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/memcontrol.h>
35 #include <linux/cleancache.h>
36 #include <linux/rmap.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/filemap.h>
43 * FIXME: remove all knowledge of the buffer layer from the core VM
45 #include <linux/buffer_head.h> /* for try_to_free_buffers */
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * Shared mappings now work. 15.8.1995 Bruno.
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
64 * ->i_mmap_mutex (truncate_pagecache)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
70 * ->i_mmap_mutex (truncate->unmap_mapping_range)
74 * ->page_table_lock or pte_lock (various, mainly in memory.c)
75 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
78 * ->lock_page (access_process_vm)
80 * ->i_mutex (generic_perform_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
84 * sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
104 * ->inode->i_lock (zap_pte_range->set_page_dirty)
105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 static void page_cache_tree_delete(struct address_space *mapping,
112 struct page *page, void *shadow)
114 struct radix_tree_node *node;
120 VM_BUG_ON(!PageLocked(page));
122 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
125 mapping->nrshadows++;
127 * Make sure the nrshadows update is committed before
128 * the nrpages update so that final truncate racing
129 * with reclaim does not see both counters 0 at the
130 * same time and miss a shadow entry.
137 /* Clear direct pointer tags in root node */
138 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
139 radix_tree_replace_slot(slot, shadow);
143 /* Clear tree tags for the removed page */
145 offset = index & RADIX_TREE_MAP_MASK;
146 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
147 if (test_bit(offset, node->tags[tag]))
148 radix_tree_tag_clear(&mapping->page_tree, index, tag);
151 /* Delete page, swap shadow entry */
152 radix_tree_replace_slot(slot, shadow);
153 workingset_node_pages_dec(node);
155 workingset_node_shadows_inc(node);
157 if (__radix_tree_delete_node(&mapping->page_tree, node))
161 * Track node that only contains shadow entries.
163 * Avoid acquiring the list_lru lock if already tracked. The
164 * list_empty() test is safe as node->private_list is
165 * protected by mapping->tree_lock.
167 if (!workingset_node_pages(node) &&
168 list_empty(&node->private_list)) {
169 node->private_data = mapping;
170 list_lru_add(&workingset_shadow_nodes, &node->private_list);
175 * Delete a page from the page cache and free it. Caller has to make
176 * sure the page is locked and that nobody else uses it - or that usage
177 * is safe. The caller must hold the mapping's tree_lock.
179 void __delete_from_page_cache(struct page *page, void *shadow)
181 struct address_space *mapping = page->mapping;
183 trace_mm_filemap_delete_from_page_cache(page);
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
189 if (PageUptodate(page) && PageMappedToDisk(page))
190 cleancache_put_page(page);
192 cleancache_invalidate_page(mapping, page);
194 page_cache_tree_delete(mapping, page, shadow);
196 page->mapping = NULL;
197 /* Leave page->index set: truncation lookup relies upon it */
199 __dec_zone_page_state(page, NR_FILE_PAGES);
200 if (PageSwapBacked(page))
201 __dec_zone_page_state(page, NR_SHMEM);
202 BUG_ON(page_mapped(page));
205 * Some filesystems seem to re-dirty the page even after
206 * the VM has canceled the dirty bit (eg ext3 journaling).
208 * Fix it up by doing a final dirty accounting check after
209 * having removed the page entirely.
211 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
212 dec_zone_page_state(page, NR_FILE_DIRTY);
213 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
218 * delete_from_page_cache - delete page from page cache
219 * @page: the page which the kernel is trying to remove from page cache
221 * This must be called only on pages that have been verified to be in the page
222 * cache and locked. It will never put the page into the free list, the caller
223 * has a reference on the page.
225 void delete_from_page_cache(struct page *page)
227 struct address_space *mapping = page->mapping;
228 void (*freepage)(struct page *);
230 BUG_ON(!PageLocked(page));
232 freepage = mapping->a_ops->freepage;
233 spin_lock_irq(&mapping->tree_lock);
234 __delete_from_page_cache(page, NULL);
235 spin_unlock_irq(&mapping->tree_lock);
236 mem_cgroup_uncharge_cache_page(page);
240 page_cache_release(page);
242 EXPORT_SYMBOL(delete_from_page_cache);
244 static int filemap_check_errors(struct address_space *mapping)
247 /* Check for outstanding write errors */
248 if (test_bit(AS_ENOSPC, &mapping->flags) &&
249 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
251 if (test_bit(AS_EIO, &mapping->flags) &&
252 test_and_clear_bit(AS_EIO, &mapping->flags))
258 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
259 * @mapping: address space structure to write
260 * @start: offset in bytes where the range starts
261 * @end: offset in bytes where the range ends (inclusive)
262 * @sync_mode: enable synchronous operation
264 * Start writeback against all of a mapping's dirty pages that lie
265 * within the byte offsets <start, end> inclusive.
267 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
268 * opposed to a regular memory cleansing writeback. The difference between
269 * these two operations is that if a dirty page/buffer is encountered, it must
270 * be waited upon, and not just skipped over.
272 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
273 loff_t end, int sync_mode)
276 struct writeback_control wbc = {
277 .sync_mode = sync_mode,
278 .nr_to_write = LONG_MAX,
279 .range_start = start,
283 if (!mapping_cap_writeback_dirty(mapping))
286 ret = do_writepages(mapping, &wbc);
290 static inline int __filemap_fdatawrite(struct address_space *mapping,
293 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
296 int filemap_fdatawrite(struct address_space *mapping)
298 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
300 EXPORT_SYMBOL(filemap_fdatawrite);
302 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
305 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
307 EXPORT_SYMBOL(filemap_fdatawrite_range);
310 * filemap_flush - mostly a non-blocking flush
311 * @mapping: target address_space
313 * This is a mostly non-blocking flush. Not suitable for data-integrity
314 * purposes - I/O may not be started against all dirty pages.
