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 VM_BUG_ON_PAGE(page_mapped(page
), page
);
199 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
202 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
203 current
->comm
, page_to_pfn(page
));
204 dump_page(page
, "still mapped when deleted");
206 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
208 mapcount
= page_mapcount(page
);
209 if (mapping_exiting(mapping
) &&
210 page_count(page
) >= mapcount
+ 2) {
212 * All vmas have already been torn down, so it's
213 * a good bet that actually the page is unmapped,
214 * and we'd prefer not to leak it: if we're wrong,
215 * some other bad page check should catch it later.
217 page_mapcount_reset(page
);
218 atomic_sub(mapcount
, &page
->_count
);
222 page_cache_tree_delete(mapping
, page
, shadow
);
224 page
->mapping
= NULL
;
225 /* Leave page->index set: truncation lookup relies upon it */
227 /* hugetlb pages do not participate in page cache accounting. */
229 __dec_zone_page_state(page
, NR_FILE_PAGES
);
230 if (PageSwapBacked(page
))
231 __dec_zone_page_state(page
, NR_SHMEM
);
234 * At this point page must be either written or cleaned by truncate.
235 * Dirty page here signals a bug and loss of unwritten data.
237 * This fixes dirty accounting after removing the page entirely but
238 * leaves PageDirty set: it has no effect for truncated page and
239 * anyway will be cleared before returning page into buddy allocator.
241 if (WARN_ON_ONCE(PageDirty(page
)))
242 account_page_cleaned(page
, mapping
, memcg
,
243 inode_to_wb(mapping
->host
));
247 * delete_from_page_cache - delete page from page cache
248 * @page: the page which the kernel is trying to remove from page cache
250 * This must be called only on pages that have been verified to be in the page
251 * cache and locked. It will never put the page into the free list, the caller
252 * has a reference on the page.
254 void delete_from_page_cache(struct page
*page
)
256 struct address_space
*mapping
= page
->mapping
;
257 struct mem_cgroup
*memcg
;
260 void (*freepage
)(struct page
*);
262 BUG_ON(!PageLocked(page
));
264 freepage
= mapping
->a_ops
->freepage
;
266 memcg
= mem_cgroup_begin_page_stat(page
);
267 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
268 __delete_from_page_cache(page
, NULL
, memcg
);
269 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
270 mem_cgroup_end_page_stat(memcg
);
274 page_cache_release(page
);
276 EXPORT_SYMBOL(delete_from_page_cache
);
278 static int filemap_check_errors(struct address_space
*mapping
)
281 /* Check for outstanding write errors */
282 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
283 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
285 if (test_bit(AS_EIO
, &mapping
->flags
) &&
286 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
292 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
293 * @mapping: address space structure to write
294 * @start: offset in bytes where the range starts
295 * @end: offset in bytes where the range ends (inclusive)
296 * @sync_mode: enable synchronous operation
298 * Start writeback against all of a mapping's dirty pages that lie
299 * within the byte offsets <start, end> inclusive.
301 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
302 * opposed to a regular memory cleansing writeback. The difference between
303 * these two operations is that if a dirty page/buffer is encountered, it must
304 * be waited upon, and not just skipped over.
306 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
307 loff_t end
, int sync_mode
)
310 struct writeback_control wbc
= {
311 .sync_mode
= sync_mode
,
312 .nr_to_write
= LONG_MAX
,
313 .range_start
= start
,
317 if (!mapping_cap_writeback_dirty(mapping
))
320 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
321 ret
= do_writepages(mapping
, &wbc
);
322 wbc_detach_inode(&wbc
);
326 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
329 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
332 int filemap_fdatawrite(struct address_space
*mapping
)
334 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
336 EXPORT_SYMBOL(filemap_fdatawrite
);
338 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
341 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
343 EXPORT_SYMBOL(filemap_fdatawrite_range
);
346 * filemap_flush - mostly a non-blocking flush
347 * @mapping: target address_space
349 * This is a mostly non-blocking flush. Not suitable for data-integrity
350 * purposes - I/O may not be started against all dirty pages.
352 int filemap_flush(struct address_space
*mapping
)
354 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
356 EXPORT_SYMBOL(filemap_flush
);
358 static int __filemap_fdatawait_range(struct address_space
*mapping
,
359 loff_t start_byte
, loff_t end_byte
)
361 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
362 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
367 if (end_byte
< start_byte
)
370 pagevec_init(&pvec
, 0);
371 while ((index
<= end
) &&
372 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
373 PAGECACHE_TAG_WRITEBACK
,
374 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
377 for (i
= 0; i
< nr_pages
; i
++) {
378 struct page
*page
= pvec
.pages
[i
];
380 /* until radix tree lookup accepts end_index */
381 if (page
->index
> end
)
384 wait_on_page_writeback(page
);
385 if (TestClearPageError(page
))
388 pagevec_release(&pvec
);
396 * filemap_fdatawait_range - wait for writeback to complete
397 * @mapping: address space structure to wait for
398 * @start_byte: offset in bytes where the range starts
399 * @end_byte: offset in bytes where the range ends (inclusive)
401 * Walk the list of under-writeback pages of the given address space
402 * in the given range and wait for all of them. Check error status of
403 * the address space and return it.
405 * Since the error status of the address space is cleared by this function,
406 * callers are responsible for checking the return value and handling and/or
407 * reporting the error.
409 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
414 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
415 ret2
= filemap_check_errors(mapping
);
421 EXPORT_SYMBOL(filemap_fdatawait_range
);
424 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
425 * @mapping: address space structure to wait for
427 * Walk the list of under-writeback pages of the given address space
428 * and wait for all of them. Unlike filemap_fdatawait(), this function
429 * does not clear error status of the address space.
431 * Use this function if callers don't handle errors themselves. Expected
432 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
435 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
437 loff_t i_size
= i_size_read(mapping
->host
);
442 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
446 * filemap_fdatawait - wait for all under-writeback pages to complete
447 * @mapping: address space structure to wait for
449 * Walk the list of under-writeback pages of the given address space
450 * and wait for all of them. Check error status of the address space
453 * Since the error status of the address space is cleared by this function,
454 * callers are responsible for checking the return value and handling and/or
455 * reporting the error.
457 int filemap_fdatawait(struct address_space
*mapping
)
459 loff_t i_size
= i_size_read(mapping
->host
);
464 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
466 EXPORT_SYMBOL(filemap_fdatawait
);
468 int filemap_write_and_wait(struct address_space
*mapping
)
472 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
473 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
474 err
= filemap_fdatawrite(mapping
);
476 * Even if the above returned error, the pages may be
477 * written partially (e.g. -ENOSPC), so we wait for it.
478 * But the -EIO is special case, it may indicate the worst
479 * thing (e.g. bug) happened, so we avoid waiting for it.
482 int err2
= filemap_fdatawait(mapping
);
487 err
= filemap_check_errors(mapping
);
491 EXPORT_SYMBOL(filemap_write_and_wait
);
494 * filemap_write_and_wait_range - write out & wait on a file range
495 * @mapping: the address_space for the pages
496 * @lstart: offset in bytes where the range starts
497 * @lend: offset in bytes where the range ends (inclusive)
499 * Write out and wait upon file offsets lstart->lend, inclusive.
501 * Note that `lend' is inclusive (describes the last byte to be written) so
502 * that this function can be used to write to the very end-of-file (end = -1).
