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/config.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/compiler.h>
17 #include <linux/aio.h>
18 #include <linux/kernel_stat.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
33 * FIXME: remove all knowledge of the buffer layer from the core VM
35 #include <linux/buffer_head.h> /* for generic_osync_inode */
37 #include <asm/uaccess.h>
41 * Shared mappings implemented 30.11.1994. It's not fully working yet,
44 * Shared mappings now work. 15.8.1995 Bruno.
46 * finished 'unifying' the page and buffer cache and SMP-threaded the
47 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
49 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
55 * ->i_mmap_lock (vmtruncate)
56 * ->private_lock (__free_pte->__set_page_dirty_buffers)
58 * ->swap_device_lock (exclusive_swap_page, others)
59 * ->mapping->tree_lock
62 * ->i_mmap_lock (truncate->unmap_mapping_range)
66 * ->page_table_lock (various places, mainly in mmap.c)
67 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
70 * ->lock_page (access_process_vm)
76 * ->i_alloc_sem (various)
79 * ->sb_lock (fs/fs-writeback.c)
80 * ->mapping->tree_lock (__sync_single_inode)
83 * ->anon_vma.lock (vma_adjust)
86 * ->page_table_lock (anon_vma_prepare and various)
89 * ->swap_device_lock (try_to_unmap_one)
90 * ->private_lock (try_to_unmap_one)
91 * ->tree_lock (try_to_unmap_one)
92 * ->zone.lru_lock (follow_page->mark_page_accessed)
93 * ->private_lock (page_remove_rmap->set_page_dirty)
94 * ->tree_lock (page_remove_rmap->set_page_dirty)
95 * ->inode_lock (page_remove_rmap->set_page_dirty)
96 * ->inode_lock (zap_pte_range->set_page_dirty)
97 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
100 * ->dcache_lock (proc_pid_lookup)
104 * Remove a page from the page cache and free it. Caller has to make
105 * sure the page is locked and that nobody else uses it - or that usage
106 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
108 void __remove_from_page_cache(struct page
*page
)
110 struct address_space
*mapping
= page
->mapping
;
112 radix_tree_delete(&mapping
->page_tree
, page
->index
);
113 page
->mapping
= NULL
;
118 void remove_from_page_cache(struct page
*page
)
120 struct address_space
*mapping
= page
->mapping
;
122 BUG_ON(!PageLocked(page
));
124 write_lock_irq(&mapping
->tree_lock
);
125 __remove_from_page_cache(page
);
126 write_unlock_irq(&mapping
->tree_lock
);
129 static int sync_page(void *word
)
131 struct address_space
*mapping
;
134 page
= container_of((page_flags_t
*)word
, struct page
, flags
);
137 * page_mapping() is being called without PG_locked held.
138 * Some knowledge of the state and use of the page is used to
139 * reduce the requirements down to a memory barrier.
140 * The danger here is of a stale page_mapping() return value
141 * indicating a struct address_space different from the one it's
142 * associated with when it is associated with one.
143 * After smp_mb(), it's either the correct page_mapping() for
144 * the page, or an old page_mapping() and the page's own
145 * page_mapping() has gone NULL.
146 * The ->sync_page() address_space operation must tolerate
147 * page_mapping() going NULL. By an amazing coincidence,
148 * this comes about because none of the users of the page
149 * in the ->sync_page() methods make essential use of the
150 * page_mapping(), merely passing the page down to the backing
151 * device's unplug functions when it's non-NULL, which in turn
152 * ignore it for all cases but swap, where only page->private is
153 * of interest. When page_mapping() does go NULL, the entire
154 * call stack gracefully ignores the page and returns.
158 mapping
= page_mapping(page
);
159 if (mapping
&& mapping
->a_ops
&& mapping
->a_ops
->sync_page
)
160 mapping
->a_ops
->sync_page(page
);
166 * filemap_fdatawrite_range - start writeback against all of a mapping's
167 * dirty pages that lie within the byte offsets <start, end>
168 * @mapping: address space structure to write
169 * @start: offset in bytes where the range starts
170 * @end: offset in bytes where the range ends
171 * @sync_mode: enable synchronous operation
173 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
174 * opposed to a regular memory * cleansing writeback. The difference between
175 * these two operations is that if a dirty page/buffer is encountered, it must
176 * be waited upon, and not just skipped over.
178 static int __filemap_fdatawrite_range(struct address_space
*mapping
,
179 loff_t start
, loff_t end
, int sync_mode
)
182 struct writeback_control wbc
= {
183 .sync_mode
= sync_mode
,
184 .nr_to_write
= mapping
->nrpages
* 2,
189 if (!mapping_cap_writeback_dirty(mapping
))
192 ret
= do_writepages(mapping
, &wbc
);
196 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
199 return __filemap_fdatawrite_range(mapping
, 0, 0, sync_mode
);
202 int filemap_fdatawrite(struct address_space
*mapping
)
204 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
206 EXPORT_SYMBOL(filemap_fdatawrite
);
208 static int filemap_fdatawrite_range(struct address_space
*mapping
,
209 loff_t start
, loff_t end
)
211 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
215 * This is a mostly non-blocking flush. Not suitable for data-integrity
216 * purposes - I/O may not be started against all dirty pages.
218 int filemap_flush(struct address_space
*mapping
)
220 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
222 EXPORT_SYMBOL(filemap_flush
);
225 * Wait for writeback to complete against pages indexed by start->end
228 static int wait_on_page_writeback_range(struct address_space
*mapping
,
229 pgoff_t start
, pgoff_t end
)
239 pagevec_init(&pvec
, 0);
241 while ((index
<= end
) &&
242 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
243 PAGECACHE_TAG_WRITEBACK
,
244 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
247 for (i
= 0; i
< nr_pages
; i
++) {
248 struct page
*page
= pvec
.pages
[i
];
250 /* until radix tree lookup accepts end_index */
251 if (page
->index
> end
)
254 wait_on_page_writeback(page
);
258 pagevec_release(&pvec
);
262 /* Check for outstanding write errors */
263 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
265 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
272 * Write and wait upon all the pages in the passed range. This is a "data
273 * integrity" operation. It waits upon in-flight writeout before starting and
274 * waiting upon new writeout. If there was an IO error, return it.
276 * We need to re-take i_sem during the generic_osync_inode list walk because
277 * it is otherwise livelockable.
279 int sync_page_range(struct inode
*inode
, struct address_space
*mapping
,
280 loff_t pos
, size_t count
)
282 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
283 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
286 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
288 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
291 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
295 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
298 EXPORT_SYMBOL(sync_page_range
);
301 * Note: Holding i_sem across sync_page_range_nolock is not a good idea
302 * as it forces O_SYNC writers to different parts of the same file
303 * to be serialised right until io completion.
