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)
110 static void page_cache_tree_delete(struct address_space
*mapping
,
111 struct page
*page
, void *shadow
)
116 slot
= radix_tree_lookup_slot(&mapping
->page_tree
, page
->index
);
117 radix_tree_replace_slot(slot
, shadow
);
118 mapping
->nrshadows
++;
120 * Make sure the nrshadows update is committed before
121 * the nrpages update so that final truncate racing
122 * with reclaim does not see both counters 0 at the
123 * same time and miss a shadow entry.
127 radix_tree_delete(&mapping
->page_tree
, page
->index
);
132 * Delete a page from the page cache and free it. Caller has to make
133 * sure the page is locked and that nobody else uses it - or that usage
134 * is safe. The caller must hold the mapping's tree_lock.
136 void __delete_from_page_cache(struct page
*page
, void *shadow
)
138 struct address_space
*mapping
= page
->mapping
;
140 trace_mm_filemap_delete_from_page_cache(page
);
142 * if we're uptodate, flush out into the cleancache, otherwise
143 * invalidate any existing cleancache entries. We can't leave
144 * stale data around in the cleancache once our page is gone
146 if (PageUptodate(page
) && PageMappedToDisk(page
))
147 cleancache_put_page(page
);
149 cleancache_invalidate_page(mapping
, page
);
151 page_cache_tree_delete(mapping
, page
, shadow
);
153 page
->mapping
= NULL
;
154 /* Leave page->index set: truncation lookup relies upon it */
156 __dec_zone_page_state(page
, NR_FILE_PAGES
);
157 if (PageSwapBacked(page
))
158 __dec_zone_page_state(page
, NR_SHMEM
);
159 BUG_ON(page_mapped(page
));
162 * Some filesystems seem to re-dirty the page even after
163 * the VM has canceled the dirty bit (eg ext3 journaling).
165 * Fix it up by doing a final dirty accounting check after
166 * having removed the page entirely.
168 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
169 dec_zone_page_state(page
, NR_FILE_DIRTY
);
170 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
175 * delete_from_page_cache - delete page from page cache
176 * @page: the page which the kernel is trying to remove from page cache
178 * This must be called only on pages that have been verified to be in the page
179 * cache and locked. It will never put the page into the free list, the caller
180 * has a reference on the page.
182 void delete_from_page_cache(struct page
*page
)
184 struct address_space
*mapping
= page
->mapping
;
185 void (*freepage
)(struct page
*);
187 BUG_ON(!PageLocked(page
));
189 freepage
= mapping
->a_ops
->freepage
;
190 spin_lock_irq(&mapping
->tree_lock
);
191 __delete_from_page_cache(page
, NULL
);
192 spin_unlock_irq(&mapping
->tree_lock
);
193 mem_cgroup_uncharge_cache_page(page
);
197 page_cache_release(page
);
199 EXPORT_SYMBOL(delete_from_page_cache
);
201 static int sleep_on_page(void *word
)
207 static int sleep_on_page_killable(void *word
)
210 return fatal_signal_pending(current
) ? -EINTR
: 0;
213 static int filemap_check_errors(struct address_space
*mapping
)
216 /* Check for outstanding write errors */
217 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
219 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
225 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
226 * @mapping: address space structure to write
227 * @start: offset in bytes where the range starts
228 * @end: offset in bytes where the range ends (inclusive)
229 * @sync_mode: enable synchronous operation
231 * Start writeback against all of a mapping's dirty pages that lie
232 * within the byte offsets <start, end> inclusive.
234 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
235 * opposed to a regular memory cleansing writeback. The difference between
236 * these two operations is that if a dirty page/buffer is encountered, it must
237 * be waited upon, and not just skipped over.
239 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
240 loff_t end
, int sync_mode
)
243 struct writeback_control wbc
= {
244 .sync_mode
= sync_mode
,
245 .nr_to_write
= LONG_MAX
,
246 .range_start
= start
,
250 if (!mapping_cap_writeback_dirty(mapping
))
253 ret
= do_writepages(mapping
, &wbc
);
257 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
260 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
263 int filemap_fdatawrite(struct address_space
*mapping
)
265 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
267 EXPORT_SYMBOL(filemap_fdatawrite
);
269 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
272 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
274 EXPORT_SYMBOL(filemap_fdatawrite_range
);
277 * filemap_flush - mostly a non-blocking flush
278 * @mapping: target address_space
280 * This is a mostly non-blocking flush. Not suitable for data-integrity
281 * purposes - I/O may not be started against all dirty pages.
283 int filemap_flush(struct address_space
*mapping
)
285 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
287 EXPORT_SYMBOL(filemap_flush
);
290 * filemap_fdatawait_range - wait for writeback to complete
291 * @mapping: address space structure to wait for
292 * @start_byte: offset in bytes where the range starts
293 * @end_byte: offset in bytes where the range ends (inclusive)
295 * Walk the list of under-writeback pages of the given address space
296 * in the given range and wait for all of them.
298 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
301 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
302 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
307 if (end_byte
< start_byte
)
310 pagevec_init(&pvec
, 0);
311 while ((index
<= end
) &&
312 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
313 PAGECACHE_TAG_WRITEBACK
,
314 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
317 for (i
= 0; i
< nr_pages
; i
++) {
318 struct page
*page
= pvec
.pages
[i
];
320 /* until radix tree lookup accepts end_index */
321 if (page
->index
> end
)
324 wait_on_page_writeback(page
);
325 if (TestClearPageError(page
))
328 pagevec_release(&pvec
);
332 ret2
= filemap_check_errors(mapping
);
338 EXPORT_SYMBOL(filemap_fdatawait_range
);
341 * filemap_fdatawait - wait for all under-writeback pages to complete
342 * @mapping: address space structure to wait for
344 * Walk the list of under-writeback pages of the given address space
345 * and wait for all of them.
347 int filemap_fdatawait(struct address_space
*mapping
)
349 loff_t i_size
= i_size_read(mapping
->host
);
354 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
356 EXPORT_SYMBOL(filemap_fdatawait
);
358 int filemap_write_and_wait(struct address_space
*mapping
)
362 if (mapping
->nrpages
) {
363 err
= filemap_fdatawrite(mapping
);
365 * Even if the above returned error, the pages may be
366 * written partially (e.g. -ENOSPC), so we wait for it.
367 * But the -EIO is special case, it may indicate the worst
368 * thing (e.g. bug) happened, so we avoid waiting for it.
371 int err2
= filemap_fdatawait(mapping
);
376 err
= filemap_check_errors(mapping
);
380 EXPORT_SYMBOL(filemap_write_and_wait
);
383 * filemap_write_and_wait_range - write out & wait on a file range
384 * @mapping: the address_space for the pages
385 * @lstart: offset in bytes where the range starts
386 * @lend: offset in bytes where the range ends (inclusive)
388 * Write out and wait upon file offsets lstart->lend, inclusive.
390 * Note that `lend' is inclusive (describes the last byte to be written) so
391 * that this function can be used to write to the very end-of-file (end = -1).
393 int filemap_write_and_wait_range(struct address_space
*mapping
,
394 loff_t lstart
, loff_t lend
)
398 if (mapping
->nrpages
) {
399 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
401 /* See comment of filemap_write_and_wait() */
403 int err2
= filemap_fdatawait_range(mapping
,
409 err
= filemap_check_errors(mapping
);
413 EXPORT_SYMBOL(filemap_write_and_wait_range
);
416 * replace_page_cache_page - replace a pagecache page with a new one
417 * @old: page to be replaced
418 * @new: page to replace with
419 * @gfp_mask: allocation mode
421 * This function replaces a page in the pagecache with a new one. On
422 * success it acquires the pagecache reference for the new page and
423 * drops it for the old page. Both the old and new pages must be
424 * locked. This function does not add the new page to the LRU, the
425 * caller must do that.
427 * The remove + add is atomic. The only way this function can fail is
428 * memory allocation failure.
