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>
15 #include <linux/uaccess.h>
16 #include <linux/aio.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_mutex (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_mutex (truncate->unmap_mapping_range)
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
79 * ->lock_page (access_process_vm)
81 * ->i_mutex (generic_perform_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
85 * sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
112 static void page_cache_tree_delete(struct address_space *mapping,
113 struct page *page, void *shadow)
115 struct radix_tree_node *node;
121 VM_BUG_ON(!PageLocked(page));
123 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
126 mapping->nrshadows++;
128 * Make sure the nrshadows update is committed before
129 * the nrpages update so that final truncate racing
130 * with reclaim does not see both counters 0 at the
131 * same time and miss a shadow entry.
138 /* Clear direct pointer tags in root node */
139 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
140 radix_tree_replace_slot(slot, shadow);
144 /* Clear tree tags for the removed page */
146 offset = index & RADIX_TREE_MAP_MASK;
147 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
148 if (test_bit(offset, node->tags[tag]))
149 radix_tree_tag_clear(&mapping->page_tree, index, tag);
152 /* Delete page, swap shadow entry */
153 radix_tree_replace_slot(slot, shadow);
154 workingset_node_pages_dec(node);
156 workingset_node_shadows_inc(node);
158 if (__radix_tree_delete_node(&mapping->page_tree, node))
162 * Track node that only contains shadow entries.
164 * Avoid acquiring the list_lru lock if already tracked. The
165 * list_empty() test is safe as node->private_list is
166 * protected by mapping->tree_lock.
168 if (!workingset_node_pages(node) &&
169 list_empty(&node->private_list)) {
170 node->private_data = mapping;
171 list_lru_add(&workingset_shadow_nodes, &node->private_list);
176 * Delete a page from the page cache and free it. Caller has to make
177 * sure the page is locked and that nobody else uses it - or that usage
178 * is safe. The caller must hold the mapping's tree_lock.
180 void __delete_from_page_cache(struct page *page, void *shadow)
182 struct address_space *mapping = page->mapping;
184 trace_mm_filemap_delete_from_page_cache(page);
186 * if we're uptodate, flush out into the cleancache, otherwise
187 * invalidate any existing cleancache entries. We can't leave
188 * stale data around in the cleancache once our page is gone
190 if (PageUptodate(page) && PageMappedToDisk(page))
191 cleancache_put_page(page);
193 cleancache_invalidate_page(mapping, page);
195 page_cache_tree_delete(mapping, page, shadow);
197 page->mapping = NULL;
198 /* Leave page->index set: truncation lookup relies upon it */
200 __dec_zone_page_state(page, NR_FILE_PAGES);
201 if (PageSwapBacked(page))
202 __dec_zone_page_state(page, NR_SHMEM);
203 BUG_ON(page_mapped(page));
206 * Some filesystems seem to re-dirty the page even after
207 * the VM has canceled the dirty bit (eg ext3 journaling).
209 * Fix it up by doing a final dirty accounting check after
210 * having removed the page entirely.
212 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
213 dec_zone_page_state(page, NR_FILE_DIRTY);
214 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
219 * delete_from_page_cache - delete page from page cache
220 * @page: the page which the kernel is trying to remove from page cache
222 * This must be called only on pages that have been verified to be in the page
223 * cache and locked. It will never put the page into the free list, the caller
224 * has a reference on the page.
226 void delete_from_page_cache(struct page *page)
228 struct address_space *mapping = page->mapping;
229 void (*freepage)(struct page *);
231 BUG_ON(!PageLocked(page));
233 freepage = mapping->a_ops->freepage;
234 spin_lock_irq(&mapping->tree_lock);
235 __delete_from_page_cache(page, NULL);
236 spin_unlock_irq(&mapping->tree_lock);
237 mem_cgroup_uncharge_cache_page(page);
241 page_cache_release(page);
243 EXPORT_SYMBOL(delete_from_page_cache);
245 static int filemap_check_errors(struct address_space *mapping)
248 /* Check for outstanding write errors */
249 if (test_bit(AS_ENOSPC, &mapping->flags) &&
250 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
252 if (test_bit(AS_EIO, &mapping->flags) &&
253 test_and_clear_bit(AS_EIO, &mapping->flags))
259 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
260 * @mapping: address space structure to write
261 * @start: offset in bytes where the range starts
262 * @end: offset in bytes where the range ends (inclusive)
263 * @sync_mode: enable synchronous operation
265 * Start writeback against all of a mapping's dirty pages that lie
266 * within the byte offsets <start, end> inclusive.
268 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
269 * opposed to a regular memory cleansing writeback. The difference between
270 * these two operations is that if a dirty page/buffer is encountered, it must
271 * be waited upon, and not just skipped over.
273 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
274 loff_t end, int sync_mode)
277 struct writeback_control wbc = {
278 .sync_mode = sync_mode,
279 .nr_to_write = LONG_MAX,
280 .range_start = start,
284 if (!mapping_cap_writeback_dirty(mapping))
287 ret = do_writepages(mapping, &wbc);
291 static inline int __filemap_fdatawrite(struct address_space *mapping,
294 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
297 int filemap_fdatawrite(struct address_space *mapping)
299 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
301 EXPORT_SYMBOL(filemap_fdatawrite);
303 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
306 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
308 EXPORT_SYMBOL(filemap_fdatawrite_range);
311 * filemap_flush - mostly a non-blocking flush
312 * @mapping: target address_space
314 * This is a mostly non-blocking flush. Not suitable for data-integrity
315 * purposes - I/O may not be started against all dirty pages.
317 int filemap_flush(struct address_space *mapping)
319 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
321 EXPORT_SYMBOL(filemap_flush);
324 * filemap_fdatawait_range - wait for writeback to complete
325 * @mapping: address space structure to wait for
326 * @start_byte: offset in bytes where the range starts
327 * @end_byte: offset in bytes where the range ends (inclusive)
329 * Walk the list of under-writeback pages of the given address space
330 * in the given range and wait for all of them.
332 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
335 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
336 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
341 if (end_byte < start_byte)
344 pagevec_init(&pvec, 0);
345 while ((index <= end) &&
346 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
347 PAGECACHE_TAG_WRITEBACK,
348 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
351 for (i = 0; i < nr_pages; i++) {
352 struct page *page = pvec.pages[i];
354 /* until radix tree lookup accepts end_index */
355 if (page->index > end)
358 wait_on_page_writeback(page);
359 if (TestClearPageError(page))
362 pagevec_release(&pvec);
366 ret2 = filemap_check_errors(mapping);
372 EXPORT_SYMBOL(filemap_fdatawait_range);
375 * filemap_fdatawait - wait for all under-writeback pages to complete
376 * @mapping: address space structure to wait for
378 * Walk the list of under-writeback pages of the given address space
379 * and wait for all of them.
381 int filemap_fdatawait(struct address_space *mapping)
383 loff_t i_size = i_size_read(mapping->host);
388 return filemap_fdatawait_range(mapping, 0, i_size - 1);
390 EXPORT_SYMBOL(filemap_fdatawait);
392 int filemap_write_and_wait(struct address_space *mapping)
396 if (mapping->nrpages) {
397 err = filemap_fdatawrite(mapping);
399 * Even if the above returned error, the pages may be
400 * written partially (e.g. -ENOSPC), so we wait for it.
401 * But the -EIO is special case, it may indicate the worst
402 * thing (e.g. bug) happened, so we avoid waiting for it.
