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_file_buffered_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
84 * sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
104 * ->inode->i_lock (zap_pte_range->set_page_dirty)
105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 static void page_cache_tree_delete(struct address_space
*mapping
,
112 struct page
*page
, void *shadow
)
114 struct radix_tree_node
*node
;
120 VM_BUG_ON(!PageLocked(page
));
122 __radix_tree_lookup(&mapping
->page_tree
, page
->index
, &node
, &slot
);
125 mapping
->nrshadows
++;
127 * Make sure the nrshadows update is committed before
128 * the nrpages update so that final truncate racing
129 * with reclaim does not see both counters 0 at the
130 * same time and miss a shadow entry.
137 /* Clear direct pointer tags in root node */
138 mapping
->page_tree
.gfp_mask
&= __GFP_BITS_MASK
;
139 radix_tree_replace_slot(slot
, shadow
);
143 /* Clear tree tags for the removed page */
145 offset
= index
& RADIX_TREE_MAP_MASK
;
146 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
147 if (test_bit(offset
, node
->tags
[tag
]))
148 radix_tree_tag_clear(&mapping
->page_tree
, index
, tag
);
151 /* Delete page, swap shadow entry */
152 radix_tree_replace_slot(slot
, shadow
);
153 workingset_node_pages_dec(node
);
155 workingset_node_shadows_inc(node
);
157 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
161 * Track node that only contains shadow entries.
163 * Avoid acquiring the list_lru lock if already tracked. The
164 * list_empty() test is safe as node->private_list is
165 * protected by mapping->tree_lock.
167 if (!workingset_node_pages(node
) &&
168 list_empty(&node
->private_list
)) {
169 node
->private_data
= mapping
;
170 list_lru_add(&workingset_shadow_nodes
, &node
->private_list
);
175 * Delete a page from the page cache and free it. Caller has to make
176 * sure the page is locked and that nobody else uses it - or that usage
177 * is safe. The caller must hold the mapping's tree_lock.
179 void __delete_from_page_cache(struct page
*page
, void *shadow
)
181 struct address_space
*mapping
= page
->mapping
;
183 trace_mm_filemap_delete_from_page_cache(page
);
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
189 if (PageUptodate(page
) && PageMappedToDisk(page
))
190 cleancache_put_page(page
);
192 cleancache_invalidate_page(mapping
, page
);
194 page_cache_tree_delete(mapping
, page
, shadow
);
196 page
->mapping
= NULL
;
197 /* Leave page->index set: truncation lookup relies upon it */
199 __dec_zone_page_state(page
, NR_FILE_PAGES
);
200 if (PageSwapBacked(page
))
201 __dec_zone_page_state(page
, NR_SHMEM
);
202 BUG_ON(page_mapped(page
));
205 * Some filesystems seem to re-dirty the page even after
206 * the VM has canceled the dirty bit (eg ext3 journaling).
208 * Fix it up by doing a final dirty accounting check after
209 * having removed the page entirely.
211 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
212 dec_zone_page_state(page
, NR_FILE_DIRTY
);
213 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
218 * delete_from_page_cache - delete page from page cache
219 * @page: the page which the kernel is trying to remove from page cache
221 * This must be called only on pages that have been verified to be in the page
222 * cache and locked. It will never put the page into the free list, the caller
223 * has a reference on the page.
225 void delete_from_page_cache(struct page
*page
)
227 struct address_space
*mapping
= page
->mapping
;
228 void (*freepage
)(struct page
*);
230 BUG_ON(!PageLocked(page
));
232 freepage
= mapping
->a_ops
->freepage
;
233 spin_lock_irq(&mapping
->tree_lock
);
234 __delete_from_page_cache(page
, NULL
);
235 spin_unlock_irq(&mapping
->tree_lock
);
236 mem_cgroup_uncharge_cache_page(page
);
240 page_cache_release(page
);
242 EXPORT_SYMBOL(delete_from_page_cache
);
244 static int sleep_on_page(void *word
)
250 static int sleep_on_page_killable(void *word
)
253 return fatal_signal_pending(current
) ? -EINTR
: 0;
256 static int filemap_check_errors(struct address_space
*mapping
)
259 /* Check for outstanding write errors */
260 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
262 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
268 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
269 * @mapping: address space structure to write
270 * @start: offset in bytes where the range starts
271 * @end: offset in bytes where the range ends (inclusive)
272 * @sync_mode: enable synchronous operation
274 * Start writeback against all of a mapping's dirty pages that lie
275 * within the byte offsets <start, end> inclusive.
277 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
278 * opposed to a regular memory cleansing writeback. The difference between
279 * these two operations is that if a dirty page/buffer is encountered, it must
280 * be waited upon, and not just skipped over.
282 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
283 loff_t end
, int sync_mode
)
286 struct writeback_control wbc
= {
287 .sync_mode
= sync_mode
,
288 .nr_to_write
= LONG_MAX
,
289 .range_start
= start
,
293 if (!mapping_cap_writeback_dirty(mapping
))
296 ret
= do_writepages(mapping
, &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
*);
486 pgoff_t offset
= old
->index
;
487 freepage
= mapping
->a_ops
->freepage
;
490 new->mapping
= mapping
;
493 spin_lock_irq(&mapping
->tree_lock
);
494 __delete_from_page_cache(old
, NULL
);
495 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
498 __inc_zone_page_state(new, NR_FILE_PAGES
);
499 if (PageSwapBacked(new))
500 __inc_zone_page_state(new, NR_SHMEM
);
501 spin_unlock_irq(&mapping
->tree_lock
);
502 /* mem_cgroup codes must not be called under tree_lock */
503 mem_cgroup_replace_page_cache(old
, new);
504 radix_tree_preload_end();
507 page_cache_release(old
);
512 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
514 static int page_cache_tree_insert(struct address_space
*mapping
,
515 struct page
*page
, void **shadowp
)
517 struct radix_tree_node
*node
;
521 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
,
528 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
529 if (!radix_tree_exceptional_entry(p
))
533 mapping
->nrshadows
--;
535 workingset_node_shadows_dec(node
);
537 radix_tree_replace_slot(slot
, page
);
540 workingset_node_pages_inc(node
);
542 * Don't track node that contains actual pages.
544 * Avoid acquiring the list_lru lock if already
545 * untracked. The list_empty() test is safe as
546 * node->private_list is protected by
547 * mapping->tree_lock.
549 if (!list_empty(&node
->private_list
))
550 list_lru_del(&workingset_shadow_nodes
,
551 &node
->private_list
);
556 static int __add_to_page_cache_locked(struct page
*page
,
557 struct address_space
*mapping
,
558 pgoff_t offset
, gfp_t gfp_mask
,
563 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
564 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
566 error
= mem_cgroup_charge_file(page
, current
->mm
,
567 gfp_mask
& GFP_RECLAIM_MASK
);
571 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
573 mem_cgroup_uncharge_cache_page(page
);
577 page_cache_get(page
);
578 page
->mapping
= mapping
;
579 page
->index
= offset
;
581 spin_lock_irq(&mapping
->tree_lock
);
582 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
583 radix_tree_preload_end();
586 __inc_zone_page_state(page
, NR_FILE_PAGES
);
587 spin_unlock_irq(&mapping
->tree_lock
);
588 trace_mm_filemap_add_to_page_cache(page
);
591 page
->mapping
= NULL
;
592 /* Leave page->index set: truncation relies upon it */
593 spin_unlock_irq(&mapping
->tree_lock
);
594 mem_cgroup_uncharge_cache_page(page
);
595 page_cache_release(page
);
600 * add_to_page_cache_locked - add a locked page to the pagecache
602 * @mapping: the page's address_space
603 * @offset: page index
604 * @gfp_mask: page allocation mode
606 * This function is used to add a page to the pagecache. It must be locked.
