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/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/aio.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.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 try_to_free_buffers */
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
50 * Shared mappings now work. 15.8.1995 Bruno.
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
61 * ->i_mmap_lock (truncate_pagecache)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
67 * ->i_mmap_lock (truncate->unmap_mapping_range)
71 * ->page_table_lock or pte_lock (various, mainly in memory.c)
72 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
75 * ->lock_page (access_process_vm)
77 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
81 * ->i_alloc_sem (various)
84 * ->sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 * (code doesn't rely on that order, so you could switch it around)
106 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 * Delete 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 __delete_from_page_cache(struct page
*page
)
117 struct address_space
*mapping
= page
->mapping
;
119 radix_tree_delete(&mapping
->page_tree
, page
->index
);
120 page
->mapping
= NULL
;
122 __dec_zone_page_state(page
, NR_FILE_PAGES
);
123 if (PageSwapBacked(page
))
124 __dec_zone_page_state(page
, NR_SHMEM
);
125 BUG_ON(page_mapped(page
));
128 * Some filesystems seem to re-dirty the page even after
129 * the VM has canceled the dirty bit (eg ext3 journaling).
131 * Fix it up by doing a final dirty accounting check after
132 * having removed the page entirely.
134 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
135 dec_zone_page_state(page
, NR_FILE_DIRTY
);
136 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
141 * delete_from_page_cache - delete page from page cache
142 * @page: the page which the kernel is trying to remove from page cache
144 * This must be called only on pages that have been verified to be in the page
145 * cache and locked. It will never put the page into the free list, the caller
146 * has a reference on the page.
148 void delete_from_page_cache(struct page
*page
)
150 struct address_space
*mapping
= page
->mapping
;
151 void (*freepage
)(struct page
*);
153 BUG_ON(!PageLocked(page
));
155 freepage
= mapping
->a_ops
->freepage
;
156 spin_lock_irq(&mapping
->tree_lock
);
157 __delete_from_page_cache(page
);
158 spin_unlock_irq(&mapping
->tree_lock
);
159 mem_cgroup_uncharge_cache_page(page
);
163 page_cache_release(page
);
165 EXPORT_SYMBOL(delete_from_page_cache
);
167 static int sync_page(void *word
)
169 struct address_space
*mapping
;
172 page
= container_of((unsigned long *)word
, struct page
, flags
);
175 * page_mapping() is being called without PG_locked held.
176 * Some knowledge of the state and use of the page is used to
177 * reduce the requirements down to a memory barrier.
178 * The danger here is of a stale page_mapping() return value
179 * indicating a struct address_space different from the one it's
180 * associated with when it is associated with one.
181 * After smp_mb(), it's either the correct page_mapping() for
182 * the page, or an old page_mapping() and the page's own
183 * page_mapping() has gone NULL.
184 * The ->sync_page() address_space operation must tolerate
185 * page_mapping() going NULL. By an amazing coincidence,
186 * this comes about because none of the users of the page
187 * in the ->sync_page() methods make essential use of the
188 * page_mapping(), merely passing the page down to the backing
189 * device's unplug functions when it's non-NULL, which in turn
190 * ignore it for all cases but swap, where only page_private(page) is
191 * of interest. When page_mapping() does go NULL, the entire
192 * call stack gracefully ignores the page and returns.
196 mapping
= page_mapping(page
);
197 if (mapping
&& mapping
->a_ops
&& mapping
->a_ops
->sync_page
)
198 mapping
->a_ops
->sync_page(page
);
203 static int sync_page_killable(void *word
)
206 return fatal_signal_pending(current
) ? -EINTR
: 0;
210 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
211 * @mapping: address space structure to write
212 * @start: offset in bytes where the range starts
213 * @end: offset in bytes where the range ends (inclusive)
214 * @sync_mode: enable synchronous operation
216 * Start writeback against all of a mapping's dirty pages that lie
217 * within the byte offsets <start, end> inclusive.
219 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
220 * opposed to a regular memory cleansing writeback. The difference between
221 * these two operations is that if a dirty page/buffer is encountered, it must
222 * be waited upon, and not just skipped over.
224 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
225 loff_t end
, int sync_mode
)
228 struct writeback_control wbc
= {
229 .sync_mode
= sync_mode
,
230 .nr_to_write
= LONG_MAX
,
231 .range_start
= start
,
235 if (!mapping_cap_writeback_dirty(mapping
))
238 ret
= do_writepages(mapping
, &wbc
);
242 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
245 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
248 int filemap_fdatawrite(struct address_space
*mapping
)
250 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
252 EXPORT_SYMBOL(filemap_fdatawrite
);
254 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
257 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
259 EXPORT_SYMBOL(filemap_fdatawrite_range
);
262 * filemap_flush - mostly a non-blocking flush
263 * @mapping: target address_space
265 * This is a mostly non-blocking flush. Not suitable for data-integrity
266 * purposes - I/O may not be started against all dirty pages.
268 int filemap_flush(struct address_space
*mapping
)
270 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
272 EXPORT_SYMBOL(filemap_flush
);
275 * filemap_fdatawait_range - wait for writeback to complete
276 * @mapping: address space structure to wait for
277 * @start_byte: offset in bytes where the range starts
278 * @end_byte: offset in bytes where the range ends (inclusive)
280 * Walk the list of under-writeback pages of the given address space
281 * in the given range and wait for all of them.
283 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
286 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
287 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
292 if (end_byte
< start_byte
)
295 pagevec_init(&pvec
, 0);
296 while ((index
<= end
) &&
297 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
298 PAGECACHE_TAG_WRITEBACK
,
299 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
302 for (i
= 0; i
< nr_pages
; i
++) {
303 struct page
*page
= pvec
.pages
[i
];
305 /* until radix tree lookup accepts end_index */
306 if (page
->index
> end
)
309 wait_on_page_writeback(page
);
310 if (TestClearPageError(page
))
313 pagevec_release(&pvec
);
317 /* Check for outstanding write errors */
318 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
320 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
325 EXPORT_SYMBOL(filemap_fdatawait_range
);
328 * filemap_fdatawait - wait for all under-writeback pages to complete
329 * @mapping: address space structure to wait for
331 * Walk the list of under-writeback pages of the given address space
332 * and wait for all of them.
334 int filemap_fdatawait(struct address_space
*mapping
)
336 loff_t i_size
= i_size_read(mapping
->host
);
341 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
343 EXPORT_SYMBOL(filemap_fdatawait
);
345 int filemap_write_and_wait(struct address_space
*mapping
)
349 if (mapping
->nrpages
) {
350 err
= filemap_fdatawrite(mapping
);
352 * Even if the above returned error, the pages may be
353 * written partially (e.g. -ENOSPC), so we wait for it.
354 * But the -EIO is special case, it may indicate the worst
355 * thing (e.g. bug) happened, so we avoid waiting for it.
358 int err2
= filemap_fdatawait(mapping
);
365 EXPORT_SYMBOL(filemap_write_and_wait
);
368 * filemap_write_and_wait_range - write out & wait on a file range
369 * @mapping: the address_space for the pages
370 * @lstart: offset in bytes where the range starts
371 * @lend: offset in bytes where the range ends (inclusive)
373 * Write out and wait upon file offsets lstart->lend, inclusive.
375 * Note that `lend' is inclusive (describes the last byte to be written) so
376 * that this function can be used to write to the very end-of-file (end = -1).
