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/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>
36 * FIXME: remove all knowledge of the buffer layer from the core VM
38 #include <linux/buffer_head.h> /* for generic_osync_inode */
40 #include <asm/uaccess.h>
44 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
45 loff_t offset
, unsigned long nr_segs
);
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * Shared mappings now work. 15.8.1995 Bruno.
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
64 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
68 * ->i_mmap_lock (truncate->unmap_mapping_range)
72 * ->page_table_lock or pte_lock (various, mainly in memory.c)
73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 * ->lock_page (access_process_vm)
82 * ->i_alloc_sem (various)
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->dcache_lock (proc_pid_lookup)
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
115 void __remove_from_page_cache(struct page
*page
)
117 struct address_space
*mapping
= page
->mapping
;
119 radix_tree_delete(&mapping
->page_tree
, page
->index
);
120 page
->mapping
= NULL
;
125 void remove_from_page_cache(struct page
*page
)
127 struct address_space
*mapping
= page
->mapping
;
129 BUG_ON(!PageLocked(page
));
131 write_lock_irq(&mapping
->tree_lock
);
132 __remove_from_page_cache(page
);
133 write_unlock_irq(&mapping
->tree_lock
);
136 static int sync_page(void *word
)
138 struct address_space
*mapping
;
141 page
= container_of((unsigned long *)word
, struct page
, flags
);
144 * page_mapping() is being called without PG_locked held.
145 * Some knowledge of the state and use of the page is used to
146 * reduce the requirements down to a memory barrier.
147 * The danger here is of a stale page_mapping() return value
148 * indicating a struct address_space different from the one it's
149 * associated with when it is associated with one.
150 * After smp_mb(), it's either the correct page_mapping() for
151 * the page, or an old page_mapping() and the page's own
152 * page_mapping() has gone NULL.
153 * The ->sync_page() address_space operation must tolerate
154 * page_mapping() going NULL. By an amazing coincidence,
155 * this comes about because none of the users of the page
156 * in the ->sync_page() methods make essential use of the
157 * page_mapping(), merely passing the page down to the backing
158 * device's unplug functions when it's non-NULL, which in turn
159 * ignore it for all cases but swap, where only page_private(page) is
160 * of interest. When page_mapping() does go NULL, the entire
161 * call stack gracefully ignores the page and returns.
165 mapping
= page_mapping(page
);
166 if (mapping
&& mapping
->a_ops
&& mapping
->a_ops
->sync_page
)
167 mapping
->a_ops
->sync_page(page
);
173 * filemap_fdatawrite_range - start writeback against all of a mapping's
174 * dirty pages that lie within the byte offsets <start, end>
175 * @mapping: address space structure to write
176 * @start: offset in bytes where the range starts
177 * @end: offset in bytes where the range ends
178 * @sync_mode: enable synchronous operation
180 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
181 * opposed to a regular memory * cleansing writeback. The difference between
182 * these two operations is that if a dirty page/buffer is encountered, it must
183 * be waited upon, and not just skipped over.
185 static int __filemap_fdatawrite_range(struct address_space
*mapping
,
186 loff_t start
, loff_t end
, int sync_mode
)
189 struct writeback_control wbc
= {
190 .sync_mode
= sync_mode
,
191 .nr_to_write
= mapping
->nrpages
* 2,
196 if (!mapping_cap_writeback_dirty(mapping
))
199 ret
= do_writepages(mapping
, &wbc
);
203 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
206 return __filemap_fdatawrite_range(mapping
, 0, 0, sync_mode
);
209 int filemap_fdatawrite(struct address_space
*mapping
)
211 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
213 EXPORT_SYMBOL(filemap_fdatawrite
);
215 static int filemap_fdatawrite_range(struct address_space
*mapping
,
216 loff_t start
, loff_t end
)
218 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
222 * This is a mostly non-blocking flush. Not suitable for data-integrity
223 * purposes - I/O may not be started against all dirty pages.
225 int filemap_flush(struct address_space
*mapping
)
227 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
229 EXPORT_SYMBOL(filemap_flush
);
232 * Wait for writeback to complete against pages indexed by start->end
235 static int wait_on_page_writeback_range(struct address_space
*mapping
,
236 pgoff_t start
, pgoff_t end
)
246 pagevec_init(&pvec
, 0);
248 while ((index
<= end
) &&
249 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
250 PAGECACHE_TAG_WRITEBACK
,
251 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
254 for (i
= 0; i
< nr_pages
; i
++) {
255 struct page
*page
= pvec
.pages
[i
];
257 /* until radix tree lookup accepts end_index */
258 if (page
->index
> end
)
261 wait_on_page_writeback(page
);
265 pagevec_release(&pvec
);
269 /* Check for outstanding write errors */
270 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
272 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
279 * Write and wait upon all the pages in the passed range. This is a "data
280 * integrity" operation. It waits upon in-flight writeout before starting and
281 * waiting upon new writeout. If there was an IO error, return it.
283 * We need to re-take i_mutex during the generic_osync_inode list walk because
284 * it is otherwise livelockable.
286 int sync_page_range(struct inode
*inode
, struct address_space
*mapping
,
287 loff_t pos
, loff_t count
)
289 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
290 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
293 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
295 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
297 mutex_lock(&inode
->i_mutex
);
298 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
299 mutex_unlock(&inode
->i_mutex
);
302 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
305 EXPORT_SYMBOL(sync_page_range
);
308 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
309 * as it forces O_SYNC writers to different parts of the same file
310 * to be serialised right until io completion.
312 int sync_page_range_nolock(struct inode
*inode
, struct address_space
*mapping
,
313 loff_t pos
, loff_t count
)
315 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
316 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
319 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
321 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
323 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
325 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
328 EXPORT_SYMBOL(sync_page_range_nolock
);
331 * filemap_fdatawait - walk the list of under-writeback pages of the given
332 * address space and wait for all of them.
334 * @mapping: address space structure to wait for
336 int filemap_fdatawait(struct address_space
*mapping
)
338 loff_t i_size
= i_size_read(mapping
->host
);
343 return wait_on_page_writeback_range(mapping
, 0,
344 (i_size
- 1) >> PAGE_CACHE_SHIFT
);
346 EXPORT_SYMBOL(filemap_fdatawait
);
348 int filemap_write_and_wait(struct address_space
*mapping
)
352 if (mapping
->nrpages
) {
353 err
= filemap_fdatawrite(mapping
);
355 * Even if the above returned error, the pages may be
356 * written partially (e.g. -ENOSPC), so we wait for it.
357 * But the -EIO is special case, it may indicate the worst
358 * thing (e.g. bug) happened, so we avoid waiting for it.
