4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/capability.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/gfp.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/backing-dev.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/cpuset.h>
32 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
33 #include <linux/hugetlb.h>
34 #include <linux/memcontrol.h>
35 #include <linux/cleancache.h>
36 #include <linux/rmap.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/filemap.h>
43 * FIXME: remove all knowledge of the buffer layer from the core VM
45 #include <linux/buffer_head.h> /* for try_to_free_buffers */
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * Shared mappings now work. 15.8.1995 Bruno.
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
64 * ->i_mmap_rwsem (truncate_pagecache)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
70 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
74 * ->page_table_lock or pte_lock (various, mainly in memory.c)
75 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
78 * ->lock_page (access_process_vm)
80 * ->i_mutex (generic_perform_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
84 * sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
103 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
112 static int page_cache_tree_insert(struct address_space
*mapping
,
113 struct page
*page
, void **shadowp
)
115 struct radix_tree_node
*node
;
119 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
,
126 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
127 if (!radix_tree_exceptional_entry(p
))
131 mapping
->nrshadows
--;
133 workingset_node_shadows_dec(node
);
135 radix_tree_replace_slot(slot
, page
);
138 workingset_node_pages_inc(node
);
140 * Don't track node that contains actual pages.
142 * Avoid acquiring the list_lru lock if already
143 * untracked. The list_empty() test is safe as
144 * node->private_list is protected by
145 * mapping->tree_lock.
147 if (!list_empty(&node
->private_list
))
148 list_lru_del(&workingset_shadow_nodes
,
149 &node
->private_list
);
154 static void page_cache_tree_delete(struct address_space
*mapping
,
155 struct page
*page
, void *shadow
)
157 struct radix_tree_node
*node
;
163 VM_BUG_ON(!PageLocked(page
));
165 __radix_tree_lookup(&mapping
->page_tree
, page
->index
, &node
, &slot
);
169 * We need a node to properly account shadow
170 * entries. Don't plant any without. XXX
176 mapping
->nrshadows
++;
178 * Make sure the nrshadows update is committed before
179 * the nrpages update so that final truncate racing
180 * with reclaim does not see both counters 0 at the
181 * same time and miss a shadow entry.
188 /* Clear direct pointer tags in root node */
189 mapping
->page_tree
.gfp_mask
&= __GFP_BITS_MASK
;
190 radix_tree_replace_slot(slot
, shadow
);
194 /* Clear tree tags for the removed page */
196 offset
= index
& RADIX_TREE_MAP_MASK
;
197 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
198 if (test_bit(offset
, node
->tags
[tag
]))
199 radix_tree_tag_clear(&mapping
->page_tree
, index
, tag
);
202 /* Delete page, swap shadow entry */
203 radix_tree_replace_slot(slot
, shadow
);
204 workingset_node_pages_dec(node
);
206 workingset_node_shadows_inc(node
);
208 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
212 * Track node that only contains shadow entries.
214 * Avoid acquiring the list_lru lock if already tracked. The
215 * list_empty() test is safe as node->private_list is
216 * protected by mapping->tree_lock.
218 if (!workingset_node_pages(node
) &&
219 list_empty(&node
->private_list
)) {
220 node
->private_data
= mapping
;
221 list_lru_add(&workingset_shadow_nodes
, &node
->private_list
);
226 * Delete a page from the page cache and free it. Caller has to make
227 * sure the page is locked and that nobody else uses it - or that usage
228 * is safe. The caller must hold the mapping's tree_lock and
229 * mem_cgroup_begin_page_stat().
231 void __delete_from_page_cache(struct page
*page
, void *shadow
,
232 struct mem_cgroup
*memcg
)
234 struct address_space
*mapping
= page
->mapping
;
236 trace_mm_filemap_delete_from_page_cache(page
);
238 * if we're uptodate, flush out into the cleancache, otherwise
239 * invalidate any existing cleancache entries. We can't leave
240 * stale data around in the cleancache once our page is gone
242 if (PageUptodate(page
) && PageMappedToDisk(page
))
243 cleancache_put_page(page
);
245 cleancache_invalidate_page(mapping
, page
);
247 page_cache_tree_delete(mapping
, page
, shadow
);
249 page
->mapping
= NULL
;
250 /* Leave page->index set: truncation lookup relies upon it */
252 /* hugetlb pages do not participate in page cache accounting. */
254 __dec_zone_page_state(page
, NR_FILE_PAGES
);
255 if (PageSwapBacked(page
))
256 __dec_zone_page_state(page
, NR_SHMEM
);
257 BUG_ON(page_mapped(page
));
260 * At this point page must be either written or cleaned by truncate.
261 * Dirty page here signals a bug and loss of unwritten data.
263 * This fixes dirty accounting after removing the page entirely but
264 * leaves PageDirty set: it has no effect for truncated page and
265 * anyway will be cleared before returning page into buddy allocator.
267 if (WARN_ON_ONCE(PageDirty(page
)))
268 account_page_cleaned(page
, mapping
, memcg
,
269 inode_to_wb(mapping
->host
));
273 * delete_from_page_cache - delete page from page cache
274 * @page: the page which the kernel is trying to remove from page cache
276 * This must be called only on pages that have been verified to be in the page
277 * cache and locked. It will never put the page into the free list, the caller
278 * has a reference on the page.
280 void delete_from_page_cache(struct page
*page
)
282 struct address_space
*mapping
= page
->mapping
;
283 struct mem_cgroup
*memcg
;
286 void (*freepage
)(struct page
*);
288 BUG_ON(!PageLocked(page
));
290 freepage
= mapping
->a_ops
->freepage
;
292 memcg
= mem_cgroup_begin_page_stat(page
);
293 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
294 __delete_from_page_cache(page
, NULL
, memcg
);
295 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
296 mem_cgroup_end_page_stat(memcg
);
300 page_cache_release(page
);
302 EXPORT_SYMBOL(delete_from_page_cache
);
304 static int filemap_check_errors(struct address_space
*mapping
)
307 /* Check for outstanding write errors */
308 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
309 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
311 if (test_bit(AS_EIO
, &mapping
->flags
) &&
312 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
318 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
319 * @mapping: address space structure to write
320 * @start: offset in bytes where the range starts
321 * @end: offset in bytes where the range ends (inclusive)
322 * @sync_mode: enable synchronous operation
324 * Start writeback against all of a mapping's dirty pages that lie
325 * within the byte offsets <start, end> inclusive.
327 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
328 * opposed to a regular memory cleansing writeback. The difference between
329 * these two operations is that if a dirty page/buffer is encountered, it must
330 * be waited upon, and not just skipped over.
332 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
333 loff_t end
, int sync_mode
)
336 struct writeback_control wbc
= {
337 .sync_mode
= sync_mode
,
338 .nr_to_write
= LONG_MAX
,
339 .range_start
= start
,
343 if (!mapping_cap_writeback_dirty(mapping
))
346 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
347 ret
= do_writepages(mapping
, &wbc
);
348 wbc_detach_inode(&wbc
);
352 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
355 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
358 int filemap_fdatawrite(struct address_space
*mapping
)
360 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
362 EXPORT_SYMBOL(filemap_fdatawrite
);
364 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
367 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
369 EXPORT_SYMBOL(filemap_fdatawrite_range
);
372 * filemap_flush - mostly a non-blocking flush
373 * @mapping: target address_space
375 * This is a mostly non-blocking flush. Not suitable for data-integrity
376 * purposes - I/O may not be started against all dirty pages.