305 int sync_page_range_nolock(struct inode
*inode
, struct address_space
*mapping
,
306 loff_t pos
, size_t count
)
308 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
309 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
312 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
314 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
316 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
318 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
321 EXPORT_SYMBOL(sync_page_range_nolock
);
324 * filemap_fdatawait - walk the list of under-writeback pages of the given
325 * address space and wait for all of them.
327 * @mapping: address space structure to wait for
329 int filemap_fdatawait(struct address_space
*mapping
)
331 loff_t i_size
= i_size_read(mapping
->host
);
336 return wait_on_page_writeback_range(mapping
, 0,
337 (i_size
- 1) >> PAGE_CACHE_SHIFT
);
339 EXPORT_SYMBOL(filemap_fdatawait
);
341 int filemap_write_and_wait(struct address_space
*mapping
)
345 if (mapping
->nrpages
) {
346 retval
= filemap_fdatawrite(mapping
);
348 retval
= filemap_fdatawait(mapping
);
353 int filemap_write_and_wait_range(struct address_space
*mapping
,
354 loff_t lstart
, loff_t lend
)
358 if (mapping
->nrpages
) {
359 retval
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
362 retval
= wait_on_page_writeback_range(mapping
,
363 lstart
>> PAGE_CACHE_SHIFT
,
364 lend
>> PAGE_CACHE_SHIFT
);
370 * This function is used to add newly allocated pagecache pages:
371 * the page is new, so we can just run SetPageLocked() against it.
372 * The other page state flags were set by rmqueue().
374 * This function does not add the page to the LRU. The caller must do that.
376 int add_to_page_cache(struct page
*page
, struct address_space
*mapping
,
377 pgoff_t offset
, int gfp_mask
)
379 int error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
382 write_lock_irq(&mapping
->tree_lock
);
383 error
= radix_tree_insert(&mapping
->page_tree
, offset
, page
);
385 page_cache_get(page
);
387 page
->mapping
= mapping
;
388 page
->index
= offset
;
392 write_unlock_irq(&mapping
->tree_lock
);
393 radix_tree_preload_end();
398 EXPORT_SYMBOL(add_to_page_cache
);
400 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
401 pgoff_t offset
, int gfp_mask
)
403 int ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
410 * In order to wait for pages to become available there must be
411 * waitqueues associated with pages. By using a hash table of
412 * waitqueues where the bucket discipline is to maintain all
413 * waiters on the same queue and wake all when any of the pages
414 * become available, and for the woken contexts to check to be
415 * sure the appropriate page became available, this saves space
416 * at a cost of "thundering herd" phenomena during rare hash
419 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
421 const struct zone
*zone
= page_zone(page
);
423 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
426 static inline void wake_up_page(struct page
*page
, int bit
)
428 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
431 void fastcall
wait_on_page_bit(struct page
*page
, int bit_nr
)
433 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
435 if (test_bit(bit_nr
, &page
->flags
))
436 __wait_on_bit(page_waitqueue(page
), &wait
, sync_page
,
437 TASK_UNINTERRUPTIBLE
);
439 EXPORT_SYMBOL(wait_on_page_bit
);
442 * unlock_page() - unlock a locked page
446 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
447 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
448 * mechananism between PageLocked pages and PageWriteback pages is shared.
449 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
451 * The first mb is necessary to safely close the critical section opened by the
452 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
453 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
454 * parallel wait_on_page_locked()).
456 void fastcall
unlock_page(struct page
*page
)
458 smp_mb__before_clear_bit();
459 if (!TestClearPageLocked(page
))
461 smp_mb__after_clear_bit();
462 wake_up_page(page
, PG_locked
);
464 EXPORT_SYMBOL(unlock_page
);
467 * End writeback against a page.
469 void end_page_writeback(struct page
*page
)
471 if (!TestClearPageReclaim(page
) || rotate_reclaimable_page(page
)) {
472 if (!test_clear_page_writeback(page
))
475 smp_mb__after_clear_bit();
476 wake_up_page(page
, PG_writeback
);
478 EXPORT_SYMBOL(end_page_writeback
);
481 * Get a lock on the page, assuming we need to sleep to get it.
483 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
484 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
485 * chances are that on the second loop, the block layer's plug list is empty,
486 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
488 void fastcall
__lock_page(struct page
*page
)
490 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
492 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sync_page
,
493 TASK_UNINTERRUPTIBLE
);
495 EXPORT_SYMBOL(__lock_page
);
498 * a rather lightweight function, finding and getting a reference to a
499 * hashed page atomically.
501 struct page
* find_get_page(struct address_space
*mapping
, unsigned long offset
)
505 read_lock_irq(&mapping
->tree_lock
);
506 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
508 page_cache_get(page
);
509 read_unlock_irq(&mapping
->tree_lock
);
513 EXPORT_SYMBOL(find_get_page
);
516 * Same as above, but trylock it instead of incrementing the count.
518 struct page
*find_trylock_page(struct address_space
*mapping
, unsigned long offset
)
522 read_lock_irq(&mapping
->tree_lock
);
523 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
524 if (page
&& TestSetPageLocked(page
))
526 read_unlock_irq(&mapping
->tree_lock
);
530 EXPORT_SYMBOL(find_trylock_page
);
533 * find_lock_page - locate, pin and lock a pagecache page
535 * @mapping: the address_space to search
536 * @offset: the page index
538 * Locates the desired pagecache page, locks it, increments its reference
539 * count and returns its address.
541 * Returns zero if the page was not present. find_lock_page() may sleep.
