);
584 * __lock_page - get a lock on the page, assuming we need to sleep to get it
585 * @page: the page to lock
587 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
588 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
589 * chances are that on the second loop, the block layer's plug list is empty,
590 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
592 void fastcall
__lock_page(struct page
*page
)
594 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
596 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sync_page
,
597 TASK_UNINTERRUPTIBLE
);
599 EXPORT_SYMBOL(__lock_page
);
602 * Variant of lock_page that does not require the caller to hold a reference
603 * on the page's mapping.
605 void fastcall
__lock_page_nosync(struct page
*page
)
607 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
608 __wait_on_bit_lock(page_waitqueue(page
), &wait
, __sleep_on_page_lock
,
609 TASK_UNINTERRUPTIBLE
);
613 * find_get_page - find and get a page reference
614 * @mapping: the address_space to search
615 * @offset: the page index
617 * Is there a pagecache struct page at the given (mapping, offset) tuple?
618 * If yes, increment its refcount and return it; if no, return NULL.
620 struct page
* find_get_page(struct address_space
*mapping
, unsigned long offset
)
624 read_lock_irq(&mapping
->tree_lock
);
625 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
627 page_cache_get(page
);
628 read_unlock_irq(&mapping
->tree_lock
);
631 EXPORT_SYMBOL(find_get_page
);
634 * find_trylock_page - find and lock a page
635 * @mapping: the address_space to search
636 * @offset: the page index
638 * Same as find_get_page(), but trylock it instead of incrementing the count.
640 struct page
*find_trylock_page(struct address_space
*mapping
, unsigned long offset
)
644 read_lock_irq(&mapping
->tree_lock
);
645 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
646 if (page
&& TestSetPageLocked(page
))
648 read_unlock_irq(&mapping
->tree_lock
);
651 EXPORT_SYMBOL(find_trylock_page
);
654 * find_lock_page - locate, pin and lock a pagecache page
655 * @mapping: the address_space to search
656 * @offset: the page index
658 * Locates the desired pagecache page, locks it, increments its reference
659 * count and returns its address.
661 * Returns zero if the page was not present. find_lock_page() may sleep.
663 struct page
*find_lock_page(struct address_space
*mapping
,
664 unsigned long offset
)
668 read_lock_irq(&mapping
->tree_lock
);
670 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
672 page_cache_get(page
);
673 if (TestSetPageLocked(page
)) {
674 read_unlock_irq(&mapping
->tree_lock
);
676 read_lock_irq(&mapping
->tree_lock
);
678 /* Has the page been truncated while we slept? */
679 #if 0 // mask by Victor Yu. 02-12-2007
680 if (unlikely(page
->mapping
!= mapping
||
681 page
->index
!= offset
)) {
683 if (unlikely(page
->u
.xx
.mapping
!= mapping
||
684 page
->index
!= offset
)) {
687 page_cache_release(page
);
692 read_unlock_irq(&mapping
->tree_lock
);
695 EXPORT_SYMBOL(find_lock_page
);
698 * find_or_create_page - locate or add a pagecache page
699 * @mapping: the page's address_space
700 * @index: the page's index into the mapping
701 * @gfp_mask: page allocation mode
703 * Locates a page in the pagecache. If the page is not present, a new page
704 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
705 * LRU list. The returned page is locked and has its reference count
708 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
711 * find_or_create_page() returns the desired page's address, or zero on
714 struct page
*find_or_create_page(struct address_space
*mapping
,
715 unsigned long index
, gfp_t gfp_mask
)
717 struct page
*page
, *cached_page
= NULL
;
720 page
= find_lock_page(mapping
, index
);
723 cached_page
= alloc_page(gfp_mask
);
727 err
= add_to_page_cache_lru(cached_page
, mapping
,
732 } else if (err
== -EEXIST
)
736 page_cache_release(cached_page
);
739 EXPORT_SYMBOL(find_or_create_page
);
742 * find_get_pages - gang pagecache lookup
743 * @mapping: The address_space to search
744 * @start: The starting page index
745 * @nr_pages: The maximum number of pages
746 * @pages: Where the resulting pages are placed
748 * find_get_pages() will search for and return a group of up to
749 * @nr_pages pages in the mapping. The pages are placed at @pages.
750 * find_get_pages() takes a reference against the returned pages.
752 * The search returns a group of mapping-contiguous pages with ascending
753 * indexes. There may be holes in the indices due to not-present pages.
755 * find_get_pages() returns the number of pages which were found.
757 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
758 unsigned int nr_pages
, struct page
**pages
)
763 read_lock_irq(&mapping
->tree_lock
);
764 ret
= radix_tree_gang_lookup(&mapping
->page_tree
,
765 (void **)pages
, start
, nr_pages
);
766 for (i
= 0; i
< ret
; i
++)
767 page_cache_get(pages
[i
]);
768 read_unlock_irq(&mapping
->tree_lock
);
773 * find_get_pages_contig - gang contiguous pagecache lookup
774 * @mapping: The address_space to search
775 * @index: The starting page index
776 * @nr_pages: The maximum number of pages
777 * @pages: Where the resulting pages are placed
779 * find_get_pages_contig() works exactly like find_get_pages(), except
780 * that the returned number of pages are guaranteed to be contiguous.
782 * find_get_pages_contig() returns the number of pages which were found.
