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/export.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/cpuset.h>
33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
34 #include <linux/memcontrol.h>
35 #include <linux/cleancache.h>
36 #include <linux/rmap.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/filemap.h>
43 * FIXME: remove all knowledge of the buffer layer from the core VM
45 #include <linux/buffer_head.h> /* for try_to_free_buffers */
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * Shared mappings now work. 15.8.1995 Bruno.
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
64 * ->i_mmap_mutex (truncate_pagecache)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
70 * ->i_mmap_mutex (truncate->unmap_mapping_range)
74 * ->page_table_lock or pte_lock (various, mainly in memory.c)
75 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
78 * ->lock_page (access_process_vm)
80 * ->i_mutex (generic_perform_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
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 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
104 * ->inode->i_lock (zap_pte_range->set_page_dirty)
105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 static void page_cache_tree_delete(struct address_space
*mapping
,
112 struct page
*page
, void *shadow
)
114 struct radix_tree_node
*node
;
120 VM_BUG_ON(!PageLocked(page
));
122 __radix_tree_lookup(&mapping
->page_tree
, page
->index
, &node
, &slot
);
125 mapping
->nrshadows
++;
127 * Make sure the nrshadows update is committed before
128 * the nrpages update so that final truncate racing
129 * with reclaim does not see both counters 0 at the
130 * same time and miss a shadow entry.
137 /* Clear direct pointer tags in root node */
138 mapping
->page_tree
.gfp_mask
&= __GFP_BITS_MASK
;
139 radix_tree_replace_slot(slot
, shadow
);
143 /* Clear tree tags for the removed page */
145 offset
= index
& RADIX_TREE_MAP_MASK
;
146 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
147 if (test_bit(offset
, node
->tags
[tag
]))
148 radix_tree_tag_clear(&mapping
->page_tree
, index
, tag
);
151 /* Delete page, swap shadow entry */
152 radix_tree_replace_slot(slot
, shadow
);
153 workingset_node_pages_dec(node
);
155 workingset_node_shadows_inc(node
);
157 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
161 * Track node that only contains shadow entries.
163 * Avoid acquiring the list_lru lock if already tracked. The
164 * list_empty() test is safe as node->private_list is
165 * protected by mapping->tree_lock.
167 if (!workingset_node_pages(node
) &&
168 list_empty(&node
->private_list
)) {
169 node
->private_data
= mapping
;
170 list_lru_add(&workingset_shadow_nodes
, &node
->private_list
);
175 * Delete a page from the page cache and free it. Caller has to make
176 * sure the page is locked and that nobody else uses it - or that usage
177 * is safe. The caller must hold the mapping's tree_lock.
179 void __delete_from_page_cache(struct page
*page
, void *shadow
)
181 struct address_space
*mapping
= page
->mapping
;
183 trace_mm_filemap_delete_from_page_cache(page
);
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
189 if (PageUptodate(page
) && PageMappedToDisk(page
))
190 cleancache_put_page(page
);
192 cleancache_invalidate_page(mapping
, page
);
194 page_cache_tree_delete(mapping
, page
, shadow
);
196 page
->mapping
= NULL
;
197 /* Leave page->index set: truncation lookup relies upon it */
199 __dec_zone_page_state(page
, NR_FILE_PAGES
);
200 if (PageSwapBacked(page
))
201 __dec_zone_page_state(page
, NR_SHMEM
);
202 BUG_ON(page_mapped(page
));
205 * Some filesystems seem to re-dirty the page even after
206 * the VM has canceled the dirty bit (eg ext3 journaling).
208 * Fix it up by doing a final dirty accounting check after
209 * having removed the page entirely.
211 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
212 dec_zone_page_state(page
, NR_FILE_DIRTY
);
213 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
218 * delete_from_page_cache - delete page from page cache
219 * @page: the page which the kernel is trying to remove from page cache
221 * This must be called only on pages that have been verified to be in the page
222 * cache and locked. It will never put the page into the free list, the caller
223 * has a reference on the page.
225 void delete_from_page_cache(struct page
*page
)
227 struct address_space
*mapping
= page
->mapping
;
228 void (*freepage
)(struct page
*);
230 BUG_ON(!PageLocked(page
));
232 freepage
= mapping
->a_ops
->freepage
;
233 spin_lock_irq(&mapping
->tree_lock
);
234 __delete_from_page_cache(page
, NULL
);
235 spin_unlock_irq(&mapping
->tree_lock
);
236 mem_cgroup_uncharge_cache_page(page
);
240 page_cache_release(page
);
242 EXPORT_SYMBOL(delete_from_page_cache
);
244 static int filemap_check_errors(struct address_space
*mapping
)
247 /* Check for outstanding write errors */
248 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
249 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
251 if (test_bit(AS_EIO
, &mapping
->flags
) &&
252 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
258 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
259 * @mapping: address space structure to write
260 * @start: offset in bytes where the range starts
261 * @end: offset in bytes where the range ends (inclusive)
262 * @sync_mode: enable synchronous operation
264 * Start writeback against all of a mapping's dirty pages that lie
265 * within the byte offsets <start, end> inclusive.
267 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
268 * opposed to a regular memory cleansing writeback. The difference between
269 * these two operations is that if a dirty page/buffer is encountered, it must
270 * be waited upon, and not just skipped over.
272 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
273 loff_t end
, int sync_mode
)
276 struct writeback_control wbc
= {
277 .sync_mode
= sync_mode
,
278 .nr_to_write
= LONG_MAX
,
279 .range_start
= start
,
283 if (!mapping_cap_writeback_dirty(mapping
))
286 ret
= do_writepages(mapping
, &wbc
);
290 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
293 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
296 int filemap_fdatawrite(struct address_space
*mapping
)
298 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
300 EXPORT_SYMBOL(filemap_fdatawrite
);
302 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
305 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
307 EXPORT_SYMBOL(filemap_fdatawrite_range
);
310 * filemap_flush - mostly a non-blocking flush
311 * @mapping: target address_space
313 * This is a mostly non-blocking flush. Not suitable for data-integrity
314 * purposes - I/O may not be started against all dirty pages.
316 int filemap_flush(struct address_space
*mapping
)
318 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
320 EXPORT_SYMBOL(filemap_flush
);
323 * filemap_fdatawait_range - wait for writeback to complete
324 * @mapping: address space structure to wait for
325 * @start_byte: offset in bytes where the range starts
326 * @end_byte: offset in bytes where the range ends (inclusive)
328 * Walk the list of under-writeback pages of the given address space
329 * in the given range and wait for all of them.
331 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
334 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
335 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
340 if (end_byte
< start_byte
)
343 pagevec_init(&pvec
, 0);
344 while ((index
<= end
) &&
345 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
346 PAGECACHE_TAG_WRITEBACK
,
347 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
350 for (i
= 0; i
< nr_pages
; i
++) {
351 struct page
*page
= pvec
.pages
[i
];
353 /* until radix tree lookup accepts end_index */
354 if (page
->index
> end
)
357 wait_on_page_writeback(page
);
358 if (TestClearPageError(page
))
361 pagevec_release(&pvec
);
365 ret2
= filemap_check_errors(mapping
);
371 EXPORT_SYMBOL(filemap_fdatawait_range
);
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 filemap_fdatawait_range(mapping
, 0, i_size
- 1);
389 EXPORT_SYMBOL(filemap_fdatawait
);
391 int filemap_write_and_wait(struct address_space
*mapping
)
395 if (mapping
->nrpages
) {
396 err
= filemap_fdatawrite(mapping
);
398 * Even if the above returned error, the pages may be
399 * written partially (e.g. -ENOSPC), so we wait for it.