378 int filemap_flush(struct address_space
*mapping
)
380 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
382 EXPORT_SYMBOL(filemap_flush
);
384 static int __filemap_fdatawait_range(struct address_space
*mapping
,
385 loff_t start_byte
, loff_t end_byte
)
387 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
388 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
393 if (end_byte
< start_byte
)
396 pagevec_init(&pvec
, 0);
397 while ((index
<= end
) &&
398 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
399 PAGECACHE_TAG_WRITEBACK
,
400 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
403 for (i
= 0; i
< nr_pages
; i
++) {
404 struct page
*page
= pvec
.pages
[i
];
406 /* until radix tree lookup accepts end_index */
407 if (page
->index
> end
)
410 wait_on_page_writeback(page
);
411 if (TestClearPageError(page
))
414 pagevec_release(&pvec
);
422 * filemap_fdatawait_range - wait for writeback to complete
423 * @mapping: address space structure to wait for
424 * @start_byte: offset in bytes where the range starts
425 * @end_byte: offset in bytes where the range ends (inclusive)
427 * Walk the list of under-writeback pages of the given address space
428 * in the given range and wait for all of them. Check error status of
429 * the address space and return it.
431 * Since the error status of the address space is cleared by this function,
432 * callers are responsible for checking the return value and handling and/or
433 * reporting the error.
435 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
440 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
441 ret2
= filemap_check_errors(mapping
);
447 EXPORT_SYMBOL(filemap_fdatawait_range
);
450 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
451 * @mapping: address space structure to wait for
453 * Walk the list of under-writeback pages of the given address space
454 * and wait for all of them. Unlike filemap_fdatawait(), this function
455 * does not clear error status of the address space.
457 * Use this function if callers don't handle errors themselves. Expected
458 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
461 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
463 loff_t i_size
= i_size_read(mapping
->host
);
468 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
472 * filemap_fdatawait - wait for all under-writeback pages to complete
473 * @mapping: address space structure to wait for
475 * Walk the list of under-writeback pages of the given address space
476 * and wait for all of them. Check error status of the address space
479 * Since the error status of the address space is cleared by this function,
480 * callers are responsible for checking the return value and handling and/or
481 * reporting the error.
483 int filemap_fdatawait(struct address_space
*mapping
)
485 loff_t i_size
= i_size_read(mapping
->host
);
490 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
492 EXPORT_SYMBOL(filemap_fdatawait
);
494 int filemap_write_and_wait(struct address_space
*mapping
)
498 if (mapping
->nrpages
) {
499 err
= filemap_fdatawrite(mapping
);
501 * Even if the above returned error, the pages may be
502 * written partially (e.g. -ENOSPC), so we wait for it.
503 * But the -EIO is special case, it may indicate the worst
504 * thing (e.g. bug) happened, so we avoid waiting for it.
507 int err2
= filemap_fdatawait(mapping
);
512 err
= filemap_check_errors(mapping
);
516 EXPORT_SYMBOL(filemap_write_and_wait
);
519 * filemap_write_and_wait_range - write out & wait on a file range
520 * @mapping: the address_space for the pages
521 * @lstart: offset in bytes where the range starts
522 * @lend: offset in bytes where the range ends (inclusive)
524 * Write out and wait upon file offsets lstart->lend, inclusive.
526 * Note that `lend' is inclusive (describes the last byte to be written) so
527 * that this function can be used to write to the very end-of-file (end = -1).
529 int filemap_write_and_wait_range(struct address_space
*mapping
,
530 loff_t lstart
, loff_t lend
)
534 if (mapping
->nrpages
) {
535 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
537 /* See comment of filemap_write_and_wait() */
539 int err2
= filemap_fdatawait_range(mapping
,
545 err
= filemap_check_errors(mapping
);
549 EXPORT_SYMBOL(filemap_write_and_wait_range
);
552 * replace_page_cache_page - replace a pagecache page with a new one
553 * @old: page to be replaced
554 * @new: page to replace with
555 * @gfp_mask: allocation mode
557 * This function replaces a page in the pagecache with a new one. On
558 * success it acquires the pagecache reference for the new page and
559 * drops it for the old page. Both the old and new pages must be
560 * locked. This function does not add the new page to the LRU, the
561 * caller must do that.
563 * The remove + add is atomic. The only way this function can fail is
564 * memory allocation failure.
566 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
570 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
571 VM_BUG_ON_PAGE(!PageLocked(new), new);
572 VM_BUG_ON_PAGE(new->mapping
, new);
574 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
576 struct address_space
*mapping
= old
->mapping
;
577 void (*freepage
)(struct page
*);
578 struct mem_cgroup
*memcg
;
581 pgoff_t offset
= old
->index
;
582 freepage
= mapping
->a_ops
->freepage
;
585 new->mapping
= mapping
;
588 memcg
= mem_cgroup_begin_page_stat(old
);
589 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
590 __delete_from_page_cache(old
, NULL
, memcg
);
591 error
= page_cache_tree_insert(mapping
, new, NULL
);
595 * hugetlb pages do not participate in page cache accounting.
598 __inc_zone_page_state(new, NR_FILE_PAGES
);
599 if (PageSwapBacked(new))
600 __inc_zone_page_state(new, NR_SHMEM
);
601 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
602 mem_cgroup_end_page_stat(memcg
);
603 mem_cgroup_replace_page(old
, new);
604 radix_tree_preload_end();
607 page_cache_release(old
);
612 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
614 static int __add_to_page_cache_locked(struct page
*page
,
615 struct address_space
*mapping
,
616 pgoff_t offset
, gfp_t gfp_mask
,
619 int huge
= PageHuge(page
);
620 struct mem_cgroup
*memcg
;
623 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
624 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
627 error
= mem_cgroup_try_charge(page
, current
->mm
,
633 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
636 mem_cgroup_cancel_charge(page
, memcg
);
640 page_cache_get(page
);
641 page
->mapping
= mapping
;
642 page
->index
= offset
;
644 spin_lock_irq(&mapping
->tree_lock
);
645 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
646 radix_tree_preload_end();
650 /* hugetlb pages do not participate in page cache accounting. */
652 __inc_zone_page_state(page
, NR_FILE_PAGES
);
653 spin_unlock_irq(&mapping
->tree_lock
);
655 mem_cgroup_commit_charge(page
, memcg
, false);
656 trace_mm_filemap_add_to_page_cache(page
);
659 page
->mapping
= NULL
;
660 /* Leave page->index set: truncation relies upon it */
661 spin_unlock_irq(&mapping
->tree_lock
);
663 mem_cgroup_cancel_charge(page
, memcg
);
664 page_cache_release(page
);
669 * add_to_page_cache_locked - add a locked page to the pagecache
671 * @mapping: the page's address_space
672 * @offset: page index
673 * @gfp_mask: page allocation mode
675 * This function is used to add a page to the pagecache. It must be locked.
676 * This function does not add the page to the LRU. The caller must do that.
678 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
679 pgoff_t offset
, gfp_t gfp_mask
)
681 return __add_to_page_cache_locked(page
, mapping
, offset
,
684 EXPORT_SYMBOL(add_to_page_cache_locked
);
686 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
687 pgoff_t offset
, gfp_t gfp_mask
)
692 __set_page_locked(page
);
693 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
696 __clear_page_locked(page
);
699 * The page might have been evicted from cache only
700 * recently, in which case it should be activated like
701 * any other repeatedly accessed page.