430 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
434 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
435 VM_BUG_ON_PAGE(!PageLocked(new), new);
436 VM_BUG_ON_PAGE(new->mapping
, new);
438 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
440 struct address_space
*mapping
= old
->mapping
;
441 void (*freepage
)(struct page
*);
443 pgoff_t offset
= old
->index
;
444 freepage
= mapping
->a_ops
->freepage
;
447 new->mapping
= mapping
;
450 spin_lock_irq(&mapping
->tree_lock
);
451 __delete_from_page_cache(old
, NULL
);
452 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
455 __inc_zone_page_state(new, NR_FILE_PAGES
);
456 if (PageSwapBacked(new))
457 __inc_zone_page_state(new, NR_SHMEM
);
458 spin_unlock_irq(&mapping
->tree_lock
);
459 /* mem_cgroup codes must not be called under tree_lock */
460 mem_cgroup_replace_page_cache(old
, new);
461 radix_tree_preload_end();
464 page_cache_release(old
);
469 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
471 static int page_cache_tree_insert(struct address_space
*mapping
,
477 slot
= radix_tree_lookup_slot(&mapping
->page_tree
, page
->index
);
481 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
482 if (!radix_tree_exceptional_entry(p
))
484 radix_tree_replace_slot(slot
, page
);
485 mapping
->nrshadows
--;
489 error
= radix_tree_insert(&mapping
->page_tree
, page
->index
, page
);
496 * add_to_page_cache_locked - add a locked page to the pagecache
498 * @mapping: the page's address_space
499 * @offset: page index
500 * @gfp_mask: page allocation mode
502 * This function is used to add a page to the pagecache. It must be locked.
503 * This function does not add the page to the LRU. The caller must do that.
505 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
506 pgoff_t offset
, gfp_t gfp_mask
)
510 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
511 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
513 error
= mem_cgroup_cache_charge(page
, current
->mm
,
514 gfp_mask
& GFP_RECLAIM_MASK
);
518 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
520 mem_cgroup_uncharge_cache_page(page
);
524 page_cache_get(page
);
525 page
->mapping
= mapping
;
526 page
->index
= offset
;
528 spin_lock_irq(&mapping
->tree_lock
);
529 error
= page_cache_tree_insert(mapping
, page
);
530 radix_tree_preload_end();
533 __inc_zone_page_state(page
, NR_FILE_PAGES
);
534 spin_unlock_irq(&mapping
->tree_lock
);
535 trace_mm_filemap_add_to_page_cache(page
);
538 page
->mapping
= NULL
;
539 /* Leave page->index set: truncation relies upon it */
540 spin_unlock_irq(&mapping
->tree_lock
);
541 mem_cgroup_uncharge_cache_page(page
);
542 page_cache_release(page
);
545 EXPORT_SYMBOL(add_to_page_cache_locked
);
547 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
548 pgoff_t offset
, gfp_t gfp_mask
)
552 ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
554 lru_cache_add_file(page
);
557 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
560 struct page
*__page_cache_alloc(gfp_t gfp
)
565 if (cpuset_do_page_mem_spread()) {
566 unsigned int cpuset_mems_cookie
;
568 cpuset_mems_cookie
= read_mems_allowed_begin();
569 n
= cpuset_mem_spread_node();
570 page
= alloc_pages_exact_node(n
, gfp
, 0);
571 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
575 return alloc_pages(gfp
, 0);
577 EXPORT_SYMBOL(__page_cache_alloc
);
581 * In order to wait for pages to become available there must be
582 * waitqueues associated with pages. By using a hash table of
583 * waitqueues where the bucket discipline is to maintain all
584 * waiters on the same queue and wake all when any of the pages
585 * become available, and for the woken contexts to check to be
586 * sure the appropriate page became available, this saves space
587 * at a cost of "thundering herd" phenomena during rare hash
590 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
592 const struct zone
*zone
= page_zone(page
);
594 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
597 static inline void wake_up_page(struct page
*page
, int bit
)
599 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
602 void wait_on_page_bit(struct page
*page
, int bit_nr
)
604 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
606 if (test_bit(bit_nr
, &page
->flags
))
607 __wait_on_bit(page_waitqueue(page
), &wait
, sleep_on_page
,
608 TASK_UNINTERRUPTIBLE
);
610 EXPORT_SYMBOL(wait_on_page_bit
);
612 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
614 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
616 if (!test_bit(bit_nr
, &page
->flags
))
619 return __wait_on_bit(page_waitqueue(page
), &wait
,
620 sleep_on_page_killable
, TASK_KILLABLE
);
624 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
625 * @page: Page defining the wait queue of interest
626 * @waiter: Waiter to add to the queue
628 * Add an arbitrary @waiter to the wait queue for the nominated @page.
630 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
632 wait_queue_head_t
*q
= page_waitqueue(page
);
635 spin_lock_irqsave(&q
->lock
, flags
);
636 __add_wait_queue(q
, waiter
);
637 spin_unlock_irqrestore(&q
->lock
, flags
);
639 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
642 * unlock_page - unlock a locked page
645 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
646 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
647 * mechananism between PageLocked pages and PageWriteback pages is shared.
648 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
650 * The mb is necessary to enforce ordering between the clear_bit and the read
651 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
653 void unlock_page(struct page
*page
)
655 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
656 clear_bit_unlock(PG_locked
, &page
->flags
);
657 smp_mb__after_clear_bit();
658 wake_up_page(page
, PG_locked
);
660 EXPORT_SYMBOL(unlock_page
);
663 * end_page_writeback - end writeback against a page
666 void end_page_writeback(struct page
*page
)
668 if (TestClearPageReclaim(page
))
669 rotate_reclaimable_page(page
);
671 if (!test_clear_page_writeback(page
))
674 smp_mb__after_clear_bit();
675 wake_up_page(page
, PG_writeback
);
677 EXPORT_SYMBOL(end_page_writeback
);
680 * __lock_page - get a lock on the page, assuming we need to sleep to get it
681 * @page: the page to lock
683 void __lock_page(struct page
*page
)
685 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
687 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sleep_on_page
,
688 TASK_UNINTERRUPTIBLE
);
690 EXPORT_SYMBOL(__lock_page
);
692 int __lock_page_killable(struct page
*page
)
694 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
696 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
697 sleep_on_page_killable
, TASK_KILLABLE
);
699 EXPORT_SYMBOL_GPL(__lock_page_killable
);
701 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
704 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
706 * CAUTION! In this case, mmap_sem is not released
707 * even though return 0.
709 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
712 up_read(&mm
->mmap_sem
);
713 if (flags
& FAULT_FLAG_KILLABLE
)
714 wait_on_page_locked_killable(page
);
716 wait_on_page_locked(page
);
719 if (flags
& FAULT_FLAG_KILLABLE
) {
722 ret
= __lock_page_killable(page
);
724 up_read(&mm
->mmap_sem
);
734 * page_cache_next_hole - find the next hole (not-present entry)
737 * @max_scan: maximum range to search
739 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
740 * lowest indexed hole.
742 * Returns: the index of the hole if found, otherwise returns an index
743 * outside of the set specified (in which case 'return - index >=
744 * max_scan' will be true). In rare cases of index wrap-around, 0 will
747 * page_cache_next_hole may be called under rcu_read_lock. However,
748 * like radix_tree_gang_lookup, this will not atomically search a
749 * snapshot of the tree at a single point in time. For example, if a
750 * hole is created at index 5, then subsequently a hole is created at
751 * index 10, page_cache_next_hole covering both indexes may return 10
752 * if called under rcu_read_lock.