400 * But the -EIO is special case, it may indicate the worst
401 * thing (e.g. bug) happened, so we avoid waiting for it.
404 int err2
= filemap_fdatawait(mapping
);
409 err
= filemap_check_errors(mapping
);
413 EXPORT_SYMBOL(filemap_write_and_wait
);
416 * filemap_write_and_wait_range - write out & wait on a file range
417 * @mapping: the address_space for the pages
418 * @lstart: offset in bytes where the range starts
419 * @lend: offset in bytes where the range ends (inclusive)
421 * Write out and wait upon file offsets lstart->lend, inclusive.
423 * Note that `lend' is inclusive (describes the last byte to be written) so
424 * that this function can be used to write to the very end-of-file (end = -1).
426 int filemap_write_and_wait_range(struct address_space
*mapping
,
427 loff_t lstart
, loff_t lend
)
431 if (mapping
->nrpages
) {
432 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
434 /* See comment of filemap_write_and_wait() */
436 int err2
= filemap_fdatawait_range(mapping
,
442 err
= filemap_check_errors(mapping
);
446 EXPORT_SYMBOL(filemap_write_and_wait_range
);
449 * replace_page_cache_page - replace a pagecache page with a new one
450 * @old: page to be replaced
451 * @new: page to replace with
452 * @gfp_mask: allocation mode
454 * This function replaces a page in the pagecache with a new one. On
455 * success it acquires the pagecache reference for the new page and
456 * drops it for the old page. Both the old and new pages must be
457 * locked. This function does not add the new page to the LRU, the
458 * caller must do that.
460 * The remove + add is atomic. The only way this function can fail is
461 * memory allocation failure.
463 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
467 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
468 VM_BUG_ON_PAGE(!PageLocked(new), new);
469 VM_BUG_ON_PAGE(new->mapping
, new);
471 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
473 struct address_space
*mapping
= old
->mapping
;
474 void (*freepage
)(struct page
*);
476 pgoff_t offset
= old
->index
;
477 freepage
= mapping
->a_ops
->freepage
;
480 new->mapping
= mapping
;
483 spin_lock_irq(&mapping
->tree_lock
);
484 __delete_from_page_cache(old
, NULL
);
485 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
488 __inc_zone_page_state(new, NR_FILE_PAGES
);
489 if (PageSwapBacked(new))
490 __inc_zone_page_state(new, NR_SHMEM
);
491 spin_unlock_irq(&mapping
->tree_lock
);
492 /* mem_cgroup codes must not be called under tree_lock */
493 mem_cgroup_replace_page_cache(old
, new);
494 radix_tree_preload_end();
497 page_cache_release(old
);
502 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
504 static int page_cache_tree_insert(struct address_space
*mapping
,
505 struct page
*page
, void **shadowp
)
507 struct radix_tree_node
*node
;
511 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
,
518 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
519 if (!radix_tree_exceptional_entry(p
))
523 mapping
->nrshadows
--;
525 workingset_node_shadows_dec(node
);
527 radix_tree_replace_slot(slot
, page
);
530 workingset_node_pages_inc(node
);
532 * Don't track node that contains actual pages.
534 * Avoid acquiring the list_lru lock if already
535 * untracked. The list_empty() test is safe as
536 * node->private_list is protected by
537 * mapping->tree_lock.
539 if (!list_empty(&node
->private_list
))
540 list_lru_del(&workingset_shadow_nodes
,
541 &node
->private_list
);
546 static int __add_to_page_cache_locked(struct page
*page
,
547 struct address_space
*mapping
,
548 pgoff_t offset
, gfp_t gfp_mask
,
553 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
554 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
556 error
= mem_cgroup_charge_file(page
, current
->mm
,
557 gfp_mask
& GFP_RECLAIM_MASK
);
561 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
563 mem_cgroup_uncharge_cache_page(page
);
567 page_cache_get(page
);
568 page
->mapping
= mapping
;
569 page
->index
= offset
;
571 spin_lock_irq(&mapping
->tree_lock
);
572 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
573 radix_tree_preload_end();
576 __inc_zone_page_state(page
, NR_FILE_PAGES
);
577 spin_unlock_irq(&mapping
->tree_lock
);
578 trace_mm_filemap_add_to_page_cache(page
);
581 page
->mapping
= NULL
;
582 /* Leave page->index set: truncation relies upon it */
583 spin_unlock_irq(&mapping
->tree_lock
);
584 mem_cgroup_uncharge_cache_page(page
);
585 page_cache_release(page
);
590 * add_to_page_cache_locked - add a locked page to the pagecache
592 * @mapping: the page's address_space
593 * @offset: page index
594 * @gfp_mask: page allocation mode
596 * This function is used to add a page to the pagecache. It must be locked.
597 * This function does not add the page to the LRU. The caller must do that.
599 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
600 pgoff_t offset
, gfp_t gfp_mask
)
602 return __add_to_page_cache_locked(page
, mapping
, offset
,
605 EXPORT_SYMBOL(add_to_page_cache_locked
);
607 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
608 pgoff_t offset
, gfp_t gfp_mask
)
613 __set_page_locked(page
);
614 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
617 __clear_page_locked(page
);
620 * The page might have been evicted from cache only
621 * recently, in which case it should be activated like
622 * any other repeatedly accessed page.
624 if (shadow
&& workingset_refault(shadow
)) {
626 workingset_activation(page
);
628 ClearPageActive(page
);
633 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
636 struct page
*__page_cache_alloc(gfp_t gfp
)
641 if (cpuset_do_page_mem_spread()) {
642 unsigned int cpuset_mems_cookie
;
644 cpuset_mems_cookie
= read_mems_allowed_begin();
645 n
= cpuset_mem_spread_node();
646 page
= alloc_pages_exact_node(n
, gfp
, 0);
647 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
651 return alloc_pages(gfp
, 0);
653 EXPORT_SYMBOL(__page_cache_alloc
);
657 * In order to wait for pages to become available there must be
658 * waitqueues associated with pages. By using a hash table of
659 * waitqueues where the bucket discipline is to maintain all
660 * waiters on the same queue and wake all when any of the pages
661 * become available, and for the woken contexts to check to be
662 * sure the appropriate page became available, this saves space
663 * at a cost of "thundering herd" phenomena during rare hash
666 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
668 const struct zone
*zone
= page_zone(page
);
670 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
673 static inline void wake_up_page(struct page
*page
, int bit
)
675 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
678 void wait_on_page_bit(struct page
*page
, int bit_nr
)
680 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
682 if (test_bit(bit_nr
, &page
->flags
))
683 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
684 TASK_UNINTERRUPTIBLE
);
686 EXPORT_SYMBOL(wait_on_page_bit
);
688 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
690 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
692 if (!test_bit(bit_nr
, &page
->flags
))
695 return __wait_on_bit(page_waitqueue(page
), &wait
,
696 bit_wait_io
, TASK_KILLABLE
);
700 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
701 * @page: Page defining the wait queue of interest
702 * @waiter: Waiter to add to the queue
704 * Add an arbitrary @waiter to the wait queue for the nominated @page.