316 int filemap_flush(struct address_space *mapping)
318 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
320 EXPORT_SYMBOL(filemap_flush);
323 * filemap_fdatawait_range - wait for writeback to complete
324 * @mapping: address space structure to wait for
325 * @start_byte: offset in bytes where the range starts
326 * @end_byte: offset in bytes where the range ends (inclusive)
328 * Walk the list of under-writeback pages of the given address space
329 * in the given range and wait for all of them.
331 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
334 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
335 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
340 if (end_byte < start_byte)
343 pagevec_init(&pvec, 0);
344 while ((index <= end) &&
345 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
346 PAGECACHE_TAG_WRITEBACK,
347 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
350 for (i = 0; i < nr_pages; i++) {
351 struct page *page = pvec.pages[i];
353 /* until radix tree lookup accepts end_index */
354 if (page->index > end)
357 wait_on_page_writeback(page);
358 if (TestClearPageError(page))
361 pagevec_release(&pvec);
365 ret2 = filemap_check_errors(mapping);
371 EXPORT_SYMBOL(filemap_fdatawait_range);
374 * filemap_fdatawait - wait for all under-writeback pages to complete
375 * @mapping: address space structure to wait for
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
380 int filemap_fdatawait(struct address_space *mapping)
382 loff_t i_size = i_size_read(mapping->host);
387 return filemap_fdatawait_range(mapping, 0, i_size - 1);
389 EXPORT_SYMBOL(filemap_fdatawait);
391 int filemap_write_and_wait(struct address_space *mapping)
395 if (mapping->nrpages) {
396 err = filemap_fdatawrite(mapping);
398 * Even if the above returned error, the pages may be
399 * written partially (e.g. -ENOSPC), so we wait for it.
400 * But the -EIO is special case, it may indicate the worst
401 * thing (e.g. bug) happened, so we avoid waiting for it.
404 int err2 = filemap_fdatawait(mapping);
409 err = filemap_check_errors(mapping);
413 EXPORT_SYMBOL(filemap_write_and_wait);
416 * filemap_write_and_wait_range - write out & wait on a file range
417 * @mapping: the address_space for the pages
418 * @lstart: offset in bytes where the range starts
419 * @lend: offset in bytes where the range ends (inclusive)
421 * Write out and wait upon file offsets lstart->lend, inclusive.
423 * Note that `lend' is inclusive (describes the last byte to be written) so
424 * that this function can be used to write to the very end-of-file (end = -1).
426 int filemap_write_and_wait_range(struct address_space *mapping,
427 loff_t lstart, loff_t lend)
431 if (mapping->nrpages) {
432 err = __filemap_fdatawrite_range(mapping, lstart, lend,
434 /* See comment of filemap_write_and_wait() */
436 int err2 = filemap_fdatawait_range(mapping,
442 err = filemap_check_errors(mapping);
446 EXPORT_SYMBOL(filemap_write_and_wait_range);
449 * replace_page_cache_page - replace a pagecache page with a new one
450 * @old: page to be replaced
451 * @new: page to replace with
452 * @gfp_mask: allocation mode
454 * This function replaces a page in the pagecache with a new one. On
455 * success it acquires the pagecache reference for the new page and
456 * drops it for the old page. Both the old and new pages must be
457 * locked. This function does not add the new page to the LRU, the
458 * caller must do that.
460 * The remove + add is atomic. The only way this function can fail is
461 * memory allocation failure.
463 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
467 VM_BUG_ON_PAGE(!PageLocked(old), old);
468 VM_BUG_ON_PAGE(!PageLocked(new), new);
469 VM_BUG_ON_PAGE(new->mapping, new);
471 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
473 struct address_space *mapping = old->mapping;
474 void (*freepage)(struct page *);
476 pgoff_t offset = old->index;
477 freepage = mapping->a_ops->freepage;
480 new->mapping = mapping;
483 spin_lock_irq(&mapping->tree_lock);
484 __delete_from_page_cache(old, NULL);
485 error = radix_tree_insert(&mapping->page_tree, offset, new);
488 __inc_zone_page_state(new, NR_FILE_PAGES);
489 if (PageSwapBacked(new))
490 __inc_zone_page_state(new, NR_SHMEM);
491 spin_unlock_irq(&mapping->tree_lock);
492 /* mem_cgroup codes must not be called under tree_lock */
493 mem_cgroup_replace_page_cache(old, new);
494 radix_tree_preload_end();
497 page_cache_release(old);
502 EXPORT_SYMBOL_GPL(replace_page_cache_page);
504 static int page_cache_tree_insert(struct address_space *mapping,
505 struct page *page, void **shadowp)
507 struct radix_tree_node *node;
511 error = __radix_tree_create(&mapping->page_tree, page->index,
518 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
519 if (!radix_tree_exceptional_entry(p))
523 mapping->nrshadows--;
525 workingset_node_shadows_dec(node);
527 radix_tree_replace_slot(slot, page);
530 workingset_node_pages_inc(node);
532 * Don't track node that contains actual pages.
534 * Avoid acquiring the list_lru lock if already
535 * untracked. The list_empty() test is safe as
536 * node->private_list is protected by
537 * mapping->tree_lock.
