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 int page_cache_tree_insert(struct address_space
*mapping
,
112 struct page
*page
, void **shadowp
)
114 struct radix_tree_node
*node
;
118 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
,
125 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
126 if (!radix_tree_exceptional_entry(p
))
130 mapping
->nrshadows
--;
132 workingset_node_shadows_dec(node
);
134 radix_tree_replace_slot(slot
, page
);
137 workingset_node_pages_inc(node
);
139 * Don't track node that contains actual pages.
141 * Avoid acquiring the list_lru lock if already
142 * untracked. The list_empty() test is safe as
143 * node->private_list is protected by
144 * mapping->tree_lock.
146 if (!list_empty(&node
->private_list
))
147 list_lru_del(&workingset_shadow_nodes
,
148 &node
->private_list
);
153 static void page_cache_tree_delete(struct address_space
*mapping
,
154 struct page
*page
, void *shadow
)
156 struct radix_tree_node
*node
;
162 VM_BUG_ON(!PageLocked(page
));
164 __radix_tree_lookup(&mapping
->page_tree
, page
->index
, &node
, &slot
);
167 * We need a node to properly account shadow
168 * entries. Don't plant any without. XXX
174 mapping
->nrshadows
++;
176 * Make sure the nrshadows update is committed before
177 * the nrpages update so that final truncate racing
178 * with reclaim does not see both counters 0 at the
179 * same time and miss a shadow entry.
186 /* Clear direct pointer tags in root node */
187 mapping
->page_tree
.gfp_mask
&= __GFP_BITS_MASK
;
188 radix_tree_replace_slot(slot
, shadow
);
192 /* Clear tree tags for the removed page */
194 offset
= index
& RADIX_TREE_MAP_MASK
;
195 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
196 if (test_bit(offset
, node
->tags
[tag
]))
197 radix_tree_tag_clear(&mapping
->page_tree
, index
, tag
);
200 /* Delete page, swap shadow entry */
201 radix_tree_replace_slot(slot
, shadow
);
202 workingset_node_pages_dec(node
);
204 workingset_node_shadows_inc(node
);
206 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
210 * Track node that only contains shadow entries.
212 * Avoid acquiring the list_lru lock if already tracked. The
213 * list_empty() test is safe as node->private_list is
214 * protected by mapping->tree_lock.
216 if (!workingset_node_pages(node
) &&
217 list_empty(&node
->private_list
)) {
218 node
->private_data
= mapping
;
219 list_lru_add(&workingset_shadow_nodes
, &node
->private_list
);
224 * Delete a page from the page cache and free it. Caller has to make
225 * sure the page is locked and that nobody else uses it - or that usage
226 * is safe. The caller must hold the mapping's tree_lock.
228 void __delete_from_page_cache(struct page
*page
, void *shadow
)
230 struct address_space
*mapping
= page
->mapping
;
232 trace_mm_filemap_delete_from_page_cache(page
);
234 * if we're uptodate, flush out into the cleancache, otherwise
235 * invalidate any existing cleancache entries. We can't leave
236 * stale data around in the cleancache once our page is gone
238 if (PageUptodate(page
) && PageMappedToDisk(page
))
239 cleancache_put_page(page
);
241 cleancache_invalidate_page(mapping
, page
);
243 page_cache_tree_delete(mapping
, page
, shadow
);
245 page
->mapping
= NULL
;
246 /* Leave page->index set: truncation lookup relies upon it */
248 __dec_zone_page_state(page
, NR_FILE_PAGES
);
249 if (PageSwapBacked(page
))
250 __dec_zone_page_state(page
, NR_SHMEM
);
251 BUG_ON(page_mapped(page
));
254 * Some filesystems seem to re-dirty the page even after
255 * the VM has canceled the dirty bit (eg ext3 journaling).
257 * Fix it up by doing a final dirty accounting check after
258 * having removed the page entirely.
260 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
261 dec_zone_page_state(page
, NR_FILE_DIRTY
);
262 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
267 * delete_from_page_cache - delete page from page cache
268 * @page: the page which the kernel is trying to remove from page cache
270 * This must be called only on pages that have been verified to be in the page
271 * cache and locked. It will never put the page into the free list, the caller
272 * has a reference on the page.
274 void delete_from_page_cache(struct page
*page
)
276 struct address_space
*mapping
= page
->mapping
;
277 void (*freepage
)(struct page
*);
279 BUG_ON(!PageLocked(page
));
281 freepage
= mapping
->a_ops
->freepage
;
282 spin_lock_irq(&mapping
->tree_lock
);
283 __delete_from_page_cache(page
, NULL
);
284 spin_unlock_irq(&mapping
->tree_lock
);
285 mem_cgroup_uncharge_cache_page(page
);
289 page_cache_release(page
);
291 EXPORT_SYMBOL(delete_from_page_cache
);
293 static int sleep_on_page(void *word
)
299 static int sleep_on_page_killable(void *word
)
302 return fatal_signal_pending(current
) ? -EINTR
: 0;
305 static int filemap_check_errors(struct address_space
*mapping
)
308 /* Check for outstanding write errors */
309 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
310 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
312 if (test_bit(AS_EIO
, &mapping
->flags
) &&
313 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
319 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
320 * @mapping: address space structure to write
321 * @start: offset in bytes where the range starts
322 * @end: offset in bytes where the range ends (inclusive)
323 * @sync_mode: enable synchronous operation
325 * Start writeback against all of a mapping's dirty pages that lie
326 * within the byte offsets <start, end> inclusive.
328 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
329 * opposed to a regular memory cleansing writeback. The difference between
330 * these two operations is that if a dirty page/buffer is encountered, it must
331 * be waited upon, and not just skipped over.
333 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
334 loff_t end
, int sync_mode
)
337 struct writeback_control wbc
= {
338 .sync_mode
= sync_mode
,
339 .nr_to_write
= LONG_MAX
,
340 .range_start
= start
,
344 if (!mapping_cap_writeback_dirty(mapping
))
347 ret
= do_writepages(mapping
, &wbc
);
351 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
354 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
357 int filemap_fdatawrite(struct address_space
*mapping
)
359 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
361 EXPORT_SYMBOL(filemap_fdatawrite
);
363 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
366 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
368 EXPORT_SYMBOL(filemap_fdatawrite_range
);
371 * filemap_flush - mostly a non-blocking flush
372 * @mapping: target address_space
374 * This is a mostly non-blocking flush. Not suitable for data-integrity
375 * purposes - I/O may not be started against all dirty pages.
377 int filemap_flush(struct address_space
*mapping
)
379 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
381 EXPORT_SYMBOL(filemap_flush
);
384 * filemap_fdatawait_range - wait for writeback to complete
385 * @mapping: address space structure to wait for
386 * @start_byte: offset in bytes where the range starts
387 * @end_byte: offset in bytes where the range ends (inclusive)
389 * Walk the list of under-writeback pages of the given address space
390 * in the given range and wait for all of them.