378 int filemap_write_and_wait_range(struct address_space
*mapping
,
379 loff_t lstart
, loff_t lend
)
383 if (mapping
->nrpages
) {
384 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
386 /* See comment of filemap_write_and_wait() */
388 int err2
= filemap_fdatawait_range(mapping
,
396 EXPORT_SYMBOL(filemap_write_and_wait_range
);
399 * replace_page_cache_page - replace a pagecache page with a new one
400 * @old: page to be replaced
401 * @new: page to replace with
402 * @gfp_mask: allocation mode
404 * This function replaces a page in the pagecache with a new one. On
405 * success it acquires the pagecache reference for the new page and
406 * drops it for the old page. Both the old and new pages must be
407 * locked. This function does not add the new page to the LRU, the
408 * caller must do that.
410 * The remove + add is atomic. The only way this function can fail is
411 * memory allocation failure.
413 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
416 struct mem_cgroup
*memcg
= NULL
;
418 VM_BUG_ON(!PageLocked(old
));
419 VM_BUG_ON(!PageLocked(new));
420 VM_BUG_ON(new->mapping
);
423 * This is not page migration, but prepare_migration and
424 * end_migration does enough work for charge replacement.
426 * In the longer term we probably want a specialized function
427 * for moving the charge from old to new in a more efficient
430 error
= mem_cgroup_prepare_migration(old
, new, &memcg
, gfp_mask
);
434 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
436 struct address_space
*mapping
= old
->mapping
;
437 void (*freepage
)(struct page
*);
439 pgoff_t offset
= old
->index
;
440 freepage
= mapping
->a_ops
->freepage
;
443 new->mapping
= mapping
;
446 spin_lock_irq(&mapping
->tree_lock
);
447 __delete_from_page_cache(old
);
448 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
451 __inc_zone_page_state(new, NR_FILE_PAGES
);
452 if (PageSwapBacked(new))
453 __inc_zone_page_state(new, NR_SHMEM
);
454 spin_unlock_irq(&mapping
->tree_lock
);
455 radix_tree_preload_end();
458 page_cache_release(old
);
459 mem_cgroup_end_migration(memcg
, old
, new, true);
461 mem_cgroup_end_migration(memcg
, old
, new, false);
466 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
469 * add_to_page_cache_locked - add a locked page to the pagecache
471 * @mapping: the page's address_space
472 * @offset: page index
473 * @gfp_mask: page allocation mode
475 * This function is used to add a page to the pagecache. It must be locked.
476 * This function does not add the page to the LRU. The caller must do that.
478 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
479 pgoff_t offset
, gfp_t gfp_mask
)
483 VM_BUG_ON(!PageLocked(page
));
485 error
= mem_cgroup_cache_charge(page
, current
->mm
,
486 gfp_mask
& GFP_RECLAIM_MASK
);
490 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
492 page_cache_get(page
);
493 page
->mapping
= mapping
;
494 page
->index
= offset
;
496 spin_lock_irq(&mapping
->tree_lock
);
497 error
= radix_tree_insert(&mapping
->page_tree
, offset
, page
);
498 if (likely(!error
)) {
500 __inc_zone_page_state(page
, NR_FILE_PAGES
);
501 if (PageSwapBacked(page
))
502 __inc_zone_page_state(page
, NR_SHMEM
);
503 spin_unlock_irq(&mapping
->tree_lock
);
505 page
->mapping
= NULL
;
506 spin_unlock_irq(&mapping
->tree_lock
);
507 mem_cgroup_uncharge_cache_page(page
);
508 page_cache_release(page
);
510 radix_tree_preload_end();
512 mem_cgroup_uncharge_cache_page(page
);
516 EXPORT_SYMBOL(add_to_page_cache_locked
);
518 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
519 pgoff_t offset
, gfp_t gfp_mask
)
524 * Splice_read and readahead add shmem/tmpfs pages into the page cache
525 * before shmem_readpage has a chance to mark them as SwapBacked: they
526 * need to go on the anon lru below, and mem_cgroup_cache_charge
527 * (called in add_to_page_cache) needs to know where they're going too.
529 if (mapping_cap_swap_backed(mapping
))
530 SetPageSwapBacked(page
);
532 ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
534 if (page_is_file_cache(page
))
535 lru_cache_add_file(page
);
537 lru_cache_add_anon(page
);
541 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
544 struct page
*__page_cache_alloc(gfp_t gfp
)
549 if (cpuset_do_page_mem_spread()) {
551 n
= cpuset_mem_spread_node();
552 page
= alloc_pages_exact_node(n
, gfp
, 0);
556 return alloc_pages(gfp
, 0);
558 EXPORT_SYMBOL(__page_cache_alloc
);
561 static int __sleep_on_page_lock(void *word
)
568 * In order to wait for pages to become available there must be
569 * waitqueues associated with pages. By using a hash table of
570 * waitqueues where the bucket discipline is to maintain all
571 * waiters on the same queue and wake all when any of the pages
572 * become available, and for the woken contexts to check to be
573 * sure the appropriate page became available, this saves space
574 * at a cost of "thundering herd" phenomena during rare hash
577 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
579 const struct zone
*zone
= page_zone(page
);
581 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
584 static inline void wake_up_page(struct page
*page
, int bit
)
586 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
589 void wait_on_page_bit(struct page
*page
, int bit_nr
)
591 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
593 if (test_bit(bit_nr
, &page
->flags
))
594 __wait_on_bit(page_waitqueue(page
), &wait
, sync_page
,
595 TASK_UNINTERRUPTIBLE
);
597 EXPORT_SYMBOL(wait_on_page_bit
);
600 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
601 * @page: Page defining the wait queue of interest
602 * @waiter: Waiter to add to the queue
604 * Add an arbitrary @waiter to the wait queue for the nominated @page.
606 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
608 wait_queue_head_t
*q
= page_waitqueue(page
);
611 spin_lock_irqsave(&q
->lock
, flags
);
612 __add_wait_queue(q
, waiter
);
613 spin_unlock_irqrestore(&q
->lock
, flags
);
615 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
618 * unlock_page - unlock a locked page
621 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
622 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
623 * mechananism between PageLocked pages and PageWriteback pages is shared.
624 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
626 * The mb is necessary to enforce ordering between the clear_bit and the read
627 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
629 void unlock_page(struct page
*page
)
631 VM_BUG_ON(!PageLocked(page
));
632 clear_bit_unlock(PG_locked
, &page
->flags
);
633 smp_mb__after_clear_bit();
634 wake_up_page(page
, PG_locked
);
636 EXPORT_SYMBOL(unlock_page
);
639 * end_page_writeback - end writeback against a page
642 void end_page_writeback(struct page
*page
)
644 if (TestClearPageReclaim(page
))
645 rotate_reclaimable_page(page
);
647 if (!test_clear_page_writeback(page
))
650 smp_mb__after_clear_bit();
651 wake_up_page(page
, PG_writeback
);
653 EXPORT_SYMBOL(end_page_writeback
);
656 * __lock_page - get a lock on the page, assuming we need to sleep to get it
657 * @page: the page to lock
659 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
660 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
661 * chances are that on the second loop, the block layer's plug list is empty,
662 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
664 void __lock_page(struct page
*page
)
666 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
668 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sync_page
,
669 TASK_UNINTERRUPTIBLE
);
671 EXPORT_SYMBOL(__lock_page
);
673 int __lock_page_killable(struct page
*page
)
675 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
677 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
678 sync_page_killable
, TASK_KILLABLE
);
680 EXPORT_SYMBOL_GPL(__lock_page_killable
);
683 * __lock_page_nosync - get a lock on the page, without calling sync_page()
684 * @page: the page to lock
686 * Variant of lock_page that does not require the caller to hold a reference
687 * on the page's mapping.
