4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/compiler.h>
16 #include <linux/uaccess.h>
17 #include <linux/aio.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/syscalls.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/memcontrol.h>
36 #include <linux/mm_inline.h> /* for page_is_file_cache() */
40 * FIXME: remove all knowledge of the buffer layer from the core VM
42 #include <linux/buffer_head.h> /* for generic_osync_inode */
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * Shared mappings now work. 15.8.1995 Bruno.
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
64 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
68 * ->i_mmap_lock (truncate->unmap_mapping_range)
72 * ->page_table_lock or pte_lock (various, mainly in memory.c)
73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 * ->lock_page (access_process_vm)
78 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
82 * ->i_alloc_sem (various)
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->dcache_lock (proc_pid_lookup)
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold the mapping's tree_lock.
115 void __remove_from_page_cache(struct page
*page
)
117 struct address_space
*mapping
= page
->mapping
;
119 mem_cgroup_uncharge_cache_page(page
);
120 radix_tree_delete(&mapping
->page_tree
, page
->index
);
121 page
->mapping
= NULL
;
123 __dec_zone_page_state(page
, NR_FILE_PAGES
);
124 BUG_ON(page_mapped(page
));
127 * Some filesystems seem to re-dirty the page even after
128 * the VM has canceled the dirty bit (eg ext3 journaling).
130 * Fix it up by doing a final dirty accounting check after
131 * having removed the page entirely.
133 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
134 dec_zone_page_state(page
, NR_FILE_DIRTY
);
135 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
139 void remove_from_page_cache(struct page
*page
)
141 struct address_space
*mapping
= page
->mapping
;
143 BUG_ON(!PageLocked(page
));
145 spin_lock_irq(&mapping
->tree_lock
);
146 __remove_from_page_cache(page
);
147 spin_unlock_irq(&mapping
->tree_lock
);
150 static int sync_page(void *word
)
152 struct address_space
*mapping
;
155 page
= container_of((unsigned long *)word
, struct page
, flags
);
158 * page_mapping() is being called without PG_locked held.
159 * Some knowledge of the state and use of the page is used to
160 * reduce the requirements down to a memory barrier.
161 * The danger here is of a stale page_mapping() return value
162 * indicating a struct address_space different from the one it's
163 * associated with when it is associated with one.
164 * After smp_mb(), it's either the correct page_mapping() for
165 * the page, or an old page_mapping() and the page's own
166 * page_mapping() has gone NULL.
167 * The ->sync_page() address_space operation must tolerate
168 * page_mapping() going NULL. By an amazing coincidence,
169 * this comes about because none of the users of the page
170 * in the ->sync_page() methods make essential use of the
171 * page_mapping(), merely passing the page down to the backing
172 * device's unplug functions when it's non-NULL, which in turn
173 * ignore it for all cases but swap, where only page_private(page) is
174 * of interest. When page_mapping() does go NULL, the entire
175 * call stack gracefully ignores the page and returns.
179 mapping
= page_mapping(page
);
180 if (mapping
&& mapping
->a_ops
&& mapping
->a_ops
->sync_page
)
181 mapping
->a_ops
->sync_page(page
);
186 static int sync_page_killable(void *word
)
189 return fatal_signal_pending(current
) ? -EINTR
: 0;
193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
194 * @mapping: address space structure to write
195 * @start: offset in bytes where the range starts
196 * @end: offset in bytes where the range ends (inclusive)
197 * @sync_mode: enable synchronous operation
199 * Start writeback against all of a mapping's dirty pages that lie
200 * within the byte offsets <start, end> inclusive.
202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
203 * opposed to a regular memory cleansing writeback. The difference between
204 * these two operations is that if a dirty page/buffer is encountered, it must
205 * be waited upon, and not just skipped over.
207 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
208 loff_t end
, int sync_mode
)
211 struct writeback_control wbc
= {
212 .sync_mode
= sync_mode
,
213 .nr_to_write
= mapping
->nrpages
* 2,
214 .range_start
= start
,
218 if (!mapping_cap_writeback_dirty(mapping
))
221 ret
= do_writepages(mapping
, &wbc
);
225 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
228 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
231 int filemap_fdatawrite(struct address_space
*mapping
)
233 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
235 EXPORT_SYMBOL(filemap_fdatawrite
);
237 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
240 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
242 EXPORT_SYMBOL(filemap_fdatawrite_range
);
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
251 int filemap_flush(struct address_space
*mapping
)
253 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
255 EXPORT_SYMBOL(filemap_flush
);
258 * wait_on_page_writeback_range - wait for writeback to complete
259 * @mapping: target address_space
260 * @start: beginning page index
261 * @end: ending page index
263 * Wait for writeback to complete against pages indexed by start->end
266 int wait_on_page_writeback_range(struct address_space
*mapping
,
267 pgoff_t start
, pgoff_t end
)
277 pagevec_init(&pvec
, 0);
279 while ((index
<= end
) &&
280 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
281 PAGECACHE_TAG_WRITEBACK
,
282 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
285 for (i
= 0; i
< nr_pages
; i
++) {
286 struct page
*page
= pvec
.pages
[i
];
288 /* until radix tree lookup accepts end_index */
289 if (page
->index
> end
)
292 wait_on_page_writeback(page
);
296 pagevec_release(&pvec
);
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
303 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
310 * sync_page_range - write and wait on all pages in the passed range
311 * @inode: target inode
312 * @mapping: target address_space
313 * @pos: beginning offset in pages to write
314 * @count: number of bytes to write
316 * Write and wait upon all the pages in the passed range. This is a "data
317 * integrity" operation. It waits upon in-flight writeout before starting and
318 * waiting upon new writeout. If there was an IO error, return it.
320 * We need to re-take i_mutex during the generic_osync_inode list walk because
321 * it is otherwise livelockable.
323 int sync_page_range(struct inode
*inode
, struct address_space
*mapping
,
324 loff_t pos
, loff_t count
)
326 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
327 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
330 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
332 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
334 mutex_lock(&inode
->i_mutex
);
335 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
336 mutex_unlock(&inode
->i_mutex
);
339 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
342 EXPORT_SYMBOL(sync_page_range
);
345 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
346 * @inode: target inode
347 * @mapping: target address_space
348 * @pos: beginning offset in pages to write
349 * @count: number of bytes to write
351 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
352 * as it forces O_SYNC writers to different parts of the same file
353 * to be serialised right until io completion.
355 int sync_page_range_nolock(struct inode
*inode
, struct address_space
*mapping
,
356 loff_t pos
, loff_t count
)
358 pgoff_t start
= pos
>> PAGE_CACHE_SHIFT
;
359 pgoff_t end
= (pos
+ count
- 1) >> PAGE_CACHE_SHIFT
;
362 if (!mapping_cap_writeback_dirty(mapping
) || !count
)
364 ret
= filemap_fdatawrite_range(mapping
, pos
, pos
+ count
- 1);
366 ret
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
368 ret
= wait_on_page_writeback_range(mapping
, start
, end
);
371 EXPORT_SYMBOL(sync_page_range_nolock
);
374 * filemap_fdatawait - wait for all under-writeback pages to complete
375 * @mapping: address space structure to wait for
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
380 int filemap_fdatawait(struct address_space
*mapping
)
382 loff_t i_size
= i_size_read(mapping
->host
);
387 return wait_on_page_writeback_range(mapping
, 0,
388 (i_size
- 1) >> PAGE_CACHE_SHIFT
);
390 EXPORT_SYMBOL(filemap_fdatawait
);
392 int filemap_write_and_wait(struct address_space
*mapping
)
396 if (mapping
->nrpages
) {
397 err
= filemap_fdatawrite(mapping
);
399 * Even if the above returned error, the pages may be
400 * written partially (e.g. -ENOSPC), so we wait for it.
