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>
14 #include <linux/dax.h>
16 #include <linux/uaccess.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/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/rmap.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/filemap.h>
44 * FIXME: remove all knowledge of the buffer layer from the core VM
46 #include <linux/buffer_head.h> /* for try_to_free_buffers */
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
54 * Shared mappings now work. 15.8.1995 Bruno.
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
65 * ->i_mmap_rwsem (truncate_pagecache)
66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
71 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
79 * ->lock_page (access_process_vm)
81 * ->i_mutex (generic_perform_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
85 * sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
104 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
110 * ->tasklist_lock (memory_failure, collect_procs_ao)
113 static void page_cache_tree_delete(struct address_space
*mapping
,
114 struct page
*page
, void *shadow
)
116 struct radix_tree_node
*node
;
122 VM_BUG_ON(!PageLocked(page
));
124 __radix_tree_lookup(&mapping
->page_tree
, page
->index
, &node
, &slot
);
127 mapping
->nrexceptional
++;
129 * Make sure the nrexceptional update is committed before
130 * the nrpages update so that final truncate racing
131 * with reclaim does not see both counters 0 at the
132 * same time and miss a shadow entry.
139 /* Clear direct pointer tags in root node */
140 mapping
->page_tree
.gfp_mask
&= __GFP_BITS_MASK
;
141 radix_tree_replace_slot(slot
, shadow
);
145 /* Clear tree tags for the removed page */
147 offset
= index
& RADIX_TREE_MAP_MASK
;
148 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
149 if (test_bit(offset
, node
->tags
[tag
]))
150 radix_tree_tag_clear(&mapping
->page_tree
, index
, tag
);
153 /* Delete page, swap shadow entry */
154 radix_tree_replace_slot(slot
, shadow
);
155 workingset_node_pages_dec(node
);
157 workingset_node_shadows_inc(node
);
159 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
163 * Track node that only contains shadow entries.
165 * Avoid acquiring the list_lru lock if already tracked. The
166 * list_empty() test is safe as node->private_list is
167 * protected by mapping->tree_lock.
169 if (!workingset_node_pages(node
) &&
170 list_empty(&node
->private_list
)) {
171 node
->private_data
= mapping
;
172 list_lru_add(&workingset_shadow_nodes
, &node
->private_list
);
177 * Delete a page from the page cache and free it. Caller has to make
178 * sure the page is locked and that nobody else uses it - or that usage
179 * is safe. The caller must hold the mapping's tree_lock and
180 * mem_cgroup_begin_page_stat().
182 void __delete_from_page_cache(struct page
*page
, void *shadow
,
183 struct mem_cgroup
*memcg
)
185 struct address_space
*mapping
= page
->mapping
;
187 trace_mm_filemap_delete_from_page_cache(page
);
189 * if we're uptodate, flush out into the cleancache, otherwise
190 * invalidate any existing cleancache entries. We can't leave
191 * stale data around in the cleancache once our page is gone
193 if (PageUptodate(page
) && PageMappedToDisk(page
))
194 cleancache_put_page(page
);
196 cleancache_invalidate_page(mapping
, page
);
198 page_cache_tree_delete(mapping
, page
, shadow
);
200 page
->mapping
= NULL
;
201 /* Leave page->index set: truncation lookup relies upon it */
203 /* hugetlb pages do not participate in page cache accounting. */
205 __dec_zone_page_state(page
, NR_FILE_PAGES
);
206 if (PageSwapBacked(page
))
207 __dec_zone_page_state(page
, NR_SHMEM
);
208 VM_BUG_ON_PAGE(page_mapped(page
), page
);
211 * At this point page must be either written or cleaned by truncate.
212 * Dirty page here signals a bug and loss of unwritten data.
214 * This fixes dirty accounting after removing the page entirely but
215 * leaves PageDirty set: it has no effect for truncated page and
216 * anyway will be cleared before returning page into buddy allocator.
218 if (WARN_ON_ONCE(PageDirty(page
)))
219 account_page_cleaned(page
, mapping
, memcg
,
220 inode_to_wb(mapping
->host
));
224 * delete_from_page_cache - delete page from page cache
225 * @page: the page which the kernel is trying to remove from page cache
227 * This must be called only on pages that have been verified to be in the page
228 * cache and locked. It will never put the page into the free list, the caller
229 * has a reference on the page.
231 void delete_from_page_cache(struct page
*page
)
233 struct address_space
*mapping
= page
->mapping
;
234 struct mem_cgroup
*memcg
;
237 void (*freepage
)(struct page
*);
239 BUG_ON(!PageLocked(page
));
241 freepage
= mapping
->a_ops
->freepage
;
243 memcg
= mem_cgroup_begin_page_stat(page
);
244 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
245 __delete_from_page_cache(page
, NULL
, memcg
);
246 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
247 mem_cgroup_end_page_stat(memcg
);
251 page_cache_release(page
);
253 EXPORT_SYMBOL(delete_from_page_cache
);
255 static int filemap_check_errors(struct address_space
*mapping
)
258 /* Check for outstanding write errors */
259 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
260 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
262 if (test_bit(AS_EIO
, &mapping
->flags
) &&
263 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
269 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
270 * @mapping: address space structure to write
271 * @start: offset in bytes where the range starts
272 * @end: offset in bytes where the range ends (inclusive)
273 * @sync_mode: enable synchronous operation
275 * Start writeback against all of a mapping's dirty pages that lie
276 * within the byte offsets <start, end> inclusive.
278 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
279 * opposed to a regular memory cleansing writeback. The difference between
280 * these two operations is that if a dirty page/buffer is encountered, it must
281 * be waited upon, and not just skipped over.
283 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
284 loff_t end
, int sync_mode
)
287 struct writeback_control wbc
= {
288 .sync_mode
= sync_mode
,
289 .nr_to_write
= LONG_MAX
,
290 .range_start
= start
,
294 if (!mapping_cap_writeback_dirty(mapping
))
297 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
298 ret
= do_writepages(mapping
, &wbc
);
299 wbc_detach_inode(&wbc
);
303 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
306 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
309 int filemap_fdatawrite(struct address_space
*mapping
)
311 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
313 EXPORT_SYMBOL(filemap_fdatawrite
);
315 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
318 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
320 EXPORT_SYMBOL(filemap_fdatawrite_range
);
323 * filemap_flush - mostly a non-blocking flush
324 * @mapping: target address_space
326 * This is a mostly non-blocking flush. Not suitable for data-integrity
327 * purposes - I/O may not be started against all dirty pages.
329 int filemap_flush(struct address_space
*mapping
)
331 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
333 EXPORT_SYMBOL(filemap_flush
);
335 static int __filemap_fdatawait_range(struct address_space
*mapping
,
336 loff_t start_byte
, loff_t end_byte
)
338 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
339 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
344 if (end_byte
< start_byte
)
347 pagevec_init(&pvec
, 0);
348 while ((index
<= end
) &&
349 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
350 PAGECACHE_TAG_WRITEBACK
,
351 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
354 for (i
= 0; i
< nr_pages
; i
++) {
355 struct page
*page
= pvec
.pages
[i
];
357 /* until radix tree lookup accepts end_index */
358 if (page
->index
> end
)
361 wait_on_page_writeback(page
);
362 if (TestClearPageError(page
))
365 pagevec_release(&pvec
);
373 * filemap_fdatawait_range - wait for writeback to complete
374 * @mapping: address space structure to wait for
375 * @start_byte: offset in bytes where the range starts
376 * @end_byte: offset in bytes where the range ends (inclusive)
378 * Walk the list of under-writeback pages of the given address space
379 * in the given range and wait for all of them. Check error status of
380 * the address space and return it.
382 * Since the error status of the address space is cleared by this function,
383 * callers are responsible for checking the return value and handling and/or
384 * reporting the error.