689 void __lock_page_nosync(struct page
*page
)
691 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
692 __wait_on_bit_lock(page_waitqueue(page
), &wait
, __sleep_on_page_lock
,
693 TASK_UNINTERRUPTIBLE
);
696 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
699 if (!(flags
& FAULT_FLAG_ALLOW_RETRY
)) {
703 if (!(flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
704 up_read(&mm
->mmap_sem
);
705 wait_on_page_locked(page
);
712 * find_get_page - find and get a page reference
713 * @mapping: the address_space to search
714 * @offset: the page index
716 * Is there a pagecache struct page at the given (mapping, offset) tuple?
717 * If yes, increment its refcount and return it; if no, return NULL.
719 struct page
*find_get_page(struct address_space
*mapping
, pgoff_t offset
)
727 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
729 page
= radix_tree_deref_slot(pagep
);
732 if (radix_tree_deref_retry(page
))
735 if (!page_cache_get_speculative(page
))
739 * Has the page moved?
740 * This is part of the lockless pagecache protocol. See
741 * include/linux/pagemap.h for details.
743 if (unlikely(page
!= *pagep
)) {
744 page_cache_release(page
);
753 EXPORT_SYMBOL(find_get_page
);
756 * find_lock_page - locate, pin and lock a pagecache page
757 * @mapping: the address_space to search
758 * @offset: the page index
760 * Locates the desired pagecache page, locks it, increments its reference
761 * count and returns its address.
763 * Returns zero if the page was not present. find_lock_page() may sleep.
765 struct page
*find_lock_page(struct address_space
*mapping
, pgoff_t offset
)
770 page
= find_get_page(mapping
, offset
);
773 /* Has the page been truncated? */
774 if (unlikely(page
->mapping
!= mapping
)) {
776 page_cache_release(page
);
779 VM_BUG_ON(page
->index
!= offset
);
783 EXPORT_SYMBOL(find_lock_page
);
786 * find_or_create_page - locate or add a pagecache page
787 * @mapping: the page's address_space
788 * @index: the page's index into the mapping
789 * @gfp_mask: page allocation mode
791 * Locates a page in the pagecache. If the page is not present, a new page
792 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
793 * LRU list. The returned page is locked and has its reference count
796 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
799 * find_or_create_page() returns the desired page's address, or zero on
802 struct page
*find_or_create_page(struct address_space
*mapping
,
803 pgoff_t index
, gfp_t gfp_mask
)
808 page
= find_lock_page(mapping
, index
);
810 page
= __page_cache_alloc(gfp_mask
);
814 * We want a regular kernel memory (not highmem or DMA etc)
815 * allocation for the radix tree nodes, but we need to honour
816 * the context-specific requirements the caller has asked for.
817 * GFP_RECLAIM_MASK collects those requirements.
819 err
= add_to_page_cache_lru(page
, mapping
, index
,
820 (gfp_mask
& GFP_RECLAIM_MASK
));
822 page_cache_release(page
);
830 EXPORT_SYMBOL(find_or_create_page
);
833 * find_get_pages - gang pagecache lookup
834 * @mapping: The address_space to search
835 * @start: The starting page index
836 * @nr_pages: The maximum number of pages
837 * @pages: Where the resulting pages are placed
839 * find_get_pages() will search for and return a group of up to
840 * @nr_pages pages in the mapping. The pages are placed at @pages.
841 * find_get_pages() takes a reference against the returned pages.
843 * The search returns a group of mapping-contiguous pages with ascending
844 * indexes. There may be holes in the indices due to not-present pages.
846 * find_get_pages() returns the number of pages which were found.
848 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
849 unsigned int nr_pages
, struct page
**pages
)
853 unsigned int nr_found
;
857 nr_found
= radix_tree_gang_lookup_slot(&mapping
->page_tree
,
858 (void ***)pages
, start
, nr_pages
);
860 for (i
= 0; i
< nr_found
; i
++) {
863 page
= radix_tree_deref_slot((void **)pages
[i
]);
868 * This can only trigger when the entry at index 0 moves out
869 * of or back to the root: none yet gotten, safe to restart.
871 if (radix_tree_deref_retry(page
)) {
876 if (!page_cache_get_speculative(page
))
879 /* Has the page moved? */
880 if (unlikely(page
!= *((void **)pages
[i
]))) {
881 page_cache_release(page
);
890 * If all entries were removed before we could secure them,
891 * try again, because callers stop trying once 0 is returned.
893 if (unlikely(!ret
&& nr_found
))
900 * find_get_pages_contig - gang contiguous pagecache lookup
901 * @mapping: The address_space to search
902 * @index: The starting page index
903 * @nr_pages: The maximum number of pages
904 * @pages: Where the resulting pages are placed
906 * find_get_pages_contig() works exactly like find_get_pages(), except
907 * that the returned number of pages are guaranteed to be contiguous.
909 * find_get_pages_contig() returns the number of pages which were found.
911 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
912 unsigned int nr_pages
, struct page
**pages
)
916 unsigned int nr_found
;
920 nr_found
= radix_tree_gang_lookup_slot(&mapping
->page_tree
,
921 (void ***)pages
, index
, nr_pages
);
923 for (i
= 0; i
< nr_found
; i
++) {
926 page
= radix_tree_deref_slot((void **)pages
[i
]);
931 * This can only trigger when the entry at index 0 moves out
932 * of or back to the root: none yet gotten, safe to restart.
934 if (radix_tree_deref_retry(page
))
937 if (!page_cache_get_speculative(page
))
940 /* Has the page moved? */
941 if (unlikely(page
!= *((void **)pages
[i
]))) {
942 page_cache_release(page
);
947 * must check mapping and index after taking the ref.
948 * otherwise we can get both false positives and false
949 * negatives, which is just confusing to the caller.
951 if (page
->mapping
== NULL
|| page
->index
!= index
) {
952 page_cache_release(page
);
963 EXPORT_SYMBOL(find_get_pages_contig
);
966 * find_get_pages_tag - find and return pages that match @tag
967 * @mapping: the address_space to search
968 * @index: the starting page index
969 * @tag: the tag index
970 * @nr_pages: the maximum number of pages
971 * @pages: where the resulting pages are placed
973 * Like find_get_pages, except we only return pages which are tagged with
974 * @tag. We update @index to index the next page for the traversal.
976 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
977 int tag
, unsigned int nr_pages
, struct page
**pages
)
981 unsigned int nr_found
;
985 nr_found
= radix_tree_gang_lookup_tag_slot(&mapping
->page_tree
,
986 (void ***)pages
, *index
, nr_pages
, tag
);
988 for (i
= 0; i
< nr_found
; i
++) {
991 page
= radix_tree_deref_slot((void **)pages
[i
]);
996 * This can only trigger when the entry at index 0 moves out
997 * of or back to the root: none yet gotten, safe to restart.
999 if (radix_tree_deref_retry(page
))
1002 if (!page_cache_get_speculative(page
))
1005 /* Has the page moved? */
1006 if (unlikely(page
!= *((void **)pages
[i
]))) {
1007 page_cache_release(page
);
1016 * If all entries were removed before we could secure them,
1017 * try again, because callers stop trying once 0 is returned.
1019 if (unlikely(!ret
&& nr_found
))
1024 *index
= pages
[ret
- 1]->index
+ 1;
1028 EXPORT_SYMBOL(find_get_pages_tag
);
1031 * grab_cache_page_nowait - returns locked page at given index in given cache
1032 * @mapping: target address_space
1033 * @index: the page index
1035 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1036 * This is intended for speculative data generators, where the data can
1037 * be regenerated if the page couldn't be grabbed. This routine should
1038 * be safe to call while holding the lock for another page.