607 * This function does not add the page to the LRU. The caller must do that.
609 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
610 pgoff_t offset
, gfp_t gfp_mask
)
612 return __add_to_page_cache_locked(page
, mapping
, offset
,
615 EXPORT_SYMBOL(add_to_page_cache_locked
);
617 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
618 pgoff_t offset
, gfp_t gfp_mask
)
623 __set_page_locked(page
);
624 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
627 __clear_page_locked(page
);
630 * The page might have been evicted from cache only
631 * recently, in which case it should be activated like
632 * any other repeatedly accessed page.
634 if (shadow
&& workingset_refault(shadow
)) {
636 workingset_activation(page
);
638 ClearPageActive(page
);
643 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
646 struct page
*__page_cache_alloc(gfp_t gfp
)
651 if (cpuset_do_page_mem_spread()) {
652 unsigned int cpuset_mems_cookie
;
654 cpuset_mems_cookie
= read_mems_allowed_begin();
655 n
= cpuset_mem_spread_node();
656 page
= alloc_pages_exact_node(n
, gfp
, 0);
657 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
661 return alloc_pages(gfp
, 0);
663 EXPORT_SYMBOL(__page_cache_alloc
);
667 * In order to wait for pages to become available there must be
668 * waitqueues associated with pages. By using a hash table of
669 * waitqueues where the bucket discipline is to maintain all
670 * waiters on the same queue and wake all when any of the pages
671 * become available, and for the woken contexts to check to be
672 * sure the appropriate page became available, this saves space
673 * at a cost of "thundering herd" phenomena during rare hash
676 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
678 const struct zone
*zone
= page_zone(page
);
680 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
683 static inline void wake_up_page(struct page
*page
, int bit
)
685 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
688 void wait_on_page_bit(struct page
*page
, int bit_nr
)
690 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
692 if (test_bit(bit_nr
, &page
->flags
))
693 __wait_on_bit(page_waitqueue(page
), &wait
, sleep_on_page
,
694 TASK_UNINTERRUPTIBLE
);
696 EXPORT_SYMBOL(wait_on_page_bit
);
698 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
700 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
702 if (!test_bit(bit_nr
, &page
->flags
))
705 return __wait_on_bit(page_waitqueue(page
), &wait
,
706 sleep_on_page_killable
, TASK_KILLABLE
);
710 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
711 * @page: Page defining the wait queue of interest
712 * @waiter: Waiter to add to the queue
714 * Add an arbitrary @waiter to the wait queue for the nominated @page.
716 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
718 wait_queue_head_t
*q
= page_waitqueue(page
);
721 spin_lock_irqsave(&q
->lock
, flags
);
722 __add_wait_queue(q
, waiter
);
723 spin_unlock_irqrestore(&q
->lock
, flags
);
725 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
728 * unlock_page - unlock a locked page
731 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
732 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
733 * mechananism between PageLocked pages and PageWriteback pages is shared.
734 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
736 * The mb is necessary to enforce ordering between the clear_bit and the read
737 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
739 void unlock_page(struct page
*page
)
741 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
742 clear_bit_unlock(PG_locked
, &page
->flags
);
743 smp_mb__after_clear_bit();
744 wake_up_page(page
, PG_locked
);
746 EXPORT_SYMBOL(unlock_page
);
749 * end_page_writeback - end writeback against a page
752 void end_page_writeback(struct page
*page
)
754 if (TestClearPageReclaim(page
))
755 rotate_reclaimable_page(page
);
757 if (!test_clear_page_writeback(page
))
760 smp_mb__after_clear_bit();
761 wake_up_page(page
, PG_writeback
);
763 EXPORT_SYMBOL(end_page_writeback
);
766 * __lock_page - get a lock on the page, assuming we need to sleep to get it
767 * @page: the page to lock
769 void __lock_page(struct page
*page
)
771 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
773 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sleep_on_page
,
774 TASK_UNINTERRUPTIBLE
);
776 EXPORT_SYMBOL(__lock_page
);
778 int __lock_page_killable(struct page
*page
)
780 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
782 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
783 sleep_on_page_killable
, TASK_KILLABLE
);
785 EXPORT_SYMBOL_GPL(__lock_page_killable
);
787 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
790 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
792 * CAUTION! In this case, mmap_sem is not released
793 * even though return 0.
795 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
798 up_read(&mm
->mmap_sem
);
799 if (flags
& FAULT_FLAG_KILLABLE
)
800 wait_on_page_locked_killable(page
);
802 wait_on_page_locked(page
);
805 if (flags
& FAULT_FLAG_KILLABLE
) {
808 ret
= __lock_page_killable(page
);
810 up_read(&mm
->mmap_sem
);
820 * page_cache_next_hole - find the next hole (not-present entry)
823 * @max_scan: maximum range to search
825 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
826 * lowest indexed hole.
828 * Returns: the index of the hole if found, otherwise returns an index
829 * outside of the set specified (in which case 'return - index >=
830 * max_scan' will be true). In rare cases of index wrap-around, 0 will
833 * page_cache_next_hole may be called under rcu_read_lock. However,
834 * like radix_tree_gang_lookup, this will not atomically search a
835 * snapshot of the tree at a single point in time. For example, if a
836 * hole is created at index 5, then subsequently a hole is created at
837 * index 10, page_cache_next_hole covering both indexes may return 10
838 * if called under rcu_read_lock.
840 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
841 pgoff_t index
, unsigned long max_scan
)
845 for (i
= 0; i
< max_scan
; i
++) {
848 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
849 if (!page
|| radix_tree_exceptional_entry(page
))
858 EXPORT_SYMBOL(page_cache_next_hole
);
861 * page_cache_prev_hole - find the prev hole (not-present entry)
864 * @max_scan: maximum range to search
866 * Search backwards in the range [max(index-max_scan+1, 0), index] for
869 * Returns: the index of the hole if found, otherwise returns an index
870 * outside of the set specified (in which case 'index - return >=
871 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
874 * page_cache_prev_hole may be called under rcu_read_lock. However,
875 * like radix_tree_gang_lookup, this will not atomically search a
876 * snapshot of the tree at a single point in time. For example, if a
877 * hole is created at index 10, then subsequently a hole is created at
878 * index 5, page_cache_prev_hole covering both indexes may return 5 if
879 * called under rcu_read_lock.
881 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
882 pgoff_t index
, unsigned long max_scan
)
886 for (i
= 0; i
< max_scan
; i
++) {
889 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
890 if (!page
|| radix_tree_exceptional_entry(page
))
893 if (index
== ULONG_MAX
)
899 EXPORT_SYMBOL(page_cache_prev_hole
);
902 * find_get_entry - find and get a page cache entry
903 * @mapping: the address_space to search
904 * @offset: the page cache index
906 * Looks up the page cache slot at @mapping & @offset. If there is a
907 * page cache page, it is returned with an increased refcount.
909 * If the slot holds a shadow entry of a previously evicted page, it
912 * Otherwise, %NULL is returned.
914 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
922 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
924 page
= radix_tree_deref_slot(pagep
);
927 if (radix_tree_exception(page
)) {
928 if (radix_tree_deref_retry(page
))
931 * Otherwise, shmem/tmpfs must be storing a swap entry
932 * here as an exceptional entry: so return it without
933 * attempting to raise page count.
937 if (!page_cache_get_speculative(page
))
941 * Has the page moved?
942 * This is part of the lockless pagecache protocol. See
943 * include/linux/pagemap.h for details.
