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(zone) (follow_page->mark_page_accessed)
99 * ->zone_lru_lock(zone) (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->lock_page_memcg)
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 int page_cache_tree_insert(struct address_space
*mapping
,
114 struct page
*page
, void **shadowp
)
116 struct radix_tree_node
*node
;
120 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
127 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
128 if (!radix_tree_exceptional_entry(p
))
131 mapping
->nrexceptional
--;
132 if (!dax_mapping(mapping
)) {
136 workingset_node_shadows_dec(node
);
138 /* DAX can replace empty locked entry with a hole */
140 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY
|
141 RADIX_DAX_ENTRY_LOCK
));
142 /* DAX accounts exceptional entries as normal pages */
144 workingset_node_pages_dec(node
);
145 /* Wakeup waiters for exceptional entry lock */
146 dax_wake_mapping_entry_waiter(mapping
, page
->index
,
150 radix_tree_replace_slot(slot
, page
);
153 workingset_node_pages_inc(node
);
155 * Don't track node that contains actual pages.
157 * Avoid acquiring the list_lru lock if already
158 * untracked. The list_empty() test is safe as
159 * node->private_list is protected by
160 * mapping->tree_lock.
162 if (!list_empty(&node
->private_list
))
163 list_lru_del(&workingset_shadow_nodes
,
164 &node
->private_list
);
169 static void page_cache_tree_delete(struct address_space
*mapping
,
170 struct page
*page
, void *shadow
)
172 int i
, nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
174 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
175 VM_BUG_ON_PAGE(PageTail(page
), page
);
176 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
178 for (i
= 0; i
< nr
; i
++) {
179 struct radix_tree_node
*node
;
182 __radix_tree_lookup(&mapping
->page_tree
, page
->index
+ i
,
185 radix_tree_clear_tags(&mapping
->page_tree
, node
, slot
);
188 VM_BUG_ON_PAGE(nr
!= 1, page
);
190 * We need a node to properly account shadow
191 * entries. Don't plant any without. XXX
196 radix_tree_replace_slot(slot
, shadow
);
201 workingset_node_pages_dec(node
);
203 workingset_node_shadows_inc(node
);
205 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
209 * Track node that only contains shadow entries. DAX mappings
210 * contain no shadow entries and may contain other exceptional
211 * entries so skip those.
213 * Avoid acquiring the list_lru lock if already tracked.
214 * The list_empty() test is safe as node->private_list is
215 * protected by mapping->tree_lock.
217 if (!dax_mapping(mapping
) && !workingset_node_pages(node
) &&
218 list_empty(&node
->private_list
)) {
219 node
->private_data
= mapping
;
220 list_lru_add(&workingset_shadow_nodes
,
221 &node
->private_list
);
226 mapping
->nrexceptional
+= nr
;
228 * Make sure the nrexceptional update is committed before
229 * the nrpages update so that final truncate racing
230 * with reclaim does not see both counters 0 at the
231 * same time and miss a shadow entry.
235 mapping
->nrpages
-= nr
;
239 * Delete a page from the page cache and free it. Caller has to make
240 * sure the page is locked and that nobody else uses it - or that usage
241 * is safe. The caller must hold the mapping's tree_lock.
243 void __delete_from_page_cache(struct page
*page
, void *shadow
)
245 struct address_space
*mapping
= page
->mapping
;
246 int nr
= hpage_nr_pages(page
);
248 trace_mm_filemap_delete_from_page_cache(page
);
250 * if we're uptodate, flush out into the cleancache, otherwise
251 * invalidate any existing cleancache entries. We can't leave
252 * stale data around in the cleancache once our page is gone
254 if (PageUptodate(page
) && PageMappedToDisk(page
))
255 cleancache_put_page(page
);
257 cleancache_invalidate_page(mapping
, page
);
259 VM_BUG_ON_PAGE(PageTail(page
), page
);
260 VM_BUG_ON_PAGE(page_mapped(page
), page
);
261 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
264 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
265 current
->comm
, page_to_pfn(page
));
266 dump_page(page
, "still mapped when deleted");
268 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
270 mapcount
= page_mapcount(page
);
271 if (mapping_exiting(mapping
) &&
272 page_count(page
) >= mapcount
+ 2) {
274 * All vmas have already been torn down, so it's
275 * a good bet that actually the page is unmapped,
276 * and we'd prefer not to leak it: if we're wrong,
277 * some other bad page check should catch it later.
279 page_mapcount_reset(page
);
280 page_ref_sub(page
, mapcount
);
284 page_cache_tree_delete(mapping
, page
, shadow
);
286 page
->mapping
= NULL
;
287 /* Leave page->index set: truncation lookup relies upon it */
289 /* hugetlb pages do not participate in page cache accounting. */
291 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
292 if (PageSwapBacked(page
)) {
293 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
294 if (PageTransHuge(page
))
295 __dec_node_page_state(page
, NR_SHMEM_THPS
);
297 VM_BUG_ON_PAGE(PageTransHuge(page
) && !PageHuge(page
), page
);
301 * At this point page must be either written or cleaned by truncate.
302 * Dirty page here signals a bug and loss of unwritten data.
304 * This fixes dirty accounting after removing the page entirely but
305 * leaves PageDirty set: it has no effect for truncated page and
306 * anyway will be cleared before returning page into buddy allocator.
308 if (WARN_ON_ONCE(PageDirty(page
)))
309 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
313 * delete_from_page_cache - delete page from page cache
314 * @page: the page which the kernel is trying to remove from page cache
316 * This must be called only on pages that have been verified to be in the page
317 * cache and locked. It will never put the page into the free list, the caller
318 * has a reference on the page.
320 void delete_from_page_cache(struct page
*page
)
322 struct address_space
*mapping
= page_mapping(page
);
324 void (*freepage
)(struct page
*);
326 BUG_ON(!PageLocked(page
));
328 freepage
= mapping
->a_ops
->freepage
;
330 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
331 __delete_from_page_cache(page
, NULL
);
332 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
337 if (PageTransHuge(page
) && !PageHuge(page
)) {
338 page_ref_sub(page
, HPAGE_PMD_NR
);
339 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
344 EXPORT_SYMBOL(delete_from_page_cache
);
346 int filemap_check_errors(struct address_space
*mapping
)
349 /* Check for outstanding write errors */
350 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
351 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
353 if (test_bit(AS_EIO
, &mapping
->flags
) &&
354 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
358 EXPORT_SYMBOL(filemap_check_errors
);
361 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
362 * @mapping: address space structure to write
363 * @start: offset in bytes where the range starts
364 * @end: offset in bytes where the range ends (inclusive)
365 * @sync_mode: enable synchronous operation
367 * Start writeback against all of a mapping's dirty pages that lie
368 * within the byte offsets <start, end> inclusive.
370 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
371 * opposed to a regular memory cleansing writeback. The difference between
372 * these two operations is that if a dirty page/buffer is encountered, it must
373 * be waited upon, and not just skipped over.
375 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
376 loff_t end
, int sync_mode
)
379 struct writeback_control wbc
= {
380 .sync_mode
= sync_mode
,
381 .nr_to_write
= LONG_MAX
,
382 .range_start
= start
,
386 if (!mapping_cap_writeback_dirty(mapping
))
389 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
390 ret
= do_writepages(mapping
, &wbc
);
391 wbc_detach_inode(&wbc
);
395 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
398 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
401 int filemap_fdatawrite(struct address_space
*mapping
)
403 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
405 EXPORT_SYMBOL(filemap_fdatawrite
);
407 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
410 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
412 EXPORT_SYMBOL(filemap_fdatawrite_range
);
415 * filemap_flush - mostly a non-blocking flush
416 * @mapping: target address_space
418 * This is a mostly non-blocking flush. Not suitable for data-integrity
419 * purposes - I/O may not be started against all dirty pages.