504 int filemap_write_and_wait_range(struct address_space
*mapping
,
505 loff_t lstart
, loff_t lend
)
509 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
510 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
511 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
513 /* See comment of filemap_write_and_wait() */
515 int err2
= filemap_fdatawait_range(mapping
,
521 err
= filemap_check_errors(mapping
);
525 EXPORT_SYMBOL(filemap_write_and_wait_range
);
528 * replace_page_cache_page - replace a pagecache page with a new one
529 * @old: page to be replaced
530 * @new: page to replace with
531 * @gfp_mask: allocation mode
533 * This function replaces a page in the pagecache with a new one. On
534 * success it acquires the pagecache reference for the new page and
535 * drops it for the old page. Both the old and new pages must be
536 * locked. This function does not add the new page to the LRU, the
537 * caller must do that.
539 * The remove + add is atomic. The only way this function can fail is
540 * memory allocation failure.
542 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
546 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
547 VM_BUG_ON_PAGE(!PageLocked(new), new);
548 VM_BUG_ON_PAGE(new->mapping
, new);
550 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
552 struct address_space
*mapping
= old
->mapping
;
553 void (*freepage
)(struct page
*);
554 struct mem_cgroup
*memcg
;
557 pgoff_t offset
= old
->index
;
558 freepage
= mapping
->a_ops
->freepage
;
561 new->mapping
= mapping
;
564 memcg
= mem_cgroup_begin_page_stat(old
);
565 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
566 __delete_from_page_cache(old
, NULL
, memcg
);
567 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
572 * hugetlb pages do not participate in page cache accounting.
575 __inc_zone_page_state(new, NR_FILE_PAGES
);
576 if (PageSwapBacked(new))
577 __inc_zone_page_state(new, NR_SHMEM
);
578 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
579 mem_cgroup_end_page_stat(memcg
);
580 mem_cgroup_replace_page(old
, new);
581 radix_tree_preload_end();
584 page_cache_release(old
);
589 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
591 static int page_cache_tree_insert(struct address_space
*mapping
,
592 struct page
*page
, void **shadowp
)
594 struct radix_tree_node
*node
;
598 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
,
605 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
606 if (!radix_tree_exceptional_entry(p
))
609 if (WARN_ON(dax_mapping(mapping
)))
614 mapping
->nrexceptional
--;
616 workingset_node_shadows_dec(node
);
618 radix_tree_replace_slot(slot
, page
);
621 workingset_node_pages_inc(node
);
623 * Don't track node that contains actual pages.
625 * Avoid acquiring the list_lru lock if already
626 * untracked. The list_empty() test is safe as
627 * node->private_list is protected by
628 * mapping->tree_lock.
630 if (!list_empty(&node
->private_list
))
631 list_lru_del(&workingset_shadow_nodes
,
632 &node
->private_list
);
637 static int __add_to_page_cache_locked(struct page
*page
,
638 struct address_space
*mapping
,
639 pgoff_t offset
, gfp_t gfp_mask
,
642 int huge
= PageHuge(page
);
643 struct mem_cgroup
*memcg
;
646 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
647 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
650 error
= mem_cgroup_try_charge(page
, current
->mm
,
651 gfp_mask
, &memcg
, false);
656 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
659 mem_cgroup_cancel_charge(page
, memcg
, false);
663 page_cache_get(page
);
664 page
->mapping
= mapping
;
665 page
->index
= offset
;
667 spin_lock_irq(&mapping
->tree_lock
);
668 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
669 radix_tree_preload_end();
673 /* hugetlb pages do not participate in page cache accounting. */
675 __inc_zone_page_state(page
, NR_FILE_PAGES
);
676 spin_unlock_irq(&mapping
->tree_lock
);
678 mem_cgroup_commit_charge(page
, memcg
, false, false);
679 trace_mm_filemap_add_to_page_cache(page
);
682 page
->mapping
= NULL
;
683 /* Leave page->index set: truncation relies upon it */
684 spin_unlock_irq(&mapping
->tree_lock
);
686 mem_cgroup_cancel_charge(page
, memcg
, false);
687 page_cache_release(page
);
692 * add_to_page_cache_locked - add a locked page to the pagecache
694 * @mapping: the page's address_space
695 * @offset: page index
696 * @gfp_mask: page allocation mode
698 * This function is used to add a page to the pagecache. It must be locked.
699 * This function does not add the page to the LRU. The caller must do that.
701 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
702 pgoff_t offset
, gfp_t gfp_mask
)
704 return __add_to_page_cache_locked(page
, mapping
, offset
,
707 EXPORT_SYMBOL(add_to_page_cache_locked
);
709 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
710 pgoff_t offset
, gfp_t gfp_mask
)
715 __SetPageLocked(page
);
716 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
719 __ClearPageLocked(page
);
722 * The page might have been evicted from cache only
723 * recently, in which case it should be activated like
724 * any other repeatedly accessed page.
726 if (shadow
&& workingset_refault(shadow
)) {
728 workingset_activation(page
);
730 ClearPageActive(page
);
735 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
738 struct page
*__page_cache_alloc(gfp_t gfp
)
743 if (cpuset_do_page_mem_spread()) {
744 unsigned int cpuset_mems_cookie
;
746 cpuset_mems_cookie
= read_mems_allowed_begin();
747 n
= cpuset_mem_spread_node();
748 page
= __alloc_pages_node(n
, gfp
, 0);
749 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
753 return alloc_pages(gfp
, 0);
755 EXPORT_SYMBOL(__page_cache_alloc
);
759 * In order to wait for pages to become available there must be
760 * waitqueues associated with pages. By using a hash table of
761 * waitqueues where the bucket discipline is to maintain all
762 * waiters on the same queue and wake all when any of the pages
763 * become available, and for the woken contexts to check to be
764 * sure the appropriate page became available, this saves space
765 * at a cost of "thundering herd" phenomena during rare hash
768 wait_queue_head_t
*page_waitqueue(struct page
*page
)
770 const struct zone
*zone
= page_zone(page
);
772 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
774 EXPORT_SYMBOL(page_waitqueue
);
776 void wait_on_page_bit(struct page
*page
, int bit_nr
)
778 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
780 if (test_bit(bit_nr
, &page
->flags
))
781 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
782 TASK_UNINTERRUPTIBLE
);
784 EXPORT_SYMBOL(wait_on_page_bit
);
786 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
788 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
790 if (!test_bit(bit_nr
, &page
->flags
))
793 return __wait_on_bit(page_waitqueue(page
), &wait
,
794 bit_wait_io
, TASK_KILLABLE
);
797 int wait_on_page_bit_killable_timeout(struct page
*page
,
798 int bit_nr
, unsigned long timeout
)
800 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
802 wait
.key
.timeout
= jiffies
+ timeout
;
803 if (!test_bit(bit_nr
, &page
->flags
))
805 return __wait_on_bit(page_waitqueue(page
), &wait
,
806 bit_wait_io_timeout
, TASK_KILLABLE
);
808 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
811 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
812 * @page: Page defining the wait queue of interest
813 * @waiter: Waiter to add to the queue
815 * Add an arbitrary @waiter to the wait queue for the nominated @page.
817 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
819 wait_queue_head_t
*q
= page_waitqueue(page
);
822 spin_lock_irqsave(&q
->lock
, flags
);
823 __add_wait_queue(q
, waiter
);
824 spin_unlock_irqrestore(&q
->lock
, flags
);
826 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
829 * unlock_page - unlock a locked page
832 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
833 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
834 * mechanism between PageLocked pages and PageWriteback pages is shared.
835 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
837 * The mb is necessary to enforce ordering between the clear_bit and the read
838 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
840 void unlock_page(struct page
*page
)
842 page
= compound_head(page
);
843 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
844 clear_bit_unlock(PG_locked
, &page
->flags
);
845 smp_mb__after_atomic();
846 wake_up_page(page
, PG_locked
);
848 EXPORT_SYMBOL(unlock_page
);
851 * end_page_writeback - end writeback against a page
854 void end_page_writeback(struct page
*page
)
857 * TestClearPageReclaim could be used here but it is an atomic
858 * operation and overkill in this particular case. Failing to
859 * shuffle a page marked for immediate reclaim is too mild to
860 * justify taking an atomic operation penalty at the end of
861 * ever page writeback.
