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/capability.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/gfp.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/backing-dev.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/cpuset.h>
32 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
33 #include <linux/hugetlb.h>
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_rwsem (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_rwsem (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 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
112 static void page_cache_tree_delete(struct address_space
*mapping
,
113 struct page
*page
, void *shadow
)
115 struct radix_tree_node
*node
;
121 VM_BUG_ON(!PageLocked(page
));
123 __radix_tree_lookup(&mapping
->page_tree
, page
->index
, &node
, &slot
);
126 mapping
->nrshadows
++;
128 * Make sure the nrshadows update is committed before
129 * the nrpages update so that final truncate racing
130 * with reclaim does not see both counters 0 at the
131 * same time and miss a shadow entry.
138 /* Clear direct pointer tags in root node */
139 mapping
->page_tree
.gfp_mask
&= __GFP_BITS_MASK
;
140 radix_tree_replace_slot(slot
, shadow
);
144 /* Clear tree tags for the removed page */
146 offset
= index
& RADIX_TREE_MAP_MASK
;
147 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
148 if (test_bit(offset
, node
->tags
[tag
]))
149 radix_tree_tag_clear(&mapping
->page_tree
, index
, tag
);
152 /* Delete page, swap shadow entry */
153 radix_tree_replace_slot(slot
, shadow
);
154 workingset_node_pages_dec(node
);
156 workingset_node_shadows_inc(node
);
158 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
162 * Track node that only contains shadow entries.
164 * Avoid acquiring the list_lru lock if already tracked. The
165 * list_empty() test is safe as node->private_list is
166 * protected by mapping->tree_lock.
168 if (!workingset_node_pages(node
) &&
169 list_empty(&node
->private_list
)) {
170 node
->private_data
= mapping
;
171 list_lru_add(&workingset_shadow_nodes
, &node
->private_list
);
176 * Delete a page from the page cache and free it. Caller has to make
177 * sure the page is locked and that nobody else uses it - or that usage
178 * is safe. The caller must hold the mapping's tree_lock and
179 * mem_cgroup_begin_page_stat().
181 void __delete_from_page_cache(struct page
*page
, void *shadow
,
182 struct mem_cgroup
*memcg
)
184 struct address_space
*mapping
= page
->mapping
;
186 trace_mm_filemap_delete_from_page_cache(page
);
188 * if we're uptodate, flush out into the cleancache, otherwise
189 * invalidate any existing cleancache entries. We can't leave
190 * stale data around in the cleancache once our page is gone
192 if (PageUptodate(page
) && PageMappedToDisk(page
))
193 cleancache_put_page(page
);
195 cleancache_invalidate_page(mapping
, page
);
197 page_cache_tree_delete(mapping
, page
, shadow
);
199 page
->mapping
= NULL
;
200 /* Leave page->index set: truncation lookup relies upon it */
202 __dec_zone_page_state(page
, NR_FILE_PAGES
);
203 if (PageSwapBacked(page
))
204 __dec_zone_page_state(page
, NR_SHMEM
);
205 BUG_ON(page_mapped(page
));
208 * At this point page must be either written or cleaned by truncate.
209 * Dirty page here signals a bug and loss of unwritten data.
211 * This fixes dirty accounting after removing the page entirely but
212 * leaves PageDirty set: it has no effect for truncated page and
213 * anyway will be cleared before returning page into buddy allocator.
215 if (WARN_ON_ONCE(PageDirty(page
)))
216 account_page_cleaned(page
, mapping
, memcg
,
217 inode_to_wb(mapping
->host
));
221 * delete_from_page_cache - delete page from page cache
222 * @page: the page which the kernel is trying to remove from page cache
224 * This must be called only on pages that have been verified to be in the page
225 * cache and locked. It will never put the page into the free list, the caller
226 * has a reference on the page.
228 void delete_from_page_cache(struct page
*page
)
230 struct address_space
*mapping
= page
->mapping
;
231 struct mem_cgroup
*memcg
;
234 void (*freepage
)(struct page
*);
236 BUG_ON(!PageLocked(page
));
238 freepage
= mapping
->a_ops
->freepage
;
240 memcg
= mem_cgroup_begin_page_stat(page
);
241 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
242 __delete_from_page_cache(page
, NULL
, memcg
);
243 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
244 mem_cgroup_end_page_stat(memcg
);
248 page_cache_release(page
);
250 EXPORT_SYMBOL(delete_from_page_cache
);
252 static int filemap_check_errors(struct address_space
*mapping
)
255 /* Check for outstanding write errors */
256 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
257 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
259 if (test_bit(AS_EIO
, &mapping
->flags
) &&
260 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
266 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
267 * @mapping: address space structure to write
268 * @start: offset in bytes where the range starts
269 * @end: offset in bytes where the range ends (inclusive)
270 * @sync_mode: enable synchronous operation
272 * Start writeback against all of a mapping's dirty pages that lie
273 * within the byte offsets <start, end> inclusive.
275 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
276 * opposed to a regular memory cleansing writeback. The difference between
277 * these two operations is that if a dirty page/buffer is encountered, it must
278 * be waited upon, and not just skipped over.
280 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
281 loff_t end
, int sync_mode
)
284 struct writeback_control wbc
= {
285 .sync_mode
= sync_mode
,
286 .nr_to_write
= LONG_MAX
,
287 .range_start
= start
,
291 if (!mapping_cap_writeback_dirty(mapping
))
294 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
295 ret
= do_writepages(mapping
, &wbc
);
296 wbc_detach_inode(&wbc
);
300 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
303 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
306 int filemap_fdatawrite(struct address_space
*mapping
)
308 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
310 EXPORT_SYMBOL(filemap_fdatawrite
);
312 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
315 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
317 EXPORT_SYMBOL(filemap_fdatawrite_range
);
320 * filemap_flush - mostly a non-blocking flush
321 * @mapping: target address_space
323 * This is a mostly non-blocking flush. Not suitable for data-integrity
324 * purposes - I/O may not be started against all dirty pages.
326 int filemap_flush(struct address_space
*mapping
)
328 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
330 EXPORT_SYMBOL(filemap_flush
);
333 * filemap_fdatawait_range - wait for writeback to complete
334 * @mapping: address space structure to wait for
335 * @start_byte: offset in bytes where the range starts
336 * @end_byte: offset in bytes where the range ends (inclusive)
338 * Walk the list of under-writeback pages of the given address space
339 * in the given range and wait for all of them.
341 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
344 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
345 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
350 if (end_byte
< start_byte
)
353 pagevec_init(&pvec
, 0);
354 while ((index
<= end
) &&
355 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
356 PAGECACHE_TAG_WRITEBACK
,
357 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
360 for (i
= 0; i
< nr_pages
; i
++) {
361 struct page
*page
= pvec
.pages
[i
];
363 /* until radix tree lookup accepts end_index */
364 if (page
->index
> end
)
367 wait_on_page_writeback(page
);
368 if (TestClearPageError(page
))
371 pagevec_release(&pvec
);
375 ret2
= filemap_check_errors(mapping
);
381 EXPORT_SYMBOL(filemap_fdatawait_range
);
384 * filemap_fdatawait - wait for all under-writeback pages to complete
385 * @mapping: address space structure to wait for
387 * Walk the list of under-writeback pages of the given address space
388 * and wait for all of them.
