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/sched/signal.h>
17 #include <linux/uaccess.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/gfp.h>
22 #include <linux/swap.h>
23 #include <linux/mman.h>
24 #include <linux/pagemap.h>
25 #include <linux/file.h>
26 #include <linux/uio.h>
27 #include <linux/hash.h>
28 #include <linux/writeback.h>
29 #include <linux/backing-dev.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/security.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/hugetlb.h>
36 #include <linux/memcontrol.h>
37 #include <linux/cleancache.h>
38 #include <linux/rmap.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/filemap.h>
45 * FIXME: remove all knowledge of the buffer layer from the core VM
47 #include <linux/buffer_head.h> /* for try_to_free_buffers */
52 * Shared mappings implemented 30.11.1994. It's not fully working yet,
55 * Shared mappings now work. 15.8.1995 Bruno.
57 * finished 'unifying' the page and buffer cache and SMP-threaded the
58 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
60 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
66 * ->i_mmap_rwsem (truncate_pagecache)
67 * ->private_lock (__free_pte->__set_page_dirty_buffers)
68 * ->swap_lock (exclusive_swap_page, others)
69 * ->mapping->tree_lock
72 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
76 * ->page_table_lock or pte_lock (various, mainly in memory.c)
77 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
80 * ->lock_page (access_process_vm)
82 * ->i_mutex (generic_perform_write)
83 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
86 * sb_lock (fs/fs-writeback.c)
87 * ->mapping->tree_lock (__sync_single_inode)
90 * ->anon_vma.lock (vma_adjust)
93 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
95 * ->page_table_lock or pte_lock
96 * ->swap_lock (try_to_unmap_one)
97 * ->private_lock (try_to_unmap_one)
98 * ->tree_lock (try_to_unmap_one)
99 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
100 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
101 * ->private_lock (page_remove_rmap->set_page_dirty)
102 * ->tree_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
104 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
105 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
106 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
107 * ->inode->i_lock (zap_pte_range->set_page_dirty)
108 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
111 * ->tasklist_lock (memory_failure, collect_procs_ao)
114 static int page_cache_tree_insert(struct address_space
*mapping
,
115 struct page
*page
, void **shadowp
)
117 struct radix_tree_node
*node
;
121 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
128 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
129 if (!radix_tree_exceptional_entry(p
))
132 mapping
->nrexceptional
--;
133 if (!dax_mapping(mapping
)) {
137 /* DAX can replace empty locked entry with a hole */
139 dax_radix_locked_entry(0, RADIX_DAX_EMPTY
));
140 /* Wakeup waiters for exceptional entry lock */
141 dax_wake_mapping_entry_waiter(mapping
, page
->index
, p
,
145 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, page
,
146 workingset_update_node
, mapping
);
151 static void page_cache_tree_delete(struct address_space
*mapping
,
152 struct page
*page
, void *shadow
)
156 /* hugetlb pages are represented by one entry in the radix tree */
157 nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
159 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
160 VM_BUG_ON_PAGE(PageTail(page
), page
);
161 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
163 for (i
= 0; i
< nr
; i
++) {
164 struct radix_tree_node
*node
;
167 __radix_tree_lookup(&mapping
->page_tree
, page
->index
+ i
,
170 VM_BUG_ON_PAGE(!node
&& nr
!= 1, page
);
172 radix_tree_clear_tags(&mapping
->page_tree
, node
, slot
);
173 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, shadow
,
174 workingset_update_node
, mapping
);
178 mapping
->nrexceptional
+= nr
;
180 * Make sure the nrexceptional update is committed before
181 * the nrpages update so that final truncate racing
182 * with reclaim does not see both counters 0 at the
183 * same time and miss a shadow entry.
187 mapping
->nrpages
-= nr
;
191 * Delete a page from the page cache and free it. Caller has to make
192 * sure the page is locked and that nobody else uses it - or that usage
193 * is safe. The caller must hold the mapping's tree_lock.
195 void __delete_from_page_cache(struct page
*page
, void *shadow
)
197 struct address_space
*mapping
= page
->mapping
;
198 int nr
= hpage_nr_pages(page
);
200 trace_mm_filemap_delete_from_page_cache(page
);
202 * if we're uptodate, flush out into the cleancache, otherwise
203 * invalidate any existing cleancache entries. We can't leave
204 * stale data around in the cleancache once our page is gone
206 if (PageUptodate(page
) && PageMappedToDisk(page
))
207 cleancache_put_page(page
);
209 cleancache_invalidate_page(mapping
, page
);
211 VM_BUG_ON_PAGE(PageTail(page
), page
);
212 VM_BUG_ON_PAGE(page_mapped(page
), page
);
213 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
216 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
217 current
->comm
, page_to_pfn(page
));
218 dump_page(page
, "still mapped when deleted");
220 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
222 mapcount
= page_mapcount(page
);
223 if (mapping_exiting(mapping
) &&
224 page_count(page
) >= mapcount
+ 2) {
226 * All vmas have already been torn down, so it's
227 * a good bet that actually the page is unmapped,
228 * and we'd prefer not to leak it: if we're wrong,
229 * some other bad page check should catch it later.
231 page_mapcount_reset(page
);
232 page_ref_sub(page
, mapcount
);
236 page_cache_tree_delete(mapping
, page
, shadow
);
238 page
->mapping
= NULL
;
239 /* Leave page->index set: truncation lookup relies upon it */
241 /* hugetlb pages do not participate in page cache accounting. */
243 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
244 if (PageSwapBacked(page
)) {
245 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
246 if (PageTransHuge(page
))
247 __dec_node_page_state(page
, NR_SHMEM_THPS
);
249 VM_BUG_ON_PAGE(PageTransHuge(page
) && !PageHuge(page
), page
);
253 * At this point page must be either written or cleaned by truncate.
254 * Dirty page here signals a bug and loss of unwritten data.
256 * This fixes dirty accounting after removing the page entirely but
257 * leaves PageDirty set: it has no effect for truncated page and
258 * anyway will be cleared before returning page into buddy allocator.
260 if (WARN_ON_ONCE(PageDirty(page
)))
261 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
265 * delete_from_page_cache - delete page from page cache
266 * @page: the page which the kernel is trying to remove from page cache
268 * This must be called only on pages that have been verified to be in the page
269 * cache and locked. It will never put the page into the free list, the caller
270 * has a reference on the page.
272 void delete_from_page_cache(struct page
*page
)
274 struct address_space
*mapping
= page_mapping(page
);
276 void (*freepage
)(struct page
*);
278 BUG_ON(!PageLocked(page
));
280 freepage
= mapping
->a_ops
->freepage
;
282 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
283 __delete_from_page_cache(page
, NULL
);
284 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
289 if (PageTransHuge(page
) && !PageHuge(page
)) {
290 page_ref_sub(page
, HPAGE_PMD_NR
);
291 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
296 EXPORT_SYMBOL(delete_from_page_cache
);
298 int filemap_check_errors(struct address_space
*mapping
)
301 /* Check for outstanding write errors */
302 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
303 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
305 if (test_bit(AS_EIO
, &mapping
->flags
) &&
306 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
310 EXPORT_SYMBOL(filemap_check_errors
);
313 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
314 * @mapping: address space structure to write
315 * @start: offset in bytes where the range starts
316 * @end: offset in bytes where the range ends (inclusive)
317 * @sync_mode: enable synchronous operation
319 * Start writeback against all of a mapping's dirty pages that lie
320 * within the byte offsets <start, end> inclusive.
322 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
323 * opposed to a regular memory cleansing writeback. The difference between
324 * these two operations is that if a dirty page/buffer is encountered, it must
325 * be waited upon, and not just skipped over.
327 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
328 loff_t end
, int sync_mode
)
331 struct writeback_control wbc
= {
332 .sync_mode
= sync_mode
,
333 .nr_to_write
= LONG_MAX
,
334 .range_start
= start
,
338 if (!mapping_cap_writeback_dirty(mapping
))
341 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
342 ret
= do_writepages(mapping
, &wbc
);
343 wbc_detach_inode(&wbc
);
347 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
350 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
353 int filemap_fdatawrite(struct address_space
*mapping
)
355 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
357 EXPORT_SYMBOL(filemap_fdatawrite
);
359 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
362 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
364 EXPORT_SYMBOL(filemap_fdatawrite_range
);
367 * filemap_flush - mostly a non-blocking flush
368 * @mapping: target address_space
370 * This is a mostly non-blocking flush. Not suitable for data-integrity
371 * purposes - I/O may not be started against all dirty pages.
373 int filemap_flush(struct address_space
*mapping
)
375 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
377 EXPORT_SYMBOL(filemap_flush
);
380 * filemap_range_has_page - check if a page exists in range.