754 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
755 pgoff_t index
, unsigned long max_scan
)
759 for (i
= 0; i
< max_scan
; i
++) {
762 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
763 if (!page
|| radix_tree_exceptional_entry(page
))
772 EXPORT_SYMBOL(page_cache_next_hole
);
775 * page_cache_prev_hole - find the prev hole (not-present entry)
778 * @max_scan: maximum range to search
780 * Search backwards in the range [max(index-max_scan+1, 0), index] for
783 * Returns: the index of the hole if found, otherwise returns an index
784 * outside of the set specified (in which case 'index - return >=
785 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
788 * page_cache_prev_hole may be called under rcu_read_lock. However,
789 * like radix_tree_gang_lookup, this will not atomically search a
790 * snapshot of the tree at a single point in time. For example, if a
791 * hole is created at index 10, then subsequently a hole is created at
792 * index 5, page_cache_prev_hole covering both indexes may return 5 if
793 * called under rcu_read_lock.
795 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
796 pgoff_t index
, unsigned long max_scan
)
800 for (i
= 0; i
< max_scan
; i
++) {
803 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
804 if (!page
|| radix_tree_exceptional_entry(page
))
807 if (index
== ULONG_MAX
)
813 EXPORT_SYMBOL(page_cache_prev_hole
);
816 * find_get_entry - find and get a page cache entry
817 * @mapping: the address_space to search
818 * @offset: the page cache index
820 * Looks up the page cache slot at @mapping & @offset. If there is a
821 * page cache page, it is returned with an increased refcount.
823 * If the slot holds a shadow entry of a previously evicted page, it
826 * Otherwise, %NULL is returned.
828 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
836 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
838 page
= radix_tree_deref_slot(pagep
);
841 if (radix_tree_exception(page
)) {
842 if (radix_tree_deref_retry(page
))
845 * Otherwise, shmem/tmpfs must be storing a swap entry
846 * here as an exceptional entry: so return it without
847 * attempting to raise page count.
851 if (!page_cache_get_speculative(page
))
855 * Has the page moved?
856 * This is part of the lockless pagecache protocol. See
857 * include/linux/pagemap.h for details.
859 if (unlikely(page
!= *pagep
)) {
860 page_cache_release(page
);
869 EXPORT_SYMBOL(find_get_entry
);
872 * find_get_page - find and get a page reference
873 * @mapping: the address_space to search
874 * @offset: the page index
876 * Looks up the page cache slot at @mapping & @offset. If there is a
877 * page cache page, it is returned with an increased refcount.
879 * Otherwise, %NULL is returned.
881 struct page
*find_get_page(struct address_space
*mapping
, pgoff_t offset
)
883 struct page
*page
= find_get_entry(mapping
, offset
);
885 if (radix_tree_exceptional_entry(page
))
889 EXPORT_SYMBOL(find_get_page
);
892 * find_lock_entry - locate, pin and lock a page cache entry
893 * @mapping: the address_space to search
894 * @offset: the page cache index
896 * Looks up the page cache slot at @mapping & @offset. If there is a
897 * page cache page, it is returned locked and with an increased
900 * If the slot holds a shadow entry of a previously evicted page, it
903 * Otherwise, %NULL is returned.
905 * find_lock_entry() may sleep.
907 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
912 page
= find_get_entry(mapping
, offset
);
913 if (page
&& !radix_tree_exception(page
)) {
915 /* Has the page been truncated? */
916 if (unlikely(page
->mapping
!= mapping
)) {
918 page_cache_release(page
);
921 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
925 EXPORT_SYMBOL(find_lock_entry
);
928 * find_lock_page - locate, pin and lock a pagecache page
929 * @mapping: the address_space to search
930 * @offset: the page index
932 * Looks up the page cache slot at @mapping & @offset. If there is a
933 * page cache page, it is returned locked and with an increased
936 * Otherwise, %NULL is returned.
938 * find_lock_page() may sleep.
940 struct page
*find_lock_page(struct address_space
*mapping
, pgoff_t offset
)
942 struct page
*page
= find_lock_entry(mapping
, offset
);
944 if (radix_tree_exceptional_entry(page
))
948 EXPORT_SYMBOL(find_lock_page
);
951 * find_or_create_page - locate or add a pagecache page
952 * @mapping: the page's address_space
953 * @index: the page's index into the mapping
954 * @gfp_mask: page allocation mode
956 * Looks up the page cache slot at @mapping & @offset. If there is a
957 * page cache page, it is returned locked and with an increased
960 * If the page is not present, a new page is allocated using @gfp_mask
961 * and added to the page cache and the VM's LRU list. The page is
962 * returned locked and with an increased refcount.
964 * On memory exhaustion, %NULL is returned.
966 * find_or_create_page() may sleep, even if @gfp_flags specifies an
969 struct page
*find_or_create_page(struct address_space
*mapping
,
970 pgoff_t index
, gfp_t gfp_mask
)
975 page
= find_lock_page(mapping
, index
);
977 page
= __page_cache_alloc(gfp_mask
);
981 * We want a regular kernel memory (not highmem or DMA etc)
982 * allocation for the radix tree nodes, but we need to honour
983 * the context-specific requirements the caller has asked for.
984 * GFP_RECLAIM_MASK collects those requirements.
986 err
= add_to_page_cache_lru(page
, mapping
, index
,
987 (gfp_mask
& GFP_RECLAIM_MASK
));
989 page_cache_release(page
);
997 EXPORT_SYMBOL(find_or_create_page
);
1000 * find_get_entries - gang pagecache lookup
1001 * @mapping: The address_space to search
1002 * @start: The starting page cache index
1003 * @nr_entries: The maximum number of entries
1004 * @entries: Where the resulting entries are placed
1005 * @indices: The cache indices corresponding to the entries in @entries
1007 * find_get_entries() will search for and return a group of up to
1008 * @nr_entries entries in the mapping. The entries are placed at
1009 * @entries. find_get_entries() takes a reference against any actual
1012 * The search returns a group of mapping-contiguous page cache entries
1013 * with ascending indexes. There may be holes in the indices due to
1014 * not-present pages.
1016 * Any shadow entries of evicted pages are included in the returned
1019 * find_get_entries() returns the number of pages and shadow entries
1022 unsigned find_get_entries(struct address_space
*mapping
,
1023 pgoff_t start
, unsigned int nr_entries
,
1024 struct page
**entries
, pgoff_t
*indices
)
1027 unsigned int ret
= 0;
1028 struct radix_tree_iter iter
;
1035 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1038 page
= radix_tree_deref_slot(slot
);
1039 if (unlikely(!page
))
1041 if (radix_tree_exception(page
)) {
1042 if (radix_tree_deref_retry(page
))
1045 * Otherwise, we must be storing a swap entry
1046 * here as an exceptional entry: so return it
1047 * without attempting to raise page count.
1051 if (!page_cache_get_speculative(page
))
1054 /* Has the page moved? */
1055 if (unlikely(page
!= *slot
)) {
1056 page_cache_release(page
);
1060 indices
[ret
] = iter
.index
;
1061 entries
[ret
] = page
;
1062 if (++ret
== nr_entries
)
1070 * find_get_pages - gang pagecache lookup
1071 * @mapping: The address_space to search
1072 * @start: The starting page index
1073 * @nr_pages: The maximum number of pages
1074 * @pages: Where the resulting pages are placed
1076 * find_get_pages() will search for and return a group of up to
1077 * @nr_pages pages in the mapping. The pages are placed at @pages.
1078 * find_get_pages() takes a reference against the returned pages.
1080 * The search returns a group of mapping-contiguous pages with ascending
1081 * indexes. There may be holes in the indices due to not-present pages.
1083 * find_get_pages() returns the number of pages which were found.
1085 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1086 unsigned int nr_pages
, struct page
**pages
)
1088 struct radix_tree_iter iter
;
1092 if (unlikely(!nr_pages
))
1097 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1100 page
= radix_tree_deref_slot(slot
);
1101 if (unlikely(!page
))
1104 if (radix_tree_exception(page
)) {
1105 if (radix_tree_deref_retry(page
)) {
1107 * Transient condition which can only trigger
1108 * when entry at index 0 moves out of or back
1109 * to root: none yet gotten, safe to restart.