784 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
785 unsigned int nr_pages
, struct page
**pages
)
790 read_lock_irq(&mapping
->tree_lock
);
791 ret
= radix_tree_gang_lookup(&mapping
->page_tree
,
792 (void **)pages
, index
, nr_pages
);
793 for (i
= 0; i
< ret
; i
++) {
794 #if 0 // mask by Victor Yu. 02-12-2007
795 if (pages
[i
]->mapping
== NULL
|| pages
[i
]->index
!= index
)
797 if (pages
[i
]->u
.xx
.mapping
== NULL
|| pages
[i
]->index
!= index
)
801 page_cache_get(pages
[i
]);
804 read_unlock_irq(&mapping
->tree_lock
);
809 * find_get_pages_tag - find and return pages that match @tag
810 * @mapping: the address_space to search
811 * @index: the starting page index
812 * @tag: the tag index
813 * @nr_pages: the maximum number of pages
814 * @pages: where the resulting pages are placed
816 * Like find_get_pages, except we only return pages which are tagged with
817 * @tag. We update @index to index the next page for the traversal.
819 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
820 int tag
, unsigned int nr_pages
, struct page
**pages
)
825 read_lock_irq(&mapping
->tree_lock
);
826 ret
= radix_tree_gang_lookup_tag(&mapping
->page_tree
,
827 (void **)pages
, *index
, nr_pages
, tag
);
828 for (i
= 0; i
< ret
; i
++)
829 page_cache_get(pages
[i
]);
831 *index
= pages
[ret
- 1]->index
+ 1;
832 read_unlock_irq(&mapping
->tree_lock
);
837 * grab_cache_page_nowait - returns locked page at given index in given cache
838 * @mapping: target address_space
839 * @index: the page index
841 * Same as grab_cache_page, but do not wait if the page is unavailable.
842 * This is intended for speculative data generators, where the data can
843 * be regenerated if the page couldn't be grabbed. This routine should
844 * be safe to call while holding the lock for another page.
846 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
847 * and deadlock against the caller's locked page.
850 grab_cache_page_nowait(struct address_space
*mapping
, unsigned long index
)
852 struct page
*page
= find_get_page(mapping
, index
);
855 if (!TestSetPageLocked(page
))
857 page_cache_release(page
);
860 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
861 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_KERNEL
)) {
862 page_cache_release(page
);
867 EXPORT_SYMBOL(grab_cache_page_nowait
);
870 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
871 * a _large_ part of the i/o request. Imagine the worst scenario:
873 * ---R__________________________________________B__________
874 * ^ reading here ^ bad block(assume 4k)
876 * read(R) => miss => readahead(R...B) => media error => frustrating retries
877 * => failing the whole request => read(R) => read(R+1) =>
878 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
879 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
880 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
882 * It is going insane. Fix it by quickly scaling down the readahead size.
884 static void shrink_readahead_size_eio(struct file
*filp
,
885 struct file_ra_state
*ra
)
894 * do_generic_mapping_read - generic file read routine
895 * @mapping: address_space to be read
896 * @_ra: file's readahead state
897 * @filp: the file to read
898 * @ppos: current file position
899 * @desc: read_descriptor
900 * @actor: read method
902 * This is a generic file read routine, and uses the
903 * mapping->a_ops->readpage() function for the actual low-level stuff.
905 * This is really ugly. But the goto's actually try to clarify some
906 * of the logic when it comes to error handling etc.
908 * Note the struct file* is only passed for the use of readpage.
911 void do_generic_mapping_read(struct address_space
*mapping
,
912 struct file_ra_state
*_ra
,
915 read_descriptor_t
*desc
,
918 struct inode
*inode
= mapping
->host
;
920 unsigned long end_index
;
921 unsigned long offset
;
922 unsigned long last_index
;
923 unsigned long next_index
;
924 unsigned long prev_index
;
926 struct page
*cached_page
;
928 struct file_ra_state ra
= *_ra
;
931 index
= *ppos
>> PAGE_CACHE_SHIFT
;
933 prev_index
= ra
.prev_page
;
934 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
935 offset
= *ppos
& ~PAGE_CACHE_MASK
;
937 isize
= i_size_read(inode
);
941 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
944 unsigned long nr
, ret
;
946 /* nr is the maximum number of bytes to copy from this page */
947 nr
= PAGE_CACHE_SIZE
;
948 if (index
>= end_index
) {
949 if (index
> end_index
)
951 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
959 if (index
== next_index
)
960 next_index
= page_cache_readahead(mapping
, &ra
, filp
,
961 index
, last_index
- index
);
964 page
= find_get_page(mapping
, index
);
965 if (unlikely(page
== NULL
)) {
966 handle_ra_miss(mapping
, &ra
, index
);
969 if (!PageUptodate(page
))
970 goto page_not_up_to_date
;
973 /* If users can be writing to this page using arbitrary
974 * virtual addresses, take care about potential aliasing
975 * before reading the page on the kernel side.
977 if (mapping_writably_mapped(mapping
))
978 flush_dcache_page(page
);
981 * When (part of) the same page is read multiple times
982 * in succession, only mark it as accessed the first time.
984 if (prev_index
!= index
)
985 mark_page_accessed(page
);
989 * Ok, we have the page, and it's up-to-date, so
990 * now we can copy it to user space...
992 * The actor routine returns how many bytes were actually used..
993 * NOTE! This may not be the same as how much of a user buffer
994 * we filled up (we may be padding etc), so we can only update
995 * "pos" here (the actor routine has to update the user buffer
996 * pointers and the remaining count).