390 int filemap_fdatawait(struct address_space
*mapping
)
392 loff_t i_size
= i_size_read(mapping
->host
);
397 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
399 EXPORT_SYMBOL(filemap_fdatawait
);
401 int filemap_write_and_wait(struct address_space
*mapping
)
405 if (mapping
->nrpages
) {
406 err
= filemap_fdatawrite(mapping
);
408 * Even if the above returned error, the pages may be
409 * written partially (e.g. -ENOSPC), so we wait for it.
410 * But the -EIO is special case, it may indicate the worst
411 * thing (e.g. bug) happened, so we avoid waiting for it.
414 int err2
= filemap_fdatawait(mapping
);
419 err
= filemap_check_errors(mapping
);
423 EXPORT_SYMBOL(filemap_write_and_wait
);
426 * filemap_write_and_wait_range - write out & wait on a file range
427 * @mapping: the address_space for the pages
428 * @lstart: offset in bytes where the range starts
429 * @lend: offset in bytes where the range ends (inclusive)
431 * Write out and wait upon file offsets lstart->lend, inclusive.
433 * Note that `lend' is inclusive (describes the last byte to be written) so
434 * that this function can be used to write to the very end-of-file (end = -1).
436 int filemap_write_and_wait_range(struct address_space
*mapping
,
437 loff_t lstart
, loff_t lend
)
441 if (mapping
->nrpages
) {
442 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
444 /* See comment of filemap_write_and_wait() */
446 int err2
= filemap_fdatawait_range(mapping
,
452 err
= filemap_check_errors(mapping
);
456 EXPORT_SYMBOL(filemap_write_and_wait_range
);
459 * replace_page_cache_page - replace a pagecache page with a new one
460 * @old: page to be replaced
461 * @new: page to replace with
462 * @gfp_mask: allocation mode
464 * This function replaces a page in the pagecache with a new one. On
465 * success it acquires the pagecache reference for the new page and
466 * drops it for the old page. Both the old and new pages must be
467 * locked. This function does not add the new page to the LRU, the
468 * caller must do that.
470 * The remove + add is atomic. The only way this function can fail is
471 * memory allocation failure.
473 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
477 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
478 VM_BUG_ON_PAGE(!PageLocked(new), new);
479 VM_BUG_ON_PAGE(new->mapping
, new);
481 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
483 struct address_space
*mapping
= old
->mapping
;
484 void (*freepage
)(struct page
*);
485 struct mem_cgroup
*memcg
;
488 pgoff_t offset
= old
->index
;
489 freepage
= mapping
->a_ops
->freepage
;
492 new->mapping
= mapping
;
495 memcg
= mem_cgroup_begin_page_stat(old
);
496 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
497 __delete_from_page_cache(old
, NULL
, memcg
);
498 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
501 __inc_zone_page_state(new, NR_FILE_PAGES
);
502 if (PageSwapBacked(new))
503 __inc_zone_page_state(new, NR_SHMEM
);
504 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
505 mem_cgroup_end_page_stat(memcg
);
506 mem_cgroup_migrate(old
, new, true);
507 radix_tree_preload_end();
510 page_cache_release(old
);
515 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
517 static int page_cache_tree_insert(struct address_space
*mapping
,
518 struct page
*page
, void **shadowp
)
520 struct radix_tree_node
*node
;
524 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
,
531 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
532 if (!radix_tree_exceptional_entry(p
))
536 mapping
->nrshadows
--;
538 workingset_node_shadows_dec(node
);
540 radix_tree_replace_slot(slot
, page
);
543 workingset_node_pages_inc(node
);
545 * Don't track node that contains actual pages.
547 * Avoid acquiring the list_lru lock if already
548 * untracked. The list_empty() test is safe as
549 * node->private_list is protected by
550 * mapping->tree_lock.
552 if (!list_empty(&node
->private_list
))
553 list_lru_del(&workingset_shadow_nodes
,
554 &node
->private_list
);
559 static int __add_to_page_cache_locked(struct page
*page
,
560 struct address_space
*mapping
,
561 pgoff_t offset
, gfp_t gfp_mask
,
564 int huge
= PageHuge(page
);
565 struct mem_cgroup
*memcg
;
568 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
569 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
572 error
= mem_cgroup_try_charge(page
, current
->mm
,
578 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
581 mem_cgroup_cancel_charge(page
, memcg
);
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
);
597 mem_cgroup_commit_charge(page
, memcg
, false);
598 trace_mm_filemap_add_to_page_cache(page
);
601 page
->mapping
= NULL
;
602 /* Leave page->index set: truncation relies upon it */
603 spin_unlock_irq(&mapping
->tree_lock
);
605 mem_cgroup_cancel_charge(page
, memcg
);
606 page_cache_release(page
);
611 * add_to_page_cache_locked - add a locked page to the pagecache
613 * @mapping: the page's address_space
614 * @offset: page index
615 * @gfp_mask: page allocation mode
617 * This function is used to add a page to the pagecache. It must be locked.
618 * This function does not add the page to the LRU. The caller must do that.
620 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
621 pgoff_t offset
, gfp_t gfp_mask
)
623 return __add_to_page_cache_locked(page
, mapping
, offset
,
626 EXPORT_SYMBOL(add_to_page_cache_locked
);
628 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
629 pgoff_t offset
, gfp_t gfp_mask
)
634 __set_page_locked(page
);
635 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
638 __clear_page_locked(page
);
641 * The page might have been evicted from cache only
642 * recently, in which case it should be activated like
643 * any other repeatedly accessed page.
645 if (shadow
&& workingset_refault(shadow
)) {
647 workingset_activation(page
);
649 ClearPageActive(page
);
654 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
657 struct page
*__page_cache_alloc(gfp_t gfp
)
662 if (cpuset_do_page_mem_spread()) {
663 unsigned int cpuset_mems_cookie
;
665 cpuset_mems_cookie
= read_mems_allowed_begin();
666 n
= cpuset_mem_spread_node();
667 page
= alloc_pages_exact_node(n
, gfp
, 0);
668 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
672 return alloc_pages(gfp
, 0);
674 EXPORT_SYMBOL(__page_cache_alloc
);
678 * In order to wait for pages to become available there must be
679 * waitqueues associated with pages. By using a hash table of
680 * waitqueues where the bucket discipline is to maintain all
681 * waiters on the same queue and wake all when any of the pages
682 * become available, and for the woken contexts to check to be
683 * sure the appropriate page became available, this saves space
684 * at a cost of "thundering herd" phenomena during rare hash
687 wait_queue_head_t
*page_waitqueue(struct page
*page
)
689 const struct zone
*zone
= page_zone(page
);
691 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
693 EXPORT_SYMBOL(page_waitqueue
);
695 void wait_on_page_bit(struct page
*page
, int bit_nr
)
697 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
699 if (test_bit(bit_nr
, &page
->flags
))
700 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
701 TASK_UNINTERRUPTIBLE
);
703 EXPORT_SYMBOL(wait_on_page_bit
);
705 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
707 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
709 if (!test_bit(bit_nr
, &page
->flags
))
712 return __wait_on_bit(page_waitqueue(page
), &wait
,
713 bit_wait_io
, TASK_KILLABLE
);
716 int wait_on_page_bit_killable_timeout(struct page
*page
,
717 int bit_nr
, unsigned long timeout
)
719 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
721 wait
.key
.timeout
= jiffies
+ timeout
;
722 if (!test_bit(bit_nr
, &page
->flags
))
724 return __wait_on_bit(page_waitqueue(page
), &wait
,
725 bit_wait_io_timeout
, TASK_KILLABLE
);
727 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
730 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
731 * @page: Page defining the wait queue of interest
732 * @waiter: Waiter to add to the queue
734 * Add an arbitrary @waiter to the wait queue for the nominated @page.