386 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
391 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
392 ret2
= filemap_check_errors(mapping
);
398 EXPORT_SYMBOL(filemap_fdatawait_range
);
401 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
402 * @mapping: address space structure to wait for
404 * Walk the list of under-writeback pages of the given address space
405 * and wait for all of them. Unlike filemap_fdatawait(), this function
406 * does not clear error status of the address space.
408 * Use this function if callers don't handle errors themselves. Expected
409 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
412 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
414 loff_t i_size
= i_size_read(mapping
->host
);
419 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
423 * filemap_fdatawait - wait for all under-writeback pages to complete
424 * @mapping: address space structure to wait for
426 * Walk the list of under-writeback pages of the given address space
427 * and wait for all of them. Check error status of the address space
430 * Since the error status of the address space is cleared by this function,
431 * callers are responsible for checking the return value and handling and/or
432 * reporting the error.
434 int filemap_fdatawait(struct address_space
*mapping
)
436 loff_t i_size
= i_size_read(mapping
->host
);
441 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
443 EXPORT_SYMBOL(filemap_fdatawait
);
445 int filemap_write_and_wait(struct address_space
*mapping
)
449 if (mapping
->nrpages
) {
450 err
= filemap_fdatawrite(mapping
);
452 * Even if the above returned error, the pages may be
453 * written partially (e.g. -ENOSPC), so we wait for it.
454 * But the -EIO is special case, it may indicate the worst
455 * thing (e.g. bug) happened, so we avoid waiting for it.
458 int err2
= filemap_fdatawait(mapping
);
463 err
= filemap_check_errors(mapping
);
467 EXPORT_SYMBOL(filemap_write_and_wait
);
470 * filemap_write_and_wait_range - write out & wait on a file range
471 * @mapping: the address_space for the pages
472 * @lstart: offset in bytes where the range starts
473 * @lend: offset in bytes where the range ends (inclusive)
475 * Write out and wait upon file offsets lstart->lend, inclusive.
477 * Note that `lend' is inclusive (describes the last byte to be written) so
478 * that this function can be used to write to the very end-of-file (end = -1).
480 int filemap_write_and_wait_range(struct address_space
*mapping
,
481 loff_t lstart
, loff_t lend
)
485 if (mapping
->nrpages
) {
486 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
488 /* See comment of filemap_write_and_wait() */
490 int err2
= filemap_fdatawait_range(mapping
,
496 err
= filemap_check_errors(mapping
);
500 EXPORT_SYMBOL(filemap_write_and_wait_range
);
503 * replace_page_cache_page - replace a pagecache page with a new one
504 * @old: page to be replaced
505 * @new: page to replace with
506 * @gfp_mask: allocation mode
508 * This function replaces a page in the pagecache with a new one. On
509 * success it acquires the pagecache reference for the new page and
510 * drops it for the old page. Both the old and new pages must be
511 * locked. This function does not add the new page to the LRU, the
512 * caller must do that.
514 * The remove + add is atomic. The only way this function can fail is
515 * memory allocation failure.
517 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
521 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
522 VM_BUG_ON_PAGE(!PageLocked(new), new);
523 VM_BUG_ON_PAGE(new->mapping
, new);
525 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
527 struct address_space
*mapping
= old
->mapping
;
528 void (*freepage
)(struct page
*);
529 struct mem_cgroup
*memcg
;
532 pgoff_t offset
= old
->index
;
533 freepage
= mapping
->a_ops
->freepage
;
536 new->mapping
= mapping
;
539 memcg
= mem_cgroup_begin_page_stat(old
);
540 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
541 __delete_from_page_cache(old
, NULL
, memcg
);
542 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
547 * hugetlb pages do not participate in page cache accounting.
550 __inc_zone_page_state(new, NR_FILE_PAGES
);
551 if (PageSwapBacked(new))
552 __inc_zone_page_state(new, NR_SHMEM
);
553 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
554 mem_cgroup_end_page_stat(memcg
);
555 mem_cgroup_replace_page(old
, new);
556 radix_tree_preload_end();
559 page_cache_release(old
);
564 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
566 static int page_cache_tree_insert(struct address_space
*mapping
,
567 struct page
*page
, void **shadowp
)
569 struct radix_tree_node
*node
;
573 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
,
580 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
581 if (!radix_tree_exceptional_entry(p
))
584 if (WARN_ON(dax_mapping(mapping
)))
589 mapping
->nrexceptional
--;
591 workingset_node_shadows_dec(node
);
593 radix_tree_replace_slot(slot
, page
);
596 workingset_node_pages_inc(node
);
598 * Don't track node that contains actual pages.
600 * Avoid acquiring the list_lru lock if already
601 * untracked. The list_empty() test is safe as
602 * node->private_list is protected by
603 * mapping->tree_lock.
605 if (!list_empty(&node
->private_list
))
606 list_lru_del(&workingset_shadow_nodes
,
607 &node
->private_list
);
612 static int __add_to_page_cache_locked(struct page
*page
,
613 struct address_space
*mapping
,
614 pgoff_t offset
, gfp_t gfp_mask
,
617 int huge
= PageHuge(page
);
618 struct mem_cgroup
*memcg
;
621 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
622 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
625 error
= mem_cgroup_try_charge(page
, current
->mm
,
626 gfp_mask
, &memcg
, false);
631 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
634 mem_cgroup_cancel_charge(page
, memcg
, false);
638 page_cache_get(page
);
639 page
->mapping
= mapping
;
640 page
->index
= offset
;
642 spin_lock_irq(&mapping
->tree_lock
);
643 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
644 radix_tree_preload_end();
648 /* hugetlb pages do not participate in page cache accounting. */
650 __inc_zone_page_state(page
, NR_FILE_PAGES
);
651 spin_unlock_irq(&mapping
->tree_lock
);
653 mem_cgroup_commit_charge(page
, memcg
, false, false);
654 trace_mm_filemap_add_to_page_cache(page
);
657 page
->mapping
= NULL
;
658 /* Leave page->index set: truncation relies upon it */
659 spin_unlock_irq(&mapping
->tree_lock
);
661 mem_cgroup_cancel_charge(page
, memcg
, false);
662 page_cache_release(page
);
667 * add_to_page_cache_locked - add a locked page to the pagecache
669 * @mapping: the page's address_space
670 * @offset: page index
671 * @gfp_mask: page allocation mode
673 * This function is used to add a page to the pagecache. It must be locked.
674 * This function does not add the page to the LRU. The caller must do that.
676 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
677 pgoff_t offset
, gfp_t gfp_mask
)
679 return __add_to_page_cache_locked(page
, mapping
, offset
,
682 EXPORT_SYMBOL(add_to_page_cache_locked
);
684 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
685 pgoff_t offset
, gfp_t gfp_mask
)
690 __SetPageLocked(page
);
691 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
694 __ClearPageLocked(page
);
697 * The page might have been evicted from cache only
698 * recently, in which case it should be activated like
699 * any other repeatedly accessed page.