703 if (shadow
&& workingset_refault(shadow
)) {
705 workingset_activation(page
);
707 ClearPageActive(page
);
712 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
715 struct page
*__page_cache_alloc(gfp_t gfp
)
720 if (cpuset_do_page_mem_spread()) {
721 unsigned int cpuset_mems_cookie
;
723 cpuset_mems_cookie
= read_mems_allowed_begin();
724 n
= cpuset_mem_spread_node();
725 page
= __alloc_pages_node(n
, gfp
, 0);
726 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
730 return alloc_pages(gfp
, 0);
732 EXPORT_SYMBOL(__page_cache_alloc
);
736 * In order to wait for pages to become available there must be
737 * waitqueues associated with pages. By using a hash table of
738 * waitqueues where the bucket discipline is to maintain all
739 * waiters on the same queue and wake all when any of the pages
740 * become available, and for the woken contexts to check to be
741 * sure the appropriate page became available, this saves space
742 * at a cost of "thundering herd" phenomena during rare hash
745 wait_queue_head_t
*page_waitqueue(struct page
*page
)
747 const struct zone
*zone
= page_zone(page
);
749 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
751 EXPORT_SYMBOL(page_waitqueue
);
753 void wait_on_page_bit(struct page
*page
, int bit_nr
)
755 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
757 if (test_bit(bit_nr
, &page
->flags
))
758 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
759 TASK_UNINTERRUPTIBLE
);
761 EXPORT_SYMBOL(wait_on_page_bit
);
763 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
765 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
767 if (!test_bit(bit_nr
, &page
->flags
))
770 return __wait_on_bit(page_waitqueue(page
), &wait
,
771 bit_wait_io
, TASK_KILLABLE
);
774 int wait_on_page_bit_killable_timeout(struct page
*page
,
775 int bit_nr
, unsigned long timeout
)
777 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
779 wait
.key
.timeout
= jiffies
+ timeout
;
780 if (!test_bit(bit_nr
, &page
->flags
))
782 return __wait_on_bit(page_waitqueue(page
), &wait
,
783 bit_wait_io_timeout
, TASK_KILLABLE
);
785 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
788 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
789 * @page: Page defining the wait queue of interest
790 * @waiter: Waiter to add to the queue
792 * Add an arbitrary @waiter to the wait queue for the nominated @page.
794 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
796 wait_queue_head_t
*q
= page_waitqueue(page
);
799 spin_lock_irqsave(&q
->lock
, flags
);
800 __add_wait_queue(q
, waiter
);
801 spin_unlock_irqrestore(&q
->lock
, flags
);
803 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
806 * unlock_page - unlock a locked page
809 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
810 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
811 * mechanism between PageLocked pages and PageWriteback pages is shared.
812 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
814 * The mb is necessary to enforce ordering between the clear_bit and the read
815 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
817 void unlock_page(struct page
*page
)
819 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
820 clear_bit_unlock(PG_locked
, &page
->flags
);
821 smp_mb__after_atomic();
822 wake_up_page(page
, PG_locked
);
824 EXPORT_SYMBOL(unlock_page
);
827 * end_page_writeback - end writeback against a page
830 void end_page_writeback(struct page
*page
)
833 * TestClearPageReclaim could be used here but it is an atomic
834 * operation and overkill in this particular case. Failing to
835 * shuffle a page marked for immediate reclaim is too mild to
836 * justify taking an atomic operation penalty at the end of
837 * ever page writeback.
839 if (PageReclaim(page
)) {
840 ClearPageReclaim(page
);
841 rotate_reclaimable_page(page
);
844 if (!test_clear_page_writeback(page
))
847 smp_mb__after_atomic();
848 wake_up_page(page
, PG_writeback
);
850 EXPORT_SYMBOL(end_page_writeback
);
853 * After completing I/O on a page, call this routine to update the page
854 * flags appropriately
856 void page_endio(struct page
*page
, int rw
, int err
)
860 SetPageUptodate(page
);
862 ClearPageUptodate(page
);
866 } else { /* rw == WRITE */
868 struct address_space
*mapping
;
871 mapping
= page_mapping(page
);
873 mapping_set_error(mapping
, err
);
875 end_page_writeback(page
);
878 EXPORT_SYMBOL_GPL(page_endio
);
881 * __lock_page - get a lock on the page, assuming we need to sleep to get it
882 * @page: the page to lock
884 void __lock_page(struct page
*page
)
886 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
888 __wait_on_bit_lock(page_waitqueue(page
), &wait
, bit_wait_io
,
889 TASK_UNINTERRUPTIBLE
);
891 EXPORT_SYMBOL(__lock_page
);
893 int __lock_page_killable(struct page
*page
)
895 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
897 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
898 bit_wait_io
, TASK_KILLABLE
);
900 EXPORT_SYMBOL_GPL(__lock_page_killable
);
904 * 1 - page is locked; mmap_sem is still held.
905 * 0 - page is not locked.
906 * mmap_sem has been released (up_read()), unless flags had both
907 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
908 * which case mmap_sem is still held.
910 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
911 * with the page locked and the mmap_sem unperturbed.
913 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
916 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
918 * CAUTION! In this case, mmap_sem is not released
919 * even though return 0.
921 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
924 up_read(&mm
->mmap_sem
);
925 if (flags
& FAULT_FLAG_KILLABLE
)
926 wait_on_page_locked_killable(page
);
928 wait_on_page_locked(page
);
931 if (flags
& FAULT_FLAG_KILLABLE
) {
934 ret
= __lock_page_killable(page
);
936 up_read(&mm
->mmap_sem
);
946 * page_cache_next_hole - find the next hole (not-present entry)
949 * @max_scan: maximum range to search
951 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
952 * lowest indexed hole.
954 * Returns: the index of the hole if found, otherwise returns an index
955 * outside of the set specified (in which case 'return - index >=
956 * max_scan' will be true). In rare cases of index wrap-around, 0 will
959 * page_cache_next_hole may be called under rcu_read_lock. However,
960 * like radix_tree_gang_lookup, this will not atomically search a
961 * snapshot of the tree at a single point in time. For example, if a
962 * hole is created at index 5, then subsequently a hole is created at
963 * index 10, page_cache_next_hole covering both indexes may return 10
964 * if called under rcu_read_lock.
966 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
967 pgoff_t index
, unsigned long max_scan
)
971 for (i
= 0; i
< max_scan
; i
++) {
974 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
975 if (!page
|| radix_tree_exceptional_entry(page
))
984 EXPORT_SYMBOL(page_cache_next_hole
);
987 * page_cache_prev_hole - find the prev hole (not-present entry)
990 * @max_scan: maximum range to search
992 * Search backwards in the range [max(index-max_scan+1, 0), index] for
995 * Returns: the index of the hole if found, otherwise returns an index
996 * outside of the set specified (in which case 'index - return >=
997 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1000 * page_cache_prev_hole may be called under rcu_read_lock. However,
1001 * like radix_tree_gang_lookup, this will not atomically search a
1002 * snapshot of the tree at a single point in time. For example, if a
1003 * hole is created at index 10, then subsequently a hole is created at
1004 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1005 * called under rcu_read_lock.
1007 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1008 pgoff_t index
, unsigned long max_scan
)
1012 for (i
= 0; i
< max_scan
; i
++) {
1015 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1016 if (!page
|| radix_tree_exceptional_entry(page
))
1019 if (index
== ULONG_MAX
)
1025 EXPORT_SYMBOL(page_cache_prev_hole
);
1028 * find_get_entry - find and get a page cache entry
1029 * @mapping: the address_space to search
1030 * @offset: the page cache index
1032 * Looks up the page cache slot at @mapping & @offset. If there is a
1033 * page cache page, it is returned with an increased refcount.
1035 * If the slot holds a shadow entry of a previously evicted page, or a
1036 * swap entry from shmem/tmpfs, it is returned.
1038 * Otherwise, %NULL is returned.