539 if (!list_empty(&node->private_list))
540 list_lru_del(&workingset_shadow_nodes,
541 &node->private_list);
546 static int __add_to_page_cache_locked(struct page *page,
547 struct address_space *mapping,
548 pgoff_t offset, gfp_t gfp_mask,
553 VM_BUG_ON_PAGE(!PageLocked(page), page);
554 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
556 error = mem_cgroup_charge_file(page, current->mm,
557 gfp_mask & GFP_RECLAIM_MASK);
561 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
563 mem_cgroup_uncharge_cache_page(page);
567 page_cache_get(page);
568 page->mapping = mapping;
569 page->index = offset;
571 spin_lock_irq(&mapping->tree_lock);
572 error = page_cache_tree_insert(mapping, page, shadowp);
573 radix_tree_preload_end();
576 __inc_zone_page_state(page, NR_FILE_PAGES);
577 spin_unlock_irq(&mapping->tree_lock);
578 trace_mm_filemap_add_to_page_cache(page);
581 page->mapping = NULL;
582 /* Leave page->index set: truncation relies upon it */
583 spin_unlock_irq(&mapping->tree_lock);
584 mem_cgroup_uncharge_cache_page(page);
585 page_cache_release(page);
590 * add_to_page_cache_locked - add a locked page to the pagecache
592 * @mapping: the page's address_space
593 * @offset: page index
594 * @gfp_mask: page allocation mode
596 * This function is used to add a page to the pagecache. It must be locked.
597 * This function does not add the page to the LRU. The caller must do that.
599 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
600 pgoff_t offset, gfp_t gfp_mask)
602 return __add_to_page_cache_locked(page, mapping, offset,
605 EXPORT_SYMBOL(add_to_page_cache_locked);
607 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
608 pgoff_t offset, gfp_t gfp_mask)
613 __set_page_locked(page);
614 ret = __add_to_page_cache_locked(page, mapping, offset,
617 __clear_page_locked(page);
620 * The page might have been evicted from cache only
621 * recently, in which case it should be activated like
622 * any other repeatedly accessed page.
624 if (shadow && workingset_refault(shadow)) {
626 workingset_activation(page);
628 ClearPageActive(page);
633 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
636 struct page *__page_cache_alloc(gfp_t gfp)
641 if (cpuset_do_page_mem_spread()) {
642 unsigned int cpuset_mems_cookie;
644 cpuset_mems_cookie = read_mems_allowed_begin();
645 n = cpuset_mem_spread_node();
646 page = alloc_pages_exact_node(n, gfp, 0);
647 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
651 return alloc_pages(gfp, 0);
653 EXPORT_SYMBOL(__page_cache_alloc);
657 * In order to wait for pages to become available there must be
658 * waitqueues associated with pages. By using a hash table of
659 * waitqueues where the bucket discipline is to maintain all
660 * waiters on the same queue and wake all when any of the pages
661 * become available, and for the woken contexts to check to be
662 * sure the appropriate page became available, this saves space
663 * at a cost of "thundering herd" phenomena during rare hash
666 static wait_queue_head_t *page_waitqueue(struct page *page)
668 const struct zone *zone = page_zone(page);
670 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
673 static inline void wake_up_page(struct page *page, int bit)
675 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
678 void wait_on_page_bit(struct page *page, int bit_nr)
680 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
682 if (test_bit(bit_nr, &page->flags))
683 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
684 TASK_UNINTERRUPTIBLE);
686 EXPORT_SYMBOL(wait_on_page_bit);
688 int wait_on_page_bit_killable(struct page *page, int bit_nr)
690 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
692 if (!test_bit(bit_nr, &page->flags))
695 return __wait_on_bit(page_waitqueue(page), &wait,
696 bit_wait_io, TASK_KILLABLE);
700 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
701 * @page: Page defining the wait queue of interest
702 * @waiter: Waiter to add to the queue
704 * Add an arbitrary @waiter to the wait queue for the nominated @page.
706 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
708 wait_queue_head_t *q = page_waitqueue(page);
711 spin_lock_irqsave(&q->lock, flags);
712 __add_wait_queue(q, waiter);
713 spin_unlock_irqrestore(&q->lock, flags);
715 EXPORT_SYMBOL_GPL(add_page_wait_queue);
718 * unlock_page - unlock a locked page
721 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
722 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
723 * mechananism between PageLocked pages and PageWriteback pages is shared.
724 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
726 * The mb is necessary to enforce ordering between the clear_bit and the read
727 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
729 void unlock_page(struct page *page)
731 VM_BUG_ON_PAGE(!PageLocked(page), page);
732 clear_bit_unlock(PG_locked, &page->flags);
733 smp_mb__after_atomic();
734 wake_up_page(page, PG_locked);
736 EXPORT_SYMBOL(unlock_page);
739 * end_page_writeback - end writeback against a page
742 void end_page_writeback(struct page *page)
745 * TestClearPageReclaim could be used here but it is an atomic
746 * operation and overkill in this particular case. Failing to
747 * shuffle a page marked for immediate reclaim is too mild to
748 * justify taking an atomic operation penalty at the end of
749 * ever page writeback.
751 if (PageReclaim(page)) {
752 ClearPageReclaim(page);
753 rotate_reclaimable_page(page);
756 if (!test_clear_page_writeback(page))
759 smp_mb__after_atomic();
760 wake_up_page(page, PG_writeback);
762 EXPORT_SYMBOL(end_page_writeback);
765 * After completing I/O on a page, call this routine to update the page
766 * flags appropriately
768 void page_endio(struct page *page, int rw, int err)
772 SetPageUptodate(page);
774 ClearPageUptodate(page);
778 } else { /* rw == WRITE */
782 mapping_set_error(page->mapping, err);
784 end_page_writeback(page);
787 EXPORT_SYMBOL_GPL(page_endio);
790 * __lock_page - get a lock on the page, assuming we need to sleep to get it
791 * @page: the page to lock
793 void __lock_page(struct page *page)
795 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
797 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
798 TASK_UNINTERRUPTIBLE);
800 EXPORT_SYMBOL(__lock_page);
802 int __lock_page_killable(struct page *page)
804 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
806 return __wait_on_bit_lock(page_waitqueue(page), &wait,
807 bit_wait_io, TASK_KILLABLE);
809 EXPORT_SYMBOL_GPL(__lock_page_killable);
811 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
814 if (flags & FAULT_FLAG_ALLOW_RETRY) {
816 * CAUTION! In this case, mmap_sem is not released
817 * even though return 0.