736 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
738 wait_queue_head_t
*q
= page_waitqueue(page
);
741 spin_lock_irqsave(&q
->lock
, flags
);
742 __add_wait_queue(q
, waiter
);
743 spin_unlock_irqrestore(&q
->lock
, flags
);
745 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
748 * unlock_page - unlock a locked page
751 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
752 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
753 * mechanism between PageLocked pages and PageWriteback pages is shared.
754 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
756 * The mb is necessary to enforce ordering between the clear_bit and the read
757 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
759 void unlock_page(struct page
*page
)
761 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
762 clear_bit_unlock(PG_locked
, &page
->flags
);
763 smp_mb__after_atomic();
764 wake_up_page(page
, PG_locked
);
766 EXPORT_SYMBOL(unlock_page
);
769 * end_page_writeback - end writeback against a page
772 void end_page_writeback(struct page
*page
)
775 * TestClearPageReclaim could be used here but it is an atomic
776 * operation and overkill in this particular case. Failing to
777 * shuffle a page marked for immediate reclaim is too mild to
778 * justify taking an atomic operation penalty at the end of
779 * ever page writeback.
781 if (PageReclaim(page
)) {
782 ClearPageReclaim(page
);
783 rotate_reclaimable_page(page
);
786 if (!test_clear_page_writeback(page
))
789 smp_mb__after_atomic();
790 wake_up_page(page
, PG_writeback
);
792 EXPORT_SYMBOL(end_page_writeback
);
795 * After completing I/O on a page, call this routine to update the page
796 * flags appropriately
798 void page_endio(struct page
*page
, int rw
, int err
)
802 SetPageUptodate(page
);
804 ClearPageUptodate(page
);
808 } else { /* rw == WRITE */
812 mapping_set_error(page
->mapping
, err
);
814 end_page_writeback(page
);
817 EXPORT_SYMBOL_GPL(page_endio
);
820 * __lock_page - get a lock on the page, assuming we need to sleep to get it
821 * @page: the page to lock
823 void __lock_page(struct page
*page
)
825 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
827 __wait_on_bit_lock(page_waitqueue(page
), &wait
, bit_wait_io
,
828 TASK_UNINTERRUPTIBLE
);
830 EXPORT_SYMBOL(__lock_page
);
832 int __lock_page_killable(struct page
*page
)
834 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
836 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
837 bit_wait_io
, TASK_KILLABLE
);
839 EXPORT_SYMBOL_GPL(__lock_page_killable
);
843 * 1 - page is locked; mmap_sem is still held.
844 * 0 - page is not locked.
845 * mmap_sem has been released (up_read()), unless flags had both
846 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
847 * which case mmap_sem is still held.
849 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
850 * with the page locked and the mmap_sem unperturbed.
852 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
855 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
857 * CAUTION! In this case, mmap_sem is not released
858 * even though return 0.
860 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
863 up_read(&mm
->mmap_sem
);
864 if (flags
& FAULT_FLAG_KILLABLE
)
865 wait_on_page_locked_killable(page
);
867 wait_on_page_locked(page
);
870 if (flags
& FAULT_FLAG_KILLABLE
) {
873 ret
= __lock_page_killable(page
);
875 up_read(&mm
->mmap_sem
);
885 * page_cache_next_hole - find the next hole (not-present entry)
888 * @max_scan: maximum range to search
890 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
891 * lowest indexed hole.
893 * Returns: the index of the hole if found, otherwise returns an index
894 * outside of the set specified (in which case 'return - index >=
895 * max_scan' will be true). In rare cases of index wrap-around, 0 will
898 * page_cache_next_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 5, then subsequently a hole is created at
902 * index 10, page_cache_next_hole covering both indexes may return 10
903 * if called under rcu_read_lock.
905 pgoff_t
page_cache_next_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
))
923 EXPORT_SYMBOL(page_cache_next_hole
);
926 * page_cache_prev_hole - find the prev hole (not-present entry)
929 * @max_scan: maximum range to search
931 * Search backwards in the range [max(index-max_scan+1, 0), index] for
934 * Returns: the index of the hole if found, otherwise returns an index
935 * outside of the set specified (in which case 'index - return >=
936 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
939 * page_cache_prev_hole may be called under rcu_read_lock. However,
940 * like radix_tree_gang_lookup, this will not atomically search a
941 * snapshot of the tree at a single point in time. For example, if a
942 * hole is created at index 10, then subsequently a hole is created at
943 * index 5, page_cache_prev_hole covering both indexes may return 5 if
944 * called under rcu_read_lock.
946 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
947 pgoff_t index
, unsigned long max_scan
)
951 for (i
= 0; i
< max_scan
; i
++) {
954 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
955 if (!page
|| radix_tree_exceptional_entry(page
))
958 if (index
== ULONG_MAX
)
964 EXPORT_SYMBOL(page_cache_prev_hole
);
967 * find_get_entry - find and get a page cache entry
968 * @mapping: the address_space to search
969 * @offset: the page cache index
971 * Looks up the page cache slot at @mapping & @offset. If there is a
972 * page cache page, it is returned with an increased refcount.
974 * If the slot holds a shadow entry of a previously evicted page, or a
975 * swap entry from shmem/tmpfs, it is returned.
977 * Otherwise, %NULL is returned.
979 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
987 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
989 page
= radix_tree_deref_slot(pagep
);
992 if (radix_tree_exception(page
)) {
993 if (radix_tree_deref_retry(page
))
996 * A shadow entry of a recently evicted page,
997 * or a swap entry from shmem/tmpfs. Return
998 * it without attempting to raise page count.
1002 if (!page_cache_get_speculative(page
))
1006 * Has the page moved?
1007 * This is part of the lockless pagecache protocol. See
1008 * include/linux/pagemap.h for details.
1010 if (unlikely(page
!= *pagep
)) {
1011 page_cache_release(page
);
1020 EXPORT_SYMBOL(find_get_entry
);
1023 * find_lock_entry - locate, pin and lock a page cache entry
1024 * @mapping: the address_space to search
1025 * @offset: the page cache index
1027 * Looks up the page cache slot at @mapping & @offset. If there is a
1028 * page cache page, it is returned locked and with an increased
1031 * If the slot holds a shadow entry of a previously evicted page, or a
1032 * swap entry from shmem/tmpfs, it is returned.
1034 * Otherwise, %NULL is returned.
1036 * find_lock_entry() may sleep.
1038 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1043 page
= find_get_entry(mapping
, offset
);
1044 if (page
&& !radix_tree_exception(page
)) {
1046 /* Has the page been truncated? */
1047 if (unlikely(page
->mapping
!= mapping
)) {
1049 page_cache_release(page
);
1052 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1056 EXPORT_SYMBOL(find_lock_entry
);
1059 * pagecache_get_page - find and get a page reference
1060 * @mapping: the address_space to search
1061 * @offset: the page index
1062 * @fgp_flags: PCG flags
1063 * @gfp_mask: gfp mask to use for the page cache data page allocation
1065 * Looks up the page cache slot at @mapping & @offset.