1040 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1048 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1050 page
= radix_tree_deref_slot(pagep
);
1051 if (unlikely(!page
))
1053 if (radix_tree_exception(page
)) {
1054 if (radix_tree_deref_retry(page
))
1057 * A shadow entry of a recently evicted page,
1058 * or a swap entry from shmem/tmpfs. Return
1059 * it without attempting to raise page count.
1063 if (!page_cache_get_speculative(page
))
1067 * Has the page moved?
1068 * This is part of the lockless pagecache protocol. See
1069 * include/linux/pagemap.h for details.
1071 if (unlikely(page
!= *pagep
)) {
1072 page_cache_release(page
);
1081 EXPORT_SYMBOL(find_get_entry
);
1084 * find_lock_entry - locate, pin and lock a page cache entry
1085 * @mapping: the address_space to search
1086 * @offset: the page cache index
1088 * Looks up the page cache slot at @mapping & @offset. If there is a
1089 * page cache page, it is returned locked and with an increased
1092 * If the slot holds a shadow entry of a previously evicted page, or a
1093 * swap entry from shmem/tmpfs, it is returned.
1095 * Otherwise, %NULL is returned.
1097 * find_lock_entry() may sleep.
1099 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1104 page
= find_get_entry(mapping
, offset
);
1105 if (page
&& !radix_tree_exception(page
)) {
1107 /* Has the page been truncated? */
1108 if (unlikely(page
->mapping
!= mapping
)) {
1110 page_cache_release(page
);
1113 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1117 EXPORT_SYMBOL(find_lock_entry
);
1120 * pagecache_get_page - find and get a page reference
1121 * @mapping: the address_space to search
1122 * @offset: the page index
1123 * @fgp_flags: PCG flags
1124 * @gfp_mask: gfp mask to use for the page cache data page allocation
1126 * Looks up the page cache slot at @mapping & @offset.
1128 * PCG flags modify how the page is returned.
1130 * FGP_ACCESSED: the page will be marked accessed
1131 * FGP_LOCK: Page is return locked
1132 * FGP_CREAT: If page is not present then a new page is allocated using
1133 * @gfp_mask and added to the page cache and the VM's LRU
1134 * list. The page is returned locked and with an increased
1135 * refcount. Otherwise, %NULL is returned.
1137 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1138 * if the GFP flags specified for FGP_CREAT are atomic.
1140 * If there is a page cache page, it is returned with an increased refcount.
1142 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1143 int fgp_flags
, gfp_t gfp_mask
)
1148 page
= find_get_entry(mapping
, offset
);
1149 if (radix_tree_exceptional_entry(page
))
1154 if (fgp_flags
& FGP_LOCK
) {
1155 if (fgp_flags
& FGP_NOWAIT
) {
1156 if (!trylock_page(page
)) {
1157 page_cache_release(page
);
1164 /* Has the page been truncated? */
1165 if (unlikely(page
->mapping
!= mapping
)) {
1167 page_cache_release(page
);
1170 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1173 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1174 mark_page_accessed(page
);
1177 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1179 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1180 gfp_mask
|= __GFP_WRITE
;
1181 if (fgp_flags
& FGP_NOFS
)
1182 gfp_mask
&= ~__GFP_FS
;
1184 page
= __page_cache_alloc(gfp_mask
);
1188 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1189 fgp_flags
|= FGP_LOCK
;
1191 /* Init accessed so avoid atomic mark_page_accessed later */
1192 if (fgp_flags
& FGP_ACCESSED
)
1193 __SetPageReferenced(page
);
1195 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1196 gfp_mask
& GFP_RECLAIM_MASK
);
1197 if (unlikely(err
)) {
1198 page_cache_release(page
);
1207 EXPORT_SYMBOL(pagecache_get_page
);
1210 * find_get_entries - gang pagecache lookup
1211 * @mapping: The address_space to search
1212 * @start: The starting page cache index
1213 * @nr_entries: The maximum number of entries
1214 * @entries: Where the resulting entries are placed
1215 * @indices: The cache indices corresponding to the entries in @entries
1217 * find_get_entries() will search for and return a group of up to
1218 * @nr_entries entries in the mapping. The entries are placed at
1219 * @entries. find_get_entries() takes a reference against any actual
1222 * The search returns a group of mapping-contiguous page cache entries
1223 * with ascending indexes. There may be holes in the indices due to
1224 * not-present pages.
1226 * Any shadow entries of evicted pages, or swap entries from
1227 * shmem/tmpfs, are included in the returned array.
1229 * find_get_entries() returns the number of pages and shadow entries
1232 unsigned find_get_entries(struct address_space
*mapping
,
1233 pgoff_t start
, unsigned int nr_entries
,
1234 struct page
**entries
, pgoff_t
*indices
)
1237 unsigned int ret
= 0;
1238 struct radix_tree_iter iter
;
1245 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1248 page
= radix_tree_deref_slot(slot
);
1249 if (unlikely(!page
))
1251 if (radix_tree_exception(page
)) {
1252 if (radix_tree_deref_retry(page
))
1255 * A shadow entry of a recently evicted page,
1256 * or a swap entry from shmem/tmpfs. Return
1257 * it without attempting to raise page count.
1261 if (!page_cache_get_speculative(page
))
1264 /* Has the page moved? */
1265 if (unlikely(page
!= *slot
)) {
1266 page_cache_release(page
);
1270 indices
[ret
] = iter
.index
;
1271 entries
[ret
] = page
;
1272 if (++ret
== nr_entries
)
1280 * find_get_pages - gang pagecache lookup
1281 * @mapping: The address_space to search
1282 * @start: The starting page index
1283 * @nr_pages: The maximum number of pages
1284 * @pages: Where the resulting pages are placed
1286 * find_get_pages() will search for and return a group of up to
1287 * @nr_pages pages in the mapping. The pages are placed at @pages.
1288 * find_get_pages() takes a reference against the returned pages.
1290 * The search returns a group of mapping-contiguous pages with ascending
1291 * indexes. There may be holes in the indices due to not-present pages.
1293 * find_get_pages() returns the number of pages which were found.
1295 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1296 unsigned int nr_pages
, struct page
**pages
)
1298 struct radix_tree_iter iter
;
1302 if (unlikely(!nr_pages
))
1307 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1310 page
= radix_tree_deref_slot(slot
);
1311 if (unlikely(!page
))
1314 if (radix_tree_exception(page
)) {
1315 if (radix_tree_deref_retry(page
)) {
1317 * Transient condition which can only trigger
1318 * when entry at index 0 moves out of or back
1319 * to root: none yet gotten, safe to restart.
1321 WARN_ON(iter
.index
);
1325 * A shadow entry of a recently evicted page,
1326 * or a swap entry from shmem/tmpfs. Skip
1332 if (!page_cache_get_speculative(page
))
1335 /* Has the page moved? */
1336 if (unlikely(page
!= *slot
)) {
1337 page_cache_release(page
);
1342 if (++ret
== nr_pages
)
1351 * find_get_pages_contig - gang contiguous pagecache lookup
1352 * @mapping: The address_space to search
1353 * @index: The starting page index
1354 * @nr_pages: The maximum number of pages
1355 * @pages: Where the resulting pages are placed
1357 * find_get_pages_contig() works exactly like find_get_pages(), except
1358 * that the returned number of pages are guaranteed to be contiguous.
1360 * find_get_pages_contig() returns the number of pages which were found.
1362 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1363 unsigned int nr_pages
, struct page
**pages
)
1365 struct radix_tree_iter iter
;
1367 unsigned int ret
= 0;
1369 if (unlikely(!nr_pages
))
1374 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1377 page
= radix_tree_deref_slot(slot
);
1378 /* The hole, there no reason to continue */
1379 if (unlikely(!page
))
1382 if (radix_tree_exception(page
)) {
1383 if (radix_tree_deref_retry(page
)) {
1385 * Transient condition which can only trigger
1386 * when entry at index 0 moves out of or back
1387 * to root: none yet gotten, safe to restart.