361 int err2
= filemap_fdatawait(mapping
);
368 EXPORT_SYMBOL(filemap_write_and_wait
);
370 int filemap_write_and_wait_range(struct address_space
*mapping
,
371 loff_t lstart
, loff_t lend
)
375 if (mapping
->nrpages
) {
376 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
378 /* See comment of filemap_write_and_wait() */
380 int err2
= wait_on_page_writeback_range(mapping
,
381 lstart
>> PAGE_CACHE_SHIFT
,
382 lend
>> PAGE_CACHE_SHIFT
);
391 * This function is used to add newly allocated pagecache pages:
392 * the page is new, so we can just run SetPageLocked() against it.
393 * The other page state flags were set by rmqueue().
395 * This function does not add the page to the LRU. The caller must do that.
397 int add_to_page_cache(struct page
*page
, struct address_space
*mapping
,
398 pgoff_t offset
, gfp_t gfp_mask
)
400 int error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
403 write_lock_irq(&mapping
->tree_lock
);
404 error
= radix_tree_insert(&mapping
->page_tree
, offset
, page
);
406 page_cache_get(page
);
408 page
->mapping
= mapping
;
409 page
->index
= offset
;
413 write_unlock_irq(&mapping
->tree_lock
);
414 radix_tree_preload_end();
419 EXPORT_SYMBOL(add_to_page_cache
);
421 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
422 pgoff_t offset
, gfp_t gfp_mask
)
424 int ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
431 * In order to wait for pages to become available there must be
432 * waitqueues associated with pages. By using a hash table of
433 * waitqueues where the bucket discipline is to maintain all
434 * waiters on the same queue and wake all when any of the pages
435 * become available, and for the woken contexts to check to be
436 * sure the appropriate page became available, this saves space
437 * at a cost of "thundering herd" phenomena during rare hash
440 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
442 const struct zone
*zone
= page_zone(page
);
444 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
447 static inline void wake_up_page(struct page
*page
, int bit
)
449 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
452 void fastcall
wait_on_page_bit(struct page
*page
, int bit_nr
)
454 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
456 if (test_bit(bit_nr
, &page
->flags
))
457 __wait_on_bit(page_waitqueue(page
), &wait
, sync_page
,
458 TASK_UNINTERRUPTIBLE
);
460 EXPORT_SYMBOL(wait_on_page_bit
);
463 * unlock_page() - unlock a locked page
467 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
468 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
469 * mechananism between PageLocked pages and PageWriteback pages is shared.
470 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
472 * The first mb is necessary to safely close the critical section opened by the
473 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
474 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
475 * parallel wait_on_page_locked()).
477 void fastcall
unlock_page(struct page
*page
)
479 smp_mb__before_clear_bit();
480 if (!TestClearPageLocked(page
))
482 smp_mb__after_clear_bit();
483 wake_up_page(page
, PG_locked
);
485 EXPORT_SYMBOL(unlock_page
);
488 * End writeback against a page.
490 void end_page_writeback(struct page
*page
)
492 if (!TestClearPageReclaim(page
) || rotate_reclaimable_page(page
)) {
493 if (!test_clear_page_writeback(page
))
496 smp_mb__after_clear_bit();
497 wake_up_page(page
, PG_writeback
);
499 EXPORT_SYMBOL(end_page_writeback
);
502 * Get a lock on the page, assuming we need to sleep to get it.
504 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
505 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
506 * chances are that on the second loop, the block layer's plug list is empty,
507 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
509 void fastcall
__lock_page(struct page
*page
)
511 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
513 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sync_page
,
514 TASK_UNINTERRUPTIBLE
);
516 EXPORT_SYMBOL(__lock_page
);
519 * a rather lightweight function, finding and getting a reference to a
520 * hashed page atomically.
522 struct page
* find_get_page(struct address_space
*mapping
, unsigned long offset
)
526 read_lock_irq(&mapping
->tree_lock
);
527 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
529 page_cache_get(page
);
530 read_unlock_irq(&mapping
->tree_lock
);
534 EXPORT_SYMBOL(find_get_page
);
537 * Same as above, but trylock it instead of incrementing the count.
539 struct page
*find_trylock_page(struct address_space
*mapping
, unsigned long offset
)
543 read_lock_irq(&mapping
->tree_lock
);
544 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
545 if (page
&& TestSetPageLocked(page
))
547 read_unlock_irq(&mapping
->tree_lock
);
551 EXPORT_SYMBOL(find_trylock_page
);
554 * find_lock_page - locate, pin and lock a pagecache page
556 * @mapping: the address_space to search
557 * @offset: the page index
559 * Locates the desired pagecache page, locks it, increments its reference
560 * count and returns its address.
562 * Returns zero if the page was not present. find_lock_page() may sleep.
564 struct page
*find_lock_page(struct address_space
*mapping
,
565 unsigned long offset
)
569 read_lock_irq(&mapping
->tree_lock
);
571 page
= radix_tree_lookup(&mapping
->page_tree
, offset
);
573 page_cache_get(page
);
574 if (TestSetPageLocked(page
)) {
575 read_unlock_irq(&mapping
->tree_lock
);
577 read_lock_irq(&mapping
->tree_lock
);
579 /* Has the page been truncated while we slept? */
580 if (unlikely(page
->mapping
!= mapping
||
581 page
->index
!= offset
)) {
583 page_cache_release(page
);
588 read_unlock_irq(&mapping
->tree_lock
);
592 EXPORT_SYMBOL(find_lock_page
);
595 * find_or_create_page - locate or add a pagecache page
597 * @mapping: the page's address_space
598 * @index: the page's index into the mapping
599 * @gfp_mask: page allocation mode
601 * Locates a page in the pagecache. If the page is not present, a new page
602 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
603 * LRU list. The returned page is locked and has its reference count
606 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
609 * find_or_create_page() returns the desired page's address, or zero on
612 struct page
*find_or_create_page(struct address_space
*mapping
,
613 unsigned long index
, gfp_t gfp_mask
)
615 struct page
*page
, *cached_page
= NULL
;
618 page
= find_lock_page(mapping
, index
);
621 cached_page
= alloc_page(gfp_mask
);
625 err
= add_to_page_cache_lru(cached_page
, mapping
,
630 } else if (err
== -EEXIST
)
634 page_cache_release(cached_page
);
638 EXPORT_SYMBOL(find_or_create_page
);
641 * find_get_pages - gang pagecache lookup
642 * @mapping: The address_space to search
643 * @start: The starting page index
644 * @nr_pages: The maximum number of pages
645 * @pages: Where the resulting pages are placed
647 * find_get_pages() will search for and return a group of up to
648 * @nr_pages pages in the mapping. The pages are placed at @pages.