392 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
395 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
396 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
401 if (end_byte
< start_byte
)
404 pagevec_init(&pvec
, 0);
405 while ((index
<= end
) &&
406 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
407 PAGECACHE_TAG_WRITEBACK
,
408 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
411 for (i
= 0; i
< nr_pages
; i
++) {
412 struct page
*page
= pvec
.pages
[i
];
414 /* until radix tree lookup accepts end_index */
415 if (page
->index
> end
)
418 wait_on_page_writeback(page
);
419 if (TestClearPageError(page
))
422 pagevec_release(&pvec
);
426 ret2
= filemap_check_errors(mapping
);
432 EXPORT_SYMBOL(filemap_fdatawait_range
);
435 * filemap_fdatawait - wait for all under-writeback pages to complete
436 * @mapping: address space structure to wait for
438 * Walk the list of under-writeback pages of the given address space
439 * and wait for all of them.
441 int filemap_fdatawait(struct address_space
*mapping
)
443 loff_t i_size
= i_size_read(mapping
->host
);
448 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
450 EXPORT_SYMBOL(filemap_fdatawait
);
452 int filemap_write_and_wait(struct address_space
*mapping
)
456 if (mapping
->nrpages
) {
457 err
= filemap_fdatawrite(mapping
);
459 * Even if the above returned error, the pages may be
460 * written partially (e.g. -ENOSPC), so we wait for it.
461 * But the -EIO is special case, it may indicate the worst
462 * thing (e.g. bug) happened, so we avoid waiting for it.
465 int err2
= filemap_fdatawait(mapping
);
470 err
= filemap_check_errors(mapping
);
474 EXPORT_SYMBOL(filemap_write_and_wait
);
477 * filemap_write_and_wait_range - write out & wait on a file range
478 * @mapping: the address_space for the pages
479 * @lstart: offset in bytes where the range starts
480 * @lend: offset in bytes where the range ends (inclusive)
482 * Write out and wait upon file offsets lstart->lend, inclusive.
484 * Note that `lend' is inclusive (describes the last byte to be written) so
485 * that this function can be used to write to the very end-of-file (end = -1).
487 int filemap_write_and_wait_range(struct address_space
*mapping
,
488 loff_t lstart
, loff_t lend
)
492 if (mapping
->nrpages
) {
493 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
495 /* See comment of filemap_write_and_wait() */
497 int err2
= filemap_fdatawait_range(mapping
,
503 err
= filemap_check_errors(mapping
);
507 EXPORT_SYMBOL(filemap_write_and_wait_range
);
510 * replace_page_cache_page - replace a pagecache page with a new one
511 * @old: page to be replaced
512 * @new: page to replace with
513 * @gfp_mask: allocation mode
515 * This function replaces a page in the pagecache with a new one. On
516 * success it acquires the pagecache reference for the new page and
517 * drops it for the old page. Both the old and new pages must be
518 * locked. This function does not add the new page to the LRU, the
519 * caller must do that.
521 * The remove + add is atomic. The only way this function can fail is
522 * memory allocation failure.
524 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
528 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
529 VM_BUG_ON_PAGE(!PageLocked(new), new);
530 VM_BUG_ON_PAGE(new->mapping
, new);
532 error
= radix_tree_preload(gfp_mask
& GFP_RECLAIM_MASK
);
534 struct address_space
*mapping
= old
->mapping
;
535 void (*freepage
)(struct page
*);
537 pgoff_t offset
= old
->index
;
538 freepage
= mapping
->a_ops
->freepage
;
541 new->mapping
= mapping
;
544 spin_lock_irq(&mapping
->tree_lock
);
545 __delete_from_page_cache(old
, NULL
);
546 error
= page_cache_tree_insert(mapping
, new, NULL
);
548 __inc_zone_page_state(new, NR_FILE_PAGES
);
549 if (PageSwapBacked(new))
550 __inc_zone_page_state(new, NR_SHMEM
);
551 spin_unlock_irq(&mapping
->tree_lock
);
552 /* mem_cgroup codes must not be called under tree_lock */
553 mem_cgroup_replace_page_cache(old
, new);
554 radix_tree_preload_end();
557 page_cache_release(old
);
562 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
564 static int __add_to_page_cache_locked(struct page
*page
,
565 struct address_space
*mapping
,
566 pgoff_t offset
, gfp_t gfp_mask
,
571 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
572 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
574 error
= mem_cgroup_charge_file(page
, current
->mm
,
575 gfp_mask
& GFP_RECLAIM_MASK
);
579 error
= radix_tree_maybe_preload(gfp_mask
& GFP_RECLAIM_MASK
);
581 mem_cgroup_uncharge_cache_page(page
);
585 page_cache_get(page
);
586 page
->mapping
= mapping
;
587 page
->index
= offset
;
589 spin_lock_irq(&mapping
->tree_lock
);
590 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
591 radix_tree_preload_end();
594 __inc_zone_page_state(page
, NR_FILE_PAGES
);
595 spin_unlock_irq(&mapping
->tree_lock
);
596 trace_mm_filemap_add_to_page_cache(page
);
599 page
->mapping
= NULL
;
600 /* Leave page->index set: truncation relies upon it */
601 spin_unlock_irq(&mapping
->tree_lock
);
602 mem_cgroup_uncharge_cache_page(page
);
603 page_cache_release(page
);
608 * add_to_page_cache_locked - add a locked page to the pagecache
610 * @mapping: the page's address_space
611 * @offset: page index
612 * @gfp_mask: page allocation mode
614 * This function is used to add a page to the pagecache. It must be locked.
615 * This function does not add the page to the LRU. The caller must do that.
617 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
618 pgoff_t offset
, gfp_t gfp_mask
)
620 return __add_to_page_cache_locked(page
, mapping
, offset
,
623 EXPORT_SYMBOL(add_to_page_cache_locked
);
625 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
626 pgoff_t offset
, gfp_t gfp_mask
)
631 __set_page_locked(page
);
632 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
635 __clear_page_locked(page
);
638 * The page might have been evicted from cache only
639 * recently, in which case it should be activated like
640 * any other repeatedly accessed page.
642 if (shadow
&& workingset_refault(shadow
)) {
644 workingset_activation(page
);
646 ClearPageActive(page
);
651 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
654 struct page
*__page_cache_alloc(gfp_t gfp
)
659 if (cpuset_do_page_mem_spread()) {
660 unsigned int cpuset_mems_cookie
;
662 cpuset_mems_cookie
= read_mems_allowed_begin();
663 n
= cpuset_mem_spread_node();
664 page
= alloc_pages_exact_node(n
, gfp
, 0);
665 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
669 return alloc_pages(gfp
, 0);
671 EXPORT_SYMBOL(__page_cache_alloc
);
675 * In order to wait for pages to become available there must be
676 * waitqueues associated with pages. By using a hash table of
677 * waitqueues where the bucket discipline is to maintain all
678 * waiters on the same queue and wake all when any of the pages
679 * become available, and for the woken contexts to check to be
680 * sure the appropriate page became available, this saves space
681 * at a cost of "thundering herd" phenomena during rare hash
684 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
686 const struct zone
*zone
= page_zone(page
);
688 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
691 static inline void wake_up_page(struct page
*page
, int bit
)
693 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
696 void wait_on_page_bit(struct page
*page
, int bit_nr
)
698 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
700 if (test_bit(bit_nr
, &page
->flags
))
701 __wait_on_bit(page_waitqueue(page
), &wait
, sleep_on_page
,
702 TASK_UNINTERRUPTIBLE
);
704 EXPORT_SYMBOL(wait_on_page_bit
);
706 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
708 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
710 if (!test_bit(bit_nr
, &page
->flags
))
713 return __wait_on_bit(page_waitqueue(page
), &wait
,
714 sleep_on_page_killable
, TASK_KILLABLE
);
718 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
719 * @page: Page defining the wait queue of interest
720 * @waiter: Waiter to add to the queue
722 * Add an arbitrary @waiter to the wait queue for the nominated @page.
