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>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/filemap.h>
42 * FIXME: remove all knowledge of the buffer layer from the core VM
44 #include <linux/buffer_head.h> /* for try_to_free_buffers */
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * Shared mappings now work. 15.8.1995 Bruno.
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
63 * ->i_mmap_mutex (truncate_pagecache)
64 * ->private_lock (__free_pte->__set_page_dirty_buffers)
65 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
69 * ->i_mmap_mutex (truncate->unmap_mapping_range)
73 * ->page_table_lock or pte_lock (various, mainly in memory.c)
74 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 * ->lock_page (access_process_vm)
79 * ->i_mutex (generic_file_buffered_write)
80 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
83 * sb_lock (fs/fs-writeback.c)
84 * ->mapping->tree_lock (__sync_single_inode)
87 * ->anon_vma.lock (vma_adjust)
90 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
92 * ->page_table_lock or pte_lock
93 * ->swap_lock (try_to_unmap_one)
94 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
97 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
98 * ->private_lock (page_remove_rmap->set_page_dirty)
99 * ->tree_lock (page_remove_rmap->set_page_dirty)
100 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
101 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
103 * ->inode->i_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 * Delete a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold the mapping's tree_lock.
115 void __delete_from_page_cache(struct page
*page
)
117 struct address_space
*mapping
= page
->mapping
;
119 trace_mm_filemap_delete_from_page_cache(page
);
121 * if we're uptodate, flush out into the cleancache, otherwise
122 * invalidate any existing cleancache entries. We can't leave
123 * stale data around in the cleancache once our page is gone
125 if (PageUptodate(page
) && PageMappedToDisk(page
))
126 cleancache_put_page(page
);
128 cleancache_invalidate_page(mapping
, page
);
130 radix_tree_delete(&mapping
->page_tree
, page
->index
);
131 page
->mapping
= NULL
;
132 /* Leave page->index set: truncation lookup relies upon it */
134 __dec_zone_page_state(page
, NR_FILE_PAGES
);
135 if (PageSwapBacked(page
))
136 __dec_zone_page_state(page
, NR_SHMEM
);
137 BUG_ON(page_mapped(page
));
140 * Some filesystems seem to re-dirty the page even after
141 * the VM has canceled the dirty bit (eg ext3 journaling).
143 * Fix it up by doing a final dirty accounting check after
144 * having removed the page entirely.
146 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
147 dec_zone_page_state(page
, NR_FILE_DIRTY
);
148 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
153 * delete_from_page_cache - delete page from page cache
154 * @page: the page which the kernel is trying to remove from page cache
156 * This must be called only on pages that have been verified to be in the page
157 * cache and locked. It will never put the page into the free list, the caller
158 * has a reference on the page.
160 void delete_from_page_cache(struct page
*page
)
162 struct address_space
*mapping
= page
->mapping
;
163 void (*freepage
)(struct page
*);
165 BUG_ON(!PageLocked(page
));
167 freepage
= mapping
->a_ops
->freepage
;
168 spin_lock_irq(&mapping
->tree_lock
);
169 __delete_from_page_cache(page
);
170 spin_unlock_irq(&mapping
->tree_lock
);
171 mem_cgroup_uncharge_cache_page(page
);
175 page_cache_release(page
);
177 EXPORT_SYMBOL(delete_from_page_cache
);
179 static int sleep_on_page(void *word
)
185 static int sleep_on_page_killable(void *word
)
188 return fatal_signal_pending(current
) ? -EINTR
: 0;
191 static int filemap_check_errors(struct address_space
*mapping
)
194 /* Check for outstanding write errors */
195 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
197 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
203 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
204 * @mapping: address space structure to write
205 * @start: offset in bytes where the range starts
206 * @end: offset in bytes where the range ends (inclusive)
207 * @sync_mode: enable synchronous operation
209 * Start writeback against all of a mapping's dirty pages that lie
210 * within the byte offsets <start, end> inclusive.
212 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
213 * opposed to a regular memory cleansing writeback. The difference between
214 * these two operations is that if a dirty page/buffer is encountered, it must
215 * be waited upon, and not just skipped over.
217 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
218 loff_t end
, int sync_mode
)
221 struct writeback_control wbc
= {
222 .sync_mode
= sync_mode
,
223 .nr_to_write
= LONG_MAX
,
224 .range_start
= start
,
228 if (!mapping_cap_writeback_dirty(mapping
))
231 ret
= do_writepages(mapping
, &wbc
);
235 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
238 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
241 int filemap_fdatawrite(struct address_space
*mapping
)
243 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
245 EXPORT_SYMBOL(filemap_fdatawrite
);
247 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
250 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
252 EXPORT_SYMBOL(filemap_fdatawrite_range
);
255 * filemap_flush - mostly a non-blocking flush
256 * @mapping: target address_space
258 * This is a mostly non-blocking flush. Not suitable for data-integrity
259 * purposes - I/O may not be started against all dirty pages.
261 int filemap_flush(struct address_space
*mapping
)
263 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
265 EXPORT_SYMBOL(filemap_flush
);
268 * filemap_fdatawait_range - wait for writeback to complete
269 * @mapping: address space structure to wait for
270 * @start_byte: offset in bytes where the range starts
271 * @end_byte: offset in bytes where the range ends (inclusive)
273 * Walk the list of under-writeback pages of the given address space
274 * in the given range and wait for all of them.
276 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
279 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
280 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
285 if (end_byte
< start_byte
)
288 pagevec_init(&pvec
, 0);
289 while ((index
<= end
) &&
290 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
291 PAGECACHE_TAG_WRITEBACK
,
292 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
295 for (i
= 0; i
< nr_pages
; i
++) {
296 struct page
*page
= pvec
.pages
[i
];
298 /* until radix tree lookup accepts end_index */
299 if (page
->index
> end
)
302 wait_on_page_writeback(page
);
303 if (TestClearPageError(page
))
306 pagevec_release(&pvec
);
310 ret2
= filemap_check_errors(mapping
);
316 EXPORT_SYMBOL(filemap_fdatawait_range
);
319 * filemap_fdatawait - wait for all under-writeback pages to complete
320 * @mapping: address space structure to wait for
322 * Walk the list of under-writeback pages of the given address space
323 * and wait for all of them.
325 int filemap_fdatawait(struct address_space
*mapping
)
327 loff_t i_size
= i_size_read(mapping
->host
);
332 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
334 EXPORT_SYMBOL(filemap_fdatawait
);
336 int filemap_write_and_wait(struct address_space
*mapping
)
340 if (mapping
->nrpages
) {
341 err
= filemap_fdatawrite(mapping
);
343 * Even if the above returned error, the pages may be
344 * written partially (e.g. -ENOSPC), so we wait for it.
345 * But the -EIO is special case, it may indicate the worst
346 * thing (e.g. bug) happened, so we avoid waiting for it.
349 int err2
= filemap_fdatawait(mapping
);
354 err
= filemap_check_errors(mapping
);
358 EXPORT_SYMBOL(filemap_write_and_wait
);
361 * filemap_write_and_wait_range - write out & wait on a file range
362 * @mapping: the address_space for the pages
363 * @lstart: offset in bytes where the range starts
364 * @lend: offset in bytes where the range ends (inclusive)
366 * Write out and wait upon file offsets lstart->lend, inclusive.
368 * Note that `lend' is inclusive (describes the last byte to be written) so
369 * that this function can be used to write to the very end-of-file (end = -1).
371 int filemap_write_and_wait_range(struct address_space
*mapping
,
372 loff_t lstart
, loff_t lend
)
376 if (mapping
->nrpages
) {
377 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
379 /* See comment of filemap_write_and_wait() */
381 int err2
= filemap_fdatawait_range(mapping
,
387 err
= filemap_check_errors(mapping
);
391 EXPORT_SYMBOL(filemap_write_and_wait_range
);
394 * replace_page_cache_page - replace a pagecache page with a new one
395 * @old: page to be replaced
396 * @new: page to replace with
397 * @gfp_mask: allocation mode
399 * This function replaces a page in the pagecache with a new one. On
400 * success it acquires the pagecache reference for the new page and
401 * drops it for the old page. Both the old and new pages must be
402 * locked. This function does not add the new page to the LRU, the
403 * caller must do that.
