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 /* DAX can replace empty locked entry with a hole */
138 dax_radix_locked_entry(0, RADIX_DAX_EMPTY
));
139 /* Wakeup waiters for exceptional entry lock */
140 dax_wake_mapping_entry_waiter(mapping
, page
->index
, p
,
144 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, page
,
145 workingset_update_node
, mapping
);
150 static void page_cache_tree_delete(struct address_space
*mapping
,
151 struct page
*page
, void *shadow
)
155 /* hugetlb pages are represented by one entry in the radix tree */
156 nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
158 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
159 VM_BUG_ON_PAGE(PageTail(page
), page
);
160 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
162 for (i
= 0; i
< nr
; i
++) {
163 struct radix_tree_node
*node
;
166 __radix_tree_lookup(&mapping
->page_tree
, page
->index
+ i
,
169 VM_BUG_ON_PAGE(!node
&& nr
!= 1, page
);
171 radix_tree_clear_tags(&mapping
->page_tree
, node
, slot
);
172 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, shadow
,
173 workingset_update_node
, mapping
);
177 mapping
->nrexceptional
+= nr
;
179 * Make sure the nrexceptional update is committed before
180 * the nrpages update so that final truncate racing
181 * with reclaim does not see both counters 0 at the
182 * same time and miss a shadow entry.
186 mapping
->nrpages
-= nr
;
190 * Delete a page from the page cache and free it. Caller has to make
191 * sure the page is locked and that nobody else uses it - or that usage
192 * is safe. The caller must hold the mapping's tree_lock.
194 void __delete_from_page_cache(struct page
*page
, void *shadow
)
196 struct address_space
*mapping
= page
->mapping
;
197 int nr
= hpage_nr_pages(page
);
199 trace_mm_filemap_delete_from_page_cache(page
);
201 * if we're uptodate, flush out into the cleancache, otherwise
202 * invalidate any existing cleancache entries. We can't leave
203 * stale data around in the cleancache once our page is gone
205 if (PageUptodate(page
) && PageMappedToDisk(page
))
206 cleancache_put_page(page
);
208 cleancache_invalidate_page(mapping
, page
);
210 VM_BUG_ON_PAGE(PageTail(page
), page
);
211 VM_BUG_ON_PAGE(page_mapped(page
), page
);
212 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
215 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
216 current
->comm
, page_to_pfn(page
));
217 dump_page(page
, "still mapped when deleted");
219 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
221 mapcount
= page_mapcount(page
);
222 if (mapping_exiting(mapping
) &&
223 page_count(page
) >= mapcount
+ 2) {
225 * All vmas have already been torn down, so it's
226 * a good bet that actually the page is unmapped,
227 * and we'd prefer not to leak it: if we're wrong,
228 * some other bad page check should catch it later.
230 page_mapcount_reset(page
);
231 page_ref_sub(page
, mapcount
);
235 page_cache_tree_delete(mapping
, page
, shadow
);
237 page
->mapping
= NULL
;
238 /* Leave page->index set: truncation lookup relies upon it */
240 /* hugetlb pages do not participate in page cache accounting. */
242 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
243 if (PageSwapBacked(page
)) {
244 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
245 if (PageTransHuge(page
))
246 __dec_node_page_state(page
, NR_SHMEM_THPS
);
248 VM_BUG_ON_PAGE(PageTransHuge(page
) && !PageHuge(page
), page
);
252 * At this point page must be either written or cleaned by truncate.
253 * Dirty page here signals a bug and loss of unwritten data.
255 * This fixes dirty accounting after removing the page entirely but
256 * leaves PageDirty set: it has no effect for truncated page and
257 * anyway will be cleared before returning page into buddy allocator.
259 if (WARN_ON_ONCE(PageDirty(page
)))
260 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
264 * delete_from_page_cache - delete page from page cache
265 * @page: the page which the kernel is trying to remove from page cache
267 * This must be called only on pages that have been verified to be in the page
268 * cache and locked. It will never put the page into the free list, the caller
269 * has a reference on the page.
271 void delete_from_page_cache(struct page
*page
)
273 struct address_space
*mapping
= page_mapping(page
);
275 void (*freepage
)(struct page
*);
277 BUG_ON(!PageLocked(page
));
279 freepage
= mapping
->a_ops
->freepage
;
281 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
282 __delete_from_page_cache(page
, NULL
);
283 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
288 if (PageTransHuge(page
) && !PageHuge(page
)) {
289 page_ref_sub(page
, HPAGE_PMD_NR
);
290 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
295 EXPORT_SYMBOL(delete_from_page_cache
);
297 int filemap_check_errors(struct address_space
*mapping
)
300 /* Check for outstanding write errors */
301 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
302 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
304 if (test_bit(AS_EIO
, &mapping
->flags
) &&
305 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
309 EXPORT_SYMBOL(filemap_check_errors
);
312 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
313 * @mapping: address space structure to write
314 * @start: offset in bytes where the range starts
315 * @end: offset in bytes where the range ends (inclusive)
316 * @sync_mode: enable synchronous operation
318 * Start writeback against all of a mapping's dirty pages that lie
319 * within the byte offsets <start, end> inclusive.
321 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
322 * opposed to a regular memory cleansing writeback. The difference between
323 * these two operations is that if a dirty page/buffer is encountered, it must
324 * be waited upon, and not just skipped over.
326 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
327 loff_t end
, int sync_mode
)
330 struct writeback_control wbc
= {
331 .sync_mode
= sync_mode
,
332 .nr_to_write
= LONG_MAX
,
333 .range_start
= start
,
337 if (!mapping_cap_writeback_dirty(mapping
))
340 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
341 ret
= do_writepages(mapping
, &wbc
);
342 wbc_detach_inode(&wbc
);
346 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
349 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
352 int filemap_fdatawrite(struct address_space
*mapping
)
354 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
356 EXPORT_SYMBOL(filemap_fdatawrite
);
358 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
361 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
363 EXPORT_SYMBOL(filemap_fdatawrite_range
);
366 * filemap_flush - mostly a non-blocking flush
367 * @mapping: target address_space
369 * This is a mostly non-blocking flush. Not suitable for data-integrity
370 * purposes - I/O may not be started against all dirty pages.
372 int filemap_flush(struct address_space
*mapping
)
374 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
376 EXPORT_SYMBOL(filemap_flush
);
378 static int __filemap_fdatawait_range(struct address_space
*mapping
,
379 loff_t start_byte
, loff_t end_byte
)
381 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
382 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
387 if (end_byte
< start_byte
)
390 pagevec_init(&pvec
, 0);
391 while ((index
<= end
) &&
392 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
393 PAGECACHE_TAG_WRITEBACK
,
394 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
397 for (i
= 0; i
< nr_pages
; i
++) {
398 struct page
*page
= pvec
.pages
[i
];
400 /* until radix tree lookup accepts end_index */
401 if (page
->index
> end
)
404 wait_on_page_writeback(page
);
405 if (TestClearPageError(page
))
408 pagevec_release(&pvec
);
416 * filemap_fdatawait_range - wait for writeback to complete
417 * @mapping: address space structure to wait for
418 * @start_byte: offset in bytes where the range starts
419 * @end_byte: offset in bytes where the range ends (inclusive)
421 * Walk the list of under-writeback pages of the given address space
422 * in the given range and wait for all of them. Check error status of
423 * the address space and return it.
425 * Since the error status of the address space is cleared by this function,
426 * callers are responsible for checking the return value and handling and/or
427 * reporting the error.
429 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
434 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
435 ret2
= filemap_check_errors(mapping
);
441 EXPORT_SYMBOL(filemap_fdatawait_range
);
444 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
445 * @mapping: address space structure to wait for
447 * Walk the list of under-writeback pages of the given address space
448 * and wait for all of them. Unlike filemap_fdatawait(), this function
449 * does not clear error status of the address space.
451 * Use this function if callers don't handle errors themselves. Expected
452 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
455 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
457 loff_t i_size
= i_size_read(mapping
->host
);
462 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
466 * filemap_fdatawait - wait for all under-writeback pages to complete
467 * @mapping: address space structure to wait for
469 * Walk the list of under-writeback pages of the given address space
470 * and wait for all of them. Check error status of the address space
473 * Since the error status of the address space is cleared by this function,
474 * callers are responsible for checking the return value and handling and/or
475 * reporting the error.