998 ret
= actor(desc
, page
, offset
, nr
);
1000 index
+= offset
>> PAGE_CACHE_SHIFT
;
1001 offset
&= ~PAGE_CACHE_MASK
;
1003 page_cache_release(page
);
1004 if (ret
== nr
&& desc
->count
)
1008 page_not_up_to_date
:
1009 /* Get exclusive access to the page ... */
1012 /* Did it get truncated before we got the lock? */
1013 #if 0 // mask by Victor Yu. 02-12-2007
1014 if (!page
->mapping
) {
1016 if (!page
->u
.xx
.mapping
) {
1019 page_cache_release(page
);
1023 /* Did somebody else fill it already? */
1024 if (PageUptodate(page
)) {
1030 /* Start the actual read. The read will unlock the page. */
1031 error
= mapping
->a_ops
->readpage(filp
, page
);
1033 if (unlikely(error
)) {
1034 if (error
== AOP_TRUNCATED_PAGE
) {
1035 page_cache_release(page
);
1038 goto readpage_error
;
1041 if (!PageUptodate(page
)) {
1043 if (!PageUptodate(page
)) {
1044 #if 0 // mask by Victor Yu. 02-12-2007
1045 if (page
->mapping
== NULL
) {
1047 if (page
->u
.xx
.mapping
== NULL
) {
1050 * invalidate_inode_pages got it
1053 page_cache_release(page
);
1058 shrink_readahead_size_eio(filp
, &ra
);
1059 goto readpage_error
;
1065 * i_size must be checked after we have done ->readpage.
1067 * Checking i_size after the readpage allows us to calculate
1068 * the correct value for "nr", which means the zero-filled
1069 * part of the page is not copied back to userspace (unless
1070 * another truncate extends the file - this is desired though).
1072 isize
= i_size_read(inode
);
1073 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1074 if (unlikely(!isize
|| index
> end_index
)) {
1075 page_cache_release(page
);
1079 /* nr is the maximum number of bytes to copy from this page */
1080 nr
= PAGE_CACHE_SIZE
;
1081 if (index
== end_index
) {
1082 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1084 page_cache_release(page
);
1092 /* UHHUH! A synchronous read error occurred. Report it */
1093 desc
->error
= error
;
1094 page_cache_release(page
);
1099 * Ok, it wasn't cached, so we need to create a new
1103 cached_page
= page_cache_alloc_cold(mapping
);
1105 desc
->error
= -ENOMEM
;
1109 error
= add_to_page_cache_lru(cached_page
, mapping
,
1112 if (error
== -EEXIST
)
1114 desc
->error
= error
;
1125 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1127 page_cache_release(cached_page
);
1129 file_accessed(filp
);
1131 EXPORT_SYMBOL(do_generic_mapping_read
);
1133 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
1134 unsigned long offset
, unsigned long size
)
1137 unsigned long left
, count
= desc
->count
;
1143 * Faults on the destination of a read are common, so do it before
1146 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
1147 kaddr
= kmap_atomic(page
, KM_USER0
);
1148 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
1149 kaddr
+ offset
, size
);
1150 kunmap_atomic(kaddr
, KM_USER0
);
1155 /* Do it the slow way */
1157 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
1162 desc
->error
= -EFAULT
;
1165 desc
->count
= count
- size
;
1166 desc
->written
+= size
;
1167 desc
->arg
.buf
+= size
;
1172 * generic_file_aio_read - generic filesystem read routine
1173 * @iocb: kernel I/O control block
1174 * @iov: io vector request
1175 * @nr_segs: number of segments in the iovec
1176 * @pos: current file position
1178 * This is the "read()" routine for all filesystems
1179 * that can use the page cache directly.
1182 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1183 unsigned long nr_segs
, loff_t pos
)
1185 struct file
*filp
= iocb
->ki_filp
;
1189 loff_t
*ppos
= &iocb
->ki_pos
;
1192 for (seg
= 0; seg
< nr_segs
; seg
++) {
1193 const struct iovec
*iv
= &iov
[seg
];
1196 * If any segment has a negative length, or the cumulative
1197 * length ever wraps negative then return -EINVAL.
1199 count
+= iv
->iov_len
;
1200 if (unlikely((ssize_t
)(count
|iv
->iov_len
) < 0))
1202 if (access_ok(VERIFY_WRITE
, iv
->iov_base
, iv
->iov_len
))
1207 count
-= iv
->iov_len
; /* This segment is no good */
1211 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1212 if (filp
->f_flags
& O_DIRECT
) {
1214 struct address_space
*mapping
;
1215 struct inode
*inode
;
1217 mapping
= filp
->f_mapping
;
1218 inode
= mapping
->host
;
1221 goto out
; /* skip atime */
1222 size
= i_size_read(inode
);
1224 retval
= generic_file_direct_IO(READ
, iocb
,
1226 if (retval
> 0 && !is_sync_kiocb(iocb
))
1227 retval
= -EIOCBQUEUED
;
1229 *ppos
= pos
+ retval
;
1231 if (likely(retval
!= 0)) {
1232 file_accessed(filp
);
1239 for (seg
= 0; seg
< nr_segs
; seg
++) {
1240 read_descriptor_t desc
;
1243 desc
.arg
.buf
= iov
[seg
].iov_base
;
1244 desc
.count
= iov
[seg
].iov_len
;
1245 if (desc
.count
== 0)
1248 do_generic_file_read(filp
,ppos
,&desc
,file_read_actor
);
1249 retval
+= desc
.written
;
1251 retval
= retval
?: desc
.error
;
1259 EXPORT_SYMBOL(generic_file_aio_read
);
1261 int file_send_actor(read_descriptor_t
* desc
, struct page
*page
, unsigned long offset
, unsigned long size
)
1264 unsigned long count
= desc
->count
;
1265 struct file
*file
= desc
->arg
.data
;
1270 written
= file
->f_op
->sendpage(file
, page
, offset
,
1271 size
, &file
->f_pos
, size
<count
);
1273 desc
->error
= written
;
1276 desc
->count
= count
- written
;
1277 desc
->written
+= written
;
1281 ssize_t
generic_file_sendfile(struct file
*in_file
, loff_t
*ppos
,
1282 size_t count
, read_actor_t actor
, void *target
)
1284 read_descriptor_t desc
;
1291 desc
.arg
.data
= target
;
1294 do_generic_file_read(in_file
, ppos
, &desc
, actor
);
1296 return desc
.written
;
1299 EXPORT_SYMBOL(generic_file_sendfile
);
1302 do_readahead(struct address_space
*mapping
, struct file
*filp
,
1303 unsigned long index
, unsigned long nr
)
1305 if (!mapping
|| !mapping
->a_ops
|| !mapping
->a_ops
->readpage
)
1308 force_page_cache_readahead(mapping
, filp
, index
,
1309 max_sane_readahead(nr
));
1313 asmlinkage ssize_t
sys_readahead(int fd
, loff_t offset
, size_t count
)
1321 if (file
->f_mode
& FMODE_READ
) {
1322 struct address_space
*mapping
= file
->f_mapping
;
1323 unsigned long start
= offset
>> PAGE_CACHE_SHIFT
;
1324 unsigned long end
= (offset
+ count
- 1) >> PAGE_CACHE_SHIFT
;
1325 unsigned long len
= end
- start
+ 1;
1326 ret
= do_readahead(mapping
, file
, start
, len
);
1334 static int FASTCALL(page_cache_read(struct file
* file
, unsigned long offset
));
1336 * page_cache_read - adds requested page to the page cache if not already there
1337 * @file: file to read
1338 * @offset: page index
1340 * This adds the requested page to the page cache if it isn't already there,
1341 * and schedules an I/O to read in its contents from disk.