405 * The remove + add is atomic. The only way this function can fail is
406 * memory allocation failure.
408 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
412 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
413 VM_BUG_ON_PAGE(!PageLocked(new), new);
414 VM_BUG_ON_PAGE(new->mapping
, new);
416 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
418 struct address_space
*mapping
= old
->mapping
;
419 void (*freepage
)(struct page
*);
421 pgoff_t offset
= old
->index
;
422 freepage
= mapping
->a_ops
->freepage
;
425 new->mapping
= mapping
;
428 spin_lock_irq(&mapping
->tree_lock
);
429 __delete_from_page_cache(old
);
430 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
433 __inc_zone_page_state(new, NR_FILE_PAGES
);
434 if (PageSwapBacked(new))
435 __inc_zone_page_state(new, NR_SHMEM
);
436 spin_unlock_irq(&mapping
->tree_lock
);
437 /* mem_cgroup codes must not be called under tree_lock */
438 mem_cgroup_replace_page_cache(old
, new);
439 radix_tree_preload_end();
442 page_cache_release(old
);
447 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
449 static int page_cache_tree_insert(struct address_space
*mapping
,
455 slot
= radix_tree_lookup_slot(&mapping
->page_tree
, page
->index
);
459 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
460 if (!radix_tree_exceptional_entry(p
))
462 radix_tree_replace_slot(slot
, page
);
466 error
= radix_tree_insert(&mapping
->page_tree
, page
->index
, page
);
473 * add_to_page_cache_locked - add a locked page to the pagecache
475 * @mapping: the page's address_space
476 * @offset: page index
477 * @gfp_mask: page allocation mode
479 * This function is used to add a page to the pagecache. It must be locked.
480 * This function does not add the page to the LRU. The caller must do that.
482 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
483 pgoff_t offset
, gfp_t gfp_mask
)
487 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
488 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
490 error
= mem_cgroup_cache_charge(page
, current
->mm
,
491 gfp_mask
& GFP_RECLAIM_MASK
);
495 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
497 mem_cgroup_uncharge_cache_page(page
);
501 page_cache_get(page
);
502 page
->mapping
= mapping
;
503 page
->index
= offset
;
505 spin_lock_irq(&mapping
->tree_lock
);
506 error
= page_cache_tree_insert(mapping
, page
);
507 radix_tree_preload_end();
510 __inc_zone_page_state(page
, NR_FILE_PAGES
);
511 spin_unlock_irq(&mapping
->tree_lock
);
512 trace_mm_filemap_add_to_page_cache(page
);
515 page
->mapping
= NULL
;
516 /* Leave page->index set: truncation relies upon it */
517 spin_unlock_irq(&mapping
->tree_lock
);
518 mem_cgroup_uncharge_cache_page(page
);
519 page_cache_release(page
);
522 EXPORT_SYMBOL(add_to_page_cache_locked
);
524 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
525 pgoff_t offset
, gfp_t gfp_mask
)
529 ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
531 lru_cache_add_file(page
);
534 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
537 struct page
*__page_cache_alloc(gfp_t gfp
)
542 if (cpuset_do_page_mem_spread()) {
543 unsigned int cpuset_mems_cookie
;
545 cpuset_mems_cookie
= read_mems_allowed_begin();
546 n
= cpuset_mem_spread_node();
547 page
= alloc_pages_exact_node(n
, gfp
, 0);
548 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
552 return alloc_pages(gfp
, 0);
554 EXPORT_SYMBOL(__page_cache_alloc
);
558 * In order to wait for pages to become available there must be
559 * waitqueues associated with pages. By using a hash table of
560 * waitqueues where the bucket discipline is to maintain all
561 * waiters on the same queue and wake all when any of the pages
562 * become available, and for the woken contexts to check to be
563 * sure the appropriate page became available, this saves space
564 * at a cost of "thundering herd" phenomena during rare hash
567 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
569 const struct zone
*zone
= page_zone(page
);
571 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
574 static inline void wake_up_page(struct page
*page
, int bit
)
576 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
579 void wait_on_page_bit(struct page
*page
, int bit_nr
)
581 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
583 if (test_bit(bit_nr
, &page
->flags
))
584 __wait_on_bit(page_waitqueue(page
), &wait
, sleep_on_page
,
585 TASK_UNINTERRUPTIBLE
);
587 EXPORT_SYMBOL(wait_on_page_bit
);
589 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
591 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
593 if (!test_bit(bit_nr
, &page
->flags
))
596 return __wait_on_bit(page_waitqueue(page
), &wait
,
597 sleep_on_page_killable
, TASK_KILLABLE
);
601 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
602 * @page: Page defining the wait queue of interest
603 * @waiter: Waiter to add to the queue
605 * Add an arbitrary @waiter to the wait queue for the nominated @page.
607 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
609 wait_queue_head_t
*q
= page_waitqueue(page
);
612 spin_lock_irqsave(&q
->lock
, flags
);
613 __add_wait_queue(q
, waiter
);
614 spin_unlock_irqrestore(&q
->lock
, flags
);
616 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
619 * unlock_page - unlock a locked page
622 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
623 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
624 * mechananism between PageLocked pages and PageWriteback pages is shared.
625 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
627 * The mb is necessary to enforce ordering between the clear_bit and the read
628 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
630 void unlock_page(struct page
*page
)
632 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
633 clear_bit_unlock(PG_locked
, &page
->flags
);
634 smp_mb__after_clear_bit();
635 wake_up_page(page
, PG_locked
);
637 EXPORT_SYMBOL(unlock_page
);
640 * end_page_writeback - end writeback against a page
643 void end_page_writeback(struct page
*page
)
645 if (TestClearPageReclaim(page
))
646 rotate_reclaimable_page(page
);
648 if (!test_clear_page_writeback(page
))
651 smp_mb__after_clear_bit();
652 wake_up_page(page
, PG_writeback
);
654 EXPORT_SYMBOL(end_page_writeback
);
657 * __lock_page - get a lock on the page, assuming we need to sleep to get it
658 * @page: the page to lock
660 void __lock_page(struct page
*page
)
662 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
664 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sleep_on_page
,
665 TASK_UNINTERRUPTIBLE
);
667 EXPORT_SYMBOL(__lock_page
);
669 int __lock_page_killable(struct page
*page
)
671 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
673 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
674 sleep_on_page_killable
, TASK_KILLABLE
);
676 EXPORT_SYMBOL_GPL(__lock_page_killable
);
678 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
681 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
683 * CAUTION! In this case, mmap_sem is not released
684 * even though return 0.
686 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
689 up_read(&mm
->mmap_sem
);
690 if (flags
& FAULT_FLAG_KILLABLE
)
691 wait_on_page_locked_killable(page
);
693 wait_on_page_locked(page
);
696 if (flags
& FAULT_FLAG_KILLABLE
) {
699 ret
= __lock_page_killable(page
);
701 up_read(&mm
->mmap_sem
);
711 * page_cache_next_hole - find the next hole (not-present entry)
714 * @max_scan: maximum range to search
716 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
717 * lowest indexed hole.
719 * Returns: the index of the hole if found, otherwise returns an index
720 * outside of the set specified (in which case 'return - index >=
721 * max_scan' will be true). In rare cases of index wrap-around, 0 will
724 * page_cache_next_hole may be called under rcu_read_lock. However,
725 * like radix_tree_gang_lookup, this will not atomically search a
726 * snapshot of the tree at a single point in time. For example, if a
727 * hole is created at index 5, then subsequently a hole is created at
728 * index 10, page_cache_next_hole covering both indexes may return 10
729 * if called under rcu_read_lock.