477 int filemap_fdatawait(struct address_space
*mapping
)
479 loff_t i_size
= i_size_read(mapping
->host
);
484 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
486 EXPORT_SYMBOL(filemap_fdatawait
);
488 int filemap_write_and_wait(struct address_space
*mapping
)
492 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
493 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
494 err
= filemap_fdatawrite(mapping
);
496 * Even if the above returned error, the pages may be
497 * written partially (e.g. -ENOSPC), so we wait for it.
498 * But the -EIO is special case, it may indicate the worst
499 * thing (e.g. bug) happened, so we avoid waiting for it.
502 int err2
= filemap_fdatawait(mapping
);
507 err
= filemap_check_errors(mapping
);
511 EXPORT_SYMBOL(filemap_write_and_wait
);
514 * filemap_write_and_wait_range - write out & wait on a file range
515 * @mapping: the address_space for the pages
516 * @lstart: offset in bytes where the range starts
517 * @lend: offset in bytes where the range ends (inclusive)
519 * Write out and wait upon file offsets lstart->lend, inclusive.
521 * Note that `lend' is inclusive (describes the last byte to be written) so
522 * that this function can be used to write to the very end-of-file (end = -1).
524 int filemap_write_and_wait_range(struct address_space
*mapping
,
525 loff_t lstart
, loff_t lend
)
529 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
530 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
531 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
533 /* See comment of filemap_write_and_wait() */
535 int err2
= filemap_fdatawait_range(mapping
,
541 err
= filemap_check_errors(mapping
);
545 EXPORT_SYMBOL(filemap_write_and_wait_range
);
548 * replace_page_cache_page - replace a pagecache page with a new one
549 * @old: page to be replaced
550 * @new: page to replace with
551 * @gfp_mask: allocation mode
553 * This function replaces a page in the pagecache with a new one. On
554 * success it acquires the pagecache reference for the new page and
555 * drops it for the old page. Both the old and new pages must be
556 * locked. This function does not add the new page to the LRU, the
557 * caller must do that.
559 * The remove + add is atomic. The only way this function can fail is
560 * memory allocation failure.
562 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
566 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
567 VM_BUG_ON_PAGE(!PageLocked(new), new);
568 VM_BUG_ON_PAGE(new->mapping
, new);
570 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
572 struct address_space
*mapping
= old
->mapping
;
573 void (*freepage
)(struct page
*);
576 pgoff_t offset
= old
->index
;
577 freepage
= mapping
->a_ops
->freepage
;
580 new->mapping
= mapping
;
583 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
584 __delete_from_page_cache(old
, NULL
);
585 error
= page_cache_tree_insert(mapping
, new, NULL
);
589 * hugetlb pages do not participate in page cache accounting.
592 __inc_node_page_state(new, NR_FILE_PAGES
);
593 if (PageSwapBacked(new))
594 __inc_node_page_state(new, NR_SHMEM
);
595 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
596 mem_cgroup_migrate(old
, new);
597 radix_tree_preload_end();
605 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
607 static int __add_to_page_cache_locked(struct page
*page
,
608 struct address_space
*mapping
,
609 pgoff_t offset
, gfp_t gfp_mask
,
612 int huge
= PageHuge(page
);
613 struct mem_cgroup
*memcg
;
616 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
617 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
620 error
= mem_cgroup_try_charge(page
, current
->mm
,
621 gfp_mask
, &memcg
, false);
626 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
629 mem_cgroup_cancel_charge(page
, memcg
, false);
634 page
->mapping
= mapping
;
635 page
->index
= offset
;
637 spin_lock_irq(&mapping
->tree_lock
);
638 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
639 radix_tree_preload_end();
643 /* hugetlb pages do not participate in page cache accounting. */
645 __inc_node_page_state(page
, NR_FILE_PAGES
);
646 spin_unlock_irq(&mapping
->tree_lock
);
648 mem_cgroup_commit_charge(page
, memcg
, false, false);
649 trace_mm_filemap_add_to_page_cache(page
);
652 page
->mapping
= NULL
;
653 /* Leave page->index set: truncation relies upon it */
654 spin_unlock_irq(&mapping
->tree_lock
);
656 mem_cgroup_cancel_charge(page
, memcg
, false);
662 * add_to_page_cache_locked - add a locked page to the pagecache
664 * @mapping: the page's address_space
665 * @offset: page index
666 * @gfp_mask: page allocation mode
668 * This function is used to add a page to the pagecache. It must be locked.
669 * This function does not add the page to the LRU. The caller must do that.
671 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
672 pgoff_t offset
, gfp_t gfp_mask
)
674 return __add_to_page_cache_locked(page
, mapping
, offset
,
677 EXPORT_SYMBOL(add_to_page_cache_locked
);
679 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
680 pgoff_t offset
, gfp_t gfp_mask
)
685 __SetPageLocked(page
);
686 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
689 __ClearPageLocked(page
);
692 * The page might have been evicted from cache only
693 * recently, in which case it should be activated like
694 * any other repeatedly accessed page.
695 * The exception is pages getting rewritten; evicting other
696 * data from the working set, only to cache data that will
697 * get overwritten with something else, is a waste of memory.
699 if (!(gfp_mask
& __GFP_WRITE
) &&
700 shadow
&& workingset_refault(shadow
)) {
702 workingset_activation(page
);
704 ClearPageActive(page
);
709 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
712 struct page
*__page_cache_alloc(gfp_t gfp
)
717 if (cpuset_do_page_mem_spread()) {
718 unsigned int cpuset_mems_cookie
;
720 cpuset_mems_cookie
= read_mems_allowed_begin();
721 n
= cpuset_mem_spread_node();
722 page
= __alloc_pages_node(n
, gfp
, 0);
723 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
727 return alloc_pages(gfp
, 0);
729 EXPORT_SYMBOL(__page_cache_alloc
);
733 * In order to wait for pages to become available there must be
734 * waitqueues associated with pages. By using a hash table of
735 * waitqueues where the bucket discipline is to maintain all
736 * waiters on the same queue and wake all when any of the pages
737 * become available, and for the woken contexts to check to be
738 * sure the appropriate page became available, this saves space
739 * at a cost of "thundering herd" phenomena during rare hash
742 #define PAGE_WAIT_TABLE_BITS 8
743 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
744 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
746 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
748 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
751 void __init
pagecache_init(void)
755 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
756 init_waitqueue_head(&page_wait_table
[i
]);
758 page_writeback_init();
761 struct wait_page_key
{
767 struct wait_page_queue
{
773 static int wake_page_function(wait_queue_t
*wait
, unsigned mode
, int sync
, void *arg
)
775 struct wait_page_key
*key
= arg
;
776 struct wait_page_queue
*wait_page
777 = container_of(wait
, struct wait_page_queue
, wait
);
779 if (wait_page
->page
!= key
->page
)
783 if (wait_page
->bit_nr
!= key
->bit_nr
)
785 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
788 return autoremove_wake_function(wait
, mode
, sync
, key
);
791 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
793 wait_queue_head_t
*q
= page_waitqueue(page
);
794 struct wait_page_key key
;
801 spin_lock_irqsave(&q
->lock
, flags
);
802 __wake_up_locked_key(q
, TASK_NORMAL
, &key
);
804 * It is possible for other pages to have collided on the waitqueue
805 * hash, so in that case check for a page match. That prevents a long-
808 * It is still possible to miss a case here, when we woke page waiters
809 * and removed them from the waitqueue, but there are still other
812 if (!waitqueue_active(q
) || !key
.page_match
) {
813 ClearPageWaiters(page
);
815 * It's possible to miss clearing Waiters here, when we woke
816 * our page waiters, but the hashed waitqueue has waiters for
819 * That's okay, it's a rare case. The next waker will clear it.