405 int err2 = filemap_fdatawait(mapping);
410 err = filemap_check_errors(mapping);
414 EXPORT_SYMBOL(filemap_write_and_wait);
417 * filemap_write_and_wait_range - write out & wait on a file range
418 * @mapping: the address_space for the pages
419 * @lstart: offset in bytes where the range starts
420 * @lend: offset in bytes where the range ends (inclusive)
422 * Write out and wait upon file offsets lstart->lend, inclusive.
424 * Note that `lend' is inclusive (describes the last byte to be written) so
425 * that this function can be used to write to the very end-of-file (end = -1).
427 int filemap_write_and_wait_range(struct address_space *mapping,
428 loff_t lstart, loff_t lend)
432 if (mapping->nrpages) {
433 err = __filemap_fdatawrite_range(mapping, lstart, lend,
435 /* See comment of filemap_write_and_wait() */
437 int err2 = filemap_fdatawait_range(mapping,
443 err = filemap_check_errors(mapping);
447 EXPORT_SYMBOL(filemap_write_and_wait_range);
450 * replace_page_cache_page - replace a pagecache page with a new one
451 * @old: page to be replaced
452 * @new: page to replace with
453 * @gfp_mask: allocation mode
455 * This function replaces a page in the pagecache with a new one. On
456 * success it acquires the pagecache reference for the new page and
457 * drops it for the old page. Both the old and new pages must be
458 * locked. This function does not add the new page to the LRU, the
459 * caller must do that.
461 * The remove + add is atomic. The only way this function can fail is
462 * memory allocation failure.
464 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
468 VM_BUG_ON_PAGE(!PageLocked(old), old);
469 VM_BUG_ON_PAGE(!PageLocked(new), new);
470 VM_BUG_ON_PAGE(new->mapping, new);
472 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
474 struct address_space *mapping = old->mapping;
475 void (*freepage)(struct page *);
477 pgoff_t offset = old->index;
478 freepage = mapping->a_ops->freepage;
481 new->mapping = mapping;
484 spin_lock_irq(&mapping->tree_lock);
485 __delete_from_page_cache(old, NULL);
486 error = radix_tree_insert(&mapping->page_tree, offset, new);
489 __inc_zone_page_state(new, NR_FILE_PAGES);
490 if (PageSwapBacked(new))
491 __inc_zone_page_state(new, NR_SHMEM);
492 spin_unlock_irq(&mapping->tree_lock);
493 /* mem_cgroup codes must not be called under tree_lock */
494 mem_cgroup_replace_page_cache(old, new);
495 radix_tree_preload_end();
498 page_cache_release(old);
503 EXPORT_SYMBOL_GPL(replace_page_cache_page);
505 static int page_cache_tree_insert(struct address_space *mapping,
506 struct page *page, void **shadowp)
508 struct radix_tree_node *node;
512 error = __radix_tree_create(&mapping->page_tree, page->index,
519 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
520 if (!radix_tree_exceptional_entry(p))
524 mapping->nrshadows--;
526 workingset_node_shadows_dec(node);
528 radix_tree_replace_slot(slot, page);
531 workingset_node_pages_inc(node);
533 * Don't track node that contains actual pages.
535 * Avoid acquiring the list_lru lock if already
536 * untracked. The list_empty() test is safe as
537 * node->private_list is protected by
538 * mapping->tree_lock.
540 if (!list_empty(&node->private_list))
541 list_lru_del(&workingset_shadow_nodes,
542 &node->private_list);
547 static int __add_to_page_cache_locked(struct page *page,
548 struct address_space *mapping,
549 pgoff_t offset, gfp_t gfp_mask,
552 int huge = PageHuge(page);
553 struct mem_cgroup *memcg;
556 VM_BUG_ON_PAGE(!PageLocked(page), page);
557 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
560 error = mem_cgroup_try_charge(page, current->mm,
566 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
569 mem_cgroup_cancel_charge(page, memcg);
573 page_cache_get(page);
574 page->mapping = mapping;
575 page->index = offset;
577 spin_lock_irq(&mapping->tree_lock);
578 error = page_cache_tree_insert(mapping, page, shadowp);
579 radix_tree_preload_end();
582 __inc_zone_page_state(page, NR_FILE_PAGES);
583 spin_unlock_irq(&mapping->tree_lock);
585 mem_cgroup_commit_charge(page, memcg, false);
586 trace_mm_filemap_add_to_page_cache(page);
589 page->mapping = NULL;
590 /* Leave page->index set: truncation relies upon it */
591 spin_unlock_irq(&mapping->tree_lock);
593 mem_cgroup_cancel_charge(page, memcg);
594 page_cache_release(page);
599 * add_to_page_cache_locked - add a locked page to the pagecache
601 * @mapping: the page's address_space
602 * @offset: page index
603 * @gfp_mask: page allocation mode
605 * This function is used to add a page to the pagecache. It must be locked.
606 * This function does not add the page to the LRU. The caller must do that.
608 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
609 pgoff_t offset, gfp_t gfp_mask)
611 return __add_to_page_cache_locked(page, mapping, offset,
614 EXPORT_SYMBOL(add_to_page_cache_locked);
616 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
617 pgoff_t offset, gfp_t gfp_mask)
622 __set_page_locked(page);
623 ret = __add_to_page_cache_locked(page, mapping, offset,
626 __clear_page_locked(page);
629 * The page might have been evicted from cache only
630 * recently, in which case it should be activated like
631 * any other repeatedly accessed page.
633 if (shadow && workingset_refault(shadow)) {
635 workingset_activation(page);
637 ClearPageActive(page);
642 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
645 struct page *__page_cache_alloc(gfp_t gfp)
650 if (cpuset_do_page_mem_spread()) {
651 unsigned int cpuset_mems_cookie;
653 cpuset_mems_cookie = read_mems_allowed_begin();
654 n = cpuset_mem_spread_node();
655 page = alloc_pages_exact_node(n, gfp, 0);
656 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
660 return alloc_pages(gfp, 0);
662 EXPORT_SYMBOL(__page_cache_alloc);
666 * In order to wait for pages to become available there must be
667 * waitqueues associated with pages. By using a hash table of
668 * waitqueues where the bucket discipline is to maintain all
669 * waiters on the same queue and wake all when any of the pages
670 * become available, and for the woken contexts to check to be
671 * sure the appropriate page became available, this saves space
672 * at a cost of "thundering herd" phenomena during rare hash
675 static wait_queue_head_t *page_waitqueue(struct page *page)
677 const struct zone *zone = page_zone(page);
679 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
682 static inline void wake_up_page(struct page *page, int bit)
684 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
687 void wait_on_page_bit(struct page *page, int bit_nr)
689 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
691 if (test_bit(bit_nr, &page->flags))
692 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
693 TASK_UNINTERRUPTIBLE);
695 EXPORT_SYMBOL(wait_on_page_bit);
697 int wait_on_page_bit_killable(struct page *page, int bit_nr)
699 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
701 if (!test_bit(bit_nr, &page->flags))
704 return __wait_on_bit(page_waitqueue(page), &wait,
705 bit_wait_io, TASK_KILLABLE);
709 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
710 * @page: Page defining the wait queue of interest
711 * @waiter: Waiter to add to the queue
713 * Add an arbitrary @waiter to the wait queue for the nominated @page.