1067 * PCG flags modify how the page is returned.
1069 * FGP_ACCESSED: the page will be marked accessed
1070 * FGP_LOCK: Page is return locked
1071 * FGP_CREAT: If page is not present then a new page is allocated using
1072 * @gfp_mask and added to the page cache and the VM's LRU
1073 * list. The page is returned locked and with an increased
1074 * refcount. Otherwise, %NULL is returned.
1076 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1077 * if the GFP flags specified for FGP_CREAT are atomic.
1079 * If there is a page cache page, it is returned with an increased refcount.
1081 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1082 int fgp_flags
, gfp_t gfp_mask
)
1087 page
= find_get_entry(mapping
, offset
);
1088 if (radix_tree_exceptional_entry(page
))
1093 if (fgp_flags
& FGP_LOCK
) {
1094 if (fgp_flags
& FGP_NOWAIT
) {
1095 if (!trylock_page(page
)) {
1096 page_cache_release(page
);
1103 /* Has the page been truncated? */
1104 if (unlikely(page
->mapping
!= mapping
)) {
1106 page_cache_release(page
);
1109 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1112 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1113 mark_page_accessed(page
);
1116 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1118 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1119 gfp_mask
|= __GFP_WRITE
;
1120 if (fgp_flags
& FGP_NOFS
)
1121 gfp_mask
&= ~__GFP_FS
;
1123 page
= __page_cache_alloc(gfp_mask
);
1127 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1128 fgp_flags
|= FGP_LOCK
;
1130 /* Init accessed so avoid atomic mark_page_accessed later */
1131 if (fgp_flags
& FGP_ACCESSED
)
1132 __SetPageReferenced(page
);
1134 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1135 gfp_mask
& GFP_RECLAIM_MASK
);
1136 if (unlikely(err
)) {
1137 page_cache_release(page
);
1146 EXPORT_SYMBOL(pagecache_get_page
);
1149 * find_get_entries - gang pagecache lookup
1150 * @mapping: The address_space to search
1151 * @start: The starting page cache index
1152 * @nr_entries: The maximum number of entries
1153 * @entries: Where the resulting entries are placed
1154 * @indices: The cache indices corresponding to the entries in @entries
1156 * find_get_entries() will search for and return a group of up to
1157 * @nr_entries entries in the mapping. The entries are placed at
1158 * @entries. find_get_entries() takes a reference against any actual
1161 * The search returns a group of mapping-contiguous page cache entries
1162 * with ascending indexes. There may be holes in the indices due to
1163 * not-present pages.
1165 * Any shadow entries of evicted pages, or swap entries from
1166 * shmem/tmpfs, are included in the returned array.
1168 * find_get_entries() returns the number of pages and shadow entries
1171 unsigned find_get_entries(struct address_space
*mapping
,
1172 pgoff_t start
, unsigned int nr_entries
,
1173 struct page
**entries
, pgoff_t
*indices
)
1176 unsigned int ret
= 0;
1177 struct radix_tree_iter iter
;
1184 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1187 page
= radix_tree_deref_slot(slot
);
1188 if (unlikely(!page
))
1190 if (radix_tree_exception(page
)) {
1191 if (radix_tree_deref_retry(page
))
1194 * A shadow entry of a recently evicted page,
1195 * or a swap entry from shmem/tmpfs. Return
1196 * it without attempting to raise page count.
1200 if (!page_cache_get_speculative(page
))
1203 /* Has the page moved? */
1204 if (unlikely(page
!= *slot
)) {
1205 page_cache_release(page
);
1209 indices
[ret
] = iter
.index
;
1210 entries
[ret
] = page
;
1211 if (++ret
== nr_entries
)
1219 * find_get_pages - gang pagecache lookup
1220 * @mapping: The address_space to search
1221 * @start: The starting page index
1222 * @nr_pages: The maximum number of pages
1223 * @pages: Where the resulting pages are placed
1225 * find_get_pages() will search for and return a group of up to
1226 * @nr_pages pages in the mapping. The pages are placed at @pages.
1227 * find_get_pages() takes a reference against the returned pages.
1229 * The search returns a group of mapping-contiguous pages with ascending
1230 * indexes. There may be holes in the indices due to not-present pages.
1232 * find_get_pages() returns the number of pages which were found.
1234 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1235 unsigned int nr_pages
, struct page
**pages
)
1237 struct radix_tree_iter iter
;
1241 if (unlikely(!nr_pages
))
1246 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1249 page
= radix_tree_deref_slot(slot
);
1250 if (unlikely(!page
))
1253 if (radix_tree_exception(page
)) {
1254 if (radix_tree_deref_retry(page
)) {
1256 * Transient condition which can only trigger
1257 * when entry at index 0 moves out of or back
1258 * to root: none yet gotten, safe to restart.
1260 WARN_ON(iter
.index
);
1264 * A shadow entry of a recently evicted page,
1265 * or a swap entry from shmem/tmpfs. Skip
1271 if (!page_cache_get_speculative(page
))
1274 /* Has the page moved? */
1275 if (unlikely(page
!= *slot
)) {
1276 page_cache_release(page
);
1281 if (++ret
== nr_pages
)
1290 * find_get_pages_contig - gang contiguous pagecache lookup
1291 * @mapping: The address_space to search
1292 * @index: The starting page index
1293 * @nr_pages: The maximum number of pages
1294 * @pages: Where the resulting pages are placed
1296 * find_get_pages_contig() works exactly like find_get_pages(), except
1297 * that the returned number of pages are guaranteed to be contiguous.
1299 * find_get_pages_contig() returns the number of pages which were found.
1301 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1302 unsigned int nr_pages
, struct page
**pages
)
1304 struct radix_tree_iter iter
;
1306 unsigned int ret
= 0;
1308 if (unlikely(!nr_pages
))
1313 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1316 page
= radix_tree_deref_slot(slot
);
1317 /* The hole, there no reason to continue */
1318 if (unlikely(!page
))
1321 if (radix_tree_exception(page
)) {
1322 if (radix_tree_deref_retry(page
)) {
1324 * Transient condition which can only trigger
1325 * when entry at index 0 moves out of or back
1326 * to root: none yet gotten, safe to restart.
1331 * A shadow entry of a recently evicted page,
1332 * or a swap entry from shmem/tmpfs. Stop
1333 * looking for contiguous pages.
1338 if (!page_cache_get_speculative(page
))
1341 /* Has the page moved? */
1342 if (unlikely(page
!= *slot
)) {
1343 page_cache_release(page
);
1348 * must check mapping and index after taking the ref.
1349 * otherwise we can get both false positives and false
1350 * negatives, which is just confusing to the caller.
1352 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1353 page_cache_release(page
);
1358 if (++ret
== nr_pages
)
1364 EXPORT_SYMBOL(find_get_pages_contig
);
1367 * find_get_pages_tag - find and return pages that match @tag
1368 * @mapping: the address_space to search
1369 * @index: the starting page index
1370 * @tag: the tag index
1371 * @nr_pages: the maximum number of pages
1372 * @pages: where the resulting pages are placed
1374 * Like find_get_pages, except we only return pages which are tagged with
1375 * @tag. We update @index to index the next page for the traversal.