945 if (unlikely(page
!= *pagep
)) {
946 page_cache_release(page
);
955 EXPORT_SYMBOL(find_get_entry
);
958 * find_get_page - find and get a page reference
959 * @mapping: the address_space to search
960 * @offset: the page index
962 * Looks up the page cache slot at @mapping & @offset. If there is a
963 * page cache page, it is returned with an increased refcount.
965 * Otherwise, %NULL is returned.
967 struct page
*find_get_page(struct address_space
*mapping
, pgoff_t offset
)
969 struct page
*page
= find_get_entry(mapping
, offset
);
971 if (radix_tree_exceptional_entry(page
))
975 EXPORT_SYMBOL(find_get_page
);
978 * find_lock_entry - locate, pin and lock a page cache entry
979 * @mapping: the address_space to search
980 * @offset: the page cache index
982 * Looks up the page cache slot at @mapping & @offset. If there is a
983 * page cache page, it is returned locked and with an increased
986 * If the slot holds a shadow entry of a previously evicted page, it
989 * Otherwise, %NULL is returned.
991 * find_lock_entry() may sleep.
993 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
998 page
= find_get_entry(mapping
, offset
);
999 if (page
&& !radix_tree_exception(page
)) {
1001 /* Has the page been truncated? */
1002 if (unlikely(page
->mapping
!= mapping
)) {
1004 page_cache_release(page
);
1007 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1011 EXPORT_SYMBOL(find_lock_entry
);
1014 * find_lock_page - locate, pin and lock a pagecache page
1015 * @mapping: the address_space to search
1016 * @offset: the page index
1018 * Looks up the page cache slot at @mapping & @offset. If there is a
1019 * page cache page, it is returned locked and with an increased
1022 * Otherwise, %NULL is returned.
1024 * find_lock_page() may sleep.
1026 struct page
*find_lock_page(struct address_space
*mapping
, pgoff_t offset
)
1028 struct page
*page
= find_lock_entry(mapping
, offset
);
1030 if (radix_tree_exceptional_entry(page
))
1034 EXPORT_SYMBOL(find_lock_page
);
1037 * find_or_create_page - locate or add a pagecache page
1038 * @mapping: the page's address_space
1039 * @index: the page's index into the mapping
1040 * @gfp_mask: page allocation mode
1042 * Looks up the page cache slot at @mapping & @offset. If there is a
1043 * page cache page, it is returned locked and with an increased
1046 * If the page is not present, a new page is allocated using @gfp_mask
1047 * and added to the page cache and the VM's LRU list. The page is
1048 * returned locked and with an increased refcount.
1050 * On memory exhaustion, %NULL is returned.
1052 * find_or_create_page() may sleep, even if @gfp_flags specifies an
1053 * atomic allocation!
1055 struct page
*find_or_create_page(struct address_space
*mapping
,
1056 pgoff_t index
, gfp_t gfp_mask
)
1061 page
= find_lock_page(mapping
, index
);
1063 page
= __page_cache_alloc(gfp_mask
);
1067 * We want a regular kernel memory (not highmem or DMA etc)
1068 * allocation for the radix tree nodes, but we need to honour
1069 * the context-specific requirements the caller has asked for.
1070 * GFP_RECLAIM_MASK collects those requirements.
1072 err
= add_to_page_cache_lru(page
, mapping
, index
,
1073 (gfp_mask
& GFP_RECLAIM_MASK
));
1074 if (unlikely(err
)) {
1075 page_cache_release(page
);
1083 EXPORT_SYMBOL(find_or_create_page
);
1086 * find_get_entries - gang pagecache lookup
1087 * @mapping: The address_space to search
1088 * @start: The starting page cache index
1089 * @nr_entries: The maximum number of entries
1090 * @entries: Where the resulting entries are placed
1091 * @indices: The cache indices corresponding to the entries in @entries
1093 * find_get_entries() will search for and return a group of up to
1094 * @nr_entries entries in the mapping. The entries are placed at
1095 * @entries. find_get_entries() takes a reference against any actual
1098 * The search returns a group of mapping-contiguous page cache entries
1099 * with ascending indexes. There may be holes in the indices due to
1100 * not-present pages.
1102 * Any shadow entries of evicted pages are included in the returned
1105 * find_get_entries() returns the number of pages and shadow entries
1108 unsigned find_get_entries(struct address_space
*mapping
,
1109 pgoff_t start
, unsigned int nr_entries
,
1110 struct page
**entries
, pgoff_t
*indices
)
1113 unsigned int ret
= 0;
1114 struct radix_tree_iter iter
;
1121 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1124 page
= radix_tree_deref_slot(slot
);
1125 if (unlikely(!page
))
1127 if (radix_tree_exception(page
)) {
1128 if (radix_tree_deref_retry(page
))
1131 * Otherwise, we must be storing a swap entry
1132 * here as an exceptional entry: so return it
1133 * without attempting to raise page count.
1137 if (!page_cache_get_speculative(page
))
1140 /* Has the page moved? */
1141 if (unlikely(page
!= *slot
)) {
1142 page_cache_release(page
);
1146 indices
[ret
] = iter
.index
;
1147 entries
[ret
] = page
;
1148 if (++ret
== nr_entries
)
1156 * find_get_pages - gang pagecache lookup
1157 * @mapping: The address_space to search
1158 * @start: The starting page index
1159 * @nr_pages: The maximum number of pages
1160 * @pages: Where the resulting pages are placed
1162 * find_get_pages() will search for and return a group of up to
1163 * @nr_pages pages in the mapping. The pages are placed at @pages.
1164 * find_get_pages() takes a reference against the returned pages.
1166 * The search returns a group of mapping-contiguous pages with ascending
1167 * indexes. There may be holes in the indices due to not-present pages.
1169 * find_get_pages() returns the number of pages which were found.
1171 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1172 unsigned int nr_pages
, struct page
**pages
)
1174 struct radix_tree_iter iter
;
1178 if (unlikely(!nr_pages
))
1183 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1186 page
= radix_tree_deref_slot(slot
);
1187 if (unlikely(!page
))
1190 if (radix_tree_exception(page
)) {
1191 if (radix_tree_deref_retry(page
)) {
1193 * Transient condition which can only trigger
1194 * when entry at index 0 moves out of or back
1195 * to root: none yet gotten, safe to restart.
1197 WARN_ON(iter
.index
);
1201 * Otherwise, shmem/tmpfs must be storing a swap entry
1202 * here as an exceptional entry: so skip over it -
1203 * we only reach this from invalidate_mapping_pages().
1208 if (!page_cache_get_speculative(page
))
1211 /* Has the page moved? */
1212 if (unlikely(page
!= *slot
)) {
1213 page_cache_release(page
);
1218 if (++ret
== nr_pages
)
1227 * find_get_pages_contig - gang contiguous pagecache lookup
1228 * @mapping: The address_space to search
1229 * @index: The starting page index
1230 * @nr_pages: The maximum number of pages
1231 * @pages: Where the resulting pages are placed
1233 * find_get_pages_contig() works exactly like find_get_pages(), except
1234 * that the returned number of pages are guaranteed to be contiguous.
1236 * find_get_pages_contig() returns the number of pages which were found.
1238 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1239 unsigned int nr_pages
, struct page
**pages
)
1241 struct radix_tree_iter iter
;
1243 unsigned int ret
= 0;
1245 if (unlikely(!nr_pages
))
1250 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1253 page
= radix_tree_deref_slot(slot
);
1254 /* The hole, there no reason to continue */
1255 if (unlikely(!page
))
1258 if (radix_tree_exception(page
)) {
1259 if (radix_tree_deref_retry(page
)) {
1261 * Transient condition which can only trigger
1262 * when entry at index 0 moves out of or back
1263 * to root: none yet gotten, safe to restart.