731 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
732 pgoff_t index
, unsigned long max_scan
)
736 for (i
= 0; i
< max_scan
; i
++) {
739 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
740 if (!page
|| radix_tree_exceptional_entry(page
))
749 EXPORT_SYMBOL(page_cache_next_hole
);
752 * page_cache_prev_hole - find the prev hole (not-present entry)
755 * @max_scan: maximum range to search
757 * Search backwards in the range [max(index-max_scan+1, 0), index] for
760 * Returns: the index of the hole if found, otherwise returns an index
761 * outside of the set specified (in which case 'index - return >=
762 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
765 * page_cache_prev_hole may be called under rcu_read_lock. However,
766 * like radix_tree_gang_lookup, this will not atomically search a
767 * snapshot of the tree at a single point in time. For example, if a
768 * hole is created at index 10, then subsequently a hole is created at
769 * index 5, page_cache_prev_hole covering both indexes may return 5 if
770 * called under rcu_read_lock.
772 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
773 pgoff_t index
, unsigned long max_scan
)
777 for (i
= 0; i
< max_scan
; i
++) {
780 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
781 if (!page
|| radix_tree_exceptional_entry(page
))
784 if (index
== ULONG_MAX
)
790 EXPORT_SYMBOL(page_cache_prev_hole
);
793 * find_get_entry - find and get a page cache entry
794 * @mapping: the address_space to search
795 * @offset: the page cache index
797 * Looks up the page cache slot at @mapping & @offset. If there is a
798 * page cache page, it is returned with an increased refcount.
800 * If the slot holds a shadow entry of a previously evicted page, it
803 * Otherwise, %NULL is returned.
805 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
813 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
815 page
= radix_tree_deref_slot(pagep
);
818 if (radix_tree_exception(page
)) {
819 if (radix_tree_deref_retry(page
))
822 * Otherwise, shmem/tmpfs must be storing a swap entry
823 * here as an exceptional entry: so return it without
824 * attempting to raise page count.
828 if (!page_cache_get_speculative(page
))
832 * Has the page moved?
833 * This is part of the lockless pagecache protocol. See
834 * include/linux/pagemap.h for details.
836 if (unlikely(page
!= *pagep
)) {
837 page_cache_release(page
);
846 EXPORT_SYMBOL(find_get_entry
);
849 * find_get_page - find and get a page reference
850 * @mapping: the address_space to search
851 * @offset: the page index
853 * Looks up the page cache slot at @mapping & @offset. If there is a
854 * page cache page, it is returned with an increased refcount.
856 * Otherwise, %NULL is returned.
858 struct page
*find_get_page(struct address_space
*mapping
, pgoff_t offset
)
860 struct page
*page
= find_get_entry(mapping
, offset
);
862 if (radix_tree_exceptional_entry(page
))
866 EXPORT_SYMBOL(find_get_page
);
869 * find_lock_entry - locate, pin and lock a page cache entry
870 * @mapping: the address_space to search
871 * @offset: the page cache index
873 * Looks up the page cache slot at @mapping & @offset. If there is a
874 * page cache page, it is returned locked and with an increased
877 * If the slot holds a shadow entry of a previously evicted page, it
880 * Otherwise, %NULL is returned.
882 * find_lock_entry() may sleep.
884 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
889 page
= find_get_entry(mapping
, offset
);
890 if (page
&& !radix_tree_exception(page
)) {
892 /* Has the page been truncated? */
893 if (unlikely(page
->mapping
!= mapping
)) {
895 page_cache_release(page
);
898 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
902 EXPORT_SYMBOL(find_lock_entry
);
905 * find_lock_page - locate, pin and lock a pagecache page
906 * @mapping: the address_space to search
907 * @offset: the page index
909 * Looks up the page cache slot at @mapping & @offset. If there is a
910 * page cache page, it is returned locked and with an increased
913 * Otherwise, %NULL is returned.
915 * find_lock_page() may sleep.
917 struct page
*find_lock_page(struct address_space
*mapping
, pgoff_t offset
)
919 struct page
*page
= find_lock_entry(mapping
, offset
);
921 if (radix_tree_exceptional_entry(page
))
925 EXPORT_SYMBOL(find_lock_page
);
928 * find_or_create_page - locate or add a pagecache page
929 * @mapping: the page's address_space
930 * @index: the page's index into the mapping
931 * @gfp_mask: page allocation mode
933 * Looks up the page cache slot at @mapping & @offset. If there is a
934 * page cache page, it is returned locked and with an increased
937 * If the page is not present, a new page is allocated using @gfp_mask
938 * and added to the page cache and the VM's LRU list. The page is
939 * returned locked and with an increased refcount.
941 * On memory exhaustion, %NULL is returned.
943 * find_or_create_page() may sleep, even if @gfp_flags specifies an
946 struct page
*find_or_create_page(struct address_space
*mapping
,
947 pgoff_t index
, gfp_t gfp_mask
)
952 page
= find_lock_page(mapping
, index
);
954 page
= __page_cache_alloc(gfp_mask
);
958 * We want a regular kernel memory (not highmem or DMA etc)
959 * allocation for the radix tree nodes, but we need to honour
960 * the context-specific requirements the caller has asked for.
961 * GFP_RECLAIM_MASK collects those requirements.
963 err
= add_to_page_cache_lru(page
, mapping
, index
,
964 (gfp_mask
& GFP_RECLAIM_MASK
));
966 page_cache_release(page
);
974 EXPORT_SYMBOL(find_or_create_page
);
977 * find_get_entries - gang pagecache lookup
978 * @mapping: The address_space to search
979 * @start: The starting page cache index
980 * @nr_entries: The maximum number of entries
981 * @entries: Where the resulting entries are placed
982 * @indices: The cache indices corresponding to the entries in @entries
984 * find_get_entries() will search for and return a group of up to
985 * @nr_entries entries in the mapping. The entries are placed at
986 * @entries. find_get_entries() takes a reference against any actual
989 * The search returns a group of mapping-contiguous page cache entries
990 * with ascending indexes. There may be holes in the indices due to
993 * Any shadow entries of evicted pages are included in the returned
996 * find_get_entries() returns the number of pages and shadow entries
999 unsigned find_get_entries(struct address_space
*mapping
,
1000 pgoff_t start
, unsigned int nr_entries
,
1001 struct page
**entries
, pgoff_t
*indices
)
1004 unsigned int ret
= 0;
1005 struct radix_tree_iter iter
;
1012 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1015 page
= radix_tree_deref_slot(slot
);
1016 if (unlikely(!page
))
1018 if (radix_tree_exception(page
)) {
1019 if (radix_tree_deref_retry(page
))
1022 * Otherwise, we must be storing a swap entry
1023 * here as an exceptional entry: so return it
1024 * without attempting to raise page count.
1028 if (!page_cache_get_speculative(page
))
1031 /* Has the page moved? */
1032 if (unlikely(page
!= *slot
)) {
1033 page_cache_release(page
);
1037 indices
[ret
] = iter
.index
;
1038 entries
[ret
] = page
;
1039 if (++ret
== nr_entries
)
1047 * find_get_pages - gang pagecache lookup
1048 * @mapping: The address_space to search
1049 * @start: The starting page index
1050 * @nr_pages: The maximum number of pages
1051 * @pages: Where the resulting pages are placed
1053 * find_get_pages() will search for and return a group of up to
1054 * @nr_pages pages in the mapping. The pages are placed at @pages.
1055 * find_get_pages() takes a reference against the returned pages.
1057 * The search returns a group of mapping-contiguous pages with ascending
1058 * indexes. There may be holes in the indices due to not-present pages.
1060 * find_get_pages() returns the number of pages which were found.
1062 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1063 unsigned int nr_pages
, struct page
**pages
)
1065 struct radix_tree_iter iter
;
1069 if (unlikely(!nr_pages
))
1074 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1077 page
= radix_tree_deref_slot(slot
);
1078 if (unlikely(!page
))
1081 if (radix_tree_exception(page
)) {
1082 if (radix_tree_deref_retry(page
)) {
1084 * Transient condition which can only trigger
1085 * when entry at index 0 moves out of or back
1086 * to root: none yet gotten, safe to restart.