819 if (flags & FAULT_FLAG_RETRY_NOWAIT)
822 up_read(&mm->mmap_sem);
823 if (flags & FAULT_FLAG_KILLABLE)
824 wait_on_page_locked_killable(page);
826 wait_on_page_locked(page);
829 if (flags & FAULT_FLAG_KILLABLE) {
832 ret = __lock_page_killable(page);
834 up_read(&mm->mmap_sem);
844 * page_cache_next_hole - find the next hole (not-present entry)
847 * @max_scan: maximum range to search
849 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
850 * lowest indexed hole.
852 * Returns: the index of the hole if found, otherwise returns an index
853 * outside of the set specified (in which case 'return - index >=
854 * max_scan' will be true). In rare cases of index wrap-around, 0 will
857 * page_cache_next_hole may be called under rcu_read_lock. However,
858 * like radix_tree_gang_lookup, this will not atomically search a
859 * snapshot of the tree at a single point in time. For example, if a
860 * hole is created at index 5, then subsequently a hole is created at
861 * index 10, page_cache_next_hole covering both indexes may return 10
862 * if called under rcu_read_lock.
864 pgoff_t page_cache_next_hole(struct address_space *mapping,
865 pgoff_t index, unsigned long max_scan)
869 for (i = 0; i < max_scan; i++) {
872 page = radix_tree_lookup(&mapping->page_tree, index);
873 if (!page || radix_tree_exceptional_entry(page))
882 EXPORT_SYMBOL(page_cache_next_hole);
885 * page_cache_prev_hole - find the prev hole (not-present entry)
888 * @max_scan: maximum range to search
890 * Search backwards in the range [max(index-max_scan+1, 0), index] for
893 * Returns: the index of the hole if found, otherwise returns an index
894 * outside of the set specified (in which case 'index - return >=
895 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
898 * page_cache_prev_hole may be called under rcu_read_lock. However,
899 * like radix_tree_gang_lookup, this will not atomically search a
900 * snapshot of the tree at a single point in time. For example, if a
901 * hole is created at index 10, then subsequently a hole is created at
902 * index 5, page_cache_prev_hole covering both indexes may return 5 if
903 * called under rcu_read_lock.
905 pgoff_t page_cache_prev_hole(struct address_space *mapping,
906 pgoff_t index, unsigned long max_scan)
910 for (i = 0; i < max_scan; i++) {
913 page = radix_tree_lookup(&mapping->page_tree, index);
914 if (!page || radix_tree_exceptional_entry(page))
917 if (index == ULONG_MAX)
923 EXPORT_SYMBOL(page_cache_prev_hole);
926 * find_get_entry - find and get a page cache entry
927 * @mapping: the address_space to search
928 * @offset: the page cache index
930 * Looks up the page cache slot at @mapping & @offset. If there is a
931 * page cache page, it is returned with an increased refcount.
933 * If the slot holds a shadow entry of a previously evicted page, or a
934 * swap entry from shmem/tmpfs, it is returned.
936 * Otherwise, %NULL is returned.
938 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
946 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
948 page = radix_tree_deref_slot(pagep);
951 if (radix_tree_exception(page)) {
952 if (radix_tree_deref_retry(page))
955 * A shadow entry of a recently evicted page,
956 * or a swap entry from shmem/tmpfs. Return
957 * it without attempting to raise page count.
961 if (!page_cache_get_speculative(page))
965 * Has the page moved?
966 * This is part of the lockless pagecache protocol. See
967 * include/linux/pagemap.h for details.
969 if (unlikely(page != *pagep)) {
970 page_cache_release(page);
979 EXPORT_SYMBOL(find_get_entry);
982 * find_lock_entry - locate, pin and lock a page cache entry
983 * @mapping: the address_space to search
984 * @offset: the page cache index
986 * Looks up the page cache slot at @mapping & @offset. If there is a
987 * page cache page, it is returned locked and with an increased
990 * If the slot holds a shadow entry of a previously evicted page, or a
991 * swap entry from shmem/tmpfs, it is returned.
993 * Otherwise, %NULL is returned.
995 * find_lock_entry() may sleep.
997 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1002 page = find_get_entry(mapping, offset);
1003 if (page && !radix_tree_exception(page)) {
1005 /* Has the page been truncated? */
1006 if (unlikely(page->mapping != mapping)) {
1008 page_cache_release(page);
1011 VM_BUG_ON_PAGE(page->index != offset, page);
1015 EXPORT_SYMBOL(find_lock_entry);
1018 * pagecache_get_page - find and get a page reference
1019 * @mapping: the address_space to search
1020 * @offset: the page index
1021 * @fgp_flags: PCG flags
1022 * @gfp_mask: gfp mask to use if a page is to be allocated
1024 * Looks up the page cache slot at @mapping & @offset.
1026 * PCG flags modify how the page is returned
1028 * FGP_ACCESSED: the page will be marked accessed
1029 * FGP_LOCK: Page is return locked
1030 * FGP_CREAT: If page is not present then a new page is allocated using
1031 * @gfp_mask and added to the page cache and the VM's LRU
1032 * list. The page is returned locked and with an increased
1033 * refcount. Otherwise, %NULL is returned.
1035 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1036 * if the GFP flags specified for FGP_CREAT are atomic.
1038 * If there is a page cache page, it is returned with an increased refcount.