701 if (shadow
&& workingset_refault(shadow
)) {
703 workingset_activation(page
);
705 ClearPageActive(page
);
710 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
713 struct page
*__page_cache_alloc(gfp_t gfp
)
718 if (cpuset_do_page_mem_spread()) {
719 unsigned int cpuset_mems_cookie
;
721 cpuset_mems_cookie
= read_mems_allowed_begin();
722 n
= cpuset_mem_spread_node();
723 page
= __alloc_pages_node(n
, gfp
, 0);
724 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
728 return alloc_pages(gfp
, 0);
730 EXPORT_SYMBOL(__page_cache_alloc
);
734 * In order to wait for pages to become available there must be
735 * waitqueues associated with pages. By using a hash table of
736 * waitqueues where the bucket discipline is to maintain all
737 * waiters on the same queue and wake all when any of the pages
738 * become available, and for the woken contexts to check to be
739 * sure the appropriate page became available, this saves space
740 * at a cost of "thundering herd" phenomena during rare hash
743 wait_queue_head_t
*page_waitqueue(struct page
*page
)
745 const struct zone
*zone
= page_zone(page
);
747 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
749 EXPORT_SYMBOL(page_waitqueue
);
751 void wait_on_page_bit(struct page
*page
, int bit_nr
)
753 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
755 if (test_bit(bit_nr
, &page
->flags
))
756 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
757 TASK_UNINTERRUPTIBLE
);
759 EXPORT_SYMBOL(wait_on_page_bit
);
761 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
763 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
765 if (!test_bit(bit_nr
, &page
->flags
))
768 return __wait_on_bit(page_waitqueue(page
), &wait
,
769 bit_wait_io
, TASK_KILLABLE
);
772 int wait_on_page_bit_killable_timeout(struct page
*page
,
773 int bit_nr
, unsigned long timeout
)
775 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
777 wait
.key
.timeout
= jiffies
+ timeout
;
778 if (!test_bit(bit_nr
, &page
->flags
))
780 return __wait_on_bit(page_waitqueue(page
), &wait
,
781 bit_wait_io_timeout
, TASK_KILLABLE
);
783 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
786 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
787 * @page: Page defining the wait queue of interest
788 * @waiter: Waiter to add to the queue
790 * Add an arbitrary @waiter to the wait queue for the nominated @page.
792 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
794 wait_queue_head_t
*q
= page_waitqueue(page
);
797 spin_lock_irqsave(&q
->lock
, flags
);
798 __add_wait_queue(q
, waiter
);
799 spin_unlock_irqrestore(&q
->lock
, flags
);
801 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
804 * unlock_page - unlock a locked page
807 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
808 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
809 * mechanism between PageLocked pages and PageWriteback pages is shared.
810 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
812 * The mb is necessary to enforce ordering between the clear_bit and the read
813 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
815 void unlock_page(struct page
*page
)
817 page
= compound_head(page
);
818 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
819 clear_bit_unlock(PG_locked
, &page
->flags
);
820 smp_mb__after_atomic();
821 wake_up_page(page
, PG_locked
);
823 EXPORT_SYMBOL(unlock_page
);
826 * end_page_writeback - end writeback against a page
829 void end_page_writeback(struct page
*page
)
832 * TestClearPageReclaim could be used here but it is an atomic
833 * operation and overkill in this particular case. Failing to
834 * shuffle a page marked for immediate reclaim is too mild to
835 * justify taking an atomic operation penalty at the end of
836 * ever page writeback.
838 if (PageReclaim(page
)) {
839 ClearPageReclaim(page
);
840 rotate_reclaimable_page(page
);
843 if (!test_clear_page_writeback(page
))
846 smp_mb__after_atomic();
847 wake_up_page(page
, PG_writeback
);
849 EXPORT_SYMBOL(end_page_writeback
);
852 * After completing I/O on a page, call this routine to update the page
853 * flags appropriately
855 void page_endio(struct page
*page
, int rw
, int err
)
859 SetPageUptodate(page
);
861 ClearPageUptodate(page
);
865 } else { /* rw == WRITE */
869 mapping_set_error(page
->mapping
, err
);
871 end_page_writeback(page
);
874 EXPORT_SYMBOL_GPL(page_endio
);
877 * __lock_page - get a lock on the page, assuming we need to sleep to get it
878 * @page: the page to lock
880 void __lock_page(struct page
*page
)
882 struct page
*page_head
= compound_head(page
);
883 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
885 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
886 TASK_UNINTERRUPTIBLE
);
888 EXPORT_SYMBOL(__lock_page
);
890 int __lock_page_killable(struct page
*page
)
892 struct page
*page_head
= compound_head(page
);
893 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
895 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
896 bit_wait_io
, TASK_KILLABLE
);
898 EXPORT_SYMBOL_GPL(__lock_page_killable
);
902 * 1 - page is locked; mmap_sem is still held.
903 * 0 - page is not locked.
904 * mmap_sem has been released (up_read()), unless flags had both
905 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
906 * which case mmap_sem is still held.
908 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
909 * with the page locked and the mmap_sem unperturbed.
911 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
914 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
916 * CAUTION! In this case, mmap_sem is not released
917 * even though return 0.
919 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
922 up_read(&mm
->mmap_sem
);
923 if (flags
& FAULT_FLAG_KILLABLE
)
924 wait_on_page_locked_killable(page
);
926 wait_on_page_locked(page
);
929 if (flags
& FAULT_FLAG_KILLABLE
) {
932 ret
= __lock_page_killable(page
);
934 up_read(&mm
->mmap_sem
);
944 * page_cache_next_hole - find the next hole (not-present entry)
947 * @max_scan: maximum range to search
949 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
950 * lowest indexed hole.
952 * Returns: the index of the hole if found, otherwise returns an index
953 * outside of the set specified (in which case 'return - index >=
954 * max_scan' will be true). In rare cases of index wrap-around, 0 will
957 * page_cache_next_hole may be called under rcu_read_lock. However,
958 * like radix_tree_gang_lookup, this will not atomically search a
959 * snapshot of the tree at a single point in time. For example, if a
960 * hole is created at index 5, then subsequently a hole is created at
961 * index 10, page_cache_next_hole covering both indexes may return 10
962 * if called under rcu_read_lock.
964 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
965 pgoff_t index
, unsigned long max_scan
)
969 for (i
= 0; i
< max_scan
; i
++) {
972 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
973 if (!page
|| radix_tree_exceptional_entry(page
))
982 EXPORT_SYMBOL(page_cache_next_hole
);
985 * page_cache_prev_hole - find the prev hole (not-present entry)
988 * @max_scan: maximum range to search
990 * Search backwards in the range [max(index-max_scan+1, 0), index] for
993 * Returns: the index of the hole if found, otherwise returns an index
994 * outside of the set specified (in which case 'index - return >=
995 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
998 * page_cache_prev_hole may be called under rcu_read_lock. However,
999 * like radix_tree_gang_lookup, this will not atomically search a
1000 * snapshot of the tree at a single point in time. For example, if a
1001 * hole is created at index 10, then subsequently a hole is created at
1002 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1003 * called under rcu_read_lock.
1005 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1006 pgoff_t index
, unsigned long max_scan
)
1010 for (i
= 0; i
< max_scan
; i
++) {
1013 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1014 if (!page
|| radix_tree_exceptional_entry(page
))
1017 if (index
== ULONG_MAX
)
1023 EXPORT_SYMBOL(page_cache_prev_hole
);
1026 * find_get_entry - find and get a page cache entry
1027 * @mapping: the address_space to search
1028 * @offset: the page cache index
1030 * Looks up the page cache slot at @mapping & @offset. If there is a
1031 * page cache page, it is returned with an increased refcount.
1033 * If the slot holds a shadow entry of a previously evicted page, or a
1034 * swap entry from shmem/tmpfs, it is returned.
1036 * Otherwise, %NULL is returned.
1038 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1046 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1048 page
= radix_tree_deref_slot(pagep
);
1049 if (unlikely(!page
))
1051 if (radix_tree_exception(page
)) {
1052 if (radix_tree_deref_retry(page
))
1055 * A shadow entry of a recently evicted page,
1056 * or a swap entry from shmem/tmpfs. Return
1057 * it without attempting to raise page count.
1061 if (!page_cache_get_speculative(page
))
1065 * Has the page moved?
1066 * This is part of the lockless pagecache protocol. See
1067 * include/linux/pagemap.h for details.