706 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
708 wait_queue_head_t
*q
= page_waitqueue(page
);
711 spin_lock_irqsave(&q
->lock
, flags
);
712 __add_wait_queue(q
, waiter
);
713 spin_unlock_irqrestore(&q
->lock
, flags
);
715 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
718 * unlock_page - unlock a locked page
721 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
722 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
723 * mechananism between PageLocked pages and PageWriteback pages is shared.
724 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
726 * The mb is necessary to enforce ordering between the clear_bit and the read
727 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
729 void unlock_page(struct page
*page
)
731 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
732 clear_bit_unlock(PG_locked
, &page
->flags
);
733 smp_mb__after_atomic();
734 wake_up_page(page
, PG_locked
);
736 EXPORT_SYMBOL(unlock_page
);
739 * end_page_writeback - end writeback against a page
742 void end_page_writeback(struct page
*page
)
745 * TestClearPageReclaim could be used here but it is an atomic
746 * operation and overkill in this particular case. Failing to
747 * shuffle a page marked for immediate reclaim is too mild to
748 * justify taking an atomic operation penalty at the end of
749 * ever page writeback.
751 if (PageReclaim(page
)) {
752 ClearPageReclaim(page
);
753 rotate_reclaimable_page(page
);
756 if (!test_clear_page_writeback(page
))
759 smp_mb__after_atomic();
760 wake_up_page(page
, PG_writeback
);
762 EXPORT_SYMBOL(end_page_writeback
);
765 * After completing I/O on a page, call this routine to update the page
766 * flags appropriately
768 void page_endio(struct page
*page
, int rw
, int err
)
772 SetPageUptodate(page
);
774 ClearPageUptodate(page
);
778 } else { /* rw == WRITE */
782 mapping_set_error(page
->mapping
, err
);
784 end_page_writeback(page
);
787 EXPORT_SYMBOL_GPL(page_endio
);
790 * __lock_page - get a lock on the page, assuming we need to sleep to get it
791 * @page: the page to lock
793 void __lock_page(struct page
*page
)
795 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
797 __wait_on_bit_lock(page_waitqueue(page
), &wait
, bit_wait_io
,
798 TASK_UNINTERRUPTIBLE
);
800 EXPORT_SYMBOL(__lock_page
);
802 int __lock_page_killable(struct page
*page
)
804 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
806 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
807 bit_wait_io
, TASK_KILLABLE
);
809 EXPORT_SYMBOL_GPL(__lock_page_killable
);
811 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
814 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
816 * CAUTION! In this case, mmap_sem is not released
817 * even though return 0.
819 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
822 up_read(&mm
->mmap_sem
);
823 if (flags
& FAULT_FLAG_KILLABLE
)
824 wait_on_page_locked_killable(page
);
826 wait_on_page_locked(page
);
829 if (flags
& FAULT_FLAG_KILLABLE
) {
832 ret
= __lock_page_killable(page
);
834 up_read(&mm
->mmap_sem
);
844 * page_cache_next_hole - find the next hole (not-present entry)
847 * @max_scan: maximum range to search
849 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
850 * lowest indexed hole.
852 * Returns: the index of the hole if found, otherwise returns an index
853 * outside of the set specified (in which case 'return - index >=
854 * max_scan' will be true). In rare cases of index wrap-around, 0 will
857 * page_cache_next_hole may be called under rcu_read_lock. However,
858 * like radix_tree_gang_lookup, this will not atomically search a
859 * snapshot of the tree at a single point in time. For example, if a
860 * hole is created at index 5, then subsequently a hole is created at
861 * index 10, page_cache_next_hole covering both indexes may return 10
862 * if called under rcu_read_lock.
864 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
865 pgoff_t index
, unsigned long max_scan
)
869 for (i
= 0; i
< max_scan
; i
++) {
872 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
873 if (!page
|| radix_tree_exceptional_entry(page
))
882 EXPORT_SYMBOL(page_cache_next_hole
);
885 * page_cache_prev_hole - find the prev hole (not-present entry)
888 * @max_scan: maximum range to search
890 * Search backwards in the range [max(index-max_scan+1, 0), index] for
893 * Returns: the index of the hole if found, otherwise returns an index
894 * outside of the set specified (in which case 'index - return >=
895 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
898 * page_cache_prev_hole may be called under rcu_read_lock. However,
899 * like radix_tree_gang_lookup, this will not atomically search a
900 * snapshot of the tree at a single point in time. For example, if a
901 * hole is created at index 10, then subsequently a hole is created at
902 * index 5, page_cache_prev_hole covering both indexes may return 5 if
903 * called under rcu_read_lock.
905 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
906 pgoff_t index
, unsigned long max_scan
)
910 for (i
= 0; i
< max_scan
; i
++) {
913 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
914 if (!page
|| radix_tree_exceptional_entry(page
))
917 if (index
== ULONG_MAX
)
923 EXPORT_SYMBOL(page_cache_prev_hole
);
926 * find_get_entry - find and get a page cache entry
927 * @mapping: the address_space to search
928 * @offset: the page cache index
930 * Looks up the page cache slot at @mapping & @offset. If there is a
931 * page cache page, it is returned with an increased refcount.
933 * If the slot holds a shadow entry of a previously evicted page, or a
934 * swap entry from shmem/tmpfs, it is returned.
936 * Otherwise, %NULL is returned.
938 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
946 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
948 page
= radix_tree_deref_slot(pagep
);
951 if (radix_tree_exception(page
)) {
952 if (radix_tree_deref_retry(page
))
955 * A shadow entry of a recently evicted page,
956 * or a swap entry from shmem/tmpfs. Return
957 * it without attempting to raise page count.
961 if (!page_cache_get_speculative(page
))
965 * Has the page moved?
966 * This is part of the lockless pagecache protocol. See
967 * include/linux/pagemap.h for details.
969 if (unlikely(page
!= *pagep
)) {
970 page_cache_release(page
);
979 EXPORT_SYMBOL(find_get_entry
);
982 * find_lock_entry - locate, pin and lock a page cache entry
983 * @mapping: the address_space to search
984 * @offset: the page cache index
986 * Looks up the page cache slot at @mapping & @offset. If there is a
987 * page cache page, it is returned locked and with an increased
990 * If the slot holds a shadow entry of a previously evicted page, or a
991 * swap entry from shmem/tmpfs, it is returned.
993 * Otherwise, %NULL is returned.
995 * find_lock_entry() may sleep.
997 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1002 page
= find_get_entry(mapping
, offset
);
1003 if (page
&& !radix_tree_exception(page
)) {
1005 /* Has the page been truncated? */
1006 if (unlikely(page
->mapping
!= mapping
)) {
1008 page_cache_release(page
);
1011 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1015 EXPORT_SYMBOL(find_lock_entry
);
1018 * pagecache_get_page - find and get a page reference
1019 * @mapping: the address_space to search
1020 * @offset: the page index
1021 * @fgp_flags: PCG flags
1022 * @cache_gfp_mask: gfp mask to use for the page cache data page allocation
1023 * @radix_gfp_mask: gfp mask to use for radix tree node allocation
1025 * Looks up the page cache slot at @mapping & @offset.
1027 * PCG flags modify how the page is returned.