1040 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1041 int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask)
1046 page = find_get_entry(mapping, offset);
1047 if (radix_tree_exceptional_entry(page))
1052 if (fgp_flags & FGP_LOCK) {
1053 if (fgp_flags & FGP_NOWAIT) {
1054 if (!trylock_page(page)) {
1055 page_cache_release(page);
1062 /* Has the page been truncated? */
1063 if (unlikely(page->mapping != mapping)) {
1065 page_cache_release(page);
1068 VM_BUG_ON_PAGE(page->index != offset, page);
1071 if (page && (fgp_flags & FGP_ACCESSED))
1072 mark_page_accessed(page);
1075 if (!page && (fgp_flags & FGP_CREAT)) {
1077 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1078 cache_gfp_mask |= __GFP_WRITE;
1079 if (fgp_flags & FGP_NOFS) {
1080 cache_gfp_mask &= ~__GFP_FS;
1081 radix_gfp_mask &= ~__GFP_FS;
1084 page = __page_cache_alloc(cache_gfp_mask);
1088 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1089 fgp_flags |= FGP_LOCK;
1091 /* Init accessed so avoit atomic mark_page_accessed later */
1092 if (fgp_flags & FGP_ACCESSED)
1093 init_page_accessed(page);
1095 err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask);
1096 if (unlikely(err)) {
1097 page_cache_release(page);
1106 EXPORT_SYMBOL(pagecache_get_page);
1109 * find_get_entries - gang pagecache lookup
1110 * @mapping: The address_space to search
1111 * @start: The starting page cache index
1112 * @nr_entries: The maximum number of entries
1113 * @entries: Where the resulting entries are placed
1114 * @indices: The cache indices corresponding to the entries in @entries
1116 * find_get_entries() will search for and return a group of up to
1117 * @nr_entries entries in the mapping. The entries are placed at
1118 * @entries. find_get_entries() takes a reference against any actual
1121 * The search returns a group of mapping-contiguous page cache entries
1122 * with ascending indexes. There may be holes in the indices due to
1123 * not-present pages.
1125 * Any shadow entries of evicted pages, or swap entries from
1126 * shmem/tmpfs, are included in the returned array.
1128 * find_get_entries() returns the number of pages and shadow entries
1131 unsigned find_get_entries(struct address_space *mapping,
1132 pgoff_t start, unsigned int nr_entries,
1133 struct page **entries, pgoff_t *indices)
1136 unsigned int ret = 0;
1137 struct radix_tree_iter iter;
1144 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1147 page = radix_tree_deref_slot(slot);
1148 if (unlikely(!page))
1150 if (radix_tree_exception(page)) {
1151 if (radix_tree_deref_retry(page))
1154 * A shadow entry of a recently evicted page,
1155 * or a swap entry from shmem/tmpfs. Return
1156 * it without attempting to raise page count.
1160 if (!page_cache_get_speculative(page))
1163 /* Has the page moved? */
1164 if (unlikely(page != *slot)) {
1165 page_cache_release(page);
1169 indices[ret] = iter.index;
1170 entries[ret] = page;
1171 if (++ret == nr_entries)
1179 * find_get_pages - gang pagecache lookup
1180 * @mapping: The address_space to search
1181 * @start: The starting page index
1182 * @nr_pages: The maximum number of pages
1183 * @pages: Where the resulting pages are placed
1185 * find_get_pages() will search for and return a group of up to
1186 * @nr_pages pages in the mapping. The pages are placed at @pages.
1187 * find_get_pages() takes a reference against the returned pages.
1189 * The search returns a group of mapping-contiguous pages with ascending
1190 * indexes. There may be holes in the indices due to not-present pages.
1192 * find_get_pages() returns the number of pages which were found.
1194 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1195 unsigned int nr_pages, struct page **pages)
1197 struct radix_tree_iter iter;
1201 if (unlikely(!nr_pages))
1206 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1209 page = radix_tree_deref_slot(slot);
1210 if (unlikely(!page))
1213 if (radix_tree_exception(page)) {
1214 if (radix_tree_deref_retry(page)) {
1216 * Transient condition which can only trigger
1217 * when entry at index 0 moves out of or back
1218 * to root: none yet gotten, safe to restart.
1220 WARN_ON(iter.index);
1224 * A shadow entry of a recently evicted page,
1225 * or a swap entry from shmem/tmpfs. Skip
1231 if (!page_cache_get_speculative(page))
1234 /* Has the page moved? */
1235 if (unlikely(page != *slot)) {
1236 page_cache_release(page);
1241 if (++ret == nr_pages)
1250 * find_get_pages_contig - gang contiguous pagecache lookup
1251 * @mapping: The address_space to search
1252 * @index: The starting page index
1253 * @nr_pages: The maximum number of pages
1254 * @pages: Where the resulting pages are placed
1256 * find_get_pages_contig() works exactly like find_get_pages(), except
1257 * that the returned number of pages are guaranteed to be contiguous.
1259 * find_get_pages_contig() returns the number of pages which were found.
1261 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1262 unsigned int nr_pages, struct page **pages)
1264 struct radix_tree_iter iter;
1266 unsigned int ret = 0;
1268 if (unlikely(!nr_pages))
1273 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1276 page = radix_tree_deref_slot(slot);
1277 /* The hole, there no reason to continue */
1278 if (unlikely(!page))
1281 if (radix_tree_exception(page)) {
1282 if (radix_tree_deref_retry(page)) {
1284 * Transient condition which can only trigger
1285 * when entry at index 0 moves out of or back
1286 * to root: none yet gotten, safe to restart.
1291 * A shadow entry of a recently evicted page,
1292 * or a swap entry from shmem/tmpfs. Stop
1293 * looking for contiguous pages.
1298 if (!page_cache_get_speculative(page))
1301 /* Has the page moved? */
1302 if (unlikely(page != *slot)) {
1303 page_cache_release(page);
1308 * must check mapping and index after taking the ref.
1309 * otherwise we can get both false positives and false
1310 * negatives, which is just confusing to the caller.