381 * @mapping: address space within which to check
382 * @start_byte: offset in bytes where the range starts
383 * @end_byte: offset in bytes where the range ends (inclusive)
385 * Find at least one page in the range supplied, usually used to check if
386 * direct writing in this range will trigger a writeback.
388 bool filemap_range_has_page(struct address_space
*mapping
,
389 loff_t start_byte
, loff_t end_byte
)
391 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
392 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
396 if (end_byte
< start_byte
)
399 if (mapping
->nrpages
== 0)
402 pagevec_init(&pvec
, 0);
403 if (!pagevec_lookup(&pvec
, mapping
, index
, 1))
405 ret
= (pvec
.pages
[0]->index
<= end
);
406 pagevec_release(&pvec
);
409 EXPORT_SYMBOL(filemap_range_has_page
);
411 static int __filemap_fdatawait_range(struct address_space
*mapping
,
412 loff_t start_byte
, loff_t end_byte
)
414 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
415 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
420 if (end_byte
< start_byte
)
423 pagevec_init(&pvec
, 0);
424 while ((index
<= end
) &&
425 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
426 PAGECACHE_TAG_WRITEBACK
,
427 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
430 for (i
= 0; i
< nr_pages
; i
++) {
431 struct page
*page
= pvec
.pages
[i
];
433 /* until radix tree lookup accepts end_index */
434 if (page
->index
> end
)
437 wait_on_page_writeback(page
);
438 if (TestClearPageError(page
))
441 pagevec_release(&pvec
);
449 * filemap_fdatawait_range - wait for writeback to complete
450 * @mapping: address space structure to wait for
451 * @start_byte: offset in bytes where the range starts
452 * @end_byte: offset in bytes where the range ends (inclusive)
454 * Walk the list of under-writeback pages of the given address space
455 * in the given range and wait for all of them. Check error status of
456 * the address space and return it.
458 * Since the error status of the address space is cleared by this function,
459 * callers are responsible for checking the return value and handling and/or
460 * reporting the error.
462 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
467 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
468 ret2
= filemap_check_errors(mapping
);
474 EXPORT_SYMBOL(filemap_fdatawait_range
);
477 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
478 * @mapping: address space structure to wait for
480 * Walk the list of under-writeback pages of the given address space
481 * and wait for all of them. Unlike filemap_fdatawait(), this function
482 * does not clear error status of the address space.
484 * Use this function if callers don't handle errors themselves. Expected
485 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
488 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
490 loff_t i_size
= i_size_read(mapping
->host
);
495 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
499 * filemap_fdatawait - wait for all under-writeback pages to complete
500 * @mapping: address space structure to wait for
502 * Walk the list of under-writeback pages of the given address space
503 * and wait for all of them. Check error status of the address space
506 * Since the error status of the address space is cleared by this function,
507 * callers are responsible for checking the return value and handling and/or
508 * reporting the error.
510 int filemap_fdatawait(struct address_space
*mapping
)
512 loff_t i_size
= i_size_read(mapping
->host
);
517 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
519 EXPORT_SYMBOL(filemap_fdatawait
);
521 int filemap_write_and_wait(struct address_space
*mapping
)
525 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
526 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
527 err
= filemap_fdatawrite(mapping
);
529 * Even if the above returned error, the pages may be
530 * written partially (e.g. -ENOSPC), so we wait for it.
531 * But the -EIO is special case, it may indicate the worst
532 * thing (e.g. bug) happened, so we avoid waiting for it.
535 int err2
= filemap_fdatawait(mapping
);
540 err
= filemap_check_errors(mapping
);
544 EXPORT_SYMBOL(filemap_write_and_wait
);
547 * filemap_write_and_wait_range - write out & wait on a file range
548 * @mapping: the address_space for the pages
549 * @lstart: offset in bytes where the range starts
550 * @lend: offset in bytes where the range ends (inclusive)
552 * Write out and wait upon file offsets lstart->lend, inclusive.
554 * Note that @lend is inclusive (describes the last byte to be written) so
555 * that this function can be used to write to the very end-of-file (end = -1).
557 int filemap_write_and_wait_range(struct address_space
*mapping
,
558 loff_t lstart
, loff_t lend
)
562 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
563 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
564 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
566 /* See comment of filemap_write_and_wait() */
568 int err2
= filemap_fdatawait_range(mapping
,
574 err
= filemap_check_errors(mapping
);
578 EXPORT_SYMBOL(filemap_write_and_wait_range
);
581 * replace_page_cache_page - replace a pagecache page with a new one
582 * @old: page to be replaced
583 * @new: page to replace with
584 * @gfp_mask: allocation mode
586 * This function replaces a page in the pagecache with a new one. On
587 * success it acquires the pagecache reference for the new page and
588 * drops it for the old page. Both the old and new pages must be
589 * locked. This function does not add the new page to the LRU, the
590 * caller must do that.
592 * The remove + add is atomic. The only way this function can fail is
593 * memory allocation failure.
595 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
599 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
600 VM_BUG_ON_PAGE(!PageLocked(new), new);
601 VM_BUG_ON_PAGE(new->mapping
, new);
603 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
605 struct address_space
*mapping
= old
->mapping
;
606 void (*freepage
)(struct page
*);
609 pgoff_t offset
= old
->index
;
610 freepage
= mapping
->a_ops
->freepage
;
613 new->mapping
= mapping
;
616 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
617 __delete_from_page_cache(old
, NULL
);
618 error
= page_cache_tree_insert(mapping
, new, NULL
);
622 * hugetlb pages do not participate in page cache accounting.
625 __inc_node_page_state(new, NR_FILE_PAGES
);
626 if (PageSwapBacked(new))
627 __inc_node_page_state(new, NR_SHMEM
);
628 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
629 mem_cgroup_migrate(old
, new);
630 radix_tree_preload_end();
638 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
640 static int __add_to_page_cache_locked(struct page
*page
,
641 struct address_space
*mapping
,
642 pgoff_t offset
, gfp_t gfp_mask
,
645 int huge
= PageHuge(page
);
646 struct mem_cgroup
*memcg
;
649 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
650 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
653 error
= mem_cgroup_try_charge(page
, current
->mm
,
654 gfp_mask
, &memcg
, false);
659 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
662 mem_cgroup_cancel_charge(page
, memcg
, false);
667 page
->mapping
= mapping
;
668 page
->index
= offset
;
670 spin_lock_irq(&mapping
->tree_lock
);
671 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
672 radix_tree_preload_end();
676 /* hugetlb pages do not participate in page cache accounting. */
678 __inc_node_page_state(page
, NR_FILE_PAGES
);
679 spin_unlock_irq(&mapping
->tree_lock
);
681 mem_cgroup_commit_charge(page
, memcg
, false, false);
682 trace_mm_filemap_add_to_page_cache(page
);
685 page
->mapping
= NULL
;
686 /* Leave page->index set: truncation relies upon it */
687 spin_unlock_irq(&mapping
->tree_lock
);
689 mem_cgroup_cancel_charge(page
, memcg
, false);
695 * add_to_page_cache_locked - add a locked page to the pagecache
697 * @mapping: the page's address_space
698 * @offset: page index
699 * @gfp_mask: page allocation mode
701 * This function is used to add a page to the pagecache. It must be locked.
702 * This function does not add the page to the LRU. The caller must do that.
704 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
705 pgoff_t offset
, gfp_t gfp_mask
)
707 return __add_to_page_cache_locked(page
, mapping
, offset
,
710 EXPORT_SYMBOL(add_to_page_cache_locked
);
712 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
713 pgoff_t offset
, gfp_t gfp_mask
)
718 __SetPageLocked(page
);
719 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
722 __ClearPageLocked(page
);
725 * The page might have been evicted from cache only
726 * recently, in which case it should be activated like
727 * any other repeatedly accessed page.
728 * The exception is pages getting rewritten; evicting other
729 * data from the working set, only to cache data that will
730 * get overwritten with something else, is a waste of memory.