724 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
726 wait_queue_head_t
*q
= page_waitqueue(page
);
729 spin_lock_irqsave(&q
->lock
, flags
);
730 __add_wait_queue(q
, waiter
);
731 spin_unlock_irqrestore(&q
->lock
, flags
);
733 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
736 * unlock_page - unlock a locked page
739 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
740 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
741 * mechananism between PageLocked pages and PageWriteback pages is shared.
742 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
744 * The mb is necessary to enforce ordering between the clear_bit and the read
745 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
747 void unlock_page(struct page
*page
)
749 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
750 clear_bit_unlock(PG_locked
, &page
->flags
);
751 smp_mb__after_atomic();
752 wake_up_page(page
, PG_locked
);
754 EXPORT_SYMBOL(unlock_page
);
757 * end_page_writeback - end writeback against a page
760 void end_page_writeback(struct page
*page
)
763 * TestClearPageReclaim could be used here but it is an atomic
764 * operation and overkill in this particular case. Failing to
765 * shuffle a page marked for immediate reclaim is too mild to
766 * justify taking an atomic operation penalty at the end of
767 * ever page writeback.
769 if (PageReclaim(page
)) {
770 ClearPageReclaim(page
);
771 rotate_reclaimable_page(page
);
774 if (!test_clear_page_writeback(page
))
777 smp_mb__after_atomic();
778 wake_up_page(page
, PG_writeback
);
780 EXPORT_SYMBOL(end_page_writeback
);
783 * After completing I/O on a page, call this routine to update the page
784 * flags appropriately
786 void page_endio(struct page
*page
, int rw
, int err
)
790 SetPageUptodate(page
);
792 ClearPageUptodate(page
);
796 } else { /* rw == WRITE */
798 struct address_space
*mapping
;
801 mapping
= page_mapping(page
);
803 mapping_set_error(mapping
, err
);
805 end_page_writeback(page
);
808 EXPORT_SYMBOL_GPL(page_endio
);
811 * __lock_page - get a lock on the page, assuming we need to sleep to get it
812 * @page: the page to lock
814 void __lock_page(struct page
*page
)
816 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
818 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sleep_on_page
,
819 TASK_UNINTERRUPTIBLE
);
821 EXPORT_SYMBOL(__lock_page
);
823 int __lock_page_killable(struct page
*page
)
825 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
827 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
828 sleep_on_page_killable
, TASK_KILLABLE
);
830 EXPORT_SYMBOL_GPL(__lock_page_killable
);
832 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
835 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
837 * CAUTION! In this case, mmap_sem is not released
838 * even though return 0.
840 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
843 up_read(&mm
->mmap_sem
);
844 if (flags
& FAULT_FLAG_KILLABLE
)
845 wait_on_page_locked_killable(page
);
847 wait_on_page_locked(page
);
850 if (flags
& FAULT_FLAG_KILLABLE
) {
853 ret
= __lock_page_killable(page
);
855 up_read(&mm
->mmap_sem
);
865 * page_cache_next_hole - find the next hole (not-present entry)
868 * @max_scan: maximum range to search
870 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
871 * lowest indexed hole.
873 * Returns: the index of the hole if found, otherwise returns an index
874 * outside of the set specified (in which case 'return - index >=
875 * max_scan' will be true). In rare cases of index wrap-around, 0 will
878 * page_cache_next_hole may be called under rcu_read_lock. However,
879 * like radix_tree_gang_lookup, this will not atomically search a
880 * snapshot of the tree at a single point in time. For example, if a
881 * hole is created at index 5, then subsequently a hole is created at
882 * index 10, page_cache_next_hole covering both indexes may return 10
883 * if called under rcu_read_lock.
885 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
886 pgoff_t index
, unsigned long max_scan
)
890 for (i
= 0; i
< max_scan
; i
++) {
893 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
894 if (!page
|| radix_tree_exceptional_entry(page
))
903 EXPORT_SYMBOL(page_cache_next_hole
);
906 * page_cache_prev_hole - find the prev hole (not-present entry)
909 * @max_scan: maximum range to search
911 * Search backwards in the range [max(index-max_scan+1, 0), index] for
914 * Returns: the index of the hole if found, otherwise returns an index
915 * outside of the set specified (in which case 'index - return >=
916 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
919 * page_cache_prev_hole may be called under rcu_read_lock. However,
920 * like radix_tree_gang_lookup, this will not atomically search a
921 * snapshot of the tree at a single point in time. For example, if a
922 * hole is created at index 10, then subsequently a hole is created at
923 * index 5, page_cache_prev_hole covering both indexes may return 5 if
924 * called under rcu_read_lock.
926 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
927 pgoff_t index
, unsigned long max_scan
)
931 for (i
= 0; i
< max_scan
; i
++) {
934 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
935 if (!page
|| radix_tree_exceptional_entry(page
))
938 if (index
== ULONG_MAX
)
944 EXPORT_SYMBOL(page_cache_prev_hole
);
947 * find_get_entry - find and get a page cache entry
948 * @mapping: the address_space to search
949 * @offset: the page cache index
951 * Looks up the page cache slot at @mapping & @offset. If there is a
952 * page cache page, it is returned with an increased refcount.
954 * If the slot holds a shadow entry of a previously evicted page, or a
955 * swap entry from shmem/tmpfs, it is returned.
957 * Otherwise, %NULL is returned.
959 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
967 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
969 page
= radix_tree_deref_slot(pagep
);
972 if (radix_tree_exception(page
)) {
973 if (radix_tree_deref_retry(page
))
976 * A shadow entry of a recently evicted page,
977 * or a swap entry from shmem/tmpfs. Return
978 * it without attempting to raise page count.
982 if (!page_cache_get_speculative(page
))
986 * Has the page moved?
987 * This is part of the lockless pagecache protocol. See
988 * include/linux/pagemap.h for details.
990 if (unlikely(page
!= *pagep
)) {
991 page_cache_release(page
);
1000 EXPORT_SYMBOL(find_get_entry
);
1003 * find_lock_entry - locate, pin and lock a page cache entry
1004 * @mapping: the address_space to search
1005 * @offset: the page cache index
1007 * Looks up the page cache slot at @mapping & @offset. If there is a
1008 * page cache page, it is returned locked and with an increased
1011 * If the slot holds a shadow entry of a previously evicted page, or a
1012 * swap entry from shmem/tmpfs, it is returned.
1014 * Otherwise, %NULL is returned.
1016 * find_lock_entry() may sleep.
1018 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1023 page
= find_get_entry(mapping
, offset
);
1024 if (page
&& !radix_tree_exception(page
)) {
1026 /* Has the page been truncated? */
1027 if (unlikely(page
->mapping
!= mapping
)) {
1029 page_cache_release(page
);
1032 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1036 EXPORT_SYMBOL(find_lock_entry
);
1039 * pagecache_get_page - find and get a page reference
1040 * @mapping: the address_space to search
1041 * @offset: the page index
1042 * @fgp_flags: PCG flags
1043 * @gfp_mask: gfp mask to use for the page cache data page allocation
1045 * Looks up the page cache slot at @mapping & @offset.