543 struct page
*find_lock_page(struct address_space
*mapping
,
544 unsigned long offset
)
548 read_lock_irq(&mapping
->tree_lock
);
550 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
552 page_cache_get(page
);
553 if (TestSetPageLocked(page
)) {
554 read_unlock_irq(&mapping
->tree_lock
);
556 read_lock_irq(&mapping
->tree_lock
);
558 /* Has the page been truncated while we slept? */
559 if (page
->mapping
!= mapping
|| page
->index
!= offset
) {
561 page_cache_release(page
);
566 read_unlock_irq(&mapping
->tree_lock
);
570 EXPORT_SYMBOL(find_lock_page
);
573 * find_or_create_page - locate or add a pagecache page
575 * @mapping: the page's address_space
576 * @index: the page's index into the mapping
577 * @gfp_mask: page allocation mode
579 * Locates a page in the pagecache. If the page is not present, a new page
580 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
581 * LRU list. The returned page is locked and has its reference count
584 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
587 * find_or_create_page() returns the desired page's address, or zero on
590 struct page
*find_or_create_page(struct address_space
*mapping
,
591 unsigned long index
, unsigned int gfp_mask
)
593 struct page
*page
, *cached_page
= NULL
;
596 page
= find_lock_page(mapping
, index
);
599 cached_page
= alloc_page(gfp_mask
);
603 err
= add_to_page_cache_lru(cached_page
, mapping
,
608 } else if (err
== -EEXIST
)
612 page_cache_release(cached_page
);
616 EXPORT_SYMBOL(find_or_create_page
);
619 * find_get_pages - gang pagecache lookup
620 * @mapping: The address_space to search
621 * @start: The starting page index
622 * @nr_pages: The maximum number of pages
623 * @pages: Where the resulting pages are placed
625 * find_get_pages() will search for and return a group of up to
626 * @nr_pages pages in the mapping. The pages are placed at @pages.
627 * find_get_pages() takes a reference against the returned pages.
629 * The search returns a group of mapping-contiguous pages with ascending
630 * indexes. There may be holes in the indices due to not-present pages.
632 * find_get_pages() returns the number of pages which were found.
634 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
635 unsigned int nr_pages
, struct page
**pages
)
640 read_lock_irq(&mapping
->tree_lock
);
641 ret
= radix_tree_gang_lookup(&mapping
->page_tree
,
642 (void **)pages
, start
, nr_pages
);
643 for (i
= 0; i
< ret
; i
++)
644 page_cache_get(pages
[i
]);
645 read_unlock_irq(&mapping
->tree_lock
);
650 * Like find_get_pages, except we only return pages which are tagged with
651 * `tag'. We update *index to index the next page for the traversal.
653 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
654 int tag
, unsigned int nr_pages
, struct page
**pages
)
659 read_lock_irq(&mapping
->tree_lock
);
660 ret
= radix_tree_gang_lookup_tag(&mapping
->page_tree
,
661 (void **)pages
, *index
, nr_pages
, tag
);
662 for (i
= 0; i
< ret
; i
++)
663 page_cache_get(pages
[i
]);
665 *index
= pages
[ret
- 1]->index
+ 1;
666 read_unlock_irq(&mapping
->tree_lock
);
671 * Same as grab_cache_page, but do not wait if the page is unavailable.
672 * This is intended for speculative data generators, where the data can
673 * be regenerated if the page couldn't be grabbed. This routine should
674 * be safe to call while holding the lock for another page.
676 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
677 * and deadlock against the caller's locked page.
680 grab_cache_page_nowait(struct address_space
*mapping
, unsigned long index
)
682 struct page
*page
= find_get_page(mapping
, index
);
683 unsigned int gfp_mask
;
686 if (!TestSetPageLocked(page
))
688 page_cache_release(page
);
691 gfp_mask
= mapping_gfp_mask(mapping
) & ~__GFP_FS
;
692 page
= alloc_pages(gfp_mask
, 0);
693 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, gfp_mask
)) {
694 page_cache_release(page
);
700 EXPORT_SYMBOL(grab_cache_page_nowait
);
703 * This is a generic file read routine, and uses the
704 * mapping->a_ops->readpage() function for the actual low-level
707 * This is really ugly. But the goto's actually try to clarify some
708 * of the logic when it comes to error handling etc.
710 * Note the struct file* is only passed for the use of readpage. It may be
713 void do_generic_mapping_read(struct address_space
*mapping
,
714 struct file_ra_state
*_ra
,
717 read_descriptor_t
*desc
,
720 struct inode
*inode
= mapping
->host
;
722 unsigned long end_index
;
723 unsigned long offset
;
724 unsigned long last_index
;
725 unsigned long next_index
;
726 unsigned long prev_index
;
728 struct page
*cached_page
;
730 struct file_ra_state ra
= *_ra
;
733 index
= *ppos
>> PAGE_CACHE_SHIFT
;
735 prev_index
= ra
.prev_page
;
736 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
737 offset
= *ppos
& ~PAGE_CACHE_MASK
;
739 isize
= i_size_read(inode
);
743 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
746 unsigned long nr
, ret
;
748 /* nr is the maximum number of bytes to copy from this page */
749 nr
= PAGE_CACHE_SIZE
;
750 if (index
>= end_index
) {
751 if (index
> end_index
)
753 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
761 if (index
== next_index
)
762 next_index
= page_cache_readahead(mapping
, &ra
, filp
,
763 index
, last_index
- index
);
766 page
= find_get_page(mapping
, index
);
767 if (unlikely(page
== NULL
)) {
768 handle_ra_miss(mapping
, &ra
, index
);
771 if (!PageUptodate(page
))
772 goto page_not_up_to_date
;
775 /* If users can be writing to this page using arbitrary
776 * virtual addresses, take care about potential aliasing
777 * before reading the page on the kernel side.
779 if (mapping_writably_mapped(mapping
))
780 flush_dcache_page(page
);
783 * When (part of) the same page is read multiple times
784 * in succession, only mark it as accessed the first time.
786 if (prev_index
!= index
)
787 mark_page_accessed(page
);
791 * Ok, we have the page, and it's up-to-date, so
792 * now we can copy it to user space...
794 * The actor routine returns how many bytes were actually used..
795 * NOTE! This may not be the same as how much of a user buffer
796 * we filled up (we may be padding etc), so we can only update
797 * "pos" here (the actor routine has to update the user buffer
798 * pointers and the remaining count).
800 ret
= actor(desc
, page
, offset
, nr
);
802 index
+= offset
>> PAGE_CACHE_SHIFT
;
803 offset
&= ~PAGE_CACHE_MASK
;
805 page_cache_release(page
);
806 if (ret
== nr
&& desc
->count
)
811 /* Get exclusive access to the page ... */
814 /* Did it get unhashed before we got the lock? */
815 if (!page
->mapping
) {
817 page_cache_release(page
);
821 /* Did somebody else fill it already? */
822 if (PageUptodate(page
)) {
828 /* Start the actual read. The read will unlock the page. */
829 error
= mapping
->a_ops
->readpage(filp
, page
);
834 if (!PageUptodate(page
)) {
836 if (!PageUptodate(page
)) {
837 if (page
->mapping
== NULL
) {
839 * invalidate_inode_pages got it
842 page_cache_release(page
);
853 * i_size must be checked after we have done ->readpage.
855 * Checking i_size after the readpage allows us to calculate
856 * the correct value for "nr", which means the zero-filled
857 * part of the page is not copied back to userspace (unless
858 * another truncate extends the file - this is desired though).