1069 if (unlikely(page
!= *pagep
)) {
1070 page_cache_release(page
);
1079 EXPORT_SYMBOL(find_get_entry
);
1082 * find_lock_entry - locate, pin and lock a page cache entry
1083 * @mapping: the address_space to search
1084 * @offset: the page cache index
1086 * Looks up the page cache slot at @mapping & @offset. If there is a
1087 * page cache page, it is returned locked and with an increased
1090 * If the slot holds a shadow entry of a previously evicted page, or a
1091 * swap entry from shmem/tmpfs, it is returned.
1093 * Otherwise, %NULL is returned.
1095 * find_lock_entry() may sleep.
1097 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1102 page
= find_get_entry(mapping
, offset
);
1103 if (page
&& !radix_tree_exception(page
)) {
1105 /* Has the page been truncated? */
1106 if (unlikely(page
->mapping
!= mapping
)) {
1108 page_cache_release(page
);
1111 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1115 EXPORT_SYMBOL(find_lock_entry
);
1118 * pagecache_get_page - find and get a page reference
1119 * @mapping: the address_space to search
1120 * @offset: the page index
1121 * @fgp_flags: PCG flags
1122 * @gfp_mask: gfp mask to use for the page cache data page allocation
1124 * Looks up the page cache slot at @mapping & @offset.
1126 * PCG flags modify how the page is returned.
1128 * FGP_ACCESSED: the page will be marked accessed
1129 * FGP_LOCK: Page is return locked
1130 * FGP_CREAT: If page is not present then a new page is allocated using
1131 * @gfp_mask and added to the page cache and the VM's LRU
1132 * list. The page is returned locked and with an increased
1133 * refcount. Otherwise, %NULL is returned.
1135 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1136 * if the GFP flags specified for FGP_CREAT are atomic.
1138 * If there is a page cache page, it is returned with an increased refcount.
1140 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1141 int fgp_flags
, gfp_t gfp_mask
)
1146 page
= find_get_entry(mapping
, offset
);
1147 if (radix_tree_exceptional_entry(page
))
1152 if (fgp_flags
& FGP_LOCK
) {
1153 if (fgp_flags
& FGP_NOWAIT
) {
1154 if (!trylock_page(page
)) {
1155 page_cache_release(page
);
1162 /* Has the page been truncated? */
1163 if (unlikely(page
->mapping
!= mapping
)) {
1165 page_cache_release(page
);
1168 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1171 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1172 mark_page_accessed(page
);
1175 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1177 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1178 gfp_mask
|= __GFP_WRITE
;
1179 if (fgp_flags
& FGP_NOFS
)
1180 gfp_mask
&= ~__GFP_FS
;
1182 page
= __page_cache_alloc(gfp_mask
);
1186 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1187 fgp_flags
|= FGP_LOCK
;
1189 /* Init accessed so avoid atomic mark_page_accessed later */
1190 if (fgp_flags
& FGP_ACCESSED
)
1191 __SetPageReferenced(page
);
1193 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1194 gfp_mask
& GFP_RECLAIM_MASK
);
1195 if (unlikely(err
)) {
1196 page_cache_release(page
);
1205 EXPORT_SYMBOL(pagecache_get_page
);
1208 * find_get_entries - gang pagecache lookup
1209 * @mapping: The address_space to search
1210 * @start: The starting page cache index
1211 * @nr_entries: The maximum number of entries
1212 * @entries: Where the resulting entries are placed
1213 * @indices: The cache indices corresponding to the entries in @entries
1215 * find_get_entries() will search for and return a group of up to
1216 * @nr_entries entries in the mapping. The entries are placed at
1217 * @entries. find_get_entries() takes a reference against any actual
1220 * The search returns a group of mapping-contiguous page cache entries
1221 * with ascending indexes. There may be holes in the indices due to
1222 * not-present pages.
1224 * Any shadow entries of evicted pages, or swap entries from
1225 * shmem/tmpfs, are included in the returned array.
1227 * find_get_entries() returns the number of pages and shadow entries
1230 unsigned find_get_entries(struct address_space
*mapping
,
1231 pgoff_t start
, unsigned int nr_entries
,
1232 struct page
**entries
, pgoff_t
*indices
)
1235 unsigned int ret
= 0;
1236 struct radix_tree_iter iter
;
1243 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1246 page
= radix_tree_deref_slot(slot
);
1247 if (unlikely(!page
))
1249 if (radix_tree_exception(page
)) {
1250 if (radix_tree_deref_retry(page
))
1253 * A shadow entry of a recently evicted page, a swap
1254 * entry from shmem/tmpfs or a DAX entry. Return it
1255 * without attempting to raise page count.
1259 if (!page_cache_get_speculative(page
))
1262 /* Has the page moved? */
1263 if (unlikely(page
!= *slot
)) {
1264 page_cache_release(page
);
1268 indices
[ret
] = iter
.index
;
1269 entries
[ret
] = page
;
1270 if (++ret
== nr_entries
)
1278 * find_get_pages - gang pagecache lookup
1279 * @mapping: The address_space to search
1280 * @start: The starting page index
1281 * @nr_pages: The maximum number of pages
1282 * @pages: Where the resulting pages are placed
1284 * find_get_pages() will search for and return a group of up to
1285 * @nr_pages pages in the mapping. The pages are placed at @pages.
1286 * find_get_pages() takes a reference against the returned pages.
1288 * The search returns a group of mapping-contiguous pages with ascending
1289 * indexes. There may be holes in the indices due to not-present pages.
1291 * find_get_pages() returns the number of pages which were found.
1293 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1294 unsigned int nr_pages
, struct page
**pages
)
1296 struct radix_tree_iter iter
;
1300 if (unlikely(!nr_pages
))
1305 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1308 page
= radix_tree_deref_slot(slot
);
1309 if (unlikely(!page
))
1312 if (radix_tree_exception(page
)) {
1313 if (radix_tree_deref_retry(page
)) {
1315 * Transient condition which can only trigger
1316 * when entry at index 0 moves out of or back
1317 * to root: none yet gotten, safe to restart.
1319 WARN_ON(iter
.index
);
1323 * A shadow entry of a recently evicted page,
1324 * or a swap entry from shmem/tmpfs. Skip
1330 if (!page_cache_get_speculative(page
))
1333 /* Has the page moved? */
1334 if (unlikely(page
!= *slot
)) {
1335 page_cache_release(page
);
1340 if (++ret
== nr_pages
)
1349 * find_get_pages_contig - gang contiguous pagecache lookup
1350 * @mapping: The address_space to search
1351 * @index: The starting page index
1352 * @nr_pages: The maximum number of pages
1353 * @pages: Where the resulting pages are placed
1355 * find_get_pages_contig() works exactly like find_get_pages(), except
1356 * that the returned number of pages are guaranteed to be contiguous.
1358 * find_get_pages_contig() returns the number of pages which were found.
1360 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1361 unsigned int nr_pages
, struct page
**pages
)
1363 struct radix_tree_iter iter
;
1365 unsigned int ret
= 0;
1367 if (unlikely(!nr_pages
))
1372 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1375 page
= radix_tree_deref_slot(slot
);
1376 /* The hole, there no reason to continue */
1377 if (unlikely(!page
))
1380 if (radix_tree_exception(page
)) {
1381 if (radix_tree_deref_retry(page
)) {
1383 * Transient condition which can only trigger
1384 * when entry at index 0 moves out of or back
1385 * to root: none yet gotten, safe to restart.
1390 * A shadow entry of a recently evicted page,
1391 * or a swap entry from shmem/tmpfs. Stop
1392 * looking for contiguous pages.
1397 if (!page_cache_get_speculative(page
))
1400 /* Has the page moved? */
1401 if (unlikely(page
!= *slot
)) {
1402 page_cache_release(page
);
1407 * must check mapping and index after taking the ref.
1408 * otherwise we can get both false positives and false
1409 * negatives, which is just confusing to the caller.