1268 * Otherwise, shmem/tmpfs must be storing a swap entry
1269 * here as an exceptional entry: so stop looking for
1275 if (!page_cache_get_speculative(page
))
1278 /* Has the page moved? */
1279 if (unlikely(page
!= *slot
)) {
1280 page_cache_release(page
);
1285 * must check mapping and index after taking the ref.
1286 * otherwise we can get both false positives and false
1287 * negatives, which is just confusing to the caller.
1289 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1290 page_cache_release(page
);
1295 if (++ret
== nr_pages
)
1301 EXPORT_SYMBOL(find_get_pages_contig
);
1304 * find_get_pages_tag - find and return pages that match @tag
1305 * @mapping: the address_space to search
1306 * @index: the starting page index
1307 * @tag: the tag index
1308 * @nr_pages: the maximum number of pages
1309 * @pages: where the resulting pages are placed
1311 * Like find_get_pages, except we only return pages which are tagged with
1312 * @tag. We update @index to index the next page for the traversal.
1314 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1315 int tag
, unsigned int nr_pages
, struct page
**pages
)
1317 struct radix_tree_iter iter
;
1321 if (unlikely(!nr_pages
))
1326 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1327 &iter
, *index
, tag
) {
1330 page
= radix_tree_deref_slot(slot
);
1331 if (unlikely(!page
))
1334 if (radix_tree_exception(page
)) {
1335 if (radix_tree_deref_retry(page
)) {
1337 * Transient condition which can only trigger
1338 * when entry at index 0 moves out of or back
1339 * to root: none yet gotten, safe to restart.
1344 * This function is never used on a shmem/tmpfs
1345 * mapping, so a swap entry won't be found here.
1350 if (!page_cache_get_speculative(page
))
1353 /* Has the page moved? */
1354 if (unlikely(page
!= *slot
)) {
1355 page_cache_release(page
);
1360 if (++ret
== nr_pages
)
1367 *index
= pages
[ret
- 1]->index
+ 1;
1371 EXPORT_SYMBOL(find_get_pages_tag
);
1374 * grab_cache_page_nowait - returns locked page at given index in given cache
1375 * @mapping: target address_space
1376 * @index: the page index
1378 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1379 * This is intended for speculative data generators, where the data can
1380 * be regenerated if the page couldn't be grabbed. This routine should
1381 * be safe to call while holding the lock for another page.
1383 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1384 * and deadlock against the caller's locked page.
1387 grab_cache_page_nowait(struct address_space
*mapping
, pgoff_t index
)
1389 struct page
*page
= find_get_page(mapping
, index
);
1392 if (trylock_page(page
))
1394 page_cache_release(page
);
1397 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
1398 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_NOFS
)) {
1399 page_cache_release(page
);
1404 EXPORT_SYMBOL(grab_cache_page_nowait
);
1407 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1408 * a _large_ part of the i/o request. Imagine the worst scenario:
1410 * ---R__________________________________________B__________
1411 * ^ reading here ^ bad block(assume 4k)
1413 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1414 * => failing the whole request => read(R) => read(R+1) =>
1415 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1416 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1417 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1419 * It is going insane. Fix it by quickly scaling down the readahead size.
1421 static void shrink_readahead_size_eio(struct file
*filp
,
1422 struct file_ra_state
*ra
)
1428 * do_generic_file_read - generic file read routine
1429 * @filp: the file to read
1430 * @ppos: current file position
1431 * @desc: read_descriptor
1433 * This is a generic file read routine, and uses the
1434 * mapping->a_ops->readpage() function for the actual low-level stuff.
1436 * This is really ugly. But the goto's actually try to clarify some
1437 * of the logic when it comes to error handling etc.
1439 static void do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1440 read_descriptor_t
*desc
)
1442 struct address_space
*mapping
= filp
->f_mapping
;
1443 struct inode
*inode
= mapping
->host
;
1444 struct file_ra_state
*ra
= &filp
->f_ra
;
1448 unsigned long offset
; /* offset into pagecache page */
1449 unsigned int prev_offset
;
1452 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1453 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1454 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1455 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1456 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1462 unsigned long nr
, ret
;
1466 page
= find_get_page(mapping
, index
);
1468 page_cache_sync_readahead(mapping
,
1470 index
, last_index
- index
);
1471 page
= find_get_page(mapping
, index
);
1472 if (unlikely(page
== NULL
))
1473 goto no_cached_page
;
1475 if (PageReadahead(page
)) {
1476 page_cache_async_readahead(mapping
,
1478 index
, last_index
- index
);
1480 if (!PageUptodate(page
)) {
1481 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1482 !mapping
->a_ops
->is_partially_uptodate
)
1483 goto page_not_up_to_date
;
1484 if (!trylock_page(page
))
1485 goto page_not_up_to_date
;
1486 /* Did it get truncated before we got the lock? */
1488 goto page_not_up_to_date_locked
;
1489 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1491 goto page_not_up_to_date_locked
;
1496 * i_size must be checked after we know the page is Uptodate.
1498 * Checking i_size after the check allows us to calculate
1499 * the correct value for "nr", which means the zero-filled
1500 * part of the page is not copied back to userspace (unless
1501 * another truncate extends the file - this is desired though).
1504 isize
= i_size_read(inode
);
1505 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1506 if (unlikely(!isize
|| index
> end_index
)) {
1507 page_cache_release(page
);
1511 /* nr is the maximum number of bytes to copy from this page */
1512 nr
= PAGE_CACHE_SIZE
;
1513 if (index
== end_index
) {
1514 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1516 page_cache_release(page
);
1522 /* If users can be writing to this page using arbitrary
1523 * virtual addresses, take care about potential aliasing
1524 * before reading the page on the kernel side.
1526 if (mapping_writably_mapped(mapping
))
1527 flush_dcache_page(page
);
1530 * When a sequential read accesses a page several times,
1531 * only mark it as accessed the first time.
1533 if (prev_index
!= index
|| offset
!= prev_offset
)
1534 mark_page_accessed(page
);
1538 * Ok, we have the page, and it's up-to-date, so
1539 * now we can copy it to user space...
1541 * The file_read_actor routine returns how many bytes were
1543 * NOTE! This may not be the same as how much of a user buffer
1544 * we filled up (we may be padding etc), so we can only update
1545 * "pos" here (the actor routine has to update the user buffer
1546 * pointers and the remaining count).
1548 ret
= file_read_actor(desc
, page
, offset
, nr
);
1550 index
+= offset
>> PAGE_CACHE_SHIFT
;
1551 offset
&= ~PAGE_CACHE_MASK
;
1552 prev_offset
= offset
;
1554 page_cache_release(page
);
1555 if (ret
== nr
&& desc
->count
)
1559 page_not_up_to_date
:
1560 /* Get exclusive access to the page ... */
1561 error
= lock_page_killable(page
);
1562 if (unlikely(error
))
1563 goto readpage_error
;
1565 page_not_up_to_date_locked
:
1566 /* Did it get truncated before we got the lock? */
1567 if (!page
->mapping
) {
1569 page_cache_release(page
);
1573 /* Did somebody else fill it already? */
1574 if (PageUptodate(page
)) {
1581 * A previous I/O error may have been due to temporary
1582 * failures, eg. multipath errors.
1583 * PG_error will be set again if readpage fails.