649 * find_get_pages() takes a reference against the returned pages.
651 * The search returns a group of mapping-contiguous pages with ascending
652 * indexes. There may be holes in the indices due to not-present pages.
654 * find_get_pages() returns the number of pages which were found.
656 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
657 unsigned int nr_pages
, struct page
**pages
)
662 read_lock_irq(&mapping
->tree_lock
);
663 ret
= radix_tree_gang_lookup(&mapping
->page_tree
,
664 (void **)pages
, start
, nr_pages
);
665 for (i
= 0; i
< ret
; i
++)
666 page_cache_get(pages
[i
]);
667 read_unlock_irq(&mapping
->tree_lock
);
672 * Like find_get_pages, except we only return pages which are tagged with
673 * `tag'. We update *index to index the next page for the traversal.
675 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
676 int tag
, unsigned int nr_pages
, struct page
**pages
)
681 read_lock_irq(&mapping
->tree_lock
);
682 ret
= radix_tree_gang_lookup_tag(&mapping
->page_tree
,
683 (void **)pages
, *index
, nr_pages
, tag
);
684 for (i
= 0; i
< ret
; i
++)
685 page_cache_get(pages
[i
]);
687 *index
= pages
[ret
- 1]->index
+ 1;
688 read_unlock_irq(&mapping
->tree_lock
);
693 * Same as grab_cache_page, but do not wait if the page is unavailable.
694 * This is intended for speculative data generators, where the data can
695 * be regenerated if the page couldn't be grabbed. This routine should
696 * be safe to call while holding the lock for another page.
698 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
699 * and deadlock against the caller's locked page.
702 grab_cache_page_nowait(struct address_space
*mapping
, unsigned long index
)
704 struct page
*page
= find_get_page(mapping
, index
);
708 if (!TestSetPageLocked(page
))
710 page_cache_release(page
);
713 gfp_mask
= mapping_gfp_mask(mapping
) & ~__GFP_FS
;
714 page
= alloc_pages(gfp_mask
, 0);
715 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, gfp_mask
)) {
716 page_cache_release(page
);
722 EXPORT_SYMBOL(grab_cache_page_nowait
);
725 * This is a generic file read routine, and uses the
726 * mapping->a_ops->readpage() function for the actual low-level
729 * This is really ugly. But the goto's actually try to clarify some
730 * of the logic when it comes to error handling etc.
732 * Note the struct file* is only passed for the use of readpage. It may be
735 void do_generic_mapping_read(struct address_space
*mapping
,
736 struct file_ra_state
*_ra
,
739 read_descriptor_t
*desc
,
742 struct inode
*inode
= mapping
->host
;
744 unsigned long end_index
;
745 unsigned long offset
;
746 unsigned long last_index
;
747 unsigned long next_index
;
748 unsigned long prev_index
;
750 struct page
*cached_page
;
752 struct file_ra_state ra
= *_ra
;
755 index
= *ppos
>> PAGE_CACHE_SHIFT
;
757 prev_index
= ra
.prev_page
;
758 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
759 offset
= *ppos
& ~PAGE_CACHE_MASK
;
761 isize
= i_size_read(inode
);
765 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
768 unsigned long nr
, ret
;
770 /* nr is the maximum number of bytes to copy from this page */
771 nr
= PAGE_CACHE_SIZE
;
772 if (index
>= end_index
) {
773 if (index
> end_index
)
775 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
783 if (index
== next_index
)
784 next_index
= page_cache_readahead(mapping
, &ra
, filp
,
785 index
, last_index
- index
);
788 page
= find_get_page(mapping
, index
);
789 if (unlikely(page
== NULL
)) {
790 handle_ra_miss(mapping
, &ra
, index
);
793 if (!PageUptodate(page
))
794 goto page_not_up_to_date
;
797 /* If users can be writing to this page using arbitrary
798 * virtual addresses, take care about potential aliasing
799 * before reading the page on the kernel side.
801 if (mapping_writably_mapped(mapping
))
802 flush_dcache_page(page
);
805 * When (part of) the same page is read multiple times
806 * in succession, only mark it as accessed the first time.
808 if (prev_index
!= index
)
809 mark_page_accessed(page
);
813 * Ok, we have the page, and it's up-to-date, so
814 * now we can copy it to user space...
816 * The actor routine returns how many bytes were actually used..
817 * NOTE! This may not be the same as how much of a user buffer
818 * we filled up (we may be padding etc), so we can only update
819 * "pos" here (the actor routine has to update the user buffer
820 * pointers and the remaining count).
822 ret
= actor(desc
, page
, offset
, nr
);
824 index
+= offset
>> PAGE_CACHE_SHIFT
;
825 offset
&= ~PAGE_CACHE_MASK
;
827 page_cache_release(page
);
828 if (ret
== nr
&& desc
->count
)
833 /* Get exclusive access to the page ... */
836 /* Did it get unhashed before we got the lock? */
837 if (!page
->mapping
) {
839 page_cache_release(page
);
843 /* Did somebody else fill it already? */
844 if (PageUptodate(page
)) {
850 /* Start the actual read. The read will unlock the page. */
851 error
= mapping
->a_ops
->readpage(filp
, page
);
853 if (unlikely(error
)) {
854 if (error
== AOP_TRUNCATED_PAGE
) {
855 page_cache_release(page
);
861 if (!PageUptodate(page
)) {
863 if (!PageUptodate(page
)) {
864 if (page
->mapping
== NULL
) {
866 * invalidate_inode_pages got it
869 page_cache_release(page
);
880 * i_size must be checked after we have done ->readpage.
882 * Checking i_size after the readpage allows us to calculate
883 * the correct value for "nr", which means the zero-filled
884 * part of the page is not copied back to userspace (unless
885 * another truncate extends the file - this is desired though).