1411 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1412 page_cache_release(page
);
1417 if (++ret
== nr_pages
)
1423 EXPORT_SYMBOL(find_get_pages_contig
);
1426 * find_get_pages_tag - find and return pages that match @tag
1427 * @mapping: the address_space to search
1428 * @index: the starting page index
1429 * @tag: the tag index
1430 * @nr_pages: the maximum number of pages
1431 * @pages: where the resulting pages are placed
1433 * Like find_get_pages, except we only return pages which are tagged with
1434 * @tag. We update @index to index the next page for the traversal.
1436 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1437 int tag
, unsigned int nr_pages
, struct page
**pages
)
1439 struct radix_tree_iter iter
;
1443 if (unlikely(!nr_pages
))
1448 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1449 &iter
, *index
, tag
) {
1452 page
= radix_tree_deref_slot(slot
);
1453 if (unlikely(!page
))
1456 if (radix_tree_exception(page
)) {
1457 if (radix_tree_deref_retry(page
)) {
1459 * Transient condition which can only trigger
1460 * when entry at index 0 moves out of or back
1461 * to root: none yet gotten, safe to restart.
1466 * A shadow entry of a recently evicted page.
1468 * Those entries should never be tagged, but
1469 * this tree walk is lockless and the tags are
1470 * looked up in bulk, one radix tree node at a
1471 * time, so there is a sizable window for page
1472 * reclaim to evict a page we saw tagged.
1479 if (!page_cache_get_speculative(page
))
1482 /* Has the page moved? */
1483 if (unlikely(page
!= *slot
)) {
1484 page_cache_release(page
);
1489 if (++ret
== nr_pages
)
1496 *index
= pages
[ret
- 1]->index
+ 1;
1500 EXPORT_SYMBOL(find_get_pages_tag
);
1503 * find_get_entries_tag - find and return entries that match @tag
1504 * @mapping: the address_space to search
1505 * @start: the starting page cache index
1506 * @tag: the tag index
1507 * @nr_entries: the maximum number of entries
1508 * @entries: where the resulting entries are placed
1509 * @indices: the cache indices corresponding to the entries in @entries
1511 * Like find_get_entries, except we only return entries which are tagged with
1514 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1515 int tag
, unsigned int nr_entries
,
1516 struct page
**entries
, pgoff_t
*indices
)
1519 unsigned int ret
= 0;
1520 struct radix_tree_iter iter
;
1527 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1528 &iter
, start
, tag
) {
1531 page
= radix_tree_deref_slot(slot
);
1532 if (unlikely(!page
))
1534 if (radix_tree_exception(page
)) {
1535 if (radix_tree_deref_retry(page
)) {
1537 * Transient condition which can only trigger
1538 * when entry at index 0 moves out of or back
1539 * to root: none yet gotten, safe to restart.
1545 * A shadow entry of a recently evicted page, a swap
1546 * entry from shmem/tmpfs or a DAX entry. Return it
1547 * without attempting to raise page count.
1551 if (!page_cache_get_speculative(page
))
1554 /* Has the page moved? */
1555 if (unlikely(page
!= *slot
)) {
1556 page_cache_release(page
);
1560 indices
[ret
] = iter
.index
;
1561 entries
[ret
] = page
;
1562 if (++ret
== nr_entries
)
1568 EXPORT_SYMBOL(find_get_entries_tag
);
1571 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1572 * a _large_ part of the i/o request. Imagine the worst scenario:
1574 * ---R__________________________________________B__________
1575 * ^ reading here ^ bad block(assume 4k)
1577 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1578 * => failing the whole request => read(R) => read(R+1) =>
1579 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1580 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1581 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1583 * It is going insane. Fix it by quickly scaling down the readahead size.
1585 static void shrink_readahead_size_eio(struct file
*filp
,
1586 struct file_ra_state
*ra
)
1592 * do_generic_file_read - generic file read routine
1593 * @filp: the file to read
1594 * @ppos: current file position
1595 * @iter: data destination
1596 * @written: already copied
1598 * This is a generic file read routine, and uses the
1599 * mapping->a_ops->readpage() function for the actual low-level stuff.
1601 * This is really ugly. But the goto's actually try to clarify some
1602 * of the logic when it comes to error handling etc.
1604 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1605 struct iov_iter
*iter
, ssize_t written
)
1607 struct address_space
*mapping
= filp
->f_mapping
;
1608 struct inode
*inode
= mapping
->host
;
1609 struct file_ra_state
*ra
= &filp
->f_ra
;
1613 unsigned long offset
; /* offset into pagecache page */
1614 unsigned int prev_offset
;
1617 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1618 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1619 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1620 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1621 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1627 unsigned long nr
, ret
;
1631 page
= find_get_page(mapping
, index
);
1633 page_cache_sync_readahead(mapping
,
1635 index
, last_index
- index
);
1636 page
= find_get_page(mapping
, index
);
1637 if (unlikely(page
== NULL
))
1638 goto no_cached_page
;
1640 if (PageReadahead(page
)) {
1641 page_cache_async_readahead(mapping
,
1643 index
, last_index
- index
);
1645 if (!PageUptodate(page
)) {
1646 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1647 !mapping
->a_ops
->is_partially_uptodate
)
1648 goto page_not_up_to_date
;
1649 if (!trylock_page(page
))
1650 goto page_not_up_to_date
;
1651 /* Did it get truncated before we got the lock? */
1653 goto page_not_up_to_date_locked
;
1654 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1655 offset
, iter
->count
))
1656 goto page_not_up_to_date_locked
;
1661 * i_size must be checked after we know the page is Uptodate.
1663 * Checking i_size after the check allows us to calculate
1664 * the correct value for "nr", which means the zero-filled
1665 * part of the page is not copied back to userspace (unless
1666 * another truncate extends the file - this is desired though).
1669 isize
= i_size_read(inode
);
1670 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1671 if (unlikely(!isize
|| index
> end_index
)) {
1672 page_cache_release(page
);
1676 /* nr is the maximum number of bytes to copy from this page */
1677 nr
= PAGE_CACHE_SIZE
;
1678 if (index
== end_index
) {
1679 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1681 page_cache_release(page
);
1687 /* If users can be writing to this page using arbitrary
1688 * virtual addresses, take care about potential aliasing
1689 * before reading the page on the kernel side.
1691 if (mapping_writably_mapped(mapping
))
1692 flush_dcache_page(page
);
1695 * When a sequential read accesses a page several times,
1696 * only mark it as accessed the first time.
1698 if (prev_index
!= index
|| offset
!= prev_offset
)
1699 mark_page_accessed(page
);
1703 * Ok, we have the page, and it's up-to-date, so
1704 * now we can copy it to user space...
1707 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1709 index
+= offset
>> PAGE_CACHE_SHIFT
;
1710 offset
&= ~PAGE_CACHE_MASK
;
1711 prev_offset
= offset
;
1713 page_cache_release(page
);
1715 if (!iov_iter_count(iter
))
1723 page_not_up_to_date
:
1724 /* Get exclusive access to the page ... */
1725 error
= lock_page_killable(page
);
1726 if (unlikely(error
))
1727 goto readpage_error
;
1729 page_not_up_to_date_locked
:
1730 /* Did it get truncated before we got the lock? */
1731 if (!page
->mapping
) {
1733 page_cache_release(page
);
1737 /* Did somebody else fill it already? */
1738 if (PageUptodate(page
)) {
1745 * A previous I/O error may have been due to temporary
1746 * failures, eg. multipath errors.
1747 * PG_error will be set again if readpage fails.