1040 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1041 * and deadlock against the caller's locked page.
1044 grab_cache_page_nowait(struct address_space
*mapping
, pgoff_t index
)
1046 struct page
*page
= find_get_page(mapping
, index
);
1049 if (trylock_page(page
))
1051 page_cache_release(page
);
1054 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
1055 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_NOFS
)) {
1056 page_cache_release(page
);
1061 EXPORT_SYMBOL(grab_cache_page_nowait
);
1064 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1065 * a _large_ part of the i/o request. Imagine the worst scenario:
1067 * ---R__________________________________________B__________
1068 * ^ reading here ^ bad block(assume 4k)
1070 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1071 * => failing the whole request => read(R) => read(R+1) =>
1072 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1073 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1074 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1076 * It is going insane. Fix it by quickly scaling down the readahead size.
1078 static void shrink_readahead_size_eio(struct file
*filp
,
1079 struct file_ra_state
*ra
)
1085 * do_generic_file_read - generic file read routine
1086 * @filp: the file to read
1087 * @ppos: current file position
1088 * @desc: read_descriptor
1089 * @actor: read method
1091 * This is a generic file read routine, and uses the
1092 * mapping->a_ops->readpage() function for the actual low-level stuff.
1094 * This is really ugly. But the goto's actually try to clarify some
1095 * of the logic when it comes to error handling etc.
1097 static void do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1098 read_descriptor_t
*desc
, read_actor_t actor
)
1100 struct address_space
*mapping
= filp
->f_mapping
;
1101 struct inode
*inode
= mapping
->host
;
1102 struct file_ra_state
*ra
= &filp
->f_ra
;
1106 unsigned long offset
; /* offset into pagecache page */
1107 unsigned int prev_offset
;
1110 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1111 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1112 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1113 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1114 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1120 unsigned long nr
, ret
;
1124 page
= find_get_page(mapping
, index
);
1126 page_cache_sync_readahead(mapping
,
1128 index
, last_index
- index
);
1129 page
= find_get_page(mapping
, index
);
1130 if (unlikely(page
== NULL
))
1131 goto no_cached_page
;
1133 if (PageReadahead(page
)) {
1134 page_cache_async_readahead(mapping
,
1136 index
, last_index
- index
);
1138 if (!PageUptodate(page
)) {
1139 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1140 !mapping
->a_ops
->is_partially_uptodate
)
1141 goto page_not_up_to_date
;
1142 if (!trylock_page(page
))
1143 goto page_not_up_to_date
;
1144 /* Did it get truncated before we got the lock? */
1146 goto page_not_up_to_date_locked
;
1147 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1149 goto page_not_up_to_date_locked
;
1154 * i_size must be checked after we know the page is Uptodate.
1156 * Checking i_size after the check allows us to calculate
1157 * the correct value for "nr", which means the zero-filled
1158 * part of the page is not copied back to userspace (unless
1159 * another truncate extends the file - this is desired though).
1162 isize
= i_size_read(inode
);
1163 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1164 if (unlikely(!isize
|| index
> end_index
)) {
1165 page_cache_release(page
);
1169 /* nr is the maximum number of bytes to copy from this page */
1170 nr
= PAGE_CACHE_SIZE
;
1171 if (index
== end_index
) {
1172 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1174 page_cache_release(page
);
1180 /* If users can be writing to this page using arbitrary
1181 * virtual addresses, take care about potential aliasing
1182 * before reading the page on the kernel side.
1184 if (mapping_writably_mapped(mapping
))
1185 flush_dcache_page(page
);
1188 * When a sequential read accesses a page several times,
1189 * only mark it as accessed the first time.
1191 if (prev_index
!= index
|| offset
!= prev_offset
)
1192 mark_page_accessed(page
);
1196 * Ok, we have the page, and it's up-to-date, so
1197 * now we can copy it to user space...
1199 * The actor routine returns how many bytes were actually used..
1200 * NOTE! This may not be the same as how much of a user buffer
1201 * we filled up (we may be padding etc), so we can only update
1202 * "pos" here (the actor routine has to update the user buffer
1203 * pointers and the remaining count).
1205 ret
= actor(desc
, page
, offset
, nr
);
1207 index
+= offset
>> PAGE_CACHE_SHIFT
;
1208 offset
&= ~PAGE_CACHE_MASK
;
1209 prev_offset
= offset
;
1211 page_cache_release(page
);
1212 if (ret
== nr
&& desc
->count
)
1216 page_not_up_to_date
:
1217 /* Get exclusive access to the page ... */
1218 error
= lock_page_killable(page
);
1219 if (unlikely(error
))
1220 goto readpage_error
;
1222 page_not_up_to_date_locked
:
1223 /* Did it get truncated before we got the lock? */
1224 if (!page
->mapping
) {
1226 page_cache_release(page
);
1230 /* Did somebody else fill it already? */
1231 if (PageUptodate(page
)) {
1238 * A previous I/O error may have been due to temporary
1239 * failures, eg. multipath errors.
1240 * PG_error will be set again if readpage fails.
1242 ClearPageError(page
);
1243 /* Start the actual read. The read will unlock the page. */
1244 error
= mapping
->a_ops
->readpage(filp
, page
);
1246 if (unlikely(error
)) {
1247 if (error
== AOP_TRUNCATED_PAGE
) {
1248 page_cache_release(page
);
1251 goto readpage_error
;
1254 if (!PageUptodate(page
)) {
1255 error
= lock_page_killable(page
);
1256 if (unlikely(error
))
1257 goto readpage_error
;
1258 if (!PageUptodate(page
)) {
1259 if (page
->mapping
== NULL
) {
1261 * invalidate_mapping_pages got it
1264 page_cache_release(page
);
1268 shrink_readahead_size_eio(filp
, ra
);
1270 goto readpage_error
;
1278 /* UHHUH! A synchronous read error occurred. Report it */
1279 desc
->error
= error
;
1280 page_cache_release(page
);
1285 * Ok, it wasn't cached, so we need to create a new
1288 page
= page_cache_alloc_cold(mapping
);
1290 desc
->error
= -ENOMEM
;
1293 error
= add_to_page_cache_lru(page
, mapping
,
1296 page_cache_release(page
);
1297 if (error
== -EEXIST
)
1299 desc
->error
= error
;
1306 ra
->prev_pos
= prev_index
;
1307 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1308 ra
->prev_pos
|= prev_offset
;
1310 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1311 file_accessed(filp
);
1314 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
1315 unsigned long offset
, unsigned long size
)
1318 unsigned long left
, count
= desc
->count
;
1324 * Faults on the destination of a read are common, so do it before
1327 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
1328 kaddr
= kmap_atomic(page
, KM_USER0
);
1329 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
1330 kaddr
+ offset
, size
);
1331 kunmap_atomic(kaddr
, KM_USER0
);
1336 /* Do it the slow way */
1338 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
1343 desc
->error
= -EFAULT
;
1346 desc
->count
= count
- size
;
1347 desc
->written
+= size
;
1348 desc
->arg
.buf
+= size
;
1353 * Performs necessary checks before doing a write
1354 * @iov: io vector request
1355 * @nr_segs: number of segments in the iovec
1356 * @count: number of bytes to write
1357 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1359 * Adjust number of segments and amount of bytes to write (nr_segs should be
1360 * properly initialized first). Returns appropriate error code that caller
1361 * should return or zero in case that write should be allowed.