1343 static int fastcall
page_cache_read(struct file
* file
, unsigned long offset
)
1345 struct address_space
*mapping
= file
->f_mapping
;
1350 page
= page_cache_alloc_cold(mapping
);
1354 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1356 ret
= mapping
->a_ops
->readpage(file
, page
);
1357 else if (ret
== -EEXIST
)
1358 ret
= 0; /* losing race to add is OK */
1360 page_cache_release(page
);
1362 } while (ret
== AOP_TRUNCATED_PAGE
);
1367 #define MMAP_LOTSAMISS (100)
1370 * filemap_nopage - read in file data for page fault handling
1371 * @area: the applicable vm_area
1372 * @address: target address to read in
1373 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1375 * filemap_nopage() is invoked via the vma operations vector for a
1376 * mapped memory region to read in file data during a page fault.
1378 * The goto's are kind of ugly, but this streamlines the normal case of having
1379 * it in the page cache, and handles the special cases reasonably without
1380 * having a lot of duplicated code.
1382 struct page
*filemap_nopage(struct vm_area_struct
*area
,
1383 unsigned long address
, int *type
)
1386 struct file
*file
= area
->vm_file
;
1387 struct address_space
*mapping
= file
->f_mapping
;
1388 struct file_ra_state
*ra
= &file
->f_ra
;
1389 struct inode
*inode
= mapping
->host
;
1391 unsigned long size
, pgoff
;
1392 int did_readaround
= 0, majmin
= VM_FAULT_MINOR
;
1394 pgoff
= ((address
-area
->vm_start
) >> PAGE_CACHE_SHIFT
) + area
->vm_pgoff
;
1397 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1399 goto outside_data_content
;
1401 /* If we don't want any read-ahead, don't bother */
1402 if (VM_RandomReadHint(area
))
1403 goto no_cached_page
;
1406 * The readahead code wants to be told about each and every page
1407 * so it can build and shrink its windows appropriately
1409 * For sequential accesses, we use the generic readahead logic.
1411 if (VM_SequentialReadHint(area
))
1412 page_cache_readahead(mapping
, ra
, file
, pgoff
, 1);
1415 * Do we have something in the page cache already?
1418 page
= find_get_page(mapping
, pgoff
);
1420 unsigned long ra_pages
;
1422 if (VM_SequentialReadHint(area
)) {
1423 handle_ra_miss(mapping
, ra
, pgoff
);
1424 goto no_cached_page
;
1429 * Do we miss much more than hit in this file? If so,
1430 * stop bothering with read-ahead. It will only hurt.
1432 if (ra
->mmap_miss
> ra
->mmap_hit
+ MMAP_LOTSAMISS
)
1433 goto no_cached_page
;
1436 * To keep the pgmajfault counter straight, we need to
1437 * check did_readaround, as this is an inner loop.
1439 if (!did_readaround
) {
1440 majmin
= VM_FAULT_MAJOR
;
1441 count_vm_event(PGMAJFAULT
);
1444 ra_pages
= max_sane_readahead(file
->f_ra
.ra_pages
);
1448 if (pgoff
> ra_pages
/ 2)
1449 start
= pgoff
- ra_pages
/ 2;
1450 do_page_cache_readahead(mapping
, file
, start
, ra_pages
);
1452 page
= find_get_page(mapping
, pgoff
);
1454 goto no_cached_page
;
1457 if (!did_readaround
)
1461 * Ok, found a page in the page cache, now we need to check
1462 * that it's up-to-date.
1464 if (!PageUptodate(page
))
1465 goto page_not_uptodate
;
1469 * Found the page and have a reference on it.
1471 mark_page_accessed(page
);
1476 outside_data_content
:
1478 * An external ptracer can access pages that normally aren't
1481 if (area
->vm_mm
== current
->mm
)
1482 return NOPAGE_SIGBUS
;
1483 /* Fall through to the non-read-ahead case */
1486 * We're only likely to ever get here if MADV_RANDOM is in
1489 error
= page_cache_read(file
, pgoff
);
1493 * The page we want has now been added to the page cache.
1494 * In the unlikely event that someone removed it in the
1495 * meantime, we'll just come back here and read it again.