1047 * PCG flags modify how the page is returned.
1049 * FGP_ACCESSED: the page will be marked accessed
1050 * FGP_LOCK: Page is return locked
1051 * FGP_CREAT: If page is not present then a new page is allocated using
1052 * @gfp_mask and added to the page cache and the VM's LRU
1053 * list. The page is returned locked and with an increased
1054 * refcount. Otherwise, %NULL is returned.
1056 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1057 * if the GFP flags specified for FGP_CREAT are atomic.
1059 * If there is a page cache page, it is returned with an increased refcount.
1061 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1062 int fgp_flags
, gfp_t gfp_mask
)
1067 page
= find_get_entry(mapping
, offset
);
1068 if (radix_tree_exceptional_entry(page
))
1073 if (fgp_flags
& FGP_LOCK
) {
1074 if (fgp_flags
& FGP_NOWAIT
) {
1075 if (!trylock_page(page
)) {
1076 page_cache_release(page
);
1083 /* Has the page been truncated? */
1084 if (unlikely(page
->mapping
!= mapping
)) {
1086 page_cache_release(page
);
1089 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1092 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1093 mark_page_accessed(page
);
1096 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1098 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1099 gfp_mask
|= __GFP_WRITE
;
1100 if (fgp_flags
& FGP_NOFS
)
1101 gfp_mask
&= ~__GFP_FS
;
1103 page
= __page_cache_alloc(gfp_mask
);
1107 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1108 fgp_flags
|= FGP_LOCK
;
1110 /* Init accessed so avoit atomic mark_page_accessed later */
1111 if (fgp_flags
& FGP_ACCESSED
)
1112 init_page_accessed(page
);
1114 err
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
1115 if (unlikely(err
)) {
1116 page_cache_release(page
);
1125 EXPORT_SYMBOL(pagecache_get_page
);
1128 * find_get_entries - gang pagecache lookup
1129 * @mapping: The address_space to search
1130 * @start: The starting page cache index
1131 * @nr_entries: The maximum number of entries
1132 * @entries: Where the resulting entries are placed
1133 * @indices: The cache indices corresponding to the entries in @entries
1135 * find_get_entries() will search for and return a group of up to
1136 * @nr_entries entries in the mapping. The entries are placed at
1137 * @entries. find_get_entries() takes a reference against any actual
1140 * The search returns a group of mapping-contiguous page cache entries
1141 * with ascending indexes. There may be holes in the indices due to
1142 * not-present pages.
1144 * Any shadow entries of evicted pages, or swap entries from
1145 * shmem/tmpfs, are included in the returned array.
1147 * find_get_entries() returns the number of pages and shadow entries
1150 unsigned find_get_entries(struct address_space
*mapping
,
1151 pgoff_t start
, unsigned int nr_entries
,
1152 struct page
**entries
, pgoff_t
*indices
)
1155 unsigned int ret
= 0;
1156 struct radix_tree_iter iter
;
1163 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1166 page
= radix_tree_deref_slot(slot
);
1167 if (unlikely(!page
))
1169 if (radix_tree_exception(page
)) {
1170 if (radix_tree_deref_retry(page
))
1173 * A shadow entry of a recently evicted page,
1174 * or a swap entry from shmem/tmpfs. Return
1175 * it without attempting to raise page count.
1179 if (!page_cache_get_speculative(page
))
1182 /* Has the page moved? */
1183 if (unlikely(page
!= *slot
)) {
1184 page_cache_release(page
);
1188 indices
[ret
] = iter
.index
;
1189 entries
[ret
] = page
;
1190 if (++ret
== nr_entries
)
1198 * find_get_pages - gang pagecache lookup
1199 * @mapping: The address_space to search
1200 * @start: The starting page index
1201 * @nr_pages: The maximum number of pages
1202 * @pages: Where the resulting pages are placed
1204 * find_get_pages() will search for and return a group of up to
1205 * @nr_pages pages in the mapping. The pages are placed at @pages.
1206 * find_get_pages() takes a reference against the returned pages.
1208 * The search returns a group of mapping-contiguous pages with ascending
1209 * indexes. There may be holes in the indices due to not-present pages.
1211 * find_get_pages() returns the number of pages which were found.
1213 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1214 unsigned int nr_pages
, struct page
**pages
)
1216 struct radix_tree_iter iter
;
1220 if (unlikely(!nr_pages
))
1225 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1228 page
= radix_tree_deref_slot(slot
);
1229 if (unlikely(!page
))
1232 if (radix_tree_exception(page
)) {
1233 if (radix_tree_deref_retry(page
)) {
1235 * Transient condition which can only trigger
1236 * when entry at index 0 moves out of or back
1237 * to root: none yet gotten, safe to restart.
1239 WARN_ON(iter
.index
);
1243 * A shadow entry of a recently evicted page,
1244 * or a swap entry from shmem/tmpfs. Skip
1250 if (!page_cache_get_speculative(page
))
1253 /* Has the page moved? */
1254 if (unlikely(page
!= *slot
)) {
1255 page_cache_release(page
);
1260 if (++ret
== nr_pages
)
1269 * find_get_pages_contig - gang contiguous pagecache lookup
1270 * @mapping: The address_space to search
1271 * @index: The starting page index
1272 * @nr_pages: The maximum number of pages
1273 * @pages: Where the resulting pages are placed
1275 * find_get_pages_contig() works exactly like find_get_pages(), except
1276 * that the returned number of pages are guaranteed to be contiguous.
1278 * find_get_pages_contig() returns the number of pages which were found.
1280 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1281 unsigned int nr_pages
, struct page
**pages
)
1283 struct radix_tree_iter iter
;
1285 unsigned int ret
= 0;
1287 if (unlikely(!nr_pages
))
1292 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1295 page
= radix_tree_deref_slot(slot
);
1296 /* The hole, there no reason to continue */
1297 if (unlikely(!page
))
1300 if (radix_tree_exception(page
)) {
1301 if (radix_tree_deref_retry(page
)) {
1303 * Transient condition which can only trigger
1304 * when entry at index 0 moves out of or back
1305 * to root: none yet gotten, safe to restart.
1310 * A shadow entry of a recently evicted page,
1311 * or a swap entry from shmem/tmpfs. Stop
1312 * looking for contiguous pages.
1317 if (!page_cache_get_speculative(page
))
1320 /* Has the page moved? */
1321 if (unlikely(page
!= *slot
)) {
1322 page_cache_release(page
);
1327 * must check mapping and index after taking the ref.
1328 * otherwise we can get both false positives and false
1329 * negatives, which is just confusing to the caller.
1331 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1332 page_cache_release(page
);
1337 if (++ret
== nr_pages
)
1343 EXPORT_SYMBOL(find_get_pages_contig
);
1346 * find_get_pages_tag - find and return pages that match @tag
1347 * @mapping: the address_space to search
1348 * @index: the starting page index
1349 * @tag: the tag index
1350 * @nr_pages: the maximum number of pages
1351 * @pages: where the resulting pages are placed
1353 * Like find_get_pages, except we only return pages which are tagged with
1354 * @tag. We update @index to index the next page for the traversal.