860 isize
= i_size_read(inode
);
861 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
862 if (unlikely(!isize
|| index
> end_index
)) {
863 page_cache_release(page
);
867 /* nr is the maximum number of bytes to copy from this page */
868 nr
= PAGE_CACHE_SIZE
;
869 if (index
== end_index
) {
870 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
872 page_cache_release(page
);
880 /* UHHUH! A synchronous read error occurred. Report it */
882 page_cache_release(page
);
887 * Ok, it wasn't cached, so we need to create a new
891 cached_page
= page_cache_alloc_cold(mapping
);
893 desc
->error
= -ENOMEM
;
897 error
= add_to_page_cache_lru(cached_page
, mapping
,
900 if (error
== -EEXIST
)
913 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
915 page_cache_release(cached_page
);
920 EXPORT_SYMBOL(do_generic_mapping_read
);
922 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
923 unsigned long offset
, unsigned long size
)
926 unsigned long left
, count
= desc
->count
;
932 * Faults on the destination of a read are common, so do it before
935 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
936 kaddr
= kmap_atomic(page
, KM_USER0
);
937 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
938 kaddr
+ offset
, size
);
939 kunmap_atomic(kaddr
, KM_USER0
);
944 /* Do it the slow way */
946 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
951 desc
->error
= -EFAULT
;
954 desc
->count
= count
- size
;
955 desc
->written
+= size
;
956 desc
->arg
.buf
+= size
;
961 * This is the "read()" routine for all filesystems
962 * that can use the page cache directly.
965 __generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
966 unsigned long nr_segs
, loff_t
*ppos
)
968 struct file
*filp
= iocb
->ki_filp
;
974 for (seg
= 0; seg
< nr_segs
; seg
++) {
975 const struct iovec
*iv
= &iov
[seg
];
978 * If any segment has a negative length, or the cumulative
979 * length ever wraps negative then return -EINVAL.
981 count
+= iv
->iov_len
;
982 if (unlikely((ssize_t
)(count
|iv
->iov_len
) < 0))
984 if (access_ok(VERIFY_WRITE
, iv
->iov_base
, iv
->iov_len
))
989 count
-= iv
->iov_len
; /* This segment is no good */
993 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
994 if (filp
->f_flags
& O_DIRECT
) {
995 loff_t pos
= *ppos
, size
;
996 struct address_space
*mapping
;
999 mapping
= filp
->f_mapping
;
1000 inode
= mapping
->host
;
1003 goto out
; /* skip atime */
1004 size
= i_size_read(inode
);
1006 retval
= generic_file_direct_IO(READ
, iocb
,
1008 if (retval
> 0 && !is_sync_kiocb(iocb
))
1009 retval
= -EIOCBQUEUED
;
1011 *ppos
= pos
+ retval
;
1013 file_accessed(filp
);
1019 for (seg
= 0; seg
< nr_segs
; seg
++) {
1020 read_descriptor_t desc
;
1023 desc
.arg
.buf
= iov
[seg
].iov_base
;
1024 desc
.count
= iov
[seg
].iov_len
;
1025 if (desc
.count
== 0)
1028 do_generic_file_read(filp
,ppos
,&desc
,file_read_actor
);
1029 retval
+= desc
.written
;
1031 retval
= desc
.error
;
1040 EXPORT_SYMBOL(__generic_file_aio_read
);
1043 generic_file_aio_read(struct kiocb
*iocb
, char __user
*buf
, size_t count
, loff_t pos
)
1045 struct iovec local_iov
= { .iov_base
= buf
, .iov_len
= count
};
1047 BUG_ON(iocb
->ki_pos
!= pos
);
1048 return __generic_file_aio_read(iocb
, &local_iov
, 1, &iocb
->ki_pos
);
1051 EXPORT_SYMBOL(generic_file_aio_read
);
1054 generic_file_read(struct file
*filp
, char __user
*buf
, size_t count
, loff_t
*ppos
)
1056 struct iovec local_iov
= { .iov_base
= buf
, .iov_len
= count
};
1060 init_sync_kiocb(&kiocb
, filp
);
1061 ret
= __generic_file_aio_read(&kiocb
, &local_iov
, 1, ppos
);
1062 if (-EIOCBQUEUED
== ret
)
1063 ret
= wait_on_sync_kiocb(&kiocb
);
1067 EXPORT_SYMBOL(generic_file_read
);
1069 int file_send_actor(read_descriptor_t
* desc
, struct page
*page
, unsigned long offset
, unsigned long size
)
1072 unsigned long count
= desc
->count
;
1073 struct file
*file
= desc
->arg
.data
;
1078 written
= file
->f_op
->sendpage(file
, page
, offset
,
1079 size
, &file
->f_pos
, size
<count
);
1081 desc
->error
= written
;
1084 desc
->count
= count
- written
;
1085 desc
->written
+= written
;
1089 ssize_t
generic_file_sendfile(struct file
*in_file
, loff_t
*ppos
,
1090 size_t count
, read_actor_t actor
, void *target
)
1092 read_descriptor_t desc
;
1099 desc
.arg
.data
= target
;
1102 do_generic_file_read(in_file
, ppos
, &desc
, actor
);
1104 return desc
.written
;
1108 EXPORT_SYMBOL(generic_file_sendfile
);
1111 do_readahead(struct address_space
*mapping
, struct file
*filp
,
1112 unsigned long index
, unsigned long nr
)
1114 if (!mapping
|| !mapping
->a_ops
|| !mapping
->a_ops
->readpage
)
1117 force_page_cache_readahead(mapping
, filp
, index
,
1118 max_sane_readahead(nr
));
1122 asmlinkage ssize_t
sys_readahead(int fd
, loff_t offset
, size_t count
)
1130 if (file
->f_mode
& FMODE_READ
) {
1131 struct address_space
*mapping
= file
->f_mapping
;
1132 unsigned long start
= offset
>> PAGE_CACHE_SHIFT
;
1133 unsigned long end
= (offset
+ count
- 1) >> PAGE_CACHE_SHIFT
;
1134 unsigned long len
= end
- start
+ 1;
1135 ret
= do_readahead(mapping
, file
, start
, len
);
1144 * This adds the requested page to the page cache if it isn't already there,
1145 * and schedules an I/O to read in its contents from disk.