421 int filemap_flush(struct address_space
*mapping
)
423 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
425 EXPORT_SYMBOL(filemap_flush
);
427 static int __filemap_fdatawait_range(struct address_space
*mapping
,
428 loff_t start_byte
, loff_t end_byte
)
430 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
431 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
436 if (end_byte
< start_byte
)
439 pagevec_init(&pvec
, 0);
440 while ((index
<= end
) &&
441 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
442 PAGECACHE_TAG_WRITEBACK
,
443 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
446 for (i
= 0; i
< nr_pages
; i
++) {
447 struct page
*page
= pvec
.pages
[i
];
449 /* until radix tree lookup accepts end_index */
450 if (page
->index
> end
)
453 wait_on_page_writeback(page
);
454 if (TestClearPageError(page
))
457 pagevec_release(&pvec
);
465 * filemap_fdatawait_range - wait for writeback to complete
466 * @mapping: address space structure to wait for
467 * @start_byte: offset in bytes where the range starts
468 * @end_byte: offset in bytes where the range ends (inclusive)
470 * Walk the list of under-writeback pages of the given address space
471 * in the given range and wait for all of them. Check error status of
472 * the address space and return it.
474 * Since the error status of the address space is cleared by this function,
475 * callers are responsible for checking the return value and handling and/or
476 * reporting the error.
478 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
483 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
484 ret2
= filemap_check_errors(mapping
);
490 EXPORT_SYMBOL(filemap_fdatawait_range
);
493 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
494 * @mapping: address space structure to wait for
496 * Walk the list of under-writeback pages of the given address space
497 * and wait for all of them. Unlike filemap_fdatawait(), this function
498 * does not clear error status of the address space.
500 * Use this function if callers don't handle errors themselves. Expected
501 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
504 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
506 loff_t i_size
= i_size_read(mapping
->host
);
511 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
515 * filemap_fdatawait - wait for all under-writeback pages to complete
516 * @mapping: address space structure to wait for
518 * Walk the list of under-writeback pages of the given address space
519 * and wait for all of them. Check error status of the address space
522 * Since the error status of the address space is cleared by this function,
523 * callers are responsible for checking the return value and handling and/or
524 * reporting the error.
526 int filemap_fdatawait(struct address_space
*mapping
)
528 loff_t i_size
= i_size_read(mapping
->host
);
533 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
535 EXPORT_SYMBOL(filemap_fdatawait
);
537 int filemap_write_and_wait(struct address_space
*mapping
)
541 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
542 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
543 err
= filemap_fdatawrite(mapping
);
545 * Even if the above returned error, the pages may be
546 * written partially (e.g. -ENOSPC), so we wait for it.
547 * But the -EIO is special case, it may indicate the worst
548 * thing (e.g. bug) happened, so we avoid waiting for it.
551 int err2
= filemap_fdatawait(mapping
);
556 err
= filemap_check_errors(mapping
);
560 EXPORT_SYMBOL(filemap_write_and_wait
);
563 * filemap_write_and_wait_range - write out & wait on a file range
564 * @mapping: the address_space for the pages
565 * @lstart: offset in bytes where the range starts
566 * @lend: offset in bytes where the range ends (inclusive)
568 * Write out and wait upon file offsets lstart->lend, inclusive.
570 * Note that `lend' is inclusive (describes the last byte to be written) so
571 * that this function can be used to write to the very end-of-file (end = -1).
573 int filemap_write_and_wait_range(struct address_space
*mapping
,
574 loff_t lstart
, loff_t lend
)
578 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
579 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
580 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
582 /* See comment of filemap_write_and_wait() */
584 int err2
= filemap_fdatawait_range(mapping
,
590 err
= filemap_check_errors(mapping
);
594 EXPORT_SYMBOL(filemap_write_and_wait_range
);
597 * replace_page_cache_page - replace a pagecache page with a new one
598 * @old: page to be replaced
599 * @new: page to replace with
600 * @gfp_mask: allocation mode
602 * This function replaces a page in the pagecache with a new one. On
603 * success it acquires the pagecache reference for the new page and
604 * drops it for the old page. Both the old and new pages must be
605 * locked. This function does not add the new page to the LRU, the
606 * caller must do that.
608 * The remove + add is atomic. The only way this function can fail is
609 * memory allocation failure.
611 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
615 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
616 VM_BUG_ON_PAGE(!PageLocked(new), new);
617 VM_BUG_ON_PAGE(new->mapping
, new);
619 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
621 struct address_space
*mapping
= old
->mapping
;
622 void (*freepage
)(struct page
*);
625 pgoff_t offset
= old
->index
;
626 freepage
= mapping
->a_ops
->freepage
;
629 new->mapping
= mapping
;
632 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
633 __delete_from_page_cache(old
, NULL
);
634 error
= page_cache_tree_insert(mapping
, new, NULL
);
639 * hugetlb pages do not participate in page cache accounting.
642 __inc_node_page_state(new, NR_FILE_PAGES
);
643 if (PageSwapBacked(new))
644 __inc_node_page_state(new, NR_SHMEM
);
645 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
646 mem_cgroup_migrate(old
, new);
647 radix_tree_preload_end();
655 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
657 static int __add_to_page_cache_locked(struct page
*page
,
658 struct address_space
*mapping
,
659 pgoff_t offset
, gfp_t gfp_mask
,
662 int huge
= PageHuge(page
);
663 struct mem_cgroup
*memcg
;
666 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
667 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
670 error
= mem_cgroup_try_charge(page
, current
->mm
,
671 gfp_mask
, &memcg
, false);
676 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
679 mem_cgroup_cancel_charge(page
, memcg
, false);
684 page
->mapping
= mapping
;
685 page
->index
= offset
;
687 spin_lock_irq(&mapping
->tree_lock
);
688 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
689 radix_tree_preload_end();
693 /* hugetlb pages do not participate in page cache accounting. */
695 __inc_node_page_state(page
, NR_FILE_PAGES
);
696 spin_unlock_irq(&mapping
->tree_lock
);
698 mem_cgroup_commit_charge(page
, memcg
, false, false);
699 trace_mm_filemap_add_to_page_cache(page
);
702 page
->mapping
= NULL
;
703 /* Leave page->index set: truncation relies upon it */
704 spin_unlock_irq(&mapping
->tree_lock
);
706 mem_cgroup_cancel_charge(page
, memcg
, false);
712 * add_to_page_cache_locked - add a locked page to the pagecache
714 * @mapping: the page's address_space
715 * @offset: page index
716 * @gfp_mask: page allocation mode
718 * This function is used to add a page to the pagecache. It must be locked.
719 * This function does not add the page to the LRU. The caller must do that.
721 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
722 pgoff_t offset
, gfp_t gfp_mask
)
724 return __add_to_page_cache_locked(page
, mapping
, offset
,
727 EXPORT_SYMBOL(add_to_page_cache_locked
);
729 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
730 pgoff_t offset
, gfp_t gfp_mask
)
735 __SetPageLocked(page
);
736 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
739 __ClearPageLocked(page
);
742 * The page might have been evicted from cache only
743 * recently, in which case it should be activated like
744 * any other repeatedly accessed page.
745 * The exception is pages getting rewritten; evicting other
746 * data from the working set, only to cache data that will
747 * get overwritten with something else, is a waste of memory.