1111 WARN_ON(iter
.index
);
1115 * Otherwise, shmem/tmpfs must be storing a swap entry
1116 * here as an exceptional entry: so skip over it -
1117 * we only reach this from invalidate_mapping_pages().
1122 if (!page_cache_get_speculative(page
))
1125 /* Has the page moved? */
1126 if (unlikely(page
!= *slot
)) {
1127 page_cache_release(page
);
1132 if (++ret
== nr_pages
)
1141 * find_get_pages_contig - gang contiguous pagecache lookup
1142 * @mapping: The address_space to search
1143 * @index: The starting page index
1144 * @nr_pages: The maximum number of pages
1145 * @pages: Where the resulting pages are placed
1147 * find_get_pages_contig() works exactly like find_get_pages(), except
1148 * that the returned number of pages are guaranteed to be contiguous.
1150 * find_get_pages_contig() returns the number of pages which were found.
1152 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1153 unsigned int nr_pages
, struct page
**pages
)
1155 struct radix_tree_iter iter
;
1157 unsigned int ret
= 0;
1159 if (unlikely(!nr_pages
))
1164 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1167 page
= radix_tree_deref_slot(slot
);
1168 /* The hole, there no reason to continue */
1169 if (unlikely(!page
))
1172 if (radix_tree_exception(page
)) {
1173 if (radix_tree_deref_retry(page
)) {
1175 * Transient condition which can only trigger
1176 * when entry at index 0 moves out of or back
1177 * to root: none yet gotten, safe to restart.
1182 * Otherwise, shmem/tmpfs must be storing a swap entry
1183 * here as an exceptional entry: so stop looking for
1189 if (!page_cache_get_speculative(page
))
1192 /* Has the page moved? */
1193 if (unlikely(page
!= *slot
)) {
1194 page_cache_release(page
);
1199 * must check mapping and index after taking the ref.
1200 * otherwise we can get both false positives and false
1201 * negatives, which is just confusing to the caller.
1203 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1204 page_cache_release(page
);
1209 if (++ret
== nr_pages
)
1215 EXPORT_SYMBOL(find_get_pages_contig
);
1218 * find_get_pages_tag - find and return pages that match @tag
1219 * @mapping: the address_space to search
1220 * @index: the starting page index
1221 * @tag: the tag index
1222 * @nr_pages: the maximum number of pages
1223 * @pages: where the resulting pages are placed
1225 * Like find_get_pages, except we only return pages which are tagged with
1226 * @tag. We update @index to index the next page for the traversal.
1228 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1229 int tag
, unsigned int nr_pages
, struct page
**pages
)
1231 struct radix_tree_iter iter
;
1235 if (unlikely(!nr_pages
))
1240 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1241 &iter
, *index
, tag
) {
1244 page
= radix_tree_deref_slot(slot
);
1245 if (unlikely(!page
))
1248 if (radix_tree_exception(page
)) {
1249 if (radix_tree_deref_retry(page
)) {
1251 * Transient condition which can only trigger
1252 * when entry at index 0 moves out of or back
1253 * to root: none yet gotten, safe to restart.
1258 * This function is never used on a shmem/tmpfs
1259 * mapping, so a swap entry won't be found here.
1264 if (!page_cache_get_speculative(page
))
1267 /* Has the page moved? */
1268 if (unlikely(page
!= *slot
)) {
1269 page_cache_release(page
);
1274 if (++ret
== nr_pages
)
1281 *index
= pages
[ret
- 1]->index
+ 1;
1285 EXPORT_SYMBOL(find_get_pages_tag
);
1288 * grab_cache_page_nowait - returns locked page at given index in given cache
1289 * @mapping: target address_space
1290 * @index: the page index
1292 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1293 * This is intended for speculative data generators, where the data can
1294 * be regenerated if the page couldn't be grabbed. This routine should
1295 * be safe to call while holding the lock for another page.
1297 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1298 * and deadlock against the caller's locked page.
1301 grab_cache_page_nowait(struct address_space
*mapping
, pgoff_t index
)
1303 struct page
*page
= find_get_page(mapping
, index
);
1306 if (trylock_page(page
))
1308 page_cache_release(page
);
1311 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
1312 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_NOFS
)) {
1313 page_cache_release(page
);
1318 EXPORT_SYMBOL(grab_cache_page_nowait
);
1321 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1322 * a _large_ part of the i/o request. Imagine the worst scenario:
1324 * ---R__________________________________________B__________
1325 * ^ reading here ^ bad block(assume 4k)
1327 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1328 * => failing the whole request => read(R) => read(R+1) =>
1329 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1330 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1331 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1333 * It is going insane. Fix it by quickly scaling down the readahead size.
1335 static void shrink_readahead_size_eio(struct file
*filp
,
1336 struct file_ra_state
*ra
)
1342 * do_generic_file_read - generic file read routine
1343 * @filp: the file to read
1344 * @ppos: current file position
1345 * @desc: read_descriptor
1347 * This is a generic file read routine, and uses the
1348 * mapping->a_ops->readpage() function for the actual low-level stuff.
1350 * This is really ugly. But the goto's actually try to clarify some
1351 * of the logic when it comes to error handling etc.
1353 static void do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1354 read_descriptor_t
*desc
)
1356 struct address_space
*mapping
= filp
->f_mapping
;
1357 struct inode
*inode
= mapping
->host
;
1358 struct file_ra_state
*ra
= &filp
->f_ra
;
1362 unsigned long offset
; /* offset into pagecache page */
1363 unsigned int prev_offset
;
1366 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1367 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1368 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1369 last_index
= (*ppos
+ desc
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1370 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1376 unsigned long nr
, ret
;
1380 page
= find_get_page(mapping
, index
);
1382 page_cache_sync_readahead(mapping
,
1384 index
, last_index
- index
);
1385 page
= find_get_page(mapping
, index
);
1386 if (unlikely(page
== NULL
))
1387 goto no_cached_page
;
1389 if (PageReadahead(page
)) {
1390 page_cache_async_readahead(mapping
,
1392 index
, last_index
- index
);
1394 if (!PageUptodate(page
)) {
1395 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1396 !mapping
->a_ops
->is_partially_uptodate
)
1397 goto page_not_up_to_date
;
1398 if (!trylock_page(page
))
1399 goto page_not_up_to_date
;
1400 /* Did it get truncated before we got the lock? */
1402 goto page_not_up_to_date_locked
;
1403 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1405 goto page_not_up_to_date_locked
;
1410 * i_size must be checked after we know the page is Uptodate.
1412 * Checking i_size after the check allows us to calculate
1413 * the correct value for "nr", which means the zero-filled
1414 * part of the page is not copied back to userspace (unless
1415 * another truncate extends the file - this is desired though).
1418 isize
= i_size_read(inode
);
1419 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1420 if (unlikely(!isize
|| index
> end_index
)) {
1421 page_cache_release(page
);
1425 /* nr is the maximum number of bytes to copy from this page */
1426 nr
= PAGE_CACHE_SIZE
;
1427 if (index
== end_index
) {
1428 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1430 page_cache_release(page
);
1436 /* If users can be writing to this page using arbitrary
1437 * virtual addresses, take care about potential aliasing
1438 * before reading the page on the kernel side.
1440 if (mapping_writably_mapped(mapping
))
1441 flush_dcache_page(page
);
1444 * When a sequential read accesses a page several times,
1445 * only mark it as accessed the first time.
1447 if (prev_index
!= index
|| offset
!= prev_offset
)
1448 mark_page_accessed(page
);
1452 * Ok, we have the page, and it's up-to-date, so
1453 * now we can copy it to user space...
1455 * The file_read_actor routine returns how many bytes were
1457 * NOTE! This may not be the same as how much of a user buffer
1458 * we filled up (we may be padding etc), so we can only update
1459 * "pos" here (the actor routine has to update the user buffer
1460 * pointers and the remaining count).