887 isize
= i_size_read(inode
);
888 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
889 if (unlikely(!isize
|| index
> end_index
)) {
890 page_cache_release(page
);
894 /* nr is the maximum number of bytes to copy from this page */
895 nr
= PAGE_CACHE_SIZE
;
896 if (index
== end_index
) {
897 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
899 page_cache_release(page
);
907 /* UHHUH! A synchronous read error occurred. Report it */
909 page_cache_release(page
);
914 * Ok, it wasn't cached, so we need to create a new
918 cached_page
= page_cache_alloc_cold(mapping
);
920 desc
->error
= -ENOMEM
;
924 error
= add_to_page_cache_lru(cached_page
, mapping
,
927 if (error
== -EEXIST
)
940 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
942 page_cache_release(cached_page
);
947 EXPORT_SYMBOL(do_generic_mapping_read
);
949 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
950 unsigned long offset
, unsigned long size
)
953 unsigned long left
, count
= desc
->count
;
959 * Faults on the destination of a read are common, so do it before
962 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
963 kaddr
= kmap_atomic(page
, KM_USER0
);
964 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
965 kaddr
+ offset
, size
);
966 kunmap_atomic(kaddr
, KM_USER0
);
971 /* Do it the slow way */
973 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
978 desc
->error
= -EFAULT
;
981 desc
->count
= count
- size
;
982 desc
->written
+= size
;
983 desc
->arg
.buf
+= size
;
988 * This is the "read()" routine for all filesystems
989 * that can use the page cache directly.
992 __generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
993 unsigned long nr_segs
, loff_t
*ppos
)
995 struct file
*filp
= iocb
->ki_filp
;
1001 for (seg
= 0; seg
< nr_segs
; seg
++) {
1002 const struct iovec
*iv
= &iov
[seg
];
1005 * If any segment has a negative length, or the cumulative
1006 * length ever wraps negative then return -EINVAL.
1008 count
+= iv
->iov_len
;
1009 if (unlikely((ssize_t
)(count
|iv
->iov_len
) < 0))
1011 if (access_ok(VERIFY_WRITE
, iv
->iov_base
, iv
->iov_len
))
1016 count
-= iv
->iov_len
; /* This segment is no good */
1020 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1021 if (filp
->f_flags
& O_DIRECT
) {
1022 loff_t pos
= *ppos
, size
;
1023 struct address_space
*mapping
;
1024 struct inode
*inode
;
1026 mapping
= filp
->f_mapping
;
1027 inode
= mapping
->host
;
1030 goto out
; /* skip atime */
1031 size
= i_size_read(inode
);
1033 retval
= generic_file_direct_IO(READ
, iocb
,
1035 if (retval
> 0 && !is_sync_kiocb(iocb
))
1036 retval
= -EIOCBQUEUED
;
1038 *ppos
= pos
+ retval
;
1040 file_accessed(filp
);
1046 for (seg
= 0; seg
< nr_segs
; seg
++) {
1047 read_descriptor_t desc
;
1050 desc
.arg
.buf
= iov
[seg
].iov_base
;
1051 desc
.count
= iov
[seg
].iov_len
;
1052 if (desc
.count
== 0)
1055 do_generic_file_read(filp
,ppos
,&desc
,file_read_actor
);
1056 retval
+= desc
.written
;
1058 retval
= retval
?: desc
.error
;
1067 EXPORT_SYMBOL(__generic_file_aio_read
);
1070 generic_file_aio_read(struct kiocb
*iocb
, char __user
*buf
, size_t count
, loff_t pos
)
1072 struct iovec local_iov
= { .iov_base
= buf
, .iov_len
= count
};
1074 BUG_ON(iocb
->ki_pos
!= pos
);
1075 return __generic_file_aio_read(iocb
, &local_iov
, 1, &iocb
->ki_pos
);
1078 EXPORT_SYMBOL(generic_file_aio_read
);
1081 generic_file_read(struct file
*filp
, char __user
*buf
, size_t count
, loff_t
*ppos
)
1083 struct iovec local_iov
= { .iov_base
= buf
, .iov_len
= count
};
1087 init_sync_kiocb(&kiocb
, filp
);
1088 ret
= __generic_file_aio_read(&kiocb
, &local_iov
, 1, ppos
);
1089 if (-EIOCBQUEUED
== ret
)
1090 ret
= wait_on_sync_kiocb(&kiocb
);
1094 EXPORT_SYMBOL(generic_file_read
);
1096 int file_send_actor(read_descriptor_t
* desc
, struct page
*page
, unsigned long offset
, unsigned long size
)
1099 unsigned long count
= desc
->count
;
1100 struct file
*file
= desc
->arg
.data
;
1105 written
= file
->f_op
->sendpage(file
, page
, offset
,
1106 size
, &file
->f_pos
, size
<count
);
1108 desc
->error
= written
;
1111 desc
->count
= count
- written
;
1112 desc
->written
+= written
;
1116 ssize_t
generic_file_sendfile(struct file
*in_file
, loff_t
*ppos
,
1117 size_t count
, read_actor_t actor
, void *target
)
1119 read_descriptor_t desc
;
1126 desc
.arg
.data
= target
;
1129 do_generic_file_read(in_file
, ppos
, &desc
, actor
);
1131 return desc
.written
;
1135 EXPORT_SYMBOL(generic_file_sendfile
);
1138 do_readahead(struct address_space
*mapping
, struct file
*filp
,
1139 unsigned long index
, unsigned long nr
)
1141 if (!mapping
|| !mapping
->a_ops
|| !mapping
->a_ops
->readpage
)
1144 force_page_cache_readahead(mapping
, filp
, index
,
1145 max_sane_readahead(nr
));
1149 asmlinkage ssize_t
sys_readahead(int fd
, loff_t offset
, size_t count
)
1157 if (file
->f_mode
& FMODE_READ
) {
1158 struct address_space
*mapping
= file
->f_mapping
;
1159 unsigned long start
= offset
>> PAGE_CACHE_SHIFT
;
1160 unsigned long end
= (offset
+ count
- 1) >> PAGE_CACHE_SHIFT
;
1161 unsigned long len
= end
- start
+ 1;
1162 ret
= do_readahead(mapping
, file
, start
, len
);
1171 * This adds the requested page to the page cache if it isn't already there,
1172 * and schedules an I/O to read in its contents from disk.
1174 static int FASTCALL(page_cache_read(struct file
* file
, unsigned long offset
));
1175 static int fastcall
page_cache_read(struct file
* file
, unsigned long offset
)
1177 struct address_space
*mapping
= file
->f_mapping
;
1182 page
= page_cache_alloc_cold(mapping
);
1186 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1188 ret
= mapping
->a_ops
->readpage(file
, page
);
1189 else if (ret
== -EEXIST
)
1190 ret
= 0; /* losing race to add is OK */
1192 page_cache_release(page
);
1194 } while (ret
== AOP_TRUNCATED_PAGE
);
1199 #define MMAP_LOTSAMISS (100)
1202 * filemap_nopage() is invoked via the vma operations vector for a
1203 * mapped memory region to read in file data during a page fault.
1205 * The goto's are kind of ugly, but this streamlines the normal case of having
1206 * it in the page cache, and handles the special cases reasonably without
1207 * having a lot of duplicated code.