749 if (!(gfp_mask
& __GFP_WRITE
) &&
750 shadow
&& workingset_refault(shadow
)) {
752 workingset_activation(page
);
754 ClearPageActive(page
);
759 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
762 struct page
*__page_cache_alloc(gfp_t gfp
)
767 if (cpuset_do_page_mem_spread()) {
768 unsigned int cpuset_mems_cookie
;
770 cpuset_mems_cookie
= read_mems_allowed_begin();
771 n
= cpuset_mem_spread_node();
772 page
= __alloc_pages_node(n
, gfp
, 0);
773 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
777 return alloc_pages(gfp
, 0);
779 EXPORT_SYMBOL(__page_cache_alloc
);
783 * In order to wait for pages to become available there must be
784 * waitqueues associated with pages. By using a hash table of
785 * waitqueues where the bucket discipline is to maintain all
786 * waiters on the same queue and wake all when any of the pages
787 * become available, and for the woken contexts to check to be
788 * sure the appropriate page became available, this saves space
789 * at a cost of "thundering herd" phenomena during rare hash
792 wait_queue_head_t
*page_waitqueue(struct page
*page
)
794 const struct zone
*zone
= page_zone(page
);
796 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
798 EXPORT_SYMBOL(page_waitqueue
);
800 void wait_on_page_bit(struct page
*page
, int bit_nr
)
802 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
804 if (test_bit(bit_nr
, &page
->flags
))
805 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
806 TASK_UNINTERRUPTIBLE
);
808 EXPORT_SYMBOL(wait_on_page_bit
);
810 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
812 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
814 if (!test_bit(bit_nr
, &page
->flags
))
817 return __wait_on_bit(page_waitqueue(page
), &wait
,
818 bit_wait_io
, TASK_KILLABLE
);
821 int wait_on_page_bit_killable_timeout(struct page
*page
,
822 int bit_nr
, unsigned long timeout
)
824 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
826 wait
.key
.timeout
= jiffies
+ timeout
;
827 if (!test_bit(bit_nr
, &page
->flags
))
829 return __wait_on_bit(page_waitqueue(page
), &wait
,
830 bit_wait_io_timeout
, TASK_KILLABLE
);
832 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
835 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
836 * @page: Page defining the wait queue of interest
837 * @waiter: Waiter to add to the queue
839 * Add an arbitrary @waiter to the wait queue for the nominated @page.
841 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
843 wait_queue_head_t
*q
= page_waitqueue(page
);
846 spin_lock_irqsave(&q
->lock
, flags
);
847 __add_wait_queue(q
, waiter
);
848 spin_unlock_irqrestore(&q
->lock
, flags
);
850 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
853 * unlock_page - unlock a locked page
856 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
857 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
858 * mechanism between PageLocked pages and PageWriteback pages is shared.
859 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
861 * The mb is necessary to enforce ordering between the clear_bit and the read
862 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
864 void unlock_page(struct page
*page
)
866 page
= compound_head(page
);
867 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
868 clear_bit_unlock(PG_locked
, &page
->flags
);
869 smp_mb__after_atomic();
870 wake_up_page(page
, PG_locked
);
872 EXPORT_SYMBOL(unlock_page
);
875 * end_page_writeback - end writeback against a page
878 void end_page_writeback(struct page
*page
)
881 * TestClearPageReclaim could be used here but it is an atomic
882 * operation and overkill in this particular case. Failing to
883 * shuffle a page marked for immediate reclaim is too mild to
884 * justify taking an atomic operation penalty at the end of
885 * ever page writeback.
887 if (PageReclaim(page
)) {
888 ClearPageReclaim(page
);
889 rotate_reclaimable_page(page
);
892 if (!test_clear_page_writeback(page
))
895 smp_mb__after_atomic();
896 wake_up_page(page
, PG_writeback
);
898 EXPORT_SYMBOL(end_page_writeback
);
901 * After completing I/O on a page, call this routine to update the page
902 * flags appropriately
904 void page_endio(struct page
*page
, bool is_write
, int err
)
908 SetPageUptodate(page
);
910 ClearPageUptodate(page
);
918 mapping_set_error(page
->mapping
, err
);
920 end_page_writeback(page
);
923 EXPORT_SYMBOL_GPL(page_endio
);
926 * __lock_page - get a lock on the page, assuming we need to sleep to get it
927 * @page: the page to lock
929 void __lock_page(struct page
*page
)
931 struct page
*page_head
= compound_head(page
);
932 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
934 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
935 TASK_UNINTERRUPTIBLE
);
937 EXPORT_SYMBOL(__lock_page
);
939 int __lock_page_killable(struct page
*page
)
941 struct page
*page_head
= compound_head(page
);
942 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
944 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
945 bit_wait_io
, TASK_KILLABLE
);
947 EXPORT_SYMBOL_GPL(__lock_page_killable
);
951 * 1 - page is locked; mmap_sem is still held.
952 * 0 - page is not locked.
953 * mmap_sem has been released (up_read()), unless flags had both
954 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
955 * which case mmap_sem is still held.
957 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
958 * with the page locked and the mmap_sem unperturbed.
960 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
963 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
965 * CAUTION! In this case, mmap_sem is not released
966 * even though return 0.
968 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
971 up_read(&mm
->mmap_sem
);
972 if (flags
& FAULT_FLAG_KILLABLE
)
973 wait_on_page_locked_killable(page
);
975 wait_on_page_locked(page
);
978 if (flags
& FAULT_FLAG_KILLABLE
) {
981 ret
= __lock_page_killable(page
);
983 up_read(&mm
->mmap_sem
);
993 * page_cache_next_hole - find the next hole (not-present entry)
996 * @max_scan: maximum range to search
998 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
999 * lowest indexed hole.
1001 * Returns: the index of the hole if found, otherwise returns an index
1002 * outside of the set specified (in which case 'return - index >=
1003 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1006 * page_cache_next_hole may be called under rcu_read_lock. However,
1007 * like radix_tree_gang_lookup, this will not atomically search a
1008 * snapshot of the tree at a single point in time. For example, if a
1009 * hole is created at index 5, then subsequently a hole is created at
1010 * index 10, page_cache_next_hole covering both indexes may return 10
1011 * if called under rcu_read_lock.
1013 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1014 pgoff_t index
, unsigned long max_scan
)
1018 for (i
= 0; i
< max_scan
; i
++) {
1021 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1022 if (!page
|| radix_tree_exceptional_entry(page
))
1031 EXPORT_SYMBOL(page_cache_next_hole
);
1034 * page_cache_prev_hole - find the prev hole (not-present entry)
1037 * @max_scan: maximum range to search
1039 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1042 * Returns: the index of the hole if found, otherwise returns an index
1043 * outside of the set specified (in which case 'index - return >=
1044 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1047 * page_cache_prev_hole may be called under rcu_read_lock. However,
1048 * like radix_tree_gang_lookup, this will not atomically search a
1049 * snapshot of the tree at a single point in time. For example, if a
1050 * hole is created at index 10, then subsequently a hole is created at
1051 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1052 * called under rcu_read_lock.
1054 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1055 pgoff_t index
, unsigned long max_scan
)
1059 for (i
= 0; i
< max_scan
; i
++) {
1062 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1063 if (!page
|| radix_tree_exceptional_entry(page
))
1066 if (index
== ULONG_MAX
)
1072 EXPORT_SYMBOL(page_cache_prev_hole
);
1075 * find_get_entry - find and get a page cache entry
1076 * @mapping: the address_space to search
1077 * @offset: the page cache index
1079 * Looks up the page cache slot at @mapping & @offset. If there is a
1080 * page cache page, it is returned with an increased refcount.
1082 * If the slot holds a shadow entry of a previously evicted page, or a
1083 * swap entry from shmem/tmpfs, it is returned.
1085 * Otherwise, %NULL is returned.
1087 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1090 struct page
*head
, *page
;
1095 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1097 page
= radix_tree_deref_slot(pagep
);
1098 if (unlikely(!page
))
1100 if (radix_tree_exception(page
)) {
1101 if (radix_tree_deref_retry(page
))
1104 * A shadow entry of a recently evicted page,
1105 * or a swap entry from shmem/tmpfs. Return
1106 * it without attempting to raise page count.
1111 head
= compound_head(page
);
1112 if (!page_cache_get_speculative(head
))
1115 /* The page was split under us? */
1116 if (compound_head(page
) != head
) {
1122 * Has the page moved?
1123 * This is part of the lockless pagecache protocol. See
1124 * include/linux/pagemap.h for details.
1126 if (unlikely(page
!= *pagep
)) {
1136 EXPORT_SYMBOL(find_get_entry
);
1139 * find_lock_entry - locate, pin and lock a page cache entry
1140 * @mapping: the address_space to search
1141 * @offset: the page cache index
1143 * Looks up the page cache slot at @mapping & @offset. If there is a
1144 * page cache page, it is returned locked and with an increased
1147 * If the slot holds a shadow entry of a previously evicted page, or a
1148 * swap entry from shmem/tmpfs, it is returned.
1150 * Otherwise, %NULL is returned.