1356 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1357 int tag
, unsigned int nr_pages
, struct page
**pages
)
1359 struct radix_tree_iter iter
;
1363 if (unlikely(!nr_pages
))
1368 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1369 &iter
, *index
, tag
) {
1372 page
= radix_tree_deref_slot(slot
);
1373 if (unlikely(!page
))
1376 if (radix_tree_exception(page
)) {
1377 if (radix_tree_deref_retry(page
)) {
1379 * Transient condition which can only trigger
1380 * when entry at index 0 moves out of or back
1381 * to root: none yet gotten, safe to restart.
1386 * A shadow entry of a recently evicted page.
1388 * Those entries should never be tagged, but
1389 * this tree walk is lockless and the tags are
1390 * looked up in bulk, one radix tree node at a
1391 * time, so there is a sizable window for page
1392 * reclaim to evict a page we saw tagged.
1399 if (!page_cache_get_speculative(page
))
1402 /* Has the page moved? */
1403 if (unlikely(page
!= *slot
)) {
1404 page_cache_release(page
);
1409 if (++ret
== nr_pages
)
1416 *index
= pages
[ret
- 1]->index
+ 1;
1420 EXPORT_SYMBOL(find_get_pages_tag
);
1423 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1424 * a _large_ part of the i/o request. Imagine the worst scenario:
1426 * ---R__________________________________________B__________
1427 * ^ reading here ^ bad block(assume 4k)
1429 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1430 * => failing the whole request => read(R) => read(R+1) =>
1431 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1432 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1433 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1435 * It is going insane. Fix it by quickly scaling down the readahead size.
1437 static void shrink_readahead_size_eio(struct file
*filp
,
1438 struct file_ra_state
*ra
)
1444 * do_generic_file_read - generic file read routine
1445 * @filp: the file to read
1446 * @ppos: current file position
1447 * @iter: data destination
1448 * @written: already copied
1450 * This is a generic file read routine, and uses the
1451 * mapping->a_ops->readpage() function for the actual low-level stuff.
1453 * This is really ugly. But the goto's actually try to clarify some
1454 * of the logic when it comes to error handling etc.
1456 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1457 struct iov_iter
*iter
, ssize_t written
)
1459 struct address_space
*mapping
= filp
->f_mapping
;
1460 struct inode
*inode
= mapping
->host
;
1461 struct file_ra_state
*ra
= &filp
->f_ra
;
1465 unsigned long offset
; /* offset into pagecache page */
1466 unsigned int prev_offset
;
1469 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
1471 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
1473 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1474 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1475 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1476 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1477 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1483 unsigned long nr
, ret
;
1487 if (fatal_signal_pending(current
)) {
1492 page
= find_get_page(mapping
, index
);
1494 page_cache_sync_readahead(mapping
,
1496 index
, last_index
- index
);
1497 page
= find_get_page(mapping
, index
);
1498 if (unlikely(page
== NULL
))
1499 goto no_cached_page
;
1501 if (PageReadahead(page
)) {
1502 page_cache_async_readahead(mapping
,
1504 index
, last_index
- index
);
1506 if (!PageUptodate(page
)) {
1507 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1508 !mapping
->a_ops
->is_partially_uptodate
)
1509 goto page_not_up_to_date
;
1510 if (!trylock_page(page
))
1511 goto page_not_up_to_date
;
1512 /* Did it get truncated before we got the lock? */
1514 goto page_not_up_to_date_locked
;
1515 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1516 offset
, iter
->count
))
1517 goto page_not_up_to_date_locked
;
1522 * i_size must be checked after we know the page is Uptodate.
1524 * Checking i_size after the check allows us to calculate
1525 * the correct value for "nr", which means the zero-filled
1526 * part of the page is not copied back to userspace (unless
1527 * another truncate extends the file - this is desired though).
1530 isize
= i_size_read(inode
);
1531 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1532 if (unlikely(!isize
|| index
> end_index
)) {
1533 page_cache_release(page
);
1537 /* nr is the maximum number of bytes to copy from this page */
1538 nr
= PAGE_CACHE_SIZE
;
1539 if (index
== end_index
) {
1540 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1542 page_cache_release(page
);
1548 /* If users can be writing to this page using arbitrary
1549 * virtual addresses, take care about potential aliasing
1550 * before reading the page on the kernel side.
1552 if (mapping_writably_mapped(mapping
))
1553 flush_dcache_page(page
);
1556 * When a sequential read accesses a page several times,
1557 * only mark it as accessed the first time.
1559 if (prev_index
!= index
|| offset
!= prev_offset
)
1560 mark_page_accessed(page
);
1564 * Ok, we have the page, and it's up-to-date, so
1565 * now we can copy it to user space...
1568 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1570 index
+= offset
>> PAGE_CACHE_SHIFT
;
1571 offset
&= ~PAGE_CACHE_MASK
;
1572 prev_offset
= offset
;
1574 page_cache_release(page
);
1576 if (!iov_iter_count(iter
))
1584 page_not_up_to_date
:
1585 /* Get exclusive access to the page ... */
1586 error
= lock_page_killable(page
);
1587 if (unlikely(error
))
1588 goto readpage_error
;
1590 page_not_up_to_date_locked
:
1591 /* Did it get truncated before we got the lock? */
1592 if (!page
->mapping
) {
1594 page_cache_release(page
);
1598 /* Did somebody else fill it already? */
1599 if (PageUptodate(page
)) {
1606 * A previous I/O error may have been due to temporary
1607 * failures, eg. multipath errors.
1608 * PG_error will be set again if readpage fails.
1610 ClearPageError(page
);
1611 /* Start the actual read. The read will unlock the page. */
1612 error
= mapping
->a_ops
->readpage(filp
, page
);
1614 if (unlikely(error
)) {
1615 if (error
== AOP_TRUNCATED_PAGE
) {
1616 page_cache_release(page
);
1620 goto readpage_error
;
1623 if (!PageUptodate(page
)) {
1624 error
= lock_page_killable(page
);
1625 if (unlikely(error
))
1626 goto readpage_error
;
1627 if (!PageUptodate(page
)) {
1628 if (page
->mapping
== NULL
) {
1630 * invalidate_mapping_pages got it
1633 page_cache_release(page
);
1637 shrink_readahead_size_eio(filp
, ra
);
1639 goto readpage_error
;
1647 /* UHHUH! A synchronous read error occurred. Report it */
1648 page_cache_release(page
);
1653 * Ok, it wasn't cached, so we need to create a new
1656 page
= page_cache_alloc_cold(mapping
);
1661 error
= add_to_page_cache_lru(page
, mapping
,
1664 page_cache_release(page
);
1665 if (error
== -EEXIST
) {
1675 ra
->prev_pos
= prev_index
;
1676 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1677 ra
->prev_pos
|= prev_offset
;
1679 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1680 file_accessed(filp
);
1681 return written
? written
: error
;
1685 * generic_file_read_iter - generic filesystem read routine
1686 * @iocb: kernel I/O control block
1687 * @iter: destination for the data read
1689 * This is the "read_iter()" routine for all filesystems
1690 * that can use the page cache directly.