1363 int generic_segment_checks(const struct iovec
*iov
,
1364 unsigned long *nr_segs
, size_t *count
, int access_flags
)
1368 for (seg
= 0; seg
< *nr_segs
; seg
++) {
1369 const struct iovec
*iv
= &iov
[seg
];
1372 * If any segment has a negative length, or the cumulative
1373 * length ever wraps negative then return -EINVAL.
1376 if (unlikely((ssize_t
)(cnt
|iv
->iov_len
) < 0))
1378 if (access_ok(access_flags
, iv
->iov_base
, iv
->iov_len
))
1383 cnt
-= iv
->iov_len
; /* This segment is no good */
1389 EXPORT_SYMBOL(generic_segment_checks
);
1392 * generic_file_aio_read - generic filesystem read routine
1393 * @iocb: kernel I/O control block
1394 * @iov: io vector request
1395 * @nr_segs: number of segments in the iovec
1396 * @pos: current file position
1398 * This is the "read()" routine for all filesystems
1399 * that can use the page cache directly.
1402 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1403 unsigned long nr_segs
, loff_t pos
)
1405 struct file
*filp
= iocb
->ki_filp
;
1407 unsigned long seg
= 0;
1409 loff_t
*ppos
= &iocb
->ki_pos
;
1412 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1416 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1417 if (filp
->f_flags
& O_DIRECT
) {
1419 struct address_space
*mapping
;
1420 struct inode
*inode
;
1422 mapping
= filp
->f_mapping
;
1423 inode
= mapping
->host
;
1425 goto out
; /* skip atime */
1426 size
= i_size_read(inode
);
1428 retval
= filemap_write_and_wait_range(mapping
, pos
,
1429 pos
+ iov_length(iov
, nr_segs
) - 1);
1431 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
,
1435 *ppos
= pos
+ retval
;
1440 * Btrfs can have a short DIO read if we encounter
1441 * compressed extents, so if there was an error, or if
1442 * we've already read everything we wanted to, or if
1443 * there was a short read because we hit EOF, go ahead
1444 * and return. Otherwise fallthrough to buffered io for
1445 * the rest of the read.
1447 if (retval
< 0 || !count
|| *ppos
>= size
) {
1448 file_accessed(filp
);
1455 for (seg
= 0; seg
< nr_segs
; seg
++) {
1456 read_descriptor_t desc
;
1460 * If we did a short DIO read we need to skip the section of the
1461 * iov that we've already read data into.
1464 if (count
> iov
[seg
].iov_len
) {
1465 count
-= iov
[seg
].iov_len
;
1473 desc
.arg
.buf
= iov
[seg
].iov_base
+ offset
;
1474 desc
.count
= iov
[seg
].iov_len
- offset
;
1475 if (desc
.count
== 0)
1478 do_generic_file_read(filp
, ppos
, &desc
, file_read_actor
);
1479 retval
+= desc
.written
;
1481 retval
= retval
?: desc
.error
;
1490 EXPORT_SYMBOL(generic_file_aio_read
);
1493 do_readahead(struct address_space
*mapping
, struct file
*filp
,
1494 pgoff_t index
, unsigned long nr
)
1496 if (!mapping
|| !mapping
->a_ops
|| !mapping
->a_ops
->readpage
)
1499 force_page_cache_readahead(mapping
, filp
, index
, nr
);
1503 SYSCALL_DEFINE(readahead
)(int fd
, loff_t offset
, size_t count
)
1511 if (file
->f_mode
& FMODE_READ
) {
1512 struct address_space
*mapping
= file
->f_mapping
;
1513 pgoff_t start
= offset
>> PAGE_CACHE_SHIFT
;
1514 pgoff_t end
= (offset
+ count
- 1) >> PAGE_CACHE_SHIFT
;
1515 unsigned long len
= end
- start
+ 1;
1516 ret
= do_readahead(mapping
, file
, start
, len
);
1522 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1523 asmlinkage
long SyS_readahead(long fd
, loff_t offset
, long count
)
1525 return SYSC_readahead((int) fd
, offset
, (size_t) count
);
1527 SYSCALL_ALIAS(sys_readahead
, SyS_readahead
);
1532 * page_cache_read - adds requested page to the page cache if not already there
1533 * @file: file to read
1534 * @offset: page index
1536 * This adds the requested page to the page cache if it isn't already there,
1537 * and schedules an I/O to read in its contents from disk.
1539 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1541 struct address_space
*mapping
= file
->f_mapping
;
1546 page
= page_cache_alloc_cold(mapping
);
1550 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1552 ret
= mapping
->a_ops
->readpage(file
, page
);
1553 else if (ret
== -EEXIST
)
1554 ret
= 0; /* losing race to add is OK */
1556 page_cache_release(page
);
1558 } while (ret
== AOP_TRUNCATED_PAGE
);
1563 #define MMAP_LOTSAMISS (100)
1566 * Synchronous readahead happens when we don't even find
1567 * a page in the page cache at all.
1569 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1570 struct file_ra_state
*ra
,
1574 unsigned long ra_pages
;
1575 struct address_space
*mapping
= file
->f_mapping
;
1577 /* If we don't want any read-ahead, don't bother */
1578 if (VM_RandomReadHint(vma
))
1581 if (VM_SequentialReadHint(vma
) ||
1582 offset
- 1 == (ra
->prev_pos
>> PAGE_CACHE_SHIFT
)) {
1583 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1588 if (ra
->mmap_miss
< INT_MAX
)
1592 * Do we miss much more than hit in this file? If so,
1593 * stop bothering with read-ahead. It will only hurt.
1595 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1601 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1603 ra
->start
= max_t(long, 0, offset
- ra_pages
/2);
1604 ra
->size
= ra_pages
;
1606 ra_submit(ra
, mapping
, file
);
1611 * Asynchronous readahead happens when we find the page and PG_readahead,
1612 * so we want to possibly extend the readahead further..
1614 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1615 struct file_ra_state
*ra
,
1620 struct address_space
*mapping
= file
->f_mapping
;
1622 /* If we don't want any read-ahead, don't bother */
1623 if (VM_RandomReadHint(vma
))
1625 if (ra
->mmap_miss
> 0)
1627 if (PageReadahead(page
))
1628 page_cache_async_readahead(mapping
, ra
, file
,
1629 page
, offset
, ra
->ra_pages
);
1633 * filemap_fault - read in file data for page fault handling
1634 * @vma: vma in which the fault was taken
1635 * @vmf: struct vm_fault containing details of the fault
1637 * filemap_fault() is invoked via the vma operations vector for a
1638 * mapped memory region to read in file data during a page fault.
1640 * The goto's are kind of ugly, but this streamlines the normal case of having
1641 * it in the page cache, and handles the special cases reasonably without
1642 * having a lot of duplicated code.
1644 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1647 struct file
*file
= vma
->vm_file
;
1648 struct address_space
*mapping
= file
->f_mapping
;
1649 struct file_ra_state
*ra
= &file
->f_ra
;
1650 struct inode
*inode
= mapping
->host
;
1651 pgoff_t offset
= vmf
->pgoff
;
1656 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1658 return VM_FAULT_SIGBUS
;
1661 * Do we have something in the page cache already?
1663 page
= find_get_page(mapping
, offset
);
1666 * We found the page, so try async readahead before
1667 * waiting for the lock.