1152 * find_lock_entry() may sleep.
1154 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1159 page
= find_get_entry(mapping
, offset
);
1160 if (page
&& !radix_tree_exception(page
)) {
1162 /* Has the page been truncated? */
1163 if (unlikely(page_mapping(page
) != mapping
)) {
1168 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1172 EXPORT_SYMBOL(find_lock_entry
);
1175 * pagecache_get_page - find and get a page reference
1176 * @mapping: the address_space to search
1177 * @offset: the page index
1178 * @fgp_flags: PCG flags
1179 * @gfp_mask: gfp mask to use for the page cache data page allocation
1181 * Looks up the page cache slot at @mapping & @offset.
1183 * PCG flags modify how the page is returned.
1185 * FGP_ACCESSED: the page will be marked accessed
1186 * FGP_LOCK: Page is return locked
1187 * FGP_CREAT: If page is not present then a new page is allocated using
1188 * @gfp_mask and added to the page cache and the VM's LRU
1189 * list. The page is returned locked and with an increased
1190 * refcount. Otherwise, %NULL is returned.
1192 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1193 * if the GFP flags specified for FGP_CREAT are atomic.
1195 * If there is a page cache page, it is returned with an increased refcount.
1197 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1198 int fgp_flags
, gfp_t gfp_mask
)
1203 page
= find_get_entry(mapping
, offset
);
1204 if (radix_tree_exceptional_entry(page
))
1209 if (fgp_flags
& FGP_LOCK
) {
1210 if (fgp_flags
& FGP_NOWAIT
) {
1211 if (!trylock_page(page
)) {
1219 /* Has the page been truncated? */
1220 if (unlikely(page
->mapping
!= mapping
)) {
1225 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1228 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1229 mark_page_accessed(page
);
1232 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1234 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1235 gfp_mask
|= __GFP_WRITE
;
1236 if (fgp_flags
& FGP_NOFS
)
1237 gfp_mask
&= ~__GFP_FS
;
1239 page
= __page_cache_alloc(gfp_mask
);
1243 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1244 fgp_flags
|= FGP_LOCK
;
1246 /* Init accessed so avoid atomic mark_page_accessed later */
1247 if (fgp_flags
& FGP_ACCESSED
)
1248 __SetPageReferenced(page
);
1250 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1251 gfp_mask
& GFP_RECLAIM_MASK
);
1252 if (unlikely(err
)) {
1262 EXPORT_SYMBOL(pagecache_get_page
);
1265 * find_get_entries - gang pagecache lookup
1266 * @mapping: The address_space to search
1267 * @start: The starting page cache index
1268 * @nr_entries: The maximum number of entries
1269 * @entries: Where the resulting entries are placed
1270 * @indices: The cache indices corresponding to the entries in @entries
1272 * find_get_entries() will search for and return a group of up to
1273 * @nr_entries entries in the mapping. The entries are placed at
1274 * @entries. find_get_entries() takes a reference against any actual
1277 * The search returns a group of mapping-contiguous page cache entries
1278 * with ascending indexes. There may be holes in the indices due to
1279 * not-present pages.
1281 * Any shadow entries of evicted pages, or swap entries from
1282 * shmem/tmpfs, are included in the returned array.
1284 * find_get_entries() returns the number of pages and shadow entries
1287 unsigned find_get_entries(struct address_space
*mapping
,
1288 pgoff_t start
, unsigned int nr_entries
,
1289 struct page
**entries
, pgoff_t
*indices
)
1292 unsigned int ret
= 0;
1293 struct radix_tree_iter iter
;
1299 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1300 struct page
*head
, *page
;
1302 page
= radix_tree_deref_slot(slot
);
1303 if (unlikely(!page
))
1305 if (radix_tree_exception(page
)) {
1306 if (radix_tree_deref_retry(page
)) {
1307 slot
= radix_tree_iter_retry(&iter
);
1311 * A shadow entry of a recently evicted page, a swap
1312 * entry from shmem/tmpfs or a DAX entry. Return it
1313 * without attempting to raise page count.
1318 head
= compound_head(page
);
1319 if (!page_cache_get_speculative(head
))
1322 /* The page was split under us? */
1323 if (compound_head(page
) != head
) {
1328 /* Has the page moved? */
1329 if (unlikely(page
!= *slot
)) {
1334 indices
[ret
] = iter
.index
;
1335 entries
[ret
] = page
;
1336 if (++ret
== nr_entries
)
1344 * find_get_pages - gang pagecache lookup
1345 * @mapping: The address_space to search
1346 * @start: The starting page index
1347 * @nr_pages: The maximum number of pages
1348 * @pages: Where the resulting pages are placed
1350 * find_get_pages() will search for and return a group of up to
1351 * @nr_pages pages in the mapping. The pages are placed at @pages.
1352 * find_get_pages() takes a reference against the returned pages.
1354 * The search returns a group of mapping-contiguous pages with ascending
1355 * indexes. There may be holes in the indices due to not-present pages.
1357 * find_get_pages() returns the number of pages which were found.
1359 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1360 unsigned int nr_pages
, struct page
**pages
)
1362 struct radix_tree_iter iter
;
1366 if (unlikely(!nr_pages
))
1370 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1371 struct page
*head
, *page
;
1373 page
= radix_tree_deref_slot(slot
);
1374 if (unlikely(!page
))
1377 if (radix_tree_exception(page
)) {
1378 if (radix_tree_deref_retry(page
)) {
1379 slot
= radix_tree_iter_retry(&iter
);
1383 * A shadow entry of a recently evicted page,
1384 * or a swap entry from shmem/tmpfs. Skip
1390 head
= compound_head(page
);
1391 if (!page_cache_get_speculative(head
))
1394 /* The page was split under us? */
1395 if (compound_head(page
) != head
) {
1400 /* Has the page moved? */
1401 if (unlikely(page
!= *slot
)) {
1407 if (++ret
== nr_pages
)
1416 * find_get_pages_contig - gang contiguous pagecache lookup
1417 * @mapping: The address_space to search
1418 * @index: The starting page index
1419 * @nr_pages: The maximum number of pages
1420 * @pages: Where the resulting pages are placed
1422 * find_get_pages_contig() works exactly like find_get_pages(), except
1423 * that the returned number of pages are guaranteed to be contiguous.
1425 * find_get_pages_contig() returns the number of pages which were found.
1427 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1428 unsigned int nr_pages
, struct page
**pages
)
1430 struct radix_tree_iter iter
;
1432 unsigned int ret
= 0;
1434 if (unlikely(!nr_pages
))
1438 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1439 struct page
*head
, *page
;
1441 page
= radix_tree_deref_slot(slot
);
1442 /* The hole, there no reason to continue */
1443 if (unlikely(!page
))
1446 if (radix_tree_exception(page
)) {
1447 if (radix_tree_deref_retry(page
)) {
1448 slot
= radix_tree_iter_retry(&iter
);
1452 * A shadow entry of a recently evicted page,
1453 * or a swap entry from shmem/tmpfs. Stop
1454 * looking for contiguous pages.
1459 head
= compound_head(page
);
1460 if (!page_cache_get_speculative(head
))
1463 /* The page was split under us? */
1464 if (compound_head(page
) != head
) {
1469 /* Has the page moved? */
1470 if (unlikely(page
!= *slot
)) {
1476 * must check mapping and index after taking the ref.
1477 * otherwise we can get both false positives and false
1478 * negatives, which is just confusing to the caller.
1480 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1486 if (++ret
== nr_pages
)
1492 EXPORT_SYMBOL(find_get_pages_contig
);
1495 * find_get_pages_tag - find and return pages that match @tag
1496 * @mapping: the address_space to search
1497 * @index: the starting page index
1498 * @tag: the tag index
1499 * @nr_pages: the maximum number of pages
1500 * @pages: where the resulting pages are placed
1502 * Like find_get_pages, except we only return pages which are tagged with
1503 * @tag. We update @index to index the next page for the traversal.