1029 * FGP_ACCESSED: the page will be marked accessed
1030 * FGP_LOCK: Page is return locked
1031 * FGP_CREAT: If page is not present then a new page is allocated using
1032 * @cache_gfp_mask and added to the page cache and the VM's LRU
1033 * list. If radix tree nodes are allocated during page cache
1034 * insertion then @radix_gfp_mask is used. The page is returned
1035 * locked and with an increased refcount. Otherwise, %NULL is
1038 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1039 * if the GFP flags specified for FGP_CREAT are atomic.
1041 * If there is a page cache page, it is returned with an increased refcount.
1043 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1044 int fgp_flags
, gfp_t cache_gfp_mask
, gfp_t radix_gfp_mask
)
1049 page
= find_get_entry(mapping
, offset
);
1050 if (radix_tree_exceptional_entry(page
))
1055 if (fgp_flags
& FGP_LOCK
) {
1056 if (fgp_flags
& FGP_NOWAIT
) {
1057 if (!trylock_page(page
)) {
1058 page_cache_release(page
);
1065 /* Has the page been truncated? */
1066 if (unlikely(page
->mapping
!= mapping
)) {
1068 page_cache_release(page
);
1071 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1074 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1075 mark_page_accessed(page
);
1078 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1080 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1081 cache_gfp_mask
|= __GFP_WRITE
;
1082 if (fgp_flags
& FGP_NOFS
) {
1083 cache_gfp_mask
&= ~__GFP_FS
;
1084 radix_gfp_mask
&= ~__GFP_FS
;
1087 page
= __page_cache_alloc(cache_gfp_mask
);
1091 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1092 fgp_flags
|= FGP_LOCK
;
1094 /* Init accessed so avoit atomic mark_page_accessed later */
1095 if (fgp_flags
& FGP_ACCESSED
)
1096 init_page_accessed(page
);
1098 err
= add_to_page_cache_lru(page
, mapping
, offset
, radix_gfp_mask
);
1099 if (unlikely(err
)) {
1100 page_cache_release(page
);
1109 EXPORT_SYMBOL(pagecache_get_page
);
1112 * find_get_entries - gang pagecache lookup
1113 * @mapping: The address_space to search
1114 * @start: The starting page cache index
1115 * @nr_entries: The maximum number of entries
1116 * @entries: Where the resulting entries are placed
1117 * @indices: The cache indices corresponding to the entries in @entries
1119 * find_get_entries() will search for and return a group of up to
1120 * @nr_entries entries in the mapping. The entries are placed at
1121 * @entries. find_get_entries() takes a reference against any actual
1124 * The search returns a group of mapping-contiguous page cache entries
1125 * with ascending indexes. There may be holes in the indices due to
1126 * not-present pages.
1128 * Any shadow entries of evicted pages, or swap entries from
1129 * shmem/tmpfs, are included in the returned array.
1131 * find_get_entries() returns the number of pages and shadow entries
1134 unsigned find_get_entries(struct address_space
*mapping
,
1135 pgoff_t start
, unsigned int nr_entries
,
1136 struct page
**entries
, pgoff_t
*indices
)
1139 unsigned int ret
= 0;
1140 struct radix_tree_iter iter
;
1147 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1150 page
= radix_tree_deref_slot(slot
);
1151 if (unlikely(!page
))
1153 if (radix_tree_exception(page
)) {
1154 if (radix_tree_deref_retry(page
))
1157 * A shadow entry of a recently evicted page,
1158 * or a swap entry from shmem/tmpfs. Return
1159 * it without attempting to raise page count.
1163 if (!page_cache_get_speculative(page
))
1166 /* Has the page moved? */
1167 if (unlikely(page
!= *slot
)) {
1168 page_cache_release(page
);
1172 indices
[ret
] = iter
.index
;
1173 entries
[ret
] = page
;
1174 if (++ret
== nr_entries
)
1182 * find_get_pages - gang pagecache lookup
1183 * @mapping: The address_space to search
1184 * @start: The starting page index
1185 * @nr_pages: The maximum number of pages
1186 * @pages: Where the resulting pages are placed
1188 * find_get_pages() will search for and return a group of up to
1189 * @nr_pages pages in the mapping. The pages are placed at @pages.
1190 * find_get_pages() takes a reference against the returned pages.
1192 * The search returns a group of mapping-contiguous pages with ascending
1193 * indexes. There may be holes in the indices due to not-present pages.
1195 * find_get_pages() returns the number of pages which were found.
1197 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1198 unsigned int nr_pages
, struct page
**pages
)
1200 struct radix_tree_iter iter
;
1204 if (unlikely(!nr_pages
))
1209 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1212 page
= radix_tree_deref_slot(slot
);
1213 if (unlikely(!page
))
1216 if (radix_tree_exception(page
)) {
1217 if (radix_tree_deref_retry(page
)) {
1219 * Transient condition which can only trigger
1220 * when entry at index 0 moves out of or back
1221 * to root: none yet gotten, safe to restart.
1223 WARN_ON(iter
.index
);
1227 * A shadow entry of a recently evicted page,
1228 * or a swap entry from shmem/tmpfs. Skip
1234 if (!page_cache_get_speculative(page
))
1237 /* Has the page moved? */
1238 if (unlikely(page
!= *slot
)) {
1239 page_cache_release(page
);
1244 if (++ret
== nr_pages
)
1253 * find_get_pages_contig - gang contiguous pagecache lookup
1254 * @mapping: The address_space to search
1255 * @index: The starting page index
1256 * @nr_pages: The maximum number of pages
1257 * @pages: Where the resulting pages are placed
1259 * find_get_pages_contig() works exactly like find_get_pages(), except
1260 * that the returned number of pages are guaranteed to be contiguous.
1262 * find_get_pages_contig() returns the number of pages which were found.
1264 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1265 unsigned int nr_pages
, struct page
**pages
)
1267 struct radix_tree_iter iter
;
1269 unsigned int ret
= 0;
1271 if (unlikely(!nr_pages
))
1276 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1279 page
= radix_tree_deref_slot(slot
);
1280 /* The hole, there no reason to continue */
1281 if (unlikely(!page
))
1284 if (radix_tree_exception(page
)) {
1285 if (radix_tree_deref_retry(page
)) {
1287 * Transient condition which can only trigger
1288 * when entry at index 0 moves out of or back
1289 * to root: none yet gotten, safe to restart.
1294 * A shadow entry of a recently evicted page,
1295 * or a swap entry from shmem/tmpfs. Stop
1296 * looking for contiguous pages.
1301 if (!page_cache_get_speculative(page
))
1304 /* Has the page moved? */
1305 if (unlikely(page
!= *slot
)) {
1306 page_cache_release(page
);
1311 * must check mapping and index after taking the ref.
1312 * otherwise we can get both false positives and false
1313 * negatives, which is just confusing to the caller.
1315 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1316 page_cache_release(page
);
1321 if (++ret
== nr_pages
)
1327 EXPORT_SYMBOL(find_get_pages_contig
);
1330 * find_get_pages_tag - find and return pages that match @tag
1331 * @mapping: the address_space to search
1332 * @index: the starting page index
1333 * @tag: the tag index
1334 * @nr_pages: the maximum number of pages
1335 * @pages: where the resulting pages are placed
1337 * Like find_get_pages, except we only return pages which are tagged with
1338 * @tag. We update @index to index the next page for the traversal.