1749 ClearPageError(page
);
1750 /* Start the actual read. The read will unlock the page. */
1751 error
= mapping
->a_ops
->readpage(filp
, page
);
1753 if (unlikely(error
)) {
1754 if (error
== AOP_TRUNCATED_PAGE
) {
1755 page_cache_release(page
);
1759 goto readpage_error
;
1762 if (!PageUptodate(page
)) {
1763 error
= lock_page_killable(page
);
1764 if (unlikely(error
))
1765 goto readpage_error
;
1766 if (!PageUptodate(page
)) {
1767 if (page
->mapping
== NULL
) {
1769 * invalidate_mapping_pages got it
1772 page_cache_release(page
);
1776 shrink_readahead_size_eio(filp
, ra
);
1778 goto readpage_error
;
1786 /* UHHUH! A synchronous read error occurred. Report it */
1787 page_cache_release(page
);
1792 * Ok, it wasn't cached, so we need to create a new
1795 page
= page_cache_alloc_cold(mapping
);
1800 error
= add_to_page_cache_lru(page
, mapping
, index
,
1801 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1803 page_cache_release(page
);
1804 if (error
== -EEXIST
) {
1814 ra
->prev_pos
= prev_index
;
1815 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1816 ra
->prev_pos
|= prev_offset
;
1818 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1819 file_accessed(filp
);
1820 return written
? written
: error
;
1824 * generic_file_read_iter - generic filesystem read routine
1825 * @iocb: kernel I/O control block
1826 * @iter: destination for the data read
1828 * This is the "read_iter()" routine for all filesystems
1829 * that can use the page cache directly.
1832 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1834 struct file
*file
= iocb
->ki_filp
;
1836 loff_t
*ppos
= &iocb
->ki_pos
;
1839 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1840 struct address_space
*mapping
= file
->f_mapping
;
1841 struct inode
*inode
= mapping
->host
;
1842 size_t count
= iov_iter_count(iter
);
1846 goto out
; /* skip atime */
1847 size
= i_size_read(inode
);
1848 retval
= filemap_write_and_wait_range(mapping
, pos
,
1851 struct iov_iter data
= *iter
;
1852 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
1856 *ppos
= pos
+ retval
;
1857 iov_iter_advance(iter
, retval
);
1861 * Btrfs can have a short DIO read if we encounter
1862 * compressed extents, so if there was an error, or if
1863 * we've already read everything we wanted to, or if
1864 * there was a short read because we hit EOF, go ahead
1865 * and return. Otherwise fallthrough to buffered io for
1866 * the rest of the read. Buffered reads will not work for
1867 * DAX files, so don't bother trying.
1869 if (retval
< 0 || !iov_iter_count(iter
) || *ppos
>= size
||
1871 file_accessed(file
);
1876 retval
= do_generic_file_read(file
, ppos
, iter
, retval
);
1880 EXPORT_SYMBOL(generic_file_read_iter
);
1884 * page_cache_read - adds requested page to the page cache if not already there
1885 * @file: file to read
1886 * @offset: page index
1888 * This adds the requested page to the page cache if it isn't already there,
1889 * and schedules an I/O to read in its contents from disk.
1891 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1893 struct address_space
*mapping
= file
->f_mapping
;
1898 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1902 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1904 ret
= mapping
->a_ops
->readpage(file
, page
);
1905 else if (ret
== -EEXIST
)
1906 ret
= 0; /* losing race to add is OK */
1908 page_cache_release(page
);
1910 } while (ret
== AOP_TRUNCATED_PAGE
);
1915 #define MMAP_LOTSAMISS (100)
1918 * Synchronous readahead happens when we don't even find
1919 * a page in the page cache at all.
1921 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1922 struct file_ra_state
*ra
,
1926 struct address_space
*mapping
= file
->f_mapping
;
1928 /* If we don't want any read-ahead, don't bother */
1929 if (vma
->vm_flags
& VM_RAND_READ
)
1934 if (vma
->vm_flags
& VM_SEQ_READ
) {
1935 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1940 /* Avoid banging the cache line if not needed */
1941 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1945 * Do we miss much more than hit in this file? If so,
1946 * stop bothering with read-ahead. It will only hurt.
1948 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1954 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
1955 ra
->size
= ra
->ra_pages
;
1956 ra
->async_size
= ra
->ra_pages
/ 4;
1957 ra_submit(ra
, mapping
, file
);
1961 * Asynchronous readahead happens when we find the page and PG_readahead,
1962 * so we want to possibly extend the readahead further..
1964 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1965 struct file_ra_state
*ra
,
1970 struct address_space
*mapping
= file
->f_mapping
;
1972 /* If we don't want any read-ahead, don't bother */
1973 if (vma
->vm_flags
& VM_RAND_READ
)
1975 if (ra
->mmap_miss
> 0)
1977 if (PageReadahead(page
))
1978 page_cache_async_readahead(mapping
, ra
, file
,
1979 page
, offset
, ra
->ra_pages
);
1983 * filemap_fault - read in file data for page fault handling
1984 * @vma: vma in which the fault was taken
1985 * @vmf: struct vm_fault containing details of the fault
1987 * filemap_fault() is invoked via the vma operations vector for a
1988 * mapped memory region to read in file data during a page fault.
1990 * The goto's are kind of ugly, but this streamlines the normal case of having
1991 * it in the page cache, and handles the special cases reasonably without
1992 * having a lot of duplicated code.
1994 * vma->vm_mm->mmap_sem must be held on entry.
1996 * If our return value has VM_FAULT_RETRY set, it's because
1997 * lock_page_or_retry() returned 0.
1998 * The mmap_sem has usually been released in this case.
1999 * See __lock_page_or_retry() for the exception.
2001 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2002 * has not been released.
2004 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2006 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2009 struct file
*file
= vma
->vm_file
;
2010 struct address_space
*mapping
= file
->f_mapping
;
2011 struct file_ra_state
*ra
= &file
->f_ra
;
2012 struct inode
*inode
= mapping
->host
;
2013 pgoff_t offset
= vmf
->pgoff
;
2018 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2019 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
2020 return VM_FAULT_SIGBUS
;
2023 * Do we have something in the page cache already?
2025 page
= find_get_page(mapping
, offset
);
2026 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2028 * We found the page, so try async readahead before
2029 * waiting for the lock.
2031 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2033 /* No page in the page cache at all */
2034 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2035 count_vm_event(PGMAJFAULT
);
2036 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2037 ret
= VM_FAULT_MAJOR
;
2039 page
= find_get_page(mapping
, offset
);
2041 goto no_cached_page
;
2044 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2045 page_cache_release(page
);
2046 return ret
| VM_FAULT_RETRY
;
2049 /* Did it get truncated? */
2050 if (unlikely(page
->mapping
!= mapping
)) {
2055 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2058 * We have a locked page in the page cache, now we need to check
2059 * that it's up-to-date. If not, it is going to be due to an error.
2061 if (unlikely(!PageUptodate(page
)))
2062 goto page_not_uptodate
;
2065 * Found the page and have a reference on it.
2066 * We must recheck i_size under page lock.
2068 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2069 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
2071 page_cache_release(page
);
2072 return VM_FAULT_SIGBUS
;
2076 return ret
| VM_FAULT_LOCKED
;
2080 * We're only likely to ever get here if MADV_RANDOM is in
2083 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2086 * The page we want has now been added to the page cache.
2087 * In the unlikely event that someone removed it in the
2088 * meantime, we'll just come back here and read it again.
2094 * An error return from page_cache_read can result if the
2095 * system is low on memory, or a problem occurs while trying
2098 if (error
== -ENOMEM
)
2099 return VM_FAULT_OOM
;
2100 return VM_FAULT_SIGBUS
;
2104 * Umm, take care of errors if the page isn't up-to-date.
2105 * Try to re-read it _once_. We do this synchronously,
2106 * because there really aren't any performance issues here
2107 * and we need to check for errors.