732 if (!(gfp_mask
& __GFP_WRITE
) &&
733 shadow
&& workingset_refault(shadow
)) {
735 workingset_activation(page
);
737 ClearPageActive(page
);
742 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
745 struct page
*__page_cache_alloc(gfp_t gfp
)
750 if (cpuset_do_page_mem_spread()) {
751 unsigned int cpuset_mems_cookie
;
753 cpuset_mems_cookie
= read_mems_allowed_begin();
754 n
= cpuset_mem_spread_node();
755 page
= __alloc_pages_node(n
, gfp
, 0);
756 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
760 return alloc_pages(gfp
, 0);
762 EXPORT_SYMBOL(__page_cache_alloc
);
766 * In order to wait for pages to become available there must be
767 * waitqueues associated with pages. By using a hash table of
768 * waitqueues where the bucket discipline is to maintain all
769 * waiters on the same queue and wake all when any of the pages
770 * become available, and for the woken contexts to check to be
771 * sure the appropriate page became available, this saves space
772 * at a cost of "thundering herd" phenomena during rare hash
775 #define PAGE_WAIT_TABLE_BITS 8
776 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
777 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
779 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
781 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
784 void __init
pagecache_init(void)
788 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
789 init_waitqueue_head(&page_wait_table
[i
]);
791 page_writeback_init();
794 struct wait_page_key
{
800 struct wait_page_queue
{
803 wait_queue_entry_t wait
;
806 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
808 struct wait_page_key
*key
= arg
;
809 struct wait_page_queue
*wait_page
810 = container_of(wait
, struct wait_page_queue
, wait
);
812 if (wait_page
->page
!= key
->page
)
816 if (wait_page
->bit_nr
!= key
->bit_nr
)
818 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
821 return autoremove_wake_function(wait
, mode
, sync
, key
);
824 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
826 wait_queue_head_t
*q
= page_waitqueue(page
);
827 struct wait_page_key key
;
834 spin_lock_irqsave(&q
->lock
, flags
);
835 __wake_up_locked_key(q
, TASK_NORMAL
, &key
);
837 * It is possible for other pages to have collided on the waitqueue
838 * hash, so in that case check for a page match. That prevents a long-
841 * It is still possible to miss a case here, when we woke page waiters
842 * and removed them from the waitqueue, but there are still other
845 if (!waitqueue_active(q
) || !key
.page_match
) {
846 ClearPageWaiters(page
);
848 * It's possible to miss clearing Waiters here, when we woke
849 * our page waiters, but the hashed waitqueue has waiters for
852 * That's okay, it's a rare case. The next waker will clear it.
855 spin_unlock_irqrestore(&q
->lock
, flags
);
858 static void wake_up_page(struct page
*page
, int bit
)
860 if (!PageWaiters(page
))
862 wake_up_page_bit(page
, bit
);
865 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
866 struct page
*page
, int bit_nr
, int state
, bool lock
)
868 struct wait_page_queue wait_page
;
869 wait_queue_entry_t
*wait
= &wait_page
.wait
;
873 wait
->func
= wake_page_function
;
874 wait_page
.page
= page
;
875 wait_page
.bit_nr
= bit_nr
;
878 spin_lock_irq(&q
->lock
);
880 if (likely(list_empty(&wait
->entry
))) {
882 __add_wait_queue_entry_tail_exclusive(q
, wait
);
884 __add_wait_queue(q
, wait
);
885 SetPageWaiters(page
);
888 set_current_state(state
);
890 spin_unlock_irq(&q
->lock
);
892 if (likely(test_bit(bit_nr
, &page
->flags
))) {
894 if (unlikely(signal_pending_state(state
, current
))) {
901 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
904 if (!test_bit(bit_nr
, &page
->flags
))
909 finish_wait(q
, wait
);
912 * A signal could leave PageWaiters set. Clearing it here if
913 * !waitqueue_active would be possible (by open-coding finish_wait),
914 * but still fail to catch it in the case of wait hash collision. We
915 * already can fail to clear wait hash collision cases, so don't
916 * bother with signals either.
922 void wait_on_page_bit(struct page
*page
, int bit_nr
)
924 wait_queue_head_t
*q
= page_waitqueue(page
);
925 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, false);
927 EXPORT_SYMBOL(wait_on_page_bit
);
929 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
931 wait_queue_head_t
*q
= page_waitqueue(page
);
932 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, false);
936 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
937 * @page: Page defining the wait queue of interest
938 * @waiter: Waiter to add to the queue
940 * Add an arbitrary @waiter to the wait queue for the nominated @page.
942 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
944 wait_queue_head_t
*q
= page_waitqueue(page
);
947 spin_lock_irqsave(&q
->lock
, flags
);
948 __add_wait_queue(q
, waiter
);
949 SetPageWaiters(page
);
950 spin_unlock_irqrestore(&q
->lock
, flags
);
952 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
954 #ifndef clear_bit_unlock_is_negative_byte
957 * PG_waiters is the high bit in the same byte as PG_lock.
959 * On x86 (and on many other architectures), we can clear PG_lock and
960 * test the sign bit at the same time. But if the architecture does
961 * not support that special operation, we just do this all by hand
964 * The read of PG_waiters has to be after (or concurrently with) PG_locked
965 * being cleared, but a memory barrier should be unneccssary since it is
966 * in the same byte as PG_locked.
968 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
970 clear_bit_unlock(nr
, mem
);
971 /* smp_mb__after_atomic(); */
972 return test_bit(PG_waiters
, mem
);
978 * unlock_page - unlock a locked page
981 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
982 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
983 * mechanism between PageLocked pages and PageWriteback pages is shared.
984 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
986 * Note that this depends on PG_waiters being the sign bit in the byte
987 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
988 * clear the PG_locked bit and test PG_waiters at the same time fairly
989 * portably (architectures that do LL/SC can test any bit, while x86 can
990 * test the sign bit).
992 void unlock_page(struct page
*page
)
994 BUILD_BUG_ON(PG_waiters
!= 7);
995 page
= compound_head(page
);
996 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
997 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
998 wake_up_page_bit(page
, PG_locked
);
1000 EXPORT_SYMBOL(unlock_page
);
1003 * end_page_writeback - end writeback against a page
1006 void end_page_writeback(struct page
*page
)
1009 * TestClearPageReclaim could be used here but it is an atomic
1010 * operation and overkill in this particular case. Failing to
1011 * shuffle a page marked for immediate reclaim is too mild to
1012 * justify taking an atomic operation penalty at the end of
1013 * ever page writeback.
1015 if (PageReclaim(page
)) {
1016 ClearPageReclaim(page
);
1017 rotate_reclaimable_page(page
);
1020 if (!test_clear_page_writeback(page
))
1023 smp_mb__after_atomic();
1024 wake_up_page(page
, PG_writeback
);
1026 EXPORT_SYMBOL(end_page_writeback
);
1029 * After completing I/O on a page, call this routine to update the page
1030 * flags appropriately
1032 void page_endio(struct page
*page
, bool is_write
, int err
)
1036 SetPageUptodate(page
);
1038 ClearPageUptodate(page
);
1044 struct address_space
*mapping
;
1047 mapping
= page_mapping(page
);
1049 mapping_set_error(mapping
, err
);
1051 end_page_writeback(page
);
1054 EXPORT_SYMBOL_GPL(page_endio
);
1057 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1058 * @__page: the page to lock
1060 void __lock_page(struct page
*__page
)
1062 struct page
*page
= compound_head(__page
);
1063 wait_queue_head_t
*q
= page_waitqueue(page
);
1064 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, true);
1066 EXPORT_SYMBOL(__lock_page
);
1068 int __lock_page_killable(struct page
*__page
)
1070 struct page
*page
= compound_head(__page
);
1071 wait_queue_head_t
*q
= page_waitqueue(page
);
1072 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
, true);
1074 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1078 * 1 - page is locked; mmap_sem is still held.
1079 * 0 - page is not locked.
1080 * mmap_sem has been released (up_read()), unless flags had both
1081 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1082 * which case mmap_sem is still held.
1084 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1085 * with the page locked and the mmap_sem unperturbed.
1087 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1090 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1092 * CAUTION! In this case, mmap_sem is not released
1093 * even though return 0.
1095 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1098 up_read(&mm
->mmap_sem
);
1099 if (flags
& FAULT_FLAG_KILLABLE
)
1100 wait_on_page_locked_killable(page
);
1102 wait_on_page_locked(page
);
1105 if (flags
& FAULT_FLAG_KILLABLE
) {
1108 ret
= __lock_page_killable(page
);
1110 up_read(&mm
->mmap_sem
);
1120 * page_cache_next_hole - find the next hole (not-present entry)
1123 * @max_scan: maximum range to search
1125 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1126 * lowest indexed hole.
1128 * Returns: the index of the hole if found, otherwise returns an index
1129 * outside of the set specified (in which case 'return - index >=
1130 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1133 * page_cache_next_hole may be called under rcu_read_lock. However,
1134 * like radix_tree_gang_lookup, this will not atomically search a
1135 * snapshot of the tree at a single point in time. For example, if a
1136 * hole is created at index 5, then subsequently a hole is created at
1137 * index 10, page_cache_next_hole covering both indexes may return 10
1138 * if called under rcu_read_lock.