401 * But the -EIO is special case, it may indicate the worst
402 * thing (e.g. bug) happened, so we avoid waiting for it.
405 int err2
= filemap_fdatawait(mapping
);
412 EXPORT_SYMBOL(filemap_write_and_wait
);
415 * filemap_write_and_wait_range - write out & wait on a file range
416 * @mapping: the address_space for the pages
417 * @lstart: offset in bytes where the range starts
418 * @lend: offset in bytes where the range ends (inclusive)
420 * Write out and wait upon file offsets lstart->lend, inclusive.
422 * Note that `lend' is inclusive (describes the last byte to be written) so
423 * that this function can be used to write to the very end-of-file (end = -1).
425 int filemap_write_and_wait_range(struct address_space
*mapping
,
426 loff_t lstart
, loff_t lend
)
430 if (mapping
->nrpages
) {
431 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
433 /* See comment of filemap_write_and_wait() */
435 int err2
= wait_on_page_writeback_range(mapping
,
436 lstart
>> PAGE_CACHE_SHIFT
,
437 lend
>> PAGE_CACHE_SHIFT
);
446 * add_to_page_cache_locked - add a locked page to the pagecache
448 * @mapping: the page's address_space
449 * @offset: page index
450 * @gfp_mask: page allocation mode
452 * This function is used to add a page to the pagecache. It must be locked.
453 * This function does not add the page to the LRU. The caller must do that.
455 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
456 pgoff_t offset
, gfp_t gfp_mask
)
460 VM_BUG_ON(!PageLocked(page
));
462 error
= mem_cgroup_cache_charge(page
, current
->mm
,
463 gfp_mask
& ~__GFP_HIGHMEM
);
467 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
469 page_cache_get(page
);
470 page
->mapping
= mapping
;
471 page
->index
= offset
;
473 spin_lock_irq(&mapping
->tree_lock
);
474 error
= radix_tree_insert(&mapping
->page_tree
, offset
, page
);
475 if (likely(!error
)) {
477 __inc_zone_page_state(page
, NR_FILE_PAGES
);
479 page
->mapping
= NULL
;
480 mem_cgroup_uncharge_cache_page(page
);
481 page_cache_release(page
);
484 spin_unlock_irq(&mapping
->tree_lock
);
485 radix_tree_preload_end();
487 mem_cgroup_uncharge_cache_page(page
);
491 EXPORT_SYMBOL(add_to_page_cache_locked
);
493 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
494 pgoff_t offset
, gfp_t gfp_mask
)
499 * Splice_read and readahead add shmem/tmpfs pages into the page cache
500 * before shmem_readpage has a chance to mark them as SwapBacked: they
501 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
502 * (called in add_to_page_cache) needs to know where they're going too.
504 if (mapping_cap_swap_backed(mapping
))
505 SetPageSwapBacked(page
);
507 ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
509 if (page_is_file_cache(page
))
510 lru_cache_add_file(page
);
512 lru_cache_add_active_anon(page
);
518 struct page
*__page_cache_alloc(gfp_t gfp
)
520 if (cpuset_do_page_mem_spread()) {
521 int n
= cpuset_mem_spread_node();
522 return alloc_pages_node(n
, gfp
, 0);
524 return alloc_pages(gfp
, 0);
526 EXPORT_SYMBOL(__page_cache_alloc
);
529 static int __sleep_on_page_lock(void *word
)
536 * In order to wait for pages to become available there must be
537 * waitqueues associated with pages. By using a hash table of
538 * waitqueues where the bucket discipline is to maintain all
539 * waiters on the same queue and wake all when any of the pages
540 * become available, and for the woken contexts to check to be
541 * sure the appropriate page became available, this saves space
542 * at a cost of "thundering herd" phenomena during rare hash
545 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
547 const struct zone
*zone
= page_zone(page
);
549 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
552 static inline void wake_up_page(struct page
*page
, int bit
)
554 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
557 void wait_on_page_bit(struct page
*page
, int bit_nr
)
559 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
561 if (test_bit(bit_nr
, &page
->flags
))
562 __wait_on_bit(page_waitqueue(page
), &wait
, sync_page
,
563 TASK_UNINTERRUPTIBLE
);
565 EXPORT_SYMBOL(wait_on_page_bit
);
568 * unlock_page - unlock a locked page
571 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
572 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
573 * mechananism between PageLocked pages and PageWriteback pages is shared.
574 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
576 * The first mb is necessary to safely close the critical section opened by the
577 * test_and_set_bit() to lock the page; the second mb is necessary to enforce
578 * ordering between the clear_bit and the read of the waitqueue (to avoid SMP
579 * races with a parallel wait_on_page_locked()).
581 void unlock_page(struct page
*page
)
583 smp_mb__before_clear_bit();
584 if (!test_and_clear_bit(PG_locked
, &page
->flags
))
586 smp_mb__after_clear_bit();
587 wake_up_page(page
, PG_locked
);
589 EXPORT_SYMBOL(unlock_page
);
592 * end_page_writeback - end writeback against a page
595 void end_page_writeback(struct page
*page
)
597 if (TestClearPageReclaim(page
))
598 rotate_reclaimable_page(page
);
600 if (!test_clear_page_writeback(page
))
603 smp_mb__after_clear_bit();
604 wake_up_page(page
, PG_writeback
);
606 EXPORT_SYMBOL(end_page_writeback
);
609 * __lock_page - get a lock on the page, assuming we need to sleep to get it
610 * @page: the page to lock
612 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
613 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
614 * chances are that on the second loop, the block layer's plug list is empty,
615 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
617 void __lock_page(struct page
*page
)
619 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
621 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sync_page
,
622 TASK_UNINTERRUPTIBLE
);
624 EXPORT_SYMBOL(__lock_page
);
626 int __lock_page_killable(struct page
*page
)
628 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
630 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
631 sync_page_killable
, TASK_KILLABLE
);
635 * __lock_page_nosync - get a lock on the page, without calling sync_page()
636 * @page: the page to lock
638 * Variant of lock_page that does not require the caller to hold a reference
639 * on the page's mapping.
641 void __lock_page_nosync(struct page
*page
)
643 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
644 __wait_on_bit_lock(page_waitqueue(page
), &wait
, __sleep_on_page_lock
,
645 TASK_UNINTERRUPTIBLE
);
649 * find_get_page - find and get a page reference
650 * @mapping: the address_space to search
651 * @offset: the page index
653 * Is there a pagecache struct page at the given (mapping, offset) tuple?
654 * If yes, increment its refcount and return it; if no, return NULL.
656 struct page
*find_get_page(struct address_space
*mapping
, pgoff_t offset
)
664 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
666 page
= radix_tree_deref_slot(pagep
);
667 if (unlikely(!page
|| page
== RADIX_TREE_RETRY
))
670 if (!page_cache_get_speculative(page
))
674 * Has the page moved?
675 * This is part of the lockless pagecache protocol. See
676 * include/linux/pagemap.h for details.
678 if (unlikely(page
!= *pagep
)) {
679 page_cache_release(page
);
687 EXPORT_SYMBOL(find_get_page
);
690 * find_lock_page - locate, pin and lock a pagecache page
691 * @mapping: the address_space to search
692 * @offset: the page index
694 * Locates the desired pagecache page, locks it, increments its reference
695 * count and returns its address.
697 * Returns zero if the page was not present. find_lock_page() may sleep.