1505 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1506 int tag
, unsigned int nr_pages
, struct page
**pages
)
1508 struct radix_tree_iter iter
;
1512 if (unlikely(!nr_pages
))
1516 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1517 &iter
, *index
, tag
) {
1518 struct page
*head
, *page
;
1520 page
= radix_tree_deref_slot(slot
);
1521 if (unlikely(!page
))
1524 if (radix_tree_exception(page
)) {
1525 if (radix_tree_deref_retry(page
)) {
1526 slot
= radix_tree_iter_retry(&iter
);
1530 * A shadow entry of a recently evicted page.
1532 * Those entries should never be tagged, but
1533 * this tree walk is lockless and the tags are
1534 * looked up in bulk, one radix tree node at a
1535 * time, so there is a sizable window for page
1536 * reclaim to evict a page we saw tagged.
1543 head
= compound_head(page
);
1544 if (!page_cache_get_speculative(head
))
1547 /* The page was split under us? */
1548 if (compound_head(page
) != head
) {
1553 /* Has the page moved? */
1554 if (unlikely(page
!= *slot
)) {
1560 if (++ret
== nr_pages
)
1567 *index
= pages
[ret
- 1]->index
+ 1;
1571 EXPORT_SYMBOL(find_get_pages_tag
);
1574 * find_get_entries_tag - find and return entries that match @tag
1575 * @mapping: the address_space to search
1576 * @start: the starting page cache index
1577 * @tag: the tag index
1578 * @nr_entries: the maximum number of entries
1579 * @entries: where the resulting entries are placed
1580 * @indices: the cache indices corresponding to the entries in @entries
1582 * Like find_get_entries, except we only return entries which are tagged with
1585 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1586 int tag
, unsigned int nr_entries
,
1587 struct page
**entries
, pgoff_t
*indices
)
1590 unsigned int ret
= 0;
1591 struct radix_tree_iter iter
;
1597 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1598 &iter
, start
, tag
) {
1599 struct page
*head
, *page
;
1601 page
= radix_tree_deref_slot(slot
);
1602 if (unlikely(!page
))
1604 if (radix_tree_exception(page
)) {
1605 if (radix_tree_deref_retry(page
)) {
1606 slot
= radix_tree_iter_retry(&iter
);
1611 * A shadow entry of a recently evicted page, a swap
1612 * entry from shmem/tmpfs or a DAX entry. Return it
1613 * without attempting to raise page count.
1618 head
= compound_head(page
);
1619 if (!page_cache_get_speculative(head
))
1622 /* The page was split under us? */
1623 if (compound_head(page
) != head
) {
1628 /* Has the page moved? */
1629 if (unlikely(page
!= *slot
)) {
1634 indices
[ret
] = iter
.index
;
1635 entries
[ret
] = page
;
1636 if (++ret
== nr_entries
)
1642 EXPORT_SYMBOL(find_get_entries_tag
);
1645 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1646 * a _large_ part of the i/o request. Imagine the worst scenario:
1648 * ---R__________________________________________B__________
1649 * ^ reading here ^ bad block(assume 4k)
1651 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1652 * => failing the whole request => read(R) => read(R+1) =>
1653 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1654 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1655 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1657 * It is going insane. Fix it by quickly scaling down the readahead size.
1659 static void shrink_readahead_size_eio(struct file
*filp
,
1660 struct file_ra_state
*ra
)
1666 * do_generic_file_read - generic file read routine
1667 * @filp: the file to read
1668 * @ppos: current file position
1669 * @iter: data destination
1670 * @written: already copied
1672 * This is a generic file read routine, and uses the
1673 * mapping->a_ops->readpage() function for the actual low-level stuff.
1675 * This is really ugly. But the goto's actually try to clarify some
1676 * of the logic when it comes to error handling etc.
1678 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1679 struct iov_iter
*iter
, ssize_t written
)
1681 struct address_space
*mapping
= filp
->f_mapping
;
1682 struct inode
*inode
= mapping
->host
;
1683 struct file_ra_state
*ra
= &filp
->f_ra
;
1687 unsigned long offset
; /* offset into pagecache page */
1688 unsigned int prev_offset
;
1691 index
= *ppos
>> PAGE_SHIFT
;
1692 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1693 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1694 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1695 offset
= *ppos
& ~PAGE_MASK
;
1701 unsigned long nr
, ret
;
1705 page
= find_get_page(mapping
, index
);
1707 page_cache_sync_readahead(mapping
,
1709 index
, last_index
- index
);
1710 page
= find_get_page(mapping
, index
);
1711 if (unlikely(page
== NULL
))
1712 goto no_cached_page
;
1714 if (PageReadahead(page
)) {
1715 page_cache_async_readahead(mapping
,
1717 index
, last_index
- index
);
1719 if (!PageUptodate(page
)) {
1721 * See comment in do_read_cache_page on why
1722 * wait_on_page_locked is used to avoid unnecessarily
1723 * serialisations and why it's safe.
1725 wait_on_page_locked_killable(page
);
1726 if (PageUptodate(page
))
1729 if (inode
->i_blkbits
== PAGE_SHIFT
||
1730 !mapping
->a_ops
->is_partially_uptodate
)
1731 goto page_not_up_to_date
;
1732 if (!trylock_page(page
))
1733 goto page_not_up_to_date
;
1734 /* Did it get truncated before we got the lock? */
1736 goto page_not_up_to_date_locked
;
1737 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1738 offset
, iter
->count
))
1739 goto page_not_up_to_date_locked
;
1744 * i_size must be checked after we know the page is Uptodate.
1746 * Checking i_size after the check allows us to calculate
1747 * the correct value for "nr", which means the zero-filled
1748 * part of the page is not copied back to userspace (unless
1749 * another truncate extends the file - this is desired though).
1752 isize
= i_size_read(inode
);
1753 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1754 if (unlikely(!isize
|| index
> end_index
)) {
1759 /* nr is the maximum number of bytes to copy from this page */
1761 if (index
== end_index
) {
1762 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1770 /* If users can be writing to this page using arbitrary
1771 * virtual addresses, take care about potential aliasing
1772 * before reading the page on the kernel side.
1774 if (mapping_writably_mapped(mapping
))
1775 flush_dcache_page(page
);
1778 * When a sequential read accesses a page several times,
1779 * only mark it as accessed the first time.
1781 if (prev_index
!= index
|| offset
!= prev_offset
)
1782 mark_page_accessed(page
);
1786 * Ok, we have the page, and it's up-to-date, so
1787 * now we can copy it to user space...
1790 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1792 index
+= offset
>> PAGE_SHIFT
;
1793 offset
&= ~PAGE_MASK
;
1794 prev_offset
= offset
;
1798 if (!iov_iter_count(iter
))
1806 page_not_up_to_date
:
1807 /* Get exclusive access to the page ... */
1808 error
= lock_page_killable(page
);
1809 if (unlikely(error
))
1810 goto readpage_error
;
1812 page_not_up_to_date_locked
:
1813 /* Did it get truncated before we got the lock? */
1814 if (!page
->mapping
) {
1820 /* Did somebody else fill it already? */
1821 if (PageUptodate(page
)) {
1828 * A previous I/O error may have been due to temporary
1829 * failures, eg. multipath errors.
1830 * PG_error will be set again if readpage fails.
1832 ClearPageError(page
);
1833 /* Start the actual read. The read will unlock the page. */
1834 error
= mapping
->a_ops
->readpage(filp
, page
);
1836 if (unlikely(error
)) {
1837 if (error
== AOP_TRUNCATED_PAGE
) {
1842 goto readpage_error
;
1845 if (!PageUptodate(page
)) {
1846 error
= lock_page_killable(page
);
1847 if (unlikely(error
))
1848 goto readpage_error
;
1849 if (!PageUptodate(page
)) {
1850 if (page
->mapping
== NULL
) {
1852 * invalidate_mapping_pages got it
1859 shrink_readahead_size_eio(filp
, ra
);
1861 goto readpage_error
;
1869 /* UHHUH! A synchronous read error occurred. Report it */
1875 * Ok, it wasn't cached, so we need to create a new
1878 page
= page_cache_alloc_cold(mapping
);
1883 error
= add_to_page_cache_lru(page
, mapping
, index
,
1884 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1887 if (error
== -EEXIST
) {
1897 ra
->prev_pos
= prev_index
;
1898 ra
->prev_pos
<<= PAGE_SHIFT
;
1899 ra
->prev_pos
|= prev_offset
;
1901 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
1902 file_accessed(filp
);
1903 return written
? written
: error
;
1907 * generic_file_read_iter - generic filesystem read routine
1908 * @iocb: kernel I/O control block
1909 * @iter: destination for the data read
1911 * This is the "read_iter()" routine for all filesystems
1912 * that can use the page cache directly.