1462 ret
= file_read_actor(desc
, page
, offset
, nr
);
1464 index
+= offset
>> PAGE_CACHE_SHIFT
;
1465 offset
&= ~PAGE_CACHE_MASK
;
1466 prev_offset
= offset
;
1468 page_cache_release(page
);
1469 if (ret
== nr
&& desc
->count
)
1473 page_not_up_to_date
:
1474 /* Get exclusive access to the page ... */
1475 error
= lock_page_killable(page
);
1476 if (unlikely(error
))
1477 goto readpage_error
;
1479 page_not_up_to_date_locked
:
1480 /* Did it get truncated before we got the lock? */
1481 if (!page
->mapping
) {
1483 page_cache_release(page
);
1487 /* Did somebody else fill it already? */
1488 if (PageUptodate(page
)) {
1495 * A previous I/O error may have been due to temporary
1496 * failures, eg. multipath errors.
1497 * PG_error will be set again if readpage fails.
1499 ClearPageError(page
);
1500 /* Start the actual read. The read will unlock the page. */
1501 error
= mapping
->a_ops
->readpage(filp
, page
);
1503 if (unlikely(error
)) {
1504 if (error
== AOP_TRUNCATED_PAGE
) {
1505 page_cache_release(page
);
1508 goto readpage_error
;
1511 if (!PageUptodate(page
)) {
1512 error
= lock_page_killable(page
);
1513 if (unlikely(error
))
1514 goto readpage_error
;
1515 if (!PageUptodate(page
)) {
1516 if (page
->mapping
== NULL
) {
1518 * invalidate_mapping_pages got it
1521 page_cache_release(page
);
1525 shrink_readahead_size_eio(filp
, ra
);
1527 goto readpage_error
;
1535 /* UHHUH! A synchronous read error occurred. Report it */
1536 desc
->error
= error
;
1537 page_cache_release(page
);
1542 * Ok, it wasn't cached, so we need to create a new
1545 page
= page_cache_alloc_cold(mapping
);
1547 desc
->error
= -ENOMEM
;
1550 error
= add_to_page_cache_lru(page
, mapping
,
1553 page_cache_release(page
);
1554 if (error
== -EEXIST
)
1556 desc
->error
= error
;
1563 ra
->prev_pos
= prev_index
;
1564 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1565 ra
->prev_pos
|= prev_offset
;
1567 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1568 file_accessed(filp
);
1571 int file_read_actor(read_descriptor_t
*desc
, struct page
*page
,
1572 unsigned long offset
, unsigned long size
)
1575 unsigned long left
, count
= desc
->count
;
1581 * Faults on the destination of a read are common, so do it before
1584 if (!fault_in_pages_writeable(desc
->arg
.buf
, size
)) {
1585 kaddr
= kmap_atomic(page
);
1586 left
= __copy_to_user_inatomic(desc
->arg
.buf
,
1587 kaddr
+ offset
, size
);
1588 kunmap_atomic(kaddr
);
1593 /* Do it the slow way */
1595 left
= __copy_to_user(desc
->arg
.buf
, kaddr
+ offset
, size
);
1600 desc
->error
= -EFAULT
;
1603 desc
->count
= count
- size
;
1604 desc
->written
+= size
;
1605 desc
->arg
.buf
+= size
;
1610 * Performs necessary checks before doing a write
1611 * @iov: io vector request
1612 * @nr_segs: number of segments in the iovec
1613 * @count: number of bytes to write
1614 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1616 * Adjust number of segments and amount of bytes to write (nr_segs should be
1617 * properly initialized first). Returns appropriate error code that caller
1618 * should return or zero in case that write should be allowed.
1620 int generic_segment_checks(const struct iovec
*iov
,
1621 unsigned long *nr_segs
, size_t *count
, int access_flags
)
1625 for (seg
= 0; seg
< *nr_segs
; seg
++) {
1626 const struct iovec
*iv
= &iov
[seg
];
1629 * If any segment has a negative length, or the cumulative
1630 * length ever wraps negative then return -EINVAL.
1633 if (unlikely((ssize_t
)(cnt
|iv
->iov_len
) < 0))
1635 if (access_ok(access_flags
, iv
->iov_base
, iv
->iov_len
))
1640 cnt
-= iv
->iov_len
; /* This segment is no good */
1646 EXPORT_SYMBOL(generic_segment_checks
);
1649 * generic_file_aio_read - generic filesystem read routine
1650 * @iocb: kernel I/O control block
1651 * @iov: io vector request
1652 * @nr_segs: number of segments in the iovec
1653 * @pos: current file position
1655 * This is the "read()" routine for all filesystems
1656 * that can use the page cache directly.
1659 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1660 unsigned long nr_segs
, loff_t pos
)
1662 struct file
*filp
= iocb
->ki_filp
;
1664 unsigned long seg
= 0;
1666 loff_t
*ppos
= &iocb
->ki_pos
;
1669 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1673 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1674 if (filp
->f_flags
& O_DIRECT
) {
1676 struct address_space
*mapping
;
1677 struct inode
*inode
;
1679 mapping
= filp
->f_mapping
;
1680 inode
= mapping
->host
;
1682 goto out
; /* skip atime */
1683 size
= i_size_read(inode
);
1684 retval
= filemap_write_and_wait_range(mapping
, pos
,
1685 pos
+ iov_length(iov
, nr_segs
) - 1);
1687 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
,
1691 *ppos
= pos
+ retval
;
1696 * Btrfs can have a short DIO read if we encounter
1697 * compressed extents, so if there was an error, or if
1698 * we've already read everything we wanted to, or if
1699 * there was a short read because we hit EOF, go ahead
1700 * and return. Otherwise fallthrough to buffered io for
1701 * the rest of the read.
1703 if (retval
< 0 || !count
|| *ppos
>= size
) {
1704 file_accessed(filp
);
1710 for (seg
= 0; seg
< nr_segs
; seg
++) {
1711 read_descriptor_t desc
;
1715 * If we did a short DIO read we need to skip the section of the
1716 * iov that we've already read data into.
1719 if (count
> iov
[seg
].iov_len
) {
1720 count
-= iov
[seg
].iov_len
;
1728 desc
.arg
.buf
= iov
[seg
].iov_base
+ offset
;
1729 desc
.count
= iov
[seg
].iov_len
- offset
;
1730 if (desc
.count
== 0)
1733 do_generic_file_read(filp
, ppos
, &desc
);
1734 retval
+= desc
.written
;
1736 retval
= retval
?: desc
.error
;
1745 EXPORT_SYMBOL(generic_file_aio_read
);
1749 * page_cache_read - adds requested page to the page cache if not already there
1750 * @file: file to read
1751 * @offset: page index
1753 * This adds the requested page to the page cache if it isn't already there,
1754 * and schedules an I/O to read in its contents from disk.
1756 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1758 struct address_space
*mapping
= file
->f_mapping
;
1763 page
= page_cache_alloc_cold(mapping
);
1767 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1769 ret
= mapping
->a_ops
->readpage(file
, page
);
1770 else if (ret
== -EEXIST
)
1771 ret
= 0; /* losing race to add is OK */
1773 page_cache_release(page
);
1775 } while (ret
== AOP_TRUNCATED_PAGE
);
1780 #define MMAP_LOTSAMISS (100)
1783 * Synchronous readahead happens when we don't even find
1784 * a page in the page cache at all.
1786 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1787 struct file_ra_state
*ra
,
1791 unsigned long ra_pages
;
1792 struct address_space
*mapping
= file
->f_mapping
;
1794 /* If we don't want any read-ahead, don't bother */
1795 if (vma
->vm_flags
& VM_RAND_READ
)
1800 if (vma
->vm_flags
& VM_SEQ_READ
) {
1801 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1806 /* Avoid banging the cache line if not needed */
1807 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1811 * Do we miss much more than hit in this file? If so,
1812 * stop bothering with read-ahead. It will only hurt.