863 if (PageReclaim(page
)) {
864 ClearPageReclaim(page
);
865 rotate_reclaimable_page(page
);
868 if (!test_clear_page_writeback(page
))
871 smp_mb__after_atomic();
872 wake_up_page(page
, PG_writeback
);
874 EXPORT_SYMBOL(end_page_writeback
);
877 * After completing I/O on a page, call this routine to update the page
878 * flags appropriately
880 void page_endio(struct page
*page
, int rw
, int err
)
884 SetPageUptodate(page
);
886 ClearPageUptodate(page
);
890 } else { /* rw == WRITE */
894 mapping_set_error(page
->mapping
, err
);
896 end_page_writeback(page
);
899 EXPORT_SYMBOL_GPL(page_endio
);
902 * __lock_page - get a lock on the page, assuming we need to sleep to get it
903 * @page: the page to lock
905 void __lock_page(struct page
*page
)
907 struct page
*page_head
= compound_head(page
);
908 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
910 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
911 TASK_UNINTERRUPTIBLE
);
913 EXPORT_SYMBOL(__lock_page
);
915 int __lock_page_killable(struct page
*page
)
917 struct page
*page_head
= compound_head(page
);
918 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
920 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
921 bit_wait_io
, TASK_KILLABLE
);
923 EXPORT_SYMBOL_GPL(__lock_page_killable
);
927 * 1 - page is locked; mmap_sem is still held.
928 * 0 - page is not locked.
929 * mmap_sem has been released (up_read()), unless flags had both
930 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
931 * which case mmap_sem is still held.
933 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
934 * with the page locked and the mmap_sem unperturbed.
936 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
939 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
941 * CAUTION! In this case, mmap_sem is not released
942 * even though return 0.
944 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
947 up_read(&mm
->mmap_sem
);
948 if (flags
& FAULT_FLAG_KILLABLE
)
949 wait_on_page_locked_killable(page
);
951 wait_on_page_locked(page
);
954 if (flags
& FAULT_FLAG_KILLABLE
) {
957 ret
= __lock_page_killable(page
);
959 up_read(&mm
->mmap_sem
);
969 * page_cache_next_hole - find the next hole (not-present entry)
972 * @max_scan: maximum range to search
974 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
975 * lowest indexed hole.
977 * Returns: the index of the hole if found, otherwise returns an index
978 * outside of the set specified (in which case 'return - index >=
979 * max_scan' will be true). In rare cases of index wrap-around, 0 will
982 * page_cache_next_hole may be called under rcu_read_lock. However,
983 * like radix_tree_gang_lookup, this will not atomically search a
984 * snapshot of the tree at a single point in time. For example, if a
985 * hole is created at index 5, then subsequently a hole is created at
986 * index 10, page_cache_next_hole covering both indexes may return 10
987 * if called under rcu_read_lock.
989 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
990 pgoff_t index
, unsigned long max_scan
)
994 for (i
= 0; i
< max_scan
; i
++) {
997 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
998 if (!page
|| radix_tree_exceptional_entry(page
))
1007 EXPORT_SYMBOL(page_cache_next_hole
);
1010 * page_cache_prev_hole - find the prev hole (not-present entry)
1013 * @max_scan: maximum range to search
1015 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1018 * Returns: the index of the hole if found, otherwise returns an index
1019 * outside of the set specified (in which case 'index - return >=
1020 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1023 * page_cache_prev_hole may be called under rcu_read_lock. However,
1024 * like radix_tree_gang_lookup, this will not atomically search a
1025 * snapshot of the tree at a single point in time. For example, if a
1026 * hole is created at index 10, then subsequently a hole is created at
1027 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1028 * called under rcu_read_lock.
1030 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1031 pgoff_t index
, unsigned long max_scan
)
1035 for (i
= 0; i
< max_scan
; i
++) {
1038 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1039 if (!page
|| radix_tree_exceptional_entry(page
))
1042 if (index
== ULONG_MAX
)
1048 EXPORT_SYMBOL(page_cache_prev_hole
);
1051 * find_get_entry - find and get a page cache entry
1052 * @mapping: the address_space to search
1053 * @offset: the page cache index
1055 * Looks up the page cache slot at @mapping & @offset. If there is a
1056 * page cache page, it is returned with an increased refcount.
1058 * If the slot holds a shadow entry of a previously evicted page, or a
1059 * swap entry from shmem/tmpfs, it is returned.
1061 * Otherwise, %NULL is returned.
1063 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1071 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1073 page
= radix_tree_deref_slot(pagep
);
1074 if (unlikely(!page
))
1076 if (radix_tree_exception(page
)) {
1077 if (radix_tree_deref_retry(page
))
1080 * A shadow entry of a recently evicted page,
1081 * or a swap entry from shmem/tmpfs. Return
1082 * it without attempting to raise page count.
1086 if (!page_cache_get_speculative(page
))
1090 * Has the page moved?
1091 * This is part of the lockless pagecache protocol. See
1092 * include/linux/pagemap.h for details.
1094 if (unlikely(page
!= *pagep
)) {
1095 page_cache_release(page
);
1104 EXPORT_SYMBOL(find_get_entry
);
1107 * find_lock_entry - locate, pin and lock a page cache entry
1108 * @mapping: the address_space to search
1109 * @offset: the page cache index
1111 * Looks up the page cache slot at @mapping & @offset. If there is a
1112 * page cache page, it is returned locked and with an increased
1115 * If the slot holds a shadow entry of a previously evicted page, or a
1116 * swap entry from shmem/tmpfs, it is returned.
1118 * Otherwise, %NULL is returned.
1120 * find_lock_entry() may sleep.
1122 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1127 page
= find_get_entry(mapping
, offset
);
1128 if (page
&& !radix_tree_exception(page
)) {
1130 /* Has the page been truncated? */
1131 if (unlikely(page
->mapping
!= mapping
)) {
1133 page_cache_release(page
);
1136 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1140 EXPORT_SYMBOL(find_lock_entry
);
1143 * pagecache_get_page - find and get a page reference
1144 * @mapping: the address_space to search
1145 * @offset: the page index
1146 * @fgp_flags: PCG flags
1147 * @gfp_mask: gfp mask to use for the page cache data page allocation
1149 * Looks up the page cache slot at @mapping & @offset.
1151 * PCG flags modify how the page is returned.
1153 * FGP_ACCESSED: the page will be marked accessed
1154 * FGP_LOCK: Page is return locked
1155 * FGP_CREAT: If page is not present then a new page is allocated using
1156 * @gfp_mask and added to the page cache and the VM's LRU
1157 * list. The page is returned locked and with an increased
1158 * refcount. Otherwise, %NULL is returned.
1160 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1161 * if the GFP flags specified for FGP_CREAT are atomic.
1163 * If there is a page cache page, it is returned with an increased refcount.