1693 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1695 struct file
*file
= iocb
->ki_filp
;
1697 loff_t
*ppos
= &iocb
->ki_pos
;
1700 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1701 if (file
->f_flags
& O_DIRECT
) {
1702 struct address_space
*mapping
= file
->f_mapping
;
1703 struct inode
*inode
= mapping
->host
;
1704 size_t count
= iov_iter_count(iter
);
1708 goto out
; /* skip atime */
1709 size
= i_size_read(inode
);
1710 retval
= filemap_write_and_wait_range(mapping
, pos
,
1713 struct iov_iter data
= *iter
;
1714 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
, &data
, pos
);
1718 *ppos
= pos
+ retval
;
1719 iov_iter_advance(iter
, retval
);
1723 * Btrfs can have a short DIO read if we encounter
1724 * compressed extents, so if there was an error, or if
1725 * we've already read everything we wanted to, or if
1726 * there was a short read because we hit EOF, go ahead
1727 * and return. Otherwise fallthrough to buffered io for
1728 * the rest of the read.
1730 if (retval
< 0 || !iov_iter_count(iter
) || *ppos
>= size
) {
1731 file_accessed(file
);
1736 retval
= do_generic_file_read(file
, ppos
, iter
, retval
);
1740 EXPORT_SYMBOL(generic_file_read_iter
);
1744 * page_cache_read - adds requested page to the page cache if not already there
1745 * @file: file to read
1746 * @offset: page index
1748 * This adds the requested page to the page cache if it isn't already there,
1749 * and schedules an I/O to read in its contents from disk.
1751 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1753 struct address_space
*mapping
= file
->f_mapping
;
1758 page
= page_cache_alloc_cold(mapping
);
1762 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1764 ret
= mapping
->a_ops
->readpage(file
, page
);
1765 else if (ret
== -EEXIST
)
1766 ret
= 0; /* losing race to add is OK */
1768 page_cache_release(page
);
1770 } while (ret
== AOP_TRUNCATED_PAGE
);
1775 #define MMAP_LOTSAMISS (100)
1778 * Synchronous readahead happens when we don't even find
1779 * a page in the page cache at all.
1781 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1782 struct file_ra_state
*ra
,
1786 unsigned long ra_pages
;
1787 struct address_space
*mapping
= file
->f_mapping
;
1789 /* If we don't want any read-ahead, don't bother */
1790 if (vma
->vm_flags
& VM_RAND_READ
)
1795 if (vma
->vm_flags
& VM_SEQ_READ
) {
1796 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1801 /* Avoid banging the cache line if not needed */
1802 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1806 * Do we miss much more than hit in this file? If so,
1807 * stop bothering with read-ahead. It will only hurt.
1809 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1815 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1816 ra
->start
= max_t(long, 0, offset
- ra_pages
/ 2);
1817 ra
->size
= ra_pages
;
1818 ra
->async_size
= ra_pages
/ 4;
1819 ra_submit(ra
, mapping
, file
);
1823 * Asynchronous readahead happens when we find the page and PG_readahead,
1824 * so we want to possibly extend the readahead further..
1826 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1827 struct file_ra_state
*ra
,
1832 struct address_space
*mapping
= file
->f_mapping
;
1834 /* If we don't want any read-ahead, don't bother */
1835 if (vma
->vm_flags
& VM_RAND_READ
)
1837 if (ra
->mmap_miss
> 0)
1839 if (PageReadahead(page
))
1840 page_cache_async_readahead(mapping
, ra
, file
,
1841 page
, offset
, ra
->ra_pages
);
1845 * filemap_fault - read in file data for page fault handling
1846 * @vma: vma in which the fault was taken
1847 * @vmf: struct vm_fault containing details of the fault
1849 * filemap_fault() is invoked via the vma operations vector for a
1850 * mapped memory region to read in file data during a page fault.
1852 * The goto's are kind of ugly, but this streamlines the normal case of having
1853 * it in the page cache, and handles the special cases reasonably without
1854 * having a lot of duplicated code.
1856 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1859 struct file
*file
= vma
->vm_file
;
1860 struct address_space
*mapping
= file
->f_mapping
;
1861 struct file_ra_state
*ra
= &file
->f_ra
;
1862 struct inode
*inode
= mapping
->host
;
1863 pgoff_t offset
= vmf
->pgoff
;
1868 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1869 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
1870 return VM_FAULT_SIGBUS
;
1873 * Do we have something in the page cache already?
1875 page
= find_get_page(mapping
, offset
);
1876 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1878 * We found the page, so try async readahead before
1879 * waiting for the lock.
1881 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1883 /* No page in the page cache at all */
1884 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1885 count_vm_event(PGMAJFAULT
);
1886 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1887 ret
= VM_FAULT_MAJOR
;
1889 page
= find_get_page(mapping
, offset
);
1891 goto no_cached_page
;
1894 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1895 page_cache_release(page
);
1896 return ret
| VM_FAULT_RETRY
;
1899 /* Did it get truncated? */
1900 if (unlikely(page
->mapping
!= mapping
)) {
1905 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1908 * We have a locked page in the page cache, now we need to check
1909 * that it's up-to-date. If not, it is going to be due to an error.
1911 if (unlikely(!PageUptodate(page
)))
1912 goto page_not_uptodate
;
1915 * Found the page and have a reference on it.
1916 * We must recheck i_size under page lock.
1918 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1919 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
1921 page_cache_release(page
);
1922 return VM_FAULT_SIGBUS
;
1926 return ret
| VM_FAULT_LOCKED
;
1930 * We're only likely to ever get here if MADV_RANDOM is in
1933 error
= page_cache_read(file
, offset
);
1936 * The page we want has now been added to the page cache.
1937 * In the unlikely event that someone removed it in the
1938 * meantime, we'll just come back here and read it again.
1944 * An error return from page_cache_read can result if the
1945 * system is low on memory, or a problem occurs while trying
1948 if (error
== -ENOMEM
)
1949 return VM_FAULT_OOM
;
1950 return VM_FAULT_SIGBUS
;
1954 * Umm, take care of errors if the page isn't up-to-date.
1955 * Try to re-read it _once_. We do this synchronously,
1956 * because there really aren't any performance issues here
1957 * and we need to check for errors.
1959 ClearPageError(page
);
1960 error
= mapping
->a_ops
->readpage(file
, page
);
1962 wait_on_page_locked(page
);
1963 if (!PageUptodate(page
))
1966 page_cache_release(page
);
1968 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1971 /* Things didn't work out. Return zero to tell the mm layer so. */
1972 shrink_readahead_size_eio(file
, ra
);
1973 return VM_FAULT_SIGBUS
;
1975 EXPORT_SYMBOL(filemap_fault
);
1977 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1979 struct radix_tree_iter iter
;
1981 struct file
*file
= vma
->vm_file
;
1982 struct address_space
*mapping
= file
->f_mapping
;
1985 unsigned long address
= (unsigned long) vmf
->virtual_address
;
1990 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
1991 if (iter
.index
> vmf
->max_pgoff
)
1994 page
= radix_tree_deref_slot(slot
);
1995 if (unlikely(!page
))
1997 if (radix_tree_exception(page
)) {
1998 if (radix_tree_deref_retry(page
))
2004 if (!page_cache_get_speculative(page
))
2007 /* Has the page moved? */
2008 if (unlikely(page
!= *slot
)) {
2009 page_cache_release(page
);
2013 if (!PageUptodate(page
) ||
2014 PageReadahead(page
) ||
2017 if (!trylock_page(page
))
2020 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2023 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
2024 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2027 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2028 if (!pte_none(*pte
))
2031 if (file
->f_ra
.mmap_miss
> 0)
2032 file
->f_ra
.mmap_miss
--;
2033 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2034 do_set_pte(vma
, addr
, page
, pte
, false, false);
2040 page_cache_release(page
);
2042 if (iter
.index
== vmf
->max_pgoff
)
2047 EXPORT_SYMBOL(filemap_map_pages
);
2049 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2051 struct page
*page
= vmf
->page
;
2052 struct inode
*inode
= file_inode(vma
->vm_file
);
2053 int ret
= VM_FAULT_LOCKED
;
2055 sb_start_pagefault(inode
->i_sb
);
2056 file_update_time(vma
->vm_file
);
2058 if (page
->mapping
!= inode
->i_mapping
) {
2060 ret
= VM_FAULT_NOPAGE
;
2064 * We mark the page dirty already here so that when freeze is in
2065 * progress, we are guaranteed that writeback during freezing will
2066 * see the dirty page and writeprotect it again.