715 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
717 wait_queue_head_t *q = page_waitqueue(page);
720 spin_lock_irqsave(&q->lock, flags);
721 __add_wait_queue(q, waiter);
722 spin_unlock_irqrestore(&q->lock, flags);
724 EXPORT_SYMBOL_GPL(add_page_wait_queue);
727 * unlock_page - unlock a locked page
730 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
731 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
732 * mechananism between PageLocked pages and PageWriteback pages is shared.
733 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
735 * The mb is necessary to enforce ordering between the clear_bit and the read
736 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
738 void unlock_page(struct page *page)
740 VM_BUG_ON_PAGE(!PageLocked(page), page);
741 clear_bit_unlock(PG_locked, &page->flags);
742 smp_mb__after_atomic();
743 wake_up_page(page, PG_locked);
745 EXPORT_SYMBOL(unlock_page);
748 * end_page_writeback - end writeback against a page
751 void end_page_writeback(struct page *page)
754 * TestClearPageReclaim could be used here but it is an atomic
755 * operation and overkill in this particular case. Failing to
756 * shuffle a page marked for immediate reclaim is too mild to
757 * justify taking an atomic operation penalty at the end of
758 * ever page writeback.
760 if (PageReclaim(page)) {
761 ClearPageReclaim(page);
762 rotate_reclaimable_page(page);
765 if (!test_clear_page_writeback(page))
768 smp_mb__after_atomic();
769 wake_up_page(page, PG_writeback);
771 EXPORT_SYMBOL(end_page_writeback);
774 * After completing I/O on a page, call this routine to update the page
775 * flags appropriately
777 void page_endio(struct page *page, int rw, int err)
781 SetPageUptodate(page);
783 ClearPageUptodate(page);
787 } else { /* rw == WRITE */
791 mapping_set_error(page->mapping, err);
793 end_page_writeback(page);
796 EXPORT_SYMBOL_GPL(page_endio);
799 * __lock_page - get a lock on the page, assuming we need to sleep to get it
800 * @page: the page to lock
802 void __lock_page(struct page *page)
804 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
806 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
807 TASK_UNINTERRUPTIBLE);
809 EXPORT_SYMBOL(__lock_page);
811 int __lock_page_killable(struct page *page)
813 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
815 return __wait_on_bit_lock(page_waitqueue(page), &wait,
816 bit_wait_io, TASK_KILLABLE);
818 EXPORT_SYMBOL_GPL(__lock_page_killable);
822 * 1 - page is locked; mmap_sem is still held.
823 * 0 - page is not locked.
824 * mmap_sem has been released (up_read()), unless flags had both
825 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
826 * which case mmap_sem is still held.
828 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
829 * with the page locked and the mmap_sem unperturbed.
831 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
834 if (flags & FAULT_FLAG_ALLOW_RETRY) {
836 * CAUTION! In this case, mmap_sem is not released
837 * even though return 0.
839 if (flags & FAULT_FLAG_RETRY_NOWAIT)
842 up_read(&mm->mmap_sem);
843 if (flags & FAULT_FLAG_KILLABLE)
844 wait_on_page_locked_killable(page);
846 wait_on_page_locked(page);
849 if (flags & FAULT_FLAG_KILLABLE) {
852 ret = __lock_page_killable(page);
854 up_read(&mm->mmap_sem);
864 * page_cache_next_hole - find the next hole (not-present entry)
867 * @max_scan: maximum range to search
869 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
870 * lowest indexed hole.
872 * Returns: the index of the hole if found, otherwise returns an index
873 * outside of the set specified (in which case 'return - index >=
874 * max_scan' will be true). In rare cases of index wrap-around, 0 will
877 * page_cache_next_hole may be called under rcu_read_lock. However,
878 * like radix_tree_gang_lookup, this will not atomically search a
879 * snapshot of the tree at a single point in time. For example, if a
880 * hole is created at index 5, then subsequently a hole is created at
881 * index 10, page_cache_next_hole covering both indexes may return 10
882 * if called under rcu_read_lock.
884 pgoff_t page_cache_next_hole(struct address_space *mapping,
885 pgoff_t index, unsigned long max_scan)
889 for (i = 0; i < max_scan; i++) {
892 page = radix_tree_lookup(&mapping->page_tree, index);
893 if (!page || radix_tree_exceptional_entry(page))
902 EXPORT_SYMBOL(page_cache_next_hole);
905 * page_cache_prev_hole - find the prev hole (not-present entry)
908 * @max_scan: maximum range to search
910 * Search backwards in the range [max(index-max_scan+1, 0), index] for
913 * Returns: the index of the hole if found, otherwise returns an index
914 * outside of the set specified (in which case 'index - return >=
915 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
918 * page_cache_prev_hole may be called under rcu_read_lock. However,
919 * like radix_tree_gang_lookup, this will not atomically search a
920 * snapshot of the tree at a single point in time. For example, if a
921 * hole is created at index 10, then subsequently a hole is created at
922 * index 5, page_cache_prev_hole covering both indexes may return 5 if
923 * called under rcu_read_lock.
925 pgoff_t page_cache_prev_hole(struct address_space *mapping,
926 pgoff_t index, unsigned long max_scan)
930 for (i = 0; i < max_scan; i++) {
933 page = radix_tree_lookup(&mapping->page_tree, index);
934 if (!page || radix_tree_exceptional_entry(page))
937 if (index == ULONG_MAX)
943 EXPORT_SYMBOL(page_cache_prev_hole);
946 * find_get_entry - find and get a page cache entry
947 * @mapping: the address_space to search
948 * @offset: the page cache index
950 * Looks up the page cache slot at @mapping & @offset. If there is a
951 * page cache page, it is returned with an increased refcount.
953 * If the slot holds a shadow entry of a previously evicted page, or a
954 * swap entry from shmem/tmpfs, it is returned.
956 * Otherwise, %NULL is returned.
958 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
966 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
968 page = radix_tree_deref_slot(pagep);
971 if (radix_tree_exception(page)) {
972 if (radix_tree_deref_retry(page))
975 * A shadow entry of a recently evicted page,
976 * or a swap entry from shmem/tmpfs. Return
977 * it without attempting to raise page count.
981 if (!page_cache_get_speculative(page))
985 * Has the page moved?
986 * This is part of the lockless pagecache protocol. See
987 * include/linux/pagemap.h for details.
989 if (unlikely(page != *pagep)) {
990 page_cache_release(page);
999 EXPORT_SYMBOL(find_get_entry);
1002 * find_lock_entry - locate, pin and lock a page cache entry
1003 * @mapping: the address_space to search
1004 * @offset: the page cache index
1006 * Looks up the page cache slot at @mapping & @offset. If there is a
1007 * page cache page, it is returned locked and with an increased
1010 * If the slot holds a shadow entry of a previously evicted page, or a
1011 * swap entry from shmem/tmpfs, it is returned.
1013 * Otherwise, %NULL is returned.
1015 * find_lock_entry() may sleep.
1017 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1022 page = find_get_entry(mapping, offset);
1023 if (page && !radix_tree_exception(page)) {
1025 /* Has the page been truncated? */
1026 if (unlikely(page->mapping != mapping)) {
1028 page_cache_release(page);
1031 VM_BUG_ON_PAGE(page->index != offset, page);
1035 EXPORT_SYMBOL(find_lock_entry);
1038 * pagecache_get_page - find and get a page reference
1039 * @mapping: the address_space to search
1040 * @offset: the page index
1041 * @fgp_flags: PCG flags
1042 * @cache_gfp_mask: gfp mask to use for the page cache data page allocation
1043 * @radix_gfp_mask: gfp mask to use for radix tree node allocation
1045 * Looks up the page cache slot at @mapping & @offset.