1165 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1166 int fgp_flags
, gfp_t gfp_mask
)
1171 page
= find_get_entry(mapping
, offset
);
1172 if (radix_tree_exceptional_entry(page
))
1177 if (fgp_flags
& FGP_LOCK
) {
1178 if (fgp_flags
& FGP_NOWAIT
) {
1179 if (!trylock_page(page
)) {
1180 page_cache_release(page
);
1187 /* Has the page been truncated? */
1188 if (unlikely(page
->mapping
!= mapping
)) {
1190 page_cache_release(page
);
1193 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1196 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1197 mark_page_accessed(page
);
1200 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1202 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1203 gfp_mask
|= __GFP_WRITE
;
1204 if (fgp_flags
& FGP_NOFS
)
1205 gfp_mask
&= ~__GFP_FS
;
1207 page
= __page_cache_alloc(gfp_mask
);
1211 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1212 fgp_flags
|= FGP_LOCK
;
1214 /* Init accessed so avoid atomic mark_page_accessed later */
1215 if (fgp_flags
& FGP_ACCESSED
)
1216 __SetPageReferenced(page
);
1218 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1219 gfp_mask
& GFP_RECLAIM_MASK
);
1220 if (unlikely(err
)) {
1221 page_cache_release(page
);
1230 EXPORT_SYMBOL(pagecache_get_page
);
1233 * find_get_entries - gang pagecache lookup
1234 * @mapping: The address_space to search
1235 * @start: The starting page cache index
1236 * @nr_entries: The maximum number of entries
1237 * @entries: Where the resulting entries are placed
1238 * @indices: The cache indices corresponding to the entries in @entries
1240 * find_get_entries() will search for and return a group of up to
1241 * @nr_entries entries in the mapping. The entries are placed at
1242 * @entries. find_get_entries() takes a reference against any actual
1245 * The search returns a group of mapping-contiguous page cache entries
1246 * with ascending indexes. There may be holes in the indices due to
1247 * not-present pages.
1249 * Any shadow entries of evicted pages, or swap entries from
1250 * shmem/tmpfs, are included in the returned array.
1252 * find_get_entries() returns the number of pages and shadow entries
1255 unsigned find_get_entries(struct address_space
*mapping
,
1256 pgoff_t start
, unsigned int nr_entries
,
1257 struct page
**entries
, pgoff_t
*indices
)
1260 unsigned int ret
= 0;
1261 struct radix_tree_iter iter
;
1268 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1271 page
= radix_tree_deref_slot(slot
);
1272 if (unlikely(!page
))
1274 if (radix_tree_exception(page
)) {
1275 if (radix_tree_deref_retry(page
))
1278 * A shadow entry of a recently evicted page, a swap
1279 * entry from shmem/tmpfs or a DAX entry. Return it
1280 * without attempting to raise page count.
1284 if (!page_cache_get_speculative(page
))
1287 /* Has the page moved? */
1288 if (unlikely(page
!= *slot
)) {
1289 page_cache_release(page
);
1293 indices
[ret
] = iter
.index
;
1294 entries
[ret
] = page
;
1295 if (++ret
== nr_entries
)
1303 * find_get_pages - gang pagecache lookup
1304 * @mapping: The address_space to search
1305 * @start: The starting page index
1306 * @nr_pages: The maximum number of pages
1307 * @pages: Where the resulting pages are placed
1309 * find_get_pages() will search for and return a group of up to
1310 * @nr_pages pages in the mapping. The pages are placed at @pages.
1311 * find_get_pages() takes a reference against the returned pages.
1313 * The search returns a group of mapping-contiguous pages with ascending
1314 * indexes. There may be holes in the indices due to not-present pages.
1316 * find_get_pages() returns the number of pages which were found.
1318 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1319 unsigned int nr_pages
, struct page
**pages
)
1321 struct radix_tree_iter iter
;
1325 if (unlikely(!nr_pages
))
1330 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1333 page
= radix_tree_deref_slot(slot
);
1334 if (unlikely(!page
))
1337 if (radix_tree_exception(page
)) {
1338 if (radix_tree_deref_retry(page
)) {
1340 * Transient condition which can only trigger
1341 * when entry at index 0 moves out of or back
1342 * to root: none yet gotten, safe to restart.
1344 WARN_ON(iter
.index
);
1348 * A shadow entry of a recently evicted page,
1349 * or a swap entry from shmem/tmpfs. Skip
1355 if (!page_cache_get_speculative(page
))
1358 /* Has the page moved? */
1359 if (unlikely(page
!= *slot
)) {
1360 page_cache_release(page
);
1365 if (++ret
== nr_pages
)
1374 * find_get_pages_contig - gang contiguous pagecache lookup
1375 * @mapping: The address_space to search
1376 * @index: The starting page index
1377 * @nr_pages: The maximum number of pages
1378 * @pages: Where the resulting pages are placed
1380 * find_get_pages_contig() works exactly like find_get_pages(), except
1381 * that the returned number of pages are guaranteed to be contiguous.
1383 * find_get_pages_contig() returns the number of pages which were found.
1385 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1386 unsigned int nr_pages
, struct page
**pages
)
1388 struct radix_tree_iter iter
;
1390 unsigned int ret
= 0;
1392 if (unlikely(!nr_pages
))
1397 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1400 page
= radix_tree_deref_slot(slot
);
1401 /* The hole, there no reason to continue */
1402 if (unlikely(!page
))
1405 if (radix_tree_exception(page
)) {
1406 if (radix_tree_deref_retry(page
)) {
1408 * Transient condition which can only trigger
1409 * when entry at index 0 moves out of or back
1410 * to root: none yet gotten, safe to restart.
1415 * A shadow entry of a recently evicted page,
1416 * or a swap entry from shmem/tmpfs. Stop
1417 * looking for contiguous pages.
1422 if (!page_cache_get_speculative(page
))
1425 /* Has the page moved? */
1426 if (unlikely(page
!= *slot
)) {
1427 page_cache_release(page
);
1432 * must check mapping and index after taking the ref.
1433 * otherwise we can get both false positives and false
1434 * negatives, which is just confusing to the caller.
1436 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1437 page_cache_release(page
);
1442 if (++ret
== nr_pages
)
1448 EXPORT_SYMBOL(find_get_pages_contig
);
1451 * find_get_pages_tag - find and return pages that match @tag
1452 * @mapping: the address_space to search
1453 * @index: the starting page index
1454 * @tag: the tag index
1455 * @nr_pages: the maximum number of pages
1456 * @pages: where the resulting pages are placed
1458 * Like find_get_pages, except we only return pages which are tagged with
1459 * @tag. We update @index to index the next page for the traversal.
1461 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1462 int tag
, unsigned int nr_pages
, struct page
**pages
)
1464 struct radix_tree_iter iter
;
1468 if (unlikely(!nr_pages
))
1473 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1474 &iter
, *index
, tag
) {
1477 page
= radix_tree_deref_slot(slot
);
1478 if (unlikely(!page
))
1481 if (radix_tree_exception(page
)) {
1482 if (radix_tree_deref_retry(page
)) {
1484 * Transient condition which can only trigger
1485 * when entry at index 0 moves out of or back
1486 * to root: none yet gotten, safe to restart.
1491 * A shadow entry of a recently evicted page.
1493 * Those entries should never be tagged, but
1494 * this tree walk is lockless and the tags are
1495 * looked up in bulk, one radix tree node at a
1496 * time, so there is a sizable window for page
1497 * reclaim to evict a page we saw tagged.
1504 if (!page_cache_get_speculative(page
))
1507 /* Has the page moved? */
1508 if (unlikely(page
!= *slot
)) {
1509 page_cache_release(page
);
1514 if (++ret
== nr_pages
)
1521 *index
= pages
[ret
- 1]->index
+ 1;
1525 EXPORT_SYMBOL(find_get_pages_tag
);
1528 * find_get_entries_tag - find and return entries that match @tag
1529 * @mapping: the address_space to search
1530 * @start: the starting page cache index
1531 * @tag: the tag index
1532 * @nr_entries: the maximum number of entries
1533 * @entries: where the resulting entries are placed
1534 * @indices: the cache indices corresponding to the entries in @entries
1536 * Like find_get_entries, except we only return entries which are tagged with
1539 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1540 int tag
, unsigned int nr_entries
,
1541 struct page
**entries
, pgoff_t
*indices
)
1544 unsigned int ret
= 0;
1545 struct radix_tree_iter iter
;
1552 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1553 &iter
, start
, tag
) {
1556 page
= radix_tree_deref_slot(slot
);
1557 if (unlikely(!page
))
1559 if (radix_tree_exception(page
)) {
1560 if (radix_tree_deref_retry(page
)) {
1562 * Transient condition which can only trigger
1563 * when entry at index 0 moves out of or back
1564 * to root: none yet gotten, safe to restart.