1501 * An error return from page_cache_read can result if the
1502 * system is low on memory, or a problem occurs while trying
1505 if (error
== -ENOMEM
)
1507 return NOPAGE_SIGBUS
;
1510 if (!did_readaround
) {
1511 majmin
= VM_FAULT_MAJOR
;
1512 count_vm_event(PGMAJFAULT
);
1516 /* Did it get unhashed while we waited for it? */
1517 if (!page
->mapping
) {
1519 page_cache_release(page
);
1523 /* Did somebody else get it up-to-date? */
1524 if (PageUptodate(page
)) {
1529 error
= mapping
->a_ops
->readpage(file
, page
);
1531 wait_on_page_locked(page
);
1532 if (PageUptodate(page
))
1534 } else if (error
== AOP_TRUNCATED_PAGE
) {
1535 page_cache_release(page
);
1540 * Umm, take care of errors if the page isn't up-to-date.
1541 * Try to re-read it _once_. We do this synchronously,
1542 * because there really aren't any performance issues here
1543 * and we need to check for errors.
1547 /* Somebody truncated the page on us? */
1548 if (!page
->mapping
) {
1550 page_cache_release(page
);
1554 /* Somebody else successfully read it in? */
1555 if (PageUptodate(page
)) {
1559 ClearPageError(page
);
1560 error
= mapping
->a_ops
->readpage(file
, page
);
1562 wait_on_page_locked(page
);
1563 if (PageUptodate(page
))
1565 } else if (error
== AOP_TRUNCATED_PAGE
) {
1566 page_cache_release(page
);
1571 * Things didn't work out. Return zero to tell the
1572 * mm layer so, possibly freeing the page cache page first.
1574 shrink_readahead_size_eio(file
, ra
);
1575 page_cache_release(page
);
1576 return NOPAGE_SIGBUS
;
1578 EXPORT_SYMBOL(filemap_nopage
);
1580 static struct page
* filemap_getpage(struct file
*file
, unsigned long pgoff
,
1583 struct address_space
*mapping
= file
->f_mapping
;
1588 * Do we have something in the page cache already?
1591 page
= find_get_page(mapping
, pgoff
);
1595 goto no_cached_page
;
1599 * Ok, found a page in the page cache, now we need to check
1600 * that it's up-to-date.
1602 if (!PageUptodate(page
)) {
1604 page_cache_release(page
);
1607 goto page_not_uptodate
;
1612 * Found the page and have a reference on it.
1614 mark_page_accessed(page
);
1618 error
= page_cache_read(file
, pgoff
);
1621 * The page we want has now been added to the page cache.
1622 * In the unlikely event that someone removed it in the
1623 * meantime, we'll just come back here and read it again.
1629 * An error return from page_cache_read can result if the
1630 * system is low on memory, or a problem occurs while trying
1638 /* Did it get truncated while we waited for it? */
1639 if (!page
->mapping
) {
1644 /* Did somebody else get it up-to-date? */
1645 if (PageUptodate(page
)) {
1650 error
= mapping
->a_ops
->readpage(file
, page
);
1652 wait_on_page_locked(page
);
1653 if (PageUptodate(page
))
1655 } else if (error
== AOP_TRUNCATED_PAGE
) {
1656 page_cache_release(page
);
1661 * Umm, take care of errors if the page isn't up-to-date.
1662 * Try to re-read it _once_. We do this synchronously,
1663 * because there really aren't any performance issues here
1664 * and we need to check for errors.
1668 /* Somebody truncated the page on us? */
1669 if (!page
->mapping
) {
1673 /* Somebody else successfully read it in? */
1674 if (PageUptodate(page
)) {
1679 ClearPageError(page
);
1680 error
= mapping
->a_ops
->readpage(file
, page
);
1682 wait_on_page_locked(page
);
1683 if (PageUptodate(page
))
1685 } else if (error
== AOP_TRUNCATED_PAGE
) {
1686 page_cache_release(page
);
1691 * Things didn't work out. Return zero to tell the
1692 * mm layer so, possibly freeing the page cache page first.
1695 page_cache_release(page
);
1700 int filemap_populate(struct vm_area_struct
*vma
, unsigned long addr
,
1701 unsigned long len
, pgprot_t prot
, unsigned long pgoff
,
1704 struct file
*file
= vma
->vm_file
;
1705 struct address_space
*mapping
= file
->f_mapping
;
1706 struct inode
*inode
= mapping
->host
;
1708 struct mm_struct
*mm
= vma
->vm_mm
;
1713 force_page_cache_readahead(mapping
, vma
->vm_file
,
1714 pgoff
, len
>> PAGE_CACHE_SHIFT
);
1717 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1718 if (pgoff
+ (len
>> PAGE_CACHE_SHIFT
) > size
)
1721 page
= filemap_getpage(file
, pgoff
, nonblock
);
1723 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1724 * done in shmem_populate calling shmem_getpage */
1725 if (!page
&& !nonblock
)
1729 err
= install_page(mm
, vma
, addr
, page
, prot
);
1731 page_cache_release(page
);
1734 } else if (vma
->vm_flags
& VM_NONLINEAR
) {
1735 /* No page was found just because we can't read it in now (being
1736 * here implies nonblock != 0), but the page may exist, so set
1737 * the PTE to fault it in later. */
1738 err
= install_file_pte(mm
, vma
, addr
, pgoff
, prot
);
1751 EXPORT_SYMBOL(filemap_populate
);
1753 struct vm_operations_struct generic_file_vm_ops
= {
1754 .nopage
= filemap_nopage
,
1755 .populate
= filemap_populate
,
1758 /* This is used for a general mmap of a disk file */
1760 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1762 struct address_space
*mapping
= file
->f_mapping
;
1764 if (!mapping
->a_ops
->readpage
)
1766 file_accessed(file
);
1767 vma
->vm_ops
= &generic_file_vm_ops
;
1772 * This is for filesystems which do not implement ->writepage.