1669 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1671 /* No page in the page cache at all */
1672 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1673 count_vm_event(PGMAJFAULT
);
1674 ret
= VM_FAULT_MAJOR
;
1676 page
= find_get_page(mapping
, offset
);
1678 goto no_cached_page
;
1681 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1682 page_cache_release(page
);
1683 return ret
| VM_FAULT_RETRY
;
1686 /* Did it get truncated? */
1687 if (unlikely(page
->mapping
!= mapping
)) {
1692 VM_BUG_ON(page
->index
!= offset
);
1695 * We have a locked page in the page cache, now we need to check
1696 * that it's up-to-date. If not, it is going to be due to an error.
1698 if (unlikely(!PageUptodate(page
)))
1699 goto page_not_uptodate
;
1702 * Found the page and have a reference on it.
1703 * We must recheck i_size under page lock.
1705 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1706 if (unlikely(offset
>= size
)) {
1708 page_cache_release(page
);
1709 return VM_FAULT_SIGBUS
;
1712 ra
->prev_pos
= (loff_t
)offset
<< PAGE_CACHE_SHIFT
;
1714 return ret
| VM_FAULT_LOCKED
;
1718 * We're only likely to ever get here if MADV_RANDOM is in
1721 error
= page_cache_read(file
, offset
);
1724 * The page we want has now been added to the page cache.
1725 * In the unlikely event that someone removed it in the
1726 * meantime, we'll just come back here and read it again.
1732 * An error return from page_cache_read can result if the
1733 * system is low on memory, or a problem occurs while trying
1736 if (error
== -ENOMEM
)
1737 return VM_FAULT_OOM
;
1738 return VM_FAULT_SIGBUS
;
1742 * Umm, take care of errors if the page isn't up-to-date.
1743 * Try to re-read it _once_. We do this synchronously,
1744 * because there really aren't any performance issues here
1745 * and we need to check for errors.
1747 ClearPageError(page
);
1748 error
= mapping
->a_ops
->readpage(file
, page
);
1750 wait_on_page_locked(page
);
1751 if (!PageUptodate(page
))
1754 page_cache_release(page
);
1756 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1759 /* Things didn't work out. Return zero to tell the mm layer so. */
1760 shrink_readahead_size_eio(file
, ra
);
1761 return VM_FAULT_SIGBUS
;
1763 EXPORT_SYMBOL(filemap_fault
);
1765 const struct vm_operations_struct generic_file_vm_ops
= {
1766 .fault
= filemap_fault
,
1769 /* This is used for a general mmap of a disk file */
1771 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1773 struct address_space
*mapping
= file
->f_mapping
;
1775 if (!mapping
->a_ops
->readpage
)
1777 file_accessed(file
);
1778 vma
->vm_ops
= &generic_file_vm_ops
;
1779 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1784 * This is for filesystems which do not implement ->writepage.
1786 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1788 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
1790 return generic_file_mmap(file
, vma
);
1793 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1797 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1801 #endif /* CONFIG_MMU */
1803 EXPORT_SYMBOL(generic_file_mmap
);
1804 EXPORT_SYMBOL(generic_file_readonly_mmap
);
1806 static struct page
*__read_cache_page(struct address_space
*mapping
,
1808 int (*filler
)(void *,struct page
*),
1815 page
= find_get_page(mapping
, index
);
1817 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
1819 return ERR_PTR(-ENOMEM
);
1820 err
= add_to_page_cache_lru(page
, mapping
, index
, GFP_KERNEL
);
1821 if (unlikely(err
)) {
1822 page_cache_release(page
);
1825 /* Presumably ENOMEM for radix tree node */
1826 return ERR_PTR(err
);
1828 err
= filler(data
, page
);
1830 page_cache_release(page
);
1831 page
= ERR_PTR(err
);
1837 static struct page
*do_read_cache_page(struct address_space
*mapping
,
1839 int (*filler
)(void *,struct page
*),
1848 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
1851 if (PageUptodate(page
))
1855 if (!page
->mapping
) {
1857 page_cache_release(page
);
1860 if (PageUptodate(page
)) {
1864 err
= filler(data
, page
);
1866 page_cache_release(page
);
1867 return ERR_PTR(err
);
1870 mark_page_accessed(page
);
1875 * read_cache_page_async - read into page cache, fill it if needed
1876 * @mapping: the page's address_space
1877 * @index: the page index
1878 * @filler: function to perform the read
1879 * @data: destination for read data
1881 * Same as read_cache_page, but don't wait for page to become unlocked
1882 * after submitting it to the filler.
1884 * Read into the page cache. If a page already exists, and PageUptodate() is
1885 * not set, try to fill the page but don't wait for it to become unlocked.
1887 * If the page does not get brought uptodate, return -EIO.
1889 struct page
*read_cache_page_async(struct address_space
*mapping
,
1891 int (*filler
)(void *,struct page
*),
1894 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
1896 EXPORT_SYMBOL(read_cache_page_async
);
1898 static struct page
*wait_on_page_read(struct page
*page
)
1900 if (!IS_ERR(page
)) {
1901 wait_on_page_locked(page
);
1902 if (!PageUptodate(page
)) {
1903 page_cache_release(page
);
1904 page
= ERR_PTR(-EIO
);
1911 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1912 * @mapping: the page's address_space
1913 * @index: the page index
1914 * @gfp: the page allocator flags to use if allocating
1916 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1917 * any new page allocations done using the specified allocation flags. Note
1918 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1919 * expect to do this atomically or anything like that - but you can pass in
1920 * other page requirements.
1922 * If the page does not get brought uptodate, return -EIO.
1924 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
1928 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
1930 return wait_on_page_read(do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
));
1932 EXPORT_SYMBOL(read_cache_page_gfp
);
1935 * read_cache_page - read into page cache, fill it if needed
1936 * @mapping: the page's address_space
1937 * @index: the page index
1938 * @filler: function to perform the read
1939 * @data: destination for read data
1941 * Read into the page cache. If a page already exists, and PageUptodate() is
1942 * not set, try to fill the page then wait for it to become unlocked.
1944 * If the page does not get brought uptodate, return -EIO.
1946 struct page
*read_cache_page(struct address_space
*mapping
,
1948 int (*filler
)(void *,struct page
*),
1951 return wait_on_page_read(read_cache_page_async(mapping
, index
, filler
, data
));
1953 EXPORT_SYMBOL(read_cache_page
);
1956 * The logic we want is
1958 * if suid or (sgid and xgrp)
1961 int should_remove_suid(struct dentry
*dentry
)
1963 mode_t mode
= dentry
->d_inode
->i_mode
;
1966 /* suid always must be killed */
1967 if (unlikely(mode
& S_ISUID
))
1968 kill
= ATTR_KILL_SUID
;
1971 * sgid without any exec bits is just a mandatory locking mark; leave
1972 * it alone. If some exec bits are set, it's a real sgid; kill it.
1974 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1975 kill
|= ATTR_KILL_SGID
;
1977 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1982 EXPORT_SYMBOL(should_remove_suid
);
1984 static int __remove_suid(struct dentry
*dentry
, int kill
)
1986 struct iattr newattrs
;
1988 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1989 return notify_change(dentry
, &newattrs
);
1992 int file_remove_suid(struct file
*file
)
1994 struct dentry
*dentry
= file
->f_path
.dentry
;
1995 int killsuid
= should_remove_suid(dentry
);
1996 int killpriv
= security_inode_need_killpriv(dentry
);
2002 error
= security_inode_killpriv(dentry
);
2003 if (!error
&& killsuid
)
2004 error
= __remove_suid(dentry
, killsuid
);
2008 EXPORT_SYMBOL(file_remove_suid
);
2010 static size_t __iovec_copy_from_user_inatomic(char *vaddr
,
2011 const struct iovec
*iov
, size_t base
, size_t bytes
)
2013 size_t copied
= 0, left
= 0;
2016 char __user
*buf
= iov
->iov_base
+ base
;
2017 int copy
= min(bytes
, iov
->iov_len
- base
);
2020 left
= __copy_from_user_inatomic(vaddr
, buf
, copy
);
2029 return copied
- left
;
2033 * Copy as much as we can into the page and return the number of bytes which
2034 * were successfully copied. If a fault is encountered then return the number of
2035 * bytes which were copied.