1047 * PCG flags modify how the page is returned.
1049 * FGP_ACCESSED: the page will be marked accessed
1050 * FGP_LOCK: Page is return locked
1051 * FGP_CREAT: If page is not present then a new page is allocated using
1052 * @cache_gfp_mask and added to the page cache and the VM's LRU
1053 * list. If radix tree nodes are allocated during page cache
1054 * insertion then @radix_gfp_mask is used. The page is returned
1055 * locked and with an increased refcount. Otherwise, %NULL is
1058 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1059 * if the GFP flags specified for FGP_CREAT are atomic.
1061 * If there is a page cache page, it is returned with an increased refcount.
1063 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1064 int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask)
1069 page = find_get_entry(mapping, offset);
1070 if (radix_tree_exceptional_entry(page))
1075 if (fgp_flags & FGP_LOCK) {
1076 if (fgp_flags & FGP_NOWAIT) {
1077 if (!trylock_page(page)) {
1078 page_cache_release(page);
1085 /* Has the page been truncated? */
1086 if (unlikely(page->mapping != mapping)) {
1088 page_cache_release(page);
1091 VM_BUG_ON_PAGE(page->index != offset, page);
1094 if (page && (fgp_flags & FGP_ACCESSED))
1095 mark_page_accessed(page);
1098 if (!page && (fgp_flags & FGP_CREAT)) {
1100 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1101 cache_gfp_mask |= __GFP_WRITE;
1102 if (fgp_flags & FGP_NOFS) {
1103 cache_gfp_mask &= ~__GFP_FS;
1104 radix_gfp_mask &= ~__GFP_FS;
1107 page = __page_cache_alloc(cache_gfp_mask);
1111 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1112 fgp_flags |= FGP_LOCK;
1114 /* Init accessed so avoid atomic mark_page_accessed later */
1115 if (fgp_flags & FGP_ACCESSED)
1116 __SetPageReferenced(page);
1118 err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask);
1119 if (unlikely(err)) {
1120 page_cache_release(page);
1129 EXPORT_SYMBOL(pagecache_get_page);
1132 * find_get_entries - gang pagecache lookup
1133 * @mapping: The address_space to search
1134 * @start: The starting page cache index
1135 * @nr_entries: The maximum number of entries
1136 * @entries: Where the resulting entries are placed
1137 * @indices: The cache indices corresponding to the entries in @entries
1139 * find_get_entries() will search for and return a group of up to
1140 * @nr_entries entries in the mapping. The entries are placed at
1141 * @entries. find_get_entries() takes a reference against any actual
1144 * The search returns a group of mapping-contiguous page cache entries
1145 * with ascending indexes. There may be holes in the indices due to
1146 * not-present pages.
1148 * Any shadow entries of evicted pages, or swap entries from
1149 * shmem/tmpfs, are included in the returned array.
1151 * find_get_entries() returns the number of pages and shadow entries
1154 unsigned find_get_entries(struct address_space *mapping,
1155 pgoff_t start, unsigned int nr_entries,
1156 struct page **entries, pgoff_t *indices)
1159 unsigned int ret = 0;
1160 struct radix_tree_iter iter;
1167 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1170 page = radix_tree_deref_slot(slot);
1171 if (unlikely(!page))
1173 if (radix_tree_exception(page)) {
1174 if (radix_tree_deref_retry(page))
1177 * A shadow entry of a recently evicted page,
1178 * or a swap entry from shmem/tmpfs. Return
1179 * it without attempting to raise page count.
1183 if (!page_cache_get_speculative(page))
1186 /* Has the page moved? */
1187 if (unlikely(page != *slot)) {
1188 page_cache_release(page);
1192 indices[ret] = iter.index;
1193 entries[ret] = page;
1194 if (++ret == nr_entries)
1202 * find_get_pages - gang pagecache lookup
1203 * @mapping: The address_space to search
1204 * @start: The starting page index
1205 * @nr_pages: The maximum number of pages
1206 * @pages: Where the resulting pages are placed
1208 * find_get_pages() will search for and return a group of up to
1209 * @nr_pages pages in the mapping. The pages are placed at @pages.
1210 * find_get_pages() takes a reference against the returned pages.
1212 * The search returns a group of mapping-contiguous pages with ascending
1213 * indexes. There may be holes in the indices due to not-present pages.
1215 * find_get_pages() returns the number of pages which were found.
1217 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1218 unsigned int nr_pages, struct page **pages)
1220 struct radix_tree_iter iter;
1224 if (unlikely(!nr_pages))
1229 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1232 page = radix_tree_deref_slot(slot);
1233 if (unlikely(!page))
1236 if (radix_tree_exception(page)) {
1237 if (radix_tree_deref_retry(page)) {
1239 * Transient condition which can only trigger
1240 * when entry at index 0 moves out of or back
1241 * to root: none yet gotten, safe to restart.
1243 WARN_ON(iter.index);
1247 * A shadow entry of a recently evicted page,
1248 * or a swap entry from shmem/tmpfs. Skip
1254 if (!page_cache_get_speculative(page))
1257 /* Has the page moved? */
1258 if (unlikely(page != *slot)) {
1259 page_cache_release(page);
1264 if (++ret == nr_pages)
1273 * find_get_pages_contig - gang contiguous pagecache lookup
1274 * @mapping: The address_space to search
1275 * @index: The starting page index
1276 * @nr_pages: The maximum number of pages
1277 * @pages: Where the resulting pages are placed
1279 * find_get_pages_contig() works exactly like find_get_pages(), except
1280 * that the returned number of pages are guaranteed to be contiguous.
1282 * find_get_pages_contig() returns the number of pages which were found.
1284 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1285 unsigned int nr_pages, struct page **pages)
1287 struct radix_tree_iter iter;
1289 unsigned int ret = 0;
1291 if (unlikely(!nr_pages))
1296 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1299 page = radix_tree_deref_slot(slot);
1300 /* The hole, there no reason to continue */
1301 if (unlikely(!page))
1304 if (radix_tree_exception(page)) {
1305 if (radix_tree_deref_retry(page)) {
1307 * Transient condition which can only trigger
1308 * when entry at index 0 moves out of or back
1309 * to root: none yet gotten, safe to restart.
1314 * A shadow entry of a recently evicted page,
1315 * or a swap entry from shmem/tmpfs. Stop
1316 * looking for contiguous pages.
1321 if (!page_cache_get_speculative(page))
1324 /* Has the page moved? */
1325 if (unlikely(page != *slot)) {
1326 page_cache_release(page);
1331 * must check mapping and index after taking the ref.
1332 * otherwise we can get both false positives and false
1333 * negatives, which is just confusing to the caller.
1335 if (page->mapping == NULL || page->index != iter.index) {
1336 page_cache_release(page);
1341 if (++ret == nr_pages)
1347 EXPORT_SYMBOL(find_get_pages_contig);
1350 * find_get_pages_tag - find and return pages that match @tag
1351 * @mapping: the address_space to search
1352 * @index: the starting page index
1353 * @tag: the tag index
1354 * @nr_pages: the maximum number of pages
1355 * @pages: where the resulting pages are placed
1357 * Like find_get_pages, except we only return pages which are tagged with
1358 * @tag. We update @index to index the next page for the traversal.