1147 static int FASTCALL(page_cache_read(struct file
* file
, unsigned long offset
));
1148 static int fastcall
page_cache_read(struct file
* file
, unsigned long offset
)
1150 struct address_space
*mapping
= file
->f_mapping
;
1154 page
= page_cache_alloc_cold(mapping
);
1158 error
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1160 error
= mapping
->a_ops
->readpage(file
, page
);
1161 page_cache_release(page
);
1166 * We arrive here in the unlikely event that someone
1167 * raced with us and added our page to the cache first
1168 * or we are out of memory for radix-tree nodes.
1170 page_cache_release(page
);
1171 return error
== -EEXIST
? 0 : error
;
1174 #define MMAP_LOTSAMISS (100)
1177 * filemap_nopage() is invoked via the vma operations vector for a
1178 * mapped memory region to read in file data during a page fault.
1180 * The goto's are kind of ugly, but this streamlines the normal case of having
1181 * it in the page cache, and handles the special cases reasonably without
1182 * having a lot of duplicated code.
1184 struct page
*filemap_nopage(struct vm_area_struct
*area
,
1185 unsigned long address
, int *type
)
1188 struct file
*file
= area
->vm_file
;
1189 struct address_space
*mapping
= file
->f_mapping
;
1190 struct file_ra_state
*ra
= &file
->f_ra
;
1191 struct inode
*inode
= mapping
->host
;
1193 unsigned long size
, pgoff
;
1194 int did_readaround
= 0, majmin
= VM_FAULT_MINOR
;
1196 pgoff
= ((address
-area
->vm_start
) >> PAGE_CACHE_SHIFT
) + area
->vm_pgoff
;
1199 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1201 goto outside_data_content
;
1203 /* If we don't want any read-ahead, don't bother */
1204 if (VM_RandomReadHint(area
))
1205 goto no_cached_page
;
1208 * The readahead code wants to be told about each and every page
1209 * so it can build and shrink its windows appropriately
1211 * For sequential accesses, we use the generic readahead logic.
1213 if (VM_SequentialReadHint(area
))
1214 page_cache_readahead(mapping
, ra
, file
, pgoff
, 1);
1217 * Do we have something in the page cache already?
1220 page
= find_get_page(mapping
, pgoff
);
1222 unsigned long ra_pages
;
1224 if (VM_SequentialReadHint(area
)) {
1225 handle_ra_miss(mapping
, ra
, pgoff
);
1226 goto no_cached_page
;
1231 * Do we miss much more than hit in this file? If so,
1232 * stop bothering with read-ahead. It will only hurt.
1234 if (ra
->mmap_miss
> ra
->mmap_hit
+ MMAP_LOTSAMISS
)
1235 goto no_cached_page
;
1238 * To keep the pgmajfault counter straight, we need to
1239 * check did_readaround, as this is an inner loop.
1241 if (!did_readaround
) {
1242 majmin
= VM_FAULT_MAJOR
;
1243 inc_page_state(pgmajfault
);
1246 ra_pages
= max_sane_readahead(file
->f_ra
.ra_pages
);
1250 if (pgoff
> ra_pages
/ 2)
1251 start
= pgoff
- ra_pages
/ 2;
1252 do_page_cache_readahead(mapping
, file
, start
, ra_pages
);
1254 page
= find_get_page(mapping
, pgoff
);
1256 goto no_cached_page
;
1259 if (!did_readaround
)
1263 * Ok, found a page in the page cache, now we need to check
1264 * that it's up-to-date.
1266 if (!PageUptodate(page
))
1267 goto page_not_uptodate
;
1271 * Found the page and have a reference on it.
1273 mark_page_accessed(page
);
1278 outside_data_content
:
1280 * An external ptracer can access pages that normally aren't
1283 if (area
->vm_mm
== current
->mm
)
1285 /* Fall through to the non-read-ahead case */
1288 * We're only likely to ever get here if MADV_RANDOM is in
1291 error
= page_cache_read(file
, pgoff
);
1295 * The page we want has now been added to the page cache.
1296 * In the unlikely event that someone removed it in the
1297 * meantime, we'll just come back here and read it again.
1303 * An error return from page_cache_read can result if the
1304 * system is low on memory, or a problem occurs while trying
1307 if (error
== -ENOMEM
)
1312 if (!did_readaround
) {
1313 majmin
= VM_FAULT_MAJOR
;
1314 inc_page_state(pgmajfault
);
1318 /* Did it get unhashed while we waited for it? */
1319 if (!page
->mapping
) {
1321 page_cache_release(page
);
1325 /* Did somebody else get it up-to-date? */
1326 if (PageUptodate(page
)) {
1331 if (!mapping
->a_ops
->readpage(file
, page
)) {
1332 wait_on_page_locked(page
);
1333 if (PageUptodate(page
))
1338 * Umm, take care of errors if the page isn't up-to-date.
1339 * Try to re-read it _once_. We do this synchronously,
1340 * because there really aren't any performance issues here
1341 * and we need to check for errors.
1345 /* Somebody truncated the page on us? */
1346 if (!page
->mapping
) {
1348 page_cache_release(page
);
1352 /* Somebody else successfully read it in? */
1353 if (PageUptodate(page
)) {
1357 ClearPageError(page
);
1358 if (!mapping
->a_ops
->readpage(file
, page
)) {
1359 wait_on_page_locked(page
);
1360 if (PageUptodate(page
))
1365 * Things didn't work out. Return zero to tell the
1366 * mm layer so, possibly freeing the page cache page first.
1368 page_cache_release(page
);
1372 EXPORT_SYMBOL(filemap_nopage
);
1374 static struct page
* filemap_getpage(struct file
*file
, unsigned long pgoff
,
1377 struct address_space
*mapping
= file
->f_mapping
;
1382 * Do we have something in the page cache already?
1385 page
= find_get_page(mapping
, pgoff
);
1389 goto no_cached_page
;
1393 * Ok, found a page in the page cache, now we need to check
1394 * that it's up-to-date.
1396 if (!PageUptodate(page
)) {
1398 page_cache_release(page
);
1401 goto page_not_uptodate
;
1406 * Found the page and have a reference on it.
1408 mark_page_accessed(page
);
1412 error
= page_cache_read(file
, pgoff
);
1415 * The page we want has now been added to the page cache.
1416 * In the unlikely event that someone removed it in the
1417 * meantime, we'll just come back here and read it again.
1423 * An error return from page_cache_read can result if the
1424 * system is low on memory, or a problem occurs while trying
1432 /* Did it get unhashed while we waited for it? */
1433 if (!page
->mapping
) {
1438 /* Did somebody else get it up-to-date? */
1439 if (PageUptodate(page
)) {
1444 if (!mapping
->a_ops
->readpage(file
, page
)) {
1445 wait_on_page_locked(page
);
1446 if (PageUptodate(page
))
1451 * Umm, take care of errors if the page isn't up-to-date.