1140 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1141 pgoff_t index
, unsigned long max_scan
)
1145 for (i
= 0; i
< max_scan
; i
++) {
1148 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1149 if (!page
|| radix_tree_exceptional_entry(page
))
1158 EXPORT_SYMBOL(page_cache_next_hole
);
1161 * page_cache_prev_hole - find the prev hole (not-present entry)
1164 * @max_scan: maximum range to search
1166 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1169 * Returns: the index of the hole if found, otherwise returns an index
1170 * outside of the set specified (in which case 'index - return >=
1171 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1174 * page_cache_prev_hole may be called under rcu_read_lock. However,
1175 * like radix_tree_gang_lookup, this will not atomically search a
1176 * snapshot of the tree at a single point in time. For example, if a
1177 * hole is created at index 10, then subsequently a hole is created at
1178 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1179 * called under rcu_read_lock.
1181 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1182 pgoff_t index
, unsigned long max_scan
)
1186 for (i
= 0; i
< max_scan
; i
++) {
1189 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1190 if (!page
|| radix_tree_exceptional_entry(page
))
1193 if (index
== ULONG_MAX
)
1199 EXPORT_SYMBOL(page_cache_prev_hole
);
1202 * find_get_entry - find and get a page cache entry
1203 * @mapping: the address_space to search
1204 * @offset: the page cache index
1206 * Looks up the page cache slot at @mapping & @offset. If there is a
1207 * page cache page, it is returned with an increased refcount.
1209 * If the slot holds a shadow entry of a previously evicted page, or a
1210 * swap entry from shmem/tmpfs, it is returned.
1212 * Otherwise, %NULL is returned.
1214 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1217 struct page
*head
, *page
;
1222 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1224 page
= radix_tree_deref_slot(pagep
);
1225 if (unlikely(!page
))
1227 if (radix_tree_exception(page
)) {
1228 if (radix_tree_deref_retry(page
))
1231 * A shadow entry of a recently evicted page,
1232 * or a swap entry from shmem/tmpfs. Return
1233 * it without attempting to raise page count.
1238 head
= compound_head(page
);
1239 if (!page_cache_get_speculative(head
))
1242 /* The page was split under us? */
1243 if (compound_head(page
) != head
) {
1249 * Has the page moved?
1250 * This is part of the lockless pagecache protocol. See
1251 * include/linux/pagemap.h for details.
1253 if (unlikely(page
!= *pagep
)) {
1263 EXPORT_SYMBOL(find_get_entry
);
1266 * find_lock_entry - locate, pin and lock a page cache entry
1267 * @mapping: the address_space to search
1268 * @offset: the page cache index
1270 * Looks up the page cache slot at @mapping & @offset. If there is a
1271 * page cache page, it is returned locked and with an increased
1274 * If the slot holds a shadow entry of a previously evicted page, or a
1275 * swap entry from shmem/tmpfs, it is returned.
1277 * Otherwise, %NULL is returned.
1279 * find_lock_entry() may sleep.
1281 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1286 page
= find_get_entry(mapping
, offset
);
1287 if (page
&& !radix_tree_exception(page
)) {
1289 /* Has the page been truncated? */
1290 if (unlikely(page_mapping(page
) != mapping
)) {
1295 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1299 EXPORT_SYMBOL(find_lock_entry
);
1302 * pagecache_get_page - find and get a page reference
1303 * @mapping: the address_space to search
1304 * @offset: the page index
1305 * @fgp_flags: PCG flags
1306 * @gfp_mask: gfp mask to use for the page cache data page allocation
1308 * Looks up the page cache slot at @mapping & @offset.
1310 * PCG flags modify how the page is returned.
1312 * @fgp_flags can be:
1314 * - FGP_ACCESSED: the page will be marked accessed
1315 * - FGP_LOCK: Page is return locked
1316 * - FGP_CREAT: If page is not present then a new page is allocated using
1317 * @gfp_mask and added to the page cache and the VM's LRU
1318 * list. The page is returned locked and with an increased
1319 * refcount. Otherwise, NULL is returned.
1321 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1322 * if the GFP flags specified for FGP_CREAT are atomic.
1324 * If there is a page cache page, it is returned with an increased refcount.
1326 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1327 int fgp_flags
, gfp_t gfp_mask
)
1332 page
= find_get_entry(mapping
, offset
);
1333 if (radix_tree_exceptional_entry(page
))
1338 if (fgp_flags
& FGP_LOCK
) {
1339 if (fgp_flags
& FGP_NOWAIT
) {
1340 if (!trylock_page(page
)) {
1348 /* Has the page been truncated? */
1349 if (unlikely(page
->mapping
!= mapping
)) {
1354 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1357 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1358 mark_page_accessed(page
);
1361 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1363 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1364 gfp_mask
|= __GFP_WRITE
;
1365 if (fgp_flags
& FGP_NOFS
)
1366 gfp_mask
&= ~__GFP_FS
;
1368 page
= __page_cache_alloc(gfp_mask
);
1372 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1373 fgp_flags
|= FGP_LOCK
;
1375 /* Init accessed so avoid atomic mark_page_accessed later */
1376 if (fgp_flags
& FGP_ACCESSED
)
1377 __SetPageReferenced(page
);
1379 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1380 gfp_mask
& GFP_RECLAIM_MASK
);
1381 if (unlikely(err
)) {
1391 EXPORT_SYMBOL(pagecache_get_page
);
1394 * find_get_entries - gang pagecache lookup
1395 * @mapping: The address_space to search
1396 * @start: The starting page cache index
1397 * @nr_entries: The maximum number of entries
1398 * @entries: Where the resulting entries are placed
1399 * @indices: The cache indices corresponding to the entries in @entries
1401 * find_get_entries() will search for and return a group of up to
1402 * @nr_entries entries in the mapping. The entries are placed at
1403 * @entries. find_get_entries() takes a reference against any actual
1406 * The search returns a group of mapping-contiguous page cache entries
1407 * with ascending indexes. There may be holes in the indices due to
1408 * not-present pages.
1410 * Any shadow entries of evicted pages, or swap entries from
1411 * shmem/tmpfs, are included in the returned array.
1413 * find_get_entries() returns the number of pages and shadow entries
1416 unsigned find_get_entries(struct address_space
*mapping
,
1417 pgoff_t start
, unsigned int nr_entries
,
1418 struct page
**entries
, pgoff_t
*indices
)
1421 unsigned int ret
= 0;
1422 struct radix_tree_iter iter
;
1428 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1429 struct page
*head
, *page
;
1431 page
= radix_tree_deref_slot(slot
);
1432 if (unlikely(!page
))
1434 if (radix_tree_exception(page
)) {
1435 if (radix_tree_deref_retry(page
)) {
1436 slot
= radix_tree_iter_retry(&iter
);
1440 * A shadow entry of a recently evicted page, a swap
1441 * entry from shmem/tmpfs or a DAX entry. Return it
1442 * without attempting to raise page count.
1447 head
= compound_head(page
);
1448 if (!page_cache_get_speculative(head
))
1451 /* The page was split under us? */
1452 if (compound_head(page
) != head
) {
1457 /* Has the page moved? */
1458 if (unlikely(page
!= *slot
)) {
1463 indices
[ret
] = iter
.index
;
1464 entries
[ret
] = page
;
1465 if (++ret
== nr_entries
)
1473 * find_get_pages - gang pagecache lookup
1474 * @mapping: The address_space to search
1475 * @start: The starting page index
1476 * @nr_pages: The maximum number of pages
1477 * @pages: Where the resulting pages are placed
1479 * find_get_pages() will search for and return a group of up to
1480 * @nr_pages pages in the mapping. The pages are placed at @pages.
1481 * find_get_pages() takes a reference against the returned pages.
1483 * The search returns a group of mapping-contiguous pages with ascending
1484 * indexes. There may be holes in the indices due to not-present pages.
1486 * find_get_pages() returns the number of pages which were found.
1488 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1489 unsigned int nr_pages
, struct page
**pages
)
1491 struct radix_tree_iter iter
;
1495 if (unlikely(!nr_pages
))
1499 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1500 struct page
*head
, *page
;
1502 page
= radix_tree_deref_slot(slot
);
1503 if (unlikely(!page
))
1506 if (radix_tree_exception(page
)) {
1507 if (radix_tree_deref_retry(page
)) {
1508 slot
= radix_tree_iter_retry(&iter
);
1512 * A shadow entry of a recently evicted page,
1513 * or a swap entry from shmem/tmpfs. Skip
1519 head
= compound_head(page
);
1520 if (!page_cache_get_speculative(head
))
1523 /* The page was split under us? */
1524 if (compound_head(page
) != head
) {
1529 /* Has the page moved? */
1530 if (unlikely(page
!= *slot
)) {
1536 if (++ret
== nr_pages
)
1545 * find_get_pages_contig - gang contiguous pagecache lookup
1546 * @mapping: The address_space to search
1547 * @index: The starting page index
1548 * @nr_pages: The maximum number of pages
1549 * @pages: Where the resulting pages are placed
1551 * find_get_pages_contig() works exactly like find_get_pages(), except
1552 * that the returned number of pages are guaranteed to be contiguous.