1340 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1341 int tag
, unsigned int nr_pages
, struct page
**pages
)
1343 struct radix_tree_iter iter
;
1347 if (unlikely(!nr_pages
))
1352 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1353 &iter
, *index
, tag
) {
1356 page
= radix_tree_deref_slot(slot
);
1357 if (unlikely(!page
))
1360 if (radix_tree_exception(page
)) {
1361 if (radix_tree_deref_retry(page
)) {
1363 * Transient condition which can only trigger
1364 * when entry at index 0 moves out of or back
1365 * to root: none yet gotten, safe to restart.
1370 * A shadow entry of a recently evicted page.
1372 * Those entries should never be tagged, but
1373 * this tree walk is lockless and the tags are
1374 * looked up in bulk, one radix tree node at a
1375 * time, so there is a sizable window for page
1376 * reclaim to evict a page we saw tagged.
1383 if (!page_cache_get_speculative(page
))
1386 /* Has the page moved? */
1387 if (unlikely(page
!= *slot
)) {
1388 page_cache_release(page
);
1393 if (++ret
== nr_pages
)
1400 *index
= pages
[ret
- 1]->index
+ 1;
1404 EXPORT_SYMBOL(find_get_pages_tag
);
1407 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1408 * a _large_ part of the i/o request. Imagine the worst scenario:
1410 * ---R__________________________________________B__________
1411 * ^ reading here ^ bad block(assume 4k)
1413 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1414 * => failing the whole request => read(R) => read(R+1) =>
1415 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1416 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1417 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1419 * It is going insane. Fix it by quickly scaling down the readahead size.
1421 static void shrink_readahead_size_eio(struct file
*filp
,
1422 struct file_ra_state
*ra
)
1428 * do_generic_file_read - generic file read routine
1429 * @filp: the file to read
1430 * @ppos: current file position
1431 * @iter: data destination
1432 * @written: already copied
1434 * This is a generic file read routine, and uses the
1435 * mapping->a_ops->readpage() function for the actual low-level stuff.
1437 * This is really ugly. But the goto's actually try to clarify some
1438 * of the logic when it comes to error handling etc.
1440 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1441 struct iov_iter
*iter
, ssize_t written
)
1443 struct address_space
*mapping
= filp
->f_mapping
;
1444 struct inode
*inode
= mapping
->host
;
1445 struct file_ra_state
*ra
= &filp
->f_ra
;
1449 unsigned long offset
; /* offset into pagecache page */
1450 unsigned int prev_offset
;
1453 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1454 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1455 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1456 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1457 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1463 unsigned long nr
, ret
;
1467 page
= find_get_page(mapping
, index
);
1469 page_cache_sync_readahead(mapping
,
1471 index
, last_index
- index
);
1472 page
= find_get_page(mapping
, index
);
1473 if (unlikely(page
== NULL
))
1474 goto no_cached_page
;
1476 if (PageReadahead(page
)) {
1477 page_cache_async_readahead(mapping
,
1479 index
, last_index
- index
);
1481 if (!PageUptodate(page
)) {
1482 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1483 !mapping
->a_ops
->is_partially_uptodate
)
1484 goto page_not_up_to_date
;
1485 if (!trylock_page(page
))
1486 goto page_not_up_to_date
;
1487 /* Did it get truncated before we got the lock? */
1489 goto page_not_up_to_date_locked
;
1490 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1491 offset
, iter
->count
))
1492 goto page_not_up_to_date_locked
;
1497 * i_size must be checked after we know the page is Uptodate.
1499 * Checking i_size after the check allows us to calculate
1500 * the correct value for "nr", which means the zero-filled
1501 * part of the page is not copied back to userspace (unless
1502 * another truncate extends the file - this is desired though).
1505 isize
= i_size_read(inode
);
1506 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1507 if (unlikely(!isize
|| index
> end_index
)) {
1508 page_cache_release(page
);
1512 /* nr is the maximum number of bytes to copy from this page */
1513 nr
= PAGE_CACHE_SIZE
;
1514 if (index
== end_index
) {
1515 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1517 page_cache_release(page
);
1523 /* If users can be writing to this page using arbitrary
1524 * virtual addresses, take care about potential aliasing
1525 * before reading the page on the kernel side.
1527 if (mapping_writably_mapped(mapping
))
1528 flush_dcache_page(page
);
1531 * When a sequential read accesses a page several times,
1532 * only mark it as accessed the first time.
1534 if (prev_index
!= index
|| offset
!= prev_offset
)
1535 mark_page_accessed(page
);
1539 * Ok, we have the page, and it's up-to-date, so
1540 * now we can copy it to user space...
1543 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1545 index
+= offset
>> PAGE_CACHE_SHIFT
;
1546 offset
&= ~PAGE_CACHE_MASK
;
1547 prev_offset
= offset
;
1549 page_cache_release(page
);
1551 if (!iov_iter_count(iter
))
1559 page_not_up_to_date
:
1560 /* Get exclusive access to the page ... */
1561 error
= lock_page_killable(page
);
1562 if (unlikely(error
))
1563 goto readpage_error
;
1565 page_not_up_to_date_locked
:
1566 /* Did it get truncated before we got the lock? */
1567 if (!page
->mapping
) {
1569 page_cache_release(page
);
1573 /* Did somebody else fill it already? */
1574 if (PageUptodate(page
)) {
1581 * A previous I/O error may have been due to temporary
1582 * failures, eg. multipath errors.
1583 * PG_error will be set again if readpage fails.
1585 ClearPageError(page
);
1586 /* Start the actual read. The read will unlock the page. */
1587 error
= mapping
->a_ops
->readpage(filp
, page
);
1589 if (unlikely(error
)) {
1590 if (error
== AOP_TRUNCATED_PAGE
) {
1591 page_cache_release(page
);
1595 goto readpage_error
;
1598 if (!PageUptodate(page
)) {
1599 error
= lock_page_killable(page
);
1600 if (unlikely(error
))
1601 goto readpage_error
;
1602 if (!PageUptodate(page
)) {
1603 if (page
->mapping
== NULL
) {
1605 * invalidate_mapping_pages got it
1608 page_cache_release(page
);
1612 shrink_readahead_size_eio(filp
, ra
);
1614 goto readpage_error
;
1622 /* UHHUH! A synchronous read error occurred. Report it */
1623 page_cache_release(page
);
1628 * Ok, it wasn't cached, so we need to create a new
1631 page
= page_cache_alloc_cold(mapping
);
1636 error
= add_to_page_cache_lru(page
, mapping
,
1639 page_cache_release(page
);
1640 if (error
== -EEXIST
) {
1650 ra
->prev_pos
= prev_index
;
1651 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1652 ra
->prev_pos
|= prev_offset
;
1654 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1655 file_accessed(filp
);
1656 return written
? written
: error
;
1660 * generic_file_read_iter - generic filesystem read routine
1661 * @iocb: kernel I/O control block
1662 * @iter: destination for the data read
1664 * This is the "read_iter()" routine for all filesystems
1665 * that can use the page cache directly.