2037 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
2038 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2043 BUG_ON(!in_atomic());
2044 kaddr
= kmap_atomic(page
, KM_USER0
);
2045 if (likely(i
->nr_segs
== 1)) {
2047 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2048 left
= __copy_from_user_inatomic(kaddr
+ offset
, buf
, bytes
);
2049 copied
= bytes
- left
;
2051 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2052 i
->iov
, i
->iov_offset
, bytes
);
2054 kunmap_atomic(kaddr
, KM_USER0
);
2058 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
2061 * This has the same sideeffects and return value as
2062 * iov_iter_copy_from_user_atomic().
2063 * The difference is that it attempts to resolve faults.
2064 * Page must not be locked.
2066 size_t iov_iter_copy_from_user(struct page
*page
,
2067 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2073 if (likely(i
->nr_segs
== 1)) {
2075 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2076 left
= __copy_from_user(kaddr
+ offset
, buf
, bytes
);
2077 copied
= bytes
- left
;
2079 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2080 i
->iov
, i
->iov_offset
, bytes
);
2085 EXPORT_SYMBOL(iov_iter_copy_from_user
);
2087 void iov_iter_advance(struct iov_iter
*i
, size_t bytes
)
2089 BUG_ON(i
->count
< bytes
);
2091 if (likely(i
->nr_segs
== 1)) {
2092 i
->iov_offset
+= bytes
;
2095 const struct iovec
*iov
= i
->iov
;
2096 size_t base
= i
->iov_offset
;
2099 * The !iov->iov_len check ensures we skip over unlikely
2100 * zero-length segments (without overruning the iovec).
2102 while (bytes
|| unlikely(i
->count
&& !iov
->iov_len
)) {
2105 copy
= min(bytes
, iov
->iov_len
- base
);
2106 BUG_ON(!i
->count
|| i
->count
< copy
);
2110 if (iov
->iov_len
== base
) {
2116 i
->iov_offset
= base
;
2119 EXPORT_SYMBOL(iov_iter_advance
);
2122 * Fault in the first iovec of the given iov_iter, to a maximum length
2123 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2124 * accessed (ie. because it is an invalid address).
2126 * writev-intensive code may want this to prefault several iovecs -- that
2127 * would be possible (callers must not rely on the fact that _only_ the
2128 * first iovec will be faulted with the current implementation).
2130 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
2132 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2133 bytes
= min(bytes
, i
->iov
->iov_len
- i
->iov_offset
);
2134 return fault_in_pages_readable(buf
, bytes
);
2136 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
2139 * Return the count of just the current iov_iter segment.
2141 size_t iov_iter_single_seg_count(struct iov_iter
*i
)
2143 const struct iovec
*iov
= i
->iov
;
2144 if (i
->nr_segs
== 1)
2147 return min(i
->count
, iov
->iov_len
- i
->iov_offset
);
2149 EXPORT_SYMBOL(iov_iter_single_seg_count
);
2152 * Performs necessary checks before doing a write
2154 * Can adjust writing position or amount of bytes to write.
2155 * Returns appropriate error code that caller should return or
2156 * zero in case that write should be allowed.
2158 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
2160 struct inode
*inode
= file
->f_mapping
->host
;
2161 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2163 if (unlikely(*pos
< 0))
2167 /* FIXME: this is for backwards compatibility with 2.4 */
2168 if (file
->f_flags
& O_APPEND
)
2169 *pos
= i_size_read(inode
);
2171 if (limit
!= RLIM_INFINITY
) {
2172 if (*pos
>= limit
) {
2173 send_sig(SIGXFSZ
, current
, 0);
2176 if (*count
> limit
- (typeof(limit
))*pos
) {
2177 *count
= limit
- (typeof(limit
))*pos
;
2185 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2186 !(file
->f_flags
& O_LARGEFILE
))) {
2187 if (*pos
>= MAX_NON_LFS
) {
2190 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2191 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2196 * Are we about to exceed the fs block limit ?
2198 * If we have written data it becomes a short write. If we have
2199 * exceeded without writing data we send a signal and return EFBIG.
2200 * Linus frestrict idea will clean these up nicely..
2202 if (likely(!isblk
)) {
2203 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2204 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2207 /* zero-length writes at ->s_maxbytes are OK */
2210 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2211 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2215 if (bdev_read_only(I_BDEV(inode
)))
2217 isize
= i_size_read(inode
);
2218 if (*pos
>= isize
) {
2219 if (*count
|| *pos
> isize
)
2223 if (*pos
+ *count
> isize
)
2224 *count
= isize
- *pos
;
2231 EXPORT_SYMBOL(generic_write_checks
);
2233 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2234 loff_t pos
, unsigned len
, unsigned flags
,
2235 struct page
**pagep
, void **fsdata
)
2237 const struct address_space_operations
*aops
= mapping
->a_ops
;
2239 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2242 EXPORT_SYMBOL(pagecache_write_begin
);
2244 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2245 loff_t pos
, unsigned len
, unsigned copied
,
2246 struct page
*page
, void *fsdata
)
2248 const struct address_space_operations
*aops
= mapping
->a_ops
;
2250 mark_page_accessed(page
);
2251 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2253 EXPORT_SYMBOL(pagecache_write_end
);
2256 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2257 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2258 size_t count
, size_t ocount
)
2260 struct file
*file
= iocb
->ki_filp
;
2261 struct address_space
*mapping
= file
->f_mapping
;
2262 struct inode
*inode
= mapping
->host
;
2267 if (count
!= ocount
)
2268 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2270 write_len
= iov_length(iov
, *nr_segs
);
2271 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2273 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2278 * After a write we want buffered reads to be sure to go to disk to get
2279 * the new data. We invalidate clean cached page from the region we're
2280 * about to write. We do this *before* the write so that we can return
2281 * without clobbering -EIOCBQUEUED from ->direct_IO().
2283 if (mapping
->nrpages
) {
2284 written
= invalidate_inode_pages2_range(mapping
,
2285 pos
>> PAGE_CACHE_SHIFT
, end
);
2287 * If a page can not be invalidated, return 0 to fall back
2288 * to buffered write.
2291 if (written
== -EBUSY
)
2297 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2300 * Finally, try again to invalidate clean pages which might have been
2301 * cached by non-direct readahead, or faulted in by get_user_pages()
2302 * if the source of the write was an mmap'ed region of the file
2303 * we're writing. Either one is a pretty crazy thing to do,
2304 * so we don't support it 100%. If this invalidation
2305 * fails, tough, the write still worked...
2307 if (mapping
->nrpages
) {
2308 invalidate_inode_pages2_range(mapping
,
2309 pos
>> PAGE_CACHE_SHIFT
, end
);
2314 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2315 i_size_write(inode
, pos
);
2316 mark_inode_dirty(inode
);
2323 EXPORT_SYMBOL(generic_file_direct_write
);
2326 * Find or create a page at the given pagecache position. Return the locked
2327 * page. This function is specifically for buffered writes.