822 spin_unlock_irqrestore(&q
->lock
, flags
);
825 static void wake_up_page(struct page
*page
, int bit
)
827 if (!PageWaiters(page
))
829 wake_up_page_bit(page
, bit
);
832 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
833 struct page
*page
, int bit_nr
, int state
, bool lock
)
835 struct wait_page_queue wait_page
;
836 wait_queue_t
*wait
= &wait_page
.wait
;
840 wait
->func
= wake_page_function
;
841 wait_page
.page
= page
;
842 wait_page
.bit_nr
= bit_nr
;
845 spin_lock_irq(&q
->lock
);
847 if (likely(list_empty(&wait
->task_list
))) {
849 __add_wait_queue_tail_exclusive(q
, wait
);
851 __add_wait_queue(q
, wait
);
852 SetPageWaiters(page
);
855 set_current_state(state
);
857 spin_unlock_irq(&q
->lock
);
859 if (likely(test_bit(bit_nr
, &page
->flags
))) {
861 if (unlikely(signal_pending_state(state
, current
))) {
868 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
871 if (!test_bit(bit_nr
, &page
->flags
))
876 finish_wait(q
, wait
);
879 * A signal could leave PageWaiters set. Clearing it here if
880 * !waitqueue_active would be possible (by open-coding finish_wait),
881 * but still fail to catch it in the case of wait hash collision. We
882 * already can fail to clear wait hash collision cases, so don't
883 * bother with signals either.
889 void wait_on_page_bit(struct page
*page
, int bit_nr
)
891 wait_queue_head_t
*q
= page_waitqueue(page
);
892 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, false);
894 EXPORT_SYMBOL(wait_on_page_bit
);
896 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
898 wait_queue_head_t
*q
= page_waitqueue(page
);
899 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, false);
903 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
904 * @page: Page defining the wait queue of interest
905 * @waiter: Waiter to add to the queue
907 * Add an arbitrary @waiter to the wait queue for the nominated @page.
909 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
911 wait_queue_head_t
*q
= page_waitqueue(page
);
914 spin_lock_irqsave(&q
->lock
, flags
);
915 __add_wait_queue(q
, waiter
);
916 SetPageWaiters(page
);
917 spin_unlock_irqrestore(&q
->lock
, flags
);
919 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
921 #ifndef clear_bit_unlock_is_negative_byte
924 * PG_waiters is the high bit in the same byte as PG_lock.
926 * On x86 (and on many other architectures), we can clear PG_lock and
927 * test the sign bit at the same time. But if the architecture does
928 * not support that special operation, we just do this all by hand
931 * The read of PG_waiters has to be after (or concurrently with) PG_locked
932 * being cleared, but a memory barrier should be unneccssary since it is
933 * in the same byte as PG_locked.
935 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
937 clear_bit_unlock(nr
, mem
);
938 /* smp_mb__after_atomic(); */
939 return test_bit(PG_waiters
, mem
);
945 * unlock_page - unlock a locked page
948 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
949 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
950 * mechanism between PageLocked pages and PageWriteback pages is shared.
951 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
953 * Note that this depends on PG_waiters being the sign bit in the byte
954 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
955 * clear the PG_locked bit and test PG_waiters at the same time fairly
956 * portably (architectures that do LL/SC can test any bit, while x86 can
957 * test the sign bit).
959 void unlock_page(struct page
*page
)
961 BUILD_BUG_ON(PG_waiters
!= 7);
962 page
= compound_head(page
);
963 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
964 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
965 wake_up_page_bit(page
, PG_locked
);
967 EXPORT_SYMBOL(unlock_page
);
970 * end_page_writeback - end writeback against a page
973 void end_page_writeback(struct page
*page
)
976 * TestClearPageReclaim could be used here but it is an atomic
977 * operation and overkill in this particular case. Failing to
978 * shuffle a page marked for immediate reclaim is too mild to
979 * justify taking an atomic operation penalty at the end of
980 * ever page writeback.
982 if (PageReclaim(page
)) {
983 ClearPageReclaim(page
);
984 rotate_reclaimable_page(page
);
987 if (!test_clear_page_writeback(page
))
990 smp_mb__after_atomic();
991 wake_up_page(page
, PG_writeback
);
993 EXPORT_SYMBOL(end_page_writeback
);
996 * After completing I/O on a page, call this routine to update the page
997 * flags appropriately
999 void page_endio(struct page
*page
, bool is_write
, int err
)
1003 SetPageUptodate(page
);
1005 ClearPageUptodate(page
);
1013 mapping_set_error(page
->mapping
, err
);
1015 end_page_writeback(page
);
1018 EXPORT_SYMBOL_GPL(page_endio
);
1021 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1022 * @__page: the page to lock
1024 void __lock_page(struct page
*__page
)
1026 struct page
*page
= compound_head(__page
);
1027 wait_queue_head_t
*q
= page_waitqueue(page
);
1028 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, true);
1030 EXPORT_SYMBOL(__lock_page
);
1032 int __lock_page_killable(struct page
*__page
)
1034 struct page
*page
= compound_head(__page
);
1035 wait_queue_head_t
*q
= page_waitqueue(page
);
1036 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
, true);
1038 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1042 * 1 - page is locked; mmap_sem is still held.
1043 * 0 - page is not locked.
1044 * mmap_sem has been released (up_read()), unless flags had both
1045 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1046 * which case mmap_sem is still held.
1048 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1049 * with the page locked and the mmap_sem unperturbed.
1051 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1054 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1056 * CAUTION! In this case, mmap_sem is not released
1057 * even though return 0.
1059 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1062 up_read(&mm
->mmap_sem
);
1063 if (flags
& FAULT_FLAG_KILLABLE
)
1064 wait_on_page_locked_killable(page
);
1066 wait_on_page_locked(page
);
1069 if (flags
& FAULT_FLAG_KILLABLE
) {
1072 ret
= __lock_page_killable(page
);
1074 up_read(&mm
->mmap_sem
);
1084 * page_cache_next_hole - find the next hole (not-present entry)
1087 * @max_scan: maximum range to search
1089 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1090 * lowest indexed hole.
1092 * Returns: the index of the hole if found, otherwise returns an index
1093 * outside of the set specified (in which case 'return - index >=
1094 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1097 * page_cache_next_hole may be called under rcu_read_lock. However,
1098 * like radix_tree_gang_lookup, this will not atomically search a
1099 * snapshot of the tree at a single point in time. For example, if a
1100 * hole is created at index 5, then subsequently a hole is created at
1101 * index 10, page_cache_next_hole covering both indexes may return 10
1102 * if called under rcu_read_lock.
1104 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1105 pgoff_t index
, unsigned long max_scan
)
1109 for (i
= 0; i
< max_scan
; i
++) {
1112 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1113 if (!page
|| radix_tree_exceptional_entry(page
))
1122 EXPORT_SYMBOL(page_cache_next_hole
);
1125 * page_cache_prev_hole - find the prev hole (not-present entry)
1128 * @max_scan: maximum range to search
1130 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1133 * Returns: the index of the hole if found, otherwise returns an index
1134 * outside of the set specified (in which case 'index - return >=
1135 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1138 * page_cache_prev_hole may be called under rcu_read_lock. However,
1139 * like radix_tree_gang_lookup, this will not atomically search a
1140 * snapshot of the tree at a single point in time. For example, if a
1141 * hole is created at index 10, then subsequently a hole is created at
1142 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1143 * called under rcu_read_lock.
1145 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1146 pgoff_t index
, unsigned long max_scan
)
1150 for (i
= 0; i
< max_scan
; i
++) {
1153 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1154 if (!page
|| radix_tree_exceptional_entry(page
))
1157 if (index
== ULONG_MAX
)
1163 EXPORT_SYMBOL(page_cache_prev_hole
);
1166 * find_get_entry - find and get a page cache entry
1167 * @mapping: the address_space to search
1168 * @offset: the page cache index
1170 * Looks up the page cache slot at @mapping & @offset. If there is a
1171 * page cache page, it is returned with an increased refcount.
1173 * If the slot holds a shadow entry of a previously evicted page, or a
1174 * swap entry from shmem/tmpfs, it is returned.
1176 * Otherwise, %NULL is returned.
1178 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1181 struct page
*head
, *page
;
1186 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1188 page
= radix_tree_deref_slot(pagep
);
1189 if (unlikely(!page
))
1191 if (radix_tree_exception(page
)) {
1192 if (radix_tree_deref_retry(page
))
1195 * A shadow entry of a recently evicted page,
1196 * or a swap entry from shmem/tmpfs. Return
1197 * it without attempting to raise page count.
1202 head
= compound_head(page
);
1203 if (!page_cache_get_speculative(head
))
1206 /* The page was split under us? */
1207 if (compound_head(page
) != head
) {
1213 * Has the page moved?
1214 * This is part of the lockless pagecache protocol. See
1215 * include/linux/pagemap.h for details.