1209 struct page
*filemap_nopage(struct vm_area_struct
*area
,
1210 unsigned long address
, int *type
)
1213 struct file
*file
= area
->vm_file
;
1214 struct address_space
*mapping
= file
->f_mapping
;
1215 struct file_ra_state
*ra
= &file
->f_ra
;
1216 struct inode
*inode
= mapping
->host
;
1218 unsigned long size
, pgoff
;
1219 int did_readaround
= 0, majmin
= VM_FAULT_MINOR
;
1221 pgoff
= ((address
-area
->vm_start
) >> PAGE_CACHE_SHIFT
) + area
->vm_pgoff
;
1224 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1226 goto outside_data_content
;
1228 /* If we don't want any read-ahead, don't bother */
1229 if (VM_RandomReadHint(area
))
1230 goto no_cached_page
;
1233 * The readahead code wants to be told about each and every page
1234 * so it can build and shrink its windows appropriately
1236 * For sequential accesses, we use the generic readahead logic.
1238 if (VM_SequentialReadHint(area
))
1239 page_cache_readahead(mapping
, ra
, file
, pgoff
, 1);
1242 * Do we have something in the page cache already?
1245 page
= find_get_page(mapping
, pgoff
);
1247 unsigned long ra_pages
;
1249 if (VM_SequentialReadHint(area
)) {
1250 handle_ra_miss(mapping
, ra
, pgoff
);
1251 goto no_cached_page
;
1256 * Do we miss much more than hit in this file? If so,
1257 * stop bothering with read-ahead. It will only hurt.
1259 if (ra
->mmap_miss
> ra
->mmap_hit
+ MMAP_LOTSAMISS
)
1260 goto no_cached_page
;
1263 * To keep the pgmajfault counter straight, we need to
1264 * check did_readaround, as this is an inner loop.
1266 if (!did_readaround
) {
1267 majmin
= VM_FAULT_MAJOR
;
1268 inc_page_state(pgmajfault
);
1271 ra_pages
= max_sane_readahead(file
->f_ra
.ra_pages
);
1275 if (pgoff
> ra_pages
/ 2)
1276 start
= pgoff
- ra_pages
/ 2;
1277 do_page_cache_readahead(mapping
, file
, start
, ra_pages
);
1279 page
= find_get_page(mapping
, pgoff
);
1281 goto no_cached_page
;
1284 if (!did_readaround
)
1288 * Ok, found a page in the page cache, now we need to check
1289 * that it's up-to-date.
1291 if (!PageUptodate(page
))
1292 goto page_not_uptodate
;
1296 * Found the page and have a reference on it.
1298 mark_page_accessed(page
);
1303 outside_data_content
:
1305 * An external ptracer can access pages that normally aren't
1308 if (area
->vm_mm
== current
->mm
)
1310 /* Fall through to the non-read-ahead case */
1313 * We're only likely to ever get here if MADV_RANDOM is in
1316 error
= page_cache_read(file
, pgoff
);
1320 * The page we want has now been added to the page cache.
1321 * In the unlikely event that someone removed it in the
1322 * meantime, we'll just come back here and read it again.
1328 * An error return from page_cache_read can result if the
1329 * system is low on memory, or a problem occurs while trying
1332 if (error
== -ENOMEM
)
1337 if (!did_readaround
) {
1338 majmin
= VM_FAULT_MAJOR
;
1339 inc_page_state(pgmajfault
);
1343 /* Did it get unhashed while we waited for it? */
1344 if (!page
->mapping
) {
1346 page_cache_release(page
);
1350 /* Did somebody else get it up-to-date? */
1351 if (PageUptodate(page
)) {
1356 error
= mapping
->a_ops
->readpage(file
, page
);
1358 wait_on_page_locked(page
);
1359 if (PageUptodate(page
))
1361 } else if (error
== AOP_TRUNCATED_PAGE
) {
1362 page_cache_release(page
);
1367 * Umm, take care of errors if the page isn't up-to-date.
1368 * Try to re-read it _once_. We do this synchronously,
1369 * because there really aren't any performance issues here
1370 * and we need to check for errors.
1374 /* Somebody truncated the page on us? */
1375 if (!page
->mapping
) {
1377 page_cache_release(page
);
1381 /* Somebody else successfully read it in? */
1382 if (PageUptodate(page
)) {
1386 ClearPageError(page
);
1387 error
= mapping
->a_ops
->readpage(file
, page
);
1389 wait_on_page_locked(page
);
1390 if (PageUptodate(page
))
1392 } else if (error
== AOP_TRUNCATED_PAGE
) {
1393 page_cache_release(page
);
1398 * Things didn't work out. Return zero to tell the
1399 * mm layer so, possibly freeing the page cache page first.
1401 page_cache_release(page
);
1405 EXPORT_SYMBOL(filemap_nopage
);
1407 static struct page
* filemap_getpage(struct file
*file
, unsigned long pgoff
,
1410 struct address_space
*mapping
= file
->f_mapping
;
1415 * Do we have something in the page cache already?
1418 page
= find_get_page(mapping
, pgoff
);
1422 goto no_cached_page
;
1426 * Ok, found a page in the page cache, now we need to check
1427 * that it's up-to-date.
1429 if (!PageUptodate(page
)) {
1431 page_cache_release(page
);
1434 goto page_not_uptodate
;
1439 * Found the page and have a reference on it.
1441 mark_page_accessed(page
);
1445 error
= page_cache_read(file
, pgoff
);
1448 * The page we want has now been added to the page cache.
1449 * In the unlikely event that someone removed it in the
1450 * meantime, we'll just come back here and read it again.
1456 * An error return from page_cache_read can result if the
1457 * system is low on memory, or a problem occurs while trying
1465 /* Did it get unhashed while we waited for it? */
1466 if (!page
->mapping
) {
1471 /* Did somebody else get it up-to-date? */
1472 if (PageUptodate(page
)) {
1477 error
= mapping
->a_ops
->readpage(file
, page
);
1479 wait_on_page_locked(page
);
1480 if (PageUptodate(page
))
1482 } else if (error
== AOP_TRUNCATED_PAGE
) {
1483 page_cache_release(page
);
1488 * Umm, take care of errors if the page isn't up-to-date.
1489 * Try to re-read it _once_. We do this synchronously,
1490 * because there really aren't any performance issues here
1491 * and we need to check for errors.