1554 * find_get_pages_contig() returns the number of pages which were found.
1556 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1557 unsigned int nr_pages
, struct page
**pages
)
1559 struct radix_tree_iter iter
;
1561 unsigned int ret
= 0;
1563 if (unlikely(!nr_pages
))
1567 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1568 struct page
*head
, *page
;
1570 page
= radix_tree_deref_slot(slot
);
1571 /* The hole, there no reason to continue */
1572 if (unlikely(!page
))
1575 if (radix_tree_exception(page
)) {
1576 if (radix_tree_deref_retry(page
)) {
1577 slot
= radix_tree_iter_retry(&iter
);
1581 * A shadow entry of a recently evicted page,
1582 * or a swap entry from shmem/tmpfs. Stop
1583 * looking for contiguous pages.
1588 head
= compound_head(page
);
1589 if (!page_cache_get_speculative(head
))
1592 /* The page was split under us? */
1593 if (compound_head(page
) != head
) {
1598 /* Has the page moved? */
1599 if (unlikely(page
!= *slot
)) {
1605 * must check mapping and index after taking the ref.
1606 * otherwise we can get both false positives and false
1607 * negatives, which is just confusing to the caller.
1609 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1615 if (++ret
== nr_pages
)
1621 EXPORT_SYMBOL(find_get_pages_contig
);
1624 * find_get_pages_tag - find and return pages that match @tag
1625 * @mapping: the address_space to search
1626 * @index: the starting page index
1627 * @tag: the tag index
1628 * @nr_pages: the maximum number of pages
1629 * @pages: where the resulting pages are placed
1631 * Like find_get_pages, except we only return pages which are tagged with
1632 * @tag. We update @index to index the next page for the traversal.
1634 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1635 int tag
, unsigned int nr_pages
, struct page
**pages
)
1637 struct radix_tree_iter iter
;
1641 if (unlikely(!nr_pages
))
1645 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1646 &iter
, *index
, tag
) {
1647 struct page
*head
, *page
;
1649 page
= radix_tree_deref_slot(slot
);
1650 if (unlikely(!page
))
1653 if (radix_tree_exception(page
)) {
1654 if (radix_tree_deref_retry(page
)) {
1655 slot
= radix_tree_iter_retry(&iter
);
1659 * A shadow entry of a recently evicted page.
1661 * Those entries should never be tagged, but
1662 * this tree walk is lockless and the tags are
1663 * looked up in bulk, one radix tree node at a
1664 * time, so there is a sizable window for page
1665 * reclaim to evict a page we saw tagged.
1672 head
= compound_head(page
);
1673 if (!page_cache_get_speculative(head
))
1676 /* The page was split under us? */
1677 if (compound_head(page
) != head
) {
1682 /* Has the page moved? */
1683 if (unlikely(page
!= *slot
)) {
1689 if (++ret
== nr_pages
)
1696 *index
= pages
[ret
- 1]->index
+ 1;
1700 EXPORT_SYMBOL(find_get_pages_tag
);
1703 * find_get_entries_tag - find and return entries that match @tag
1704 * @mapping: the address_space to search
1705 * @start: the starting page cache index
1706 * @tag: the tag index
1707 * @nr_entries: the maximum number of entries
1708 * @entries: where the resulting entries are placed
1709 * @indices: the cache indices corresponding to the entries in @entries
1711 * Like find_get_entries, except we only return entries which are tagged with
1714 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1715 int tag
, unsigned int nr_entries
,
1716 struct page
**entries
, pgoff_t
*indices
)
1719 unsigned int ret
= 0;
1720 struct radix_tree_iter iter
;
1726 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1727 &iter
, start
, tag
) {
1728 struct page
*head
, *page
;
1730 page
= radix_tree_deref_slot(slot
);
1731 if (unlikely(!page
))
1733 if (radix_tree_exception(page
)) {
1734 if (radix_tree_deref_retry(page
)) {
1735 slot
= radix_tree_iter_retry(&iter
);
1740 * A shadow entry of a recently evicted page, a swap
1741 * entry from shmem/tmpfs or a DAX entry. Return it
1742 * without attempting to raise page count.
1747 head
= compound_head(page
);
1748 if (!page_cache_get_speculative(head
))
1751 /* The page was split under us? */
1752 if (compound_head(page
) != head
) {
1757 /* Has the page moved? */
1758 if (unlikely(page
!= *slot
)) {
1763 indices
[ret
] = iter
.index
;
1764 entries
[ret
] = page
;
1765 if (++ret
== nr_entries
)
1771 EXPORT_SYMBOL(find_get_entries_tag
);
1774 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1775 * a _large_ part of the i/o request. Imagine the worst scenario:
1777 * ---R__________________________________________B__________
1778 * ^ reading here ^ bad block(assume 4k)
1780 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1781 * => failing the whole request => read(R) => read(R+1) =>
1782 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1783 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1784 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1786 * It is going insane. Fix it by quickly scaling down the readahead size.
1788 static void shrink_readahead_size_eio(struct file
*filp
,
1789 struct file_ra_state
*ra
)
1795 * do_generic_file_read - generic file read routine
1796 * @filp: the file to read
1797 * @ppos: current file position
1798 * @iter: data destination
1799 * @written: already copied
1801 * This is a generic file read routine, and uses the
1802 * mapping->a_ops->readpage() function for the actual low-level stuff.
1804 * This is really ugly. But the goto's actually try to clarify some
1805 * of the logic when it comes to error handling etc.
1807 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1808 struct iov_iter
*iter
, ssize_t written
)
1810 struct address_space
*mapping
= filp
->f_mapping
;
1811 struct inode
*inode
= mapping
->host
;
1812 struct file_ra_state
*ra
= &filp
->f_ra
;
1816 unsigned long offset
; /* offset into pagecache page */
1817 unsigned int prev_offset
;
1820 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
1822 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
1824 index
= *ppos
>> PAGE_SHIFT
;
1825 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1826 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1827 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1828 offset
= *ppos
& ~PAGE_MASK
;
1834 unsigned long nr
, ret
;
1838 if (fatal_signal_pending(current
)) {
1843 page
= find_get_page(mapping
, index
);
1845 page_cache_sync_readahead(mapping
,
1847 index
, last_index
- index
);
1848 page
= find_get_page(mapping
, index
);
1849 if (unlikely(page
== NULL
))
1850 goto no_cached_page
;
1852 if (PageReadahead(page
)) {
1853 page_cache_async_readahead(mapping
,
1855 index
, last_index
- index
);
1857 if (!PageUptodate(page
)) {
1859 * See comment in do_read_cache_page on why
1860 * wait_on_page_locked is used to avoid unnecessarily
1861 * serialisations and why it's safe.
1863 error
= wait_on_page_locked_killable(page
);
1864 if (unlikely(error
))
1865 goto readpage_error
;
1866 if (PageUptodate(page
))
1869 if (inode
->i_blkbits
== PAGE_SHIFT
||
1870 !mapping
->a_ops
->is_partially_uptodate
)
1871 goto page_not_up_to_date
;
1872 /* pipes can't handle partially uptodate pages */
1873 if (unlikely(iter
->type
& ITER_PIPE
))
1874 goto page_not_up_to_date
;
1875 if (!trylock_page(page
))
1876 goto page_not_up_to_date
;
1877 /* Did it get truncated before we got the lock? */
1879 goto page_not_up_to_date_locked
;
1880 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1881 offset
, iter
->count
))
1882 goto page_not_up_to_date_locked
;
1887 * i_size must be checked after we know the page is Uptodate.
1889 * Checking i_size after the check allows us to calculate
1890 * the correct value for "nr", which means the zero-filled
1891 * part of the page is not copied back to userspace (unless
1892 * another truncate extends the file - this is desired though).
1895 isize
= i_size_read(inode
);
1896 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1897 if (unlikely(!isize
|| index
> end_index
)) {
1902 /* nr is the maximum number of bytes to copy from this page */
1904 if (index
== end_index
) {
1905 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1913 /* If users can be writing to this page using arbitrary
1914 * virtual addresses, take care about potential aliasing
1915 * before reading the page on the kernel side.
1917 if (mapping_writably_mapped(mapping
))
1918 flush_dcache_page(page
);
1921 * When a sequential read accesses a page several times,
1922 * only mark it as accessed the first time.
1924 if (prev_index
!= index
|| offset
!= prev_offset
)
1925 mark_page_accessed(page
);
1929 * Ok, we have the page, and it's up-to-date, so
1930 * now we can copy it to user space...