1585 ClearPageError(page
);
1586 /* Start the actual read. The read will unlock the page. */
1587 error
= mapping
->a_ops
->readpage(filp
, page
);
1589 if (unlikely(error
)) {
1590 if (error
== AOP_TRUNCATED_PAGE
) {
1591 page_cache_release(page
);
1594 goto readpage_error
;
1597 if (!PageUptodate(page
)) {
1598 error
= lock_page_killable(page
);
1599 if (unlikely(error
))
1600 goto readpage_error
;
1601 if (!PageUptodate(page
)) {
1602 if (page
->mapping
== NULL
) {
1604 * invalidate_mapping_pages got it
1607 page_cache_release(page
);
1611 shrink_readahead_size_eio(filp
, ra
);
1613 goto readpage_error
;
1621 /* UHHUH! A synchronous read error occurred. Report it */
1622 desc
->error
= error
;
1623 page_cache_release(page
);
1628 * Ok, it wasn't cached, so we need to create a new
1631 page
= page_cache_alloc_cold(mapping
);
1633 desc
->error
= -ENOMEM
;
1636 error
= add_to_page_cache_lru(page
, mapping
,
1639 page_cache_release(page
);
1640 if (error
== -EEXIST
)
1642 desc
->error
= error
;
1649 ra
->prev_pos
= prev_index
;
1650 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1651 ra
->prev_pos
|= prev_offset
;
1653 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1654 file_accessed(filp
);
1657 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
1658 unsigned long offset
, unsigned long size
)
1661 unsigned long left
, count
= desc
->count
;
1667 * Faults on the destination of a read are common, so do it before
1670 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
1671 kaddr
= kmap_atomic(page
);
1672 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
1673 kaddr
+ offset
, size
);
1674 kunmap_atomic(kaddr
);
1679 /* Do it the slow way */
1681 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
1686 desc
->error
= -EFAULT
;
1689 desc
->count
= count
- size
;
1690 desc
->written
+= size
;
1691 desc
->arg
.buf
+= size
;
1696 * Performs necessary checks before doing a write
1697 * @iov: io vector request
1698 * @nr_segs: number of segments in the iovec
1699 * @count: number of bytes to write
1700 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1702 * Adjust number of segments and amount of bytes to write (nr_segs should be
1703 * properly initialized first). Returns appropriate error code that caller
1704 * should return or zero in case that write should be allowed.
1706 int generic_segment_checks(const struct iovec
*iov
,
1707 unsigned long *nr_segs
, size_t *count
, int access_flags
)
1711 for (seg
= 0; seg
< *nr_segs
; seg
++) {
1712 const struct iovec
*iv
= &iov
[seg
];
1715 * If any segment has a negative length, or the cumulative
1716 * length ever wraps negative then return -EINVAL.
1719 if (unlikely((ssize_t
)(cnt
|iv
->iov_len
) < 0))
1721 if (access_ok(access_flags
, iv
->iov_base
, iv
->iov_len
))
1726 cnt
-= iv
->iov_len
; /* This segment is no good */
1732 EXPORT_SYMBOL(generic_segment_checks
);
1735 * generic_file_aio_read - generic filesystem read routine
1736 * @iocb: kernel I/O control block
1737 * @iov: io vector request
1738 * @nr_segs: number of segments in the iovec
1739 * @pos: current file position
1741 * This is the "read()" routine for all filesystems
1742 * that can use the page cache directly.
1745 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1746 unsigned long nr_segs
, loff_t pos
)
1748 struct file
*filp
= iocb
->ki_filp
;
1750 unsigned long seg
= 0;
1752 loff_t
*ppos
= &iocb
->ki_pos
;
1755 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1759 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1760 if (filp
->f_flags
& O_DIRECT
) {
1762 struct address_space
*mapping
;
1763 struct inode
*inode
;
1765 mapping
= filp
->f_mapping
;
1766 inode
= mapping
->host
;
1768 goto out
; /* skip atime */
1769 size
= i_size_read(inode
);
1770 retval
= filemap_write_and_wait_range(mapping
, pos
,
1771 pos
+ iov_length(iov
, nr_segs
) - 1);
1773 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
,
1777 *ppos
= pos
+ retval
;
1782 * Btrfs can have a short DIO read if we encounter
1783 * compressed extents, so if there was an error, or if
1784 * we've already read everything we wanted to, or if
1785 * there was a short read because we hit EOF, go ahead
1786 * and return. Otherwise fallthrough to buffered io for
1787 * the rest of the read.
1789 if (retval
< 0 || !count
|| *ppos
>= size
) {
1790 file_accessed(filp
);
1796 for (seg
= 0; seg
< nr_segs
; seg
++) {
1797 read_descriptor_t desc
;
1801 * If we did a short DIO read we need to skip the section of the
1802 * iov that we've already read data into.
1805 if (count
> iov
[seg
].iov_len
) {
1806 count
-= iov
[seg
].iov_len
;
1814 desc
.arg
.buf
= iov
[seg
].iov_base
+ offset
;
1815 desc
.count
= iov
[seg
].iov_len
- offset
;
1816 if (desc
.count
== 0)
1819 do_generic_file_read(filp
, ppos
, &desc
);
1820 retval
+= desc
.written
;
1822 retval
= retval
?: desc
.error
;
1831 EXPORT_SYMBOL(generic_file_aio_read
);
1835 * page_cache_read - adds requested page to the page cache if not already there
1836 * @file: file to read
1837 * @offset: page index
1839 * This adds the requested page to the page cache if it isn't already there,
1840 * and schedules an I/O to read in its contents from disk.
1842 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1844 struct address_space
*mapping
= file
->f_mapping
;
1849 page
= page_cache_alloc_cold(mapping
);
1853 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1855 ret
= mapping
->a_ops
->readpage(file
, page
);
1856 else if (ret
== -EEXIST
)
1857 ret
= 0; /* losing race to add is OK */
1859 page_cache_release(page
);
1861 } while (ret
== AOP_TRUNCATED_PAGE
);
1866 #define MMAP_LOTSAMISS (100)
1869 * Synchronous readahead happens when we don't even find
1870 * a page in the page cache at all.
1872 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1873 struct file_ra_state
*ra
,
1877 unsigned long ra_pages
;
1878 struct address_space
*mapping
= file
->f_mapping
;
1880 /* If we don't want any read-ahead, don't bother */
1881 if (vma
->vm_flags
& VM_RAND_READ
)
1886 if (vma
->vm_flags
& VM_SEQ_READ
) {
1887 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1892 /* Avoid banging the cache line if not needed */
1893 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1897 * Do we miss much more than hit in this file? If so,
1898 * stop bothering with read-ahead. It will only hurt.
1900 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1906 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1907 ra
->start
= max_t(long, 0, offset
- ra_pages
/ 2);
1908 ra
->size
= ra_pages
;
1909 ra
->async_size
= ra_pages
/ 4;
1910 ra_submit(ra
, mapping
, file
);
1914 * Asynchronous readahead happens when we find the page and PG_readahead,
1915 * so we want to possibly extend the readahead further..
1917 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1918 struct file_ra_state
*ra
,
1923 struct address_space
*mapping
= file
->f_mapping
;
1925 /* If we don't want any read-ahead, don't bother */
1926 if (vma
->vm_flags
& VM_RAND_READ
)
1928 if (ra
->mmap_miss
> 0)
1930 if (PageReadahead(page
))
1931 page_cache_async_readahead(mapping
, ra
, file
,
1932 page
, offset
, ra
->ra_pages
);
1936 * filemap_fault - read in file data for page fault handling
1937 * @vma: vma in which the fault was taken
1938 * @vmf: struct vm_fault containing details of the fault
1940 * filemap_fault() is invoked via the vma operations vector for a
1941 * mapped memory region to read in file data during a page fault.
1943 * The goto's are kind of ugly, but this streamlines the normal case of having
1944 * it in the page cache, and handles the special cases reasonably without
1945 * having a lot of duplicated code.