1392 * A shadow entry of a recently evicted page,
1393 * or a swap entry from shmem/tmpfs. Stop
1394 * looking for contiguous pages.
1399 if (!page_cache_get_speculative(page
))
1402 /* Has the page moved? */
1403 if (unlikely(page
!= *slot
)) {
1404 page_cache_release(page
);
1409 * must check mapping and index after taking the ref.
1410 * otherwise we can get both false positives and false
1411 * negatives, which is just confusing to the caller.
1413 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1414 page_cache_release(page
);
1419 if (++ret
== nr_pages
)
1425 EXPORT_SYMBOL(find_get_pages_contig
);
1428 * find_get_pages_tag - find and return pages that match @tag
1429 * @mapping: the address_space to search
1430 * @index: the starting page index
1431 * @tag: the tag index
1432 * @nr_pages: the maximum number of pages
1433 * @pages: where the resulting pages are placed
1435 * Like find_get_pages, except we only return pages which are tagged with
1436 * @tag. We update @index to index the next page for the traversal.
1438 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1439 int tag
, unsigned int nr_pages
, struct page
**pages
)
1441 struct radix_tree_iter iter
;
1445 if (unlikely(!nr_pages
))
1450 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1451 &iter
, *index
, tag
) {
1454 page
= radix_tree_deref_slot(slot
);
1455 if (unlikely(!page
))
1458 if (radix_tree_exception(page
)) {
1459 if (radix_tree_deref_retry(page
)) {
1461 * Transient condition which can only trigger
1462 * when entry at index 0 moves out of or back
1463 * to root: none yet gotten, safe to restart.
1468 * A shadow entry of a recently evicted page.
1470 * Those entries should never be tagged, but
1471 * this tree walk is lockless and the tags are
1472 * looked up in bulk, one radix tree node at a
1473 * time, so there is a sizable window for page
1474 * reclaim to evict a page we saw tagged.
1481 if (!page_cache_get_speculative(page
))
1484 /* Has the page moved? */
1485 if (unlikely(page
!= *slot
)) {
1486 page_cache_release(page
);
1491 if (++ret
== nr_pages
)
1498 *index
= pages
[ret
- 1]->index
+ 1;
1502 EXPORT_SYMBOL(find_get_pages_tag
);
1505 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1506 * a _large_ part of the i/o request. Imagine the worst scenario:
1508 * ---R__________________________________________B__________
1509 * ^ reading here ^ bad block(assume 4k)
1511 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1512 * => failing the whole request => read(R) => read(R+1) =>
1513 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1514 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1515 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1517 * It is going insane. Fix it by quickly scaling down the readahead size.
1519 static void shrink_readahead_size_eio(struct file
*filp
,
1520 struct file_ra_state
*ra
)
1526 * do_generic_file_read - generic file read routine
1527 * @filp: the file to read
1528 * @ppos: current file position
1529 * @iter: data destination
1530 * @written: already copied
1532 * This is a generic file read routine, and uses the
1533 * mapping->a_ops->readpage() function for the actual low-level stuff.
1535 * This is really ugly. But the goto's actually try to clarify some
1536 * of the logic when it comes to error handling etc.
1538 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1539 struct iov_iter
*iter
, ssize_t written
)
1541 struct address_space
*mapping
= filp
->f_mapping
;
1542 struct inode
*inode
= mapping
->host
;
1543 struct file_ra_state
*ra
= &filp
->f_ra
;
1547 unsigned long offset
; /* offset into pagecache page */
1548 unsigned int prev_offset
;
1551 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1552 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1553 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1554 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1555 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1561 unsigned long nr
, ret
;
1565 if (fatal_signal_pending(current
)) {
1570 page
= find_get_page(mapping
, index
);
1572 page_cache_sync_readahead(mapping
,
1574 index
, last_index
- index
);
1575 page
= find_get_page(mapping
, index
);
1576 if (unlikely(page
== NULL
))
1577 goto no_cached_page
;
1579 if (PageReadahead(page
)) {
1580 page_cache_async_readahead(mapping
,
1582 index
, last_index
- index
);
1584 if (!PageUptodate(page
)) {
1585 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1586 !mapping
->a_ops
->is_partially_uptodate
)
1587 goto page_not_up_to_date
;
1588 if (!trylock_page(page
))
1589 goto page_not_up_to_date
;
1590 /* Did it get truncated before we got the lock? */
1592 goto page_not_up_to_date_locked
;
1593 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1594 offset
, iter
->count
))
1595 goto page_not_up_to_date_locked
;
1600 * i_size must be checked after we know the page is Uptodate.
1602 * Checking i_size after the check allows us to calculate
1603 * the correct value for "nr", which means the zero-filled
1604 * part of the page is not copied back to userspace (unless
1605 * another truncate extends the file - this is desired though).
1608 isize
= i_size_read(inode
);
1609 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1610 if (unlikely(!isize
|| index
> end_index
)) {
1611 page_cache_release(page
);
1615 /* nr is the maximum number of bytes to copy from this page */
1616 nr
= PAGE_CACHE_SIZE
;
1617 if (index
== end_index
) {
1618 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1620 page_cache_release(page
);
1626 /* If users can be writing to this page using arbitrary
1627 * virtual addresses, take care about potential aliasing
1628 * before reading the page on the kernel side.
1630 if (mapping_writably_mapped(mapping
))
1631 flush_dcache_page(page
);
1634 * When a sequential read accesses a page several times,
1635 * only mark it as accessed the first time.
1637 if (prev_index
!= index
|| offset
!= prev_offset
)
1638 mark_page_accessed(page
);
1642 * Ok, we have the page, and it's up-to-date, so
1643 * now we can copy it to user space...
1646 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1648 index
+= offset
>> PAGE_CACHE_SHIFT
;
1649 offset
&= ~PAGE_CACHE_MASK
;
1650 prev_offset
= offset
;
1652 page_cache_release(page
);
1654 if (!iov_iter_count(iter
))
1662 page_not_up_to_date
:
1663 /* Get exclusive access to the page ... */
1664 error
= lock_page_killable(page
);
1665 if (unlikely(error
))
1666 goto readpage_error
;
1668 page_not_up_to_date_locked
:
1669 /* Did it get truncated before we got the lock? */
1670 if (!page
->mapping
) {
1672 page_cache_release(page
);
1676 /* Did somebody else fill it already? */
1677 if (PageUptodate(page
)) {
1684 * A previous I/O error may have been due to temporary
1685 * failures, eg. multipath errors.
1686 * PG_error will be set again if readpage fails.
1688 ClearPageError(page
);
1689 /* Start the actual read. The read will unlock the page. */
1690 error
= mapping
->a_ops
->readpage(filp
, page
);
1692 if (unlikely(error
)) {
1693 if (error
== AOP_TRUNCATED_PAGE
) {
1694 page_cache_release(page
);
1698 goto readpage_error
;
1701 if (!PageUptodate(page
)) {
1702 error
= lock_page_killable(page
);
1703 if (unlikely(error
))
1704 goto readpage_error
;
1705 if (!PageUptodate(page
)) {
1706 if (page
->mapping
== NULL
) {
1708 * invalidate_mapping_pages got it
1711 page_cache_release(page
);
1715 shrink_readahead_size_eio(filp
, ra
);
1717 goto readpage_error
;
1725 /* UHHUH! A synchronous read error occurred. Report it */
1726 page_cache_release(page
);
1731 * Ok, it wasn't cached, so we need to create a new
1734 page
= page_cache_alloc_cold(mapping
);
1739 error
= add_to_page_cache_lru(page
, mapping
, index
,
1740 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1742 page_cache_release(page
);
1743 if (error
== -EEXIST
) {
1753 ra
->prev_pos
= prev_index
;
1754 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1755 ra
->prev_pos
|= prev_offset
;
1757 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1758 file_accessed(filp
);
1759 return written
? written
: error
;
1763 * generic_file_read_iter - generic filesystem read routine
1764 * @iocb: kernel I/O control block
1765 * @iter: destination for the data read
1767 * This is the "read_iter()" routine for all filesystems
1768 * that can use the page cache directly.