1217 if (unlikely(page
!= *pagep
)) {
1227 EXPORT_SYMBOL(find_get_entry
);
1230 * find_lock_entry - locate, pin and lock a page cache entry
1231 * @mapping: the address_space to search
1232 * @offset: the page cache index
1234 * Looks up the page cache slot at @mapping & @offset. If there is a
1235 * page cache page, it is returned locked and with an increased
1238 * If the slot holds a shadow entry of a previously evicted page, or a
1239 * swap entry from shmem/tmpfs, it is returned.
1241 * Otherwise, %NULL is returned.
1243 * find_lock_entry() may sleep.
1245 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1250 page
= find_get_entry(mapping
, offset
);
1251 if (page
&& !radix_tree_exception(page
)) {
1253 /* Has the page been truncated? */
1254 if (unlikely(page_mapping(page
) != mapping
)) {
1259 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1263 EXPORT_SYMBOL(find_lock_entry
);
1266 * pagecache_get_page - find and get a page reference
1267 * @mapping: the address_space to search
1268 * @offset: the page index
1269 * @fgp_flags: PCG flags
1270 * @gfp_mask: gfp mask to use for the page cache data page allocation
1272 * Looks up the page cache slot at @mapping & @offset.
1274 * PCG flags modify how the page is returned.
1276 * FGP_ACCESSED: the page will be marked accessed
1277 * FGP_LOCK: Page is return locked
1278 * FGP_CREAT: If page is not present then a new page is allocated using
1279 * @gfp_mask and added to the page cache and the VM's LRU
1280 * list. The page is returned locked and with an increased
1281 * refcount. Otherwise, %NULL is returned.
1283 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1284 * if the GFP flags specified for FGP_CREAT are atomic.
1286 * If there is a page cache page, it is returned with an increased refcount.
1288 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1289 int fgp_flags
, gfp_t gfp_mask
)
1294 page
= find_get_entry(mapping
, offset
);
1295 if (radix_tree_exceptional_entry(page
))
1300 if (fgp_flags
& FGP_LOCK
) {
1301 if (fgp_flags
& FGP_NOWAIT
) {
1302 if (!trylock_page(page
)) {
1310 /* Has the page been truncated? */
1311 if (unlikely(page
->mapping
!= mapping
)) {
1316 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1319 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1320 mark_page_accessed(page
);
1323 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1325 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1326 gfp_mask
|= __GFP_WRITE
;
1327 if (fgp_flags
& FGP_NOFS
)
1328 gfp_mask
&= ~__GFP_FS
;
1330 page
= __page_cache_alloc(gfp_mask
);
1334 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1335 fgp_flags
|= FGP_LOCK
;
1337 /* Init accessed so avoid atomic mark_page_accessed later */
1338 if (fgp_flags
& FGP_ACCESSED
)
1339 __SetPageReferenced(page
);
1341 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1342 gfp_mask
& GFP_RECLAIM_MASK
);
1343 if (unlikely(err
)) {
1353 EXPORT_SYMBOL(pagecache_get_page
);
1356 * find_get_entries - gang pagecache lookup
1357 * @mapping: The address_space to search
1358 * @start: The starting page cache index
1359 * @nr_entries: The maximum number of entries
1360 * @entries: Where the resulting entries are placed
1361 * @indices: The cache indices corresponding to the entries in @entries
1363 * find_get_entries() will search for and return a group of up to
1364 * @nr_entries entries in the mapping. The entries are placed at
1365 * @entries. find_get_entries() takes a reference against any actual
1368 * The search returns a group of mapping-contiguous page cache entries
1369 * with ascending indexes. There may be holes in the indices due to
1370 * not-present pages.
1372 * Any shadow entries of evicted pages, or swap entries from
1373 * shmem/tmpfs, are included in the returned array.
1375 * find_get_entries() returns the number of pages and shadow entries
1378 unsigned find_get_entries(struct address_space
*mapping
,
1379 pgoff_t start
, unsigned int nr_entries
,
1380 struct page
**entries
, pgoff_t
*indices
)
1383 unsigned int ret
= 0;
1384 struct radix_tree_iter iter
;
1390 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1391 struct page
*head
, *page
;
1393 page
= radix_tree_deref_slot(slot
);
1394 if (unlikely(!page
))
1396 if (radix_tree_exception(page
)) {
1397 if (radix_tree_deref_retry(page
)) {
1398 slot
= radix_tree_iter_retry(&iter
);
1402 * A shadow entry of a recently evicted page, a swap
1403 * entry from shmem/tmpfs or a DAX entry. Return it
1404 * without attempting to raise page count.
1409 head
= compound_head(page
);
1410 if (!page_cache_get_speculative(head
))
1413 /* The page was split under us? */
1414 if (compound_head(page
) != head
) {
1419 /* Has the page moved? */
1420 if (unlikely(page
!= *slot
)) {
1425 indices
[ret
] = iter
.index
;
1426 entries
[ret
] = page
;
1427 if (++ret
== nr_entries
)
1435 * find_get_pages - gang pagecache lookup
1436 * @mapping: The address_space to search
1437 * @start: The starting page index
1438 * @nr_pages: The maximum number of pages
1439 * @pages: Where the resulting pages are placed
1441 * find_get_pages() will search for and return a group of up to
1442 * @nr_pages pages in the mapping. The pages are placed at @pages.
1443 * find_get_pages() takes a reference against the returned pages.
1445 * The search returns a group of mapping-contiguous pages with ascending
1446 * indexes. There may be holes in the indices due to not-present pages.
1448 * find_get_pages() returns the number of pages which were found.
1450 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1451 unsigned int nr_pages
, struct page
**pages
)
1453 struct radix_tree_iter iter
;
1457 if (unlikely(!nr_pages
))
1461 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1462 struct page
*head
, *page
;
1464 page
= radix_tree_deref_slot(slot
);
1465 if (unlikely(!page
))
1468 if (radix_tree_exception(page
)) {
1469 if (radix_tree_deref_retry(page
)) {
1470 slot
= radix_tree_iter_retry(&iter
);
1474 * A shadow entry of a recently evicted page,
1475 * or a swap entry from shmem/tmpfs. Skip
1481 head
= compound_head(page
);
1482 if (!page_cache_get_speculative(head
))
1485 /* The page was split under us? */
1486 if (compound_head(page
) != head
) {
1491 /* Has the page moved? */
1492 if (unlikely(page
!= *slot
)) {
1498 if (++ret
== nr_pages
)
1507 * find_get_pages_contig - gang contiguous pagecache lookup
1508 * @mapping: The address_space to search
1509 * @index: The starting page index
1510 * @nr_pages: The maximum number of pages
1511 * @pages: Where the resulting pages are placed
1513 * find_get_pages_contig() works exactly like find_get_pages(), except
1514 * that the returned number of pages are guaranteed to be contiguous.
1516 * find_get_pages_contig() returns the number of pages which were found.
1518 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1519 unsigned int nr_pages
, struct page
**pages
)
1521 struct radix_tree_iter iter
;
1523 unsigned int ret
= 0;
1525 if (unlikely(!nr_pages
))
1529 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1530 struct page
*head
, *page
;
1532 page
= radix_tree_deref_slot(slot
);
1533 /* The hole, there no reason to continue */
1534 if (unlikely(!page
))
1537 if (radix_tree_exception(page
)) {
1538 if (radix_tree_deref_retry(page
)) {
1539 slot
= radix_tree_iter_retry(&iter
);
1543 * A shadow entry of a recently evicted page,
1544 * or a swap entry from shmem/tmpfs. Stop
1545 * looking for contiguous pages.
1550 head
= compound_head(page
);
1551 if (!page_cache_get_speculative(head
))
1554 /* The page was split under us? */
1555 if (compound_head(page
) != head
) {
1560 /* Has the page moved? */
1561 if (unlikely(page
!= *slot
)) {
1567 * must check mapping and index after taking the ref.
1568 * otherwise we can get both false positives and false
1569 * negatives, which is just confusing to the caller.
1571 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1577 if (++ret
== nr_pages
)
1583 EXPORT_SYMBOL(find_get_pages_contig
);
1586 * find_get_pages_tag - find and return pages that match @tag
1587 * @mapping: the address_space to search
1588 * @index: the starting page index
1589 * @tag: the tag index
1590 * @nr_pages: the maximum number of pages
1591 * @pages: where the resulting pages are placed
1593 * Like find_get_pages, except we only return pages which are tagged with
1594 * @tag. We update @index to index the next page for the traversal.