1377 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1378 int tag
, unsigned int nr_pages
, struct page
**pages
)
1380 struct radix_tree_iter iter
;
1384 if (unlikely(!nr_pages
))
1389 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1390 &iter
, *index
, tag
) {
1393 page
= radix_tree_deref_slot(slot
);
1394 if (unlikely(!page
))
1397 if (radix_tree_exception(page
)) {
1398 if (radix_tree_deref_retry(page
)) {
1400 * Transient condition which can only trigger
1401 * when entry at index 0 moves out of or back
1402 * to root: none yet gotten, safe to restart.
1407 * A shadow entry of a recently evicted page.
1409 * Those entries should never be tagged, but
1410 * this tree walk is lockless and the tags are
1411 * looked up in bulk, one radix tree node at a
1412 * time, so there is a sizable window for page
1413 * reclaim to evict a page we saw tagged.
1420 if (!page_cache_get_speculative(page
))
1423 /* Has the page moved? */
1424 if (unlikely(page
!= *slot
)) {
1425 page_cache_release(page
);
1430 if (++ret
== nr_pages
)
1437 *index
= pages
[ret
- 1]->index
+ 1;
1441 EXPORT_SYMBOL(find_get_pages_tag
);
1444 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1445 * a _large_ part of the i/o request. Imagine the worst scenario:
1447 * ---R__________________________________________B__________
1448 * ^ reading here ^ bad block(assume 4k)
1450 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1451 * => failing the whole request => read(R) => read(R+1) =>
1452 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1453 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1454 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1456 * It is going insane. Fix it by quickly scaling down the readahead size.
1458 static void shrink_readahead_size_eio(struct file
*filp
,
1459 struct file_ra_state
*ra
)
1465 * do_generic_file_read - generic file read routine
1466 * @filp: the file to read
1467 * @ppos: current file position
1468 * @iter: data destination
1469 * @written: already copied
1471 * This is a generic file read routine, and uses the
1472 * mapping->a_ops->readpage() function for the actual low-level stuff.
1474 * This is really ugly. But the goto's actually try to clarify some
1475 * of the logic when it comes to error handling etc.
1477 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1478 struct iov_iter
*iter
, ssize_t written
)
1480 struct address_space
*mapping
= filp
->f_mapping
;
1481 struct inode
*inode
= mapping
->host
;
1482 struct file_ra_state
*ra
= &filp
->f_ra
;
1486 unsigned long offset
; /* offset into pagecache page */
1487 unsigned int prev_offset
;
1490 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1491 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1492 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1493 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1494 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1500 unsigned long nr
, ret
;
1504 page
= find_get_page(mapping
, index
);
1506 page_cache_sync_readahead(mapping
,
1508 index
, last_index
- index
);
1509 page
= find_get_page(mapping
, index
);
1510 if (unlikely(page
== NULL
))
1511 goto no_cached_page
;
1513 if (PageReadahead(page
)) {
1514 page_cache_async_readahead(mapping
,
1516 index
, last_index
- index
);
1518 if (!PageUptodate(page
)) {
1519 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1520 !mapping
->a_ops
->is_partially_uptodate
)
1521 goto page_not_up_to_date
;
1522 if (!trylock_page(page
))
1523 goto page_not_up_to_date
;
1524 /* Did it get truncated before we got the lock? */
1526 goto page_not_up_to_date_locked
;
1527 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1528 offset
, iter
->count
))
1529 goto page_not_up_to_date_locked
;
1534 * i_size must be checked after we know the page is Uptodate.
1536 * Checking i_size after the check allows us to calculate
1537 * the correct value for "nr", which means the zero-filled
1538 * part of the page is not copied back to userspace (unless
1539 * another truncate extends the file - this is desired though).
1542 isize
= i_size_read(inode
);
1543 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1544 if (unlikely(!isize
|| index
> end_index
)) {
1545 page_cache_release(page
);
1549 /* nr is the maximum number of bytes to copy from this page */
1550 nr
= PAGE_CACHE_SIZE
;
1551 if (index
== end_index
) {
1552 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1554 page_cache_release(page
);
1560 /* If users can be writing to this page using arbitrary
1561 * virtual addresses, take care about potential aliasing
1562 * before reading the page on the kernel side.
1564 if (mapping_writably_mapped(mapping
))
1565 flush_dcache_page(page
);
1568 * When a sequential read accesses a page several times,
1569 * only mark it as accessed the first time.
1571 if (prev_index
!= index
|| offset
!= prev_offset
)
1572 mark_page_accessed(page
);
1576 * Ok, we have the page, and it's up-to-date, so
1577 * now we can copy it to user space...
1580 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1582 index
+= offset
>> PAGE_CACHE_SHIFT
;
1583 offset
&= ~PAGE_CACHE_MASK
;
1584 prev_offset
= offset
;
1586 page_cache_release(page
);
1588 if (!iov_iter_count(iter
))
1596 page_not_up_to_date
:
1597 /* Get exclusive access to the page ... */
1598 error
= lock_page_killable(page
);
1599 if (unlikely(error
))
1600 goto readpage_error
;
1602 page_not_up_to_date_locked
:
1603 /* Did it get truncated before we got the lock? */
1604 if (!page
->mapping
) {
1606 page_cache_release(page
);
1610 /* Did somebody else fill it already? */
1611 if (PageUptodate(page
)) {
1618 * A previous I/O error may have been due to temporary
1619 * failures, eg. multipath errors.
1620 * PG_error will be set again if readpage fails.
1622 ClearPageError(page
);
1623 /* Start the actual read. The read will unlock the page. */
1624 error
= mapping
->a_ops
->readpage(filp
, page
);
1626 if (unlikely(error
)) {
1627 if (error
== AOP_TRUNCATED_PAGE
) {
1628 page_cache_release(page
);
1632 goto readpage_error
;
1635 if (!PageUptodate(page
)) {
1636 error
= lock_page_killable(page
);
1637 if (unlikely(error
))
1638 goto readpage_error
;
1639 if (!PageUptodate(page
)) {
1640 if (page
->mapping
== NULL
) {
1642 * invalidate_mapping_pages got it
1645 page_cache_release(page
);
1649 shrink_readahead_size_eio(filp
, ra
);
1651 goto readpage_error
;
1659 /* UHHUH! A synchronous read error occurred. Report it */
1660 page_cache_release(page
);
1665 * Ok, it wasn't cached, so we need to create a new
1668 page
= page_cache_alloc_cold(mapping
);
1673 error
= add_to_page_cache_lru(page
, mapping
,
1676 page_cache_release(page
);
1677 if (error
== -EEXIST
) {
1687 ra
->prev_pos
= prev_index
;
1688 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1689 ra
->prev_pos
|= prev_offset
;
1691 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1692 file_accessed(filp
);
1693 return written
? written
: error
;
1697 * generic_file_read_iter - generic filesystem read routine
1698 * @iocb: kernel I/O control block
1699 * @iter: destination for the data read
1701 * This is the "read_iter()" routine for all filesystems
1702 * that can use the page cache directly.