1570 * A shadow entry of a recently evicted page, a swap
1571 * entry from shmem/tmpfs or a DAX entry. Return it
1572 * without attempting to raise page count.
1576 if (!page_cache_get_speculative(page
))
1579 /* Has the page moved? */
1580 if (unlikely(page
!= *slot
)) {
1581 page_cache_release(page
);
1585 indices
[ret
] = iter
.index
;
1586 entries
[ret
] = page
;
1587 if (++ret
== nr_entries
)
1593 EXPORT_SYMBOL(find_get_entries_tag
);
1596 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1597 * a _large_ part of the i/o request. Imagine the worst scenario:
1599 * ---R__________________________________________B__________
1600 * ^ reading here ^ bad block(assume 4k)
1602 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1603 * => failing the whole request => read(R) => read(R+1) =>
1604 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1605 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1606 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1608 * It is going insane. Fix it by quickly scaling down the readahead size.
1610 static void shrink_readahead_size_eio(struct file
*filp
,
1611 struct file_ra_state
*ra
)
1617 * do_generic_file_read - generic file read routine
1618 * @filp: the file to read
1619 * @ppos: current file position
1620 * @iter: data destination
1621 * @written: already copied
1623 * This is a generic file read routine, and uses the
1624 * mapping->a_ops->readpage() function for the actual low-level stuff.
1626 * This is really ugly. But the goto's actually try to clarify some
1627 * of the logic when it comes to error handling etc.
1629 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1630 struct iov_iter
*iter
, ssize_t written
)
1632 struct address_space
*mapping
= filp
->f_mapping
;
1633 struct inode
*inode
= mapping
->host
;
1634 struct file_ra_state
*ra
= &filp
->f_ra
;
1638 unsigned long offset
; /* offset into pagecache page */
1639 unsigned int prev_offset
;
1642 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1643 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1644 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1645 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1646 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1652 unsigned long nr
, ret
;
1656 page
= find_get_page(mapping
, index
);
1658 page_cache_sync_readahead(mapping
,
1660 index
, last_index
- index
);
1661 page
= find_get_page(mapping
, index
);
1662 if (unlikely(page
== NULL
))
1663 goto no_cached_page
;
1665 if (PageReadahead(page
)) {
1666 page_cache_async_readahead(mapping
,
1668 index
, last_index
- index
);
1670 if (!PageUptodate(page
)) {
1672 * See comment in do_read_cache_page on why
1673 * wait_on_page_locked is used to avoid unnecessarily
1674 * serialisations and why it's safe.
1676 wait_on_page_locked_killable(page
);
1677 if (PageUptodate(page
))
1680 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1681 !mapping
->a_ops
->is_partially_uptodate
)
1682 goto page_not_up_to_date
;
1683 if (!trylock_page(page
))
1684 goto page_not_up_to_date
;
1685 /* Did it get truncated before we got the lock? */
1687 goto page_not_up_to_date_locked
;
1688 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1689 offset
, iter
->count
))
1690 goto page_not_up_to_date_locked
;
1695 * i_size must be checked after we know the page is Uptodate.
1697 * Checking i_size after the check allows us to calculate
1698 * the correct value for "nr", which means the zero-filled
1699 * part of the page is not copied back to userspace (unless
1700 * another truncate extends the file - this is desired though).
1703 isize
= i_size_read(inode
);
1704 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1705 if (unlikely(!isize
|| index
> end_index
)) {
1706 page_cache_release(page
);
1710 /* nr is the maximum number of bytes to copy from this page */
1711 nr
= PAGE_CACHE_SIZE
;
1712 if (index
== end_index
) {
1713 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1715 page_cache_release(page
);
1721 /* If users can be writing to this page using arbitrary
1722 * virtual addresses, take care about potential aliasing
1723 * before reading the page on the kernel side.
1725 if (mapping_writably_mapped(mapping
))
1726 flush_dcache_page(page
);
1729 * When a sequential read accesses a page several times,
1730 * only mark it as accessed the first time.
1732 if (prev_index
!= index
|| offset
!= prev_offset
)
1733 mark_page_accessed(page
);
1737 * Ok, we have the page, and it's up-to-date, so
1738 * now we can copy it to user space...
1741 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1743 index
+= offset
>> PAGE_CACHE_SHIFT
;
1744 offset
&= ~PAGE_CACHE_MASK
;
1745 prev_offset
= offset
;
1747 page_cache_release(page
);
1749 if (!iov_iter_count(iter
))
1757 page_not_up_to_date
:
1758 /* Get exclusive access to the page ... */
1759 error
= lock_page_killable(page
);
1760 if (unlikely(error
))
1761 goto readpage_error
;
1763 page_not_up_to_date_locked
:
1764 /* Did it get truncated before we got the lock? */
1765 if (!page
->mapping
) {
1767 page_cache_release(page
);
1771 /* Did somebody else fill it already? */
1772 if (PageUptodate(page
)) {
1779 * A previous I/O error may have been due to temporary
1780 * failures, eg. multipath errors.
1781 * PG_error will be set again if readpage fails.
1783 ClearPageError(page
);
1784 /* Start the actual read. The read will unlock the page. */
1785 error
= mapping
->a_ops
->readpage(filp
, page
);
1787 if (unlikely(error
)) {
1788 if (error
== AOP_TRUNCATED_PAGE
) {
1789 page_cache_release(page
);
1793 goto readpage_error
;
1796 if (!PageUptodate(page
)) {
1797 error
= lock_page_killable(page
);
1798 if (unlikely(error
))
1799 goto readpage_error
;
1800 if (!PageUptodate(page
)) {
1801 if (page
->mapping
== NULL
) {
1803 * invalidate_mapping_pages got it
1806 page_cache_release(page
);
1810 shrink_readahead_size_eio(filp
, ra
);
1812 goto readpage_error
;
1820 /* UHHUH! A synchronous read error occurred. Report it */
1821 page_cache_release(page
);
1826 * Ok, it wasn't cached, so we need to create a new
1829 page
= page_cache_alloc_cold(mapping
);
1834 error
= add_to_page_cache_lru(page
, mapping
, index
,
1835 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1837 page_cache_release(page
);
1838 if (error
== -EEXIST
) {
1848 ra
->prev_pos
= prev_index
;
1849 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1850 ra
->prev_pos
|= prev_offset
;
1852 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1853 file_accessed(filp
);
1854 return written
? written
: error
;
1858 * generic_file_read_iter - generic filesystem read routine
1859 * @iocb: kernel I/O control block
1860 * @iter: destination for the data read
1862 * This is the "read_iter()" routine for all filesystems
1863 * that can use the page cache directly.