699 struct page
*find_lock_page(struct address_space
*mapping
, pgoff_t offset
)
704 page
= find_get_page(mapping
, offset
);
707 /* Has the page been truncated? */
708 if (unlikely(page
->mapping
!= mapping
)) {
710 page_cache_release(page
);
713 VM_BUG_ON(page
->index
!= offset
);
717 EXPORT_SYMBOL(find_lock_page
);
720 * find_or_create_page - locate or add a pagecache page
721 * @mapping: the page's address_space
722 * @index: the page's index into the mapping
723 * @gfp_mask: page allocation mode
725 * Locates a page in the pagecache. If the page is not present, a new page
726 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
727 * LRU list. The returned page is locked and has its reference count
730 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
733 * find_or_create_page() returns the desired page's address, or zero on
736 struct page
*find_or_create_page(struct address_space
*mapping
,
737 pgoff_t index
, gfp_t gfp_mask
)
742 page
= find_lock_page(mapping
, index
);
744 page
= __page_cache_alloc(gfp_mask
);
747 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp_mask
);
749 page_cache_release(page
);
757 EXPORT_SYMBOL(find_or_create_page
);
760 * find_get_pages - gang pagecache lookup
761 * @mapping: The address_space to search
762 * @start: The starting page index
763 * @nr_pages: The maximum number of pages
764 * @pages: Where the resulting pages are placed
766 * find_get_pages() will search for and return a group of up to
767 * @nr_pages pages in the mapping. The pages are placed at @pages.
768 * find_get_pages() takes a reference against the returned pages.
770 * The search returns a group of mapping-contiguous pages with ascending
771 * indexes. There may be holes in the indices due to not-present pages.
773 * find_get_pages() returns the number of pages which were found.
775 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
776 unsigned int nr_pages
, struct page
**pages
)
780 unsigned int nr_found
;
784 nr_found
= radix_tree_gang_lookup_slot(&mapping
->page_tree
,
785 (void ***)pages
, start
, nr_pages
);
787 for (i
= 0; i
< nr_found
; i
++) {
790 page
= radix_tree_deref_slot((void **)pages
[i
]);
794 * this can only trigger if nr_found == 1, making livelock
797 if (unlikely(page
== RADIX_TREE_RETRY
))
800 if (!page_cache_get_speculative(page
))
803 /* Has the page moved? */
804 if (unlikely(page
!= *((void **)pages
[i
]))) {
805 page_cache_release(page
);
817 * find_get_pages_contig - gang contiguous pagecache lookup
818 * @mapping: The address_space to search
819 * @index: The starting page index
820 * @nr_pages: The maximum number of pages
821 * @pages: Where the resulting pages are placed
823 * find_get_pages_contig() works exactly like find_get_pages(), except
824 * that the returned number of pages are guaranteed to be contiguous.
826 * find_get_pages_contig() returns the number of pages which were found.
828 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
829 unsigned int nr_pages
, struct page
**pages
)
833 unsigned int nr_found
;
837 nr_found
= radix_tree_gang_lookup_slot(&mapping
->page_tree
,
838 (void ***)pages
, index
, nr_pages
);
840 for (i
= 0; i
< nr_found
; i
++) {
843 page
= radix_tree_deref_slot((void **)pages
[i
]);
847 * this can only trigger if nr_found == 1, making livelock
850 if (unlikely(page
== RADIX_TREE_RETRY
))
853 if (page
->mapping
== NULL
|| page
->index
!= index
)
856 if (!page_cache_get_speculative(page
))
859 /* Has the page moved? */
860 if (unlikely(page
!= *((void **)pages
[i
]))) {
861 page_cache_release(page
);
872 EXPORT_SYMBOL(find_get_pages_contig
);
875 * find_get_pages_tag - find and return pages that match @tag
876 * @mapping: the address_space to search
877 * @index: the starting page index
878 * @tag: the tag index
879 * @nr_pages: the maximum number of pages
880 * @pages: where the resulting pages are placed
882 * Like find_get_pages, except we only return pages which are tagged with
883 * @tag. We update @index to index the next page for the traversal.
885 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
886 int tag
, unsigned int nr_pages
, struct page
**pages
)
890 unsigned int nr_found
;
894 nr_found
= radix_tree_gang_lookup_tag_slot(&mapping
->page_tree
,
895 (void ***)pages
, *index
, nr_pages
, tag
);
897 for (i
= 0; i
< nr_found
; i
++) {
900 page
= radix_tree_deref_slot((void **)pages
[i
]);
904 * this can only trigger if nr_found == 1, making livelock
907 if (unlikely(page
== RADIX_TREE_RETRY
))
910 if (!page_cache_get_speculative(page
))
913 /* Has the page moved? */
914 if (unlikely(page
!= *((void **)pages
[i
]))) {
915 page_cache_release(page
);
925 *index
= pages
[ret
- 1]->index
+ 1;
929 EXPORT_SYMBOL(find_get_pages_tag
);
932 * grab_cache_page_nowait - returns locked page at given index in given cache
933 * @mapping: target address_space
934 * @index: the page index
936 * Same as grab_cache_page(), but do not wait if the page is unavailable.
937 * This is intended for speculative data generators, where the data can
938 * be regenerated if the page couldn't be grabbed. This routine should
939 * be safe to call while holding the lock for another page.
941 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
942 * and deadlock against the caller's locked page.
945 grab_cache_page_nowait(struct address_space
*mapping
, pgoff_t index
)
947 struct page
*page
= find_get_page(mapping
, index
);
950 if (trylock_page(page
))
952 page_cache_release(page
);
955 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
956 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_KERNEL
)) {
957 page_cache_release(page
);
962 EXPORT_SYMBOL(grab_cache_page_nowait
);
965 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
966 * a _large_ part of the i/o request. Imagine the worst scenario:
968 * ---R__________________________________________B__________
969 * ^ reading here ^ bad block(assume 4k)
971 * read(R) => miss => readahead(R...B) => media error => frustrating retries
972 * => failing the whole request => read(R) => read(R+1) =>
973 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
974 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
975 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
977 * It is going insane. Fix it by quickly scaling down the readahead size.
979 static void shrink_readahead_size_eio(struct file
*filp
,
980 struct file_ra_state
*ra
)
989 * do_generic_file_read - generic file read routine
990 * @filp: the file to read
991 * @ppos: current file position
992 * @desc: read_descriptor
993 * @actor: read method
995 * This is a generic file read routine, and uses the
996 * mapping->a_ops->readpage() function for the actual low-level stuff.
998 * This is really ugly. But the goto's actually try to clarify some
999 * of the logic when it comes to error handling etc.
1001 static void do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1002 read_descriptor_t
*desc
, read_actor_t actor
)
1004 struct address_space
*mapping
= filp
->f_mapping
;
1005 struct inode
*inode
= mapping
->host
;
1006 struct file_ra_state
*ra
= &filp
->f_ra
;
1010 unsigned long offset
; /* offset into pagecache page */
1011 unsigned int prev_offset
;
1014 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1015 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1016 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1017 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1018 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1024 unsigned long nr
, ret
;
1028 page
= find_get_page(mapping
, index
);
1030 page_cache_sync_readahead(mapping
,
1032 index
, last_index
- index
);
1033 page
= find_get_page(mapping
, index
);
1034 if (unlikely(page
== NULL
))
1035 goto no_cached_page
;
1037 if (PageReadahead(page
)) {
1038 page_cache_async_readahead(mapping
,
1040 index
, last_index
- index
);
1042 if (!PageUptodate(page
)) {
1043 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1044 !mapping
->a_ops
->is_partially_uptodate
)
1045 goto page_not_up_to_date
;
1046 if (!trylock_page(page
))
1047 goto page_not_up_to_date
;
1048 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1050 goto page_not_up_to_date_locked
;
1055 * i_size must be checked after we know the page is Uptodate.
1057 * Checking i_size after the check allows us to calculate
1058 * the correct value for "nr", which means the zero-filled
1059 * part of the page is not copied back to userspace (unless
1060 * another truncate extends the file - this is desired though).
1063 isize
= i_size_read(inode
);
1064 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1065 if (unlikely(!isize
|| index
> end_index
)) {
1066 page_cache_release(page
);
1070 /* nr is the maximum number of bytes to copy from this page */
1071 nr
= PAGE_CACHE_SIZE
;
1072 if (index
== end_index
) {
1073 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1075 page_cache_release(page
);
1081 /* If users can be writing to this page using arbitrary
1082 * virtual addresses, take care about potential aliasing
1083 * before reading the page on the kernel side.