1596 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1597 int tag
, unsigned int nr_pages
, struct page
**pages
)
1599 struct radix_tree_iter iter
;
1603 if (unlikely(!nr_pages
))
1607 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1608 &iter
, *index
, tag
) {
1609 struct page
*head
, *page
;
1611 page
= radix_tree_deref_slot(slot
);
1612 if (unlikely(!page
))
1615 if (radix_tree_exception(page
)) {
1616 if (radix_tree_deref_retry(page
)) {
1617 slot
= radix_tree_iter_retry(&iter
);
1621 * A shadow entry of a recently evicted page.
1623 * Those entries should never be tagged, but
1624 * this tree walk is lockless and the tags are
1625 * looked up in bulk, one radix tree node at a
1626 * time, so there is a sizable window for page
1627 * reclaim to evict a page we saw tagged.
1634 head
= compound_head(page
);
1635 if (!page_cache_get_speculative(head
))
1638 /* The page was split under us? */
1639 if (compound_head(page
) != head
) {
1644 /* Has the page moved? */
1645 if (unlikely(page
!= *slot
)) {
1651 if (++ret
== nr_pages
)
1658 *index
= pages
[ret
- 1]->index
+ 1;
1662 EXPORT_SYMBOL(find_get_pages_tag
);
1665 * find_get_entries_tag - find and return entries that match @tag
1666 * @mapping: the address_space to search
1667 * @start: the starting page cache index
1668 * @tag: the tag index
1669 * @nr_entries: the maximum number of entries
1670 * @entries: where the resulting entries are placed
1671 * @indices: the cache indices corresponding to the entries in @entries
1673 * Like find_get_entries, except we only return entries which are tagged with
1676 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1677 int tag
, unsigned int nr_entries
,
1678 struct page
**entries
, pgoff_t
*indices
)
1681 unsigned int ret
= 0;
1682 struct radix_tree_iter iter
;
1688 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1689 &iter
, start
, tag
) {
1690 struct page
*head
, *page
;
1692 page
= radix_tree_deref_slot(slot
);
1693 if (unlikely(!page
))
1695 if (radix_tree_exception(page
)) {
1696 if (radix_tree_deref_retry(page
)) {
1697 slot
= radix_tree_iter_retry(&iter
);
1702 * A shadow entry of a recently evicted page, a swap
1703 * entry from shmem/tmpfs or a DAX entry. Return it
1704 * without attempting to raise page count.
1709 head
= compound_head(page
);
1710 if (!page_cache_get_speculative(head
))
1713 /* The page was split under us? */
1714 if (compound_head(page
) != head
) {
1719 /* Has the page moved? */
1720 if (unlikely(page
!= *slot
)) {
1725 indices
[ret
] = iter
.index
;
1726 entries
[ret
] = page
;
1727 if (++ret
== nr_entries
)
1733 EXPORT_SYMBOL(find_get_entries_tag
);
1736 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1737 * a _large_ part of the i/o request. Imagine the worst scenario:
1739 * ---R__________________________________________B__________
1740 * ^ reading here ^ bad block(assume 4k)
1742 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1743 * => failing the whole request => read(R) => read(R+1) =>
1744 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1745 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1746 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1748 * It is going insane. Fix it by quickly scaling down the readahead size.
1750 static void shrink_readahead_size_eio(struct file
*filp
,
1751 struct file_ra_state
*ra
)
1757 * do_generic_file_read - generic file read routine
1758 * @filp: the file to read
1759 * @ppos: current file position
1760 * @iter: data destination
1761 * @written: already copied
1763 * This is a generic file read routine, and uses the
1764 * mapping->a_ops->readpage() function for the actual low-level stuff.
1766 * This is really ugly. But the goto's actually try to clarify some
1767 * of the logic when it comes to error handling etc.
1769 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1770 struct iov_iter
*iter
, ssize_t written
)
1772 struct address_space
*mapping
= filp
->f_mapping
;
1773 struct inode
*inode
= mapping
->host
;
1774 struct file_ra_state
*ra
= &filp
->f_ra
;
1778 unsigned long offset
; /* offset into pagecache page */
1779 unsigned int prev_offset
;
1782 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
1784 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
1786 index
= *ppos
>> PAGE_SHIFT
;
1787 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1788 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1789 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1790 offset
= *ppos
& ~PAGE_MASK
;
1796 unsigned long nr
, ret
;
1800 if (fatal_signal_pending(current
)) {
1805 page
= find_get_page(mapping
, index
);
1807 page_cache_sync_readahead(mapping
,
1809 index
, last_index
- index
);
1810 page
= find_get_page(mapping
, index
);
1811 if (unlikely(page
== NULL
))
1812 goto no_cached_page
;
1814 if (PageReadahead(page
)) {
1815 page_cache_async_readahead(mapping
,
1817 index
, last_index
- index
);
1819 if (!PageUptodate(page
)) {
1821 * See comment in do_read_cache_page on why
1822 * wait_on_page_locked is used to avoid unnecessarily
1823 * serialisations and why it's safe.
1825 error
= wait_on_page_locked_killable(page
);
1826 if (unlikely(error
))
1827 goto readpage_error
;
1828 if (PageUptodate(page
))
1831 if (inode
->i_blkbits
== PAGE_SHIFT
||
1832 !mapping
->a_ops
->is_partially_uptodate
)
1833 goto page_not_up_to_date
;
1834 /* pipes can't handle partially uptodate pages */
1835 if (unlikely(iter
->type
& ITER_PIPE
))
1836 goto page_not_up_to_date
;
1837 if (!trylock_page(page
))
1838 goto page_not_up_to_date
;
1839 /* Did it get truncated before we got the lock? */
1841 goto page_not_up_to_date_locked
;
1842 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1843 offset
, iter
->count
))
1844 goto page_not_up_to_date_locked
;
1849 * i_size must be checked after we know the page is Uptodate.
1851 * Checking i_size after the check allows us to calculate
1852 * the correct value for "nr", which means the zero-filled
1853 * part of the page is not copied back to userspace (unless
1854 * another truncate extends the file - this is desired though).
1857 isize
= i_size_read(inode
);
1858 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1859 if (unlikely(!isize
|| index
> end_index
)) {
1864 /* nr is the maximum number of bytes to copy from this page */
1866 if (index
== end_index
) {
1867 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1875 /* If users can be writing to this page using arbitrary
1876 * virtual addresses, take care about potential aliasing
1877 * before reading the page on the kernel side.
1879 if (mapping_writably_mapped(mapping
))
1880 flush_dcache_page(page
);
1883 * When a sequential read accesses a page several times,
1884 * only mark it as accessed the first time.
1886 if (prev_index
!= index
|| offset
!= prev_offset
)
1887 mark_page_accessed(page
);
1891 * Ok, we have the page, and it's up-to-date, so
1892 * now we can copy it to user space...
1895 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1897 index
+= offset
>> PAGE_SHIFT
;
1898 offset
&= ~PAGE_MASK
;
1899 prev_offset
= offset
;
1903 if (!iov_iter_count(iter
))
1911 page_not_up_to_date
:
1912 /* Get exclusive access to the page ... */
1913 error
= lock_page_killable(page
);
1914 if (unlikely(error
))
1915 goto readpage_error
;
1917 page_not_up_to_date_locked
:
1918 /* Did it get truncated before we got the lock? */
1919 if (!page
->mapping
) {
1925 /* Did somebody else fill it already? */
1926 if (PageUptodate(page
)) {
1933 * A previous I/O error may have been due to temporary
1934 * failures, eg. multipath errors.
1935 * PG_error will be set again if readpage fails.