1915 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1917 struct file
*file
= iocb
->ki_filp
;
1919 size_t count
= iov_iter_count(iter
);
1922 goto out
; /* skip atime */
1924 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1925 struct address_space
*mapping
= file
->f_mapping
;
1926 struct inode
*inode
= mapping
->host
;
1929 size
= i_size_read(inode
);
1930 retval
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
1931 iocb
->ki_pos
+ count
- 1);
1933 struct iov_iter data
= *iter
;
1934 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
);
1938 iocb
->ki_pos
+= retval
;
1939 iov_iter_advance(iter
, retval
);
1943 * Btrfs can have a short DIO read if we encounter
1944 * compressed extents, so if there was an error, or if
1945 * we've already read everything we wanted to, or if
1946 * there was a short read because we hit EOF, go ahead
1947 * and return. Otherwise fallthrough to buffered io for
1948 * the rest of the read. Buffered reads will not work for
1949 * DAX files, so don't bother trying.
1951 if (retval
< 0 || !iov_iter_count(iter
) || iocb
->ki_pos
>= size
||
1953 file_accessed(file
);
1958 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
1962 EXPORT_SYMBOL(generic_file_read_iter
);
1966 * page_cache_read - adds requested page to the page cache if not already there
1967 * @file: file to read
1968 * @offset: page index
1969 * @gfp_mask: memory allocation flags
1971 * This adds the requested page to the page cache if it isn't already there,
1972 * and schedules an I/O to read in its contents from disk.
1974 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1976 struct address_space
*mapping
= file
->f_mapping
;
1981 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1985 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1987 ret
= mapping
->a_ops
->readpage(file
, page
);
1988 else if (ret
== -EEXIST
)
1989 ret
= 0; /* losing race to add is OK */
1993 } while (ret
== AOP_TRUNCATED_PAGE
);
1998 #define MMAP_LOTSAMISS (100)
2001 * Synchronous readahead happens when we don't even find
2002 * a page in the page cache at all.
2004 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2005 struct file_ra_state
*ra
,
2009 struct address_space
*mapping
= file
->f_mapping
;
2011 /* If we don't want any read-ahead, don't bother */
2012 if (vma
->vm_flags
& VM_RAND_READ
)
2017 if (vma
->vm_flags
& VM_SEQ_READ
) {
2018 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2023 /* Avoid banging the cache line if not needed */
2024 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2028 * Do we miss much more than hit in this file? If so,
2029 * stop bothering with read-ahead. It will only hurt.
2031 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2037 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2038 ra
->size
= ra
->ra_pages
;
2039 ra
->async_size
= ra
->ra_pages
/ 4;
2040 ra_submit(ra
, mapping
, file
);
2044 * Asynchronous readahead happens when we find the page and PG_readahead,
2045 * so we want to possibly extend the readahead further..
2047 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2048 struct file_ra_state
*ra
,
2053 struct address_space
*mapping
= file
->f_mapping
;
2055 /* If we don't want any read-ahead, don't bother */
2056 if (vma
->vm_flags
& VM_RAND_READ
)
2058 if (ra
->mmap_miss
> 0)
2060 if (PageReadahead(page
))
2061 page_cache_async_readahead(mapping
, ra
, file
,
2062 page
, offset
, ra
->ra_pages
);
2066 * filemap_fault - read in file data for page fault handling
2067 * @vma: vma in which the fault was taken
2068 * @vmf: struct vm_fault containing details of the fault
2070 * filemap_fault() is invoked via the vma operations vector for a
2071 * mapped memory region to read in file data during a page fault.
2073 * The goto's are kind of ugly, but this streamlines the normal case of having
2074 * it in the page cache, and handles the special cases reasonably without
2075 * having a lot of duplicated code.
2077 * vma->vm_mm->mmap_sem must be held on entry.
2079 * If our return value has VM_FAULT_RETRY set, it's because
2080 * lock_page_or_retry() returned 0.
2081 * The mmap_sem has usually been released in this case.
2082 * See __lock_page_or_retry() for the exception.
2084 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2085 * has not been released.
2087 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2089 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2092 struct file
*file
= vma
->vm_file
;
2093 struct address_space
*mapping
= file
->f_mapping
;
2094 struct file_ra_state
*ra
= &file
->f_ra
;
2095 struct inode
*inode
= mapping
->host
;
2096 pgoff_t offset
= vmf
->pgoff
;
2101 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2102 if (offset
>= size
>> PAGE_SHIFT
)
2103 return VM_FAULT_SIGBUS
;
2106 * Do we have something in the page cache already?
2108 page
= find_get_page(mapping
, offset
);
2109 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2111 * We found the page, so try async readahead before
2112 * waiting for the lock.
2114 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2116 /* No page in the page cache at all */
2117 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2118 count_vm_event(PGMAJFAULT
);
2119 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2120 ret
= VM_FAULT_MAJOR
;
2122 page
= find_get_page(mapping
, offset
);
2124 goto no_cached_page
;
2127 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2129 return ret
| VM_FAULT_RETRY
;
2132 /* Did it get truncated? */
2133 if (unlikely(page
->mapping
!= mapping
)) {
2138 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2141 * We have a locked page in the page cache, now we need to check
2142 * that it's up-to-date. If not, it is going to be due to an error.
2144 if (unlikely(!PageUptodate(page
)))
2145 goto page_not_uptodate
;
2148 * Found the page and have a reference on it.
2149 * We must recheck i_size under page lock.
2151 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2152 if (unlikely(offset
>= size
>> PAGE_SHIFT
)) {
2155 return VM_FAULT_SIGBUS
;
2159 return ret
| VM_FAULT_LOCKED
;
2163 * We're only likely to ever get here if MADV_RANDOM is in
2166 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2169 * The page we want has now been added to the page cache.
2170 * In the unlikely event that someone removed it in the
2171 * meantime, we'll just come back here and read it again.
2177 * An error return from page_cache_read can result if the
2178 * system is low on memory, or a problem occurs while trying
2181 if (error
== -ENOMEM
)
2182 return VM_FAULT_OOM
;
2183 return VM_FAULT_SIGBUS
;
2187 * Umm, take care of errors if the page isn't up-to-date.
2188 * Try to re-read it _once_. We do this synchronously,
2189 * because there really aren't any performance issues here
2190 * and we need to check for errors.