1085 if (mapping_writably_mapped(mapping
))
1086 flush_dcache_page(page
);
1089 * When a sequential read accesses a page several times,
1090 * only mark it as accessed the first time.
1092 if (prev_index
!= index
|| offset
!= prev_offset
)
1093 mark_page_accessed(page
);
1097 * Ok, we have the page, and it's up-to-date, so
1098 * now we can copy it to user space...
1100 * The actor routine returns how many bytes were actually used..
1101 * NOTE! This may not be the same as how much of a user buffer
1102 * we filled up (we may be padding etc), so we can only update
1103 * "pos" here (the actor routine has to update the user buffer
1104 * pointers and the remaining count).
1106 ret
= actor(desc
, page
, offset
, nr
);
1108 index
+= offset
>> PAGE_CACHE_SHIFT
;
1109 offset
&= ~PAGE_CACHE_MASK
;
1110 prev_offset
= offset
;
1112 page_cache_release(page
);
1113 if (ret
== nr
&& desc
->count
)
1117 page_not_up_to_date
:
1118 /* Get exclusive access to the page ... */
1119 error
= lock_page_killable(page
);
1120 if (unlikely(error
))
1121 goto readpage_error
;
1123 page_not_up_to_date_locked
:
1124 /* Did it get truncated before we got the lock? */
1125 if (!page
->mapping
) {
1127 page_cache_release(page
);
1131 /* Did somebody else fill it already? */
1132 if (PageUptodate(page
)) {
1138 /* Start the actual read. The read will unlock the page. */
1139 error
= mapping
->a_ops
->readpage(filp
, page
);
1141 if (unlikely(error
)) {
1142 if (error
== AOP_TRUNCATED_PAGE
) {
1143 page_cache_release(page
);
1146 goto readpage_error
;
1149 if (!PageUptodate(page
)) {
1150 error
= lock_page_killable(page
);
1151 if (unlikely(error
))
1152 goto readpage_error
;
1153 if (!PageUptodate(page
)) {
1154 if (page
->mapping
== NULL
) {
1156 * invalidate_inode_pages got it
1159 page_cache_release(page
);
1163 shrink_readahead_size_eio(filp
, ra
);
1165 goto readpage_error
;
1173 /* UHHUH! A synchronous read error occurred. Report it */
1174 desc
->error
= error
;
1175 page_cache_release(page
);
1180 * Ok, it wasn't cached, so we need to create a new
1183 page
= page_cache_alloc_cold(mapping
);
1185 desc
->error
= -ENOMEM
;
1188 error
= add_to_page_cache_lru(page
, mapping
,
1191 page_cache_release(page
);
1192 if (error
== -EEXIST
)
1194 desc
->error
= error
;
1201 ra
->prev_pos
= prev_index
;
1202 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1203 ra
->prev_pos
|= prev_offset
;
1205 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1206 file_accessed(filp
);
1209 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
1210 unsigned long offset
, unsigned long size
)
1213 unsigned long left
, count
= desc
->count
;
1219 * Faults on the destination of a read are common, so do it before
1222 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
1223 kaddr
= kmap_atomic(page
, KM_USER0
);
1224 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
1225 kaddr
+ offset
, size
);
1226 kunmap_atomic(kaddr
, KM_USER0
);
1231 /* Do it the slow way */
1233 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
1238 desc
->error
= -EFAULT
;
1241 desc
->count
= count
- size
;
1242 desc
->written
+= size
;
1243 desc
->arg
.buf
+= size
;
1248 * Performs necessary checks before doing a write
1249 * @iov: io vector request
1250 * @nr_segs: number of segments in the iovec
1251 * @count: number of bytes to write
1252 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1254 * Adjust number of segments and amount of bytes to write (nr_segs should be
1255 * properly initialized first). Returns appropriate error code that caller
1256 * should return or zero in case that write should be allowed.
1258 int generic_segment_checks(const struct iovec
*iov
,
1259 unsigned long *nr_segs
, size_t *count
, int access_flags
)
1263 for (seg
= 0; seg
< *nr_segs
; seg
++) {
1264 const struct iovec
*iv
= &iov
[seg
];
1267 * If any segment has a negative length, or the cumulative
1268 * length ever wraps negative then return -EINVAL.
1271 if (unlikely((ssize_t
)(cnt
|iv
->iov_len
) < 0))
1273 if (access_ok(access_flags
, iv
->iov_base
, iv
->iov_len
))
1278 cnt
-= iv
->iov_len
; /* This segment is no good */
1284 EXPORT_SYMBOL(generic_segment_checks
);
1287 * generic_file_aio_read - generic filesystem read routine
1288 * @iocb: kernel I/O control block
1289 * @iov: io vector request
1290 * @nr_segs: number of segments in the iovec
1291 * @pos: current file position
1293 * This is the "read()" routine for all filesystems
1294 * that can use the page cache directly.
1297 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1298 unsigned long nr_segs
, loff_t pos
)
1300 struct file
*filp
= iocb
->ki_filp
;
1304 loff_t
*ppos
= &iocb
->ki_pos
;
1307 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1311 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1312 if (filp
->f_flags
& O_DIRECT
) {
1314 struct address_space
*mapping
;
1315 struct inode
*inode
;
1317 mapping
= filp
->f_mapping
;
1318 inode
= mapping
->host
;
1320 goto out
; /* skip atime */
1321 size
= i_size_read(inode
);
1323 retval
= filemap_write_and_wait(mapping
);
1325 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
,
1329 *ppos
= pos
+ retval
;
1331 file_accessed(filp
);
1337 for (seg
= 0; seg
< nr_segs
; seg
++) {
1338 read_descriptor_t desc
;
1341 desc
.arg
.buf
= iov
[seg
].iov_base
;
1342 desc
.count
= iov
[seg
].iov_len
;
1343 if (desc
.count
== 0)
1346 do_generic_file_read(filp
, ppos
, &desc
, file_read_actor
);
1347 retval
+= desc
.written
;
1349 retval
= retval
?: desc
.error
;
1358 EXPORT_SYMBOL(generic_file_aio_read
);
1361 do_readahead(struct address_space
*mapping
, struct file
*filp
,
1362 pgoff_t index
, unsigned long nr
)
1364 if (!mapping
|| !mapping
->a_ops
|| !mapping
->a_ops
->readpage
)
1367 force_page_cache_readahead(mapping
, filp
, index
,
1368 max_sane_readahead(nr
));
1372 asmlinkage ssize_t
sys_readahead(int fd
, loff_t offset
, size_t count
)
1380 if (file
->f_mode
& FMODE_READ
) {
1381 struct address_space
*mapping
= file
->f_mapping
;
1382 pgoff_t start
= offset
>> PAGE_CACHE_SHIFT
;
1383 pgoff_t end
= (offset
+ count
- 1) >> PAGE_CACHE_SHIFT
;
1384 unsigned long len
= end
- start
+ 1;
1385 ret
= do_readahead(mapping
, file
, start
, len
);
1394 * page_cache_read - adds requested page to the page cache if not already there
1395 * @file: file to read
1396 * @offset: page index
1398 * This adds the requested page to the page cache if it isn't already there,
1399 * and schedules an I/O to read in its contents from disk.
1401 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1403 struct address_space
*mapping
= file
->f_mapping
;
1408 page
= page_cache_alloc_cold(mapping
);
1412 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1414 ret
= mapping
->a_ops
->readpage(file
, page
);
1415 else if (ret
== -EEXIST
)
1416 ret
= 0; /* losing race to add is OK */
1418 page_cache_release(page
);
1420 } while (ret
== AOP_TRUNCATED_PAGE
);
1425 #define MMAP_LOTSAMISS (100)
1428 * filemap_fault - read in file data for page fault handling
1429 * @vma: vma in which the fault was taken
1430 * @vmf: struct vm_fault containing details of the fault
1432 * filemap_fault() is invoked via the vma operations vector for a
1433 * mapped memory region to read in file data during a page fault.