1088 WARN_ON(iter
.index
);
1092 * Otherwise, shmem/tmpfs must be storing a swap entry
1093 * here as an exceptional entry: so skip over it -
1094 * we only reach this from invalidate_mapping_pages().
1099 if (!page_cache_get_speculative(page
))
1102 /* Has the page moved? */
1103 if (unlikely(page
!= *slot
)) {
1104 page_cache_release(page
);
1109 if (++ret
== nr_pages
)
1118 * find_get_pages_contig - gang contiguous pagecache lookup
1119 * @mapping: The address_space to search
1120 * @index: The starting page index
1121 * @nr_pages: The maximum number of pages
1122 * @pages: Where the resulting pages are placed
1124 * find_get_pages_contig() works exactly like find_get_pages(), except
1125 * that the returned number of pages are guaranteed to be contiguous.
1127 * find_get_pages_contig() returns the number of pages which were found.
1129 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1130 unsigned int nr_pages
, struct page
**pages
)
1132 struct radix_tree_iter iter
;
1134 unsigned int ret
= 0;
1136 if (unlikely(!nr_pages
))
1141 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1144 page
= radix_tree_deref_slot(slot
);
1145 /* The hole, there no reason to continue */
1146 if (unlikely(!page
))
1149 if (radix_tree_exception(page
)) {
1150 if (radix_tree_deref_retry(page
)) {
1152 * Transient condition which can only trigger
1153 * when entry at index 0 moves out of or back
1154 * to root: none yet gotten, safe to restart.
1159 * Otherwise, shmem/tmpfs must be storing a swap entry
1160 * here as an exceptional entry: so stop looking for
1166 if (!page_cache_get_speculative(page
))
1169 /* Has the page moved? */
1170 if (unlikely(page
!= *slot
)) {
1171 page_cache_release(page
);
1176 * must check mapping and index after taking the ref.
1177 * otherwise we can get both false positives and false
1178 * negatives, which is just confusing to the caller.
1180 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1181 page_cache_release(page
);
1186 if (++ret
== nr_pages
)
1192 EXPORT_SYMBOL(find_get_pages_contig
);
1195 * find_get_pages_tag - find and return pages that match @tag
1196 * @mapping: the address_space to search
1197 * @index: the starting page index
1198 * @tag: the tag index
1199 * @nr_pages: the maximum number of pages
1200 * @pages: where the resulting pages are placed
1202 * Like find_get_pages, except we only return pages which are tagged with
1203 * @tag. We update @index to index the next page for the traversal.
1205 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1206 int tag
, unsigned int nr_pages
, struct page
**pages
)
1208 struct radix_tree_iter iter
;
1212 if (unlikely(!nr_pages
))
1217 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1218 &iter
, *index
, tag
) {
1221 page
= radix_tree_deref_slot(slot
);
1222 if (unlikely(!page
))
1225 if (radix_tree_exception(page
)) {
1226 if (radix_tree_deref_retry(page
)) {
1228 * Transient condition which can only trigger
1229 * when entry at index 0 moves out of or back
1230 * to root: none yet gotten, safe to restart.
1235 * This function is never used on a shmem/tmpfs
1236 * mapping, so a swap entry won't be found here.
1241 if (!page_cache_get_speculative(page
))
1244 /* Has the page moved? */
1245 if (unlikely(page
!= *slot
)) {
1246 page_cache_release(page
);
1251 if (++ret
== nr_pages
)
1258 *index
= pages
[ret
- 1]->index
+ 1;
1262 EXPORT_SYMBOL(find_get_pages_tag
);
1265 * grab_cache_page_nowait - returns locked page at given index in given cache
1266 * @mapping: target address_space
1267 * @index: the page index
1269 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1270 * This is intended for speculative data generators, where the data can
1271 * be regenerated if the page couldn't be grabbed. This routine should
1272 * be safe to call while holding the lock for another page.
1274 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1275 * and deadlock against the caller's locked page.
1278 grab_cache_page_nowait(struct address_space
*mapping
, pgoff_t index
)
1280 struct page
*page
= find_get_page(mapping
, index
);
1283 if (trylock_page(page
))
1285 page_cache_release(page
);
1288 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
1289 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_NOFS
)) {
1290 page_cache_release(page
);
1295 EXPORT_SYMBOL(grab_cache_page_nowait
);
1298 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1299 * a _large_ part of the i/o request. Imagine the worst scenario:
1301 * ---R__________________________________________B__________
1302 * ^ reading here ^ bad block(assume 4k)
1304 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1305 * => failing the whole request => read(R) => read(R+1) =>
1306 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1307 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1308 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1310 * It is going insane. Fix it by quickly scaling down the readahead size.
1312 static void shrink_readahead_size_eio(struct file
*filp
,
1313 struct file_ra_state
*ra
)
1319 * do_generic_file_read - generic file read routine
1320 * @filp: the file to read
1321 * @ppos: current file position
1322 * @desc: read_descriptor
1324 * This is a generic file read routine, and uses the
1325 * mapping->a_ops->readpage() function for the actual low-level stuff.
1327 * This is really ugly. But the goto's actually try to clarify some
1328 * of the logic when it comes to error handling etc.
1330 static void do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1331 read_descriptor_t
*desc
)
1333 struct address_space
*mapping
= filp
->f_mapping
;
1334 struct inode
*inode
= mapping
->host
;
1335 struct file_ra_state
*ra
= &filp
->f_ra
;
1339 unsigned long offset
; /* offset into pagecache page */
1340 unsigned int prev_offset
;
1343 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1344 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1345 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1346 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1347 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1353 unsigned long nr
, ret
;
1357 page
= find_get_page(mapping
, index
);
1359 page_cache_sync_readahead(mapping
,
1361 index
, last_index
- index
);
1362 page
= find_get_page(mapping
, index
);
1363 if (unlikely(page
== NULL
))
1364 goto no_cached_page
;
1366 if (PageReadahead(page
)) {
1367 page_cache_async_readahead(mapping
,
1369 index
, last_index
- index
);
1371 if (!PageUptodate(page
)) {
1372 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1373 !mapping
->a_ops
->is_partially_uptodate
)
1374 goto page_not_up_to_date
;
1375 if (!trylock_page(page
))
1376 goto page_not_up_to_date
;
1377 /* Did it get truncated before we got the lock? */
1379 goto page_not_up_to_date_locked
;
1380 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1382 goto page_not_up_to_date_locked
;
1387 * i_size must be checked after we know the page is Uptodate.
1389 * Checking i_size after the check allows us to calculate
1390 * the correct value for "nr", which means the zero-filled
1391 * part of the page is not copied back to userspace (unless
1392 * another truncate extends the file - this is desired though).
1395 isize
= i_size_read(inode
);
1396 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1397 if (unlikely(!isize
|| index
> end_index
)) {
1398 page_cache_release(page
);
1402 /* nr is the maximum number of bytes to copy from this page */
1403 nr
= PAGE_CACHE_SIZE
;
1404 if (index
== end_index
) {
1405 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1407 page_cache_release(page
);
1413 /* If users can be writing to this page using arbitrary
1414 * virtual addresses, take care about potential aliasing
1415 * before reading the page on the kernel side.
1417 if (mapping_writably_mapped(mapping
))
1418 flush_dcache_page(page
);
1421 * When a sequential read accesses a page several times,
1422 * only mark it as accessed the first time.
1424 if (prev_index
!= index
|| offset
!= prev_offset
)
1425 mark_page_accessed(page
);
1429 * Ok, we have the page, and it's up-to-date, so
1430 * now we can copy it to user space...