1452 * Try to re-read it _once_. We do this synchronously,
1453 * because there really aren't any performance issues here
1454 * and we need to check for errors.
1458 /* Somebody truncated the page on us? */
1459 if (!page
->mapping
) {
1463 /* Somebody else successfully read it in? */
1464 if (PageUptodate(page
)) {
1469 ClearPageError(page
);
1470 if (!mapping
->a_ops
->readpage(file
, page
)) {
1471 wait_on_page_locked(page
);
1472 if (PageUptodate(page
))
1477 * Things didn't work out. Return zero to tell the
1478 * mm layer so, possibly freeing the page cache page first.
1481 page_cache_release(page
);
1486 int filemap_populate(struct vm_area_struct
*vma
, unsigned long addr
,
1487 unsigned long len
, pgprot_t prot
, unsigned long pgoff
,
1490 struct file
*file
= vma
->vm_file
;
1491 struct address_space
*mapping
= file
->f_mapping
;
1492 struct inode
*inode
= mapping
->host
;
1494 struct mm_struct
*mm
= vma
->vm_mm
;
1499 force_page_cache_readahead(mapping
, vma
->vm_file
,
1500 pgoff
, len
>> PAGE_CACHE_SHIFT
);
1503 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1504 if (pgoff
+ (len
>> PAGE_CACHE_SHIFT
) > size
)
1507 page
= filemap_getpage(file
, pgoff
, nonblock
);
1508 if (!page
&& !nonblock
)
1511 err
= install_page(mm
, vma
, addr
, page
, prot
);
1513 page_cache_release(page
);
1517 err
= install_file_pte(mm
, vma
, addr
, pgoff
, prot
);
1531 struct vm_operations_struct generic_file_vm_ops
= {
1532 .nopage
= filemap_nopage
,
1533 .populate
= filemap_populate
,
1536 /* This is used for a general mmap of a disk file */
1538 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1540 struct address_space
*mapping
= file
->f_mapping
;
1542 if (!mapping
->a_ops
->readpage
)
1544 file_accessed(file
);
1545 vma
->vm_ops
= &generic_file_vm_ops
;
1548 EXPORT_SYMBOL(filemap_populate
);
1551 * This is for filesystems which do not implement ->writepage.
1553 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1555 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
1557 return generic_file_mmap(file
, vma
);
1560 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1564 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1568 #endif /* CONFIG_MMU */
1570 EXPORT_SYMBOL(generic_file_mmap
);
1571 EXPORT_SYMBOL(generic_file_readonly_mmap
);
1573 static inline struct page
*__read_cache_page(struct address_space
*mapping
,
1574 unsigned long index
,
1575 int (*filler
)(void *,struct page
*),
1578 struct page
*page
, *cached_page
= NULL
;
1581 page
= find_get_page(mapping
, index
);
1584 cached_page
= page_cache_alloc_cold(mapping
);
1586 return ERR_PTR(-ENOMEM
);
1588 err
= add_to_page_cache_lru(cached_page
, mapping
,
1593 /* Presumably ENOMEM for radix tree node */
1594 page_cache_release(cached_page
);
1595 return ERR_PTR(err
);
1599 err
= filler(data
, page
);
1601 page_cache_release(page
);
1602 page
= ERR_PTR(err
);
1606 page_cache_release(cached_page
);
1611 * Read into the page cache. If a page already exists,
1612 * and PageUptodate() is not set, try to fill the page.
1614 struct page
*read_cache_page(struct address_space
*mapping
,
1615 unsigned long index
,
1616 int (*filler
)(void *,struct page
*),
1623 page
= __read_cache_page(mapping
, index
, filler
, data
);
1626 mark_page_accessed(page
);
1627 if (PageUptodate(page
))
1631 if (!page
->mapping
) {
1633 page_cache_release(page
);
1636 if (PageUptodate(page
)) {
1640 err
= filler(data
, page
);
1642 page_cache_release(page
);
1643 page
= ERR_PTR(err
);
1649 EXPORT_SYMBOL(read_cache_page
);
1652 * If the page was newly created, increment its refcount and add it to the
1653 * caller's lru-buffering pagevec. This function is specifically for
1654 * generic_file_write().
1656 static inline struct page
*
1657 __grab_cache_page(struct address_space
*mapping
, unsigned long index
,
1658 struct page
**cached_page
, struct pagevec
*lru_pvec
)
1663 page
= find_lock_page(mapping
, index
);
1665 if (!*cached_page
) {
1666 *cached_page
= page_cache_alloc(mapping
);
1670 err
= add_to_page_cache(*cached_page
, mapping
,
1675 page
= *cached_page
;
1676 page_cache_get(page
);
1677 if (!pagevec_add(lru_pvec
, page
))
1678 __pagevec_lru_add(lru_pvec
);
1679 *cached_page
= NULL
;
1686 * The logic we want is
1688 * if suid or (sgid and xgrp)
1691 int remove_suid(struct dentry
*dentry
)
1693 mode_t mode
= dentry
->d_inode
->i_mode
;
1697 /* suid always must be killed */
1698 if (unlikely(mode
& S_ISUID
))
1699 kill
= ATTR_KILL_SUID
;
1702 * sgid without any exec bits is just a mandatory locking mark; leave
1703 * it alone. If some exec bits are set, it's a real sgid; kill it.
1705 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1706 kill
|= ATTR_KILL_SGID
;
1708 if (unlikely(kill
&& !capable(CAP_FSETID
))) {
1709 struct iattr newattrs
;
1711 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1712 result
= notify_change(dentry
, &newattrs
);
1716 EXPORT_SYMBOL(remove_suid
);
1719 __filemap_copy_from_user_iovec(char *vaddr
,
1720 const struct iovec
*iov
, size_t base
, size_t bytes
)
1722 size_t copied
= 0, left
= 0;
1725 char __user
*buf
= iov
->iov_base
+ base
;
1726 int copy
= min(bytes
, iov
->iov_len
- base
);
1729 left
= __copy_from_user_inatomic(vaddr
, buf
, copy
);
1735 if (unlikely(left
)) {
1736 /* zero the rest of the target like __copy_from_user */
1738 memset(vaddr
, 0, bytes
);
1742 return copied
- left
;
1746 * Performs necessary checks before doing a write
1748 * Can adjust writing position aor amount of bytes to write.
1749 * Returns appropriate error code that caller should return or
1750 * zero in case that write should be allowed.