1814 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1820 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1821 ra
->start
= max_t(long, 0, offset
- ra_pages
/ 2);
1822 ra
->size
= ra_pages
;
1823 ra
->async_size
= ra_pages
/ 4;
1824 ra_submit(ra
, mapping
, file
);
1828 * Asynchronous readahead happens when we find the page and PG_readahead,
1829 * so we want to possibly extend the readahead further..
1831 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1832 struct file_ra_state
*ra
,
1837 struct address_space
*mapping
= file
->f_mapping
;
1839 /* If we don't want any read-ahead, don't bother */
1840 if (vma
->vm_flags
& VM_RAND_READ
)
1842 if (ra
->mmap_miss
> 0)
1844 if (PageReadahead(page
))
1845 page_cache_async_readahead(mapping
, ra
, file
,
1846 page
, offset
, ra
->ra_pages
);
1850 * filemap_fault - read in file data for page fault handling
1851 * @vma: vma in which the fault was taken
1852 * @vmf: struct vm_fault containing details of the fault
1854 * filemap_fault() is invoked via the vma operations vector for a
1855 * mapped memory region to read in file data during a page fault.
1857 * The goto's are kind of ugly, but this streamlines the normal case of having
1858 * it in the page cache, and handles the special cases reasonably without
1859 * having a lot of duplicated code.
1861 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1864 struct file
*file
= vma
->vm_file
;
1865 struct address_space
*mapping
= file
->f_mapping
;
1866 struct file_ra_state
*ra
= &file
->f_ra
;
1867 struct inode
*inode
= mapping
->host
;
1868 pgoff_t offset
= vmf
->pgoff
;
1873 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1875 return VM_FAULT_SIGBUS
;
1878 * Do we have something in the page cache already?
1880 page
= find_get_page(mapping
, offset
);
1881 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1883 * We found the page, so try async readahead before
1884 * waiting for the lock.
1886 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1888 /* No page in the page cache at all */
1889 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1890 count_vm_event(PGMAJFAULT
);
1891 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1892 ret
= VM_FAULT_MAJOR
;
1894 page
= find_get_page(mapping
, offset
);
1896 goto no_cached_page
;
1899 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1900 page_cache_release(page
);
1901 return ret
| VM_FAULT_RETRY
;
1904 /* Did it get truncated? */
1905 if (unlikely(page
->mapping
!= mapping
)) {
1910 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1913 * We have a locked page in the page cache, now we need to check
1914 * that it's up-to-date. If not, it is going to be due to an error.
1916 if (unlikely(!PageUptodate(page
)))
1917 goto page_not_uptodate
;
1920 * Found the page and have a reference on it.
1921 * We must recheck i_size under page lock.
1923 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1924 if (unlikely(offset
>= size
)) {
1926 page_cache_release(page
);
1927 return VM_FAULT_SIGBUS
;
1931 return ret
| VM_FAULT_LOCKED
;
1935 * We're only likely to ever get here if MADV_RANDOM is in
1938 error
= page_cache_read(file
, offset
);
1941 * The page we want has now been added to the page cache.
1942 * In the unlikely event that someone removed it in the
1943 * meantime, we'll just come back here and read it again.
1949 * An error return from page_cache_read can result if the
1950 * system is low on memory, or a problem occurs while trying
1953 if (error
== -ENOMEM
)
1954 return VM_FAULT_OOM
;
1955 return VM_FAULT_SIGBUS
;
1959 * Umm, take care of errors if the page isn't up-to-date.
1960 * Try to re-read it _once_. We do this synchronously,
1961 * because there really aren't any performance issues here
1962 * and we need to check for errors.
1964 ClearPageError(page
);
1965 error
= mapping
->a_ops
->readpage(file
, page
);
1967 wait_on_page_locked(page
);
1968 if (!PageUptodate(page
))
1971 page_cache_release(page
);
1973 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1976 /* Things didn't work out. Return zero to tell the mm layer so. */
1977 shrink_readahead_size_eio(file
, ra
);
1978 return VM_FAULT_SIGBUS
;
1980 EXPORT_SYMBOL(filemap_fault
);
1982 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1984 struct page
*page
= vmf
->page
;
1985 struct inode
*inode
= file_inode(vma
->vm_file
);
1986 int ret
= VM_FAULT_LOCKED
;
1988 sb_start_pagefault(inode
->i_sb
);
1989 file_update_time(vma
->vm_file
);
1991 if (page
->mapping
!= inode
->i_mapping
) {
1993 ret
= VM_FAULT_NOPAGE
;
1997 * We mark the page dirty already here so that when freeze is in
1998 * progress, we are guaranteed that writeback during freezing will
1999 * see the dirty page and writeprotect it again.
2001 set_page_dirty(page
);
2002 wait_for_stable_page(page
);
2004 sb_end_pagefault(inode
->i_sb
);
2007 EXPORT_SYMBOL(filemap_page_mkwrite
);
2009 const struct vm_operations_struct generic_file_vm_ops
= {
2010 .fault
= filemap_fault
,
2011 .page_mkwrite
= filemap_page_mkwrite
,
2012 .remap_pages
= generic_file_remap_pages
,
2015 /* This is used for a general mmap of a disk file */
2017 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2019 struct address_space
*mapping
= file
->f_mapping
;
2021 if (!mapping
->a_ops
->readpage
)
2023 file_accessed(file
);
2024 vma
->vm_ops
= &generic_file_vm_ops
;
2029 * This is for filesystems which do not implement ->writepage.
2031 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2033 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2035 return generic_file_mmap(file
, vma
);
2038 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2042 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2046 #endif /* CONFIG_MMU */
2048 EXPORT_SYMBOL(generic_file_mmap
);
2049 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2051 static struct page
*__read_cache_page(struct address_space
*mapping
,
2053 int (*filler
)(void *, struct page
*),
2060 page
= find_get_page(mapping
, index
);
2062 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2064 return ERR_PTR(-ENOMEM
);
2065 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2066 if (unlikely(err
)) {
2067 page_cache_release(page
);
2070 /* Presumably ENOMEM for radix tree node */
2071 return ERR_PTR(err
);
2073 err
= filler(data
, page
);
2075 page_cache_release(page
);
2076 page
= ERR_PTR(err
);
2082 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2084 int (*filler
)(void *, struct page
*),
2093 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2096 if (PageUptodate(page
))
2100 if (!page
->mapping
) {
2102 page_cache_release(page
);
2105 if (PageUptodate(page
)) {
2109 err
= filler(data
, page
);
2111 page_cache_release(page
);
2112 return ERR_PTR(err
);
2115 mark_page_accessed(page
);
2120 * read_cache_page_async - read into page cache, fill it if needed
2121 * @mapping: the page's address_space
2122 * @index: the page index
2123 * @filler: function to perform the read
2124 * @data: first arg to filler(data, page) function, often left as NULL
2126 * Same as read_cache_page, but don't wait for page to become unlocked
2127 * after submitting it to the filler.
2129 * Read into the page cache. If a page already exists, and PageUptodate() is
2130 * not set, try to fill the page but don't wait for it to become unlocked.
2132 * If the page does not get brought uptodate, return -EIO.
2134 struct page
*read_cache_page_async(struct address_space
*mapping
,
2136 int (*filler
)(void *, struct page
*),
2139 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2141 EXPORT_SYMBOL(read_cache_page_async
);
2143 static struct page
*wait_on_page_read(struct page
*page
)
2145 if (!IS_ERR(page
)) {
2146 wait_on_page_locked(page
);
2147 if (!PageUptodate(page
)) {
2148 page_cache_release(page
);
2149 page
= ERR_PTR(-EIO
);
2156 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2157 * @mapping: the page's address_space
2158 * @index: the page index
2159 * @gfp: the page allocator flags to use if allocating
2161 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2162 * any new page allocations done using the specified allocation flags.
2164 * If the page does not get brought uptodate, return -EIO.