1771 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1773 struct file
*file
= iocb
->ki_filp
;
1775 loff_t
*ppos
= &iocb
->ki_pos
;
1778 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1779 struct address_space
*mapping
= file
->f_mapping
;
1780 struct inode
*inode
= mapping
->host
;
1781 size_t count
= iov_iter_count(iter
);
1785 goto out
; /* skip atime */
1786 size
= i_size_read(inode
);
1787 retval
= filemap_write_and_wait_range(mapping
, pos
,
1790 struct iov_iter data
= *iter
;
1791 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
1795 *ppos
= pos
+ retval
;
1796 iov_iter_advance(iter
, retval
);
1800 * Btrfs can have a short DIO read if we encounter
1801 * compressed extents, so if there was an error, or if
1802 * we've already read everything we wanted to, or if
1803 * there was a short read because we hit EOF, go ahead
1804 * and return. Otherwise fallthrough to buffered io for
1805 * the rest of the read. Buffered reads will not work for
1806 * DAX files, so don't bother trying.
1808 if (retval
< 0 || !iov_iter_count(iter
) || *ppos
>= size
||
1810 file_accessed(file
);
1815 retval
= do_generic_file_read(file
, ppos
, iter
, retval
);
1819 EXPORT_SYMBOL(generic_file_read_iter
);
1823 * page_cache_read - adds requested page to the page cache if not already there
1824 * @file: file to read
1825 * @offset: page index
1827 * This adds the requested page to the page cache if it isn't already there,
1828 * and schedules an I/O to read in its contents from disk.
1830 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1832 struct address_space
*mapping
= file
->f_mapping
;
1837 page
= page_cache_alloc_cold(mapping
);
1841 ret
= add_to_page_cache_lru(page
, mapping
, offset
,
1842 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1844 ret
= mapping
->a_ops
->readpage(file
, page
);
1845 else if (ret
== -EEXIST
)
1846 ret
= 0; /* losing race to add is OK */
1848 page_cache_release(page
);
1850 } while (ret
== AOP_TRUNCATED_PAGE
);
1855 #define MMAP_LOTSAMISS (100)
1858 * Synchronous readahead happens when we don't even find
1859 * a page in the page cache at all.
1861 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1862 struct file_ra_state
*ra
,
1866 struct address_space
*mapping
= file
->f_mapping
;
1868 /* If we don't want any read-ahead, don't bother */
1869 if (vma
->vm_flags
& VM_RAND_READ
)
1874 if (vma
->vm_flags
& VM_SEQ_READ
) {
1875 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1880 /* Avoid banging the cache line if not needed */
1881 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1885 * Do we miss much more than hit in this file? If so,
1886 * stop bothering with read-ahead. It will only hurt.
1888 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1894 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
1895 ra
->size
= ra
->ra_pages
;
1896 ra
->async_size
= ra
->ra_pages
/ 4;
1897 ra_submit(ra
, mapping
, file
);
1901 * Asynchronous readahead happens when we find the page and PG_readahead,
1902 * so we want to possibly extend the readahead further..
1904 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1905 struct file_ra_state
*ra
,
1910 struct address_space
*mapping
= file
->f_mapping
;
1912 /* If we don't want any read-ahead, don't bother */
1913 if (vma
->vm_flags
& VM_RAND_READ
)
1915 if (ra
->mmap_miss
> 0)
1917 if (PageReadahead(page
))
1918 page_cache_async_readahead(mapping
, ra
, file
,
1919 page
, offset
, ra
->ra_pages
);
1923 * filemap_fault - read in file data for page fault handling
1924 * @vma: vma in which the fault was taken
1925 * @vmf: struct vm_fault containing details of the fault
1927 * filemap_fault() is invoked via the vma operations vector for a
1928 * mapped memory region to read in file data during a page fault.
1930 * The goto's are kind of ugly, but this streamlines the normal case of having
1931 * it in the page cache, and handles the special cases reasonably without
1932 * having a lot of duplicated code.
1934 * vma->vm_mm->mmap_sem must be held on entry.
1936 * If our return value has VM_FAULT_RETRY set, it's because
1937 * lock_page_or_retry() returned 0.
1938 * The mmap_sem has usually been released in this case.
1939 * See __lock_page_or_retry() for the exception.
1941 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1942 * has not been released.
1944 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1946 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1949 struct file
*file
= vma
->vm_file
;
1950 struct address_space
*mapping
= file
->f_mapping
;
1951 struct file_ra_state
*ra
= &file
->f_ra
;
1952 struct inode
*inode
= mapping
->host
;
1953 pgoff_t offset
= vmf
->pgoff
;
1958 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
1959 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
1960 return VM_FAULT_SIGBUS
;
1963 * Do we have something in the page cache already?
1965 page
= find_get_page(mapping
, offset
);
1966 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1968 * We found the page, so try async readahead before
1969 * waiting for the lock.
1971 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1973 /* No page in the page cache at all */
1974 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1975 count_vm_event(PGMAJFAULT
);
1976 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1977 ret
= VM_FAULT_MAJOR
;
1979 page
= find_get_page(mapping
, offset
);
1981 goto no_cached_page
;
1984 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1985 page_cache_release(page
);
1986 return ret
| VM_FAULT_RETRY
;
1989 /* Did it get truncated? */
1990 if (unlikely(page
->mapping
!= mapping
)) {
1995 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1998 * We have a locked page in the page cache, now we need to check
1999 * that it's up-to-date. If not, it is going to be due to an error.
2001 if (unlikely(!PageUptodate(page
)))
2002 goto page_not_uptodate
;
2005 * Found the page and have a reference on it.
2006 * We must recheck i_size under page lock.
2008 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2009 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
2011 page_cache_release(page
);
2012 return VM_FAULT_SIGBUS
;
2016 return ret
| VM_FAULT_LOCKED
;
2020 * We're only likely to ever get here if MADV_RANDOM is in
2023 error
= page_cache_read(file
, offset
);
2026 * The page we want has now been added to the page cache.
2027 * In the unlikely event that someone removed it in the
2028 * meantime, we'll just come back here and read it again.
2034 * An error return from page_cache_read can result if the
2035 * system is low on memory, or a problem occurs while trying
2038 if (error
== -ENOMEM
)
2039 return VM_FAULT_OOM
;
2040 return VM_FAULT_SIGBUS
;
2044 * Umm, take care of errors if the page isn't up-to-date.
2045 * Try to re-read it _once_. We do this synchronously,
2046 * because there really aren't any performance issues here
2047 * and we need to check for errors.