1752 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
1754 struct inode
*inode
= file
->f_mapping
->host
;
1755 unsigned long limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1757 if (unlikely(*pos
< 0))
1761 /* FIXME: this is for backwards compatibility with 2.4 */
1762 if (file
->f_flags
& O_APPEND
)
1763 *pos
= i_size_read(inode
);
1765 if (limit
!= RLIM_INFINITY
) {
1766 if (*pos
>= limit
) {
1767 send_sig(SIGXFSZ
, current
, 0);
1770 if (*count
> limit
- (typeof(limit
))*pos
) {
1771 *count
= limit
- (typeof(limit
))*pos
;
1779 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
1780 !(file
->f_flags
& O_LARGEFILE
))) {
1781 if (*pos
>= MAX_NON_LFS
) {
1782 send_sig(SIGXFSZ
, current
, 0);
1785 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
1786 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
1791 * Are we about to exceed the fs block limit ?
1793 * If we have written data it becomes a short write. If we have
1794 * exceeded without writing data we send a signal and return EFBIG.
1795 * Linus frestrict idea will clean these up nicely..
1797 if (likely(!isblk
)) {
1798 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
1799 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
1800 send_sig(SIGXFSZ
, current
, 0);
1803 /* zero-length writes at ->s_maxbytes are OK */
1806 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
1807 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
1810 if (bdev_read_only(I_BDEV(inode
)))
1812 isize
= i_size_read(inode
);
1813 if (*pos
>= isize
) {
1814 if (*count
|| *pos
> isize
)
1818 if (*pos
+ *count
> isize
)
1819 *count
= isize
- *pos
;
1823 EXPORT_SYMBOL(generic_write_checks
);
1826 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1827 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
1828 size_t count
, size_t ocount
)
1830 struct file
*file
= iocb
->ki_filp
;
1831 struct address_space
*mapping
= file
->f_mapping
;
1832 struct inode
*inode
= mapping
->host
;
1835 if (count
!= ocount
)
1836 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
1838 written
= generic_file_direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
1840 loff_t end
= pos
+ written
;
1841 if (end
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
1842 i_size_write(inode
, end
);
1843 mark_inode_dirty(inode
);
1849 * Sync the fs metadata but not the minor inode changes and
1850 * of course not the data as we did direct DMA for the IO.
1851 * i_sem is held, which protects generic_osync_inode() from
1854 if (written
>= 0 && file
->f_flags
& O_SYNC
)
1855 generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
1856 if (written
== count
&& !is_sync_kiocb(iocb
))
1857 written
= -EIOCBQUEUED
;
1860 EXPORT_SYMBOL(generic_file_direct_write
);
1863 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1864 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
1865 size_t count
, ssize_t written
)
1867 struct file
*file
= iocb
->ki_filp
;
1868 struct address_space
* mapping
= file
->f_mapping
;
1869 struct address_space_operations
*a_ops
= mapping
->a_ops
;
1870 struct inode
*inode
= mapping
->host
;
1873 struct page
*cached_page
= NULL
;
1875 struct pagevec lru_pvec
;
1876 const struct iovec
*cur_iov
= iov
; /* current iovec */
1877 size_t iov_base
= 0; /* offset in the current iovec */
1880 pagevec_init(&lru_pvec
, 0);
1883 * handle partial DIO write. Adjust cur_iov if needed.
1885 if (likely(nr_segs
== 1))
1886 buf
= iov
->iov_base
+ written
;
1888 filemap_set_next_iovec(&cur_iov
, &iov_base
, written
);
1889 buf
= cur_iov
->iov_base
+ iov_base
;
1893 unsigned long index
;
1894 unsigned long offset
;
1895 unsigned long maxlen
;
1898 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
1899 index
= pos
>> PAGE_CACHE_SHIFT
;
1900 bytes
= PAGE_CACHE_SIZE
- offset
;
1905 * Bring in the user page that we will copy from _first_.
1906 * Otherwise there's a nasty deadlock on copying from the
1907 * same page as we're writing to, without it being marked
1910 maxlen
= cur_iov
->iov_len
- iov_base
;
1913 fault_in_pages_readable(buf
, maxlen
);
1915 page
= __grab_cache_page(mapping
,index
,&cached_page
,&lru_pvec
);
1921 status
= a_ops
->prepare_write(file
, page
, offset
, offset
+bytes
);
1922 if (unlikely(status
)) {
1923 loff_t isize
= i_size_read(inode
);
1925 * prepare_write() may have instantiated a few blocks
1926 * outside i_size. Trim these off again.
1929 page_cache_release(page
);
1930 if (pos
+ bytes
> isize
)
1931 vmtruncate(inode
, isize
);
1934 if (likely(nr_segs
== 1))
1935 copied
= filemap_copy_from_user(page
, offset
,
1938 copied
= filemap_copy_from_user_iovec(page
, offset
,
1939 cur_iov
, iov_base
, bytes
);
1940 flush_dcache_page(page
);
1941 status
= a_ops
->commit_write(file
, page
, offset
, offset
+bytes
);
1942 if (likely(copied
> 0)) {
1951 if (unlikely(nr_segs
> 1)) {
1952 filemap_set_next_iovec(&cur_iov
,
1954 buf
= cur_iov
->iov_base
+ iov_base
;
1960 if (unlikely(copied
!= bytes
))
1964 mark_page_accessed(page
);
1965 page_cache_release(page
);
1968 balance_dirty_pages_ratelimited(mapping
);
1974 page_cache_release(cached_page
);
1977 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
1979 if (likely(status
>= 0)) {
1980 if (unlikely((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
1981 if (!a_ops
->writepage
|| !is_sync_kiocb(iocb
))
1982 status
= generic_osync_inode(inode
, mapping
,
1983 OSYNC_METADATA
|OSYNC_DATA
);
1988 * If we get here for O_DIRECT writes then we must have fallen through
1989 * to buffered writes (block instantiation inside i_size). So we sync
1990 * the file data here, to try to honour O_DIRECT expectations.
1992 if (unlikely(file
->f_flags
& O_DIRECT
) && written
)
1993 status
= filemap_write_and_wait(mapping
);
1995 pagevec_lru_add(&lru_pvec
);
1996 return written
? written
: status
;
1998 EXPORT_SYMBOL(generic_file_buffered_write
);
2001 __generic_file_aio_write_nolock(struct kiocb
*iocb
, const struct iovec
*iov
,
2002 unsigned long nr_segs
, loff_t
*ppos
)
2004 struct file
*file
= iocb
->ki_filp
;
2005 struct address_space
* mapping
= file
->f_mapping
;
2006 size_t ocount
; /* original count */
2007 size_t count
; /* after file limit checks */
2008 struct inode
*inode
= mapping
->host
;
2015 for (seg
= 0; seg
< nr_segs
; seg
++) {
2016 const struct iovec
*iv
= &iov
[seg
];
2019 * If any segment has a negative length, or the cumulative
2020 * length ever wraps negative then return -EINVAL.