1435 * The goto's are kind of ugly, but this streamlines the normal case of having
1436 * it in the page cache, and handles the special cases reasonably without
1437 * having a lot of duplicated code.
1439 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1442 struct file
*file
= vma
->vm_file
;
1443 struct address_space
*mapping
= file
->f_mapping
;
1444 struct file_ra_state
*ra
= &file
->f_ra
;
1445 struct inode
*inode
= mapping
->host
;
1448 int did_readaround
= 0;
1451 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1452 if (vmf
->pgoff
>= size
)
1453 return VM_FAULT_SIGBUS
;
1455 /* If we don't want any read-ahead, don't bother */
1456 if (VM_RandomReadHint(vma
))
1457 goto no_cached_page
;
1460 * Do we have something in the page cache already?
1463 page
= find_lock_page(mapping
, vmf
->pgoff
);
1465 * For sequential accesses, we use the generic readahead logic.
1467 if (VM_SequentialReadHint(vma
)) {
1469 page_cache_sync_readahead(mapping
, ra
, file
,
1471 page
= find_lock_page(mapping
, vmf
->pgoff
);
1473 goto no_cached_page
;
1475 if (PageReadahead(page
)) {
1476 page_cache_async_readahead(mapping
, ra
, file
, page
,
1482 unsigned long ra_pages
;
1487 * Do we miss much more than hit in this file? If so,
1488 * stop bothering with read-ahead. It will only hurt.
1490 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1491 goto no_cached_page
;
1494 * To keep the pgmajfault counter straight, we need to
1495 * check did_readaround, as this is an inner loop.
1497 if (!did_readaround
) {
1498 ret
= VM_FAULT_MAJOR
;
1499 count_vm_event(PGMAJFAULT
);
1502 ra_pages
= max_sane_readahead(file
->f_ra
.ra_pages
);
1506 if (vmf
->pgoff
> ra_pages
/ 2)
1507 start
= vmf
->pgoff
- ra_pages
/ 2;
1508 do_page_cache_readahead(mapping
, file
, start
, ra_pages
);
1510 page
= find_lock_page(mapping
, vmf
->pgoff
);
1512 goto no_cached_page
;
1515 if (!did_readaround
)
1519 * We have a locked page in the page cache, now we need to check
1520 * that it's up-to-date. If not, it is going to be due to an error.
1522 if (unlikely(!PageUptodate(page
)))
1523 goto page_not_uptodate
;
1525 /* Must recheck i_size under page lock */
1526 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1527 if (unlikely(vmf
->pgoff
>= size
)) {
1529 page_cache_release(page
);
1530 return VM_FAULT_SIGBUS
;
1534 * Found the page and have a reference on it.
1536 mark_page_accessed(page
);
1537 ra
->prev_pos
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
1539 return ret
| VM_FAULT_LOCKED
;
1543 * We're only likely to ever get here if MADV_RANDOM is in
1546 error
= page_cache_read(file
, vmf
->pgoff
);
1549 * The page we want has now been added to the page cache.
1550 * In the unlikely event that someone removed it in the
1551 * meantime, we'll just come back here and read it again.
1557 * An error return from page_cache_read can result if the
1558 * system is low on memory, or a problem occurs while trying
1561 if (error
== -ENOMEM
)
1562 return VM_FAULT_OOM
;
1563 return VM_FAULT_SIGBUS
;
1567 if (!did_readaround
) {
1568 ret
= VM_FAULT_MAJOR
;
1569 count_vm_event(PGMAJFAULT
);
1573 * Umm, take care of errors if the page isn't up-to-date.
1574 * Try to re-read it _once_. We do this synchronously,
1575 * because there really aren't any performance issues here
1576 * and we need to check for errors.
1578 ClearPageError(page
);
1579 error
= mapping
->a_ops
->readpage(file
, page
);
1581 wait_on_page_locked(page
);
1582 if (!PageUptodate(page
))
1585 page_cache_release(page
);
1587 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1590 /* Things didn't work out. Return zero to tell the mm layer so. */
1591 shrink_readahead_size_eio(file
, ra
);
1592 return VM_FAULT_SIGBUS
;
1594 EXPORT_SYMBOL(filemap_fault
);
1596 struct vm_operations_struct generic_file_vm_ops
= {
1597 .fault
= filemap_fault
,
1600 /* This is used for a general mmap of a disk file */
1602 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1604 struct address_space
*mapping
= file
->f_mapping
;
1606 if (!mapping
->a_ops
->readpage
)
1608 file_accessed(file
);
1609 vma
->vm_ops
= &generic_file_vm_ops
;
1610 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1615 * This is for filesystems which do not implement ->writepage.
1617 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1619 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
1621 return generic_file_mmap(file
, vma
);
1624 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1628 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1632 #endif /* CONFIG_MMU */
1634 EXPORT_SYMBOL(generic_file_mmap
);
1635 EXPORT_SYMBOL(generic_file_readonly_mmap
);
1637 static struct page
*__read_cache_page(struct address_space
*mapping
,
1639 int (*filler
)(void *,struct page
*),
1645 page
= find_get_page(mapping
, index
);
1647 page
= page_cache_alloc_cold(mapping
);
1649 return ERR_PTR(-ENOMEM
);
1650 err
= add_to_page_cache_lru(page
, mapping
, index
, GFP_KERNEL
);
1651 if (unlikely(err
)) {
1652 page_cache_release(page
);
1655 /* Presumably ENOMEM for radix tree node */
1656 return ERR_PTR(err
);
1658 err
= filler(data
, page
);
1660 page_cache_release(page
);
1661 page
= ERR_PTR(err
);
1668 * read_cache_page_async - read into page cache, fill it if needed
1669 * @mapping: the page's address_space
1670 * @index: the page index
1671 * @filler: function to perform the read
1672 * @data: destination for read data
1674 * Same as read_cache_page, but don't wait for page to become unlocked
1675 * after submitting it to the filler.
1677 * Read into the page cache. If a page already exists, and PageUptodate() is
1678 * not set, try to fill the page but don't wait for it to become unlocked.
1680 * If the page does not get brought uptodate, return -EIO.
1682 struct page
*read_cache_page_async(struct address_space
*mapping
,
1684 int (*filler
)(void *,struct page
*),
1691 page
= __read_cache_page(mapping
, index
, filler
, data
);
1694 if (PageUptodate(page
))
1698 if (!page
->mapping
) {
1700 page_cache_release(page
);
1703 if (PageUptodate(page
)) {
1707 err
= filler(data
, page
);
1709 page_cache_release(page
);
1710 return ERR_PTR(err
);
1713 mark_page_accessed(page
);
1716 EXPORT_SYMBOL(read_cache_page_async
);
1719 * read_cache_page - read into page cache, fill it if needed
1720 * @mapping: the page's address_space
1721 * @index: the page index
1722 * @filler: function to perform the read
1723 * @data: destination for read data
1725 * Read into the page cache. If a page already exists, and PageUptodate() is
1726 * not set, try to fill the page then wait for it to become unlocked.
1728 * If the page does not get brought uptodate, return -EIO.