1866 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1868 struct file
*file
= iocb
->ki_filp
;
1870 loff_t
*ppos
= &iocb
->ki_pos
;
1873 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1874 struct address_space
*mapping
= file
->f_mapping
;
1875 struct inode
*inode
= mapping
->host
;
1876 size_t count
= iov_iter_count(iter
);
1880 goto out
; /* skip atime */
1881 size
= i_size_read(inode
);
1882 retval
= filemap_write_and_wait_range(mapping
, pos
,
1885 struct iov_iter data
= *iter
;
1886 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
1890 *ppos
= pos
+ retval
;
1891 iov_iter_advance(iter
, retval
);
1895 * Btrfs can have a short DIO read if we encounter
1896 * compressed extents, so if there was an error, or if
1897 * we've already read everything we wanted to, or if
1898 * there was a short read because we hit EOF, go ahead
1899 * and return. Otherwise fallthrough to buffered io for
1900 * the rest of the read. Buffered reads will not work for
1901 * DAX files, so don't bother trying.
1903 if (retval
< 0 || !iov_iter_count(iter
) || *ppos
>= size
||
1905 file_accessed(file
);
1910 retval
= do_generic_file_read(file
, ppos
, iter
, retval
);
1914 EXPORT_SYMBOL(generic_file_read_iter
);
1918 * page_cache_read - adds requested page to the page cache if not already there
1919 * @file: file to read
1920 * @offset: page index
1921 * @gfp_mask: memory allocation flags
1923 * This adds the requested page to the page cache if it isn't already there,
1924 * and schedules an I/O to read in its contents from disk.
1926 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1928 struct address_space
*mapping
= file
->f_mapping
;
1933 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1937 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1939 ret
= mapping
->a_ops
->readpage(file
, page
);
1940 else if (ret
== -EEXIST
)
1941 ret
= 0; /* losing race to add is OK */
1943 page_cache_release(page
);
1945 } while (ret
== AOP_TRUNCATED_PAGE
);
1950 #define MMAP_LOTSAMISS (100)
1953 * Synchronous readahead happens when we don't even find
1954 * a page in the page cache at all.
1956 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1957 struct file_ra_state
*ra
,
1961 struct address_space
*mapping
= file
->f_mapping
;
1963 /* If we don't want any read-ahead, don't bother */
1964 if (vma
->vm_flags
& VM_RAND_READ
)
1969 if (vma
->vm_flags
& VM_SEQ_READ
) {
1970 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1975 /* Avoid banging the cache line if not needed */
1976 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1980 * Do we miss much more than hit in this file? If so,
1981 * stop bothering with read-ahead. It will only hurt.
1983 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1989 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
1990 ra
->size
= ra
->ra_pages
;
1991 ra
->async_size
= ra
->ra_pages
/ 4;
1992 ra_submit(ra
, mapping
, file
);
1996 * Asynchronous readahead happens when we find the page and PG_readahead,
1997 * so we want to possibly extend the readahead further..
1999 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2000 struct file_ra_state
*ra
,
2005 struct address_space
*mapping
= file
->f_mapping
;
2007 /* If we don't want any read-ahead, don't bother */
2008 if (vma
->vm_flags
& VM_RAND_READ
)
2010 if (ra
->mmap_miss
> 0)
2012 if (PageReadahead(page
))
2013 page_cache_async_readahead(mapping
, ra
, file
,
2014 page
, offset
, ra
->ra_pages
);
2018 * filemap_fault - read in file data for page fault handling
2019 * @vma: vma in which the fault was taken
2020 * @vmf: struct vm_fault containing details of the fault
2022 * filemap_fault() is invoked via the vma operations vector for a
2023 * mapped memory region to read in file data during a page fault.
2025 * The goto's are kind of ugly, but this streamlines the normal case of having
2026 * it in the page cache, and handles the special cases reasonably without
2027 * having a lot of duplicated code.
2029 * vma->vm_mm->mmap_sem must be held on entry.
2031 * If our return value has VM_FAULT_RETRY set, it's because
2032 * lock_page_or_retry() returned 0.
2033 * The mmap_sem has usually been released in this case.
2034 * See __lock_page_or_retry() for the exception.
2036 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2037 * has not been released.
2039 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2041 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2044 struct file
*file
= vma
->vm_file
;
2045 struct address_space
*mapping
= file
->f_mapping
;
2046 struct file_ra_state
*ra
= &file
->f_ra
;
2047 struct inode
*inode
= mapping
->host
;
2048 pgoff_t offset
= vmf
->pgoff
;
2053 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2054 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
2055 return VM_FAULT_SIGBUS
;
2058 * Do we have something in the page cache already?
2060 page
= find_get_page(mapping
, offset
);
2061 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2063 * We found the page, so try async readahead before
2064 * waiting for the lock.
2066 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2068 /* No page in the page cache at all */
2069 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2070 count_vm_event(PGMAJFAULT
);
2071 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2072 ret
= VM_FAULT_MAJOR
;
2074 page
= find_get_page(mapping
, offset
);
2076 goto no_cached_page
;
2079 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2080 page_cache_release(page
);
2081 return ret
| VM_FAULT_RETRY
;
2084 /* Did it get truncated? */
2085 if (unlikely(page
->mapping
!= mapping
)) {
2090 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2093 * We have a locked page in the page cache, now we need to check
2094 * that it's up-to-date. If not, it is going to be due to an error.
2096 if (unlikely(!PageUptodate(page
)))
2097 goto page_not_uptodate
;
2100 * Found the page and have a reference on it.
2101 * We must recheck i_size under page lock.
2103 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2104 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
2106 page_cache_release(page
);
2107 return VM_FAULT_SIGBUS
;
2111 return ret
| VM_FAULT_LOCKED
;
2115 * We're only likely to ever get here if MADV_RANDOM is in
2118 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2121 * The page we want has now been added to the page cache.
2122 * In the unlikely event that someone removed it in the
2123 * meantime, we'll just come back here and read it again.
2129 * An error return from page_cache_read can result if the
2130 * system is low on memory, or a problem occurs while trying
2133 if (error
== -ENOMEM
)
2134 return VM_FAULT_OOM
;
2135 return VM_FAULT_SIGBUS
;
2139 * Umm, take care of errors if the page isn't up-to-date.
2140 * Try to re-read it _once_. We do this synchronously,
2141 * because there really aren't any performance issues here
2142 * and we need to check for errors.
2144 ClearPageError(page
);
2145 error
= mapping
->a_ops
->readpage(file
, page
);
2147 wait_on_page_locked(page
);
2148 if (!PageUptodate(page
))
2151 page_cache_release(page
);
2153 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2156 /* Things didn't work out. Return zero to tell the mm layer so. */
2157 shrink_readahead_size_eio(file
, ra
);
2158 return VM_FAULT_SIGBUS
;
2160 EXPORT_SYMBOL(filemap_fault
);
2162 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2164 struct radix_tree_iter iter
;
2166 struct file
*file
= vma
->vm_file
;
2167 struct address_space
*mapping
= file
->f_mapping
;
2170 unsigned long address
= (unsigned long) vmf
->virtual_address
;
2175 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
2176 if (iter
.index
> vmf
->max_pgoff
)
2179 page
= radix_tree_deref_slot(slot
);
2180 if (unlikely(!page
))
2182 if (radix_tree_exception(page
)) {
2183 if (radix_tree_deref_retry(page
))
2189 if (!page_cache_get_speculative(page
))
2192 /* Has the page moved? */
2193 if (unlikely(page
!= *slot
)) {
2194 page_cache_release(page
);
2198 if (!PageUptodate(page
) ||
2199 PageReadahead(page
) ||
2202 if (!trylock_page(page
))
2205 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2208 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
2209 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2212 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2213 if (!pte_none(*pte
))
2216 if (file
->f_ra
.mmap_miss
> 0)
2217 file
->f_ra
.mmap_miss
--;
2218 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2219 do_set_pte(vma
, addr
, page
, pte
, false, false);
2225 page_cache_release(page
);
2227 if (iter
.index
== vmf
->max_pgoff
)
2232 EXPORT_SYMBOL(filemap_map_pages
);
2234 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2236 struct page
*page
= vmf
->page
;
2237 struct inode
*inode
= file_inode(vma
->vm_file
);
2238 int ret
= VM_FAULT_LOCKED
;
2240 sb_start_pagefault(inode
->i_sb
);
2241 file_update_time(vma
->vm_file
);
2243 if (page
->mapping
!= inode
->i_mapping
) {
2245 ret
= VM_FAULT_NOPAGE
;
2249 * We mark the page dirty already here so that when freeze is in
2250 * progress, we are guaranteed that writeback during freezing will
2251 * see the dirty page and writeprotect it again.