2109 ClearPageError(page
);
2110 error
= mapping
->a_ops
->readpage(file
, page
);
2112 wait_on_page_locked(page
);
2113 if (!PageUptodate(page
))
2116 page_cache_release(page
);
2118 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2121 /* Things didn't work out. Return zero to tell the mm layer so. */
2122 shrink_readahead_size_eio(file
, ra
);
2123 return VM_FAULT_SIGBUS
;
2125 EXPORT_SYMBOL(filemap_fault
);
2127 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2129 struct radix_tree_iter iter
;
2131 struct file
*file
= vma
->vm_file
;
2132 struct address_space
*mapping
= file
->f_mapping
;
2135 unsigned long address
= (unsigned long) vmf
->virtual_address
;
2140 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
2141 if (iter
.index
> vmf
->max_pgoff
)
2144 page
= radix_tree_deref_slot(slot
);
2145 if (unlikely(!page
))
2147 if (radix_tree_exception(page
)) {
2148 if (radix_tree_deref_retry(page
))
2154 if (!page_cache_get_speculative(page
))
2157 /* Has the page moved? */
2158 if (unlikely(page
!= *slot
)) {
2159 page_cache_release(page
);
2163 if (!PageUptodate(page
) ||
2164 PageReadahead(page
) ||
2167 if (!trylock_page(page
))
2170 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2173 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
2174 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2177 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2178 if (!pte_none(*pte
))
2181 if (file
->f_ra
.mmap_miss
> 0)
2182 file
->f_ra
.mmap_miss
--;
2183 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2184 do_set_pte(vma
, addr
, page
, pte
, false, false);
2190 page_cache_release(page
);
2192 if (iter
.index
== vmf
->max_pgoff
)
2197 EXPORT_SYMBOL(filemap_map_pages
);
2199 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2201 struct page
*page
= vmf
->page
;
2202 struct inode
*inode
= file_inode(vma
->vm_file
);
2203 int ret
= VM_FAULT_LOCKED
;
2205 sb_start_pagefault(inode
->i_sb
);
2206 file_update_time(vma
->vm_file
);
2208 if (page
->mapping
!= inode
->i_mapping
) {
2210 ret
= VM_FAULT_NOPAGE
;
2214 * We mark the page dirty already here so that when freeze is in
2215 * progress, we are guaranteed that writeback during freezing will
2216 * see the dirty page and writeprotect it again.
2218 set_page_dirty(page
);
2219 wait_for_stable_page(page
);
2221 sb_end_pagefault(inode
->i_sb
);
2224 EXPORT_SYMBOL(filemap_page_mkwrite
);
2226 const struct vm_operations_struct generic_file_vm_ops
= {
2227 .fault
= filemap_fault
,
2228 .map_pages
= filemap_map_pages
,
2229 .page_mkwrite
= filemap_page_mkwrite
,
2232 /* This is used for a general mmap of a disk file */
2234 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2236 struct address_space
*mapping
= file
->f_mapping
;
2238 if (!mapping
->a_ops
->readpage
)
2240 file_accessed(file
);
2241 vma
->vm_ops
= &generic_file_vm_ops
;
2246 * This is for filesystems which do not implement ->writepage.
2248 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2250 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2252 return generic_file_mmap(file
, vma
);
2255 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2259 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2263 #endif /* CONFIG_MMU */
2265 EXPORT_SYMBOL(generic_file_mmap
);
2266 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2268 static struct page
*wait_on_page_read(struct page
*page
)
2270 if (!IS_ERR(page
)) {
2271 wait_on_page_locked(page
);
2272 if (!PageUptodate(page
)) {
2273 page_cache_release(page
);
2274 page
= ERR_PTR(-EIO
);
2280 static struct page
*__read_cache_page(struct address_space
*mapping
,
2282 int (*filler
)(void *, struct page
*),
2289 page
= find_get_page(mapping
, index
);
2291 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2293 return ERR_PTR(-ENOMEM
);
2294 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2295 if (unlikely(err
)) {
2296 page_cache_release(page
);
2299 /* Presumably ENOMEM for radix tree node */
2300 return ERR_PTR(err
);
2302 err
= filler(data
, page
);
2304 page_cache_release(page
);
2305 page
= ERR_PTR(err
);
2307 page
= wait_on_page_read(page
);
2313 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2315 int (*filler
)(void *, struct page
*),
2324 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
2327 if (PageUptodate(page
))
2331 if (!page
->mapping
) {
2333 page_cache_release(page
);
2336 if (PageUptodate(page
)) {
2340 err
= filler(data
, page
);
2342 page_cache_release(page
);
2343 return ERR_PTR(err
);
2345 page
= wait_on_page_read(page
);
2350 mark_page_accessed(page
);
2355 * read_cache_page - read into page cache, fill it if needed
2356 * @mapping: the page's address_space
2357 * @index: the page index
2358 * @filler: function to perform the read
2359 * @data: first arg to filler(data, page) function, often left as NULL
2361 * Read into the page cache. If a page already exists, and PageUptodate() is
2362 * not set, try to fill the page and wait for it to become unlocked.
2364 * If the page does not get brought uptodate, return -EIO.
2366 struct page
*read_cache_page(struct address_space
*mapping
,
2368 int (*filler
)(void *, struct page
*),
2371 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2373 EXPORT_SYMBOL(read_cache_page
);
2376 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2377 * @mapping: the page's address_space
2378 * @index: the page index
2379 * @gfp: the page allocator flags to use if allocating
2381 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2382 * any new page allocations done using the specified allocation flags.
2384 * If the page does not get brought uptodate, return -EIO.
2386 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2390 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2392 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2394 EXPORT_SYMBOL(read_cache_page_gfp
);
2397 * Performs necessary checks before doing a write
2399 * Can adjust writing position or amount of bytes to write.
2400 * Returns appropriate error code that caller should return or
2401 * zero in case that write should be allowed.
2403 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2405 struct file
*file
= iocb
->ki_filp
;
2406 struct inode
*inode
= file
->f_mapping
->host
;
2407 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2410 if (!iov_iter_count(from
))
2413 /* FIXME: this is for backwards compatibility with 2.4 */
2414 if (iocb
->ki_flags
& IOCB_APPEND
)
2415 iocb
->ki_pos
= i_size_read(inode
);
2419 if (limit
!= RLIM_INFINITY
) {
2420 if (iocb
->ki_pos
>= limit
) {
2421 send_sig(SIGXFSZ
, current
, 0);
2424 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2430 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2431 !(file
->f_flags
& O_LARGEFILE
))) {
2432 if (pos
>= MAX_NON_LFS
)
2434 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2438 * Are we about to exceed the fs block limit ?
2440 * If we have written data it becomes a short write. If we have
2441 * exceeded without writing data we send a signal and return EFBIG.
2442 * Linus frestrict idea will clean these up nicely..
2444 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2447 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2448 return iov_iter_count(from
);
2450 EXPORT_SYMBOL(generic_write_checks
);
2452 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2453 loff_t pos
, unsigned len
, unsigned flags
,
2454 struct page
**pagep
, void **fsdata
)
2456 const struct address_space_operations
*aops
= mapping
->a_ops
;
2458 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2461 EXPORT_SYMBOL(pagecache_write_begin
);
2463 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2464 loff_t pos
, unsigned len
, unsigned copied
,
2465 struct page
*page
, void *fsdata
)
2467 const struct address_space_operations
*aops
= mapping
->a_ops
;
2469 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2471 EXPORT_SYMBOL(pagecache_write_end
);
2474 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
2476 struct file
*file
= iocb
->ki_filp
;
2477 struct address_space
*mapping
= file
->f_mapping
;
2478 struct inode
*inode
= mapping
->host
;
2482 struct iov_iter data
;
2484 write_len
= iov_iter_count(from
);
2485 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2487 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2492 * After a write we want buffered reads to be sure to go to disk to get
2493 * the new data. We invalidate clean cached page from the region we're
2494 * about to write. We do this *before* the write so that we can return
2495 * without clobbering -EIOCBQUEUED from ->direct_IO().