1360 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1361 int tag, unsigned int nr_pages, struct page **pages)
1363 struct radix_tree_iter iter;
1367 if (unlikely(!nr_pages))
1372 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1373 &iter, *index, tag) {
1376 page = radix_tree_deref_slot(slot);
1377 if (unlikely(!page))
1380 if (radix_tree_exception(page)) {
1381 if (radix_tree_deref_retry(page)) {
1383 * Transient condition which can only trigger
1384 * when entry at index 0 moves out of or back
1385 * to root: none yet gotten, safe to restart.
1390 * A shadow entry of a recently evicted page.
1392 * Those entries should never be tagged, but
1393 * this tree walk is lockless and the tags are
1394 * looked up in bulk, one radix tree node at a
1395 * time, so there is a sizable window for page
1396 * reclaim to evict a page we saw tagged.
1403 if (!page_cache_get_speculative(page))
1406 /* Has the page moved? */
1407 if (unlikely(page != *slot)) {
1408 page_cache_release(page);
1413 if (++ret == nr_pages)
1420 *index = pages[ret - 1]->index + 1;
1424 EXPORT_SYMBOL(find_get_pages_tag);
1427 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1428 * a _large_ part of the i/o request. Imagine the worst scenario:
1430 * ---R__________________________________________B__________
1431 * ^ reading here ^ bad block(assume 4k)
1433 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1434 * => failing the whole request => read(R) => read(R+1) =>
1435 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1436 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1437 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1439 * It is going insane. Fix it by quickly scaling down the readahead size.
1441 static void shrink_readahead_size_eio(struct file *filp,
1442 struct file_ra_state *ra)
1448 * do_generic_file_read - generic file read routine
1449 * @filp: the file to read
1450 * @ppos: current file position
1451 * @iter: data destination
1452 * @written: already copied
1454 * This is a generic file read routine, and uses the
1455 * mapping->a_ops->readpage() function for the actual low-level stuff.
1457 * This is really ugly. But the goto's actually try to clarify some
1458 * of the logic when it comes to error handling etc.
1460 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1461 struct iov_iter *iter, ssize_t written)
1463 struct address_space *mapping = filp->f_mapping;
1464 struct inode *inode = mapping->host;
1465 struct file_ra_state *ra = &filp->f_ra;
1469 unsigned long offset; /* offset into pagecache page */
1470 unsigned int prev_offset;
1473 index = *ppos >> PAGE_CACHE_SHIFT;
1474 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1475 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1476 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1477 offset = *ppos & ~PAGE_CACHE_MASK;
1483 unsigned long nr, ret;
1487 page = find_get_page(mapping, index);
1489 page_cache_sync_readahead(mapping,
1491 index, last_index - index);
1492 page = find_get_page(mapping, index);
1493 if (unlikely(page == NULL))
1494 goto no_cached_page;
1496 if (PageReadahead(page)) {
1497 page_cache_async_readahead(mapping,
1499 index, last_index - index);
1501 if (!PageUptodate(page)) {
1502 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1503 !mapping->a_ops->is_partially_uptodate)
1504 goto page_not_up_to_date;
1505 if (!trylock_page(page))
1506 goto page_not_up_to_date;
1507 /* Did it get truncated before we got the lock? */
1509 goto page_not_up_to_date_locked;
1510 if (!mapping->a_ops->is_partially_uptodate(page,
1511 offset, iter->count))
1512 goto page_not_up_to_date_locked;
1517 * i_size must be checked after we know the page is Uptodate.
1519 * Checking i_size after the check allows us to calculate
1520 * the correct value for "nr", which means the zero-filled
1521 * part of the page is not copied back to userspace (unless
1522 * another truncate extends the file - this is desired though).
1525 isize = i_size_read(inode);
1526 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1527 if (unlikely(!isize || index > end_index)) {
1528 page_cache_release(page);
1532 /* nr is the maximum number of bytes to copy from this page */
1533 nr = PAGE_CACHE_SIZE;
1534 if (index == end_index) {
1535 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1537 page_cache_release(page);
1543 /* If users can be writing to this page using arbitrary
1544 * virtual addresses, take care about potential aliasing
1545 * before reading the page on the kernel side.
1547 if (mapping_writably_mapped(mapping))
1548 flush_dcache_page(page);
1551 * When a sequential read accesses a page several times,
1552 * only mark it as accessed the first time.
1554 if (prev_index != index || offset != prev_offset)
1555 mark_page_accessed(page);
1559 * Ok, we have the page, and it's up-to-date, so
1560 * now we can copy it to user space...
1563 ret = copy_page_to_iter(page, offset, nr, iter);
1565 index += offset >> PAGE_CACHE_SHIFT;
1566 offset &= ~PAGE_CACHE_MASK;
1567 prev_offset = offset;
1569 page_cache_release(page);
1571 if (!iov_iter_count(iter))
1579 page_not_up_to_date:
1580 /* Get exclusive access to the page ... */
1581 error = lock_page_killable(page);
1582 if (unlikely(error))
1583 goto readpage_error;
1585 page_not_up_to_date_locked:
1586 /* Did it get truncated before we got the lock? */
1587 if (!page->mapping) {
1589 page_cache_release(page);
1593 /* Did somebody else fill it already? */
1594 if (PageUptodate(page)) {
1601 * A previous I/O error may have been due to temporary
1602 * failures, eg. multipath errors.
1603 * PG_error will be set again if readpage fails.
1605 ClearPageError(page);
1606 /* Start the actual read. The read will unlock the page. */
1607 error = mapping->a_ops->readpage(filp, page);
1609 if (unlikely(error)) {
1610 if (error == AOP_TRUNCATED_PAGE) {
1611 page_cache_release(page);
1615 goto readpage_error;
1618 if (!PageUptodate(page)) {
1619 error = lock_page_killable(page);
1620 if (unlikely(error))
1621 goto readpage_error;
1622 if (!PageUptodate(page)) {
1623 if (page->mapping == NULL) {
1625 * invalidate_mapping_pages got it
1628 page_cache_release(page);
1632 shrink_readahead_size_eio(filp, ra);
1634 goto readpage_error;
1642 /* UHHUH! A synchronous read error occurred. Report it */
1643 page_cache_release(page);
1648 * Ok, it wasn't cached, so we need to create a new
1651 page = page_cache_alloc_cold(mapping);
1656 error = add_to_page_cache_lru(page, mapping,
1659 page_cache_release(page);
1660 if (error == -EEXIST) {
1670 ra->prev_pos = prev_index;
1671 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1672 ra->prev_pos |= prev_offset;
1674 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1675 file_accessed(filp);
1676 return written ? written : error;
1680 * generic_file_read_iter - generic filesystem read routine
1681 * @iocb: kernel I/O control block
1682 * @iter: destination for the data read
1684 * This is the "read_iter()" routine for all filesystems
1685 * that can use the page cache directly.
1688 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1690 struct file *file = iocb->ki_filp;
1692 loff_t *ppos = &iocb->ki_pos;
1695 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1696 if (file->f_flags & O_DIRECT) {
1697 struct address_space *mapping = file->f_mapping;
1698 struct inode *inode = mapping->host;
1699 size_t count = iov_iter_count(iter);
1703 goto out; /* skip atime */
1704 size = i_size_read(inode);
1705 retval = filemap_write_and_wait_range(mapping, pos,
1708 struct iov_iter data = *iter;
1709 retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos);
1713 *ppos = pos + retval;
1714 iov_iter_advance(iter, retval);
1718 * Btrfs can have a short DIO read if we encounter
1719 * compressed extents, so if there was an error, or if
1720 * we've already read everything we wanted to, or if
1721 * there was a short read because we hit EOF, go ahead
1722 * and return. Otherwise fallthrough to buffered io for
1723 * the rest of the read.