2166 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2170 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2172 return wait_on_page_read(do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
));
2174 EXPORT_SYMBOL(read_cache_page_gfp
);
2177 * read_cache_page - read into page cache, fill it if needed
2178 * @mapping: the page's address_space
2179 * @index: the page index
2180 * @filler: function to perform the read
2181 * @data: first arg to filler(data, page) function, often left as NULL
2183 * Read into the page cache. If a page already exists, and PageUptodate() is
2184 * not set, try to fill the page then wait for it to become unlocked.
2186 * If the page does not get brought uptodate, return -EIO.
2188 struct page
*read_cache_page(struct address_space
*mapping
,
2190 int (*filler
)(void *, struct page
*),
2193 return wait_on_page_read(read_cache_page_async(mapping
, index
, filler
, data
));
2195 EXPORT_SYMBOL(read_cache_page
);
2197 static size_t __iovec_copy_from_user_inatomic(char *vaddr
,
2198 const struct iovec
*iov
, size_t base
, size_t bytes
)
2200 size_t copied
= 0, left
= 0;
2203 char __user
*buf
= iov
->iov_base
+ base
;
2204 int copy
= min(bytes
, iov
->iov_len
- base
);
2207 left
= __copy_from_user_inatomic(vaddr
, buf
, copy
);
2216 return copied
- left
;
2220 * Copy as much as we can into the page and return the number of bytes which
2221 * were successfully copied. If a fault is encountered then return the number of
2222 * bytes which were copied.
2224 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
2225 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2230 BUG_ON(!in_atomic());
2231 kaddr
= kmap_atomic(page
);
2232 if (likely(i
->nr_segs
== 1)) {
2234 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2235 left
= __copy_from_user_inatomic(kaddr
+ offset
, buf
, bytes
);
2236 copied
= bytes
- left
;
2238 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2239 i
->iov
, i
->iov_offset
, bytes
);
2241 kunmap_atomic(kaddr
);
2245 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
2248 * This has the same sideeffects and return value as
2249 * iov_iter_copy_from_user_atomic().
2250 * The difference is that it attempts to resolve faults.
2251 * Page must not be locked.
2253 size_t iov_iter_copy_from_user(struct page
*page
,
2254 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2260 if (likely(i
->nr_segs
== 1)) {
2262 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2263 left
= __copy_from_user(kaddr
+ offset
, buf
, bytes
);
2264 copied
= bytes
- left
;
2266 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2267 i
->iov
, i
->iov_offset
, bytes
);
2272 EXPORT_SYMBOL(iov_iter_copy_from_user
);
2274 void iov_iter_advance(struct iov_iter
*i
, size_t bytes
)
2276 BUG_ON(i
->count
< bytes
);
2278 if (likely(i
->nr_segs
== 1)) {
2279 i
->iov_offset
+= bytes
;
2282 const struct iovec
*iov
= i
->iov
;
2283 size_t base
= i
->iov_offset
;
2284 unsigned long nr_segs
= i
->nr_segs
;
2287 * The !iov->iov_len check ensures we skip over unlikely
2288 * zero-length segments (without overruning the iovec).
2290 while (bytes
|| unlikely(i
->count
&& !iov
->iov_len
)) {
2293 copy
= min(bytes
, iov
->iov_len
- base
);
2294 BUG_ON(!i
->count
|| i
->count
< copy
);
2298 if (iov
->iov_len
== base
) {
2305 i
->iov_offset
= base
;
2306 i
->nr_segs
= nr_segs
;
2309 EXPORT_SYMBOL(iov_iter_advance
);
2312 * Fault in the first iovec of the given iov_iter, to a maximum length
2313 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2314 * accessed (ie. because it is an invalid address).
2316 * writev-intensive code may want this to prefault several iovecs -- that
2317 * would be possible (callers must not rely on the fact that _only_ the
2318 * first iovec will be faulted with the current implementation).
2320 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
2322 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2323 bytes
= min(bytes
, i
->iov
->iov_len
- i
->iov_offset
);
2324 return fault_in_pages_readable(buf
, bytes
);
2326 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
2329 * Return the count of just the current iov_iter segment.
2331 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
2333 const struct iovec
*iov
= i
->iov
;
2334 if (i
->nr_segs
== 1)
2337 return min(i
->count
, iov
->iov_len
- i
->iov_offset
);
2339 EXPORT_SYMBOL(iov_iter_single_seg_count
);
2342 * Performs necessary checks before doing a write
2344 * Can adjust writing position or amount of bytes to write.
2345 * Returns appropriate error code that caller should return or
2346 * zero in case that write should be allowed.
2348 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
2350 struct inode
*inode
= file
->f_mapping
->host
;
2351 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2353 if (unlikely(*pos
< 0))
2357 /* FIXME: this is for backwards compatibility with 2.4 */
2358 if (file
->f_flags
& O_APPEND
)
2359 *pos
= i_size_read(inode
);
2361 if (limit
!= RLIM_INFINITY
) {
2362 if (*pos
>= limit
) {
2363 send_sig(SIGXFSZ
, current
, 0);
2366 if (*count
> limit
- (typeof(limit
))*pos
) {
2367 *count
= limit
- (typeof(limit
))*pos
;
2375 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2376 !(file
->f_flags
& O_LARGEFILE
))) {
2377 if (*pos
>= MAX_NON_LFS
) {
2380 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2381 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2386 * Are we about to exceed the fs block limit ?
2388 * If we have written data it becomes a short write. If we have
2389 * exceeded without writing data we send a signal and return EFBIG.
2390 * Linus frestrict idea will clean these up nicely..
2392 if (likely(!isblk
)) {
2393 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2394 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2397 /* zero-length writes at ->s_maxbytes are OK */
2400 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2401 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2405 if (bdev_read_only(I_BDEV(inode
)))
2407 isize
= i_size_read(inode
);
2408 if (*pos
>= isize
) {
2409 if (*count
|| *pos
> isize
)
2413 if (*pos
+ *count
> isize
)
2414 *count
= isize
- *pos
;
2421 EXPORT_SYMBOL(generic_write_checks
);
2423 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2424 loff_t pos
, unsigned len
, unsigned flags
,
2425 struct page
**pagep
, void **fsdata
)
2427 const struct address_space_operations
*aops
= mapping
->a_ops
;
2429 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2432 EXPORT_SYMBOL(pagecache_write_begin
);
2434 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2435 loff_t pos
, unsigned len
, unsigned copied
,
2436 struct page
*page
, void *fsdata
)
2438 const struct address_space_operations
*aops
= mapping
->a_ops
;
2440 mark_page_accessed(page
);
2441 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2443 EXPORT_SYMBOL(pagecache_write_end
);
2446 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2447 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2448 size_t count
, size_t ocount
)
2450 struct file
*file
= iocb
->ki_filp
;
2451 struct address_space
*mapping
= file
->f_mapping
;
2452 struct inode
*inode
= mapping
->host
;
2457 if (count
!= ocount
)
2458 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2460 write_len
= iov_length(iov
, *nr_segs
);
2461 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2463 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2468 * After a write we want buffered reads to be sure to go to disk to get
2469 * the new data. We invalidate clean cached page from the region we're
2470 * about to write. We do this *before* the write so that we can return
2471 * without clobbering -EIOCBQUEUED from ->direct_IO().
2473 if (mapping
->nrpages
) {
2474 written
= invalidate_inode_pages2_range(mapping
,
2475 pos
>> PAGE_CACHE_SHIFT
, end
);
2477 * If a page can not be invalidated, return 0 to fall back
2478 * to buffered write.
2481 if (written
== -EBUSY
)
2487 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2490 * Finally, try again to invalidate clean pages which might have been
2491 * cached by non-direct readahead, or faulted in by get_user_pages()
2492 * if the source of the write was an mmap'ed region of the file
2493 * we're writing. Either one is a pretty crazy thing to do,
2494 * so we don't support it 100%. If this invalidation
2495 * fails, tough, the write still worked...