1432 * The file_read_actor routine returns how many bytes were
1434 * NOTE! This may not be the same as how much of a user buffer
1435 * we filled up (we may be padding etc), so we can only update
1436 * "pos" here (the actor routine has to update the user buffer
1437 * pointers and the remaining count).
1439 ret
= file_read_actor(desc
, page
, offset
, nr
);
1441 index
+= offset
>> PAGE_CACHE_SHIFT
;
1442 offset
&= ~PAGE_CACHE_MASK
;
1443 prev_offset
= offset
;
1445 page_cache_release(page
);
1446 if (ret
== nr
&& desc
->count
)
1450 page_not_up_to_date
:
1451 /* Get exclusive access to the page ... */
1452 error
= lock_page_killable(page
);
1453 if (unlikely(error
))
1454 goto readpage_error
;
1456 page_not_up_to_date_locked
:
1457 /* Did it get truncated before we got the lock? */
1458 if (!page
->mapping
) {
1460 page_cache_release(page
);
1464 /* Did somebody else fill it already? */
1465 if (PageUptodate(page
)) {
1472 * A previous I/O error may have been due to temporary
1473 * failures, eg. multipath errors.
1474 * PG_error will be set again if readpage fails.
1476 ClearPageError(page
);
1477 /* Start the actual read. The read will unlock the page. */
1478 error
= mapping
->a_ops
->readpage(filp
, page
);
1480 if (unlikely(error
)) {
1481 if (error
== AOP_TRUNCATED_PAGE
) {
1482 page_cache_release(page
);
1485 goto readpage_error
;
1488 if (!PageUptodate(page
)) {
1489 error
= lock_page_killable(page
);
1490 if (unlikely(error
))
1491 goto readpage_error
;
1492 if (!PageUptodate(page
)) {
1493 if (page
->mapping
== NULL
) {
1495 * invalidate_mapping_pages got it
1498 page_cache_release(page
);
1502 shrink_readahead_size_eio(filp
, ra
);
1504 goto readpage_error
;
1512 /* UHHUH! A synchronous read error occurred. Report it */
1513 desc
->error
= error
;
1514 page_cache_release(page
);
1519 * Ok, it wasn't cached, so we need to create a new
1522 page
= page_cache_alloc_cold(mapping
);
1524 desc
->error
= -ENOMEM
;
1527 error
= add_to_page_cache_lru(page
, mapping
,
1530 page_cache_release(page
);
1531 if (error
== -EEXIST
)
1533 desc
->error
= error
;
1540 ra
->prev_pos
= prev_index
;
1541 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1542 ra
->prev_pos
|= prev_offset
;
1544 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1545 file_accessed(filp
);
1548 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
1549 unsigned long offset
, unsigned long size
)
1552 unsigned long left
, count
= desc
->count
;
1558 * Faults on the destination of a read are common, so do it before
1561 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
1562 kaddr
= kmap_atomic(page
);
1563 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
1564 kaddr
+ offset
, size
);
1565 kunmap_atomic(kaddr
);
1570 /* Do it the slow way */
1572 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
1577 desc
->error
= -EFAULT
;
1580 desc
->count
= count
- size
;
1581 desc
->written
+= size
;
1582 desc
->arg
.buf
+= size
;
1587 * Performs necessary checks before doing a write
1588 * @iov: io vector request
1589 * @nr_segs: number of segments in the iovec
1590 * @count: number of bytes to write
1591 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1593 * Adjust number of segments and amount of bytes to write (nr_segs should be
1594 * properly initialized first). Returns appropriate error code that caller
1595 * should return or zero in case that write should be allowed.
1597 int generic_segment_checks(const struct iovec
*iov
,
1598 unsigned long *nr_segs
, size_t *count
, int access_flags
)
1602 for (seg
= 0; seg
< *nr_segs
; seg
++) {
1603 const struct iovec
*iv
= &iov
[seg
];
1606 * If any segment has a negative length, or the cumulative
1607 * length ever wraps negative then return -EINVAL.
1610 if (unlikely((ssize_t
)(cnt
|iv
->iov_len
) < 0))
1612 if (access_ok(access_flags
, iv
->iov_base
, iv
->iov_len
))
1617 cnt
-= iv
->iov_len
; /* This segment is no good */
1623 EXPORT_SYMBOL(generic_segment_checks
);
1626 * generic_file_aio_read - generic filesystem read routine
1627 * @iocb: kernel I/O control block
1628 * @iov: io vector request
1629 * @nr_segs: number of segments in the iovec
1630 * @pos: current file position
1632 * This is the "read()" routine for all filesystems
1633 * that can use the page cache directly.
1636 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1637 unsigned long nr_segs
, loff_t pos
)
1639 struct file
*filp
= iocb
->ki_filp
;
1641 unsigned long seg
= 0;
1643 loff_t
*ppos
= &iocb
->ki_pos
;
1646 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1650 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1651 if (filp
->f_flags
& O_DIRECT
) {
1653 struct address_space
*mapping
;
1654 struct inode
*inode
;
1656 mapping
= filp
->f_mapping
;
1657 inode
= mapping
->host
;
1659 goto out
; /* skip atime */
1660 size
= i_size_read(inode
);
1661 retval
= filemap_write_and_wait_range(mapping
, pos
,
1662 pos
+ iov_length(iov
, nr_segs
) - 1);
1664 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
,
1668 *ppos
= pos
+ retval
;
1673 * Btrfs can have a short DIO read if we encounter
1674 * compressed extents, so if there was an error, or if
1675 * we've already read everything we wanted to, or if
1676 * there was a short read because we hit EOF, go ahead
1677 * and return. Otherwise fallthrough to buffered io for
1678 * the rest of the read.
1680 if (retval
< 0 || !count
|| *ppos
>= size
) {
1681 file_accessed(filp
);
1687 for (seg
= 0; seg
< nr_segs
; seg
++) {
1688 read_descriptor_t desc
;
1692 * If we did a short DIO read we need to skip the section of the
1693 * iov that we've already read data into.
1696 if (count
> iov
[seg
].iov_len
) {
1697 count
-= iov
[seg
].iov_len
;
1705 desc
.arg
.buf
= iov
[seg
].iov_base
+ offset
;
1706 desc
.count
= iov
[seg
].iov_len
- offset
;
1707 if (desc
.count
== 0)
1710 do_generic_file_read(filp
, ppos
, &desc
);
1711 retval
+= desc
.written
;
1713 retval
= retval
?: desc
.error
;
1722 EXPORT_SYMBOL(generic_file_aio_read
);
1726 * page_cache_read - adds requested page to the page cache if not already there
1727 * @file: file to read
1728 * @offset: page index
1730 * This adds the requested page to the page cache if it isn't already there,
1731 * and schedules an I/O to read in its contents from disk.
1733 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1735 struct address_space
*mapping
= file
->f_mapping
;
1740 page
= page_cache_alloc_cold(mapping
);
1744 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1746 ret
= mapping
->a_ops
->readpage(file
, page
);
1747 else if (ret
== -EEXIST
)
1748 ret
= 0; /* losing race to add is OK */
1750 page_cache_release(page
);
1752 } while (ret
== AOP_TRUNCATED_PAGE
);
1757 #define MMAP_LOTSAMISS (100)
1760 * Synchronous readahead happens when we don't even find
1761 * a page in the page cache at all.
1763 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1764 struct file_ra_state
*ra
,
1768 unsigned long ra_pages
;
1769 struct address_space
*mapping
= file
->f_mapping
;
1771 /* If we don't want any read-ahead, don't bother */
1772 if (vma
->vm_flags
& VM_RAND_READ
)
1777 if (vma
->vm_flags
& VM_SEQ_READ
) {
1778 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1783 /* Avoid banging the cache line if not needed */
1784 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1788 * Do we miss much more than hit in this file? If so,
1789 * stop bothering with read-ahead. It will only hurt.