1774 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1776 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
1778 return generic_file_mmap(file
, vma
);
1781 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1785 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1789 #endif /* CONFIG_MMU */
1791 EXPORT_SYMBOL(generic_file_mmap
);
1792 EXPORT_SYMBOL(generic_file_readonly_mmap
);
1794 static inline struct page
*__read_cache_page(struct address_space
*mapping
,
1795 unsigned long index
,
1796 int (*filler
)(void *,struct page
*),
1799 struct page
*page
, *cached_page
= NULL
;
1802 page
= find_get_page(mapping
, index
);
1805 cached_page
= page_cache_alloc_cold(mapping
);
1807 return ERR_PTR(-ENOMEM
);
1809 err
= add_to_page_cache_lru(cached_page
, mapping
,
1814 /* Presumably ENOMEM for radix tree node */
1815 page_cache_release(cached_page
);
1816 return ERR_PTR(err
);
1820 err
= filler(data
, page
);
1822 page_cache_release(page
);
1823 page
= ERR_PTR(err
);
1827 page_cache_release(cached_page
);
1832 * read_cache_page - read into page cache, fill it if needed
1833 * @mapping: the page's address_space
1834 * @index: the page index
1835 * @filler: function to perform the read
1836 * @data: destination for read data
1838 * Read into the page cache. If a page already exists,
1839 * and PageUptodate() is not set, try to fill the page.
1841 struct page
*read_cache_page(struct address_space
*mapping
,
1842 unsigned long index
,
1843 int (*filler
)(void *,struct page
*),
1850 page
= __read_cache_page(mapping
, index
, filler
, data
);
1853 mark_page_accessed(page
);
1854 if (PageUptodate(page
))
1858 #if 0 // mask by Victor Yu. 02-12-2007
1859 if (!page
->mapping
) {
1861 if (!page
->u
.xx
.mapping
) {
1864 page_cache_release(page
);
1867 if (PageUptodate(page
)) {
1871 err
= filler(data
, page
);
1873 page_cache_release(page
);
1874 page
= ERR_PTR(err
);
1879 EXPORT_SYMBOL(read_cache_page
);
1882 * If the page was newly created, increment its refcount and add it to the
1883 * caller's lru-buffering pagevec. This function is specifically for
1884 * generic_file_write().
1886 static inline struct page
*
1887 __grab_cache_page(struct address_space
*mapping
, unsigned long index
,
1888 struct page
**cached_page
, struct pagevec
*lru_pvec
)
1893 page
= find_lock_page(mapping
, index
);
1895 if (!*cached_page
) {
1896 *cached_page
= page_cache_alloc(mapping
);
1900 err
= add_to_page_cache(*cached_page
, mapping
,
1905 page
= *cached_page
;
1906 page_cache_get(page
);
1907 if (!pagevec_add(lru_pvec
, page
))
1908 __pagevec_lru_add(lru_pvec
);
1909 *cached_page
= NULL
;
1916 * The logic we want is
1918 * if suid or (sgid and xgrp)
1921 int should_remove_suid(struct dentry
*dentry
)
1923 mode_t mode
= dentry
->d_inode
->i_mode
;
1926 /* suid always must be killed */
1927 if (unlikely(mode
& S_ISUID
))
1928 kill
= ATTR_KILL_SUID
;
1931 * sgid without any exec bits is just a mandatory locking mark; leave
1932 * it alone. If some exec bits are set, it's a real sgid; kill it.
1934 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1935 kill
|= ATTR_KILL_SGID
;
1937 if (unlikely(kill
&& !capable(CAP_FSETID
)))
1943 int __remove_suid(struct dentry
*dentry
, int kill
)
1945 struct iattr newattrs
;
1947 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1948 return notify_change(dentry
, &newattrs
);
1951 int remove_suid(struct dentry
*dentry
)
1953 int kill
= should_remove_suid(dentry
);
1956 return __remove_suid(dentry
, kill
);
1960 EXPORT_SYMBOL(remove_suid
);
1963 __filemap_copy_from_user_iovec_inatomic(char *vaddr
,
1964 const struct iovec
*iov
, size_t base
, size_t bytes
)
1966 size_t copied
= 0, left
= 0;
1969 char __user
*buf
= iov
->iov_base
+ base
;
1970 int copy
= min(bytes
, iov
->iov_len
- base
);
1973 left
= __copy_from_user_inatomic_nocache(vaddr
, buf
, copy
);
1982 return copied
- left
;
1986 * Performs necessary checks before doing a write
1988 * Can adjust writing position or amount of bytes to write.
1989 * Returns appropriate error code that caller should return or
1990 * zero in case that write should be allowed.
1992 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
1994 struct inode
*inode
= file
->f_mapping
->host
;
1995 unsigned long limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1997 if (unlikely(*pos
< 0))
2001 /* FIXME: this is for backwards compatibility with 2.4 */
2002 if (file
->f_flags
& O_APPEND
)
2003 *pos
= i_size_read(inode
);
2005 if (limit
!= RLIM_INFINITY
) {
2006 if (*pos
>= limit
) {
2007 send_sig(SIGXFSZ
, current
, 0);
2010 if (*count
> limit
- (typeof(limit
))*pos
) {
2011 *count
= limit
- (typeof(limit
))*pos
;
2019 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2020 !(file
->f_flags
& O_LARGEFILE
))) {
2021 if (*pos
>= MAX_NON_LFS
) {
2022 send_sig(SIGXFSZ
, current
, 0);
2025 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2026 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2031 * Are we about to exceed the fs block limit ?
2033 * If we have written data it becomes a short write. If we have
2034 * exceeded without writing data we send a signal and return EFBIG.
2035 * Linus frestrict idea will clean these up nicely..