2049 ClearPageError(page
);
2050 error
= mapping
->a_ops
->readpage(file
, page
);
2052 wait_on_page_locked(page
);
2053 if (!PageUptodate(page
))
2056 page_cache_release(page
);
2058 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2061 /* Things didn't work out. Return zero to tell the mm layer so. */
2062 shrink_readahead_size_eio(file
, ra
);
2063 return VM_FAULT_SIGBUS
;
2065 EXPORT_SYMBOL(filemap_fault
);
2067 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2069 struct radix_tree_iter iter
;
2071 struct file
*file
= vma
->vm_file
;
2072 struct address_space
*mapping
= file
->f_mapping
;
2075 unsigned long address
= (unsigned long) vmf
->virtual_address
;
2080 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
2081 if (iter
.index
> vmf
->max_pgoff
)
2084 page
= radix_tree_deref_slot(slot
);
2085 if (unlikely(!page
))
2087 if (radix_tree_exception(page
)) {
2088 if (radix_tree_deref_retry(page
))
2094 if (!page_cache_get_speculative(page
))
2097 /* Has the page moved? */
2098 if (unlikely(page
!= *slot
)) {
2099 page_cache_release(page
);
2103 if (!PageUptodate(page
) ||
2104 PageReadahead(page
) ||
2107 if (!trylock_page(page
))
2110 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2113 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
2114 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2117 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2118 if (!pte_none(*pte
))
2121 if (file
->f_ra
.mmap_miss
> 0)
2122 file
->f_ra
.mmap_miss
--;
2123 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2124 do_set_pte(vma
, addr
, page
, pte
, false, false);
2130 page_cache_release(page
);
2132 if (iter
.index
== vmf
->max_pgoff
)
2137 EXPORT_SYMBOL(filemap_map_pages
);
2139 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2141 struct page
*page
= vmf
->page
;
2142 struct inode
*inode
= file_inode(vma
->vm_file
);
2143 int ret
= VM_FAULT_LOCKED
;
2145 sb_start_pagefault(inode
->i_sb
);
2146 vma_file_update_time(vma
);
2148 if (page
->mapping
!= inode
->i_mapping
) {
2150 ret
= VM_FAULT_NOPAGE
;
2154 * We mark the page dirty already here so that when freeze is in
2155 * progress, we are guaranteed that writeback during freezing will
2156 * see the dirty page and writeprotect it again.
2158 set_page_dirty(page
);
2159 wait_for_stable_page(page
);
2161 sb_end_pagefault(inode
->i_sb
);
2164 EXPORT_SYMBOL(filemap_page_mkwrite
);
2166 const struct vm_operations_struct generic_file_vm_ops
= {
2167 .fault
= filemap_fault
,
2168 .map_pages
= filemap_map_pages
,
2169 .page_mkwrite
= filemap_page_mkwrite
,
2172 /* This is used for a general mmap of a disk file */
2174 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2176 struct address_space
*mapping
= file
->f_mapping
;
2178 if (!mapping
->a_ops
->readpage
)
2180 file_accessed(file
);
2181 vma
->vm_ops
= &generic_file_vm_ops
;
2186 * This is for filesystems which do not implement ->writepage.
2188 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2190 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2192 return generic_file_mmap(file
, vma
);
2195 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2199 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2203 #endif /* CONFIG_MMU */
2205 EXPORT_SYMBOL(generic_file_mmap
);
2206 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2208 static struct page
*wait_on_page_read(struct page
*page
)
2210 if (!IS_ERR(page
)) {
2211 wait_on_page_locked(page
);
2212 if (!PageUptodate(page
)) {
2213 page_cache_release(page
);
2214 page
= ERR_PTR(-EIO
);
2220 static struct page
*__read_cache_page(struct address_space
*mapping
,
2222 int (*filler
)(void *, struct page
*),
2229 page
= find_get_page(mapping
, index
);
2231 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2233 return ERR_PTR(-ENOMEM
);
2234 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2235 if (unlikely(err
)) {
2236 page_cache_release(page
);
2239 /* Presumably ENOMEM for radix tree node */
2240 return ERR_PTR(err
);
2242 err
= filler(data
, page
);
2244 page_cache_release(page
);
2245 page
= ERR_PTR(err
);
2247 page
= wait_on_page_read(page
);
2253 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2255 int (*filler
)(void *, struct page
*),
2264 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2267 if (PageUptodate(page
))
2271 if (!page
->mapping
) {
2273 page_cache_release(page
);
2276 if (PageUptodate(page
)) {
2280 err
= filler(data
, page
);
2282 page_cache_release(page
);
2283 return ERR_PTR(err
);
2285 page
= wait_on_page_read(page
);
2290 mark_page_accessed(page
);
2295 * read_cache_page - read into page cache, fill it if needed
2296 * @mapping: the page's address_space
2297 * @index: the page index
2298 * @filler: function to perform the read
2299 * @data: first arg to filler(data, page) function, often left as NULL
2301 * Read into the page cache. If a page already exists, and PageUptodate() is
2302 * not set, try to fill the page and wait for it to become unlocked.
2304 * If the page does not get brought uptodate, return -EIO.
2306 struct page
*read_cache_page(struct address_space
*mapping
,
2308 int (*filler
)(void *, struct page
*),
2311 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2313 EXPORT_SYMBOL(read_cache_page
);
2316 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2317 * @mapping: the page's address_space
2318 * @index: the page index
2319 * @gfp: the page allocator flags to use if allocating
2321 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2322 * any new page allocations done using the specified allocation flags.
2324 * If the page does not get brought uptodate, return -EIO.
2326 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2330 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2332 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2334 EXPORT_SYMBOL(read_cache_page_gfp
);
2337 * Performs necessary checks before doing a write
2339 * Can adjust writing position or amount of bytes to write.
2340 * Returns appropriate error code that caller should return or
2341 * zero in case that write should be allowed.
2343 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2345 struct file
*file
= iocb
->ki_filp
;
2346 struct inode
*inode
= file
->f_mapping
->host
;
2347 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2350 if (!iov_iter_count(from
))
2353 /* FIXME: this is for backwards compatibility with 2.4 */
2354 if (iocb
->ki_flags
& IOCB_APPEND
)
2355 iocb
->ki_pos
= i_size_read(inode
);
2359 if (limit
!= RLIM_INFINITY
) {
2360 if (iocb
->ki_pos
>= limit
) {
2361 send_sig(SIGXFSZ
, current
, 0);
2364 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2370 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2371 !(file
->f_flags
& O_LARGEFILE
))) {
2372 if (pos
>= MAX_NON_LFS
)
2374 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2378 * Are we about to exceed the fs block limit ?
2380 * If we have written data it becomes a short write. If we have
2381 * exceeded without writing data we send a signal and return EFBIG.
2382 * Linus frestrict idea will clean these up nicely..
2384 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2387 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2388 return iov_iter_count(from
);
2390 EXPORT_SYMBOL(generic_write_checks
);
2392 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2393 loff_t pos
, unsigned len
, unsigned flags
,
2394 struct page
**pagep
, void **fsdata
)
2396 const struct address_space_operations
*aops
= mapping
->a_ops
;
2398 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2401 EXPORT_SYMBOL(pagecache_write_begin
);
2403 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2404 loff_t pos
, unsigned len
, unsigned copied
,
2405 struct page
*page
, void *fsdata
)
2407 const struct address_space_operations
*aops
= mapping
->a_ops
;
2409 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2411 EXPORT_SYMBOL(pagecache_write_end
);
2414 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
2416 struct file
*file
= iocb
->ki_filp
;
2417 struct address_space
*mapping
= file
->f_mapping
;
2418 struct inode
*inode
= mapping
->host
;
2422 struct iov_iter data
;
2424 write_len
= iov_iter_count(from
);
2425 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2427 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2432 * After a write we want buffered reads to be sure to go to disk to get
2433 * the new data. We invalidate clean cached page from the region we're
2434 * about to write. We do this *before* the write so that we can return
2435 * without clobbering -EIOCBQUEUED from ->direct_IO().