1933 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1935 index
+= offset
>> PAGE_SHIFT
;
1936 offset
&= ~PAGE_MASK
;
1937 prev_offset
= offset
;
1941 if (!iov_iter_count(iter
))
1949 page_not_up_to_date
:
1950 /* Get exclusive access to the page ... */
1951 error
= lock_page_killable(page
);
1952 if (unlikely(error
))
1953 goto readpage_error
;
1955 page_not_up_to_date_locked
:
1956 /* Did it get truncated before we got the lock? */
1957 if (!page
->mapping
) {
1963 /* Did somebody else fill it already? */
1964 if (PageUptodate(page
)) {
1971 * A previous I/O error may have been due to temporary
1972 * failures, eg. multipath errors.
1973 * PG_error will be set again if readpage fails.
1975 ClearPageError(page
);
1976 /* Start the actual read. The read will unlock the page. */
1977 error
= mapping
->a_ops
->readpage(filp
, page
);
1979 if (unlikely(error
)) {
1980 if (error
== AOP_TRUNCATED_PAGE
) {
1985 goto readpage_error
;
1988 if (!PageUptodate(page
)) {
1989 error
= lock_page_killable(page
);
1990 if (unlikely(error
))
1991 goto readpage_error
;
1992 if (!PageUptodate(page
)) {
1993 if (page
->mapping
== NULL
) {
1995 * invalidate_mapping_pages got it
2002 shrink_readahead_size_eio(filp
, ra
);
2004 goto readpage_error
;
2012 /* UHHUH! A synchronous read error occurred. Report it */
2018 * Ok, it wasn't cached, so we need to create a new
2021 page
= page_cache_alloc_cold(mapping
);
2026 error
= add_to_page_cache_lru(page
, mapping
, index
,
2027 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2030 if (error
== -EEXIST
) {
2040 ra
->prev_pos
= prev_index
;
2041 ra
->prev_pos
<<= PAGE_SHIFT
;
2042 ra
->prev_pos
|= prev_offset
;
2044 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2045 file_accessed(filp
);
2046 return written
? written
: error
;
2050 * generic_file_read_iter - generic filesystem read routine
2051 * @iocb: kernel I/O control block
2052 * @iter: destination for the data read
2054 * This is the "read_iter()" routine for all filesystems
2055 * that can use the page cache directly.
2058 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2060 struct file
*file
= iocb
->ki_filp
;
2062 size_t count
= iov_iter_count(iter
);
2065 goto out
; /* skip atime */
2067 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2068 struct address_space
*mapping
= file
->f_mapping
;
2069 struct inode
*inode
= mapping
->host
;
2072 size
= i_size_read(inode
);
2073 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2074 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2075 iocb
->ki_pos
+ count
- 1))
2078 retval
= filemap_write_and_wait_range(mapping
,
2080 iocb
->ki_pos
+ count
- 1);
2085 file_accessed(file
);
2087 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2089 iocb
->ki_pos
+= retval
;
2092 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2095 * Btrfs can have a short DIO read if we encounter
2096 * compressed extents, so if there was an error, or if
2097 * we've already read everything we wanted to, or if
2098 * there was a short read because we hit EOF, go ahead
2099 * and return. Otherwise fallthrough to buffered io for
2100 * the rest of the read. Buffered reads will not work for
2101 * DAX files, so don't bother trying.
2103 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2108 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
2112 EXPORT_SYMBOL(generic_file_read_iter
);
2116 * page_cache_read - adds requested page to the page cache if not already there
2117 * @file: file to read
2118 * @offset: page index
2119 * @gfp_mask: memory allocation flags
2121 * This adds the requested page to the page cache if it isn't already there,
2122 * and schedules an I/O to read in its contents from disk.
2124 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
2126 struct address_space
*mapping
= file
->f_mapping
;
2131 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
2135 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
2137 ret
= mapping
->a_ops
->readpage(file
, page
);
2138 else if (ret
== -EEXIST
)
2139 ret
= 0; /* losing race to add is OK */
2143 } while (ret
== AOP_TRUNCATED_PAGE
);
2148 #define MMAP_LOTSAMISS (100)
2151 * Synchronous readahead happens when we don't even find
2152 * a page in the page cache at all.
2154 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2155 struct file_ra_state
*ra
,
2159 struct address_space
*mapping
= file
->f_mapping
;
2161 /* If we don't want any read-ahead, don't bother */
2162 if (vma
->vm_flags
& VM_RAND_READ
)
2167 if (vma
->vm_flags
& VM_SEQ_READ
) {
2168 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2173 /* Avoid banging the cache line if not needed */
2174 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2178 * Do we miss much more than hit in this file? If so,
2179 * stop bothering with read-ahead. It will only hurt.
2181 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2187 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2188 ra
->size
= ra
->ra_pages
;
2189 ra
->async_size
= ra
->ra_pages
/ 4;
2190 ra_submit(ra
, mapping
, file
);
2194 * Asynchronous readahead happens when we find the page and PG_readahead,
2195 * so we want to possibly extend the readahead further..
2197 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2198 struct file_ra_state
*ra
,
2203 struct address_space
*mapping
= file
->f_mapping
;
2205 /* If we don't want any read-ahead, don't bother */
2206 if (vma
->vm_flags
& VM_RAND_READ
)
2208 if (ra
->mmap_miss
> 0)
2210 if (PageReadahead(page
))
2211 page_cache_async_readahead(mapping
, ra
, file
,
2212 page
, offset
, ra
->ra_pages
);
2216 * filemap_fault - read in file data for page fault handling
2217 * @vmf: struct vm_fault containing details of the fault
2219 * filemap_fault() is invoked via the vma operations vector for a
2220 * mapped memory region to read in file data during a page fault.
2222 * The goto's are kind of ugly, but this streamlines the normal case of having
2223 * it in the page cache, and handles the special cases reasonably without
2224 * having a lot of duplicated code.
2226 * vma->vm_mm->mmap_sem must be held on entry.
2228 * If our return value has VM_FAULT_RETRY set, it's because
2229 * lock_page_or_retry() returned 0.
2230 * The mmap_sem has usually been released in this case.
2231 * See __lock_page_or_retry() for the exception.
2233 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2234 * has not been released.
2236 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2238 int filemap_fault(struct vm_fault
*vmf
)
2241 struct file
*file
= vmf
->vma
->vm_file
;
2242 struct address_space
*mapping
= file
->f_mapping
;
2243 struct file_ra_state
*ra
= &file
->f_ra
;
2244 struct inode
*inode
= mapping
->host
;
2245 pgoff_t offset
= vmf
->pgoff
;
2250 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2251 if (unlikely(offset
>= max_off
))
2252 return VM_FAULT_SIGBUS
;
2255 * Do we have something in the page cache already?
2257 page
= find_get_page(mapping
, offset
);
2258 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2260 * We found the page, so try async readahead before
2261 * waiting for the lock.
2263 do_async_mmap_readahead(vmf
->vma
, ra
, file
, page
, offset
);
2265 /* No page in the page cache at all */
2266 do_sync_mmap_readahead(vmf
->vma
, ra
, file
, offset
);
2267 count_vm_event(PGMAJFAULT
);
2268 mem_cgroup_count_vm_event(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2269 ret
= VM_FAULT_MAJOR
;
2271 page
= find_get_page(mapping
, offset
);
2273 goto no_cached_page
;
2276 if (!lock_page_or_retry(page
, vmf
->vma
->vm_mm
, vmf
->flags
)) {
2278 return ret
| VM_FAULT_RETRY
;
2281 /* Did it get truncated? */
2282 if (unlikely(page
->mapping
!= mapping
)) {
2287 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2290 * We have a locked page in the page cache, now we need to check
2291 * that it's up-to-date. If not, it is going to be due to an error.
2293 if (unlikely(!PageUptodate(page
)))
2294 goto page_not_uptodate
;
2297 * Found the page and have a reference on it.
2298 * We must recheck i_size under page lock.
2300 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2301 if (unlikely(offset
>= max_off
)) {
2304 return VM_FAULT_SIGBUS
;
2308 return ret
| VM_FAULT_LOCKED
;
2312 * We're only likely to ever get here if MADV_RANDOM is in
2315 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2318 * The page we want has now been added to the page cache.
2319 * In the unlikely event that someone removed it in the
2320 * meantime, we'll just come back here and read it again.
2326 * An error return from page_cache_read can result if the
2327 * system is low on memory, or a problem occurs while trying
2330 if (error
== -ENOMEM
)
2331 return VM_FAULT_OOM
;
2332 return VM_FAULT_SIGBUS
;
2336 * Umm, take care of errors if the page isn't up-to-date.