2037 if (likely(!isblk
)) {
2038 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2039 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2040 send_sig(SIGXFSZ
, current
, 0);
2043 /* zero-length writes at ->s_maxbytes are OK */
2046 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2047 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2051 if (bdev_read_only(I_BDEV(inode
)))
2053 isize
= i_size_read(inode
);
2054 if (*pos
>= isize
) {
2055 if (*count
|| *pos
> isize
)
2059 if (*pos
+ *count
> isize
)
2060 *count
= isize
- *pos
;
2067 EXPORT_SYMBOL(generic_write_checks
);
2070 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2071 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2072 size_t count
, size_t ocount
)
2074 struct file
*file
= iocb
->ki_filp
;
2075 struct address_space
*mapping
= file
->f_mapping
;
2076 struct inode
*inode
= mapping
->host
;
2079 if (count
!= ocount
)
2080 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2082 written
= generic_file_direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2084 loff_t end
= pos
+ written
;
2085 if (end
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2086 i_size_write(inode
, end
);
2087 mark_inode_dirty(inode
);
2093 * Sync the fs metadata but not the minor inode changes and
2094 * of course not the data as we did direct DMA for the IO.
2095 * i_mutex is held, which protects generic_osync_inode() from
2098 if (written
>= 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2099 int err
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
2103 if (written
== count
&& !is_sync_kiocb(iocb
))
2104 written
= -EIOCBQUEUED
;
2107 EXPORT_SYMBOL(generic_file_direct_write
);
2110 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2111 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2112 size_t count
, ssize_t written
)
2114 struct file
*file
= iocb
->ki_filp
;
2115 struct address_space
* mapping
= file
->f_mapping
;
2116 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2117 struct inode
*inode
= mapping
->host
;
2120 struct page
*cached_page
= NULL
;
2122 struct pagevec lru_pvec
;
2123 const struct iovec
*cur_iov
= iov
; /* current iovec */
2124 size_t iov_base
= 0; /* offset in the current iovec */
2127 pagevec_init(&lru_pvec
, 0);
2130 * handle partial DIO write. Adjust cur_iov if needed.
2132 if (likely(nr_segs
== 1))
2133 buf
= iov
->iov_base
+ written
;
2135 filemap_set_next_iovec(&cur_iov
, &iov_base
, written
);
2136 buf
= cur_iov
->iov_base
+ iov_base
;
2140 unsigned long index
;
2141 unsigned long offset
;
2144 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
2145 index
= pos
>> PAGE_CACHE_SHIFT
;
2146 bytes
= PAGE_CACHE_SIZE
- offset
;
2148 /* Limit the size of the copy to the caller's write size */
2149 bytes
= min(bytes
, count
);
2152 * Limit the size of the copy to that of the current segment,
2153 * because fault_in_pages_readable() doesn't know how to walk
2156 bytes
= min(bytes
, cur_iov
->iov_len
- iov_base
);
2159 * Bring in the user page that we will copy from _first_.
2160 * Otherwise there's a nasty deadlock on copying from the
2161 * same page as we're writing to, without it being marked
2164 fault_in_pages_readable(buf
, bytes
);
2166 page
= __grab_cache_page(mapping
,index
,&cached_page
,&lru_pvec
);
2172 if (unlikely(bytes
== 0)) {
2175 goto zero_length_segment
;
2178 status
= a_ops
->prepare_write(file
, page
, offset
, offset
+bytes
);
2179 if (unlikely(status
)) {
2180 loff_t isize
= i_size_read(inode
);
2182 if (status
!= AOP_TRUNCATED_PAGE
)
2184 page_cache_release(page
);
2185 if (status
== AOP_TRUNCATED_PAGE
)
2188 * prepare_write() may have instantiated a few blocks
2189 * outside i_size. Trim these off again.
2191 if (pos
+ bytes
> isize
)
2192 vmtruncate(inode
, isize
);
2195 if (likely(nr_segs
== 1))
2196 copied
= filemap_copy_from_user(page
, offset
,
2199 copied
= filemap_copy_from_user_iovec(page
, offset
,
2200 cur_iov
, iov_base
, bytes
);
2201 flush_dcache_page(page
);
2202 status
= a_ops
->commit_write(file
, page
, offset
, offset
+bytes
);
2203 if (status
== AOP_TRUNCATED_PAGE
) {
2204 page_cache_release(page
);
2207 zero_length_segment
:
2208 if (likely(copied
>= 0)) {
2217 if (unlikely(nr_segs
> 1)) {
2218 filemap_set_next_iovec(&cur_iov
,
2221 buf
= cur_iov
->iov_base
+
2228 if (unlikely(copied
!= bytes
))
2232 mark_page_accessed(page
);
2233 page_cache_release(page
);
2236 balance_dirty_pages_ratelimited(mapping
);
2242 page_cache_release(cached_page
);
2245 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2247 if (likely(status
>= 0)) {
2248 if (unlikely((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2249 if (!a_ops
->writepage
|| !is_sync_kiocb(iocb
))
2250 status
= generic_osync_inode(inode
, mapping
,
2251 OSYNC_METADATA
|OSYNC_DATA
);
2256 * If we get here for O_DIRECT writes then we must have fallen through
2257 * to buffered writes (block instantiation inside i_size). So we sync
2258 * the file data here, to try to honour O_DIRECT expectations.
2260 if (unlikely(file
->f_flags
& O_DIRECT
) && written
)
2261 status
= filemap_write_and_wait(mapping
);
2263 pagevec_lru_add(&lru_pvec
);
2264 return written
? written
: status
;
2266 EXPORT_SYMBOL(generic_file_buffered_write
);
2269 __generic_file_aio_write_nolock(struct kiocb
*iocb
, const struct iovec
*iov
,
2270 unsigned long nr_segs
, loff_t
*ppos
)
2272 struct file
*file
= iocb
->ki_filp
;
2273 struct address_space
* mapping
= file
->f_mapping
;
2274 size_t ocount
; /* original count */
2275 size_t count
; /* after file limit checks */
2276 struct inode
*inode
= mapping
->host
;
2283 for (seg
= 0; seg
< nr_segs
; seg
++) {
2284 const struct iovec
*iv
= &iov
[seg
];
2287 * If any segment has a negative length, or the cumulative
2288 * length ever wraps negative then return -EINVAL.