1791 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1797 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1798 ra
->start
= max_t(long, 0, offset
- ra_pages
/ 2);
1799 ra
->size
= ra_pages
;
1800 ra
->async_size
= ra_pages
/ 4;
1801 ra_submit(ra
, mapping
, file
);
1805 * Asynchronous readahead happens when we find the page and PG_readahead,
1806 * so we want to possibly extend the readahead further..
1808 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1809 struct file_ra_state
*ra
,
1814 struct address_space
*mapping
= file
->f_mapping
;
1816 /* If we don't want any read-ahead, don't bother */
1817 if (vma
->vm_flags
& VM_RAND_READ
)
1819 if (ra
->mmap_miss
> 0)
1821 if (PageReadahead(page
))
1822 page_cache_async_readahead(mapping
, ra
, file
,
1823 page
, offset
, ra
->ra_pages
);
1827 * filemap_fault - read in file data for page fault handling
1828 * @vma: vma in which the fault was taken
1829 * @vmf: struct vm_fault containing details of the fault
1831 * filemap_fault() is invoked via the vma operations vector for a
1832 * mapped memory region to read in file data during a page fault.
1834 * The goto's are kind of ugly, but this streamlines the normal case of having
1835 * it in the page cache, and handles the special cases reasonably without
1836 * having a lot of duplicated code.
1838 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1841 struct file
*file
= vma
->vm_file
;
1842 struct address_space
*mapping
= file
->f_mapping
;
1843 struct file_ra_state
*ra
= &file
->f_ra
;
1844 struct inode
*inode
= mapping
->host
;
1845 pgoff_t offset
= vmf
->pgoff
;
1850 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1852 return VM_FAULT_SIGBUS
;
1855 * Do we have something in the page cache already?
1857 page
= find_get_page(mapping
, offset
);
1858 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1860 * We found the page, so try async readahead before
1861 * waiting for the lock.
1863 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1865 /* No page in the page cache at all */
1866 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1867 count_vm_event(PGMAJFAULT
);
1868 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1869 ret
= VM_FAULT_MAJOR
;
1871 page
= find_get_page(mapping
, offset
);
1873 goto no_cached_page
;
1876 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1877 page_cache_release(page
);
1878 return ret
| VM_FAULT_RETRY
;
1881 /* Did it get truncated? */
1882 if (unlikely(page
->mapping
!= mapping
)) {
1887 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1890 * We have a locked page in the page cache, now we need to check
1891 * that it's up-to-date. If not, it is going to be due to an error.
1893 if (unlikely(!PageUptodate(page
)))
1894 goto page_not_uptodate
;
1897 * Found the page and have a reference on it.
1898 * We must recheck i_size under page lock.
1900 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1901 if (unlikely(offset
>= size
)) {
1903 page_cache_release(page
);
1904 return VM_FAULT_SIGBUS
;
1908 return ret
| VM_FAULT_LOCKED
;
1912 * We're only likely to ever get here if MADV_RANDOM is in
1915 error
= page_cache_read(file
, offset
);
1918 * The page we want has now been added to the page cache.
1919 * In the unlikely event that someone removed it in the
1920 * meantime, we'll just come back here and read it again.
1926 * An error return from page_cache_read can result if the
1927 * system is low on memory, or a problem occurs while trying
1930 if (error
== -ENOMEM
)
1931 return VM_FAULT_OOM
;
1932 return VM_FAULT_SIGBUS
;
1936 * Umm, take care of errors if the page isn't up-to-date.
1937 * Try to re-read it _once_. We do this synchronously,
1938 * because there really aren't any performance issues here
1939 * and we need to check for errors.
1941 ClearPageError(page
);
1942 error
= mapping
->a_ops
->readpage(file
, page
);
1944 wait_on_page_locked(page
);
1945 if (!PageUptodate(page
))
1948 page_cache_release(page
);
1950 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1953 /* Things didn't work out. Return zero to tell the mm layer so. */
1954 shrink_readahead_size_eio(file
, ra
);
1955 return VM_FAULT_SIGBUS
;
1957 EXPORT_SYMBOL(filemap_fault
);
1959 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1961 struct page
*page
= vmf
->page
;
1962 struct inode
*inode
= file_inode(vma
->vm_file
);
1963 int ret
= VM_FAULT_LOCKED
;
1965 sb_start_pagefault(inode
->i_sb
);
1966 file_update_time(vma
->vm_file
);
1968 if (page
->mapping
!= inode
->i_mapping
) {
1970 ret
= VM_FAULT_NOPAGE
;
1974 * We mark the page dirty already here so that when freeze is in
1975 * progress, we are guaranteed that writeback during freezing will
1976 * see the dirty page and writeprotect it again.
1978 set_page_dirty(page
);
1979 wait_for_stable_page(page
);
1981 sb_end_pagefault(inode
->i_sb
);
1984 EXPORT_SYMBOL(filemap_page_mkwrite
);
1986 const struct vm_operations_struct generic_file_vm_ops
= {
1987 .fault
= filemap_fault
,
1988 .page_mkwrite
= filemap_page_mkwrite
,
1989 .remap_pages
= generic_file_remap_pages
,
1992 /* This is used for a general mmap of a disk file */
1994 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1996 struct address_space
*mapping
= file
->f_mapping
;
1998 if (!mapping
->a_ops
->readpage
)
2000 file_accessed(file
);
2001 vma
->vm_ops
= &generic_file_vm_ops
;
2006 * This is for filesystems which do not implement ->writepage.
2008 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2010 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2012 return generic_file_mmap(file
, vma
);
2015 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2019 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2023 #endif /* CONFIG_MMU */
2025 EXPORT_SYMBOL(generic_file_mmap
);
2026 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2028 static struct page
*__read_cache_page(struct address_space
*mapping
,
2030 int (*filler
)(void *, struct page
*),
2037 page
= find_get_page(mapping
, index
);
2039 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2041 return ERR_PTR(-ENOMEM
);
2042 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2043 if (unlikely(err
)) {
2044 page_cache_release(page
);
2047 /* Presumably ENOMEM for radix tree node */
2048 return ERR_PTR(err
);
2050 err
= filler(data
, page
);
2052 page_cache_release(page
);
2053 page
= ERR_PTR(err
);
2059 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2061 int (*filler
)(void *, struct page
*),
2070 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2073 if (PageUptodate(page
))
2077 if (!page
->mapping
) {
2079 page_cache_release(page
);
2082 if (PageUptodate(page
)) {
2086 err
= filler(data
, page
);
2088 page_cache_release(page
);
2089 return ERR_PTR(err
);
2092 mark_page_accessed(page
);
2097 * read_cache_page_async - read into page cache, fill it if needed
2098 * @mapping: the page's address_space
2099 * @index: the page index
2100 * @filler: function to perform the read
2101 * @data: first arg to filler(data, page) function, often left as NULL
2103 * Same as read_cache_page, but don't wait for page to become unlocked
2104 * after submitting it to the filler.
2106 * Read into the page cache. If a page already exists, and PageUptodate() is
2107 * not set, try to fill the page but don't wait for it to become unlocked.
2109 * If the page does not get brought uptodate, return -EIO.
2111 struct page
*read_cache_page_async(struct address_space
*mapping
,
2113 int (*filler
)(void *, struct page
*),
2116 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2118 EXPORT_SYMBOL(read_cache_page_async
);
2120 static struct page
*wait_on_page_read(struct page
*page
)
2122 if (!IS_ERR(page
)) {
2123 wait_on_page_locked(page
);
2124 if (!PageUptodate(page
)) {
2125 page_cache_release(page
);
2126 page
= ERR_PTR(-EIO
);
2133 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2134 * @mapping: the page's address_space
2135 * @index: the page index
2136 * @gfp: the page allocator flags to use if allocating
2138 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2139 * any new page allocations done using the specified allocation flags.
2141 * If the page does not get brought uptodate, return -EIO.