2337 * Try to re-read it _once_. We do this synchronously,
2338 * because there really aren't any performance issues here
2339 * and we need to check for errors.
2341 ClearPageError(page
);
2342 error
= mapping
->a_ops
->readpage(file
, page
);
2344 wait_on_page_locked(page
);
2345 if (!PageUptodate(page
))
2350 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2353 /* Things didn't work out. Return zero to tell the mm layer so. */
2354 shrink_readahead_size_eio(file
, ra
);
2355 return VM_FAULT_SIGBUS
;
2357 EXPORT_SYMBOL(filemap_fault
);
2359 void filemap_map_pages(struct vm_fault
*vmf
,
2360 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2362 struct radix_tree_iter iter
;
2364 struct file
*file
= vmf
->vma
->vm_file
;
2365 struct address_space
*mapping
= file
->f_mapping
;
2366 pgoff_t last_pgoff
= start_pgoff
;
2367 unsigned long max_idx
;
2368 struct page
*head
, *page
;
2371 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2373 if (iter
.index
> end_pgoff
)
2376 page
= radix_tree_deref_slot(slot
);
2377 if (unlikely(!page
))
2379 if (radix_tree_exception(page
)) {
2380 if (radix_tree_deref_retry(page
)) {
2381 slot
= radix_tree_iter_retry(&iter
);
2387 head
= compound_head(page
);
2388 if (!page_cache_get_speculative(head
))
2391 /* The page was split under us? */
2392 if (compound_head(page
) != head
) {
2397 /* Has the page moved? */
2398 if (unlikely(page
!= *slot
)) {
2403 if (!PageUptodate(page
) ||
2404 PageReadahead(page
) ||
2407 if (!trylock_page(page
))
2410 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2413 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2414 if (page
->index
>= max_idx
)
2417 if (file
->f_ra
.mmap_miss
> 0)
2418 file
->f_ra
.mmap_miss
--;
2420 vmf
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2422 vmf
->pte
+= iter
.index
- last_pgoff
;
2423 last_pgoff
= iter
.index
;
2424 if (alloc_set_pte(vmf
, NULL
, page
))
2433 /* Huge page is mapped? No need to proceed. */
2434 if (pmd_trans_huge(*vmf
->pmd
))
2436 if (iter
.index
== end_pgoff
)
2441 EXPORT_SYMBOL(filemap_map_pages
);
2443 int filemap_page_mkwrite(struct vm_fault
*vmf
)
2445 struct page
*page
= vmf
->page
;
2446 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2447 int ret
= VM_FAULT_LOCKED
;
2449 sb_start_pagefault(inode
->i_sb
);
2450 file_update_time(vmf
->vma
->vm_file
);
2452 if (page
->mapping
!= inode
->i_mapping
) {
2454 ret
= VM_FAULT_NOPAGE
;
2458 * We mark the page dirty already here so that when freeze is in
2459 * progress, we are guaranteed that writeback during freezing will
2460 * see the dirty page and writeprotect it again.
2462 set_page_dirty(page
);
2463 wait_for_stable_page(page
);
2465 sb_end_pagefault(inode
->i_sb
);
2468 EXPORT_SYMBOL(filemap_page_mkwrite
);
2470 const struct vm_operations_struct generic_file_vm_ops
= {
2471 .fault
= filemap_fault
,
2472 .map_pages
= filemap_map_pages
,
2473 .page_mkwrite
= filemap_page_mkwrite
,
2476 /* This is used for a general mmap of a disk file */
2478 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2480 struct address_space
*mapping
= file
->f_mapping
;
2482 if (!mapping
->a_ops
->readpage
)
2484 file_accessed(file
);
2485 vma
->vm_ops
= &generic_file_vm_ops
;
2490 * This is for filesystems which do not implement ->writepage.
2492 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2494 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2496 return generic_file_mmap(file
, vma
);
2499 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2503 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2507 #endif /* CONFIG_MMU */
2509 EXPORT_SYMBOL(generic_file_mmap
);
2510 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2512 static struct page
*wait_on_page_read(struct page
*page
)
2514 if (!IS_ERR(page
)) {
2515 wait_on_page_locked(page
);
2516 if (!PageUptodate(page
)) {
2518 page
= ERR_PTR(-EIO
);
2524 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2526 int (*filler
)(void *, struct page
*),
2533 page
= find_get_page(mapping
, index
);
2535 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2537 return ERR_PTR(-ENOMEM
);
2538 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2539 if (unlikely(err
)) {
2543 /* Presumably ENOMEM for radix tree node */
2544 return ERR_PTR(err
);
2548 err
= filler(data
, page
);
2551 return ERR_PTR(err
);
2554 page
= wait_on_page_read(page
);
2559 if (PageUptodate(page
))
2563 * Page is not up to date and may be locked due one of the following
2564 * case a: Page is being filled and the page lock is held
2565 * case b: Read/write error clearing the page uptodate status
2566 * case c: Truncation in progress (page locked)
2567 * case d: Reclaim in progress
2569 * Case a, the page will be up to date when the page is unlocked.
2570 * There is no need to serialise on the page lock here as the page
2571 * is pinned so the lock gives no additional protection. Even if the
2572 * the page is truncated, the data is still valid if PageUptodate as
2573 * it's a race vs truncate race.
2574 * Case b, the page will not be up to date
2575 * Case c, the page may be truncated but in itself, the data may still
2576 * be valid after IO completes as it's a read vs truncate race. The
2577 * operation must restart if the page is not uptodate on unlock but
2578 * otherwise serialising on page lock to stabilise the mapping gives
2579 * no additional guarantees to the caller as the page lock is
2580 * released before return.
2581 * Case d, similar to truncation. If reclaim holds the page lock, it
2582 * will be a race with remove_mapping that determines if the mapping
2583 * is valid on unlock but otherwise the data is valid and there is
2584 * no need to serialise with page lock.
2586 * As the page lock gives no additional guarantee, we optimistically
2587 * wait on the page to be unlocked and check if it's up to date and
2588 * use the page if it is. Otherwise, the page lock is required to
2589 * distinguish between the different cases. The motivation is that we
2590 * avoid spurious serialisations and wakeups when multiple processes
2591 * wait on the same page for IO to complete.
2593 wait_on_page_locked(page
);
2594 if (PageUptodate(page
))
2597 /* Distinguish between all the cases under the safety of the lock */
2600 /* Case c or d, restart the operation */
2601 if (!page
->mapping
) {
2607 /* Someone else locked and filled the page in a very small window */
2608 if (PageUptodate(page
)) {
2615 mark_page_accessed(page
);
2620 * read_cache_page - read into page cache, fill it if needed
2621 * @mapping: the page's address_space
2622 * @index: the page index
2623 * @filler: function to perform the read
2624 * @data: first arg to filler(data, page) function, often left as NULL
2626 * Read into the page cache. If a page already exists, and PageUptodate() is
2627 * not set, try to fill the page and wait for it to become unlocked.
2629 * If the page does not get brought uptodate, return -EIO.
2631 struct page
*read_cache_page(struct address_space
*mapping
,
2633 int (*filler
)(void *, struct page
*),
2636 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2638 EXPORT_SYMBOL(read_cache_page
);
2641 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2642 * @mapping: the page's address_space
2643 * @index: the page index
2644 * @gfp: the page allocator flags to use if allocating
2646 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2647 * any new page allocations done using the specified allocation flags.
2649 * If the page does not get brought uptodate, return -EIO.
2651 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2655 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2657 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2659 EXPORT_SYMBOL(read_cache_page_gfp
);
2662 * Performs necessary checks before doing a write
2664 * Can adjust writing position or amount of bytes to write.
2665 * Returns appropriate error code that caller should return or
2666 * zero in case that write should be allowed.
2668 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2670 struct file
*file
= iocb
->ki_filp
;
2671 struct inode
*inode
= file
->f_mapping
->host
;
2672 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2675 if (!iov_iter_count(from
))
2678 /* FIXME: this is for backwards compatibility with 2.4 */
2679 if (iocb
->ki_flags
& IOCB_APPEND
)
2680 iocb
->ki_pos
= i_size_read(inode
);
2684 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
2687 if (limit
!= RLIM_INFINITY
) {
2688 if (iocb
->ki_pos
>= limit
) {
2689 send_sig(SIGXFSZ
, current
, 0);
2692 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2698 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2699 !(file
->f_flags
& O_LARGEFILE
))) {
2700 if (pos
>= MAX_NON_LFS
)
2702 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2706 * Are we about to exceed the fs block limit ?
2708 * If we have written data it becomes a short write. If we have
2709 * exceeded without writing data we send a signal and return EFBIG.
2710 * Linus frestrict idea will clean these up nicely..