2290 ocount
+= iv
->iov_len
;
2291 if (unlikely((ssize_t
)(ocount
|iv
->iov_len
) < 0))
2293 if (access_ok(VERIFY_READ
, iv
->iov_base
, iv
->iov_len
))
2298 ocount
-= iv
->iov_len
; /* This segment is no good */
2305 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2307 /* We can write back this queue in page reclaim */
2308 current
->backing_dev_info
= mapping
->backing_dev_info
;
2311 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2318 err
= remove_suid(file
->f_dentry
);
2322 file_update_time(file
);
2324 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2325 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2327 ssize_t written_buffered
;
2329 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2330 ppos
, count
, ocount
);
2331 if (written
< 0 || written
== count
)
2334 * direct-io write to a hole: fall through to buffered I/O
2335 * for completing the rest of the request.
2339 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2340 nr_segs
, pos
, ppos
, count
,
2343 * If generic_file_buffered_write() retuned a synchronous error
2344 * then we want to return the number of bytes which were
2345 * direct-written, or the error code if that was zero. Note
2346 * that this differs from normal direct-io semantics, which
2347 * will return -EFOO even if some bytes were written.
2349 if (written_buffered
< 0) {
2350 err
= written_buffered
;
2355 * We need to ensure that the page cache pages are written to
2356 * disk and invalidated to preserve the expected O_DIRECT
2359 endbyte
= pos
+ written_buffered
- written
- 1;
2360 err
= do_sync_file_range(file
, pos
, endbyte
,
2361 SYNC_FILE_RANGE_WAIT_BEFORE
|
2362 SYNC_FILE_RANGE_WRITE
|
2363 SYNC_FILE_RANGE_WAIT_AFTER
);
2365 written
= written_buffered
;
2366 invalidate_mapping_pages(mapping
,
2367 pos
>> PAGE_CACHE_SHIFT
,
2368 endbyte
>> PAGE_CACHE_SHIFT
);
2371 * We don't know how much we wrote, so just return
2372 * the number of bytes which were direct-written
2376 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2377 pos
, ppos
, count
, written
);
2380 current
->backing_dev_info
= NULL
;
2381 return written
? written
: err
;
2384 ssize_t
generic_file_aio_write_nolock(struct kiocb
*iocb
,
2385 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
2387 struct file
*file
= iocb
->ki_filp
;
2388 struct address_space
*mapping
= file
->f_mapping
;
2389 struct inode
*inode
= mapping
->host
;
2392 BUG_ON(iocb
->ki_pos
!= pos
);
2394 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
,
2397 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2400 err
= sync_page_range_nolock(inode
, mapping
, pos
, ret
);
2406 EXPORT_SYMBOL(generic_file_aio_write_nolock
);
2408 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2409 unsigned long nr_segs
, loff_t pos
)
2411 struct file
*file
= iocb
->ki_filp
;
2412 struct address_space
*mapping
= file
->f_mapping
;
2413 struct inode
*inode
= mapping
->host
;
2416 BUG_ON(iocb
->ki_pos
!= pos
);
2418 mutex_lock(&inode
->i_mutex
);
2419 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
,
2421 mutex_unlock(&inode
->i_mutex
);
2423 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2426 err
= sync_page_range(inode
, mapping
, pos
, ret
);
2432 EXPORT_SYMBOL(generic_file_aio_write
);
2434 #ifdef CONFIG_DIRECTIO
2436 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2437 * went wrong during pagecache shootdown.
2440 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
2441 loff_t offset
, unsigned long nr_segs
)
2443 struct file
*file
= iocb
->ki_filp
;
2444 struct address_space
*mapping
= file
->f_mapping
;
2446 size_t write_len
= 0;
2449 * If it's a write, unmap all mmappings of the file up-front. This
2450 * will cause any pte dirty bits to be propagated into the pageframes
2451 * for the subsequent filemap_write_and_wait().
2454 write_len
= iov_length(iov
, nr_segs
);
2455 if (mapping_mapped(mapping
))
2456 unmap_mapping_range(mapping
, offset
, write_len
, 0);
2459 retval
= filemap_write_and_wait(mapping
);
2461 retval
= mapping
->a_ops
->direct_IO(rw
, iocb
, iov
,
2463 if (rw
== WRITE
&& mapping
->nrpages
) {
2464 pgoff_t end
= (offset
+ write_len
- 1)
2465 >> PAGE_CACHE_SHIFT
;
2466 int err
= invalidate_inode_pages2_range(mapping
,
2467 offset
>> PAGE_CACHE_SHIFT
, end
);
2477 * try_to_release_page() - release old fs-specific metadata on a page
2479 * @page: the page which the kernel is trying to free
2480 * @gfp_mask: memory allocation flags (and I/O mode)
2482 * The address_space is to try to release any data against the page
2483 * (presumably at page->private). If the release was successful, return `1'.
2484 * Otherwise return zero.
2486 * The @gfp_mask argument specifies whether I/O may be performed to release
2487 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2489 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2491 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2493 #if 0 // mask by Victor Yu. 02-12-2007
2494 struct address_space
* const mapping
= page
->mapping
;
2496 struct address_space
* const mapping
= page
->u
.xx
.mapping
;
2499 BUG_ON(!PageLocked(page
));
2500 if (PageWriteback(page
))
2503 if (mapping
&& mapping
->a_ops
->releasepage
)
2504 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2505 return try_to_free_buffers(page
);
2508 EXPORT_SYMBOL(try_to_release_page
);