1312 if (page->mapping == NULL || page->index != iter.index) {
1313 page_cache_release(page);
1318 if (++ret == nr_pages)
1324 EXPORT_SYMBOL(find_get_pages_contig);
1327 * find_get_pages_tag - find and return pages that match @tag
1328 * @mapping: the address_space to search
1329 * @index: the starting page index
1330 * @tag: the tag index
1331 * @nr_pages: the maximum number of pages
1332 * @pages: where the resulting pages are placed
1334 * Like find_get_pages, except we only return pages which are tagged with
1335 * @tag. We update @index to index the next page for the traversal.
1337 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1338 int tag, unsigned int nr_pages, struct page **pages)
1340 struct radix_tree_iter iter;
1344 if (unlikely(!nr_pages))
1349 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1350 &iter, *index, tag) {
1353 page = radix_tree_deref_slot(slot);
1354 if (unlikely(!page))
1357 if (radix_tree_exception(page)) {
1358 if (radix_tree_deref_retry(page)) {
1360 * Transient condition which can only trigger
1361 * when entry at index 0 moves out of or back
1362 * to root: none yet gotten, safe to restart.
1367 * A shadow entry of a recently evicted page.
1369 * Those entries should never be tagged, but
1370 * this tree walk is lockless and the tags are
1371 * looked up in bulk, one radix tree node at a
1372 * time, so there is a sizable window for page
1373 * reclaim to evict a page we saw tagged.
1380 if (!page_cache_get_speculative(page))
1383 /* Has the page moved? */
1384 if (unlikely(page != *slot)) {
1385 page_cache_release(page);
1390 if (++ret == nr_pages)
1397 *index = pages[ret - 1]->index + 1;
1401 EXPORT_SYMBOL(find_get_pages_tag);
1404 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1405 * a _large_ part of the i/o request. Imagine the worst scenario:
1407 * ---R__________________________________________B__________
1408 * ^ reading here ^ bad block(assume 4k)
1410 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1411 * => failing the whole request => read(R) => read(R+1) =>
1412 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1413 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1414 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1416 * It is going insane. Fix it by quickly scaling down the readahead size.
1418 static void shrink_readahead_size_eio(struct file *filp,
1419 struct file_ra_state *ra)
1425 * do_generic_file_read - generic file read routine
1426 * @filp: the file to read
1427 * @ppos: current file position
1428 * @iter: data destination
1429 * @written: already copied
1431 * This is a generic file read routine, and uses the
1432 * mapping->a_ops->readpage() function for the actual low-level stuff.
1434 * This is really ugly. But the goto's actually try to clarify some
1435 * of the logic when it comes to error handling etc.
1437 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1438 struct iov_iter *iter, ssize_t written)
1440 struct address_space *mapping = filp->f_mapping;
1441 struct inode *inode = mapping->host;
1442 struct file_ra_state *ra = &filp->f_ra;
1446 unsigned long offset; /* offset into pagecache page */
1447 unsigned int prev_offset;
1450 index = *ppos >> PAGE_CACHE_SHIFT;
1451 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1452 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1453 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1454 offset = *ppos & ~PAGE_CACHE_MASK;
1460 unsigned long nr, ret;
1464 page = find_get_page(mapping, index);
1466 page_cache_sync_readahead(mapping,
1468 index, last_index - index);
1469 page = find_get_page(mapping, index);
1470 if (unlikely(page == NULL))
1471 goto no_cached_page;
1473 if (PageReadahead(page)) {
1474 page_cache_async_readahead(mapping,
1476 index, last_index - index);
1478 if (!PageUptodate(page)) {
1479 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1480 !mapping->a_ops->is_partially_uptodate)
1481 goto page_not_up_to_date;
1482 if (!trylock_page(page))
1483 goto page_not_up_to_date;
1484 /* Did it get truncated before we got the lock? */
1486 goto page_not_up_to_date_locked;
1487 if (!mapping->a_ops->is_partially_uptodate(page,
1488 offset, iter->count))
1489 goto page_not_up_to_date_locked;
1494 * i_size must be checked after we know the page is Uptodate.
1496 * Checking i_size after the check allows us to calculate
1497 * the correct value for "nr", which means the zero-filled
1498 * part of the page is not copied back to userspace (unless
1499 * another truncate extends the file - this is desired though).
1502 isize = i_size_read(inode);
1503 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1504 if (unlikely(!isize || index > end_index)) {
1505 page_cache_release(page);
1509 /* nr is the maximum number of bytes to copy from this page */
1510 nr = PAGE_CACHE_SIZE;
1511 if (index == end_index) {
1512 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1514 page_cache_release(page);
1520 /* If users can be writing to this page using arbitrary
1521 * virtual addresses, take care about potential aliasing
1522 * before reading the page on the kernel side.
1524 if (mapping_writably_mapped(mapping))
1525 flush_dcache_page(page);
1528 * When a sequential read accesses a page several times,
1529 * only mark it as accessed the first time.
1531 if (prev_index != index || offset != prev_offset)
1532 mark_page_accessed(page);
1536 * Ok, we have the page, and it's up-to-date, so
1537 * now we can copy it to user space...
1540 ret = copy_page_to_iter(page, offset, nr, iter);
1542 index += offset >> PAGE_CACHE_SHIFT;
1543 offset &= ~PAGE_CACHE_MASK;
1544 prev_offset = offset;
1546 page_cache_release(page);
1548 if (!iov_iter_count(iter))
1556 page_not_up_to_date:
1557 /* Get exclusive access to the page ... */
1558 error = lock_page_killable(page);
1559 if (unlikely(error))
1560 goto readpage_error;
1562 page_not_up_to_date_locked:
1563 /* Did it get truncated before we got the lock? */
1564 if (!page->mapping) {
1566 page_cache_release(page);
1570 /* Did somebody else fill it already? */
1571 if (PageUptodate(page)) {
1578 * A previous I/O error may have been due to temporary
1579 * failures, eg. multipath errors.