2712 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2715 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2716 return iov_iter_count(from
);
2718 EXPORT_SYMBOL(generic_write_checks
);
2720 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2721 loff_t pos
, unsigned len
, unsigned flags
,
2722 struct page
**pagep
, void **fsdata
)
2724 const struct address_space_operations
*aops
= mapping
->a_ops
;
2726 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2729 EXPORT_SYMBOL(pagecache_write_begin
);
2731 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2732 loff_t pos
, unsigned len
, unsigned copied
,
2733 struct page
*page
, void *fsdata
)
2735 const struct address_space_operations
*aops
= mapping
->a_ops
;
2737 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2739 EXPORT_SYMBOL(pagecache_write_end
);
2742 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2744 struct file
*file
= iocb
->ki_filp
;
2745 struct address_space
*mapping
= file
->f_mapping
;
2746 struct inode
*inode
= mapping
->host
;
2747 loff_t pos
= iocb
->ki_pos
;
2752 write_len
= iov_iter_count(from
);
2753 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2755 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2756 /* If there are pages to writeback, return */
2757 if (filemap_range_has_page(inode
->i_mapping
, pos
,
2758 pos
+ iov_iter_count(from
)))
2761 written
= filemap_write_and_wait_range(mapping
, pos
,
2762 pos
+ write_len
- 1);
2768 * After a write we want buffered reads to be sure to go to disk to get
2769 * the new data. We invalidate clean cached page from the region we're
2770 * about to write. We do this *before* the write so that we can return
2771 * without clobbering -EIOCBQUEUED from ->direct_IO().
2773 written
= invalidate_inode_pages2_range(mapping
,
2774 pos
>> PAGE_SHIFT
, end
);
2776 * If a page can not be invalidated, return 0 to fall back
2777 * to buffered write.
2780 if (written
== -EBUSY
)
2785 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
2788 * Finally, try again to invalidate clean pages which might have been
2789 * cached by non-direct readahead, or faulted in by get_user_pages()
2790 * if the source of the write was an mmap'ed region of the file
2791 * we're writing. Either one is a pretty crazy thing to do,
2792 * so we don't support it 100%. If this invalidation
2793 * fails, tough, the write still worked...
2795 invalidate_inode_pages2_range(mapping
,
2796 pos
>> PAGE_SHIFT
, end
);
2800 write_len
-= written
;
2801 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2802 i_size_write(inode
, pos
);
2803 mark_inode_dirty(inode
);
2807 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
2811 EXPORT_SYMBOL(generic_file_direct_write
);
2814 * Find or create a page at the given pagecache position. Return the locked
2815 * page. This function is specifically for buffered writes.
2817 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2818 pgoff_t index
, unsigned flags
)
2821 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2823 if (flags
& AOP_FLAG_NOFS
)
2824 fgp_flags
|= FGP_NOFS
;
2826 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2827 mapping_gfp_mask(mapping
));
2829 wait_for_stable_page(page
);
2833 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2835 ssize_t
generic_perform_write(struct file
*file
,
2836 struct iov_iter
*i
, loff_t pos
)
2838 struct address_space
*mapping
= file
->f_mapping
;
2839 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2841 ssize_t written
= 0;
2842 unsigned int flags
= 0;
2846 unsigned long offset
; /* Offset into pagecache page */
2847 unsigned long bytes
; /* Bytes to write to page */
2848 size_t copied
; /* Bytes copied from user */
2851 offset
= (pos
& (PAGE_SIZE
- 1));
2852 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2857 * Bring in the user page that we will copy from _first_.
2858 * Otherwise there's a nasty deadlock on copying from the
2859 * same page as we're writing to, without it being marked
2862 * Not only is this an optimisation, but it is also required
2863 * to check that the address is actually valid, when atomic
2864 * usercopies are used, below.
2866 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2871 if (fatal_signal_pending(current
)) {
2876 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2878 if (unlikely(status
< 0))
2881 if (mapping_writably_mapped(mapping
))
2882 flush_dcache_page(page
);
2884 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2885 flush_dcache_page(page
);
2887 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2889 if (unlikely(status
< 0))
2895 iov_iter_advance(i
, copied
);
2896 if (unlikely(copied
== 0)) {
2898 * If we were unable to copy any data at all, we must
2899 * fall back to a single segment length write.
2901 * If we didn't fallback here, we could livelock
2902 * because not all segments in the iov can be copied at
2903 * once without a pagefault.
2905 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2906 iov_iter_single_seg_count(i
));
2912 balance_dirty_pages_ratelimited(mapping
);
2913 } while (iov_iter_count(i
));
2915 return written
? written
: status
;
2917 EXPORT_SYMBOL(generic_perform_write
);
2920 * __generic_file_write_iter - write data to a file
2921 * @iocb: IO state structure (file, offset, etc.)
2922 * @from: iov_iter with data to write
2924 * This function does all the work needed for actually writing data to a
2925 * file. It does all basic checks, removes SUID from the file, updates
2926 * modification times and calls proper subroutines depending on whether we
2927 * do direct IO or a standard buffered write.
2929 * It expects i_mutex to be grabbed unless we work on a block device or similar
2930 * object which does not need locking at all.
2932 * This function does *not* take care of syncing data in case of O_SYNC write.
2933 * A caller has to handle it. This is mainly due to the fact that we want to
2934 * avoid syncing under i_mutex.
2936 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2938 struct file
*file
= iocb
->ki_filp
;
2939 struct address_space
* mapping
= file
->f_mapping
;
2940 struct inode
*inode
= mapping
->host
;
2941 ssize_t written
= 0;
2945 /* We can write back this queue in page reclaim */
2946 current
->backing_dev_info
= inode_to_bdi(inode
);
2947 err
= file_remove_privs(file
);
2951 err
= file_update_time(file
);
2955 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2956 loff_t pos
, endbyte
;
2958 written
= generic_file_direct_write(iocb
, from
);
2960 * If the write stopped short of completing, fall back to
2961 * buffered writes. Some filesystems do this for writes to
2962 * holes, for example. For DAX files, a buffered write will
2963 * not succeed (even if it did, DAX does not handle dirty
2964 * page-cache pages correctly).
2966 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2969 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2971 * If generic_perform_write() returned a synchronous error
2972 * then we want to return the number of bytes which were
2973 * direct-written, or the error code if that was zero. Note
2974 * that this differs from normal direct-io semantics, which
2975 * will return -EFOO even if some bytes were written.
2977 if (unlikely(status
< 0)) {
2982 * We need to ensure that the page cache pages are written to
2983 * disk and invalidated to preserve the expected O_DIRECT
2986 endbyte
= pos
+ status
- 1;
2987 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2989 iocb
->ki_pos
= endbyte
+ 1;
2991 invalidate_mapping_pages(mapping
,
2993 endbyte
>> PAGE_SHIFT
);
2996 * We don't know how much we wrote, so just return
2997 * the number of bytes which were direct-written
3001 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3002 if (likely(written
> 0))
3003 iocb
->ki_pos
+= written
;
3006 current
->backing_dev_info
= NULL
;
3007 return written
? written
: err
;
3009 EXPORT_SYMBOL(__generic_file_write_iter
);
3012 * generic_file_write_iter - write data to a file
3013 * @iocb: IO state structure
3014 * @from: iov_iter with data to write
3016 * This is a wrapper around __generic_file_write_iter() to be used by most
3017 * filesystems. It takes care of syncing the file in case of O_SYNC file
3018 * and acquires i_mutex as needed.
3020 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3022 struct file
*file
= iocb
->ki_filp
;
3023 struct inode
*inode
= file
->f_mapping
->host
;
3027 ret
= generic_write_checks(iocb
, from
);
3029 ret
= __generic_file_write_iter(iocb
, from
);
3030 inode_unlock(inode
);
3033 ret
= generic_write_sync(iocb
, ret
);
3036 EXPORT_SYMBOL(generic_file_write_iter
);
3039 * try_to_release_page() - release old fs-specific metadata on a page
3041 * @page: the page which the kernel is trying to free
3042 * @gfp_mask: memory allocation flags (and I/O mode)
3044 * The address_space is to try to release any data against the page
3045 * (presumably at page->private). If the release was successful, return '1'.
3046 * Otherwise return zero.
3048 * This may also be called if PG_fscache is set on a page, indicating that the
3049 * page is known to the local caching routines.
3051 * The @gfp_mask argument specifies whether I/O may be performed to release
3052 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3055 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3057 struct address_space
* const mapping
= page
->mapping
;
3059 BUG_ON(!PageLocked(page
));
3060 if (PageWriteback(page
))
3063 if (mapping
&& mapping
->a_ops
->releasepage
)
3064 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3065 return try_to_free_buffers(page
);
3068 EXPORT_SYMBOL(try_to_release_page
);