1495 /* Somebody truncated the page on us? */
1496 if (!page
->mapping
) {
1500 /* Somebody else successfully read it in? */
1501 if (PageUptodate(page
)) {
1506 ClearPageError(page
);
1507 error
= mapping
->a_ops
->readpage(file
, page
);
1509 wait_on_page_locked(page
);
1510 if (PageUptodate(page
))
1512 } else if (error
== AOP_TRUNCATED_PAGE
) {
1513 page_cache_release(page
);
1518 * Things didn't work out. Return zero to tell the
1519 * mm layer so, possibly freeing the page cache page first.
1522 page_cache_release(page
);
1527 int filemap_populate(struct vm_area_struct
*vma
, unsigned long addr
,
1528 unsigned long len
, pgprot_t prot
, unsigned long pgoff
,
1531 struct file
*file
= vma
->vm_file
;
1532 struct address_space
*mapping
= file
->f_mapping
;
1533 struct inode
*inode
= mapping
->host
;
1535 struct mm_struct
*mm
= vma
->vm_mm
;
1540 force_page_cache_readahead(mapping
, vma
->vm_file
,
1541 pgoff
, len
>> PAGE_CACHE_SHIFT
);
1544 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1545 if (pgoff
+ (len
>> PAGE_CACHE_SHIFT
) > size
)
1548 page
= filemap_getpage(file
, pgoff
, nonblock
);
1550 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1551 * done in shmem_populate calling shmem_getpage */
1552 if (!page
&& !nonblock
)
1556 err
= install_page(mm
, vma
, addr
, page
, prot
);
1558 page_cache_release(page
);
1561 } else if (vma
->vm_flags
& VM_NONLINEAR
) {
1562 /* No page was found just because we can't read it in now (being
1563 * here implies nonblock != 0), but the page may exist, so set
1564 * the PTE to fault it in later. */
1565 err
= install_file_pte(mm
, vma
, addr
, pgoff
, prot
);
1578 EXPORT_SYMBOL(filemap_populate
);
1580 struct vm_operations_struct generic_file_vm_ops
= {
1581 .nopage
= filemap_nopage
,
1582 .populate
= filemap_populate
,
1585 /* This is used for a general mmap of a disk file */
1587 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1589 struct address_space
*mapping
= file
->f_mapping
;
1591 if (!mapping
->a_ops
->readpage
)
1593 file_accessed(file
);
1594 vma
->vm_ops
= &generic_file_vm_ops
;
1599 * This is for filesystems which do not implement ->writepage.
1601 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1603 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
1605 return generic_file_mmap(file
, vma
);
1608 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1612 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1616 #endif /* CONFIG_MMU */
1618 EXPORT_SYMBOL(generic_file_mmap
);
1619 EXPORT_SYMBOL(generic_file_readonly_mmap
);
1621 static inline struct page
*__read_cache_page(struct address_space
*mapping
,
1622 unsigned long index
,
1623 int (*filler
)(void *,struct page
*),
1626 struct page
*page
, *cached_page
= NULL
;
1629 page
= find_get_page(mapping
, index
);
1632 cached_page
= page_cache_alloc_cold(mapping
);
1634 return ERR_PTR(-ENOMEM
);
1636 err
= add_to_page_cache_lru(cached_page
, mapping
,
1641 /* Presumably ENOMEM for radix tree node */
1642 page_cache_release(cached_page
);
1643 return ERR_PTR(err
);
1647 err
= filler(data
, page
);
1649 page_cache_release(page
);
1650 page
= ERR_PTR(err
);
1654 page_cache_release(cached_page
);
1659 * Read into the page cache. If a page already exists,
1660 * and PageUptodate() is not set, try to fill the page.
1662 struct page
*read_cache_page(struct address_space
*mapping
,
1663 unsigned long index
,
1664 int (*filler
)(void *,struct page
*),
1671 page
= __read_cache_page(mapping
, index
, filler
, data
);
1674 mark_page_accessed(page
);
1675 if (PageUptodate(page
))
1679 if (!page
->mapping
) {
1681 page_cache_release(page
);
1684 if (PageUptodate(page
)) {
1688 err
= filler(data
, page
);
1690 page_cache_release(page
);
1691 page
= ERR_PTR(err
);
1697 EXPORT_SYMBOL(read_cache_page
);
1700 * If the page was newly created, increment its refcount and add it to the
1701 * caller's lru-buffering pagevec. This function is specifically for
1702 * generic_file_write().
1704 static inline struct page
*
1705 __grab_cache_page(struct address_space
*mapping
, unsigned long index
,
1706 struct page
**cached_page
, struct pagevec
*lru_pvec
)
1711 page
= find_lock_page(mapping
, index
);
1713 if (!*cached_page
) {
1714 *cached_page
= page_cache_alloc(mapping
);
1718 err
= add_to_page_cache(*cached_page
, mapping
,
1723 page
= *cached_page
;
1724 page_cache_get(page
);
1725 if (!pagevec_add(lru_pvec
, page
))
1726 __pagevec_lru_add(lru_pvec
);
1727 *cached_page
= NULL
;
1734 * The logic we want is
1736 * if suid or (sgid and xgrp)
1739 int remove_suid(struct dentry
*dentry
)
1741 mode_t mode
= dentry
->d_inode
->i_mode
;
1745 /* suid always must be killed */
1746 if (unlikely(mode
& S_ISUID
))
1747 kill
= ATTR_KILL_SUID
;
1750 * sgid without any exec bits is just a mandatory locking mark; leave
1751 * it alone. If some exec bits are set, it's a real sgid; kill it.
1753 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1754 kill
|= ATTR_KILL_SGID
;
1756 if (unlikely(kill
&& !capable(CAP_FSETID
))) {
1757 struct iattr newattrs
;
1759 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1760 result
= notify_change(dentry
, &newattrs
);
1764 EXPORT_SYMBOL(remove_suid
);
1767 __filemap_copy_from_user_iovec(char *vaddr
,
1768 const struct iovec
*iov
, size_t base
, size_t bytes
)
1770 size_t copied
= 0, left
= 0;
1773 char __user
*buf
= iov
->iov_base
+ base
;
1774 int copy
= min(bytes
, iov
->iov_len
- base
);
1777 left
= __copy_from_user_inatomic(vaddr
, buf
, copy
);
1783 if (unlikely(left
)) {
1784 /* zero the rest of the target like __copy_from_user */
1786 memset(vaddr
, 0, bytes
);
1790 return copied
- left
;
1794 * Performs necessary checks before doing a write
1796 * Can adjust writing position aor amount of bytes to write.
1797 * Returns appropriate error code that caller should return or
1798 * zero in case that write should be allowed.