2253 set_page_dirty(page
);
2254 wait_for_stable_page(page
);
2256 sb_end_pagefault(inode
->i_sb
);
2259 EXPORT_SYMBOL(filemap_page_mkwrite
);
2261 const struct vm_operations_struct generic_file_vm_ops
= {
2262 .fault
= filemap_fault
,
2263 .map_pages
= filemap_map_pages
,
2264 .page_mkwrite
= filemap_page_mkwrite
,
2267 /* This is used for a general mmap of a disk file */
2269 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2271 struct address_space
*mapping
= file
->f_mapping
;
2273 if (!mapping
->a_ops
->readpage
)
2275 file_accessed(file
);
2276 vma
->vm_ops
= &generic_file_vm_ops
;
2281 * This is for filesystems which do not implement ->writepage.
2283 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2285 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2287 return generic_file_mmap(file
, vma
);
2290 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2294 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2298 #endif /* CONFIG_MMU */
2300 EXPORT_SYMBOL(generic_file_mmap
);
2301 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2303 static struct page
*wait_on_page_read(struct page
*page
)
2305 if (!IS_ERR(page
)) {
2306 wait_on_page_locked(page
);
2307 if (!PageUptodate(page
)) {
2308 page_cache_release(page
);
2309 page
= ERR_PTR(-EIO
);
2315 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2317 int (*filler
)(void *, struct page
*),
2324 page
= find_get_page(mapping
, index
);
2326 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2328 return ERR_PTR(-ENOMEM
);
2329 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2330 if (unlikely(err
)) {
2331 page_cache_release(page
);
2334 /* Presumably ENOMEM for radix tree node */
2335 return ERR_PTR(err
);
2339 err
= filler(data
, page
);
2341 page_cache_release(page
);
2342 return ERR_PTR(err
);
2345 page
= wait_on_page_read(page
);
2350 if (PageUptodate(page
))
2354 * Page is not up to date and may be locked due one of the following
2355 * case a: Page is being filled and the page lock is held
2356 * case b: Read/write error clearing the page uptodate status
2357 * case c: Truncation in progress (page locked)
2358 * case d: Reclaim in progress
2360 * Case a, the page will be up to date when the page is unlocked.
2361 * There is no need to serialise on the page lock here as the page
2362 * is pinned so the lock gives no additional protection. Even if the
2363 * the page is truncated, the data is still valid if PageUptodate as
2364 * it's a race vs truncate race.
2365 * Case b, the page will not be up to date
2366 * Case c, the page may be truncated but in itself, the data may still
2367 * be valid after IO completes as it's a read vs truncate race. The
2368 * operation must restart if the page is not uptodate on unlock but
2369 * otherwise serialising on page lock to stabilise the mapping gives
2370 * no additional guarantees to the caller as the page lock is
2371 * released before return.
2372 * Case d, similar to truncation. If reclaim holds the page lock, it
2373 * will be a race with remove_mapping that determines if the mapping
2374 * is valid on unlock but otherwise the data is valid and there is
2375 * no need to serialise with page lock.
2377 * As the page lock gives no additional guarantee, we optimistically
2378 * wait on the page to be unlocked and check if it's up to date and
2379 * use the page if it is. Otherwise, the page lock is required to
2380 * distinguish between the different cases. The motivation is that we
2381 * avoid spurious serialisations and wakeups when multiple processes
2382 * wait on the same page for IO to complete.
2384 wait_on_page_locked(page
);
2385 if (PageUptodate(page
))
2388 /* Distinguish between all the cases under the safety of the lock */
2391 /* Case c or d, restart the operation */
2392 if (!page
->mapping
) {
2394 page_cache_release(page
);
2398 /* Someone else locked and filled the page in a very small window */
2399 if (PageUptodate(page
)) {
2406 mark_page_accessed(page
);
2411 * read_cache_page - read into page cache, fill it if needed
2412 * @mapping: the page's address_space
2413 * @index: the page index
2414 * @filler: function to perform the read
2415 * @data: first arg to filler(data, page) function, often left as NULL
2417 * Read into the page cache. If a page already exists, and PageUptodate() is
2418 * not set, try to fill the page and wait for it to become unlocked.
2420 * If the page does not get brought uptodate, return -EIO.
2422 struct page
*read_cache_page(struct address_space
*mapping
,
2424 int (*filler
)(void *, struct page
*),
2427 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2429 EXPORT_SYMBOL(read_cache_page
);
2432 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2433 * @mapping: the page's address_space
2434 * @index: the page index
2435 * @gfp: the page allocator flags to use if allocating
2437 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2438 * any new page allocations done using the specified allocation flags.
2440 * If the page does not get brought uptodate, return -EIO.
2442 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2446 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2448 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2450 EXPORT_SYMBOL(read_cache_page_gfp
);
2453 * Performs necessary checks before doing a write
2455 * Can adjust writing position or amount of bytes to write.
2456 * Returns appropriate error code that caller should return or
2457 * zero in case that write should be allowed.
2459 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2461 struct file
*file
= iocb
->ki_filp
;
2462 struct inode
*inode
= file
->f_mapping
->host
;
2463 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2466 if (!iov_iter_count(from
))
2469 /* FIXME: this is for backwards compatibility with 2.4 */
2470 if (iocb
->ki_flags
& IOCB_APPEND
)
2471 iocb
->ki_pos
= i_size_read(inode
);
2475 if (limit
!= RLIM_INFINITY
) {
2476 if (iocb
->ki_pos
>= limit
) {
2477 send_sig(SIGXFSZ
, current
, 0);
2480 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2486 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2487 !(file
->f_flags
& O_LARGEFILE
))) {
2488 if (pos
>= MAX_NON_LFS
)
2490 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2494 * Are we about to exceed the fs block limit ?
2496 * If we have written data it becomes a short write. If we have
2497 * exceeded without writing data we send a signal and return EFBIG.
2498 * Linus frestrict idea will clean these up nicely..
2500 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2503 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2504 return iov_iter_count(from
);
2506 EXPORT_SYMBOL(generic_write_checks
);
2508 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2509 loff_t pos
, unsigned len
, unsigned flags
,
2510 struct page
**pagep
, void **fsdata
)
2512 const struct address_space_operations
*aops
= mapping
->a_ops
;
2514 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2517 EXPORT_SYMBOL(pagecache_write_begin
);
2519 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2520 loff_t pos
, unsigned len
, unsigned copied
,
2521 struct page
*page
, void *fsdata
)
2523 const struct address_space_operations
*aops
= mapping
->a_ops
;
2525 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2527 EXPORT_SYMBOL(pagecache_write_end
);
2530 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
2532 struct file
*file
= iocb
->ki_filp
;
2533 struct address_space
*mapping
= file
->f_mapping
;
2534 struct inode
*inode
= mapping
->host
;
2538 struct iov_iter data
;
2540 write_len
= iov_iter_count(from
);
2541 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2543 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2548 * After a write we want buffered reads to be sure to go to disk to get
2549 * the new data. We invalidate clean cached page from the region we're
2550 * about to write. We do this *before* the write so that we can return
2551 * without clobbering -EIOCBQUEUED from ->direct_IO().