1668 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1670 struct file
*file
= iocb
->ki_filp
;
1672 loff_t
*ppos
= &iocb
->ki_pos
;
1675 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1676 if (file
->f_flags
& O_DIRECT
) {
1677 struct address_space
*mapping
= file
->f_mapping
;
1678 struct inode
*inode
= mapping
->host
;
1679 size_t count
= iov_iter_count(iter
);
1683 goto out
; /* skip atime */
1684 size
= i_size_read(inode
);
1685 retval
= filemap_write_and_wait_range(mapping
, pos
,
1688 struct iov_iter data
= *iter
;
1689 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
, &data
, pos
);
1693 *ppos
= pos
+ retval
;
1694 iov_iter_advance(iter
, retval
);
1698 * Btrfs can have a short DIO read if we encounter
1699 * compressed extents, so if there was an error, or if
1700 * we've already read everything we wanted to, or if
1701 * there was a short read because we hit EOF, go ahead
1702 * and return. Otherwise fallthrough to buffered io for
1703 * the rest of the read.
1705 if (retval
< 0 || !iov_iter_count(iter
) || *ppos
>= size
) {
1706 file_accessed(file
);
1711 retval
= do_generic_file_read(file
, ppos
, iter
, retval
);
1715 EXPORT_SYMBOL(generic_file_read_iter
);
1719 * page_cache_read - adds requested page to the page cache if not already there
1720 * @file: file to read
1721 * @offset: page index
1723 * This adds the requested page to the page cache if it isn't already there,
1724 * and schedules an I/O to read in its contents from disk.
1726 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1728 struct address_space
*mapping
= file
->f_mapping
;
1733 page
= page_cache_alloc_cold(mapping
);
1737 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1739 ret
= mapping
->a_ops
->readpage(file
, page
);
1740 else if (ret
== -EEXIST
)
1741 ret
= 0; /* losing race to add is OK */
1743 page_cache_release(page
);
1745 } while (ret
== AOP_TRUNCATED_PAGE
);
1750 #define MMAP_LOTSAMISS (100)
1753 * Synchronous readahead happens when we don't even find
1754 * a page in the page cache at all.
1756 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1757 struct file_ra_state
*ra
,
1761 unsigned long ra_pages
;
1762 struct address_space
*mapping
= file
->f_mapping
;
1764 /* If we don't want any read-ahead, don't bother */
1765 if (vma
->vm_flags
& VM_RAND_READ
)
1770 if (vma
->vm_flags
& VM_SEQ_READ
) {
1771 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1776 /* Avoid banging the cache line if not needed */
1777 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1781 * Do we miss much more than hit in this file? If so,
1782 * stop bothering with read-ahead. It will only hurt.
1784 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1790 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1791 ra
->start
= max_t(long, 0, offset
- ra_pages
/ 2);
1792 ra
->size
= ra_pages
;
1793 ra
->async_size
= ra_pages
/ 4;
1794 ra_submit(ra
, mapping
, file
);
1798 * Asynchronous readahead happens when we find the page and PG_readahead,
1799 * so we want to possibly extend the readahead further..
1801 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1802 struct file_ra_state
*ra
,
1807 struct address_space
*mapping
= file
->f_mapping
;
1809 /* If we don't want any read-ahead, don't bother */
1810 if (vma
->vm_flags
& VM_RAND_READ
)
1812 if (ra
->mmap_miss
> 0)
1814 if (PageReadahead(page
))
1815 page_cache_async_readahead(mapping
, ra
, file
,
1816 page
, offset
, ra
->ra_pages
);
1820 * filemap_fault - read in file data for page fault handling
1821 * @vma: vma in which the fault was taken
1822 * @vmf: struct vm_fault containing details of the fault
1824 * filemap_fault() is invoked via the vma operations vector for a
1825 * mapped memory region to read in file data during a page fault.
1827 * The goto's are kind of ugly, but this streamlines the normal case of having
1828 * it in the page cache, and handles the special cases reasonably without
1829 * having a lot of duplicated code.
1831 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1834 struct file
*file
= vma
->vm_file
;
1835 struct address_space
*mapping
= file
->f_mapping
;
1836 struct file_ra_state
*ra
= &file
->f_ra
;
1837 struct inode
*inode
= mapping
->host
;
1838 pgoff_t offset
= vmf
->pgoff
;
1843 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1844 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
1845 return VM_FAULT_SIGBUS
;
1848 * Do we have something in the page cache already?
1850 page
= find_get_page(mapping
, offset
);
1851 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1853 * We found the page, so try async readahead before
1854 * waiting for the lock.
1856 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1858 /* No page in the page cache at all */
1859 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1860 count_vm_event(PGMAJFAULT
);
1861 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1862 ret
= VM_FAULT_MAJOR
;
1864 page
= find_get_page(mapping
, offset
);
1866 goto no_cached_page
;
1869 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1870 page_cache_release(page
);
1871 return ret
| VM_FAULT_RETRY
;
1874 /* Did it get truncated? */
1875 if (unlikely(page
->mapping
!= mapping
)) {
1880 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1883 * We have a locked page in the page cache, now we need to check
1884 * that it's up-to-date. If not, it is going to be due to an error.
1886 if (unlikely(!PageUptodate(page
)))
1887 goto page_not_uptodate
;
1890 * Found the page and have a reference on it.
1891 * We must recheck i_size under page lock.
1893 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1894 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
1896 page_cache_release(page
);
1897 return VM_FAULT_SIGBUS
;
1901 return ret
| VM_FAULT_LOCKED
;
1905 * We're only likely to ever get here if MADV_RANDOM is in
1908 error
= page_cache_read(file
, offset
);
1911 * The page we want has now been added to the page cache.
1912 * In the unlikely event that someone removed it in the
1913 * meantime, we'll just come back here and read it again.
1919 * An error return from page_cache_read can result if the
1920 * system is low on memory, or a problem occurs while trying
1923 if (error
== -ENOMEM
)
1924 return VM_FAULT_OOM
;
1925 return VM_FAULT_SIGBUS
;
1929 * Umm, take care of errors if the page isn't up-to-date.
1930 * Try to re-read it _once_. We do this synchronously,
1931 * because there really aren't any performance issues here
1932 * and we need to check for errors.
1934 ClearPageError(page
);
1935 error
= mapping
->a_ops
->readpage(file
, page
);
1937 wait_on_page_locked(page
);
1938 if (!PageUptodate(page
))
1941 page_cache_release(page
);
1943 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1946 /* Things didn't work out. Return zero to tell the mm layer so. */
1947 shrink_readahead_size_eio(file
, ra
);
1948 return VM_FAULT_SIGBUS
;
1950 EXPORT_SYMBOL(filemap_fault
);
1952 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1954 struct radix_tree_iter iter
;
1956 struct file
*file
= vma
->vm_file
;
1957 struct address_space
*mapping
= file
->f_mapping
;
1960 unsigned long address
= (unsigned long) vmf
->virtual_address
;
1965 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
1966 if (iter
.index
> vmf
->max_pgoff
)
1969 page
= radix_tree_deref_slot(slot
);
1970 if (unlikely(!page
))
1972 if (radix_tree_exception(page
)) {
1973 if (radix_tree_deref_retry(page
))
1979 if (!page_cache_get_speculative(page
))
1982 /* Has the page moved? */
1983 if (unlikely(page
!= *slot
)) {
1984 page_cache_release(page
);
1988 if (!PageUptodate(page
) ||
1989 PageReadahead(page
) ||
1992 if (!trylock_page(page
))
1995 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
1998 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
1999 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2002 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2003 if (!pte_none(*pte
))
2006 if (file
->f_ra
.mmap_miss
> 0)
2007 file
->f_ra
.mmap_miss
--;
2008 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2009 do_set_pte(vma
, addr
, page
, pte
, false, false);
2015 page_cache_release(page
);
2017 if (iter
.index
== vmf
->max_pgoff
)
2022 EXPORT_SYMBOL(filemap_map_pages
);
2024 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2026 struct page
*page
= vmf
->page
;
2027 struct inode
*inode
= file_inode(vma
->vm_file
);
2028 int ret
= VM_FAULT_LOCKED
;
2030 sb_start_pagefault(inode
->i_sb
);
2031 file_update_time(vma
->vm_file
);
2033 if (page
->mapping
!= inode
->i_mapping
) {
2035 ret
= VM_FAULT_NOPAGE
;
2039 * We mark the page dirty already here so that when freeze is in
2040 * progress, we are guaranteed that writeback during freezing will
2041 * see the dirty page and writeprotect it again.