1705 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1707 struct file
*file
= iocb
->ki_filp
;
1709 loff_t
*ppos
= &iocb
->ki_pos
;
1712 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1713 struct address_space
*mapping
= file
->f_mapping
;
1714 struct inode
*inode
= mapping
->host
;
1715 size_t count
= iov_iter_count(iter
);
1719 goto out
; /* skip atime */
1720 size
= i_size_read(inode
);
1721 retval
= filemap_write_and_wait_range(mapping
, pos
,
1724 struct iov_iter data
= *iter
;
1725 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
1729 *ppos
= pos
+ retval
;
1730 iov_iter_advance(iter
, retval
);
1734 * Btrfs can have a short DIO read if we encounter
1735 * compressed extents, so if there was an error, or if
1736 * we've already read everything we wanted to, or if
1737 * there was a short read because we hit EOF, go ahead
1738 * and return. Otherwise fallthrough to buffered io for
1739 * the rest of the read. Buffered reads will not work for
1740 * DAX files, so don't bother trying.
1742 if (retval
< 0 || !iov_iter_count(iter
) || *ppos
>= size
||
1744 file_accessed(file
);
1749 retval
= do_generic_file_read(file
, ppos
, iter
, retval
);
1753 EXPORT_SYMBOL(generic_file_read_iter
);
1757 * page_cache_read - adds requested page to the page cache if not already there
1758 * @file: file to read
1759 * @offset: page index
1761 * This adds the requested page to the page cache if it isn't already there,
1762 * and schedules an I/O to read in its contents from disk.
1764 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1766 struct address_space
*mapping
= file
->f_mapping
;
1771 page
= page_cache_alloc_cold(mapping
);
1775 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1777 ret
= mapping
->a_ops
->readpage(file
, page
);
1778 else if (ret
== -EEXIST
)
1779 ret
= 0; /* losing race to add is OK */
1781 page_cache_release(page
);
1783 } while (ret
== AOP_TRUNCATED_PAGE
);
1788 #define MMAP_LOTSAMISS (100)
1791 * Synchronous readahead happens when we don't even find
1792 * a page in the page cache at all.
1794 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1795 struct file_ra_state
*ra
,
1799 unsigned long ra_pages
;
1800 struct address_space
*mapping
= file
->f_mapping
;
1802 /* If we don't want any read-ahead, don't bother */
1803 if (vma
->vm_flags
& VM_RAND_READ
)
1808 if (vma
->vm_flags
& VM_SEQ_READ
) {
1809 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1814 /* Avoid banging the cache line if not needed */
1815 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1819 * Do we miss much more than hit in this file? If so,
1820 * stop bothering with read-ahead. It will only hurt.
1822 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1828 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1829 ra
->start
= max_t(long, 0, offset
- ra_pages
/ 2);
1830 ra
->size
= ra_pages
;
1831 ra
->async_size
= ra_pages
/ 4;
1832 ra_submit(ra
, mapping
, file
);
1836 * Asynchronous readahead happens when we find the page and PG_readahead,
1837 * so we want to possibly extend the readahead further..
1839 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1840 struct file_ra_state
*ra
,
1845 struct address_space
*mapping
= file
->f_mapping
;
1847 /* If we don't want any read-ahead, don't bother */
1848 if (vma
->vm_flags
& VM_RAND_READ
)
1850 if (ra
->mmap_miss
> 0)
1852 if (PageReadahead(page
))
1853 page_cache_async_readahead(mapping
, ra
, file
,
1854 page
, offset
, ra
->ra_pages
);
1858 * filemap_fault - read in file data for page fault handling
1859 * @vma: vma in which the fault was taken
1860 * @vmf: struct vm_fault containing details of the fault
1862 * filemap_fault() is invoked via the vma operations vector for a
1863 * mapped memory region to read in file data during a page fault.
1865 * The goto's are kind of ugly, but this streamlines the normal case of having
1866 * it in the page cache, and handles the special cases reasonably without
1867 * having a lot of duplicated code.
1869 * vma->vm_mm->mmap_sem must be held on entry.
1871 * If our return value has VM_FAULT_RETRY set, it's because
1872 * lock_page_or_retry() returned 0.
1873 * The mmap_sem has usually been released in this case.
1874 * See __lock_page_or_retry() for the exception.
1876 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1877 * has not been released.
1879 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1881 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1884 struct file
*file
= vma
->vm_file
;
1885 struct address_space
*mapping
= file
->f_mapping
;
1886 struct file_ra_state
*ra
= &file
->f_ra
;
1887 struct inode
*inode
= mapping
->host
;
1888 pgoff_t offset
= vmf
->pgoff
;
1893 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1894 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
1895 return VM_FAULT_SIGBUS
;
1898 * Do we have something in the page cache already?
1900 page
= find_get_page(mapping
, offset
);
1901 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1903 * We found the page, so try async readahead before
1904 * waiting for the lock.
1906 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1908 /* No page in the page cache at all */
1909 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1910 count_vm_event(PGMAJFAULT
);
1911 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1912 ret
= VM_FAULT_MAJOR
;
1914 page
= find_get_page(mapping
, offset
);
1916 goto no_cached_page
;
1919 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1920 page_cache_release(page
);
1921 return ret
| VM_FAULT_RETRY
;
1924 /* Did it get truncated? */
1925 if (unlikely(page
->mapping
!= mapping
)) {
1930 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1933 * We have a locked page in the page cache, now we need to check
1934 * that it's up-to-date. If not, it is going to be due to an error.
1936 if (unlikely(!PageUptodate(page
)))
1937 goto page_not_uptodate
;
1940 * Found the page and have a reference on it.
1941 * We must recheck i_size under page lock.
1943 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1944 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
1946 page_cache_release(page
);
1947 return VM_FAULT_SIGBUS
;
1951 return ret
| VM_FAULT_LOCKED
;
1955 * We're only likely to ever get here if MADV_RANDOM is in
1958 error
= page_cache_read(file
, offset
);
1961 * The page we want has now been added to the page cache.
1962 * In the unlikely event that someone removed it in the
1963 * meantime, we'll just come back here and read it again.
1969 * An error return from page_cache_read can result if the
1970 * system is low on memory, or a problem occurs while trying
1973 if (error
== -ENOMEM
)
1974 return VM_FAULT_OOM
;
1975 return VM_FAULT_SIGBUS
;
1979 * Umm, take care of errors if the page isn't up-to-date.
1980 * Try to re-read it _once_. We do this synchronously,
1981 * because there really aren't any performance issues here
1982 * and we need to check for errors.