1580 * PG_error will be set again if readpage fails.
1582 ClearPageError(page);
1583 /* Start the actual read. The read will unlock the page. */
1584 error = mapping->a_ops->readpage(filp, page);
1586 if (unlikely(error)) {
1587 if (error == AOP_TRUNCATED_PAGE) {
1588 page_cache_release(page);
1592 goto readpage_error;
1595 if (!PageUptodate(page)) {
1596 error = lock_page_killable(page);
1597 if (unlikely(error))
1598 goto readpage_error;
1599 if (!PageUptodate(page)) {
1600 if (page->mapping == NULL) {
1602 * invalidate_mapping_pages got it
1605 page_cache_release(page);
1609 shrink_readahead_size_eio(filp, ra);
1611 goto readpage_error;
1619 /* UHHUH! A synchronous read error occurred. Report it */
1620 page_cache_release(page);
1625 * Ok, it wasn't cached, so we need to create a new
1628 page = page_cache_alloc_cold(mapping);
1633 error = add_to_page_cache_lru(page, mapping,
1636 page_cache_release(page);
1637 if (error == -EEXIST) {
1647 ra->prev_pos = prev_index;
1648 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1649 ra->prev_pos |= prev_offset;
1651 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1652 file_accessed(filp);
1653 return written ? written : error;
1657 * generic_file_read_iter - generic filesystem read routine
1658 * @iocb: kernel I/O control block
1659 * @iter: destination for the data read
1661 * This is the "read_iter()" routine for all filesystems
1662 * that can use the page cache directly.
1665 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1667 struct file *file = iocb->ki_filp;
1669 loff_t *ppos = &iocb->ki_pos;
1672 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1673 if (file->f_flags & O_DIRECT) {
1674 struct address_space *mapping = file->f_mapping;
1675 struct inode *inode = mapping->host;
1676 size_t count = iov_iter_count(iter);
1680 goto out; /* skip atime */
1681 size = i_size_read(inode);
1682 retval = filemap_write_and_wait_range(mapping, pos,
1685 struct iov_iter data = *iter;
1686 retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos);
1690 *ppos = pos + retval;
1691 iov_iter_advance(iter, retval);
1695 * Btrfs can have a short DIO read if we encounter
1696 * compressed extents, so if there was an error, or if
1697 * we've already read everything we wanted to, or if
1698 * there was a short read because we hit EOF, go ahead
1699 * and return. Otherwise fallthrough to buffered io for
1700 * the rest of the read.
1702 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) {
1703 file_accessed(file);
1708 retval = do_generic_file_read(file, ppos, iter, retval);
1712 EXPORT_SYMBOL(generic_file_read_iter);
1716 * page_cache_read - adds requested page to the page cache if not already there
1717 * @file: file to read
1718 * @offset: page index
1720 * This adds the requested page to the page cache if it isn't already there,
1721 * and schedules an I/O to read in its contents from disk.
1723 static int page_cache_read(struct file *file, pgoff_t offset)
1725 struct address_space *mapping = file->f_mapping;
1730 page = page_cache_alloc_cold(mapping);
1734 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1736 ret = mapping->a_ops->readpage(file, page);
1737 else if (ret == -EEXIST)
1738 ret = 0; /* losing race to add is OK */
1740 page_cache_release(page);
1742 } while (ret == AOP_TRUNCATED_PAGE);
1747 #define MMAP_LOTSAMISS (100)
1750 * Synchronous readahead happens when we don't even find
1751 * a page in the page cache at all.
1753 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1754 struct file_ra_state *ra,
1758 unsigned long ra_pages;
1759 struct address_space *mapping = file->f_mapping;
1761 /* If we don't want any read-ahead, don't bother */
1762 if (vma->vm_flags & VM_RAND_READ)
1767 if (vma->vm_flags & VM_SEQ_READ) {
1768 page_cache_sync_readahead(mapping, ra, file, offset,
1773 /* Avoid banging the cache line if not needed */
1774 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1778 * Do we miss much more than hit in this file? If so,
1779 * stop bothering with read-ahead. It will only hurt.
1781 if (ra->mmap_miss > MMAP_LOTSAMISS)
1787 ra_pages = max_sane_readahead(ra->ra_pages);
1788 ra->start = max_t(long, 0, offset - ra_pages / 2);
1789 ra->size = ra_pages;
1790 ra->async_size = ra_pages / 4;
1791 ra_submit(ra, mapping, file);
1795 * Asynchronous readahead happens when we find the page and PG_readahead,
1796 * so we want to possibly extend the readahead further..
1798 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1799 struct file_ra_state *ra,
1804 struct address_space *mapping = file->f_mapping;
1806 /* If we don't want any read-ahead, don't bother */
1807 if (vma->vm_flags & VM_RAND_READ)
1809 if (ra->mmap_miss > 0)
1811 if (PageReadahead(page))
1812 page_cache_async_readahead(mapping, ra, file,
1813 page, offset, ra->ra_pages);
1817 * filemap_fault - read in file data for page fault handling
1818 * @vma: vma in which the fault was taken
1819 * @vmf: struct vm_fault containing details of the fault
1821 * filemap_fault() is invoked via the vma operations vector for a
1822 * mapped memory region to read in file data during a page fault.
1824 * The goto's are kind of ugly, but this streamlines the normal case of having
1825 * it in the page cache, and handles the special cases reasonably without
1826 * having a lot of duplicated code.
1828 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1831 struct file *file = vma->vm_file;
1832 struct address_space *mapping = file->f_mapping;
1833 struct file_ra_state *ra = &file->f_ra;
1834 struct inode *inode = mapping->host;
1835 pgoff_t offset = vmf->pgoff;
1840 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1841 if (offset >= size >> PAGE_CACHE_SHIFT)
1842 return VM_FAULT_SIGBUS;
1845 * Do we have something in the page cache already?