2553 if (mapping
->nrpages
) {
2554 written
= invalidate_inode_pages2_range(mapping
,
2555 pos
>> PAGE_CACHE_SHIFT
, end
);
2557 * If a page can not be invalidated, return 0 to fall back
2558 * to buffered write.
2561 if (written
== -EBUSY
)
2568 written
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
2571 * Finally, try again to invalidate clean pages which might have been
2572 * cached by non-direct readahead, or faulted in by get_user_pages()
2573 * if the source of the write was an mmap'ed region of the file
2574 * we're writing. Either one is a pretty crazy thing to do,
2575 * so we don't support it 100%. If this invalidation
2576 * fails, tough, the write still worked...
2578 if (mapping
->nrpages
) {
2579 invalidate_inode_pages2_range(mapping
,
2580 pos
>> PAGE_CACHE_SHIFT
, end
);
2585 iov_iter_advance(from
, written
);
2586 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2587 i_size_write(inode
, pos
);
2588 mark_inode_dirty(inode
);
2595 EXPORT_SYMBOL(generic_file_direct_write
);
2598 * Find or create a page at the given pagecache position. Return the locked
2599 * page. This function is specifically for buffered writes.
2601 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2602 pgoff_t index
, unsigned flags
)
2605 int fgp_flags
= FGP_LOCK
|FGP_ACCESSED
|FGP_WRITE
|FGP_CREAT
;
2607 if (flags
& AOP_FLAG_NOFS
)
2608 fgp_flags
|= FGP_NOFS
;
2610 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2611 mapping_gfp_mask(mapping
));
2613 wait_for_stable_page(page
);
2617 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2619 ssize_t
generic_perform_write(struct file
*file
,
2620 struct iov_iter
*i
, loff_t pos
)
2622 struct address_space
*mapping
= file
->f_mapping
;
2623 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2625 ssize_t written
= 0;
2626 unsigned int flags
= 0;
2629 * Copies from kernel address space cannot fail (NFSD is a big user).
2631 if (!iter_is_iovec(i
))
2632 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2636 unsigned long offset
; /* Offset into pagecache page */
2637 unsigned long bytes
; /* Bytes to write to page */
2638 size_t copied
; /* Bytes copied from user */
2641 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2642 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2647 * Bring in the user page that we will copy from _first_.
2648 * Otherwise there's a nasty deadlock on copying from the
2649 * same page as we're writing to, without it being marked
2652 * Not only is this an optimisation, but it is also required
2653 * to check that the address is actually valid, when atomic
2654 * usercopies are used, below.
2656 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2661 if (fatal_signal_pending(current
)) {
2666 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2668 if (unlikely(status
< 0))
2671 if (mapping_writably_mapped(mapping
))
2672 flush_dcache_page(page
);
2674 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2675 flush_dcache_page(page
);
2677 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2679 if (unlikely(status
< 0))
2685 iov_iter_advance(i
, copied
);
2686 if (unlikely(copied
== 0)) {
2688 * If we were unable to copy any data at all, we must
2689 * fall back to a single segment length write.
2691 * If we didn't fallback here, we could livelock
2692 * because not all segments in the iov can be copied at
2693 * once without a pagefault.
2695 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2696 iov_iter_single_seg_count(i
));
2702 balance_dirty_pages_ratelimited(mapping
);
2703 } while (iov_iter_count(i
));
2705 return written
? written
: status
;
2707 EXPORT_SYMBOL(generic_perform_write
);
2710 * __generic_file_write_iter - write data to a file
2711 * @iocb: IO state structure (file, offset, etc.)
2712 * @from: iov_iter with data to write
2714 * This function does all the work needed for actually writing data to a
2715 * file. It does all basic checks, removes SUID from the file, updates
2716 * modification times and calls proper subroutines depending on whether we
2717 * do direct IO or a standard buffered write.
2719 * It expects i_mutex to be grabbed unless we work on a block device or similar
2720 * object which does not need locking at all.
2722 * This function does *not* take care of syncing data in case of O_SYNC write.
2723 * A caller has to handle it. This is mainly due to the fact that we want to
2724 * avoid syncing under i_mutex.
2726 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2728 struct file
*file
= iocb
->ki_filp
;
2729 struct address_space
* mapping
= file
->f_mapping
;
2730 struct inode
*inode
= mapping
->host
;
2731 ssize_t written
= 0;
2735 /* We can write back this queue in page reclaim */
2736 current
->backing_dev_info
= inode_to_bdi(inode
);
2737 err
= file_remove_privs(file
);
2741 err
= file_update_time(file
);
2745 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2746 loff_t pos
, endbyte
;
2748 written
= generic_file_direct_write(iocb
, from
, iocb
->ki_pos
);
2750 * If the write stopped short of completing, fall back to
2751 * buffered writes. Some filesystems do this for writes to
2752 * holes, for example. For DAX files, a buffered write will
2753 * not succeed (even if it did, DAX does not handle dirty
2754 * page-cache pages correctly).
2756 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2759 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2761 * If generic_perform_write() returned a synchronous error
2762 * then we want to return the number of bytes which were
2763 * direct-written, or the error code if that was zero. Note
2764 * that this differs from normal direct-io semantics, which
2765 * will return -EFOO even if some bytes were written.
2767 if (unlikely(status
< 0)) {
2772 * We need to ensure that the page cache pages are written to
2773 * disk and invalidated to preserve the expected O_DIRECT
2776 endbyte
= pos
+ status
- 1;
2777 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2779 iocb
->ki_pos
= endbyte
+ 1;
2781 invalidate_mapping_pages(mapping
,
2782 pos
>> PAGE_CACHE_SHIFT
,
2783 endbyte
>> PAGE_CACHE_SHIFT
);
2786 * We don't know how much we wrote, so just return
2787 * the number of bytes which were direct-written
2791 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2792 if (likely(written
> 0))
2793 iocb
->ki_pos
+= written
;
2796 current
->backing_dev_info
= NULL
;
2797 return written
? written
: err
;
2799 EXPORT_SYMBOL(__generic_file_write_iter
);
2802 * generic_file_write_iter - write data to a file
2803 * @iocb: IO state structure
2804 * @from: iov_iter with data to write
2806 * This is a wrapper around __generic_file_write_iter() to be used by most
2807 * filesystems. It takes care of syncing the file in case of O_SYNC file
2808 * and acquires i_mutex as needed.
2810 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2812 struct file
*file
= iocb
->ki_filp
;
2813 struct inode
*inode
= file
->f_mapping
->host
;
2817 ret
= generic_write_checks(iocb
, from
);
2819 ret
= __generic_file_write_iter(iocb
, from
);
2820 inode_unlock(inode
);
2825 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2831 EXPORT_SYMBOL(generic_file_write_iter
);
2834 * try_to_release_page() - release old fs-specific metadata on a page
2836 * @page: the page which the kernel is trying to free
2837 * @gfp_mask: memory allocation flags (and I/O mode)
2839 * The address_space is to try to release any data against the page
2840 * (presumably at page->private). If the release was successful, return `1'.
2841 * Otherwise return zero.
2843 * This may also be called if PG_fscache is set on a page, indicating that the
2844 * page is known to the local caching routines.
2846 * The @gfp_mask argument specifies whether I/O may be performed to release
2847 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2850 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2852 struct address_space
* const mapping
= page
->mapping
;
2854 BUG_ON(!PageLocked(page
));
2855 if (PageWriteback(page
))
2858 if (mapping
&& mapping
->a_ops
->releasepage
)
2859 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2860 return try_to_free_buffers(page
);
2863 EXPORT_SYMBOL(try_to_release_page
);