1725 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) {
1726 file_accessed(file);
1731 retval = do_generic_file_read(file, ppos, iter, retval);
1735 EXPORT_SYMBOL(generic_file_read_iter);
1739 * page_cache_read - adds requested page to the page cache if not already there
1740 * @file: file to read
1741 * @offset: page index
1743 * This adds the requested page to the page cache if it isn't already there,
1744 * and schedules an I/O to read in its contents from disk.
1746 static int page_cache_read(struct file *file, pgoff_t offset)
1748 struct address_space *mapping = file->f_mapping;
1753 page = page_cache_alloc_cold(mapping);
1757 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1759 ret = mapping->a_ops->readpage(file, page);
1760 else if (ret == -EEXIST)
1761 ret = 0; /* losing race to add is OK */
1763 page_cache_release(page);
1765 } while (ret == AOP_TRUNCATED_PAGE);
1770 #define MMAP_LOTSAMISS (100)
1773 * Synchronous readahead happens when we don't even find
1774 * a page in the page cache at all.
1776 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1777 struct file_ra_state *ra,
1781 unsigned long ra_pages;
1782 struct address_space *mapping = file->f_mapping;
1784 /* If we don't want any read-ahead, don't bother */
1785 if (vma->vm_flags & VM_RAND_READ)
1790 if (vma->vm_flags & VM_SEQ_READ) {
1791 page_cache_sync_readahead(mapping, ra, file, offset,
1796 /* Avoid banging the cache line if not needed */
1797 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1801 * Do we miss much more than hit in this file? If so,
1802 * stop bothering with read-ahead. It will only hurt.
1804 if (ra->mmap_miss > MMAP_LOTSAMISS)
1810 ra_pages = max_sane_readahead(ra->ra_pages);
1811 ra->start = max_t(long, 0, offset - ra_pages / 2);
1812 ra->size = ra_pages;
1813 ra->async_size = ra_pages / 4;
1814 ra_submit(ra, mapping, file);
1818 * Asynchronous readahead happens when we find the page and PG_readahead,
1819 * so we want to possibly extend the readahead further..
1821 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1822 struct file_ra_state *ra,
1827 struct address_space *mapping = file->f_mapping;
1829 /* If we don't want any read-ahead, don't bother */
1830 if (vma->vm_flags & VM_RAND_READ)
1832 if (ra->mmap_miss > 0)
1834 if (PageReadahead(page))
1835 page_cache_async_readahead(mapping, ra, file,
1836 page, offset, ra->ra_pages);
1840 * filemap_fault - read in file data for page fault handling
1841 * @vma: vma in which the fault was taken
1842 * @vmf: struct vm_fault containing details of the fault
1844 * filemap_fault() is invoked via the vma operations vector for a
1845 * mapped memory region to read in file data during a page fault.
1847 * The goto's are kind of ugly, but this streamlines the normal case of having
1848 * it in the page cache, and handles the special cases reasonably without
1849 * having a lot of duplicated code.
1851 * vma->vm_mm->mmap_sem must be held on entry.
1853 * If our return value has VM_FAULT_RETRY set, it's because
1854 * lock_page_or_retry() returned 0.
1855 * The mmap_sem has usually been released in this case.
1856 * See __lock_page_or_retry() for the exception.
1858 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1859 * has not been released.
1861 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1863 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1866 struct file *file = vma->vm_file;
1867 struct address_space *mapping = file->f_mapping;
1868 struct file_ra_state *ra = &file->f_ra;
1869 struct inode *inode = mapping->host;
1870 pgoff_t offset = vmf->pgoff;
1875 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1876 if (offset >= size >> PAGE_CACHE_SHIFT)
1877 return VM_FAULT_SIGBUS;
1880 * Do we have something in the page cache already?
1882 page = find_get_page(mapping, offset);
1883 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1885 * We found the page, so try async readahead before
1886 * waiting for the lock.
1888 do_async_mmap_readahead(vma, ra, file, page, offset);
1890 /* No page in the page cache at all */
1891 do_sync_mmap_readahead(vma, ra, file, offset);
1892 count_vm_event(PGMAJFAULT);
1893 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1894 ret = VM_FAULT_MAJOR;
1896 page = find_get_page(mapping, offset);
1898 goto no_cached_page;
1901 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1902 page_cache_release(page);
1903 return ret | VM_FAULT_RETRY;
1906 /* Did it get truncated? */
1907 if (unlikely(page->mapping != mapping)) {
1912 VM_BUG_ON_PAGE(page->index != offset, page);
1915 * We have a locked page in the page cache, now we need to check
1916 * that it's up-to-date. If not, it is going to be due to an error.
1918 if (unlikely(!PageUptodate(page)))
1919 goto page_not_uptodate;
1922 * Found the page and have a reference on it.
1923 * We must recheck i_size under page lock.
1925 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1926 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1928 page_cache_release(page);
1929 return VM_FAULT_SIGBUS;
1933 return ret | VM_FAULT_LOCKED;
1937 * We're only likely to ever get here if MADV_RANDOM is in
1940 error = page_cache_read(file, offset);
1943 * The page we want has now been added to the page cache.
1944 * In the unlikely event that someone removed it in the
1945 * meantime, we'll just come back here and read it again.
1951 * An error return from page_cache_read can result if the
1952 * system is low on memory, or a problem occurs while trying
1955 if (error == -ENOMEM)
1956 return VM_FAULT_OOM;
1957 return VM_FAULT_SIGBUS;
1961 * Umm, take care of errors if the page isn't up-to-date.
1962 * Try to re-read it _once_. We do this synchronously,
1963 * because there really aren't any performance issues here
1964 * and we need to check for errors.
1966 ClearPageError(page);
1967 error = mapping->a_ops->readpage(file, page);
1969 wait_on_page_locked(page);
1970 if (!PageUptodate(page))
1973 page_cache_release(page);
1975 if (!error || error == AOP_TRUNCATED_PAGE)
1978 /* Things didn't work out. Return zero to tell the mm layer so. */
1979 shrink_readahead_size_eio(file, ra);
1980 return VM_FAULT_SIGBUS;
1982 EXPORT_SYMBOL(filemap_fault);
1984 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1986 struct radix_tree_iter iter;
1988 struct file *file = vma->vm_file;
1989 struct address_space *mapping = file->f_mapping;
1992 unsigned long address = (unsigned long) vmf->virtual_address;
1997 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
1998 if (iter.index > vmf->max_pgoff)
2001 page = radix_tree_deref_slot(slot);
2002 if (unlikely(!page))
2004 if (radix_tree_exception(page)) {
2005 if (radix_tree_deref_retry(page))
2011 if (!page_cache_get_speculative(page))
2014 /* Has the page moved? */
2015 if (unlikely(page != *slot)) {
2016 page_cache_release(page);
2020 if (!PageUptodate(page) ||
2021 PageReadahead(page) ||
2024 if (!trylock_page(page))
2027 if (page->mapping != mapping || !PageUptodate(page))
2030 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2031 if (page->index >= size >> PAGE_CACHE_SHIFT)
2034 pte = vmf->pte + page->index - vmf->pgoff;
2035 if (!pte_none(*pte))
2038 if (file->f_ra.mmap_miss > 0)
2039 file->f_ra.mmap_miss--;
2040 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2041 do_set_pte(vma, addr, page, pte, false, false);
2047 page_cache_release(page);
2049 if (iter.index == vmf->max_pgoff)
2054 EXPORT_SYMBOL(filemap_map_pages);
2056 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2058 struct page *page = vmf->page;
2059 struct inode *inode = file_inode(vma->vm_file);
2060 int ret = VM_FAULT_LOCKED;
2062 sb_start_pagefault(inode->i_sb);
2063 file_update_time(vma->vm_file);
2065 if (page->mapping != inode->i_mapping) {
2067 ret = VM_FAULT_NOPAGE;
2071 * We mark the page dirty already here so that when freeze is in
2072 * progress, we are guaranteed that writeback during freezing will
2073 * see the dirty page and writeprotect it again.