2143 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2147 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2149 return wait_on_page_read(do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
));
2151 EXPORT_SYMBOL(read_cache_page_gfp
);
2154 * read_cache_page - read into page cache, fill it if needed
2155 * @mapping: the page's address_space
2156 * @index: the page index
2157 * @filler: function to perform the read
2158 * @data: first arg to filler(data, page) function, often left as NULL
2160 * Read into the page cache. If a page already exists, and PageUptodate() is
2161 * not set, try to fill the page then wait for it to become unlocked.
2163 * If the page does not get brought uptodate, return -EIO.
2165 struct page
*read_cache_page(struct address_space
*mapping
,
2167 int (*filler
)(void *, struct page
*),
2170 return wait_on_page_read(read_cache_page_async(mapping
, index
, filler
, data
));
2172 EXPORT_SYMBOL(read_cache_page
);
2174 static size_t __iovec_copy_from_user_inatomic(char *vaddr
,
2175 const struct iovec
*iov
, size_t base
, size_t bytes
)
2177 size_t copied
= 0, left
= 0;
2180 char __user
*buf
= iov
->iov_base
+ base
;
2181 int copy
= min(bytes
, iov
->iov_len
- base
);
2184 left
= __copy_from_user_inatomic(vaddr
, buf
, copy
);
2193 return copied
- left
;
2197 * Copy as much as we can into the page and return the number of bytes which
2198 * were successfully copied. If a fault is encountered then return the number of
2199 * bytes which were copied.
2201 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
2202 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2207 BUG_ON(!in_atomic());
2208 kaddr
= kmap_atomic(page
);
2209 if (likely(i
->nr_segs
== 1)) {
2211 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2212 left
= __copy_from_user_inatomic(kaddr
+ offset
, buf
, bytes
);
2213 copied
= bytes
- left
;
2215 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2216 i
->iov
, i
->iov_offset
, bytes
);
2218 kunmap_atomic(kaddr
);
2222 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
2225 * This has the same sideeffects and return value as
2226 * iov_iter_copy_from_user_atomic().
2227 * The difference is that it attempts to resolve faults.
2228 * Page must not be locked.
2230 size_t iov_iter_copy_from_user(struct page
*page
,
2231 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2237 if (likely(i
->nr_segs
== 1)) {
2239 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2240 left
= __copy_from_user(kaddr
+ offset
, buf
, bytes
);
2241 copied
= bytes
- left
;
2243 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2244 i
->iov
, i
->iov_offset
, bytes
);
2249 EXPORT_SYMBOL(iov_iter_copy_from_user
);
2251 void iov_iter_advance(struct iov_iter
*i
, size_t bytes
)
2253 BUG_ON(i
->count
< bytes
);
2255 if (likely(i
->nr_segs
== 1)) {
2256 i
->iov_offset
+= bytes
;
2259 const struct iovec
*iov
= i
->iov
;
2260 size_t base
= i
->iov_offset
;
2261 unsigned long nr_segs
= i
->nr_segs
;
2264 * The !iov->iov_len check ensures we skip over unlikely
2265 * zero-length segments (without overruning the iovec).
2267 while (bytes
|| unlikely(i
->count
&& !iov
->iov_len
)) {
2270 copy
= min(bytes
, iov
->iov_len
- base
);
2271 BUG_ON(!i
->count
|| i
->count
< copy
);
2275 if (iov
->iov_len
== base
) {
2282 i
->iov_offset
= base
;
2283 i
->nr_segs
= nr_segs
;
2286 EXPORT_SYMBOL(iov_iter_advance
);
2289 * Fault in the first iovec of the given iov_iter, to a maximum length
2290 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2291 * accessed (ie. because it is an invalid address).
2293 * writev-intensive code may want this to prefault several iovecs -- that
2294 * would be possible (callers must not rely on the fact that _only_ the
2295 * first iovec will be faulted with the current implementation).
2297 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
2299 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2300 bytes
= min(bytes
, i
->iov
->iov_len
- i
->iov_offset
);
2301 return fault_in_pages_readable(buf
, bytes
);
2303 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
2306 * Return the count of just the current iov_iter segment.
2308 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
2310 const struct iovec
*iov
= i
->iov
;
2311 if (i
->nr_segs
== 1)
2314 return min(i
->count
, iov
->iov_len
- i
->iov_offset
);
2316 EXPORT_SYMBOL(iov_iter_single_seg_count
);
2319 * Performs necessary checks before doing a write
2321 * Can adjust writing position or amount of bytes to write.
2322 * Returns appropriate error code that caller should return or
2323 * zero in case that write should be allowed.
2325 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
2327 struct inode
*inode
= file
->f_mapping
->host
;
2328 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2330 if (unlikely(*pos
< 0))
2334 /* FIXME: this is for backwards compatibility with 2.4 */
2335 if (file
->f_flags
& O_APPEND
)
2336 *pos
= i_size_read(inode
);
2338 if (limit
!= RLIM_INFINITY
) {
2339 if (*pos
>= limit
) {
2340 send_sig(SIGXFSZ
, current
, 0);
2343 if (*count
> limit
- (typeof(limit
))*pos
) {
2344 *count
= limit
- (typeof(limit
))*pos
;
2352 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2353 !(file
->f_flags
& O_LARGEFILE
))) {
2354 if (*pos
>= MAX_NON_LFS
) {
2357 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2358 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2363 * Are we about to exceed the fs block limit ?
2365 * If we have written data it becomes a short write. If we have
2366 * exceeded without writing data we send a signal and return EFBIG.
2367 * Linus frestrict idea will clean these up nicely..
2369 if (likely(!isblk
)) {
2370 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2371 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2374 /* zero-length writes at ->s_maxbytes are OK */
2377 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2378 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2382 if (bdev_read_only(I_BDEV(inode
)))
2384 isize
= i_size_read(inode
);
2385 if (*pos
>= isize
) {
2386 if (*count
|| *pos
> isize
)
2390 if (*pos
+ *count
> isize
)
2391 *count
= isize
- *pos
;
2398 EXPORT_SYMBOL(generic_write_checks
);
2400 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2401 loff_t pos
, unsigned len
, unsigned flags
,
2402 struct page
**pagep
, void **fsdata
)
2404 const struct address_space_operations
*aops
= mapping
->a_ops
;
2406 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2409 EXPORT_SYMBOL(pagecache_write_begin
);
2411 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2412 loff_t pos
, unsigned len
, unsigned copied
,
2413 struct page
*page
, void *fsdata
)
2415 const struct address_space_operations
*aops
= mapping
->a_ops
;
2417 mark_page_accessed(page
);
2418 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2420 EXPORT_SYMBOL(pagecache_write_end
);
2423 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2424 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2425 size_t count
, size_t ocount
)
2427 struct file
*file
= iocb
->ki_filp
;
2428 struct address_space
*mapping
= file
->f_mapping
;
2429 struct inode
*inode
= mapping
->host
;
2434 if (count
!= ocount
)
2435 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2437 write_len
= iov_length(iov
, *nr_segs
);
2438 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2440 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2445 * After a write we want buffered reads to be sure to go to disk to get
2446 * the new data. We invalidate clean cached page from the region we're
2447 * about to write. We do this *before* the write so that we can return
2448 * without clobbering -EIOCBQUEUED from ->direct_IO().
2450 if (mapping
->nrpages
) {
2451 written
= invalidate_inode_pages2_range(mapping
,
2452 pos
>> PAGE_CACHE_SHIFT
, end
);
2454 * If a page can not be invalidated, return 0 to fall back
2455 * to buffered write.
2458 if (written
== -EBUSY
)
2464 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2467 * Finally, try again to invalidate clean pages which might have been
2468 * cached by non-direct readahead, or faulted in by get_user_pages()
2469 * if the source of the write was an mmap'ed region of the file
2470 * we're writing. Either one is a pretty crazy thing to do,
2471 * so we don't support it 100%. If this invalidation
2472 * fails, tough, the write still worked...