1730 struct page
*read_cache_page(struct address_space
*mapping
,
1732 int (*filler
)(void *,struct page
*),
1737 page
= read_cache_page_async(mapping
, index
, filler
, data
);
1740 wait_on_page_locked(page
);
1741 if (!PageUptodate(page
)) {
1742 page_cache_release(page
);
1743 page
= ERR_PTR(-EIO
);
1748 EXPORT_SYMBOL(read_cache_page
);
1751 * The logic we want is
1753 * if suid or (sgid and xgrp)
1756 int should_remove_suid(struct dentry
*dentry
)
1758 mode_t mode
= dentry
->d_inode
->i_mode
;
1761 /* suid always must be killed */
1762 if (unlikely(mode
& S_ISUID
))
1763 kill
= ATTR_KILL_SUID
;
1766 * sgid without any exec bits is just a mandatory locking mark; leave
1767 * it alone. If some exec bits are set, it's a real sgid; kill it.
1769 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1770 kill
|= ATTR_KILL_SGID
;
1772 if (unlikely(kill
&& !capable(CAP_FSETID
)))
1777 EXPORT_SYMBOL(should_remove_suid
);
1779 static int __remove_suid(struct dentry
*dentry
, int kill
)
1781 struct iattr newattrs
;
1783 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1784 return notify_change(dentry
, &newattrs
);
1787 int file_remove_suid(struct file
*file
)
1789 struct dentry
*dentry
= file
->f_path
.dentry
;
1790 int killsuid
= should_remove_suid(dentry
);
1791 int killpriv
= security_inode_need_killpriv(dentry
);
1797 error
= security_inode_killpriv(dentry
);
1798 if (!error
&& killsuid
)
1799 error
= __remove_suid(dentry
, killsuid
);
1803 EXPORT_SYMBOL(file_remove_suid
);
1805 static size_t __iovec_copy_from_user_inatomic(char *vaddr
,
1806 const struct iovec
*iov
, size_t base
, size_t bytes
)
1808 size_t copied
= 0, left
= 0;
1811 char __user
*buf
= iov
->iov_base
+ base
;
1812 int copy
= min(bytes
, iov
->iov_len
- base
);
1815 left
= __copy_from_user_inatomic_nocache(vaddr
, buf
, copy
);
1824 return copied
- left
;
1828 * Copy as much as we can into the page and return the number of bytes which
1829 * were sucessfully copied. If a fault is encountered then return the number of
1830 * bytes which were copied.
1832 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
1833 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
1838 BUG_ON(!in_atomic());
1839 kaddr
= kmap_atomic(page
, KM_USER0
);
1840 if (likely(i
->nr_segs
== 1)) {
1842 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
1843 left
= __copy_from_user_inatomic_nocache(kaddr
+ offset
,
1845 copied
= bytes
- left
;
1847 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
1848 i
->iov
, i
->iov_offset
, bytes
);
1850 kunmap_atomic(kaddr
, KM_USER0
);
1854 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
1857 * This has the same sideeffects and return value as
1858 * iov_iter_copy_from_user_atomic().
1859 * The difference is that it attempts to resolve faults.
1860 * Page must not be locked.
1862 size_t iov_iter_copy_from_user(struct page
*page
,
1863 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
1869 if (likely(i
->nr_segs
== 1)) {
1871 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
1872 left
= __copy_from_user_nocache(kaddr
+ offset
, buf
, bytes
);
1873 copied
= bytes
- left
;
1875 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
1876 i
->iov
, i
->iov_offset
, bytes
);
1881 EXPORT_SYMBOL(iov_iter_copy_from_user
);
1883 void iov_iter_advance(struct iov_iter
*i
, size_t bytes
)
1885 BUG_ON(i
->count
< bytes
);
1887 if (likely(i
->nr_segs
== 1)) {
1888 i
->iov_offset
+= bytes
;
1891 const struct iovec
*iov
= i
->iov
;
1892 size_t base
= i
->iov_offset
;
1895 * The !iov->iov_len check ensures we skip over unlikely
1896 * zero-length segments (without overruning the iovec).
1898 while (bytes
|| unlikely(i
->count
&& !iov
->iov_len
)) {
1901 copy
= min(bytes
, iov
->iov_len
- base
);
1902 BUG_ON(!i
->count
|| i
->count
< copy
);
1906 if (iov
->iov_len
== base
) {
1912 i
->iov_offset
= base
;
1915 EXPORT_SYMBOL(iov_iter_advance
);
1918 * Fault in the first iovec of the given iov_iter, to a maximum length
1919 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1920 * accessed (ie. because it is an invalid address).
1922 * writev-intensive code may want this to prefault several iovecs -- that
1923 * would be possible (callers must not rely on the fact that _only_ the
1924 * first iovec will be faulted with the current implementation).
1926 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
1928 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
1929 bytes
= min(bytes
, i
->iov
->iov_len
- i
->iov_offset
);
1930 return fault_in_pages_readable(buf
, bytes
);
1932 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
1935 * Return the count of just the current iov_iter segment.
1937 size_t iov_iter_single_seg_count(struct iov_iter
*i
)
1939 const struct iovec
*iov
= i
->iov
;
1940 if (i
->nr_segs
== 1)
1943 return min(i
->count
, iov
->iov_len
- i
->iov_offset
);
1945 EXPORT_SYMBOL(iov_iter_single_seg_count
);
1948 * Performs necessary checks before doing a write
1950 * Can adjust writing position or amount of bytes to write.
1951 * Returns appropriate error code that caller should return or
1952 * zero in case that write should be allowed.
1954 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
1956 struct inode
*inode
= file
->f_mapping
->host
;
1957 unsigned long limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1959 if (unlikely(*pos
< 0))
1963 /* FIXME: this is for backwards compatibility with 2.4 */
1964 if (file
->f_flags
& O_APPEND
)
1965 *pos
= i_size_read(inode
);
1967 if (limit
!= RLIM_INFINITY
) {
1968 if (*pos
>= limit
) {
1969 send_sig(SIGXFSZ
, current
, 0);
1972 if (*count
> limit
- (typeof(limit
))*pos
) {
1973 *count
= limit
- (typeof(limit
))*pos
;
1981 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
1982 !(file
->f_flags
& O_LARGEFILE
))) {
1983 if (*pos
>= MAX_NON_LFS
) {
1986 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
1987 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
1992 * Are we about to exceed the fs block limit ?
1994 * If we have written data it becomes a short write. If we have
1995 * exceeded without writing data we send a signal and return EFBIG.
1996 * Linus frestrict idea will clean these up nicely..
1998 if (likely(!isblk
)) {
1999 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2000 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2003 /* zero-length writes at ->s_maxbytes are OK */
2006 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2007 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2011 if (bdev_read_only(I_BDEV(inode
)))
2013 isize
= i_size_read(inode
);
2014 if (*pos
>= isize
) {
2015 if (*count
|| *pos
> isize
)
2019 if (*pos
+ *count
> isize
)
2020 *count
= isize
- *pos
;
2027 EXPORT_SYMBOL(generic_write_checks
);
2029 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2030 loff_t pos
, unsigned len
, unsigned flags
,
2031 struct page
**pagep
, void **fsdata
)
2033 const struct address_space_operations
*aops
= mapping
->a_ops
;
2035 if (aops
->write_begin
) {
2036 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2040 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
2041 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
2042 struct inode
*inode
= mapping
->host
;
2045 page
= __grab_cache_page(mapping
, index
);
2050 if (flags
& AOP_FLAG_UNINTERRUPTIBLE
&& !PageUptodate(page
)) {
2052 * There is no way to resolve a short write situation
2053 * for a !Uptodate page (except by double copying in
2054 * the caller done by generic_perform_write_2copy).
2056 * Instead, we have to bring it uptodate here.