2075 set_page_dirty(page);
2076 wait_for_stable_page(page);
2078 sb_end_pagefault(inode->i_sb);
2081 EXPORT_SYMBOL(filemap_page_mkwrite);
2083 const struct vm_operations_struct generic_file_vm_ops = {
2084 .fault = filemap_fault,
2085 .map_pages = filemap_map_pages,
2086 .page_mkwrite = filemap_page_mkwrite,
2087 .remap_pages = generic_file_remap_pages,
2090 /* This is used for a general mmap of a disk file */
2092 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2094 struct address_space *mapping = file->f_mapping;
2096 if (!mapping->a_ops->readpage)
2098 file_accessed(file);
2099 vma->vm_ops = &generic_file_vm_ops;
2104 * This is for filesystems which do not implement ->writepage.
2106 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2108 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2110 return generic_file_mmap(file, vma);
2113 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2117 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2121 #endif /* CONFIG_MMU */
2123 EXPORT_SYMBOL(generic_file_mmap);
2124 EXPORT_SYMBOL(generic_file_readonly_mmap);
2126 static struct page *wait_on_page_read(struct page *page)
2128 if (!IS_ERR(page)) {
2129 wait_on_page_locked(page);
2130 if (!PageUptodate(page)) {
2131 page_cache_release(page);
2132 page = ERR_PTR(-EIO);
2138 static struct page *__read_cache_page(struct address_space *mapping,
2140 int (*filler)(void *, struct page *),
2147 page = find_get_page(mapping, index);
2149 page = __page_cache_alloc(gfp | __GFP_COLD);
2151 return ERR_PTR(-ENOMEM);
2152 err = add_to_page_cache_lru(page, mapping, index, gfp);
2153 if (unlikely(err)) {
2154 page_cache_release(page);
2157 /* Presumably ENOMEM for radix tree node */
2158 return ERR_PTR(err);
2160 err = filler(data, page);
2162 page_cache_release(page);
2163 page = ERR_PTR(err);
2165 page = wait_on_page_read(page);
2171 static struct page *do_read_cache_page(struct address_space *mapping,
2173 int (*filler)(void *, struct page *),
2182 page = __read_cache_page(mapping, index, filler, data, gfp);
2185 if (PageUptodate(page))
2189 if (!page->mapping) {
2191 page_cache_release(page);
2194 if (PageUptodate(page)) {
2198 err = filler(data, page);
2200 page_cache_release(page);
2201 return ERR_PTR(err);
2203 page = wait_on_page_read(page);
2208 mark_page_accessed(page);
2213 * read_cache_page - read into page cache, fill it if needed
2214 * @mapping: the page's address_space
2215 * @index: the page index
2216 * @filler: function to perform the read
2217 * @data: first arg to filler(data, page) function, often left as NULL
2219 * Read into the page cache. If a page already exists, and PageUptodate() is
2220 * not set, try to fill the page and wait for it to become unlocked.
2222 * If the page does not get brought uptodate, return -EIO.
2224 struct page *read_cache_page(struct address_space *mapping,
2226 int (*filler)(void *, struct page *),
2229 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2231 EXPORT_SYMBOL(read_cache_page);
2234 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2235 * @mapping: the page's address_space
2236 * @index: the page index
2237 * @gfp: the page allocator flags to use if allocating
2239 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2240 * any new page allocations done using the specified allocation flags.
2242 * If the page does not get brought uptodate, return -EIO.
2244 struct page *read_cache_page_gfp(struct address_space *mapping,
2248 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2250 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2252 EXPORT_SYMBOL(read_cache_page_gfp);
2255 * Performs necessary checks before doing a write
2257 * Can adjust writing position or amount of bytes to write.
2258 * Returns appropriate error code that caller should return or
2259 * zero in case that write should be allowed.
2261 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2263 struct inode *inode = file->f_mapping->host;
2264 unsigned long limit = rlimit(RLIMIT_FSIZE);
2266 if (unlikely(*pos < 0))
2270 /* FIXME: this is for backwards compatibility with 2.4 */
2271 if (file->f_flags & O_APPEND)
2272 *pos = i_size_read(inode);
2274 if (limit != RLIM_INFINITY) {
2275 if (*pos >= limit) {
2276 send_sig(SIGXFSZ, current, 0);
2279 if (*count > limit - (typeof(limit))*pos) {
2280 *count = limit - (typeof(limit))*pos;
2288 if (unlikely(*pos + *count > MAX_NON_LFS &&
2289 !(file->f_flags & O_LARGEFILE))) {
2290 if (*pos >= MAX_NON_LFS) {
2293 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2294 *count = MAX_NON_LFS - (unsigned long)*pos;
2299 * Are we about to exceed the fs block limit ?
2301 * If we have written data it becomes a short write. If we have
2302 * exceeded without writing data we send a signal and return EFBIG.
2303 * Linus frestrict idea will clean these up nicely..
2305 if (likely(!isblk)) {
2306 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2307 if (*count || *pos > inode->i_sb->s_maxbytes) {
2310 /* zero-length writes at ->s_maxbytes are OK */
2313 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2314 *count = inode->i_sb->s_maxbytes - *pos;
2318 if (bdev_read_only(I_BDEV(inode)))
2320 isize = i_size_read(inode);
2321 if (*pos >= isize) {
2322 if (*count || *pos > isize)
2326 if (*pos + *count > isize)
2327 *count = isize - *pos;
2334 EXPORT_SYMBOL(generic_write_checks);
2336 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2337 loff_t pos, unsigned len, unsigned flags,
2338 struct page **pagep, void **fsdata)
2340 const struct address_space_operations *aops = mapping->a_ops;
2342 return aops->write_begin(file, mapping, pos, len, flags,
2345 EXPORT_SYMBOL(pagecache_write_begin);
2347 int pagecache_write_end(struct file *file, struct address_space *mapping,
2348 loff_t pos, unsigned len, unsigned copied,
2349 struct page *page, void *fsdata)
2351 const struct address_space_operations *aops = mapping->a_ops;
2353 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2355 EXPORT_SYMBOL(pagecache_write_end);
2358 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2360 struct file *file = iocb->ki_filp;
2361 struct address_space *mapping = file->f_mapping;
2362 struct inode *inode = mapping->host;
2366 struct iov_iter data;
2368 write_len = iov_iter_count(from);
2369 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2371 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2376 * After a write we want buffered reads to be sure to go to disk to get
2377 * the new data. We invalidate clean cached page from the region we're
2378 * about to write. We do this *before* the write so that we can return
2379 * without clobbering -EIOCBQUEUED from ->direct_IO().
2381 if (mapping->nrpages) {
2382 written = invalidate_inode_pages2_range(mapping,
2383 pos >> PAGE_CACHE_SHIFT, end);
2385 * If a page can not be invalidated, return 0 to fall back
2386 * to buffered write.