1800 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
1802 struct inode
*inode
= file
->f_mapping
->host
;
1803 unsigned long limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1805 if (unlikely(*pos
< 0))
1809 /* FIXME: this is for backwards compatibility with 2.4 */
1810 if (file
->f_flags
& O_APPEND
)
1811 *pos
= i_size_read(inode
);
1813 if (limit
!= RLIM_INFINITY
) {
1814 if (*pos
>= limit
) {
1815 send_sig(SIGXFSZ
, current
, 0);
1818 if (*count
> limit
- (typeof(limit
))*pos
) {
1819 *count
= limit
- (typeof(limit
))*pos
;
1827 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
1828 !(file
->f_flags
& O_LARGEFILE
))) {
1829 if (*pos
>= MAX_NON_LFS
) {
1830 send_sig(SIGXFSZ
, current
, 0);
1833 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
1834 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
1839 * Are we about to exceed the fs block limit ?
1841 * If we have written data it becomes a short write. If we have
1842 * exceeded without writing data we send a signal and return EFBIG.
1843 * Linus frestrict idea will clean these up nicely..
1845 if (likely(!isblk
)) {
1846 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
1847 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
1848 send_sig(SIGXFSZ
, current
, 0);
1851 /* zero-length writes at ->s_maxbytes are OK */
1854 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
1855 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
1858 if (bdev_read_only(I_BDEV(inode
)))
1860 isize
= i_size_read(inode
);
1861 if (*pos
>= isize
) {
1862 if (*count
|| *pos
> isize
)
1866 if (*pos
+ *count
> isize
)
1867 *count
= isize
- *pos
;
1871 EXPORT_SYMBOL(generic_write_checks
);
1874 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1875 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
1876 size_t count
, size_t ocount
)
1878 struct file
*file
= iocb
->ki_filp
;
1879 struct address_space
*mapping
= file
->f_mapping
;
1880 struct inode
*inode
= mapping
->host
;
1883 if (count
!= ocount
)
1884 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
1886 written
= generic_file_direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
1888 loff_t end
= pos
+ written
;
1889 if (end
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
1890 i_size_write(inode
, end
);
1891 mark_inode_dirty(inode
);
1897 * Sync the fs metadata but not the minor inode changes and
1898 * of course not the data as we did direct DMA for the IO.
1899 * i_mutex is held, which protects generic_osync_inode() from
1902 if (written
>= 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
1903 int err
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
1907 if (written
== count
&& !is_sync_kiocb(iocb
))
1908 written
= -EIOCBQUEUED
;
1911 EXPORT_SYMBOL(generic_file_direct_write
);
1914 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1915 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
1916 size_t count
, ssize_t written
)
1918 struct file
*file
= iocb
->ki_filp
;
1919 struct address_space
* mapping
= file
->f_mapping
;
1920 struct address_space_operations
*a_ops
= mapping
->a_ops
;
1921 struct inode
*inode
= mapping
->host
;
1924 struct page
*cached_page
= NULL
;
1926 struct pagevec lru_pvec
;
1927 const struct iovec
*cur_iov
= iov
; /* current iovec */
1928 size_t iov_base
= 0; /* offset in the current iovec */
1931 pagevec_init(&lru_pvec
, 0);
1934 * handle partial DIO write. Adjust cur_iov if needed.
1936 if (likely(nr_segs
== 1))
1937 buf
= iov
->iov_base
+ written
;
1939 filemap_set_next_iovec(&cur_iov
, &iov_base
, written
);
1940 buf
= cur_iov
->iov_base
+ iov_base
;
1944 unsigned long index
;
1945 unsigned long offset
;
1946 unsigned long maxlen
;
1949 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
1950 index
= pos
>> PAGE_CACHE_SHIFT
;
1951 bytes
= PAGE_CACHE_SIZE
- offset
;
1956 * Bring in the user page that we will copy from _first_.
1957 * Otherwise there's a nasty deadlock on copying from the
1958 * same page as we're writing to, without it being marked
1961 maxlen
= cur_iov
->iov_len
- iov_base
;
1964 fault_in_pages_readable(buf
, maxlen
);
1966 page
= __grab_cache_page(mapping
,index
,&cached_page
,&lru_pvec
);
1972 status
= a_ops
->prepare_write(file
, page
, offset
, offset
+bytes
);
1973 if (unlikely(status
)) {
1974 loff_t isize
= i_size_read(inode
);
1976 if (status
!= AOP_TRUNCATED_PAGE
)
1978 page_cache_release(page
);
1979 if (status
== AOP_TRUNCATED_PAGE
)
1982 * prepare_write() may have instantiated a few blocks
1983 * outside i_size. Trim these off again.
1985 if (pos
+ bytes
> isize
)
1986 vmtruncate(inode
, isize
);
1989 if (likely(nr_segs
== 1))
1990 copied
= filemap_copy_from_user(page
, offset
,
1993 copied
= filemap_copy_from_user_iovec(page
, offset
,
1994 cur_iov
, iov_base
, bytes
);
1995 flush_dcache_page(page
);
1996 status
= a_ops
->commit_write(file
, page
, offset
, offset
+bytes
);
1997 if (status
== AOP_TRUNCATED_PAGE
) {
1998 page_cache_release(page
);
2001 if (likely(copied
> 0)) {
2010 if (unlikely(nr_segs
> 1)) {
2011 filemap_set_next_iovec(&cur_iov
,
2014 buf
= cur_iov
->iov_base
+
2021 if (unlikely(copied
!= bytes
))
2025 mark_page_accessed(page
);
2026 page_cache_release(page
);
2029 balance_dirty_pages_ratelimited(mapping
);
2035 page_cache_release(cached_page
);
2038 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2040 if (likely(status
>= 0)) {
2041 if (unlikely((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2042 if (!a_ops
->writepage
|| !is_sync_kiocb(iocb
))
2043 status
= generic_osync_inode(inode
, mapping
,
2044 OSYNC_METADATA
|OSYNC_DATA
);
2049 * If we get here for O_DIRECT writes then we must have fallen through
2050 * to buffered writes (block instantiation inside i_size). So we sync
2051 * the file data here, to try to honour O_DIRECT expectations.
2053 if (unlikely(file
->f_flags
& O_DIRECT
) && written
)
2054 status
= filemap_write_and_wait(mapping
);
2056 pagevec_lru_add(&lru_pvec
);
2057 return written
? written
: status
;
2059 EXPORT_SYMBOL(generic_file_buffered_write
);
2062 __generic_file_aio_write_nolock(struct kiocb
*iocb
, const struct iovec
*iov
,
2063 unsigned long nr_segs
, loff_t
*ppos
)
2065 struct file
*file
= iocb
->ki_filp
;
2066 struct address_space
* mapping
= file
->f_mapping
;
2067 size_t ocount
; /* original count */
2068 size_t count
; /* after file limit checks */
2069 struct inode
*inode
= mapping
->host
;
2076 for (seg
= 0; seg
< nr_segs
; seg
++) {
2077 const struct iovec
*iv
= &iov
[seg
];
2080 * If any segment has a negative length, or the cumulative
2081 * length ever wraps negative then return -EINVAL.