2043 set_page_dirty(page
);
2044 wait_for_stable_page(page
);
2046 sb_end_pagefault(inode
->i_sb
);
2049 EXPORT_SYMBOL(filemap_page_mkwrite
);
2051 const struct vm_operations_struct generic_file_vm_ops
= {
2052 .fault
= filemap_fault
,
2053 .map_pages
= filemap_map_pages
,
2054 .page_mkwrite
= filemap_page_mkwrite
,
2055 .remap_pages
= generic_file_remap_pages
,
2058 /* This is used for a general mmap of a disk file */
2060 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2062 struct address_space
*mapping
= file
->f_mapping
;
2064 if (!mapping
->a_ops
->readpage
)
2066 file_accessed(file
);
2067 vma
->vm_ops
= &generic_file_vm_ops
;
2072 * This is for filesystems which do not implement ->writepage.
2074 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2076 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2078 return generic_file_mmap(file
, vma
);
2081 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2085 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2089 #endif /* CONFIG_MMU */
2091 EXPORT_SYMBOL(generic_file_mmap
);
2092 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2094 static struct page
*wait_on_page_read(struct page
*page
)
2096 if (!IS_ERR(page
)) {
2097 wait_on_page_locked(page
);
2098 if (!PageUptodate(page
)) {
2099 page_cache_release(page
);
2100 page
= ERR_PTR(-EIO
);
2106 static struct page
*__read_cache_page(struct address_space
*mapping
,
2108 int (*filler
)(void *, struct page
*),
2115 page
= find_get_page(mapping
, index
);
2117 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2119 return ERR_PTR(-ENOMEM
);
2120 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2121 if (unlikely(err
)) {
2122 page_cache_release(page
);
2125 /* Presumably ENOMEM for radix tree node */
2126 return ERR_PTR(err
);
2128 err
= filler(data
, page
);
2130 page_cache_release(page
);
2131 page
= ERR_PTR(err
);
2133 page
= wait_on_page_read(page
);
2139 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2141 int (*filler
)(void *, struct page
*),
2150 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2153 if (PageUptodate(page
))
2157 if (!page
->mapping
) {
2159 page_cache_release(page
);
2162 if (PageUptodate(page
)) {
2166 err
= filler(data
, page
);
2168 page_cache_release(page
);
2169 return ERR_PTR(err
);
2171 page
= wait_on_page_read(page
);
2176 mark_page_accessed(page
);
2181 * read_cache_page - read into page cache, fill it if needed
2182 * @mapping: the page's address_space
2183 * @index: the page index
2184 * @filler: function to perform the read
2185 * @data: first arg to filler(data, page) function, often left as NULL
2187 * Read into the page cache. If a page already exists, and PageUptodate() is
2188 * not set, try to fill the page and wait for it to become unlocked.
2190 * If the page does not get brought uptodate, return -EIO.
2192 struct page
*read_cache_page(struct address_space
*mapping
,
2194 int (*filler
)(void *, struct page
*),
2197 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2199 EXPORT_SYMBOL(read_cache_page
);
2202 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2203 * @mapping: the page's address_space
2204 * @index: the page index
2205 * @gfp: the page allocator flags to use if allocating
2207 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2208 * any new page allocations done using the specified allocation flags.
2210 * If the page does not get brought uptodate, return -EIO.
2212 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2216 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2218 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2220 EXPORT_SYMBOL(read_cache_page_gfp
);
2223 * Performs necessary checks before doing a write
2225 * Can adjust writing position or amount of bytes to write.
2226 * Returns appropriate error code that caller should return or
2227 * zero in case that write should be allowed.
2229 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
2231 struct inode
*inode
= file
->f_mapping
->host
;
2232 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2234 if (unlikely(*pos
< 0))
2238 /* FIXME: this is for backwards compatibility with 2.4 */
2239 if (file
->f_flags
& O_APPEND
)
2240 *pos
= i_size_read(inode
);
2242 if (limit
!= RLIM_INFINITY
) {
2243 if (*pos
>= limit
) {
2244 send_sig(SIGXFSZ
, current
, 0);
2247 if (*count
> limit
- (typeof(limit
))*pos
) {
2248 *count
= limit
- (typeof(limit
))*pos
;
2256 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2257 !(file
->f_flags
& O_LARGEFILE
))) {
2258 if (*pos
>= MAX_NON_LFS
) {
2261 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2262 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2267 * Are we about to exceed the fs block limit ?
2269 * If we have written data it becomes a short write. If we have
2270 * exceeded without writing data we send a signal and return EFBIG.
2271 * Linus frestrict idea will clean these up nicely..
2273 if (likely(!isblk
)) {
2274 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2275 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2278 /* zero-length writes at ->s_maxbytes are OK */
2281 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2282 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2286 if (bdev_read_only(I_BDEV(inode
)))
2288 isize
= i_size_read(inode
);
2289 if (*pos
>= isize
) {
2290 if (*count
|| *pos
> isize
)
2294 if (*pos
+ *count
> isize
)
2295 *count
= isize
- *pos
;
2302 EXPORT_SYMBOL(generic_write_checks
);
2304 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2305 loff_t pos
, unsigned len
, unsigned flags
,
2306 struct page
**pagep
, void **fsdata
)
2308 const struct address_space_operations
*aops
= mapping
->a_ops
;
2310 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2313 EXPORT_SYMBOL(pagecache_write_begin
);
2315 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2316 loff_t pos
, unsigned len
, unsigned copied
,
2317 struct page
*page
, void *fsdata
)
2319 const struct address_space_operations
*aops
= mapping
->a_ops
;
2321 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2323 EXPORT_SYMBOL(pagecache_write_end
);
2326 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
2328 struct file
*file
= iocb
->ki_filp
;
2329 struct address_space
*mapping
= file
->f_mapping
;
2330 struct inode
*inode
= mapping
->host
;
2334 struct iov_iter data
;
2336 write_len
= iov_iter_count(from
);
2337 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2339 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2344 * After a write we want buffered reads to be sure to go to disk to get
2345 * the new data. We invalidate clean cached page from the region we're
2346 * about to write. We do this *before* the write so that we can return
2347 * without clobbering -EIOCBQUEUED from ->direct_IO().
2349 if (mapping
->nrpages
) {
2350 written
= invalidate_inode_pages2_range(mapping
,
2351 pos
>> PAGE_CACHE_SHIFT
, end
);
2353 * If a page can not be invalidated, return 0 to fall back
2354 * to buffered write.