2389 if (written == -EBUSY)
2396 written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos);
2399 * Finally, try again to invalidate clean pages which might have been
2400 * cached by non-direct readahead, or faulted in by get_user_pages()
2401 * if the source of the write was an mmap'ed region of the file
2402 * we're writing. Either one is a pretty crazy thing to do,
2403 * so we don't support it 100%. If this invalidation
2404 * fails, tough, the write still worked...
2406 if (mapping->nrpages) {
2407 invalidate_inode_pages2_range(mapping,
2408 pos >> PAGE_CACHE_SHIFT, end);
2413 iov_iter_advance(from, written);
2414 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2415 i_size_write(inode, pos);
2416 mark_inode_dirty(inode);
2423 EXPORT_SYMBOL(generic_file_direct_write);
2426 * Find or create a page at the given pagecache position. Return the locked
2427 * page. This function is specifically for buffered writes.
2429 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2430 pgoff_t index, unsigned flags)
2433 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2435 if (flags & AOP_FLAG_NOFS)
2436 fgp_flags |= FGP_NOFS;
2438 page = pagecache_get_page(mapping, index, fgp_flags,
2439 mapping_gfp_mask(mapping),
2442 wait_for_stable_page(page);
2446 EXPORT_SYMBOL(grab_cache_page_write_begin);
2448 ssize_t generic_perform_write(struct file *file,
2449 struct iov_iter *i, loff_t pos)
2451 struct address_space *mapping = file->f_mapping;
2452 const struct address_space_operations *a_ops = mapping->a_ops;
2454 ssize_t written = 0;
2455 unsigned int flags = 0;
2458 * Copies from kernel address space cannot fail (NFSD is a big user).
2460 if (segment_eq(get_fs(), KERNEL_DS))
2461 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2465 unsigned long offset; /* Offset into pagecache page */
2466 unsigned long bytes; /* Bytes to write to page */
2467 size_t copied; /* Bytes copied from user */
2470 offset = (pos & (PAGE_CACHE_SIZE - 1));
2471 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2476 * Bring in the user page that we will copy from _first_.
2477 * Otherwise there's a nasty deadlock on copying from the
2478 * same page as we're writing to, without it being marked
2481 * Not only is this an optimisation, but it is also required
2482 * to check that the address is actually valid, when atomic
2483 * usercopies are used, below.
2485 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2490 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2492 if (unlikely(status < 0))
2495 if (mapping_writably_mapped(mapping))
2496 flush_dcache_page(page);
2498 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2499 flush_dcache_page(page);
2501 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2503 if (unlikely(status < 0))
2509 iov_iter_advance(i, copied);
2510 if (unlikely(copied == 0)) {
2512 * If we were unable to copy any data at all, we must
2513 * fall back to a single segment length write.
2515 * If we didn't fallback here, we could livelock
2516 * because not all segments in the iov can be copied at
2517 * once without a pagefault.
2519 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2520 iov_iter_single_seg_count(i));
2526 balance_dirty_pages_ratelimited(mapping);
2527 if (fatal_signal_pending(current)) {
2531 } while (iov_iter_count(i));
2533 return written ? written : status;
2535 EXPORT_SYMBOL(generic_perform_write);
2538 * __generic_file_write_iter - write data to a file
2539 * @iocb: IO state structure (file, offset, etc.)
2540 * @from: iov_iter with data to write
2542 * This function does all the work needed for actually writing data to a
2543 * file. It does all basic checks, removes SUID from the file, updates
2544 * modification times and calls proper subroutines depending on whether we
2545 * do direct IO or a standard buffered write.
2547 * It expects i_mutex to be grabbed unless we work on a block device or similar
2548 * object which does not need locking at all.
2550 * This function does *not* take care of syncing data in case of O_SYNC write.
2551 * A caller has to handle it. This is mainly due to the fact that we want to
2552 * avoid syncing under i_mutex.
2554 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2556 struct file *file = iocb->ki_filp;
2557 struct address_space * mapping = file->f_mapping;
2558 struct inode *inode = mapping->host;
2559 loff_t pos = iocb->ki_pos;
2560 ssize_t written = 0;
2563 size_t count = iov_iter_count(from);
2565 /* We can write back this queue in page reclaim */
2566 current->backing_dev_info = mapping->backing_dev_info;
2567 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2574 iov_iter_truncate(from, count);
2576 err = file_remove_suid(file);
2580 err = file_update_time(file);
2584 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2585 if (unlikely(file->f_flags & O_DIRECT)) {
2588 written = generic_file_direct_write(iocb, from, pos);
2589 if (written < 0 || written == count)
2593 * direct-io write to a hole: fall through to buffered I/O
2594 * for completing the rest of the request.
2599 status = generic_perform_write(file, from, pos);
2601 * If generic_perform_write() returned a synchronous error
2602 * then we want to return the number of bytes which were
2603 * direct-written, or the error code if that was zero. Note
2604 * that this differs from normal direct-io semantics, which
2605 * will return -EFOO even if some bytes were written.
2607 if (unlikely(status < 0) && !written) {
2611 iocb->ki_pos = pos + status;
2613 * We need to ensure that the page cache pages are written to
2614 * disk and invalidated to preserve the expected O_DIRECT
2617 endbyte = pos + status - 1;
2618 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2621 invalidate_mapping_pages(mapping,
2622 pos >> PAGE_CACHE_SHIFT,
2623 endbyte >> PAGE_CACHE_SHIFT);
2626 * We don't know how much we wrote, so just return
2627 * the number of bytes which were direct-written
2631 written = generic_perform_write(file, from, pos);
2632 if (likely(written >= 0))
2633 iocb->ki_pos = pos + written;
2636 current->backing_dev_info = NULL;
2637 return written ? written : err;
2639 EXPORT_SYMBOL(__generic_file_write_iter);
2642 * generic_file_write_iter - write data to a file
2643 * @iocb: IO state structure
2644 * @from: iov_iter with data to write
2646 * This is a wrapper around __generic_file_write_iter() to be used by most
2647 * filesystems. It takes care of syncing the file in case of O_SYNC file
2648 * and acquires i_mutex as needed.
2650 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2652 struct file *file = iocb->ki_filp;
2653 struct inode *inode = file->f_mapping->host;
2656 mutex_lock(&inode->i_mutex);
2657 ret = __generic_file_write_iter(iocb, from);
2658 mutex_unlock(&inode->i_mutex);
2663 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2669 EXPORT_SYMBOL(generic_file_write_iter);
2672 * try_to_release_page() - release old fs-specific metadata on a page
2674 * @page: the page which the kernel is trying to free
2675 * @gfp_mask: memory allocation flags (and I/O mode)
2677 * The address_space is to try to release any data against the page
2678 * (presumably at page->private). If the release was successful, return `1'.
2679 * Otherwise return zero.
2681 * This may also be called if PG_fscache is set on a page, indicating that the
2682 * page is known to the local caching routines.
2684 * The @gfp_mask argument specifies whether I/O may be performed to release
2685 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2688 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2690 struct address_space * const mapping = page->mapping;
2692 BUG_ON(!PageLocked(page));
2693 if (PageWriteback(page))
2696 if (mapping && mapping->a_ops->releasepage)
2697 return mapping->a_ops->releasepage(page, gfp_mask);
2698 return try_to_free_buffers(page);
2701 EXPORT_SYMBOL(try_to_release_page);