1937 ClearPageError(page
);
1938 /* Start the actual read. The read will unlock the page. */
1939 error
= mapping
->a_ops
->readpage(filp
, page
);
1941 if (unlikely(error
)) {
1942 if (error
== AOP_TRUNCATED_PAGE
) {
1947 goto readpage_error
;
1950 if (!PageUptodate(page
)) {
1951 error
= lock_page_killable(page
);
1952 if (unlikely(error
))
1953 goto readpage_error
;
1954 if (!PageUptodate(page
)) {
1955 if (page
->mapping
== NULL
) {
1957 * invalidate_mapping_pages got it
1964 shrink_readahead_size_eio(filp
, ra
);
1966 goto readpage_error
;
1974 /* UHHUH! A synchronous read error occurred. Report it */
1980 * Ok, it wasn't cached, so we need to create a new
1983 page
= page_cache_alloc_cold(mapping
);
1988 error
= add_to_page_cache_lru(page
, mapping
, index
,
1989 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1992 if (error
== -EEXIST
) {
2002 ra
->prev_pos
= prev_index
;
2003 ra
->prev_pos
<<= PAGE_SHIFT
;
2004 ra
->prev_pos
|= prev_offset
;
2006 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2007 file_accessed(filp
);
2008 return written
? written
: error
;
2012 * generic_file_read_iter - generic filesystem read routine
2013 * @iocb: kernel I/O control block
2014 * @iter: destination for the data read
2016 * This is the "read_iter()" routine for all filesystems
2017 * that can use the page cache directly.
2020 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2022 struct file
*file
= iocb
->ki_filp
;
2024 size_t count
= iov_iter_count(iter
);
2027 goto out
; /* skip atime */
2029 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2030 struct address_space
*mapping
= file
->f_mapping
;
2031 struct inode
*inode
= mapping
->host
;
2032 struct iov_iter data
= *iter
;
2035 size
= i_size_read(inode
);
2036 retval
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
2037 iocb
->ki_pos
+ count
- 1);
2041 file_accessed(file
);
2043 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2045 iocb
->ki_pos
+= retval
;
2046 iov_iter_advance(iter
, retval
);
2050 * Btrfs can have a short DIO read if we encounter
2051 * compressed extents, so if there was an error, or if
2052 * we've already read everything we wanted to, or if
2053 * there was a short read because we hit EOF, go ahead
2054 * and return. Otherwise fallthrough to buffered io for
2055 * the rest of the read. Buffered reads will not work for
2056 * DAX files, so don't bother trying.
2058 if (retval
< 0 || !iov_iter_count(iter
) || iocb
->ki_pos
>= size
||
2063 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
2067 EXPORT_SYMBOL(generic_file_read_iter
);
2071 * page_cache_read - adds requested page to the page cache if not already there
2072 * @file: file to read
2073 * @offset: page index
2074 * @gfp_mask: memory allocation flags
2076 * This adds the requested page to the page cache if it isn't already there,
2077 * and schedules an I/O to read in its contents from disk.
2079 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
2081 struct address_space
*mapping
= file
->f_mapping
;
2086 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
2090 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
2092 ret
= mapping
->a_ops
->readpage(file
, page
);
2093 else if (ret
== -EEXIST
)
2094 ret
= 0; /* losing race to add is OK */
2098 } while (ret
== AOP_TRUNCATED_PAGE
);
2103 #define MMAP_LOTSAMISS (100)
2106 * Synchronous readahead happens when we don't even find
2107 * a page in the page cache at all.
2109 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2110 struct file_ra_state
*ra
,
2114 struct address_space
*mapping
= file
->f_mapping
;
2116 /* If we don't want any read-ahead, don't bother */
2117 if (vma
->vm_flags
& VM_RAND_READ
)
2122 if (vma
->vm_flags
& VM_SEQ_READ
) {
2123 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2128 /* Avoid banging the cache line if not needed */
2129 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2133 * Do we miss much more than hit in this file? If so,
2134 * stop bothering with read-ahead. It will only hurt.
2136 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2142 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2143 ra
->size
= ra
->ra_pages
;
2144 ra
->async_size
= ra
->ra_pages
/ 4;
2145 ra_submit(ra
, mapping
, file
);
2149 * Asynchronous readahead happens when we find the page and PG_readahead,
2150 * so we want to possibly extend the readahead further..
2152 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2153 struct file_ra_state
*ra
,
2158 struct address_space
*mapping
= file
->f_mapping
;
2160 /* If we don't want any read-ahead, don't bother */
2161 if (vma
->vm_flags
& VM_RAND_READ
)
2163 if (ra
->mmap_miss
> 0)
2165 if (PageReadahead(page
))
2166 page_cache_async_readahead(mapping
, ra
, file
,
2167 page
, offset
, ra
->ra_pages
);
2171 * filemap_fault - read in file data for page fault handling
2172 * @vmf: struct vm_fault containing details of the fault
2174 * filemap_fault() is invoked via the vma operations vector for a
2175 * mapped memory region to read in file data during a page fault.
2177 * The goto's are kind of ugly, but this streamlines the normal case of having
2178 * it in the page cache, and handles the special cases reasonably without
2179 * having a lot of duplicated code.
2181 * vma->vm_mm->mmap_sem must be held on entry.
2183 * If our return value has VM_FAULT_RETRY set, it's because
2184 * lock_page_or_retry() returned 0.
2185 * The mmap_sem has usually been released in this case.
2186 * See __lock_page_or_retry() for the exception.
2188 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2189 * has not been released.
2191 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2193 int filemap_fault(struct vm_fault
*vmf
)
2196 struct file
*file
= vmf
->vma
->vm_file
;
2197 struct address_space
*mapping
= file
->f_mapping
;
2198 struct file_ra_state
*ra
= &file
->f_ra
;
2199 struct inode
*inode
= mapping
->host
;
2200 pgoff_t offset
= vmf
->pgoff
;
2205 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2206 if (offset
>= size
>> PAGE_SHIFT
)
2207 return VM_FAULT_SIGBUS
;
2210 * Do we have something in the page cache already?
2212 page
= find_get_page(mapping
, offset
);
2213 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2215 * We found the page, so try async readahead before
2216 * waiting for the lock.
2218 do_async_mmap_readahead(vmf
->vma
, ra
, file
, page
, offset
);
2220 /* No page in the page cache at all */
2221 do_sync_mmap_readahead(vmf
->vma
, ra
, file
, offset
);
2222 count_vm_event(PGMAJFAULT
);
2223 mem_cgroup_count_vm_event(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2224 ret
= VM_FAULT_MAJOR
;
2226 page
= find_get_page(mapping
, offset
);
2228 goto no_cached_page
;
2231 if (!lock_page_or_retry(page
, vmf
->vma
->vm_mm
, vmf
->flags
)) {
2233 return ret
| VM_FAULT_RETRY
;
2236 /* Did it get truncated? */
2237 if (unlikely(page
->mapping
!= mapping
)) {
2242 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2245 * We have a locked page in the page cache, now we need to check
2246 * that it's up-to-date. If not, it is going to be due to an error.
2248 if (unlikely(!PageUptodate(page
)))
2249 goto page_not_uptodate
;
2252 * Found the page and have a reference on it.
2253 * We must recheck i_size under page lock.
2255 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2256 if (unlikely(offset
>= size
>> PAGE_SHIFT
)) {
2259 return VM_FAULT_SIGBUS
;
2263 return ret
| VM_FAULT_LOCKED
;
2267 * We're only likely to ever get here if MADV_RANDOM is in
2270 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2273 * The page we want has now been added to the page cache.
2274 * In the unlikely event that someone removed it in the
2275 * meantime, we'll just come back here and read it again.
2281 * An error return from page_cache_read can result if the
2282 * system is low on memory, or a problem occurs while trying
2285 if (error
== -ENOMEM
)
2286 return VM_FAULT_OOM
;
2287 return VM_FAULT_SIGBUS
;
2291 * Umm, take care of errors if the page isn't up-to-date.
2292 * Try to re-read it _once_. We do this synchronously,
2293 * because there really aren't any performance issues here
2294 * and we need to check for errors.