2083 ocount
+= iv
->iov_len
;
2084 if (unlikely((ssize_t
)(ocount
|iv
->iov_len
) < 0))
2086 if (access_ok(VERIFY_READ
, iv
->iov_base
, iv
->iov_len
))
2091 ocount
-= iv
->iov_len
; /* This segment is no good */
2098 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2100 /* We can write back this queue in page reclaim */
2101 current
->backing_dev_info
= mapping
->backing_dev_info
;
2104 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2111 err
= remove_suid(file
->f_dentry
);
2115 file_update_time(file
);
2117 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2118 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2119 written
= generic_file_direct_write(iocb
, iov
,
2120 &nr_segs
, pos
, ppos
, count
, ocount
);
2121 if (written
< 0 || written
== count
)
2124 * direct-io write to a hole: fall through to buffered I/O
2125 * for completing the rest of the request.
2131 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2132 pos
, ppos
, count
, written
);
2134 current
->backing_dev_info
= NULL
;
2135 return written
? written
: err
;
2137 EXPORT_SYMBOL(generic_file_aio_write_nolock
);
2140 generic_file_aio_write_nolock(struct kiocb
*iocb
, const struct iovec
*iov
,
2141 unsigned long nr_segs
, loff_t
*ppos
)
2143 struct file
*file
= iocb
->ki_filp
;
2144 struct address_space
*mapping
= file
->f_mapping
;
2145 struct inode
*inode
= mapping
->host
;
2149 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
, ppos
);
2151 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2154 err
= sync_page_range_nolock(inode
, mapping
, pos
, ret
);
2162 __generic_file_write_nolock(struct file
*file
, const struct iovec
*iov
,
2163 unsigned long nr_segs
, loff_t
*ppos
)
2168 init_sync_kiocb(&kiocb
, file
);
2169 ret
= __generic_file_aio_write_nolock(&kiocb
, iov
, nr_segs
, ppos
);
2170 if (ret
== -EIOCBQUEUED
)
2171 ret
= wait_on_sync_kiocb(&kiocb
);
2176 generic_file_write_nolock(struct file
*file
, const struct iovec
*iov
,
2177 unsigned long nr_segs
, loff_t
*ppos
)
2182 init_sync_kiocb(&kiocb
, file
);
2183 ret
= generic_file_aio_write_nolock(&kiocb
, iov
, nr_segs
, ppos
);
2184 if (-EIOCBQUEUED
== ret
)
2185 ret
= wait_on_sync_kiocb(&kiocb
);
2188 EXPORT_SYMBOL(generic_file_write_nolock
);
2190 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const char __user
*buf
,
2191 size_t count
, loff_t pos
)
2193 struct file
*file
= iocb
->ki_filp
;
2194 struct address_space
*mapping
= file
->f_mapping
;
2195 struct inode
*inode
= mapping
->host
;
2197 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
2200 BUG_ON(iocb
->ki_pos
!= pos
);
2202 mutex_lock(&inode
->i_mutex
);
2203 ret
= __generic_file_aio_write_nolock(iocb
, &local_iov
, 1,
2205 mutex_unlock(&inode
->i_mutex
);
2207 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2210 err
= sync_page_range(inode
, mapping
, pos
, ret
);
2216 EXPORT_SYMBOL(generic_file_aio_write
);
2218 ssize_t
generic_file_write(struct file
*file
, const char __user
*buf
,
2219 size_t count
, loff_t
*ppos
)
2221 struct address_space
*mapping
= file
->f_mapping
;
2222 struct inode
*inode
= mapping
->host
;
2224 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
2227 mutex_lock(&inode
->i_mutex
);
2228 ret
= __generic_file_write_nolock(file
, &local_iov
, 1, ppos
);
2229 mutex_unlock(&inode
->i_mutex
);
2231 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2234 err
= sync_page_range(inode
, mapping
, *ppos
- ret
, ret
);
2240 EXPORT_SYMBOL(generic_file_write
);
2242 ssize_t
generic_file_readv(struct file
*filp
, const struct iovec
*iov
,
2243 unsigned long nr_segs
, loff_t
*ppos
)
2248 init_sync_kiocb(&kiocb
, filp
);
2249 ret
= __generic_file_aio_read(&kiocb
, iov
, nr_segs
, ppos
);
2250 if (-EIOCBQUEUED
== ret
)
2251 ret
= wait_on_sync_kiocb(&kiocb
);
2254 EXPORT_SYMBOL(generic_file_readv
);
2256 ssize_t
generic_file_writev(struct file
*file
, const struct iovec
*iov
,
2257 unsigned long nr_segs
, loff_t
*ppos
)
2259 struct address_space
*mapping
= file
->f_mapping
;
2260 struct inode
*inode
= mapping
->host
;
2263 mutex_lock(&inode
->i_mutex
);
2264 ret
= __generic_file_write_nolock(file
, iov
, nr_segs
, ppos
);
2265 mutex_unlock(&inode
->i_mutex
);
2267 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2270 err
= sync_page_range(inode
, mapping
, *ppos
- ret
, ret
);
2276 EXPORT_SYMBOL(generic_file_writev
);
2279 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2280 * went wrong during pagecache shootdown.
2283 generic_file_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
2284 loff_t offset
, unsigned long nr_segs
)
2286 struct file
*file
= iocb
->ki_filp
;
2287 struct address_space
*mapping
= file
->f_mapping
;
2289 size_t write_len
= 0;
2292 * If it's a write, unmap all mmappings of the file up-front. This
2293 * will cause any pte dirty bits to be propagated into the pageframes
2294 * for the subsequent filemap_write_and_wait().
2297 write_len
= iov_length(iov
, nr_segs
);
2298 if (mapping_mapped(mapping
))
2299 unmap_mapping_range(mapping
, offset
, write_len
, 0);
2302 retval
= filemap_write_and_wait(mapping
);
2304 retval
= mapping
->a_ops
->direct_IO(rw
, iocb
, iov
,
2306 if (rw
== WRITE
&& mapping
->nrpages
) {
2307 pgoff_t end
= (offset
+ write_len
- 1)
2308 >> PAGE_CACHE_SHIFT
;
2309 int err
= invalidate_inode_pages2_range(mapping
,
2310 offset
>> PAGE_CACHE_SHIFT
, end
);