2022 ocount
+= iv
->iov_len
;
2023 if (unlikely((ssize_t
)(ocount
|iv
->iov_len
) < 0))
2025 if (access_ok(VERIFY_READ
, iv
->iov_base
, iv
->iov_len
))
2030 ocount
-= iv
->iov_len
; /* This segment is no good */
2037 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2039 /* We can write back this queue in page reclaim */
2040 current
->backing_dev_info
= mapping
->backing_dev_info
;
2043 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2050 err
= remove_suid(file
->f_dentry
);
2054 inode_update_time(inode
, 1);
2056 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2057 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2058 written
= generic_file_direct_write(iocb
, iov
,
2059 &nr_segs
, pos
, ppos
, count
, ocount
);
2060 if (written
< 0 || written
== count
)
2063 * direct-io write to a hole: fall through to buffered I/O
2064 * for completing the rest of the request.
2070 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2071 pos
, ppos
, count
, written
);
2073 current
->backing_dev_info
= NULL
;
2074 return written
? written
: err
;
2076 EXPORT_SYMBOL(generic_file_aio_write_nolock
);
2079 generic_file_aio_write_nolock(struct kiocb
*iocb
, const struct iovec
*iov
,
2080 unsigned long nr_segs
, loff_t
*ppos
)
2082 struct file
*file
= iocb
->ki_filp
;
2083 struct address_space
*mapping
= file
->f_mapping
;
2084 struct inode
*inode
= mapping
->host
;
2088 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
, ppos
);
2090 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2093 err
= sync_page_range_nolock(inode
, mapping
, pos
, ret
);
2101 __generic_file_write_nolock(struct file
*file
, const struct iovec
*iov
,
2102 unsigned long nr_segs
, loff_t
*ppos
)
2107 init_sync_kiocb(&kiocb
, file
);
2108 ret
= __generic_file_aio_write_nolock(&kiocb
, iov
, nr_segs
, ppos
);
2109 if (ret
== -EIOCBQUEUED
)
2110 ret
= wait_on_sync_kiocb(&kiocb
);
2115 generic_file_write_nolock(struct file
*file
, const struct iovec
*iov
,
2116 unsigned long nr_segs
, loff_t
*ppos
)
2121 init_sync_kiocb(&kiocb
, file
);
2122 ret
= generic_file_aio_write_nolock(&kiocb
, iov
, nr_segs
, ppos
);
2123 if (-EIOCBQUEUED
== ret
)
2124 ret
= wait_on_sync_kiocb(&kiocb
);
2127 EXPORT_SYMBOL(generic_file_write_nolock
);
2129 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const char __user
*buf
,
2130 size_t count
, loff_t pos
)
2132 struct file
*file
= iocb
->ki_filp
;
2133 struct address_space
*mapping
= file
->f_mapping
;
2134 struct inode
*inode
= mapping
->host
;
2136 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
2139 BUG_ON(iocb
->ki_pos
!= pos
);
2141 down(&inode
->i_sem
);
2142 ret
= __generic_file_aio_write_nolock(iocb
, &local_iov
, 1,
2146 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2149 err
= sync_page_range(inode
, mapping
, pos
, ret
);
2155 EXPORT_SYMBOL(generic_file_aio_write
);
2157 ssize_t
generic_file_write(struct file
*file
, const char __user
*buf
,
2158 size_t count
, loff_t
*ppos
)
2160 struct address_space
*mapping
= file
->f_mapping
;
2161 struct inode
*inode
= mapping
->host
;
2163 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
2166 down(&inode
->i_sem
);
2167 ret
= __generic_file_write_nolock(file
, &local_iov
, 1, ppos
);
2170 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2173 err
= sync_page_range(inode
, mapping
, *ppos
- ret
, ret
);
2179 EXPORT_SYMBOL(generic_file_write
);
2181 ssize_t
generic_file_readv(struct file
*filp
, const struct iovec
*iov
,
2182 unsigned long nr_segs
, loff_t
*ppos
)
2187 init_sync_kiocb(&kiocb
, filp
);
2188 ret
= __generic_file_aio_read(&kiocb
, iov
, nr_segs
, ppos
);
2189 if (-EIOCBQUEUED
== ret
)
2190 ret
= wait_on_sync_kiocb(&kiocb
);
2193 EXPORT_SYMBOL(generic_file_readv
);
2195 ssize_t
generic_file_writev(struct file
*file
, const struct iovec
*iov
,
2196 unsigned long nr_segs
, loff_t
*ppos
)
2198 struct address_space
*mapping
= file
->f_mapping
;
2199 struct inode
*inode
= mapping
->host
;
2202 down(&inode
->i_sem
);
2203 ret
= __generic_file_write_nolock(file
, iov
, nr_segs
, ppos
);
2206 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2209 err
= sync_page_range(inode
, mapping
, *ppos
- ret
, ret
);
2215 EXPORT_SYMBOL(generic_file_writev
);
2218 * Called under i_sem for writes to S_ISREG files. Returns -EIO if something
2219 * went wrong during pagecache shootdown.
2222 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
2223 loff_t offset
, unsigned long nr_segs
)
2225 struct file
*file
= iocb
->ki_filp
;
2226 struct address_space
*mapping
= file
->f_mapping
;
2228 size_t write_len
= 0;
2231 * If it's a write, unmap all mmappings of the file up-front. This
2232 * will cause any pte dirty bits to be propagated into the pageframes
2233 * for the subsequent filemap_write_and_wait().
2236 write_len
= iov_length(iov
, nr_segs
);
2237 if (mapping_mapped(mapping
))
2238 unmap_mapping_range(mapping
, offset
, write_len
, 0);
2241 retval
= filemap_write_and_wait(mapping
);
2243 retval
= mapping
->a_ops
->direct_IO(rw
, iocb
, iov
,
2245 if (rw
== WRITE
&& mapping
->nrpages
) {
2246 pgoff_t end
= (offset
+ write_len
- 1)
2247 >> PAGE_CACHE_SHIFT
;
2248 int err
= invalidate_inode_pages2_range(mapping
,
2249 offset
>> PAGE_CACHE_SHIFT
, end
);
2256 EXPORT_SYMBOL_GPL(generic_file_direct_IO
);