2068 set_page_dirty(page
);
2069 wait_for_stable_page(page
);
2071 sb_end_pagefault(inode
->i_sb
);
2074 EXPORT_SYMBOL(filemap_page_mkwrite
);
2076 const struct vm_operations_struct generic_file_vm_ops
= {
2077 .fault
= filemap_fault
,
2078 .map_pages
= filemap_map_pages
,
2079 .page_mkwrite
= filemap_page_mkwrite
,
2082 /* This is used for a general mmap of a disk file */
2084 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2086 struct address_space
*mapping
= file
->f_mapping
;
2088 if (!mapping
->a_ops
->readpage
)
2090 file_accessed(file
);
2091 vma
->vm_ops
= &generic_file_vm_ops
;
2096 * This is for filesystems which do not implement ->writepage.
2098 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2100 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2102 return generic_file_mmap(file
, vma
);
2105 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2109 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2113 #endif /* CONFIG_MMU */
2115 EXPORT_SYMBOL(generic_file_mmap
);
2116 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2118 static struct page
*wait_on_page_read(struct page
*page
)
2120 if (!IS_ERR(page
)) {
2121 wait_on_page_locked(page
);
2122 if (!PageUptodate(page
)) {
2123 page_cache_release(page
);
2124 page
= ERR_PTR(-EIO
);
2130 static struct page
*__read_cache_page(struct address_space
*mapping
,
2132 int (*filler
)(void *, struct page
*),
2139 page
= find_get_page(mapping
, index
);
2141 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2143 return ERR_PTR(-ENOMEM
);
2144 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2145 if (unlikely(err
)) {
2146 page_cache_release(page
);
2149 /* Presumably ENOMEM for radix tree node */
2150 return ERR_PTR(err
);
2152 err
= filler(data
, page
);
2154 page_cache_release(page
);
2155 page
= ERR_PTR(err
);
2157 page
= wait_on_page_read(page
);
2163 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2165 int (*filler
)(void *, struct page
*),
2174 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2177 if (PageUptodate(page
))
2181 if (!page
->mapping
) {
2183 page_cache_release(page
);
2186 if (PageUptodate(page
)) {
2190 err
= filler(data
, page
);
2192 page_cache_release(page
);
2193 return ERR_PTR(err
);
2195 page
= wait_on_page_read(page
);
2200 mark_page_accessed(page
);
2205 * read_cache_page - read into page cache, fill it if needed
2206 * @mapping: the page's address_space
2207 * @index: the page index
2208 * @filler: function to perform the read
2209 * @data: first arg to filler(data, page) function, often left as NULL
2211 * Read into the page cache. If a page already exists, and PageUptodate() is
2212 * not set, try to fill the page and wait for it to become unlocked.
2214 * If the page does not get brought uptodate, return -EIO.
2216 struct page
*read_cache_page(struct address_space
*mapping
,
2218 int (*filler
)(void *, struct page
*),
2221 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2223 EXPORT_SYMBOL(read_cache_page
);
2226 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2227 * @mapping: the page's address_space
2228 * @index: the page index
2229 * @gfp: the page allocator flags to use if allocating
2231 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2232 * any new page allocations done using the specified allocation flags.
2234 * If the page does not get brought uptodate, return -EIO.
2236 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2240 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2242 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2244 EXPORT_SYMBOL(read_cache_page_gfp
);
2247 * Performs necessary checks before doing a write
2249 * Can adjust writing position or amount of bytes to write.
2250 * Returns appropriate error code that caller should return or
2251 * zero in case that write should be allowed.
2253 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
2255 struct inode
*inode
= file
->f_mapping
->host
;
2256 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2258 if (unlikely(*pos
< 0))
2262 /* FIXME: this is for backwards compatibility with 2.4 */
2263 if (file
->f_flags
& O_APPEND
)
2264 *pos
= i_size_read(inode
);
2266 if (limit
!= RLIM_INFINITY
) {
2267 if (*pos
>= limit
) {
2268 send_sig(SIGXFSZ
, current
, 0);
2271 if (*count
> limit
- (typeof(limit
))*pos
) {
2272 *count
= limit
- (typeof(limit
))*pos
;
2280 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2281 !(file
->f_flags
& O_LARGEFILE
))) {
2282 if (*pos
>= MAX_NON_LFS
) {
2285 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2286 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2291 * Are we about to exceed the fs block limit ?
2293 * If we have written data it becomes a short write. If we have
2294 * exceeded without writing data we send a signal and return EFBIG.
2295 * Linus frestrict idea will clean these up nicely..
2297 if (likely(!isblk
)) {
2298 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2299 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2302 /* zero-length writes at ->s_maxbytes are OK */
2305 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2306 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2310 if (bdev_read_only(I_BDEV(inode
)))
2312 isize
= i_size_read(inode
);
2313 if (*pos
>= isize
) {
2314 if (*count
|| *pos
> isize
)
2318 if (*pos
+ *count
> isize
)
2319 *count
= isize
- *pos
;
2326 EXPORT_SYMBOL(generic_write_checks
);
2328 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2329 loff_t pos
, unsigned len
, unsigned flags
,
2330 struct page
**pagep
, void **fsdata
)
2332 const struct address_space_operations
*aops
= mapping
->a_ops
;
2334 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2337 EXPORT_SYMBOL(pagecache_write_begin
);
2339 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2340 loff_t pos
, unsigned len
, unsigned copied
,
2341 struct page
*page
, void *fsdata
)
2343 const struct address_space_operations
*aops
= mapping
->a_ops
;
2345 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2347 EXPORT_SYMBOL(pagecache_write_end
);
2350 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
2352 struct file
*file
= iocb
->ki_filp
;
2353 struct address_space
*mapping
= file
->f_mapping
;
2354 struct inode
*inode
= mapping
->host
;
2358 struct iov_iter data
;
2360 write_len
= iov_iter_count(from
);
2361 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2363 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2368 * After a write we want buffered reads to be sure to go to disk to get
2369 * the new data. We invalidate clean cached page from the region we're
2370 * about to write. We do this *before* the write so that we can return
2371 * without clobbering -EIOCBQUEUED from ->direct_IO().
2373 if (mapping
->nrpages
) {
2374 written
= invalidate_inode_pages2_range(mapping
,
2375 pos
>> PAGE_CACHE_SHIFT
, end
);
2377 * If a page can not be invalidated, return 0 to fall back
2378 * to buffered write.