1947 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1950 struct file
*file
= vma
->vm_file
;
1951 struct address_space
*mapping
= file
->f_mapping
;
1952 struct file_ra_state
*ra
= &file
->f_ra
;
1953 struct inode
*inode
= mapping
->host
;
1954 pgoff_t offset
= vmf
->pgoff
;
1959 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1960 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
1961 return VM_FAULT_SIGBUS
;
1964 * Do we have something in the page cache already?
1966 page
= find_get_page(mapping
, offset
);
1967 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1969 * We found the page, so try async readahead before
1970 * waiting for the lock.
1972 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1974 /* No page in the page cache at all */
1975 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1976 count_vm_event(PGMAJFAULT
);
1977 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1978 ret
= VM_FAULT_MAJOR
;
1980 page
= find_get_page(mapping
, offset
);
1982 goto no_cached_page
;
1985 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1986 page_cache_release(page
);
1987 return ret
| VM_FAULT_RETRY
;
1990 /* Did it get truncated? */
1991 if (unlikely(page
->mapping
!= mapping
)) {
1996 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1999 * We have a locked page in the page cache, now we need to check
2000 * that it's up-to-date. If not, it is going to be due to an error.
2002 if (unlikely(!PageUptodate(page
)))
2003 goto page_not_uptodate
;
2006 * Found the page and have a reference on it.
2007 * We must recheck i_size under page lock.
2009 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2010 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
2012 page_cache_release(page
);
2013 return VM_FAULT_SIGBUS
;
2017 return ret
| VM_FAULT_LOCKED
;
2021 * We're only likely to ever get here if MADV_RANDOM is in
2024 error
= page_cache_read(file
, offset
);
2027 * The page we want has now been added to the page cache.
2028 * In the unlikely event that someone removed it in the
2029 * meantime, we'll just come back here and read it again.
2035 * An error return from page_cache_read can result if the
2036 * system is low on memory, or a problem occurs while trying
2039 if (error
== -ENOMEM
)
2040 return VM_FAULT_OOM
;
2041 return VM_FAULT_SIGBUS
;
2045 * Umm, take care of errors if the page isn't up-to-date.
2046 * Try to re-read it _once_. We do this synchronously,
2047 * because there really aren't any performance issues here
2048 * and we need to check for errors.
2050 ClearPageError(page
);
2051 error
= mapping
->a_ops
->readpage(file
, page
);
2053 wait_on_page_locked(page
);
2054 if (!PageUptodate(page
))
2057 page_cache_release(page
);
2059 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2062 /* Things didn't work out. Return zero to tell the mm layer so. */
2063 shrink_readahead_size_eio(file
, ra
);
2064 return VM_FAULT_SIGBUS
;
2066 EXPORT_SYMBOL(filemap_fault
);
2068 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2070 struct radix_tree_iter iter
;
2072 struct file
*file
= vma
->vm_file
;
2073 struct address_space
*mapping
= file
->f_mapping
;
2076 unsigned long address
= (unsigned long) vmf
->virtual_address
;
2081 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
2082 if (iter
.index
> vmf
->max_pgoff
)
2085 page
= radix_tree_deref_slot(slot
);
2086 if (unlikely(!page
))
2088 if (radix_tree_exception(page
)) {
2089 if (radix_tree_deref_retry(page
))
2095 if (!page_cache_get_speculative(page
))
2098 /* Has the page moved? */
2099 if (unlikely(page
!= *slot
)) {
2100 page_cache_release(page
);
2104 if (!PageUptodate(page
) ||
2105 PageReadahead(page
) ||
2108 if (!trylock_page(page
))
2111 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2114 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
2115 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2118 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2119 if (!pte_none(*pte
))
2122 if (file
->f_ra
.mmap_miss
> 0)
2123 file
->f_ra
.mmap_miss
--;
2124 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2125 do_set_pte(vma
, addr
, page
, pte
, false, false);
2131 page_cache_release(page
);
2133 if (iter
.index
== vmf
->max_pgoff
)
2138 EXPORT_SYMBOL(filemap_map_pages
);
2140 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2142 struct page
*page
= vmf
->page
;
2143 struct inode
*inode
= file_inode(vma
->vm_file
);
2144 int ret
= VM_FAULT_LOCKED
;
2146 sb_start_pagefault(inode
->i_sb
);
2147 file_update_time(vma
->vm_file
);
2149 if (page
->mapping
!= inode
->i_mapping
) {
2151 ret
= VM_FAULT_NOPAGE
;
2155 * We mark the page dirty already here so that when freeze is in
2156 * progress, we are guaranteed that writeback during freezing will
2157 * see the dirty page and writeprotect it again.
2159 set_page_dirty(page
);
2160 wait_for_stable_page(page
);
2162 sb_end_pagefault(inode
->i_sb
);
2165 EXPORT_SYMBOL(filemap_page_mkwrite
);
2167 const struct vm_operations_struct generic_file_vm_ops
= {
2168 .fault
= filemap_fault
,
2169 .map_pages
= filemap_map_pages
,
2170 .page_mkwrite
= filemap_page_mkwrite
,
2171 .remap_pages
= generic_file_remap_pages
,
2174 /* This is used for a general mmap of a disk file */
2176 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2178 struct address_space
*mapping
= file
->f_mapping
;
2180 if (!mapping
->a_ops
->readpage
)
2182 file_accessed(file
);
2183 vma
->vm_ops
= &generic_file_vm_ops
;
2188 * This is for filesystems which do not implement ->writepage.
2190 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2192 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2194 return generic_file_mmap(file
, vma
);
2197 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2201 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2205 #endif /* CONFIG_MMU */
2207 EXPORT_SYMBOL(generic_file_mmap
);
2208 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2210 static struct page
*wait_on_page_read(struct page
*page
)
2212 if (!IS_ERR(page
)) {
2213 wait_on_page_locked(page
);
2214 if (!PageUptodate(page
)) {
2215 page_cache_release(page
);
2216 page
= ERR_PTR(-EIO
);
2222 static struct page
*__read_cache_page(struct address_space
*mapping
,
2224 int (*filler
)(void *, struct page
*),
2231 page
= find_get_page(mapping
, index
);
2233 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2235 return ERR_PTR(-ENOMEM
);
2236 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2237 if (unlikely(err
)) {
2238 page_cache_release(page
);
2241 /* Presumably ENOMEM for radix tree node */
2242 return ERR_PTR(err
);
2244 err
= filler(data
, page
);
2246 page_cache_release(page
);
2247 page
= ERR_PTR(err
);
2249 page
= wait_on_page_read(page
);
2255 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2257 int (*filler
)(void *, struct page
*),
2266 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2269 if (PageUptodate(page
))
2273 if (!page
->mapping
) {
2275 page_cache_release(page
);
2278 if (PageUptodate(page
)) {
2282 err
= filler(data
, page
);
2284 page_cache_release(page
);
2285 return ERR_PTR(err
);
2287 page
= wait_on_page_read(page
);
2292 mark_page_accessed(page
);
2297 * read_cache_page - read into page cache, fill it if needed
2298 * @mapping: the page's address_space
2299 * @index: the page index
2300 * @filler: function to perform the read
2301 * @data: first arg to filler(data, page) function, often left as NULL
2303 * Read into the page cache. If a page already exists, and PageUptodate() is
2304 * not set, try to fill the page and wait for it to become unlocked.
2306 * If the page does not get brought uptodate, return -EIO.
2308 struct page
*read_cache_page(struct address_space
*mapping
,
2310 int (*filler
)(void *, struct page
*),
2313 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2315 EXPORT_SYMBOL(read_cache_page
);
2318 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2319 * @mapping: the page's address_space
2320 * @index: the page index
2321 * @gfp: the page allocator flags to use if allocating
2323 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2324 * any new page allocations done using the specified allocation flags.