2192 ClearPageError(page
);
2193 error
= mapping
->a_ops
->readpage(file
, page
);
2195 wait_on_page_locked(page
);
2196 if (!PageUptodate(page
))
2201 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2204 /* Things didn't work out. Return zero to tell the mm layer so. */
2205 shrink_readahead_size_eio(file
, ra
);
2206 return VM_FAULT_SIGBUS
;
2208 EXPORT_SYMBOL(filemap_fault
);
2210 void filemap_map_pages(struct fault_env
*fe
,
2211 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2213 struct radix_tree_iter iter
;
2215 struct file
*file
= fe
->vma
->vm_file
;
2216 struct address_space
*mapping
= file
->f_mapping
;
2217 pgoff_t last_pgoff
= start_pgoff
;
2219 struct page
*head
, *page
;
2222 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2224 if (iter
.index
> end_pgoff
)
2227 page
= radix_tree_deref_slot(slot
);
2228 if (unlikely(!page
))
2230 if (radix_tree_exception(page
)) {
2231 if (radix_tree_deref_retry(page
)) {
2232 slot
= radix_tree_iter_retry(&iter
);
2238 head
= compound_head(page
);
2239 if (!page_cache_get_speculative(head
))
2242 /* The page was split under us? */
2243 if (compound_head(page
) != head
) {
2248 /* Has the page moved? */
2249 if (unlikely(page
!= *slot
)) {
2254 if (!PageUptodate(page
) ||
2255 PageReadahead(page
) ||
2258 if (!trylock_page(page
))
2261 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2264 size
= round_up(i_size_read(mapping
->host
), PAGE_SIZE
);
2265 if (page
->index
>= size
>> PAGE_SHIFT
)
2268 if (file
->f_ra
.mmap_miss
> 0)
2269 file
->f_ra
.mmap_miss
--;
2271 fe
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2273 fe
->pte
+= iter
.index
- last_pgoff
;
2274 last_pgoff
= iter
.index
;
2275 if (alloc_set_pte(fe
, NULL
, page
))
2284 /* Huge page is mapped? No need to proceed. */
2285 if (pmd_trans_huge(*fe
->pmd
))
2287 if (iter
.index
== end_pgoff
)
2292 EXPORT_SYMBOL(filemap_map_pages
);
2294 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2296 struct page
*page
= vmf
->page
;
2297 struct inode
*inode
= file_inode(vma
->vm_file
);
2298 int ret
= VM_FAULT_LOCKED
;
2300 sb_start_pagefault(inode
->i_sb
);
2301 file_update_time(vma
->vm_file
);
2303 if (page
->mapping
!= inode
->i_mapping
) {
2305 ret
= VM_FAULT_NOPAGE
;
2309 * We mark the page dirty already here so that when freeze is in
2310 * progress, we are guaranteed that writeback during freezing will
2311 * see the dirty page and writeprotect it again.
2313 set_page_dirty(page
);
2314 wait_for_stable_page(page
);
2316 sb_end_pagefault(inode
->i_sb
);
2319 EXPORT_SYMBOL(filemap_page_mkwrite
);
2321 const struct vm_operations_struct generic_file_vm_ops
= {
2322 .fault
= filemap_fault
,
2323 .map_pages
= filemap_map_pages
,
2324 .page_mkwrite
= filemap_page_mkwrite
,
2327 /* This is used for a general mmap of a disk file */
2329 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2331 struct address_space
*mapping
= file
->f_mapping
;
2333 if (!mapping
->a_ops
->readpage
)
2335 file_accessed(file
);
2336 vma
->vm_ops
= &generic_file_vm_ops
;
2341 * This is for filesystems which do not implement ->writepage.
2343 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2345 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2347 return generic_file_mmap(file
, vma
);
2350 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2354 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2358 #endif /* CONFIG_MMU */
2360 EXPORT_SYMBOL(generic_file_mmap
);
2361 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2363 static struct page
*wait_on_page_read(struct page
*page
)
2365 if (!IS_ERR(page
)) {
2366 wait_on_page_locked(page
);
2367 if (!PageUptodate(page
)) {
2369 page
= ERR_PTR(-EIO
);
2375 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2377 int (*filler
)(void *, struct page
*),
2384 page
= find_get_page(mapping
, index
);
2386 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2388 return ERR_PTR(-ENOMEM
);
2389 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2390 if (unlikely(err
)) {
2394 /* Presumably ENOMEM for radix tree node */
2395 return ERR_PTR(err
);
2399 err
= filler(data
, page
);
2402 return ERR_PTR(err
);
2405 page
= wait_on_page_read(page
);
2410 if (PageUptodate(page
))
2414 * Page is not up to date and may be locked due one of the following
2415 * case a: Page is being filled and the page lock is held
2416 * case b: Read/write error clearing the page uptodate status
2417 * case c: Truncation in progress (page locked)
2418 * case d: Reclaim in progress
2420 * Case a, the page will be up to date when the page is unlocked.
2421 * There is no need to serialise on the page lock here as the page
2422 * is pinned so the lock gives no additional protection. Even if the
2423 * the page is truncated, the data is still valid if PageUptodate as
2424 * it's a race vs truncate race.
2425 * Case b, the page will not be up to date
2426 * Case c, the page may be truncated but in itself, the data may still
2427 * be valid after IO completes as it's a read vs truncate race. The
2428 * operation must restart if the page is not uptodate on unlock but
2429 * otherwise serialising on page lock to stabilise the mapping gives
2430 * no additional guarantees to the caller as the page lock is
2431 * released before return.
2432 * Case d, similar to truncation. If reclaim holds the page lock, it
2433 * will be a race with remove_mapping that determines if the mapping
2434 * is valid on unlock but otherwise the data is valid and there is
2435 * no need to serialise with page lock.
2437 * As the page lock gives no additional guarantee, we optimistically
2438 * wait on the page to be unlocked and check if it's up to date and
2439 * use the page if it is. Otherwise, the page lock is required to
2440 * distinguish between the different cases. The motivation is that we
2441 * avoid spurious serialisations and wakeups when multiple processes
2442 * wait on the same page for IO to complete.
2444 wait_on_page_locked(page
);
2445 if (PageUptodate(page
))
2448 /* Distinguish between all the cases under the safety of the lock */
2451 /* Case c or d, restart the operation */
2452 if (!page
->mapping
) {
2458 /* Someone else locked and filled the page in a very small window */
2459 if (PageUptodate(page
)) {
2466 mark_page_accessed(page
);
2471 * read_cache_page - read into page cache, fill it if needed
2472 * @mapping: the page's address_space
2473 * @index: the page index
2474 * @filler: function to perform the read
2475 * @data: first arg to filler(data, page) function, often left as NULL
2477 * Read into the page cache. If a page already exists, and PageUptodate() is
2478 * not set, try to fill the page and wait for it to become unlocked.
2480 * If the page does not get brought uptodate, return -EIO.
2482 struct page
*read_cache_page(struct address_space
*mapping
,
2484 int (*filler
)(void *, struct page
*),
2487 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2489 EXPORT_SYMBOL(read_cache_page
);
2492 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2493 * @mapping: the page's address_space
2494 * @index: the page index
2495 * @gfp: the page allocator flags to use if allocating
2497 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2498 * any new page allocations done using the specified allocation flags.
2500 * If the page does not get brought uptodate, return -EIO.
2502 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2506 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2508 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2510 EXPORT_SYMBOL(read_cache_page_gfp
);
2513 * Performs necessary checks before doing a write
2515 * Can adjust writing position or amount of bytes to write.
2516 * Returns appropriate error code that caller should return or
2517 * zero in case that write should be allowed.
2519 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2521 struct file
*file
= iocb
->ki_filp
;
2522 struct inode
*inode
= file
->f_mapping
->host
;
2523 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2526 if (!iov_iter_count(from
))
2529 /* FIXME: this is for backwards compatibility with 2.4 */
2530 if (iocb
->ki_flags
& IOCB_APPEND
)
2531 iocb
->ki_pos
= i_size_read(inode
);
2535 if (limit
!= RLIM_INFINITY
) {
2536 if (iocb
->ki_pos
>= limit
) {
2537 send_sig(SIGXFSZ
, current
, 0);
2540 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2546 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2547 !(file
->f_flags
& O_LARGEFILE
))) {
2548 if (pos
>= MAX_NON_LFS
)
2550 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2554 * Are we about to exceed the fs block limit ?
2556 * If we have written data it becomes a short write. If we have
2557 * exceeded without writing data we send a signal and return EFBIG.
2558 * Linus frestrict idea will clean these up nicely..
2560 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2563 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2564 return iov_iter_count(from
);
2566 EXPORT_SYMBOL(generic_write_checks
);
2568 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2569 loff_t pos
, unsigned len
, unsigned flags
,
2570 struct page
**pagep
, void **fsdata
)
2572 const struct address_space_operations
*aops
= mapping
->a_ops
;
2574 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2577 EXPORT_SYMBOL(pagecache_write_begin
);
2579 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2580 loff_t pos
, unsigned len
, unsigned copied
,
2581 struct page
*page
, void *fsdata
)
2583 const struct address_space_operations
*aops
= mapping
->a_ops
;
2585 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2587 EXPORT_SYMBOL(pagecache_write_end
);
2590 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2592 struct file
*file
= iocb
->ki_filp
;
2593 struct address_space
*mapping
= file
->f_mapping
;
2594 struct inode
*inode
= mapping
->host
;
2595 loff_t pos
= iocb
->ki_pos
;
2599 struct iov_iter data
;
2601 write_len
= iov_iter_count(from
);
2602 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2604 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2609 * After a write we want buffered reads to be sure to go to disk to get
2610 * the new data. We invalidate clean cached page from the region we're
2611 * about to write. We do this *before* the write so that we can return
2612 * without clobbering -EIOCBQUEUED from ->direct_IO().