2497 if (mapping
->nrpages
) {
2498 invalidate_inode_pages2_range(mapping
,
2499 pos
>> PAGE_CACHE_SHIFT
, end
);
2504 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2505 i_size_write(inode
, pos
);
2506 mark_inode_dirty(inode
);
2513 EXPORT_SYMBOL(generic_file_direct_write
);
2516 * Find or create a page at the given pagecache position. Return the locked
2517 * page. This function is specifically for buffered writes.
2519 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2520 pgoff_t index
, unsigned flags
)
2525 gfp_t gfp_notmask
= 0;
2527 gfp_mask
= mapping_gfp_mask(mapping
);
2528 if (mapping_cap_account_dirty(mapping
))
2529 gfp_mask
|= __GFP_WRITE
;
2530 if (flags
& AOP_FLAG_NOFS
)
2531 gfp_notmask
= __GFP_FS
;
2533 page
= find_lock_page(mapping
, index
);
2537 page
= __page_cache_alloc(gfp_mask
& ~gfp_notmask
);
2540 status
= add_to_page_cache_lru(page
, mapping
, index
,
2541 GFP_KERNEL
& ~gfp_notmask
);
2542 if (unlikely(status
)) {
2543 page_cache_release(page
);
2544 if (status
== -EEXIST
)
2549 wait_for_stable_page(page
);
2552 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2554 static ssize_t
generic_perform_write(struct file
*file
,
2555 struct iov_iter
*i
, loff_t pos
)
2557 struct address_space
*mapping
= file
->f_mapping
;
2558 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2560 ssize_t written
= 0;
2561 unsigned int flags
= 0;
2564 * Copies from kernel address space cannot fail (NFSD is a big user).
2566 if (segment_eq(get_fs(), KERNEL_DS
))
2567 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2571 unsigned long offset
; /* Offset into pagecache page */
2572 unsigned long bytes
; /* Bytes to write to page */
2573 size_t copied
; /* Bytes copied from user */
2576 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2577 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2582 * Bring in the user page that we will copy from _first_.
2583 * Otherwise there's a nasty deadlock on copying from the
2584 * same page as we're writing to, without it being marked
2587 * Not only is this an optimisation, but it is also required
2588 * to check that the address is actually valid, when atomic
2589 * usercopies are used, below.
2591 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2596 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2598 if (unlikely(status
))
2601 if (mapping_writably_mapped(mapping
))
2602 flush_dcache_page(page
);
2604 pagefault_disable();
2605 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2607 flush_dcache_page(page
);
2609 mark_page_accessed(page
);
2610 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2612 if (unlikely(status
< 0))
2618 iov_iter_advance(i
, copied
);
2619 if (unlikely(copied
== 0)) {
2621 * If we were unable to copy any data at all, we must
2622 * fall back to a single segment length write.
2624 * If we didn't fallback here, we could livelock
2625 * because not all segments in the iov can be copied at
2626 * once without a pagefault.
2628 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2629 iov_iter_single_seg_count(i
));
2635 balance_dirty_pages_ratelimited(mapping
);
2636 if (fatal_signal_pending(current
)) {
2640 } while (iov_iter_count(i
));
2642 return written
? written
: status
;
2646 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2647 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2648 size_t count
, ssize_t written
)
2650 struct file
*file
= iocb
->ki_filp
;
2654 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
2655 status
= generic_perform_write(file
, &i
, pos
);
2657 if (likely(status
>= 0)) {
2659 *ppos
= pos
+ status
;
2662 return written
? written
: status
;
2664 EXPORT_SYMBOL(generic_file_buffered_write
);
2667 * __generic_file_aio_write - write data to a file
2668 * @iocb: IO state structure (file, offset, etc.)
2669 * @iov: vector with data to write
2670 * @nr_segs: number of segments in the vector
2671 * @ppos: position where to write
2673 * This function does all the work needed for actually writing data to a
2674 * file. It does all basic checks, removes SUID from the file, updates
2675 * modification times and calls proper subroutines depending on whether we
2676 * do direct IO or a standard buffered write.
2678 * It expects i_mutex to be grabbed unless we work on a block device or similar
2679 * object which does not need locking at all.
2681 * This function does *not* take care of syncing data in case of O_SYNC write.
2682 * A caller has to handle it. This is mainly due to the fact that we want to
2683 * avoid syncing under i_mutex.
2685 ssize_t
__generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2686 unsigned long nr_segs
, loff_t
*ppos
)
2688 struct file
*file
= iocb
->ki_filp
;
2689 struct address_space
* mapping
= file
->f_mapping
;
2690 size_t ocount
; /* original count */
2691 size_t count
; /* after file limit checks */
2692 struct inode
*inode
= mapping
->host
;
2698 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
2705 /* We can write back this queue in page reclaim */
2706 current
->backing_dev_info
= mapping
->backing_dev_info
;
2709 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2716 err
= file_remove_suid(file
);
2720 err
= file_update_time(file
);
2724 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2725 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2727 ssize_t written_buffered
;
2729 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2730 ppos
, count
, ocount
);
2731 if (written
< 0 || written
== count
)
2734 * direct-io write to a hole: fall through to buffered I/O
2735 * for completing the rest of the request.
2739 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2740 nr_segs
, pos
, ppos
, count
,
2743 * If generic_file_buffered_write() retuned a synchronous error
2744 * then we want to return the number of bytes which were
2745 * direct-written, or the error code if that was zero. Note
2746 * that this differs from normal direct-io semantics, which
2747 * will return -EFOO even if some bytes were written.
2749 if (written_buffered
< 0) {
2750 err
= written_buffered
;
2755 * We need to ensure that the page cache pages are written to
2756 * disk and invalidated to preserve the expected O_DIRECT
2759 endbyte
= pos
+ written_buffered
- written
- 1;
2760 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
2762 written
= written_buffered
;
2763 invalidate_mapping_pages(mapping
,
2764 pos
>> PAGE_CACHE_SHIFT
,
2765 endbyte
>> PAGE_CACHE_SHIFT
);
2768 * We don't know how much we wrote, so just return
2769 * the number of bytes which were direct-written
2773 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2774 pos
, ppos
, count
, written
);
2777 current
->backing_dev_info
= NULL
;
2778 return written
? written
: err
;
2780 EXPORT_SYMBOL(__generic_file_aio_write
);
2783 * generic_file_aio_write - write data to a file
2784 * @iocb: IO state structure
2785 * @iov: vector with data to write
2786 * @nr_segs: number of segments in the vector
2787 * @pos: position in file where to write
2789 * This is a wrapper around __generic_file_aio_write() to be used by most
2790 * filesystems. It takes care of syncing the file in case of O_SYNC file
2791 * and acquires i_mutex as needed.
2793 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2794 unsigned long nr_segs
, loff_t pos
)
2796 struct file
*file
= iocb
->ki_filp
;
2797 struct inode
*inode
= file
->f_mapping
->host
;
2800 BUG_ON(iocb
->ki_pos
!= pos
);
2802 mutex_lock(&inode
->i_mutex
);
2803 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
, &iocb
->ki_pos
);
2804 mutex_unlock(&inode
->i_mutex
);
2809 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2815 EXPORT_SYMBOL(generic_file_aio_write
);
2818 * try_to_release_page() - release old fs-specific metadata on a page
2820 * @page: the page which the kernel is trying to free
2821 * @gfp_mask: memory allocation flags (and I/O mode)
2823 * The address_space is to try to release any data against the page
2824 * (presumably at page->private). If the release was successful, return `1'.
2825 * Otherwise return zero.
2827 * This may also be called if PG_fscache is set on a page, indicating that the
2828 * page is known to the local caching routines.
2830 * The @gfp_mask argument specifies whether I/O may be performed to release
2831 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2834 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2836 struct address_space
* const mapping
= page
->mapping
;
2838 BUG_ON(!PageLocked(page
));
2839 if (PageWriteback(page
))
2842 if (mapping
&& mapping
->a_ops
->releasepage
)
2843 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2844 return try_to_free_buffers(page
);
2847 EXPORT_SYMBOL(try_to_release_page
);