2357 if (written
== -EBUSY
)
2364 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, &data
, pos
);
2367 * Finally, try again to invalidate clean pages which might have been
2368 * cached by non-direct readahead, or faulted in by get_user_pages()
2369 * if the source of the write was an mmap'ed region of the file
2370 * we're writing. Either one is a pretty crazy thing to do,
2371 * so we don't support it 100%. If this invalidation
2372 * fails, tough, the write still worked...
2374 if (mapping
->nrpages
) {
2375 invalidate_inode_pages2_range(mapping
,
2376 pos
>> PAGE_CACHE_SHIFT
, end
);
2381 iov_iter_advance(from
, written
);
2382 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2383 i_size_write(inode
, pos
);
2384 mark_inode_dirty(inode
);
2391 EXPORT_SYMBOL(generic_file_direct_write
);
2394 * Find or create a page at the given pagecache position. Return the locked
2395 * page. This function is specifically for buffered writes.
2397 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2398 pgoff_t index
, unsigned flags
)
2401 int fgp_flags
= FGP_LOCK
|FGP_ACCESSED
|FGP_WRITE
|FGP_CREAT
;
2403 if (flags
& AOP_FLAG_NOFS
)
2404 fgp_flags
|= FGP_NOFS
;
2406 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2407 mapping_gfp_mask(mapping
),
2410 wait_for_stable_page(page
);
2414 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2416 ssize_t
generic_perform_write(struct file
*file
,
2417 struct iov_iter
*i
, loff_t pos
)
2419 struct address_space
*mapping
= file
->f_mapping
;
2420 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2422 ssize_t written
= 0;
2423 unsigned int flags
= 0;
2426 * Copies from kernel address space cannot fail (NFSD is a big user).
2428 if (segment_eq(get_fs(), KERNEL_DS
))
2429 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2433 unsigned long offset
; /* Offset into pagecache page */
2434 unsigned long bytes
; /* Bytes to write to page */
2435 size_t copied
; /* Bytes copied from user */
2438 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2439 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2444 * Bring in the user page that we will copy from _first_.
2445 * Otherwise there's a nasty deadlock on copying from the
2446 * same page as we're writing to, without it being marked
2449 * Not only is this an optimisation, but it is also required
2450 * to check that the address is actually valid, when atomic
2451 * usercopies are used, below.
2453 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2458 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2460 if (unlikely(status
< 0))
2463 if (mapping_writably_mapped(mapping
))
2464 flush_dcache_page(page
);
2466 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2467 flush_dcache_page(page
);
2469 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2471 if (unlikely(status
< 0))
2477 iov_iter_advance(i
, copied
);
2478 if (unlikely(copied
== 0)) {
2480 * If we were unable to copy any data at all, we must
2481 * fall back to a single segment length write.
2483 * If we didn't fallback here, we could livelock
2484 * because not all segments in the iov can be copied at
2485 * once without a pagefault.
2487 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2488 iov_iter_single_seg_count(i
));
2494 balance_dirty_pages_ratelimited(mapping
);
2495 if (fatal_signal_pending(current
)) {
2499 } while (iov_iter_count(i
));
2501 return written
? written
: status
;
2503 EXPORT_SYMBOL(generic_perform_write
);
2506 * __generic_file_write_iter - write data to a file
2507 * @iocb: IO state structure (file, offset, etc.)
2508 * @from: iov_iter with data to write
2510 * This function does all the work needed for actually writing data to a
2511 * file. It does all basic checks, removes SUID from the file, updates
2512 * modification times and calls proper subroutines depending on whether we
2513 * do direct IO or a standard buffered write.
2515 * It expects i_mutex to be grabbed unless we work on a block device or similar
2516 * object which does not need locking at all.
2518 * This function does *not* take care of syncing data in case of O_SYNC write.
2519 * A caller has to handle it. This is mainly due to the fact that we want to
2520 * avoid syncing under i_mutex.
2522 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2524 struct file
*file
= iocb
->ki_filp
;
2525 struct address_space
* mapping
= file
->f_mapping
;
2526 struct inode
*inode
= mapping
->host
;
2527 loff_t pos
= iocb
->ki_pos
;
2528 ssize_t written
= 0;
2531 size_t count
= iov_iter_count(from
);
2533 /* We can write back this queue in page reclaim */
2534 current
->backing_dev_info
= mapping
->backing_dev_info
;
2535 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2542 iov_iter_truncate(from
, count
);
2544 err
= file_remove_suid(file
);
2548 err
= file_update_time(file
);
2552 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2553 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2556 written
= generic_file_direct_write(iocb
, from
, pos
);
2557 if (written
< 0 || written
== count
)
2561 * direct-io write to a hole: fall through to buffered I/O
2562 * for completing the rest of the request.
2567 status
= generic_perform_write(file
, from
, pos
);
2569 * If generic_perform_write() returned a synchronous error
2570 * then we want to return the number of bytes which were
2571 * direct-written, or the error code if that was zero. Note
2572 * that this differs from normal direct-io semantics, which
2573 * will return -EFOO even if some bytes were written.
2575 if (unlikely(status
< 0) && !written
) {
2579 iocb
->ki_pos
= pos
+ status
;
2581 * We need to ensure that the page cache pages are written to
2582 * disk and invalidated to preserve the expected O_DIRECT
2585 endbyte
= pos
+ status
- 1;
2586 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
2589 invalidate_mapping_pages(mapping
,
2590 pos
>> PAGE_CACHE_SHIFT
,
2591 endbyte
>> PAGE_CACHE_SHIFT
);
2594 * We don't know how much we wrote, so just return
2595 * the number of bytes which were direct-written
2599 written
= generic_perform_write(file
, from
, pos
);
2600 if (likely(written
>= 0))
2601 iocb
->ki_pos
= pos
+ written
;
2604 current
->backing_dev_info
= NULL
;
2605 return written
? written
: err
;
2607 EXPORT_SYMBOL(__generic_file_write_iter
);
2610 * generic_file_write_iter - write data to a file
2611 * @iocb: IO state structure
2612 * @from: iov_iter with data to write
2614 * This is a wrapper around __generic_file_write_iter() to be used by most
2615 * filesystems. It takes care of syncing the file in case of O_SYNC file
2616 * and acquires i_mutex as needed.
2618 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2620 struct file
*file
= iocb
->ki_filp
;
2621 struct inode
*inode
= file
->f_mapping
->host
;
2624 mutex_lock(&inode
->i_mutex
);
2625 ret
= __generic_file_write_iter(iocb
, from
);
2626 mutex_unlock(&inode
->i_mutex
);
2631 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2637 EXPORT_SYMBOL(generic_file_write_iter
);
2640 * try_to_release_page() - release old fs-specific metadata on a page
2642 * @page: the page which the kernel is trying to free
2643 * @gfp_mask: memory allocation flags (and I/O mode)
2645 * The address_space is to try to release any data against the page
2646 * (presumably at page->private). If the release was successful, return `1'.
2647 * Otherwise return zero.
2649 * This may also be called if PG_fscache is set on a page, indicating that the
2650 * page is known to the local caching routines.
2652 * The @gfp_mask argument specifies whether I/O may be performed to release
2653 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2656 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2658 struct address_space
* const mapping
= page
->mapping
;
2660 BUG_ON(!PageLocked(page
));
2661 if (PageWriteback(page
))
2664 if (mapping
&& mapping
->a_ops
->releasepage
)
2665 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2666 return try_to_free_buffers(page
);
2669 EXPORT_SYMBOL(try_to_release_page
);