2058 ret
= aops
->readpage(file
, page
);
2059 page_cache_release(page
);
2061 if (ret
== AOP_TRUNCATED_PAGE
)
2068 ret
= aops
->prepare_write(file
, page
, offset
, offset
+len
);
2071 page_cache_release(page
);
2072 if (pos
+ len
> inode
->i_size
)
2073 vmtruncate(inode
, inode
->i_size
);
2078 EXPORT_SYMBOL(pagecache_write_begin
);
2080 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2081 loff_t pos
, unsigned len
, unsigned copied
,
2082 struct page
*page
, void *fsdata
)
2084 const struct address_space_operations
*aops
= mapping
->a_ops
;
2087 if (aops
->write_end
) {
2088 mark_page_accessed(page
);
2089 ret
= aops
->write_end(file
, mapping
, pos
, len
, copied
,
2092 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
2093 struct inode
*inode
= mapping
->host
;
2095 flush_dcache_page(page
);
2096 ret
= aops
->commit_write(file
, page
, offset
, offset
+len
);
2098 mark_page_accessed(page
);
2099 page_cache_release(page
);
2102 if (pos
+ len
> inode
->i_size
)
2103 vmtruncate(inode
, inode
->i_size
);
2105 ret
= min_t(size_t, copied
, ret
);
2112 EXPORT_SYMBOL(pagecache_write_end
);
2115 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2116 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2117 size_t count
, size_t ocount
)
2119 struct file
*file
= iocb
->ki_filp
;
2120 struct address_space
*mapping
= file
->f_mapping
;
2121 struct inode
*inode
= mapping
->host
;
2126 if (count
!= ocount
)
2127 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2130 * Unmap all mmappings of the file up-front.
2132 * This will cause any pte dirty bits to be propagated into the
2133 * pageframes for the subsequent filemap_write_and_wait().
2135 write_len
= iov_length(iov
, *nr_segs
);
2136 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2137 if (mapping_mapped(mapping
))
2138 unmap_mapping_range(mapping
, pos
, write_len
, 0);
2140 written
= filemap_write_and_wait(mapping
);
2145 * After a write we want buffered reads to be sure to go to disk to get
2146 * the new data. We invalidate clean cached page from the region we're
2147 * about to write. We do this *before* the write so that we can return
2148 * without clobbering -EIOCBQUEUED from ->direct_IO().
2150 if (mapping
->nrpages
) {
2151 written
= invalidate_inode_pages2_range(mapping
,
2152 pos
>> PAGE_CACHE_SHIFT
, end
);
2154 * If a page can not be invalidated, return 0 to fall back
2155 * to buffered write.
2158 if (written
== -EBUSY
)
2164 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2167 * Finally, try again to invalidate clean pages which might have been
2168 * cached by non-direct readahead, or faulted in by get_user_pages()
2169 * if the source of the write was an mmap'ed region of the file
2170 * we're writing. Either one is a pretty crazy thing to do,
2171 * so we don't support it 100%. If this invalidation
2172 * fails, tough, the write still worked...
2174 if (mapping
->nrpages
) {
2175 invalidate_inode_pages2_range(mapping
,
2176 pos
>> PAGE_CACHE_SHIFT
, end
);
2180 loff_t end
= pos
+ written
;
2181 if (end
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2182 i_size_write(inode
, end
);
2183 mark_inode_dirty(inode
);
2189 * Sync the fs metadata but not the minor inode changes and
2190 * of course not the data as we did direct DMA for the IO.
2191 * i_mutex is held, which protects generic_osync_inode() from
2192 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
2195 if ((written
>= 0 || written
== -EIOCBQUEUED
) &&
2196 ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2197 int err
= generic_osync_inode(inode
, mapping
, OSYNC_METADATA
);
2203 EXPORT_SYMBOL(generic_file_direct_write
);
2206 * Find or create a page at the given pagecache position. Return the locked
2207 * page. This function is specifically for buffered writes.
2209 struct page
*__grab_cache_page(struct address_space
*mapping
, pgoff_t index
)
2214 page
= find_lock_page(mapping
, index
);
2218 page
= page_cache_alloc(mapping
);
2221 status
= add_to_page_cache_lru(page
, mapping
, index
, GFP_KERNEL
);
2222 if (unlikely(status
)) {
2223 page_cache_release(page
);
2224 if (status
== -EEXIST
)
2230 EXPORT_SYMBOL(__grab_cache_page
);
2232 static ssize_t
generic_perform_write_2copy(struct file
*file
,
2233 struct iov_iter
*i
, loff_t pos
)
2235 struct address_space
*mapping
= file
->f_mapping
;
2236 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2237 struct inode
*inode
= mapping
->host
;
2239 ssize_t written
= 0;
2242 struct page
*src_page
;
2244 pgoff_t index
; /* Pagecache index for current page */
2245 unsigned long offset
; /* Offset into pagecache page */
2246 unsigned long bytes
; /* Bytes to write to page */
2247 size_t copied
; /* Bytes copied from user */
2249 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2250 index
= pos
>> PAGE_CACHE_SHIFT
;
2251 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2255 * a non-NULL src_page indicates that we're doing the
2256 * copy via get_user_pages and kmap.
2261 * Bring in the user page that we will copy from _first_.
2262 * Otherwise there's a nasty deadlock on copying from the
2263 * same page as we're writing to, without it being marked
2266 * Not only is this an optimisation, but it is also required
2267 * to check that the address is actually valid, when atomic
2268 * usercopies are used, below.
2270 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2275 page
= __grab_cache_page(mapping
, index
);
2282 * non-uptodate pages cannot cope with short copies, and we
2283 * cannot take a pagefault with the destination page locked.
2284 * So pin the source page to copy it.
2286 if (!PageUptodate(page
) && !segment_eq(get_fs(), KERNEL_DS
)) {
2289 src_page
= alloc_page(GFP_KERNEL
);
2291 page_cache_release(page
);
2297 * Cannot get_user_pages with a page locked for the
2298 * same reason as we can't take a page fault with a
2299 * page locked (as explained below).
2301 copied
= iov_iter_copy_from_user(src_page
, i
,
2303 if (unlikely(copied
== 0)) {
2305 page_cache_release(page
);
2306 page_cache_release(src_page
);
2313 * Can't handle the page going uptodate here, because
2314 * that means we would use non-atomic usercopies, which
2315 * zero out the tail of the page, which can cause
2316 * zeroes to become transiently visible. We could just
2317 * use a non-zeroing copy, but the APIs aren't too
2320 if (unlikely(!page
->mapping
|| PageUptodate(page
))) {
2322 page_cache_release(page
);
2323 page_cache_release(src_page
);
2328 status
= a_ops
->prepare_write(file
, page
, offset
, offset
+bytes
);
2329 if (unlikely(status
))
2330 goto fs_write_aop_error
;
2334 * Must not enter the pagefault handler here, because
2335 * we hold the page lock, so we might recursively
2336 * deadlock on the same lock, or get an ABBA deadlock
2337 * against a different lock, or against the mmap_sem
2338 * (which nests outside the page lock). So increment
2339 * preempt count, and use _atomic usercopies.
2341 * The page is uptodate so we are OK to encounter a
2342 * short copy: if unmodified parts of the page are
2343 * marked dirty and written out to disk, it doesn't
2346 pagefault_disable();
2347 copied
= iov_iter_copy_from_user_atomic(page
, i
,
2352 src
= kmap_atomic(src_page
, KM_USER0
);
2353 dst
= kmap_atomic(page
, KM_USER1
);
2354 memcpy(dst
+ offset
, src
+ offset
, bytes
);
2355 kunmap_atomic(dst
, KM_USER1
);
2356 kunmap_atomic(src
, KM_USER0
);
2359 flush_dcache_page(page
);
2361 status
= a_ops
->commit_write(file
, page
, offset
, offset
+bytes
);
2362 if (unlikely(status
< 0))
2363 goto fs_write_aop_error
;
2364 if (unlikely(status
> 0)) /* filesystem did partial write */
2365 copied
= min_t(size_t, copied
, status
);
2368 mark_page_accessed(page
);
2369 page_cache_release(page
);
2371 page_cache_release(src_page
);
2373 iov_iter_advance(i
, copied
);
2377 balance_dirty_pages_ratelimited(mapping
);
2383 page_cache_release(page
);
2385 page_cache_release(src_page
);
2388 * prepare_write() may have instantiated a few blocks
2389 * outside i_size. Trim these off again. Don't need
2390 * i_size_read because we hold i_mutex.