1984 ClearPageError(page
);
1985 error
= mapping
->a_ops
->readpage(file
, page
);
1987 wait_on_page_locked(page
);
1988 if (!PageUptodate(page
))
1991 page_cache_release(page
);
1993 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1996 /* Things didn't work out. Return zero to tell the mm layer so. */
1997 shrink_readahead_size_eio(file
, ra
);
1998 return VM_FAULT_SIGBUS
;
2000 EXPORT_SYMBOL(filemap_fault
);
2002 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2004 struct radix_tree_iter iter
;
2006 struct file
*file
= vma
->vm_file
;
2007 struct address_space
*mapping
= file
->f_mapping
;
2010 unsigned long address
= (unsigned long) vmf
->virtual_address
;
2015 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
2016 if (iter
.index
> vmf
->max_pgoff
)
2019 page
= radix_tree_deref_slot(slot
);
2020 if (unlikely(!page
))
2022 if (radix_tree_exception(page
)) {
2023 if (radix_tree_deref_retry(page
))
2029 if (!page_cache_get_speculative(page
))
2032 /* Has the page moved? */
2033 if (unlikely(page
!= *slot
)) {
2034 page_cache_release(page
);
2038 if (!PageUptodate(page
) ||
2039 PageReadahead(page
) ||
2042 if (!trylock_page(page
))
2045 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2048 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
2049 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2052 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2053 if (!pte_none(*pte
))
2056 if (file
->f_ra
.mmap_miss
> 0)
2057 file
->f_ra
.mmap_miss
--;
2058 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2059 do_set_pte(vma
, addr
, page
, pte
, false, false);
2065 page_cache_release(page
);
2067 if (iter
.index
== vmf
->max_pgoff
)
2072 EXPORT_SYMBOL(filemap_map_pages
);
2074 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2076 struct page
*page
= vmf
->page
;
2077 struct inode
*inode
= file_inode(vma
->vm_file
);
2078 int ret
= VM_FAULT_LOCKED
;
2080 sb_start_pagefault(inode
->i_sb
);
2081 file_update_time(vma
->vm_file
);
2083 if (page
->mapping
!= inode
->i_mapping
) {
2085 ret
= VM_FAULT_NOPAGE
;
2089 * We mark the page dirty already here so that when freeze is in
2090 * progress, we are guaranteed that writeback during freezing will
2091 * see the dirty page and writeprotect it again.
2093 set_page_dirty(page
);
2094 wait_for_stable_page(page
);
2096 sb_end_pagefault(inode
->i_sb
);
2099 EXPORT_SYMBOL(filemap_page_mkwrite
);
2101 const struct vm_operations_struct generic_file_vm_ops
= {
2102 .fault
= filemap_fault
,
2103 .map_pages
= filemap_map_pages
,
2104 .page_mkwrite
= filemap_page_mkwrite
,
2107 /* This is used for a general mmap of a disk file */
2109 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2111 struct address_space
*mapping
= file
->f_mapping
;
2113 if (!mapping
->a_ops
->readpage
)
2115 file_accessed(file
);
2116 vma
->vm_ops
= &generic_file_vm_ops
;
2121 * This is for filesystems which do not implement ->writepage.
2123 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2125 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2127 return generic_file_mmap(file
, vma
);
2130 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2134 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2138 #endif /* CONFIG_MMU */
2140 EXPORT_SYMBOL(generic_file_mmap
);
2141 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2143 static struct page
*wait_on_page_read(struct page
*page
)
2145 if (!IS_ERR(page
)) {
2146 wait_on_page_locked(page
);
2147 if (!PageUptodate(page
)) {
2148 page_cache_release(page
);
2149 page
= ERR_PTR(-EIO
);
2155 static struct page
*__read_cache_page(struct address_space
*mapping
,
2157 int (*filler
)(void *, struct page
*),
2164 page
= find_get_page(mapping
, index
);
2166 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2168 return ERR_PTR(-ENOMEM
);
2169 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2170 if (unlikely(err
)) {
2171 page_cache_release(page
);
2174 /* Presumably ENOMEM for radix tree node */
2175 return ERR_PTR(err
);
2177 err
= filler(data
, page
);
2179 page_cache_release(page
);
2180 page
= ERR_PTR(err
);
2182 page
= wait_on_page_read(page
);
2188 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2190 int (*filler
)(void *, struct page
*),
2199 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2202 if (PageUptodate(page
))
2206 if (!page
->mapping
) {
2208 page_cache_release(page
);
2211 if (PageUptodate(page
)) {
2215 err
= filler(data
, page
);
2217 page_cache_release(page
);
2218 return ERR_PTR(err
);
2220 page
= wait_on_page_read(page
);
2225 mark_page_accessed(page
);
2230 * read_cache_page - read into page cache, fill it if needed
2231 * @mapping: the page's address_space
2232 * @index: the page index
2233 * @filler: function to perform the read
2234 * @data: first arg to filler(data, page) function, often left as NULL
2236 * Read into the page cache. If a page already exists, and PageUptodate() is
2237 * not set, try to fill the page and wait for it to become unlocked.
2239 * If the page does not get brought uptodate, return -EIO.
2241 struct page
*read_cache_page(struct address_space
*mapping
,
2243 int (*filler
)(void *, struct page
*),
2246 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2248 EXPORT_SYMBOL(read_cache_page
);
2251 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2252 * @mapping: the page's address_space
2253 * @index: the page index
2254 * @gfp: the page allocator flags to use if allocating
2256 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2257 * any new page allocations done using the specified allocation flags.
2259 * If the page does not get brought uptodate, return -EIO.
2261 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2265 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2267 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2269 EXPORT_SYMBOL(read_cache_page_gfp
);
2272 * Performs necessary checks before doing a write
2274 * Can adjust writing position or amount of bytes to write.
2275 * Returns appropriate error code that caller should return or
2276 * zero in case that write should be allowed.
2278 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2280 struct file
*file
= iocb
->ki_filp
;
2281 struct inode
*inode
= file
->f_mapping
->host
;
2282 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2285 if (!iov_iter_count(from
))
2288 /* FIXME: this is for backwards compatibility with 2.4 */
2289 if (iocb
->ki_flags
& IOCB_APPEND
)
2290 iocb
->ki_pos
= i_size_read(inode
);
2294 if (limit
!= RLIM_INFINITY
) {
2295 if (iocb
->ki_pos
>= limit
) {
2296 send_sig(SIGXFSZ
, current
, 0);
2299 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2305 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2306 !(file
->f_flags
& O_LARGEFILE
))) {
2307 if (pos
>= MAX_NON_LFS
)
2309 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2313 * Are we about to exceed the fs block limit ?
2315 * If we have written data it becomes a short write. If we have
2316 * exceeded without writing data we send a signal and return EFBIG.
2317 * Linus frestrict idea will clean these up nicely..
2319 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2322 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2323 return iov_iter_count(from
);
2325 EXPORT_SYMBOL(generic_write_checks
);
2327 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2328 loff_t pos
, unsigned len
, unsigned flags
,
2329 struct page
**pagep
, void **fsdata
)
2331 const struct address_space_operations
*aops
= mapping
->a_ops
;
2333 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2336 EXPORT_SYMBOL(pagecache_write_begin
);
2338 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2339 loff_t pos
, unsigned len
, unsigned copied
,
2340 struct page
*page
, void *fsdata
)
2342 const struct address_space_operations
*aops
= mapping
->a_ops
;
2344 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2346 EXPORT_SYMBOL(pagecache_write_end
);
2349 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
2351 struct file
*file
= iocb
->ki_filp
;
2352 struct address_space
*mapping
= file
->f_mapping
;
2353 struct inode
*inode
= mapping
->host
;
2357 struct iov_iter data
;
2359 write_len
= iov_iter_count(from
);
2360 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2362 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2367 * After a write we want buffered reads to be sure to go to disk to get
2368 * the new data. We invalidate clean cached page from the region we're
2369 * about to write. We do this *before* the write so that we can return
2370 * without clobbering -EIOCBQUEUED from ->direct_IO().
2372 if (mapping
->nrpages
) {
2373 written
= invalidate_inode_pages2_range(mapping
,
2374 pos
>> PAGE_CACHE_SHIFT
, end
);
2376 * If a page can not be invalidated, return 0 to fall back
2377 * to buffered write.