2329 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2330 pgoff_t index
, unsigned flags
)
2334 gfp_t gfp_notmask
= 0;
2335 if (flags
& AOP_FLAG_NOFS
)
2336 gfp_notmask
= __GFP_FS
;
2338 page
= find_lock_page(mapping
, index
);
2342 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~gfp_notmask
);
2345 status
= add_to_page_cache_lru(page
, mapping
, index
,
2346 GFP_KERNEL
& ~gfp_notmask
);
2347 if (unlikely(status
)) {
2348 page_cache_release(page
);
2349 if (status
== -EEXIST
)
2355 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2357 static ssize_t
generic_perform_write(struct file
*file
,
2358 struct iov_iter
*i
, loff_t pos
)
2360 struct address_space
*mapping
= file
->f_mapping
;
2361 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2363 ssize_t written
= 0;
2364 unsigned int flags
= 0;
2367 * Copies from kernel address space cannot fail (NFSD is a big user).
2369 if (segment_eq(get_fs(), KERNEL_DS
))
2370 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2374 unsigned long offset
; /* Offset into pagecache page */
2375 unsigned long bytes
; /* Bytes to write to page */
2376 size_t copied
; /* Bytes copied from user */
2379 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2380 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2386 * Bring in the user page that we will copy from _first_.
2387 * Otherwise there's a nasty deadlock on copying from the
2388 * same page as we're writing to, without it being marked
2391 * Not only is this an optimisation, but it is also required
2392 * to check that the address is actually valid, when atomic
2393 * usercopies are used, below.
2395 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2400 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2402 if (unlikely(status
))
2405 if (mapping_writably_mapped(mapping
))
2406 flush_dcache_page(page
);
2408 pagefault_disable();
2409 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2411 flush_dcache_page(page
);
2413 mark_page_accessed(page
);
2414 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2416 if (unlikely(status
< 0))
2422 iov_iter_advance(i
, copied
);
2423 if (unlikely(copied
== 0)) {
2425 * If we were unable to copy any data at all, we must
2426 * fall back to a single segment length write.
2428 * If we didn't fallback here, we could livelock
2429 * because not all segments in the iov can be copied at
2430 * once without a pagefault.
2432 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2433 iov_iter_single_seg_count(i
));
2439 balance_dirty_pages_ratelimited(mapping
);
2441 } while (iov_iter_count(i
));
2443 return written
? written
: status
;
2447 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2448 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2449 size_t count
, ssize_t written
)
2451 struct file
*file
= iocb
->ki_filp
;
2455 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
2456 status
= generic_perform_write(file
, &i
, pos
);
2458 if (likely(status
>= 0)) {
2460 *ppos
= pos
+ status
;
2463 return written
? written
: status
;
2465 EXPORT_SYMBOL(generic_file_buffered_write
);
2468 * __generic_file_aio_write - write data to a file
2469 * @iocb: IO state structure (file, offset, etc.)
2470 * @iov: vector with data to write
2471 * @nr_segs: number of segments in the vector
2472 * @ppos: position where to write
2474 * This function does all the work needed for actually writing data to a
2475 * file. It does all basic checks, removes SUID from the file, updates
2476 * modification times and calls proper subroutines depending on whether we
2477 * do direct IO or a standard buffered write.
2479 * It expects i_mutex to be grabbed unless we work on a block device or similar
2480 * object which does not need locking at all.
2482 * This function does *not* take care of syncing data in case of O_SYNC write.
2483 * A caller has to handle it. This is mainly due to the fact that we want to
2484 * avoid syncing under i_mutex.
2486 ssize_t
__generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2487 unsigned long nr_segs
, loff_t
*ppos
)
2489 struct file
*file
= iocb
->ki_filp
;
2490 struct address_space
* mapping
= file
->f_mapping
;
2491 size_t ocount
; /* original count */
2492 size_t count
; /* after file limit checks */
2493 struct inode
*inode
= mapping
->host
;
2499 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
2506 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2508 /* We can write back this queue in page reclaim */
2509 current
->backing_dev_info
= mapping
->backing_dev_info
;
2512 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2519 err
= file_remove_suid(file
);
2523 file_update_time(file
);
2525 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2526 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2528 ssize_t written_buffered
;
2530 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2531 ppos
, count
, ocount
);
2532 if (written
< 0 || written
== count
)
2535 * direct-io write to a hole: fall through to buffered I/O
2536 * for completing the rest of the request.
2540 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2541 nr_segs
, pos
, ppos
, count
,
2544 * If generic_file_buffered_write() retuned a synchronous error
2545 * then we want to return the number of bytes which were
2546 * direct-written, or the error code if that was zero. Note
2547 * that this differs from normal direct-io semantics, which
2548 * will return -EFOO even if some bytes were written.
2550 if (written_buffered
< 0) {
2551 err
= written_buffered
;
2556 * We need to ensure that the page cache pages are written to
2557 * disk and invalidated to preserve the expected O_DIRECT
2560 endbyte
= pos
+ written_buffered
- written
- 1;
2561 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
2563 written
= written_buffered
;
2564 invalidate_mapping_pages(mapping
,
2565 pos
>> PAGE_CACHE_SHIFT
,
2566 endbyte
>> PAGE_CACHE_SHIFT
);
2569 * We don't know how much we wrote, so just return
2570 * the number of bytes which were direct-written
2574 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2575 pos
, ppos
, count
, written
);
2578 current
->backing_dev_info
= NULL
;
2579 return written
? written
: err
;
2581 EXPORT_SYMBOL(__generic_file_aio_write
);
2584 * generic_file_aio_write - write data to a file
2585 * @iocb: IO state structure
2586 * @iov: vector with data to write
2587 * @nr_segs: number of segments in the vector
2588 * @pos: position in file where to write
2590 * This is a wrapper around __generic_file_aio_write() to be used by most
2591 * filesystems. It takes care of syncing the file in case of O_SYNC file
2592 * and acquires i_mutex as needed.
2594 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2595 unsigned long nr_segs
, loff_t pos
)
2597 struct file
*file
= iocb
->ki_filp
;
2598 struct inode
*inode
= file
->f_mapping
->host
;
2601 BUG_ON(iocb
->ki_pos
!= pos
);
2603 mutex_lock(&inode
->i_mutex
);
2604 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
, &iocb
->ki_pos
);
2605 mutex_unlock(&inode
->i_mutex
);
2607 if (ret
> 0 || ret
== -EIOCBQUEUED
) {
2610 err
= generic_write_sync(file
, pos
, ret
);
2611 if (err
< 0 && ret
> 0)
2616 EXPORT_SYMBOL(generic_file_aio_write
);
2619 * try_to_release_page() - release old fs-specific metadata on a page
2621 * @page: the page which the kernel is trying to free
2622 * @gfp_mask: memory allocation flags (and I/O mode)
2624 * The address_space is to try to release any data against the page
2625 * (presumably at page->private). If the release was successful, return `1'.
2626 * Otherwise return zero.
2628 * This may also be called if PG_fscache is set on a page, indicating that the
2629 * page is known to the local caching routines.
2631 * The @gfp_mask argument specifies whether I/O may be performed to release
2632 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2635 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2637 struct address_space
* const mapping
= page
->mapping
;
2639 BUG_ON(!PageLocked(page
));
2640 if (PageWriteback(page
))
2643 if (mapping
&& mapping
->a_ops
->releasepage
)
2644 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2645 return try_to_free_buffers(page
);
2648 EXPORT_SYMBOL(try_to_release_page
);