2392 if (pos
+ bytes
> inode
->i_size
)
2393 vmtruncate(inode
, inode
->i_size
);
2395 } while (iov_iter_count(i
));
2397 return written
? written
: status
;
2400 static ssize_t
generic_perform_write(struct file
*file
,
2401 struct iov_iter
*i
, loff_t pos
)
2403 struct address_space
*mapping
= file
->f_mapping
;
2404 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2406 ssize_t written
= 0;
2407 unsigned int flags
= 0;
2410 * Copies from kernel address space cannot fail (NFSD is a big user).
2412 if (segment_eq(get_fs(), KERNEL_DS
))
2413 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2417 pgoff_t index
; /* Pagecache index for current page */
2418 unsigned long offset
; /* Offset into pagecache page */
2419 unsigned long bytes
; /* Bytes to write to page */
2420 size_t copied
; /* Bytes copied from user */
2423 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2424 index
= pos
>> PAGE_CACHE_SHIFT
;
2425 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2431 * Bring in the user page that we will copy from _first_.
2432 * Otherwise there's a nasty deadlock on copying from the
2433 * same page as we're writing to, without it being marked
2436 * Not only is this an optimisation, but it is also required
2437 * to check that the address is actually valid, when atomic
2438 * usercopies are used, below.
2440 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2445 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2447 if (unlikely(status
))
2450 pagefault_disable();
2451 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2453 flush_dcache_page(page
);
2455 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2457 if (unlikely(status
< 0))
2463 iov_iter_advance(i
, copied
);
2464 if (unlikely(copied
== 0)) {
2466 * If we were unable to copy any data at all, we must
2467 * fall back to a single segment length write.
2469 * If we didn't fallback here, we could livelock
2470 * because not all segments in the iov can be copied at
2471 * once without a pagefault.
2473 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2474 iov_iter_single_seg_count(i
));
2480 balance_dirty_pages_ratelimited(mapping
);
2482 } while (iov_iter_count(i
));
2484 return written
? written
: status
;
2488 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2489 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2490 size_t count
, ssize_t written
)
2492 struct file
*file
= iocb
->ki_filp
;
2493 struct address_space
*mapping
= file
->f_mapping
;
2494 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2495 struct inode
*inode
= mapping
->host
;
2499 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
2500 if (a_ops
->write_begin
)
2501 status
= generic_perform_write(file
, &i
, pos
);
2503 status
= generic_perform_write_2copy(file
, &i
, pos
);
2505 if (likely(status
>= 0)) {
2507 *ppos
= pos
+ status
;
2510 * For now, when the user asks for O_SYNC, we'll actually give
2513 if (unlikely((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2514 if (!a_ops
->writepage
|| !is_sync_kiocb(iocb
))
2515 status
= generic_osync_inode(inode
, mapping
,
2516 OSYNC_METADATA
|OSYNC_DATA
);
2521 * If we get here for O_DIRECT writes then we must have fallen through
2522 * to buffered writes (block instantiation inside i_size). So we sync
2523 * the file data here, to try to honour O_DIRECT expectations.
2525 if (unlikely(file
->f_flags
& O_DIRECT
) && written
)
2526 status
= filemap_write_and_wait(mapping
);
2528 return written
? written
: status
;
2530 EXPORT_SYMBOL(generic_file_buffered_write
);
2533 __generic_file_aio_write_nolock(struct kiocb
*iocb
, const struct iovec
*iov
,
2534 unsigned long nr_segs
, loff_t
*ppos
)
2536 struct file
*file
= iocb
->ki_filp
;
2537 struct address_space
* mapping
= file
->f_mapping
;
2538 size_t ocount
; /* original count */
2539 size_t count
; /* after file limit checks */
2540 struct inode
*inode
= mapping
->host
;
2546 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
2553 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2555 /* We can write back this queue in page reclaim */
2556 current
->backing_dev_info
= mapping
->backing_dev_info
;
2559 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2566 err
= file_remove_suid(file
);
2570 file_update_time(file
);
2572 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2573 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2575 ssize_t written_buffered
;
2577 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2578 ppos
, count
, ocount
);
2579 if (written
< 0 || written
== count
)
2582 * direct-io write to a hole: fall through to buffered I/O
2583 * for completing the rest of the request.
2587 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2588 nr_segs
, pos
, ppos
, count
,
2591 * If generic_file_buffered_write() retuned a synchronous error
2592 * then we want to return the number of bytes which were
2593 * direct-written, or the error code if that was zero. Note
2594 * that this differs from normal direct-io semantics, which
2595 * will return -EFOO even if some bytes were written.
2597 if (written_buffered
< 0) {
2598 err
= written_buffered
;
2603 * We need to ensure that the page cache pages are written to
2604 * disk and invalidated to preserve the expected O_DIRECT
2607 endbyte
= pos
+ written_buffered
- written
- 1;
2608 err
= do_sync_mapping_range(file
->f_mapping
, pos
, endbyte
,
2609 SYNC_FILE_RANGE_WAIT_BEFORE
|
2610 SYNC_FILE_RANGE_WRITE
|
2611 SYNC_FILE_RANGE_WAIT_AFTER
);
2613 written
= written_buffered
;
2614 invalidate_mapping_pages(mapping
,
2615 pos
>> PAGE_CACHE_SHIFT
,
2616 endbyte
>> PAGE_CACHE_SHIFT
);
2619 * We don't know how much we wrote, so just return
2620 * the number of bytes which were direct-written
2624 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2625 pos
, ppos
, count
, written
);
2628 current
->backing_dev_info
= NULL
;
2629 return written
? written
: err
;
2632 ssize_t
generic_file_aio_write_nolock(struct kiocb
*iocb
,
2633 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
2635 struct file
*file
= iocb
->ki_filp
;
2636 struct address_space
*mapping
= file
->f_mapping
;
2637 struct inode
*inode
= mapping
->host
;
2640 BUG_ON(iocb
->ki_pos
!= pos
);
2642 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
,
2645 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2648 err
= sync_page_range_nolock(inode
, mapping
, pos
, ret
);
2654 EXPORT_SYMBOL(generic_file_aio_write_nolock
);
2656 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2657 unsigned long nr_segs
, loff_t pos
)
2659 struct file
*file
= iocb
->ki_filp
;
2660 struct address_space
*mapping
= file
->f_mapping
;
2661 struct inode
*inode
= mapping
->host
;
2664 BUG_ON(iocb
->ki_pos
!= pos
);
2666 mutex_lock(&inode
->i_mutex
);
2667 ret
= __generic_file_aio_write_nolock(iocb
, iov
, nr_segs
,
2669 mutex_unlock(&inode
->i_mutex
);
2671 if (ret
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2674 err
= sync_page_range(inode
, mapping
, pos
, ret
);
2680 EXPORT_SYMBOL(generic_file_aio_write
);
2683 * try_to_release_page() - release old fs-specific metadata on a page
2685 * @page: the page which the kernel is trying to free
2686 * @gfp_mask: memory allocation flags (and I/O mode)
2688 * The address_space is to try to release any data against the page
2689 * (presumably at page->private). If the release was successful, return `1'.
2690 * Otherwise return zero.
2692 * The @gfp_mask argument specifies whether I/O may be performed to release
2693 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2696 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2698 struct address_space
* const mapping
= page
->mapping
;
2700 BUG_ON(!PageLocked(page
));
2701 if (PageWriteback(page
))
2704 if (mapping
&& mapping
->a_ops
->releasepage
)
2705 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2706 return try_to_free_buffers(page
);
2709 EXPORT_SYMBOL(try_to_release_page
);