2381 if (written
== -EBUSY
)
2388 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, &data
, pos
);
2391 * Finally, try again to invalidate clean pages which might have been
2392 * cached by non-direct readahead, or faulted in by get_user_pages()
2393 * if the source of the write was an mmap'ed region of the file
2394 * we're writing. Either one is a pretty crazy thing to do,
2395 * so we don't support it 100%. If this invalidation
2396 * fails, tough, the write still worked...
2398 if (mapping
->nrpages
) {
2399 invalidate_inode_pages2_range(mapping
,
2400 pos
>> PAGE_CACHE_SHIFT
, end
);
2405 iov_iter_advance(from
, written
);
2406 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2407 i_size_write(inode
, pos
);
2408 mark_inode_dirty(inode
);
2415 EXPORT_SYMBOL(generic_file_direct_write
);
2418 * Find or create a page at the given pagecache position. Return the locked
2419 * page. This function is specifically for buffered writes.
2421 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2422 pgoff_t index
, unsigned flags
)
2425 int fgp_flags
= FGP_LOCK
|FGP_ACCESSED
|FGP_WRITE
|FGP_CREAT
;
2427 if (flags
& AOP_FLAG_NOFS
)
2428 fgp_flags
|= FGP_NOFS
;
2430 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2431 mapping_gfp_mask(mapping
));
2433 wait_for_stable_page(page
);
2437 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2439 ssize_t
generic_perform_write(struct file
*file
,
2440 struct iov_iter
*i
, loff_t pos
)
2442 struct address_space
*mapping
= file
->f_mapping
;
2443 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2445 ssize_t written
= 0;
2446 unsigned int flags
= 0;
2449 * Copies from kernel address space cannot fail (NFSD is a big user).
2451 if (segment_eq(get_fs(), KERNEL_DS
))
2452 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2456 unsigned long offset
; /* Offset into pagecache page */
2457 unsigned long bytes
; /* Bytes to write to page */
2458 size_t copied
; /* Bytes copied from user */
2461 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2462 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2467 * Bring in the user page that we will copy from _first_.
2468 * Otherwise there's a nasty deadlock on copying from the
2469 * same page as we're writing to, without it being marked
2472 * Not only is this an optimisation, but it is also required
2473 * to check that the address is actually valid, when atomic
2474 * usercopies are used, below.
2476 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2481 if (fatal_signal_pending(current
)) {
2486 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2488 if (unlikely(status
< 0))
2491 if (mapping_writably_mapped(mapping
))
2492 flush_dcache_page(page
);
2494 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2495 flush_dcache_page(page
);
2497 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2499 if (unlikely(status
< 0))
2505 iov_iter_advance(i
, copied
);
2506 if (unlikely(copied
== 0)) {
2508 * If we were unable to copy any data at all, we must
2509 * fall back to a single segment length write.
2511 * If we didn't fallback here, we could livelock
2512 * because not all segments in the iov can be copied at
2513 * once without a pagefault.
2515 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2516 iov_iter_single_seg_count(i
));
2522 balance_dirty_pages_ratelimited(mapping
);
2523 } while (iov_iter_count(i
));
2525 return written
? written
: status
;
2527 EXPORT_SYMBOL(generic_perform_write
);
2530 * __generic_file_write_iter - write data to a file
2531 * @iocb: IO state structure (file, offset, etc.)
2532 * @from: iov_iter with data to write
2534 * This function does all the work needed for actually writing data to a
2535 * file. It does all basic checks, removes SUID from the file, updates
2536 * modification times and calls proper subroutines depending on whether we
2537 * do direct IO or a standard buffered write.
2539 * It expects i_mutex to be grabbed unless we work on a block device or similar
2540 * object which does not need locking at all.
2542 * This function does *not* take care of syncing data in case of O_SYNC write.
2543 * A caller has to handle it. This is mainly due to the fact that we want to
2544 * avoid syncing under i_mutex.
2546 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2548 struct file
*file
= iocb
->ki_filp
;
2549 struct address_space
* mapping
= file
->f_mapping
;
2550 struct inode
*inode
= mapping
->host
;
2551 loff_t pos
= iocb
->ki_pos
;
2552 ssize_t written
= 0;
2555 size_t count
= iov_iter_count(from
);
2557 /* We can write back this queue in page reclaim */
2558 current
->backing_dev_info
= mapping
->backing_dev_info
;
2559 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2566 iov_iter_truncate(from
, count
);
2568 err
= file_remove_suid(file
);
2572 err
= file_update_time(file
);
2576 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2577 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2580 written
= generic_file_direct_write(iocb
, from
, pos
);
2581 if (written
< 0 || written
== count
)
2585 * direct-io write to a hole: fall through to buffered I/O
2586 * for completing the rest of the request.
2591 status
= generic_perform_write(file
, from
, pos
);
2593 * If generic_perform_write() returned a synchronous error
2594 * then we want to return the number of bytes which were
2595 * direct-written, or the error code if that was zero. Note
2596 * that this differs from normal direct-io semantics, which
2597 * will return -EFOO even if some bytes were written.
2599 if (unlikely(status
< 0)) {
2603 iocb
->ki_pos
= pos
+ status
;
2605 * We need to ensure that the page cache pages are written to
2606 * disk and invalidated to preserve the expected O_DIRECT
2609 endbyte
= pos
+ status
- 1;
2610 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
2613 invalidate_mapping_pages(mapping
,
2614 pos
>> PAGE_CACHE_SHIFT
,
2615 endbyte
>> PAGE_CACHE_SHIFT
);
2618 * We don't know how much we wrote, so just return
2619 * the number of bytes which were direct-written
2623 written
= generic_perform_write(file
, from
, pos
);
2624 if (likely(written
>= 0))
2625 iocb
->ki_pos
= pos
+ written
;
2628 current
->backing_dev_info
= NULL
;
2629 return written
? written
: err
;
2631 EXPORT_SYMBOL(__generic_file_write_iter
);
2634 * generic_file_write_iter - write data to a file
2635 * @iocb: IO state structure
2636 * @from: iov_iter with data to write
2638 * This is a wrapper around __generic_file_write_iter() to be used by most
2639 * filesystems. It takes care of syncing the file in case of O_SYNC file
2640 * and acquires i_mutex as needed.
2642 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2644 struct file
*file
= iocb
->ki_filp
;
2645 struct inode
*inode
= file
->f_mapping
->host
;
2648 mutex_lock(&inode
->i_mutex
);
2649 ret
= __generic_file_write_iter(iocb
, from
);
2650 mutex_unlock(&inode
->i_mutex
);
2655 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2661 EXPORT_SYMBOL(generic_file_write_iter
);
2664 * try_to_release_page() - release old fs-specific metadata on a page
2666 * @page: the page which the kernel is trying to free
2667 * @gfp_mask: memory allocation flags (and I/O mode)
2669 * The address_space is to try to release any data against the page
2670 * (presumably at page->private). If the release was successful, return `1'.
2671 * Otherwise return zero.
2673 * This may also be called if PG_fscache is set on a page, indicating that the
2674 * page is known to the local caching routines.
2676 * The @gfp_mask argument specifies whether I/O may be performed to release
2677 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2680 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2682 struct address_space
* const mapping
= page
->mapping
;
2684 BUG_ON(!PageLocked(page
));
2685 if (PageWriteback(page
))
2688 if (mapping
&& mapping
->a_ops
->releasepage
)
2689 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2690 return try_to_free_buffers(page
);
2693 EXPORT_SYMBOL(try_to_release_page
);