2296 ClearPageError(page
);
2297 error
= mapping
->a_ops
->readpage(file
, page
);
2299 wait_on_page_locked(page
);
2300 if (!PageUptodate(page
))
2305 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2308 /* Things didn't work out. Return zero to tell the mm layer so. */
2309 shrink_readahead_size_eio(file
, ra
);
2310 return VM_FAULT_SIGBUS
;
2312 EXPORT_SYMBOL(filemap_fault
);
2314 void filemap_map_pages(struct vm_fault
*vmf
,
2315 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2317 struct radix_tree_iter iter
;
2319 struct file
*file
= vmf
->vma
->vm_file
;
2320 struct address_space
*mapping
= file
->f_mapping
;
2321 pgoff_t last_pgoff
= start_pgoff
;
2323 struct page
*head
, *page
;
2326 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2328 if (iter
.index
> end_pgoff
)
2331 page
= radix_tree_deref_slot(slot
);
2332 if (unlikely(!page
))
2334 if (radix_tree_exception(page
)) {
2335 if (radix_tree_deref_retry(page
)) {
2336 slot
= radix_tree_iter_retry(&iter
);
2342 head
= compound_head(page
);
2343 if (!page_cache_get_speculative(head
))
2346 /* The page was split under us? */
2347 if (compound_head(page
) != head
) {
2352 /* Has the page moved? */
2353 if (unlikely(page
!= *slot
)) {
2358 if (!PageUptodate(page
) ||
2359 PageReadahead(page
) ||
2362 if (!trylock_page(page
))
2365 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2368 size
= round_up(i_size_read(mapping
->host
), PAGE_SIZE
);
2369 if (page
->index
>= size
>> PAGE_SHIFT
)
2372 if (file
->f_ra
.mmap_miss
> 0)
2373 file
->f_ra
.mmap_miss
--;
2375 vmf
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2377 vmf
->pte
+= iter
.index
- last_pgoff
;
2378 last_pgoff
= iter
.index
;
2379 if (alloc_set_pte(vmf
, NULL
, page
))
2388 /* Huge page is mapped? No need to proceed. */
2389 if (pmd_trans_huge(*vmf
->pmd
))
2391 if (iter
.index
== end_pgoff
)
2396 EXPORT_SYMBOL(filemap_map_pages
);
2398 int filemap_page_mkwrite(struct vm_fault
*vmf
)
2400 struct page
*page
= vmf
->page
;
2401 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2402 int ret
= VM_FAULT_LOCKED
;
2404 sb_start_pagefault(inode
->i_sb
);
2405 file_update_time(vmf
->vma
->vm_file
);
2407 if (page
->mapping
!= inode
->i_mapping
) {
2409 ret
= VM_FAULT_NOPAGE
;
2413 * We mark the page dirty already here so that when freeze is in
2414 * progress, we are guaranteed that writeback during freezing will
2415 * see the dirty page and writeprotect it again.
2417 set_page_dirty(page
);
2418 wait_for_stable_page(page
);
2420 sb_end_pagefault(inode
->i_sb
);
2423 EXPORT_SYMBOL(filemap_page_mkwrite
);
2425 const struct vm_operations_struct generic_file_vm_ops
= {
2426 .fault
= filemap_fault
,
2427 .map_pages
= filemap_map_pages
,
2428 .page_mkwrite
= filemap_page_mkwrite
,
2431 /* This is used for a general mmap of a disk file */
2433 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2435 struct address_space
*mapping
= file
->f_mapping
;
2437 if (!mapping
->a_ops
->readpage
)
2439 file_accessed(file
);
2440 vma
->vm_ops
= &generic_file_vm_ops
;
2445 * This is for filesystems which do not implement ->writepage.
2447 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2449 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2451 return generic_file_mmap(file
, vma
);
2454 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2458 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2462 #endif /* CONFIG_MMU */
2464 EXPORT_SYMBOL(generic_file_mmap
);
2465 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2467 static struct page
*wait_on_page_read(struct page
*page
)
2469 if (!IS_ERR(page
)) {
2470 wait_on_page_locked(page
);
2471 if (!PageUptodate(page
)) {
2473 page
= ERR_PTR(-EIO
);
2479 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2481 int (*filler
)(void *, struct page
*),
2488 page
= find_get_page(mapping
, index
);
2490 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2492 return ERR_PTR(-ENOMEM
);
2493 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2494 if (unlikely(err
)) {
2498 /* Presumably ENOMEM for radix tree node */
2499 return ERR_PTR(err
);
2503 err
= filler(data
, page
);
2506 return ERR_PTR(err
);
2509 page
= wait_on_page_read(page
);
2514 if (PageUptodate(page
))
2518 * Page is not up to date and may be locked due one of the following
2519 * case a: Page is being filled and the page lock is held
2520 * case b: Read/write error clearing the page uptodate status
2521 * case c: Truncation in progress (page locked)
2522 * case d: Reclaim in progress
2524 * Case a, the page will be up to date when the page is unlocked.
2525 * There is no need to serialise on the page lock here as the page
2526 * is pinned so the lock gives no additional protection. Even if the
2527 * the page is truncated, the data is still valid if PageUptodate as
2528 * it's a race vs truncate race.
2529 * Case b, the page will not be up to date
2530 * Case c, the page may be truncated but in itself, the data may still
2531 * be valid after IO completes as it's a read vs truncate race. The
2532 * operation must restart if the page is not uptodate on unlock but
2533 * otherwise serialising on page lock to stabilise the mapping gives
2534 * no additional guarantees to the caller as the page lock is
2535 * released before return.
2536 * Case d, similar to truncation. If reclaim holds the page lock, it
2537 * will be a race with remove_mapping that determines if the mapping
2538 * is valid on unlock but otherwise the data is valid and there is
2539 * no need to serialise with page lock.
2541 * As the page lock gives no additional guarantee, we optimistically
2542 * wait on the page to be unlocked and check if it's up to date and
2543 * use the page if it is. Otherwise, the page lock is required to
2544 * distinguish between the different cases. The motivation is that we
2545 * avoid spurious serialisations and wakeups when multiple processes
2546 * wait on the same page for IO to complete.
2548 wait_on_page_locked(page
);
2549 if (PageUptodate(page
))
2552 /* Distinguish between all the cases under the safety of the lock */
2555 /* Case c or d, restart the operation */
2556 if (!page
->mapping
) {
2562 /* Someone else locked and filled the page in a very small window */
2563 if (PageUptodate(page
)) {
2570 mark_page_accessed(page
);
2575 * read_cache_page - read into page cache, fill it if needed
2576 * @mapping: the page's address_space
2577 * @index: the page index
2578 * @filler: function to perform the read
2579 * @data: first arg to filler(data, page) function, often left as NULL
2581 * Read into the page cache. If a page already exists, and PageUptodate() is
2582 * not set, try to fill the page and wait for it to become unlocked.
2584 * If the page does not get brought uptodate, return -EIO.
2586 struct page
*read_cache_page(struct address_space
*mapping
,
2588 int (*filler
)(void *, struct page
*),
2591 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2593 EXPORT_SYMBOL(read_cache_page
);
2596 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2597 * @mapping: the page's address_space
2598 * @index: the page index
2599 * @gfp: the page allocator flags to use if allocating
2601 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2602 * any new page allocations done using the specified allocation flags.
2604 * If the page does not get brought uptodate, return -EIO.
2606 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2610 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2612 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2614 EXPORT_SYMBOL(read_cache_page_gfp
);
2617 * Performs necessary checks before doing a write
2619 * Can adjust writing position or amount of bytes to write.
2620 * Returns appropriate error code that caller should return or
2621 * zero in case that write should be allowed.
2623 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2625 struct file
*file
= iocb
->ki_filp
;
2626 struct inode
*inode
= file
->f_mapping
->host
;
2627 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2630 if (!iov_iter_count(from
))
2633 /* FIXME: this is for backwards compatibility with 2.4 */
2634 if (iocb
->ki_flags
& IOCB_APPEND
)
2635 iocb
->ki_pos
= i_size_read(inode
);
2639 if (limit
!= RLIM_INFINITY
) {
2640 if (iocb
->ki_pos
>= limit
) {
2641 send_sig(SIGXFSZ
, current
, 0);
2644 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2650 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2651 !(file
->f_flags
& O_LARGEFILE
))) {
2652 if (pos
>= MAX_NON_LFS
)
2654 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2658 * Are we about to exceed the fs block limit ?
2660 * If we have written data it becomes a short write. If we have
2661 * exceeded without writing data we send a signal and return EFBIG.
2662 * Linus frestrict idea will clean these up nicely..