2437 if (mapping
->nrpages
) {
2438 written
= invalidate_inode_pages2_range(mapping
,
2439 pos
>> PAGE_CACHE_SHIFT
, end
);
2441 * If a page can not be invalidated, return 0 to fall back
2442 * to buffered write.
2445 if (written
== -EBUSY
)
2452 written
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
2455 * Finally, try again to invalidate clean pages which might have been
2456 * cached by non-direct readahead, or faulted in by get_user_pages()
2457 * if the source of the write was an mmap'ed region of the file
2458 * we're writing. Either one is a pretty crazy thing to do,
2459 * so we don't support it 100%. If this invalidation
2460 * fails, tough, the write still worked...
2462 if (mapping
->nrpages
) {
2463 invalidate_inode_pages2_range(mapping
,
2464 pos
>> PAGE_CACHE_SHIFT
, end
);
2469 iov_iter_advance(from
, written
);
2470 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2471 i_size_write(inode
, pos
);
2472 mark_inode_dirty(inode
);
2479 EXPORT_SYMBOL(generic_file_direct_write
);
2482 * Find or create a page at the given pagecache position. Return the locked
2483 * page. This function is specifically for buffered writes.
2485 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2486 pgoff_t index
, unsigned flags
)
2489 int fgp_flags
= FGP_LOCK
|FGP_ACCESSED
|FGP_WRITE
|FGP_CREAT
;
2491 if (flags
& AOP_FLAG_NOFS
)
2492 fgp_flags
|= FGP_NOFS
;
2494 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2495 mapping_gfp_mask(mapping
));
2497 wait_for_stable_page(page
);
2501 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2503 ssize_t
generic_perform_write(struct file
*file
,
2504 struct iov_iter
*i
, loff_t pos
)
2506 struct address_space
*mapping
= file
->f_mapping
;
2507 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2509 ssize_t written
= 0;
2510 unsigned int flags
= 0;
2513 * Copies from kernel address space cannot fail (NFSD is a big user).
2515 if (!iter_is_iovec(i
))
2516 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2520 unsigned long offset
; /* Offset into pagecache page */
2521 unsigned long bytes
; /* Bytes to write to page */
2522 size_t copied
; /* Bytes copied from user */
2525 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2526 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2531 * Bring in the user page that we will copy from _first_.
2532 * Otherwise there's a nasty deadlock on copying from the
2533 * same page as we're writing to, without it being marked
2536 * Not only is this an optimisation, but it is also required
2537 * to check that the address is actually valid, when atomic
2538 * usercopies are used, below.
2540 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2545 if (fatal_signal_pending(current
)) {
2550 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2552 if (unlikely(status
< 0))
2555 if (mapping_writably_mapped(mapping
))
2556 flush_dcache_page(page
);
2558 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2559 flush_dcache_page(page
);
2561 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2563 if (unlikely(status
< 0))
2569 iov_iter_advance(i
, copied
);
2570 if (unlikely(copied
== 0)) {
2572 * If we were unable to copy any data at all, we must
2573 * fall back to a single segment length write.
2575 * If we didn't fallback here, we could livelock
2576 * because not all segments in the iov can be copied at
2577 * once without a pagefault.
2579 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2580 iov_iter_single_seg_count(i
));
2586 balance_dirty_pages_ratelimited(mapping
);
2587 } while (iov_iter_count(i
));
2589 return written
? written
: status
;
2591 EXPORT_SYMBOL(generic_perform_write
);
2594 * __generic_file_write_iter - write data to a file
2595 * @iocb: IO state structure (file, offset, etc.)
2596 * @from: iov_iter with data to write
2598 * This function does all the work needed for actually writing data to a
2599 * file. It does all basic checks, removes SUID from the file, updates
2600 * modification times and calls proper subroutines depending on whether we
2601 * do direct IO or a standard buffered write.
2603 * It expects i_mutex to be grabbed unless we work on a block device or similar
2604 * object which does not need locking at all.
2606 * This function does *not* take care of syncing data in case of O_SYNC write.
2607 * A caller has to handle it. This is mainly due to the fact that we want to
2608 * avoid syncing under i_mutex.
2610 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2612 struct file
*file
= iocb
->ki_filp
;
2613 struct address_space
* mapping
= file
->f_mapping
;
2614 struct inode
*inode
= mapping
->host
;
2615 ssize_t written
= 0;
2619 /* We can write back this queue in page reclaim */
2620 current
->backing_dev_info
= inode_to_bdi(inode
);
2621 err
= file_remove_privs(file
);
2625 err
= file_update_time(file
);
2629 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2630 loff_t pos
, endbyte
;
2632 written
= generic_file_direct_write(iocb
, from
, iocb
->ki_pos
);
2634 * If the write stopped short of completing, fall back to
2635 * buffered writes. Some filesystems do this for writes to
2636 * holes, for example. For DAX files, a buffered write will
2637 * not succeed (even if it did, DAX does not handle dirty
2638 * page-cache pages correctly).
2640 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2643 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2645 * If generic_perform_write() returned a synchronous error
2646 * then we want to return the number of bytes which were
2647 * direct-written, or the error code if that was zero. Note
2648 * that this differs from normal direct-io semantics, which
2649 * will return -EFOO even if some bytes were written.
2651 if (unlikely(status
< 0)) {
2656 * We need to ensure that the page cache pages are written to
2657 * disk and invalidated to preserve the expected O_DIRECT
2660 endbyte
= pos
+ status
- 1;
2661 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2663 iocb
->ki_pos
= endbyte
+ 1;
2665 invalidate_mapping_pages(mapping
,
2666 pos
>> PAGE_CACHE_SHIFT
,
2667 endbyte
>> PAGE_CACHE_SHIFT
);
2670 * We don't know how much we wrote, so just return
2671 * the number of bytes which were direct-written
2675 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2676 if (likely(written
> 0))
2677 iocb
->ki_pos
+= written
;
2680 current
->backing_dev_info
= NULL
;
2681 return written
? written
: err
;
2683 EXPORT_SYMBOL(__generic_file_write_iter
);
2686 * generic_file_write_iter - write data to a file
2687 * @iocb: IO state structure
2688 * @from: iov_iter with data to write
2690 * This is a wrapper around __generic_file_write_iter() to be used by most
2691 * filesystems. It takes care of syncing the file in case of O_SYNC file
2692 * and acquires i_mutex as needed.
2694 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2696 struct file
*file
= iocb
->ki_filp
;
2697 struct inode
*inode
= file
->f_mapping
->host
;
2700 mutex_lock(&inode
->i_mutex
);
2701 ret
= generic_write_checks(iocb
, from
);
2703 ret
= __generic_file_write_iter(iocb
, from
);
2704 mutex_unlock(&inode
->i_mutex
);
2709 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2715 EXPORT_SYMBOL(generic_file_write_iter
);
2718 * try_to_release_page() - release old fs-specific metadata on a page
2720 * @page: the page which the kernel is trying to free
2721 * @gfp_mask: memory allocation flags (and I/O mode)
2723 * The address_space is to try to release any data against the page
2724 * (presumably at page->private). If the release was successful, return `1'.
2725 * Otherwise return zero.
2727 * This may also be called if PG_fscache is set on a page, indicating that the
2728 * page is known to the local caching routines.
2730 * The @gfp_mask argument specifies whether I/O may be performed to release
2731 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2734 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2736 struct address_space
* const mapping
= page
->mapping
;
2738 BUG_ON(!PageLocked(page
));
2739 if (PageWriteback(page
))
2742 if (mapping
&& mapping
->a_ops
->releasepage
)
2743 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2744 return try_to_free_buffers(page
);
2747 EXPORT_SYMBOL(try_to_release_page
);