2497 if (mapping
->nrpages
) {
2498 written
= invalidate_inode_pages2_range(mapping
,
2499 pos
>> PAGE_CACHE_SHIFT
, end
);
2501 * If a page can not be invalidated, return 0 to fall back
2502 * to buffered write.
2505 if (written
== -EBUSY
)
2512 written
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
2515 * Finally, try again to invalidate clean pages which might have been
2516 * cached by non-direct readahead, or faulted in by get_user_pages()
2517 * if the source of the write was an mmap'ed region of the file
2518 * we're writing. Either one is a pretty crazy thing to do,
2519 * so we don't support it 100%. If this invalidation
2520 * fails, tough, the write still worked...
2522 if (mapping
->nrpages
) {
2523 invalidate_inode_pages2_range(mapping
,
2524 pos
>> PAGE_CACHE_SHIFT
, end
);
2529 iov_iter_advance(from
, written
);
2530 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2531 i_size_write(inode
, pos
);
2532 mark_inode_dirty(inode
);
2539 EXPORT_SYMBOL(generic_file_direct_write
);
2542 * Find or create a page at the given pagecache position. Return the locked
2543 * page. This function is specifically for buffered writes.
2545 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2546 pgoff_t index
, unsigned flags
)
2549 int fgp_flags
= FGP_LOCK
|FGP_ACCESSED
|FGP_WRITE
|FGP_CREAT
;
2551 if (flags
& AOP_FLAG_NOFS
)
2552 fgp_flags
|= FGP_NOFS
;
2554 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2555 mapping_gfp_mask(mapping
));
2557 wait_for_stable_page(page
);
2561 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2563 ssize_t
generic_perform_write(struct file
*file
,
2564 struct iov_iter
*i
, loff_t pos
)
2566 struct address_space
*mapping
= file
->f_mapping
;
2567 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2569 ssize_t written
= 0;
2570 unsigned int flags
= 0;
2573 * Copies from kernel address space cannot fail (NFSD is a big user).
2575 if (!iter_is_iovec(i
))
2576 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2580 unsigned long offset
; /* Offset into pagecache page */
2581 unsigned long bytes
; /* Bytes to write to page */
2582 size_t copied
; /* Bytes copied from user */
2585 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2586 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2591 * Bring in the user page that we will copy from _first_.
2592 * Otherwise there's a nasty deadlock on copying from the
2593 * same page as we're writing to, without it being marked
2596 * Not only is this an optimisation, but it is also required
2597 * to check that the address is actually valid, when atomic
2598 * usercopies are used, below.
2600 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2605 if (fatal_signal_pending(current
)) {
2610 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2612 if (unlikely(status
< 0))
2615 if (mapping_writably_mapped(mapping
))
2616 flush_dcache_page(page
);
2618 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2619 flush_dcache_page(page
);
2621 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2623 if (unlikely(status
< 0))
2629 iov_iter_advance(i
, copied
);
2630 if (unlikely(copied
== 0)) {
2632 * If we were unable to copy any data at all, we must
2633 * fall back to a single segment length write.
2635 * If we didn't fallback here, we could livelock
2636 * because not all segments in the iov can be copied at
2637 * once without a pagefault.
2639 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2640 iov_iter_single_seg_count(i
));
2646 balance_dirty_pages_ratelimited(mapping
);
2647 } while (iov_iter_count(i
));
2649 return written
? written
: status
;
2651 EXPORT_SYMBOL(generic_perform_write
);
2654 * __generic_file_write_iter - write data to a file
2655 * @iocb: IO state structure (file, offset, etc.)
2656 * @from: iov_iter with data to write
2658 * This function does all the work needed for actually writing data to a
2659 * file. It does all basic checks, removes SUID from the file, updates
2660 * modification times and calls proper subroutines depending on whether we
2661 * do direct IO or a standard buffered write.
2663 * It expects i_mutex to be grabbed unless we work on a block device or similar
2664 * object which does not need locking at all.
2666 * This function does *not* take care of syncing data in case of O_SYNC write.
2667 * A caller has to handle it. This is mainly due to the fact that we want to
2668 * avoid syncing under i_mutex.
2670 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2672 struct file
*file
= iocb
->ki_filp
;
2673 struct address_space
* mapping
= file
->f_mapping
;
2674 struct inode
*inode
= mapping
->host
;
2675 ssize_t written
= 0;
2679 /* We can write back this queue in page reclaim */
2680 current
->backing_dev_info
= inode_to_bdi(inode
);
2681 err
= file_remove_privs(file
);
2685 err
= file_update_time(file
);
2689 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2690 loff_t pos
, endbyte
;
2692 written
= generic_file_direct_write(iocb
, from
, iocb
->ki_pos
);
2694 * If the write stopped short of completing, fall back to
2695 * buffered writes. Some filesystems do this for writes to
2696 * holes, for example. For DAX files, a buffered write will
2697 * not succeed (even if it did, DAX does not handle dirty
2698 * page-cache pages correctly).
2700 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2703 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2705 * If generic_perform_write() returned a synchronous error
2706 * then we want to return the number of bytes which were
2707 * direct-written, or the error code if that was zero. Note
2708 * that this differs from normal direct-io semantics, which
2709 * will return -EFOO even if some bytes were written.
2711 if (unlikely(status
< 0)) {
2716 * We need to ensure that the page cache pages are written to
2717 * disk and invalidated to preserve the expected O_DIRECT
2720 endbyte
= pos
+ status
- 1;
2721 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2723 iocb
->ki_pos
= endbyte
+ 1;
2725 invalidate_mapping_pages(mapping
,
2726 pos
>> PAGE_CACHE_SHIFT
,
2727 endbyte
>> PAGE_CACHE_SHIFT
);
2730 * We don't know how much we wrote, so just return
2731 * the number of bytes which were direct-written
2735 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2736 if (likely(written
> 0))
2737 iocb
->ki_pos
+= written
;
2740 current
->backing_dev_info
= NULL
;
2741 return written
? written
: err
;
2743 EXPORT_SYMBOL(__generic_file_write_iter
);
2746 * generic_file_write_iter - write data to a file
2747 * @iocb: IO state structure
2748 * @from: iov_iter with data to write
2750 * This is a wrapper around __generic_file_write_iter() to be used by most
2751 * filesystems. It takes care of syncing the file in case of O_SYNC file
2752 * and acquires i_mutex as needed.
2754 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2756 struct file
*file
= iocb
->ki_filp
;
2757 struct inode
*inode
= file
->f_mapping
->host
;
2760 mutex_lock(&inode
->i_mutex
);
2761 ret
= generic_write_checks(iocb
, from
);
2763 ret
= __generic_file_write_iter(iocb
, from
);
2764 mutex_unlock(&inode
->i_mutex
);
2769 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2775 EXPORT_SYMBOL(generic_file_write_iter
);
2778 * try_to_release_page() - release old fs-specific metadata on a page
2780 * @page: the page which the kernel is trying to free
2781 * @gfp_mask: memory allocation flags (and I/O mode)
2783 * The address_space is to try to release any data against the page
2784 * (presumably at page->private). If the release was successful, return `1'.
2785 * Otherwise return zero.
2787 * This may also be called if PG_fscache is set on a page, indicating that the
2788 * page is known to the local caching routines.
2790 * The @gfp_mask argument specifies whether I/O may be performed to release
2791 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2794 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2796 struct address_space
* const mapping
= page
->mapping
;
2798 BUG_ON(!PageLocked(page
));
2799 if (PageWriteback(page
))
2802 if (mapping
&& mapping
->a_ops
->releasepage
)
2803 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2804 return try_to_free_buffers(page
);
2807 EXPORT_SYMBOL(try_to_release_page
);