2664 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2667 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2668 return iov_iter_count(from
);
2670 EXPORT_SYMBOL(generic_write_checks
);
2672 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2673 loff_t pos
, unsigned len
, unsigned flags
,
2674 struct page
**pagep
, void **fsdata
)
2676 const struct address_space_operations
*aops
= mapping
->a_ops
;
2678 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2681 EXPORT_SYMBOL(pagecache_write_begin
);
2683 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2684 loff_t pos
, unsigned len
, unsigned copied
,
2685 struct page
*page
, void *fsdata
)
2687 const struct address_space_operations
*aops
= mapping
->a_ops
;
2689 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2691 EXPORT_SYMBOL(pagecache_write_end
);
2694 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2696 struct file
*file
= iocb
->ki_filp
;
2697 struct address_space
*mapping
= file
->f_mapping
;
2698 struct inode
*inode
= mapping
->host
;
2699 loff_t pos
= iocb
->ki_pos
;
2703 struct iov_iter data
;
2705 write_len
= iov_iter_count(from
);
2706 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2708 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2713 * After a write we want buffered reads to be sure to go to disk to get
2714 * the new data. We invalidate clean cached page from the region we're
2715 * about to write. We do this *before* the write so that we can return
2716 * without clobbering -EIOCBQUEUED from ->direct_IO().
2718 if (mapping
->nrpages
) {
2719 written
= invalidate_inode_pages2_range(mapping
,
2720 pos
>> PAGE_SHIFT
, end
);
2722 * If a page can not be invalidated, return 0 to fall back
2723 * to buffered write.
2726 if (written
== -EBUSY
)
2733 written
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2736 * Finally, try again to invalidate clean pages which might have been
2737 * cached by non-direct readahead, or faulted in by get_user_pages()
2738 * if the source of the write was an mmap'ed region of the file
2739 * we're writing. Either one is a pretty crazy thing to do,
2740 * so we don't support it 100%. If this invalidation
2741 * fails, tough, the write still worked...
2743 if (mapping
->nrpages
) {
2744 invalidate_inode_pages2_range(mapping
,
2745 pos
>> PAGE_SHIFT
, end
);
2750 iov_iter_advance(from
, written
);
2751 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2752 i_size_write(inode
, pos
);
2753 mark_inode_dirty(inode
);
2760 EXPORT_SYMBOL(generic_file_direct_write
);
2763 * Find or create a page at the given pagecache position. Return the locked
2764 * page. This function is specifically for buffered writes.
2766 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2767 pgoff_t index
, unsigned flags
)
2770 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2772 if (flags
& AOP_FLAG_NOFS
)
2773 fgp_flags
|= FGP_NOFS
;
2775 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2776 mapping_gfp_mask(mapping
));
2778 wait_for_stable_page(page
);
2782 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2784 ssize_t
generic_perform_write(struct file
*file
,
2785 struct iov_iter
*i
, loff_t pos
)
2787 struct address_space
*mapping
= file
->f_mapping
;
2788 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2790 ssize_t written
= 0;
2791 unsigned int flags
= 0;
2794 * Copies from kernel address space cannot fail (NFSD is a big user).
2796 if (!iter_is_iovec(i
))
2797 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2801 unsigned long offset
; /* Offset into pagecache page */
2802 unsigned long bytes
; /* Bytes to write to page */
2803 size_t copied
; /* Bytes copied from user */
2806 offset
= (pos
& (PAGE_SIZE
- 1));
2807 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2812 * Bring in the user page that we will copy from _first_.
2813 * Otherwise there's a nasty deadlock on copying from the
2814 * same page as we're writing to, without it being marked
2817 * Not only is this an optimisation, but it is also required
2818 * to check that the address is actually valid, when atomic
2819 * usercopies are used, below.
2821 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2826 if (fatal_signal_pending(current
)) {
2831 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2833 if (unlikely(status
< 0))
2836 if (mapping_writably_mapped(mapping
))
2837 flush_dcache_page(page
);
2839 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2840 flush_dcache_page(page
);
2842 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2844 if (unlikely(status
< 0))
2850 iov_iter_advance(i
, copied
);
2851 if (unlikely(copied
== 0)) {
2853 * If we were unable to copy any data at all, we must
2854 * fall back to a single segment length write.
2856 * If we didn't fallback here, we could livelock
2857 * because not all segments in the iov can be copied at
2858 * once without a pagefault.
2860 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2861 iov_iter_single_seg_count(i
));
2867 balance_dirty_pages_ratelimited(mapping
);
2868 } while (iov_iter_count(i
));
2870 return written
? written
: status
;
2872 EXPORT_SYMBOL(generic_perform_write
);
2875 * __generic_file_write_iter - write data to a file
2876 * @iocb: IO state structure (file, offset, etc.)
2877 * @from: iov_iter with data to write
2879 * This function does all the work needed for actually writing data to a
2880 * file. It does all basic checks, removes SUID from the file, updates
2881 * modification times and calls proper subroutines depending on whether we
2882 * do direct IO or a standard buffered write.
2884 * It expects i_mutex to be grabbed unless we work on a block device or similar
2885 * object which does not need locking at all.
2887 * This function does *not* take care of syncing data in case of O_SYNC write.
2888 * A caller has to handle it. This is mainly due to the fact that we want to
2889 * avoid syncing under i_mutex.
2891 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2893 struct file
*file
= iocb
->ki_filp
;
2894 struct address_space
* mapping
= file
->f_mapping
;
2895 struct inode
*inode
= mapping
->host
;
2896 ssize_t written
= 0;
2900 /* We can write back this queue in page reclaim */
2901 current
->backing_dev_info
= inode_to_bdi(inode
);
2902 err
= file_remove_privs(file
);
2906 err
= file_update_time(file
);
2910 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2911 loff_t pos
, endbyte
;
2913 written
= generic_file_direct_write(iocb
, from
);
2915 * If the write stopped short of completing, fall back to
2916 * buffered writes. Some filesystems do this for writes to
2917 * holes, for example. For DAX files, a buffered write will
2918 * not succeed (even if it did, DAX does not handle dirty
2919 * page-cache pages correctly).
2921 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2924 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2926 * If generic_perform_write() returned a synchronous error
2927 * then we want to return the number of bytes which were
2928 * direct-written, or the error code if that was zero. Note
2929 * that this differs from normal direct-io semantics, which
2930 * will return -EFOO even if some bytes were written.
2932 if (unlikely(status
< 0)) {
2937 * We need to ensure that the page cache pages are written to
2938 * disk and invalidated to preserve the expected O_DIRECT
2941 endbyte
= pos
+ status
- 1;
2942 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2944 iocb
->ki_pos
= endbyte
+ 1;
2946 invalidate_mapping_pages(mapping
,
2948 endbyte
>> PAGE_SHIFT
);
2951 * We don't know how much we wrote, so just return
2952 * the number of bytes which were direct-written
2956 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2957 if (likely(written
> 0))
2958 iocb
->ki_pos
+= written
;
2961 current
->backing_dev_info
= NULL
;
2962 return written
? written
: err
;
2964 EXPORT_SYMBOL(__generic_file_write_iter
);
2967 * generic_file_write_iter - write data to a file
2968 * @iocb: IO state structure
2969 * @from: iov_iter with data to write
2971 * This is a wrapper around __generic_file_write_iter() to be used by most
2972 * filesystems. It takes care of syncing the file in case of O_SYNC file
2973 * and acquires i_mutex as needed.
2975 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2977 struct file
*file
= iocb
->ki_filp
;
2978 struct inode
*inode
= file
->f_mapping
->host
;
2982 ret
= generic_write_checks(iocb
, from
);
2984 ret
= __generic_file_write_iter(iocb
, from
);
2985 inode_unlock(inode
);
2988 ret
= generic_write_sync(iocb
, ret
);
2991 EXPORT_SYMBOL(generic_file_write_iter
);
2994 * try_to_release_page() - release old fs-specific metadata on a page
2996 * @page: the page which the kernel is trying to free
2997 * @gfp_mask: memory allocation flags (and I/O mode)
2999 * The address_space is to try to release any data against the page
3000 * (presumably at page->private). If the release was successful, return `1'.
3001 * Otherwise return zero.
3003 * This may also be called if PG_fscache is set on a page, indicating that the
3004 * page is known to the local caching routines.
3006 * The @gfp_mask argument specifies whether I/O may be performed to release
3007 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3010 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3012 struct address_space
* const mapping
= page
->mapping
;
3014 BUG_ON(!PageLocked(page
));
3015 if (PageWriteback(page
))
3018 if (mapping
&& mapping
->a_ops
->releasepage
)
3019 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3020 return try_to_free_buffers(page
);
3023 EXPORT_SYMBOL(try_to_release_page
);