2474 if (mapping
->nrpages
) {
2475 invalidate_inode_pages2_range(mapping
,
2476 pos
>> PAGE_CACHE_SHIFT
, end
);
2481 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2482 i_size_write(inode
, pos
);
2483 mark_inode_dirty(inode
);
2490 EXPORT_SYMBOL(generic_file_direct_write
);
2493 * Find or create a page at the given pagecache position. Return the locked
2494 * page. This function is specifically for buffered writes.
2496 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2497 pgoff_t index
, unsigned flags
)
2502 gfp_t gfp_notmask
= 0;
2504 gfp_mask
= mapping_gfp_mask(mapping
);
2505 if (mapping_cap_account_dirty(mapping
))
2506 gfp_mask
|= __GFP_WRITE
;
2507 if (flags
& AOP_FLAG_NOFS
)
2508 gfp_notmask
= __GFP_FS
;
2510 page
= find_lock_page(mapping
, index
);
2514 page
= __page_cache_alloc(gfp_mask
& ~gfp_notmask
);
2517 status
= add_to_page_cache_lru(page
, mapping
, index
,
2518 GFP_KERNEL
& ~gfp_notmask
);
2519 if (unlikely(status
)) {
2520 page_cache_release(page
);
2521 if (status
== -EEXIST
)
2526 wait_for_stable_page(page
);
2529 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2531 static ssize_t
generic_perform_write(struct file
*file
,
2532 struct iov_iter
*i
, loff_t pos
)
2534 struct address_space
*mapping
= file
->f_mapping
;
2535 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2537 ssize_t written
= 0;
2538 unsigned int flags
= 0;
2541 * Copies from kernel address space cannot fail (NFSD is a big user).
2543 if (segment_eq(get_fs(), KERNEL_DS
))
2544 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2548 unsigned long offset
; /* Offset into pagecache page */
2549 unsigned long bytes
; /* Bytes to write to page */
2550 size_t copied
; /* Bytes copied from user */
2553 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2554 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2559 * Bring in the user page that we will copy from _first_.
2560 * Otherwise there's a nasty deadlock on copying from the
2561 * same page as we're writing to, without it being marked
2564 * Not only is this an optimisation, but it is also required
2565 * to check that the address is actually valid, when atomic
2566 * usercopies are used, below.
2568 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2573 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2575 if (unlikely(status
))
2578 if (mapping_writably_mapped(mapping
))
2579 flush_dcache_page(page
);
2581 pagefault_disable();
2582 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2584 flush_dcache_page(page
);
2586 mark_page_accessed(page
);
2587 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2589 if (unlikely(status
< 0))
2595 iov_iter_advance(i
, copied
);
2596 if (unlikely(copied
== 0)) {
2598 * If we were unable to copy any data at all, we must
2599 * fall back to a single segment length write.
2601 * If we didn't fallback here, we could livelock
2602 * because not all segments in the iov can be copied at
2603 * once without a pagefault.
2605 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2606 iov_iter_single_seg_count(i
));
2612 balance_dirty_pages_ratelimited(mapping
);
2613 if (fatal_signal_pending(current
)) {
2617 } while (iov_iter_count(i
));
2619 return written
? written
: status
;
2623 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2624 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2625 size_t count
, ssize_t written
)
2627 struct file
*file
= iocb
->ki_filp
;
2631 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
2632 status
= generic_perform_write(file
, &i
, pos
);
2634 if (likely(status
>= 0)) {
2636 *ppos
= pos
+ status
;
2639 return written
? written
: status
;
2641 EXPORT_SYMBOL(generic_file_buffered_write
);
2644 * __generic_file_aio_write - write data to a file
2645 * @iocb: IO state structure (file, offset, etc.)
2646 * @iov: vector with data to write
2647 * @nr_segs: number of segments in the vector
2648 * @ppos: position where to write
2650 * This function does all the work needed for actually writing data to a
2651 * file. It does all basic checks, removes SUID from the file, updates
2652 * modification times and calls proper subroutines depending on whether we
2653 * do direct IO or a standard buffered write.
2655 * It expects i_mutex to be grabbed unless we work on a block device or similar
2656 * object which does not need locking at all.
2658 * This function does *not* take care of syncing data in case of O_SYNC write.
2659 * A caller has to handle it. This is mainly due to the fact that we want to
2660 * avoid syncing under i_mutex.
2662 ssize_t
__generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2663 unsigned long nr_segs
, loff_t
*ppos
)
2665 struct file
*file
= iocb
->ki_filp
;
2666 struct address_space
* mapping
= file
->f_mapping
;
2667 size_t ocount
; /* original count */
2668 size_t count
; /* after file limit checks */
2669 struct inode
*inode
= mapping
->host
;
2675 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
2682 /* We can write back this queue in page reclaim */
2683 current
->backing_dev_info
= mapping
->backing_dev_info
;
2686 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2693 err
= file_remove_suid(file
);
2697 err
= file_update_time(file
);
2701 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2702 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2704 ssize_t written_buffered
;
2706 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2707 ppos
, count
, ocount
);
2708 if (written
< 0 || written
== count
)
2711 * direct-io write to a hole: fall through to buffered I/O
2712 * for completing the rest of the request.
2716 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2717 nr_segs
, pos
, ppos
, count
,
2720 * If generic_file_buffered_write() retuned a synchronous error
2721 * then we want to return the number of bytes which were
2722 * direct-written, or the error code if that was zero. Note
2723 * that this differs from normal direct-io semantics, which
2724 * will return -EFOO even if some bytes were written.
2726 if (written_buffered
< 0) {
2727 err
= written_buffered
;
2732 * We need to ensure that the page cache pages are written to
2733 * disk and invalidated to preserve the expected O_DIRECT
2736 endbyte
= pos
+ written_buffered
- written
- 1;
2737 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
2739 written
= written_buffered
;
2740 invalidate_mapping_pages(mapping
,
2741 pos
>> PAGE_CACHE_SHIFT
,
2742 endbyte
>> PAGE_CACHE_SHIFT
);
2745 * We don't know how much we wrote, so just return
2746 * the number of bytes which were direct-written
2750 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2751 pos
, ppos
, count
, written
);
2754 current
->backing_dev_info
= NULL
;
2755 return written
? written
: err
;
2757 EXPORT_SYMBOL(__generic_file_aio_write
);
2760 * generic_file_aio_write - write data to a file
2761 * @iocb: IO state structure
2762 * @iov: vector with data to write
2763 * @nr_segs: number of segments in the vector
2764 * @pos: position in file where to write
2766 * This is a wrapper around __generic_file_aio_write() to be used by most
2767 * filesystems. It takes care of syncing the file in case of O_SYNC file
2768 * and acquires i_mutex as needed.
2770 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2771 unsigned long nr_segs
, loff_t pos
)
2773 struct file
*file
= iocb
->ki_filp
;
2774 struct inode
*inode
= file
->f_mapping
->host
;
2777 BUG_ON(iocb
->ki_pos
!= pos
);
2779 mutex_lock(&inode
->i_mutex
);
2780 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
, &iocb
->ki_pos
);
2781 mutex_unlock(&inode
->i_mutex
);
2786 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2792 EXPORT_SYMBOL(generic_file_aio_write
);
2795 * try_to_release_page() - release old fs-specific metadata on a page
2797 * @page: the page which the kernel is trying to free
2798 * @gfp_mask: memory allocation flags (and I/O mode)
2800 * The address_space is to try to release any data against the page
2801 * (presumably at page->private). If the release was successful, return `1'.
2802 * Otherwise return zero.
2804 * This may also be called if PG_fscache is set on a page, indicating that the
2805 * page is known to the local caching routines.
2807 * The @gfp_mask argument specifies whether I/O may be performed to release
2808 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2811 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2813 struct address_space
* const mapping
= page
->mapping
;
2815 BUG_ON(!PageLocked(page
));
2816 if (PageWriteback(page
))
2819 if (mapping
&& mapping
->a_ops
->releasepage
)
2820 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2821 return try_to_free_buffers(page
);
2824 EXPORT_SYMBOL(try_to_release_page
);