2326 * If the page does not get brought uptodate, return -EIO.
2328 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2332 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2334 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2336 EXPORT_SYMBOL(read_cache_page_gfp
);
2338 static size_t __iovec_copy_from_user_inatomic(char *vaddr
,
2339 const struct iovec
*iov
, size_t base
, size_t bytes
)
2341 size_t copied
= 0, left
= 0;
2344 char __user
*buf
= iov
->iov_base
+ base
;
2345 int copy
= min(bytes
, iov
->iov_len
- base
);
2348 left
= __copy_from_user_inatomic(vaddr
, buf
, copy
);
2357 return copied
- left
;
2361 * Copy as much as we can into the page and return the number of bytes which
2362 * were successfully copied. If a fault is encountered then return the number of
2363 * bytes which were copied.
2365 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
2366 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2371 BUG_ON(!in_atomic());
2372 kaddr
= kmap_atomic(page
);
2373 if (likely(i
->nr_segs
== 1)) {
2375 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2376 left
= __copy_from_user_inatomic(kaddr
+ offset
, buf
, bytes
);
2377 copied
= bytes
- left
;
2379 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2380 i
->iov
, i
->iov_offset
, bytes
);
2382 kunmap_atomic(kaddr
);
2386 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
2389 * This has the same sideeffects and return value as
2390 * iov_iter_copy_from_user_atomic().
2391 * The difference is that it attempts to resolve faults.
2392 * Page must not be locked.
2394 size_t iov_iter_copy_from_user(struct page
*page
,
2395 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2401 if (likely(i
->nr_segs
== 1)) {
2403 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2404 left
= __copy_from_user(kaddr
+ offset
, buf
, bytes
);
2405 copied
= bytes
- left
;
2407 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2408 i
->iov
, i
->iov_offset
, bytes
);
2413 EXPORT_SYMBOL(iov_iter_copy_from_user
);
2415 void iov_iter_advance(struct iov_iter
*i
, size_t bytes
)
2417 BUG_ON(i
->count
< bytes
);
2419 if (likely(i
->nr_segs
== 1)) {
2420 i
->iov_offset
+= bytes
;
2423 const struct iovec
*iov
= i
->iov
;
2424 size_t base
= i
->iov_offset
;
2425 unsigned long nr_segs
= i
->nr_segs
;
2428 * The !iov->iov_len check ensures we skip over unlikely
2429 * zero-length segments (without overruning the iovec).
2431 while (bytes
|| unlikely(i
->count
&& !iov
->iov_len
)) {
2434 copy
= min(bytes
, iov
->iov_len
- base
);
2435 BUG_ON(!i
->count
|| i
->count
< copy
);
2439 if (iov
->iov_len
== base
) {
2446 i
->iov_offset
= base
;
2447 i
->nr_segs
= nr_segs
;
2450 EXPORT_SYMBOL(iov_iter_advance
);
2453 * Fault in the first iovec of the given iov_iter, to a maximum length
2454 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2455 * accessed (ie. because it is an invalid address).
2457 * writev-intensive code may want this to prefault several iovecs -- that
2458 * would be possible (callers must not rely on the fact that _only_ the
2459 * first iovec will be faulted with the current implementation).
2461 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
2463 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2464 bytes
= min(bytes
, i
->iov
->iov_len
- i
->iov_offset
);
2465 return fault_in_pages_readable(buf
, bytes
);
2467 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
2470 * Return the count of just the current iov_iter segment.
2472 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
2474 const struct iovec
*iov
= i
->iov
;
2475 if (i
->nr_segs
== 1)
2478 return min(i
->count
, iov
->iov_len
- i
->iov_offset
);
2480 EXPORT_SYMBOL(iov_iter_single_seg_count
);
2483 * Performs necessary checks before doing a write
2485 * Can adjust writing position or amount of bytes to write.
2486 * Returns appropriate error code that caller should return or
2487 * zero in case that write should be allowed.
2489 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
2491 struct inode
*inode
= file
->f_mapping
->host
;
2492 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2494 if (unlikely(*pos
< 0))
2498 /* FIXME: this is for backwards compatibility with 2.4 */
2499 if (file
->f_flags
& O_APPEND
)
2500 *pos
= i_size_read(inode
);
2502 if (limit
!= RLIM_INFINITY
) {
2503 if (*pos
>= limit
) {
2504 send_sig(SIGXFSZ
, current
, 0);
2507 if (*count
> limit
- (typeof(limit
))*pos
) {
2508 *count
= limit
- (typeof(limit
))*pos
;
2516 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2517 !(file
->f_flags
& O_LARGEFILE
))) {
2518 if (*pos
>= MAX_NON_LFS
) {
2521 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2522 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2527 * Are we about to exceed the fs block limit ?
2529 * If we have written data it becomes a short write. If we have
2530 * exceeded without writing data we send a signal and return EFBIG.
2531 * Linus frestrict idea will clean these up nicely..
2533 if (likely(!isblk
)) {
2534 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2535 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2538 /* zero-length writes at ->s_maxbytes are OK */
2541 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2542 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2546 if (bdev_read_only(I_BDEV(inode
)))
2548 isize
= i_size_read(inode
);
2549 if (*pos
>= isize
) {
2550 if (*count
|| *pos
> isize
)
2554 if (*pos
+ *count
> isize
)
2555 *count
= isize
- *pos
;
2562 EXPORT_SYMBOL(generic_write_checks
);
2564 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2565 loff_t pos
, unsigned len
, unsigned flags
,
2566 struct page
**pagep
, void **fsdata
)
2568 const struct address_space_operations
*aops
= mapping
->a_ops
;
2570 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2573 EXPORT_SYMBOL(pagecache_write_begin
);
2575 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2576 loff_t pos
, unsigned len
, unsigned copied
,
2577 struct page
*page
, void *fsdata
)
2579 const struct address_space_operations
*aops
= mapping
->a_ops
;
2581 mark_page_accessed(page
);
2582 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2584 EXPORT_SYMBOL(pagecache_write_end
);
2587 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2588 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2589 size_t count
, size_t ocount
)
2591 struct file
*file
= iocb
->ki_filp
;
2592 struct address_space
*mapping
= file
->f_mapping
;
2593 struct inode
*inode
= mapping
->host
;
2598 if (count
!= ocount
)
2599 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2601 write_len
= iov_length(iov
, *nr_segs
);
2602 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2604 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2609 * After a write we want buffered reads to be sure to go to disk to get
2610 * the new data. We invalidate clean cached page from the region we're
2611 * about to write. We do this *before* the write so that we can return
2612 * without clobbering -EIOCBQUEUED from ->direct_IO().
2614 if (mapping
->nrpages
) {
2615 written
= invalidate_inode_pages2_range(mapping
,
2616 pos
>> PAGE_CACHE_SHIFT
, end
);
2618 * If a page can not be invalidated, return 0 to fall back
2619 * to buffered write.
2622 if (written
== -EBUSY
)
2628 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2631 * Finally, try again to invalidate clean pages which might have been
2632 * cached by non-direct readahead, or faulted in by get_user_pages()
2633 * if the source of the write was an mmap'ed region of the file
2634 * we're writing. Either one is a pretty crazy thing to do,
2635 * so we don't support it 100%. If this invalidation
2636 * fails, tough, the write still worked...
2638 if (mapping
->nrpages
) {
2639 invalidate_inode_pages2_range(mapping
,
2640 pos
>> PAGE_CACHE_SHIFT
, end
);
2645 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2646 i_size_write(inode
, pos
);
2647 mark_inode_dirty(inode
);
2654 EXPORT_SYMBOL(generic_file_direct_write
);
2657 * Find or create a page at the given pagecache position. Return the locked
2658 * page. This function is specifically for buffered writes.