2614 if (mapping
->nrpages
) {
2615 written
= invalidate_inode_pages2_range(mapping
,
2616 pos
>> PAGE_SHIFT
, end
);
2618 * If a page can not be invalidated, return 0 to fall back
2619 * to buffered write.
2622 if (written
== -EBUSY
)
2629 written
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2632 * Finally, try again to invalidate clean pages which might have been
2633 * cached by non-direct readahead, or faulted in by get_user_pages()
2634 * if the source of the write was an mmap'ed region of the file
2635 * we're writing. Either one is a pretty crazy thing to do,
2636 * so we don't support it 100%. If this invalidation
2637 * fails, tough, the write still worked...
2639 if (mapping
->nrpages
) {
2640 invalidate_inode_pages2_range(mapping
,
2641 pos
>> PAGE_SHIFT
, end
);
2646 iov_iter_advance(from
, written
);
2647 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2648 i_size_write(inode
, pos
);
2649 mark_inode_dirty(inode
);
2656 EXPORT_SYMBOL(generic_file_direct_write
);
2659 * Find or create a page at the given pagecache position. Return the locked
2660 * page. This function is specifically for buffered writes.
2662 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2663 pgoff_t index
, unsigned flags
)
2666 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2668 if (flags
& AOP_FLAG_NOFS
)
2669 fgp_flags
|= FGP_NOFS
;
2671 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2672 mapping_gfp_mask(mapping
));
2674 wait_for_stable_page(page
);
2678 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2680 ssize_t
generic_perform_write(struct file
*file
,
2681 struct iov_iter
*i
, loff_t pos
)
2683 struct address_space
*mapping
= file
->f_mapping
;
2684 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2686 ssize_t written
= 0;
2687 unsigned int flags
= 0;
2690 * Copies from kernel address space cannot fail (NFSD is a big user).
2692 if (!iter_is_iovec(i
))
2693 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2697 unsigned long offset
; /* Offset into pagecache page */
2698 unsigned long bytes
; /* Bytes to write to page */
2699 size_t copied
; /* Bytes copied from user */
2702 offset
= (pos
& (PAGE_SIZE
- 1));
2703 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2708 * Bring in the user page that we will copy from _first_.
2709 * Otherwise there's a nasty deadlock on copying from the
2710 * same page as we're writing to, without it being marked
2713 * Not only is this an optimisation, but it is also required
2714 * to check that the address is actually valid, when atomic
2715 * usercopies are used, below.
2717 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2722 if (fatal_signal_pending(current
)) {
2727 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2729 if (unlikely(status
< 0))
2732 if (mapping_writably_mapped(mapping
))
2733 flush_dcache_page(page
);
2735 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2736 flush_dcache_page(page
);
2738 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2740 if (unlikely(status
< 0))
2746 iov_iter_advance(i
, copied
);
2747 if (unlikely(copied
== 0)) {
2749 * If we were unable to copy any data at all, we must
2750 * fall back to a single segment length write.
2752 * If we didn't fallback here, we could livelock
2753 * because not all segments in the iov can be copied at
2754 * once without a pagefault.
2756 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2757 iov_iter_single_seg_count(i
));
2763 balance_dirty_pages_ratelimited(mapping
);
2764 } while (iov_iter_count(i
));
2766 return written
? written
: status
;
2768 EXPORT_SYMBOL(generic_perform_write
);
2771 * __generic_file_write_iter - write data to a file
2772 * @iocb: IO state structure (file, offset, etc.)
2773 * @from: iov_iter with data to write
2775 * This function does all the work needed for actually writing data to a
2776 * file. It does all basic checks, removes SUID from the file, updates
2777 * modification times and calls proper subroutines depending on whether we
2778 * do direct IO or a standard buffered write.
2780 * It expects i_mutex to be grabbed unless we work on a block device or similar
2781 * object which does not need locking at all.
2783 * This function does *not* take care of syncing data in case of O_SYNC write.
2784 * A caller has to handle it. This is mainly due to the fact that we want to
2785 * avoid syncing under i_mutex.
2787 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2789 struct file
*file
= iocb
->ki_filp
;
2790 struct address_space
* mapping
= file
->f_mapping
;
2791 struct inode
*inode
= mapping
->host
;
2792 ssize_t written
= 0;
2796 /* We can write back this queue in page reclaim */
2797 current
->backing_dev_info
= inode_to_bdi(inode
);
2798 err
= file_remove_privs(file
);
2802 err
= file_update_time(file
);
2806 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2807 loff_t pos
, endbyte
;
2809 written
= generic_file_direct_write(iocb
, from
);
2811 * If the write stopped short of completing, fall back to
2812 * buffered writes. Some filesystems do this for writes to
2813 * holes, for example. For DAX files, a buffered write will
2814 * not succeed (even if it did, DAX does not handle dirty
2815 * page-cache pages correctly).
2817 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2820 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2822 * If generic_perform_write() returned a synchronous error
2823 * then we want to return the number of bytes which were
2824 * direct-written, or the error code if that was zero. Note
2825 * that this differs from normal direct-io semantics, which
2826 * will return -EFOO even if some bytes were written.
2828 if (unlikely(status
< 0)) {
2833 * We need to ensure that the page cache pages are written to
2834 * disk and invalidated to preserve the expected O_DIRECT
2837 endbyte
= pos
+ status
- 1;
2838 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2840 iocb
->ki_pos
= endbyte
+ 1;
2842 invalidate_mapping_pages(mapping
,
2844 endbyte
>> PAGE_SHIFT
);
2847 * We don't know how much we wrote, so just return
2848 * the number of bytes which were direct-written
2852 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2853 if (likely(written
> 0))
2854 iocb
->ki_pos
+= written
;
2857 current
->backing_dev_info
= NULL
;
2858 return written
? written
: err
;
2860 EXPORT_SYMBOL(__generic_file_write_iter
);
2863 * generic_file_write_iter - write data to a file
2864 * @iocb: IO state structure
2865 * @from: iov_iter with data to write
2867 * This is a wrapper around __generic_file_write_iter() to be used by most
2868 * filesystems. It takes care of syncing the file in case of O_SYNC file
2869 * and acquires i_mutex as needed.
2871 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2873 struct file
*file
= iocb
->ki_filp
;
2874 struct inode
*inode
= file
->f_mapping
->host
;
2878 ret
= generic_write_checks(iocb
, from
);
2880 ret
= __generic_file_write_iter(iocb
, from
);
2881 inode_unlock(inode
);
2884 ret
= generic_write_sync(iocb
, ret
);
2887 EXPORT_SYMBOL(generic_file_write_iter
);
2890 * try_to_release_page() - release old fs-specific metadata on a page
2892 * @page: the page which the kernel is trying to free
2893 * @gfp_mask: memory allocation flags (and I/O mode)
2895 * The address_space is to try to release any data against the page
2896 * (presumably at page->private). If the release was successful, return `1'.
2897 * Otherwise return zero.
2899 * This may also be called if PG_fscache is set on a page, indicating that the
2900 * page is known to the local caching routines.
2902 * The @gfp_mask argument specifies whether I/O may be performed to release
2903 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2906 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2908 struct address_space
* const mapping
= page
->mapping
;
2910 BUG_ON(!PageLocked(page
));
2911 if (PageWriteback(page
))
2914 if (mapping
&& mapping
->a_ops
->releasepage
)
2915 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2916 return try_to_free_buffers(page
);
2919 EXPORT_SYMBOL(try_to_release_page
);