2380 if (written
== -EBUSY
)
2387 written
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
2390 * Finally, try again to invalidate clean pages which might have been
2391 * cached by non-direct readahead, or faulted in by get_user_pages()
2392 * if the source of the write was an mmap'ed region of the file
2393 * we're writing. Either one is a pretty crazy thing to do,
2394 * so we don't support it 100%. If this invalidation
2395 * fails, tough, the write still worked...
2397 if (mapping
->nrpages
) {
2398 invalidate_inode_pages2_range(mapping
,
2399 pos
>> PAGE_CACHE_SHIFT
, end
);
2404 iov_iter_advance(from
, written
);
2405 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2406 i_size_write(inode
, pos
);
2407 mark_inode_dirty(inode
);
2414 EXPORT_SYMBOL(generic_file_direct_write
);
2417 * Find or create a page at the given pagecache position. Return the locked
2418 * page. This function is specifically for buffered writes.
2420 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2421 pgoff_t index
, unsigned flags
)
2424 int fgp_flags
= FGP_LOCK
|FGP_ACCESSED
|FGP_WRITE
|FGP_CREAT
;
2426 if (flags
& AOP_FLAG_NOFS
)
2427 fgp_flags
|= FGP_NOFS
;
2429 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2430 mapping_gfp_mask(mapping
));
2432 wait_for_stable_page(page
);
2436 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2438 ssize_t
generic_perform_write(struct file
*file
,
2439 struct iov_iter
*i
, loff_t pos
)
2441 struct address_space
*mapping
= file
->f_mapping
;
2442 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2444 ssize_t written
= 0;
2445 unsigned int flags
= 0;
2448 * Copies from kernel address space cannot fail (NFSD is a big user).
2450 if (!iter_is_iovec(i
))
2451 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2455 unsigned long offset
; /* Offset into pagecache page */
2456 unsigned long bytes
; /* Bytes to write to page */
2457 size_t copied
; /* Bytes copied from user */
2460 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2461 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2466 * Bring in the user page that we will copy from _first_.
2467 * Otherwise there's a nasty deadlock on copying from the
2468 * same page as we're writing to, without it being marked
2471 * Not only is this an optimisation, but it is also required
2472 * to check that the address is actually valid, when atomic
2473 * usercopies are used, below.
2475 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2480 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2482 if (unlikely(status
< 0))
2485 if (mapping_writably_mapped(mapping
))
2486 flush_dcache_page(page
);
2488 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2489 flush_dcache_page(page
);
2491 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2493 if (unlikely(status
< 0))
2499 iov_iter_advance(i
, copied
);
2500 if (unlikely(copied
== 0)) {
2502 * If we were unable to copy any data at all, we must
2503 * fall back to a single segment length write.
2505 * If we didn't fallback here, we could livelock
2506 * because not all segments in the iov can be copied at
2507 * once without a pagefault.
2509 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2510 iov_iter_single_seg_count(i
));
2516 balance_dirty_pages_ratelimited(mapping
);
2517 if (fatal_signal_pending(current
)) {
2521 } while (iov_iter_count(i
));
2523 return written
? written
: status
;
2525 EXPORT_SYMBOL(generic_perform_write
);
2528 * __generic_file_write_iter - write data to a file
2529 * @iocb: IO state structure (file, offset, etc.)
2530 * @from: iov_iter with data to write
2532 * This function does all the work needed for actually writing data to a
2533 * file. It does all basic checks, removes SUID from the file, updates
2534 * modification times and calls proper subroutines depending on whether we
2535 * do direct IO or a standard buffered write.
2537 * It expects i_mutex to be grabbed unless we work on a block device or similar
2538 * object which does not need locking at all.
2540 * This function does *not* take care of syncing data in case of O_SYNC write.
2541 * A caller has to handle it. This is mainly due to the fact that we want to
2542 * avoid syncing under i_mutex.
2544 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2546 struct file
*file
= iocb
->ki_filp
;
2547 struct address_space
* mapping
= file
->f_mapping
;
2548 struct inode
*inode
= mapping
->host
;
2549 ssize_t written
= 0;
2553 /* We can write back this queue in page reclaim */
2554 current
->backing_dev_info
= inode_to_bdi(inode
);
2555 err
= file_remove_suid(file
);
2559 err
= file_update_time(file
);
2563 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2564 loff_t pos
, endbyte
;
2566 written
= generic_file_direct_write(iocb
, from
, iocb
->ki_pos
);
2568 * If the write stopped short of completing, fall back to
2569 * buffered writes. Some filesystems do this for writes to
2570 * holes, for example. For DAX files, a buffered write will
2571 * not succeed (even if it did, DAX does not handle dirty
2572 * page-cache pages correctly).
2574 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2577 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2579 * If generic_perform_write() returned a synchronous error
2580 * then we want to return the number of bytes which were
2581 * direct-written, or the error code if that was zero. Note
2582 * that this differs from normal direct-io semantics, which
2583 * will return -EFOO even if some bytes were written.
2585 if (unlikely(status
< 0)) {
2590 * We need to ensure that the page cache pages are written to
2591 * disk and invalidated to preserve the expected O_DIRECT
2594 endbyte
= pos
+ status
- 1;
2595 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2597 iocb
->ki_pos
= endbyte
+ 1;
2599 invalidate_mapping_pages(mapping
,
2600 pos
>> PAGE_CACHE_SHIFT
,
2601 endbyte
>> PAGE_CACHE_SHIFT
);
2604 * We don't know how much we wrote, so just return
2605 * the number of bytes which were direct-written
2609 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2610 if (likely(written
> 0))
2611 iocb
->ki_pos
+= written
;
2614 current
->backing_dev_info
= NULL
;
2615 return written
? written
: err
;
2617 EXPORT_SYMBOL(__generic_file_write_iter
);
2620 * generic_file_write_iter - write data to a file
2621 * @iocb: IO state structure
2622 * @from: iov_iter with data to write
2624 * This is a wrapper around __generic_file_write_iter() to be used by most
2625 * filesystems. It takes care of syncing the file in case of O_SYNC file
2626 * and acquires i_mutex as needed.
2628 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2630 struct file
*file
= iocb
->ki_filp
;
2631 struct inode
*inode
= file
->f_mapping
->host
;
2634 mutex_lock(&inode
->i_mutex
);
2635 ret
= generic_write_checks(iocb
, from
);
2637 ret
= __generic_file_write_iter(iocb
, from
);
2638 mutex_unlock(&inode
->i_mutex
);
2643 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2649 EXPORT_SYMBOL(generic_file_write_iter
);
2652 * try_to_release_page() - release old fs-specific metadata on a page
2654 * @page: the page which the kernel is trying to free
2655 * @gfp_mask: memory allocation flags (and I/O mode)
2657 * The address_space is to try to release any data against the page
2658 * (presumably at page->private). If the release was successful, return `1'.
2659 * Otherwise return zero.
2661 * This may also be called if PG_fscache is set on a page, indicating that the
2662 * page is known to the local caching routines.
2664 * The @gfp_mask argument specifies whether I/O may be performed to release
2665 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2668 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2670 struct address_space
* const mapping
= page
->mapping
;
2672 BUG_ON(!PageLocked(page
));
2673 if (PageWriteback(page
))
2676 if (mapping
&& mapping
->a_ops
->releasepage
)
2677 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2678 return try_to_free_buffers(page
);
2681 EXPORT_SYMBOL(try_to_release_page
);