2660 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2661 pgoff_t index
, unsigned flags
)
2666 gfp_t gfp_notmask
= 0;
2668 gfp_mask
= mapping_gfp_mask(mapping
);
2669 if (mapping_cap_account_dirty(mapping
))
2670 gfp_mask
|= __GFP_WRITE
;
2671 if (flags
& AOP_FLAG_NOFS
)
2672 gfp_notmask
= __GFP_FS
;
2674 page
= find_lock_page(mapping
, index
);
2678 page
= __page_cache_alloc(gfp_mask
& ~gfp_notmask
);
2681 status
= add_to_page_cache_lru(page
, mapping
, index
,
2682 GFP_KERNEL
& ~gfp_notmask
);
2683 if (unlikely(status
)) {
2684 page_cache_release(page
);
2685 if (status
== -EEXIST
)
2690 wait_for_stable_page(page
);
2693 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2695 static ssize_t
generic_perform_write(struct file
*file
,
2696 struct iov_iter
*i
, loff_t pos
)
2698 struct address_space
*mapping
= file
->f_mapping
;
2699 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2701 ssize_t written
= 0;
2702 unsigned int flags
= 0;
2705 * Copies from kernel address space cannot fail (NFSD is a big user).
2707 if (segment_eq(get_fs(), KERNEL_DS
))
2708 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2712 unsigned long offset
; /* Offset into pagecache page */
2713 unsigned long bytes
; /* Bytes to write to page */
2714 size_t copied
; /* Bytes copied from user */
2717 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2718 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2723 * Bring in the user page that we will copy from _first_.
2724 * Otherwise there's a nasty deadlock on copying from the
2725 * same page as we're writing to, without it being marked
2728 * Not only is this an optimisation, but it is also required
2729 * to check that the address is actually valid, when atomic
2730 * usercopies are used, below.
2732 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2737 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2739 if (unlikely(status
))
2742 if (mapping_writably_mapped(mapping
))
2743 flush_dcache_page(page
);
2745 pagefault_disable();
2746 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2748 flush_dcache_page(page
);
2750 mark_page_accessed(page
);
2751 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2753 if (unlikely(status
< 0))
2759 iov_iter_advance(i
, copied
);
2760 if (unlikely(copied
== 0)) {
2762 * If we were unable to copy any data at all, we must
2763 * fall back to a single segment length write.
2765 * If we didn't fallback here, we could livelock
2766 * because not all segments in the iov can be copied at
2767 * once without a pagefault.
2769 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2770 iov_iter_single_seg_count(i
));
2776 balance_dirty_pages_ratelimited(mapping
);
2777 if (fatal_signal_pending(current
)) {
2781 } while (iov_iter_count(i
));
2783 return written
? written
: status
;
2787 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2788 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2789 size_t count
, ssize_t written
)
2791 struct file
*file
= iocb
->ki_filp
;
2795 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
2796 status
= generic_perform_write(file
, &i
, pos
);
2798 if (likely(status
>= 0)) {
2800 *ppos
= pos
+ status
;
2803 return written
? written
: status
;
2805 EXPORT_SYMBOL(generic_file_buffered_write
);
2808 * __generic_file_aio_write - write data to a file
2809 * @iocb: IO state structure (file, offset, etc.)
2810 * @iov: vector with data to write
2811 * @nr_segs: number of segments in the vector
2812 * @ppos: position where to write
2814 * This function does all the work needed for actually writing data to a
2815 * file. It does all basic checks, removes SUID from the file, updates
2816 * modification times and calls proper subroutines depending on whether we
2817 * do direct IO or a standard buffered write.
2819 * It expects i_mutex to be grabbed unless we work on a block device or similar
2820 * object which does not need locking at all.
2822 * This function does *not* take care of syncing data in case of O_SYNC write.
2823 * A caller has to handle it. This is mainly due to the fact that we want to
2824 * avoid syncing under i_mutex.
2826 ssize_t
__generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2827 unsigned long nr_segs
, loff_t
*ppos
)
2829 struct file
*file
= iocb
->ki_filp
;
2830 struct address_space
* mapping
= file
->f_mapping
;
2831 size_t ocount
; /* original count */
2832 size_t count
; /* after file limit checks */
2833 struct inode
*inode
= mapping
->host
;
2839 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
2846 /* We can write back this queue in page reclaim */
2847 current
->backing_dev_info
= mapping
->backing_dev_info
;
2850 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2857 err
= file_remove_suid(file
);
2861 err
= file_update_time(file
);
2865 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2866 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2868 ssize_t written_buffered
;
2870 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2871 ppos
, count
, ocount
);
2872 if (written
< 0 || written
== count
)
2875 * direct-io write to a hole: fall through to buffered I/O
2876 * for completing the rest of the request.
2880 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2881 nr_segs
, pos
, ppos
, count
,
2884 * If generic_file_buffered_write() retuned a synchronous error
2885 * then we want to return the number of bytes which were
2886 * direct-written, or the error code if that was zero. Note
2887 * that this differs from normal direct-io semantics, which
2888 * will return -EFOO even if some bytes were written.
2890 if (written_buffered
< 0) {
2891 err
= written_buffered
;
2896 * We need to ensure that the page cache pages are written to
2897 * disk and invalidated to preserve the expected O_DIRECT
2900 endbyte
= pos
+ written_buffered
- written
- 1;
2901 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
2903 written
= written_buffered
;
2904 invalidate_mapping_pages(mapping
,
2905 pos
>> PAGE_CACHE_SHIFT
,
2906 endbyte
>> PAGE_CACHE_SHIFT
);
2909 * We don't know how much we wrote, so just return
2910 * the number of bytes which were direct-written
2914 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2915 pos
, ppos
, count
, written
);
2918 current
->backing_dev_info
= NULL
;
2919 return written
? written
: err
;
2921 EXPORT_SYMBOL(__generic_file_aio_write
);
2924 * generic_file_aio_write - write data to a file
2925 * @iocb: IO state structure
2926 * @iov: vector with data to write
2927 * @nr_segs: number of segments in the vector
2928 * @pos: position in file where to write
2930 * This is a wrapper around __generic_file_aio_write() to be used by most
2931 * filesystems. It takes care of syncing the file in case of O_SYNC file
2932 * and acquires i_mutex as needed.
2934 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2935 unsigned long nr_segs
, loff_t pos
)
2937 struct file
*file
= iocb
->ki_filp
;
2938 struct inode
*inode
= file
->f_mapping
->host
;
2941 BUG_ON(iocb
->ki_pos
!= pos
);
2943 mutex_lock(&inode
->i_mutex
);
2944 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
, &iocb
->ki_pos
);
2945 mutex_unlock(&inode
->i_mutex
);
2950 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2956 EXPORT_SYMBOL(generic_file_aio_write
);
2959 * try_to_release_page() - release old fs-specific metadata on a page
2961 * @page: the page which the kernel is trying to free
2962 * @gfp_mask: memory allocation flags (and I/O mode)
2964 * The address_space is to try to release any data against the page
2965 * (presumably at page->private). If the release was successful, return `1'.
2966 * Otherwise return zero.
2968 * This may also be called if PG_fscache is set on a page, indicating that the
2969 * page is known to the local caching routines.
2971 * The @gfp_mask argument specifies whether I/O may be performed to release
2972 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2975 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2977 struct address_space
* const mapping
= page
->mapping
;
2979 BUG_ON(!PageLocked(page
));
2980 if (PageWriteback(page
))
2983 if (mapping
&& mapping
->a_ops
->releasepage
)
2984 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2985 return try_to_free_buffers(page
);
2988 EXPORT_SYMBOL(try_to_release_page
);