1847 page = find_get_page(mapping, offset);
1848 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1850 * We found the page, so try async readahead before
1851 * waiting for the lock.
1853 do_async_mmap_readahead(vma, ra, file, page, offset);
1855 /* No page in the page cache at all */
1856 do_sync_mmap_readahead(vma, ra, file, offset);
1857 count_vm_event(PGMAJFAULT);
1858 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1859 ret = VM_FAULT_MAJOR;
1861 page = find_get_page(mapping, offset);
1863 goto no_cached_page;
1866 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1867 page_cache_release(page);
1868 return ret | VM_FAULT_RETRY;
1871 /* Did it get truncated? */
1872 if (unlikely(page->mapping != mapping)) {
1877 VM_BUG_ON_PAGE(page->index != offset, page);
1880 * We have a locked page in the page cache, now we need to check
1881 * that it's up-to-date. If not, it is going to be due to an error.
1883 if (unlikely(!PageUptodate(page)))
1884 goto page_not_uptodate;
1887 * Found the page and have a reference on it.
1888 * We must recheck i_size under page lock.
1890 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1891 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1893 page_cache_release(page);
1894 return VM_FAULT_SIGBUS;
1898 return ret | VM_FAULT_LOCKED;
1902 * We're only likely to ever get here if MADV_RANDOM is in
1905 error = page_cache_read(file, offset);
1908 * The page we want has now been added to the page cache.
1909 * In the unlikely event that someone removed it in the
1910 * meantime, we'll just come back here and read it again.
1916 * An error return from page_cache_read can result if the
1917 * system is low on memory, or a problem occurs while trying
1920 if (error == -ENOMEM)
1921 return VM_FAULT_OOM;
1922 return VM_FAULT_SIGBUS;
1926 * Umm, take care of errors if the page isn't up-to-date.
1927 * Try to re-read it _once_. We do this synchronously,
1928 * because there really aren't any performance issues here
1929 * and we need to check for errors.
1931 ClearPageError(page);
1932 error = mapping->a_ops->readpage(file, page);
1934 wait_on_page_locked(page);
1935 if (!PageUptodate(page))
1938 page_cache_release(page);
1940 if (!error || error == AOP_TRUNCATED_PAGE)
1943 /* Things didn't work out. Return zero to tell the mm layer so. */
1944 shrink_readahead_size_eio(file, ra);
1945 return VM_FAULT_SIGBUS;
1947 EXPORT_SYMBOL(filemap_fault);
1949 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1951 struct radix_tree_iter iter;
1953 struct file *file = vma->vm_file;
1954 struct address_space *mapping = file->f_mapping;
1957 unsigned long address = (unsigned long) vmf->virtual_address;
1962 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
1963 if (iter.index > vmf->max_pgoff)
1966 page = radix_tree_deref_slot(slot);
1967 if (unlikely(!page))
1969 if (radix_tree_exception(page)) {
1970 if (radix_tree_deref_retry(page))
1976 if (!page_cache_get_speculative(page))
1979 /* Has the page moved? */
1980 if (unlikely(page != *slot)) {
1981 page_cache_release(page);
1985 if (!PageUptodate(page) ||
1986 PageReadahead(page) ||
1989 if (!trylock_page(page))
1992 if (page->mapping != mapping || !PageUptodate(page))
1995 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
1996 if (page->index >= size >> PAGE_CACHE_SHIFT)
1999 pte = vmf->pte + page->index - vmf->pgoff;
2000 if (!pte_none(*pte))
2003 if (file->f_ra.mmap_miss > 0)
2004 file->f_ra.mmap_miss--;
2005 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2006 do_set_pte(vma, addr, page, pte, false, false);
2012 page_cache_release(page);
2014 if (iter.index == vmf->max_pgoff)
2019 EXPORT_SYMBOL(filemap_map_pages);
2021 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2023 struct page *page = vmf->page;
2024 struct inode *inode = file_inode(vma->vm_file);
2025 int ret = VM_FAULT_LOCKED;
2027 sb_start_pagefault(inode->i_sb);
2028 file_update_time(vma->vm_file);
2030 if (page->mapping != inode->i_mapping) {
2032 ret = VM_FAULT_NOPAGE;
2036 * We mark the page dirty already here so that when freeze is in
2037 * progress, we are guaranteed that writeback during freezing will
2038 * see the dirty page and writeprotect it again.
2040 set_page_dirty(page);
2041 wait_for_stable_page(page);
2043 sb_end_pagefault(inode->i_sb);
2046 EXPORT_SYMBOL(filemap_page_mkwrite);
2048 const struct vm_operations_struct generic_file_vm_ops = {
2049 .fault = filemap_fault,
2050 .map_pages = filemap_map_pages,
2051 .page_mkwrite = filemap_page_mkwrite,
2052 .remap_pages = generic_file_remap_pages,
2055 /* This is used for a general mmap of a disk file */
2057 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2059 struct address_space *mapping = file->f_mapping;
2061 if (!mapping->a_ops->readpage)
2063 file_accessed(file);
2064 vma->vm_ops = &generic_file_vm_ops;
2069 * This is for filesystems which do not implement ->writepage.
2071 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2073 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2075 return generic_file_mmap(file, vma);
2078 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2082 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2086 #endif /* CONFIG_MMU */
2088 EXPORT_SYMBOL(generic_file_mmap);
2089 EXPORT_SYMBOL(generic_file_readonly_mmap);
2091 static struct page *wait_on_page_read(struct page *page)
2093 if (!IS_ERR(page)) {
2094 wait_on_page_locked(page);
2095 if (!PageUptodate(page)) {
2096 page_cache_release(page);
2097 page = ERR_PTR(-EIO);
2103 static struct page *__read_cache_page(struct address_space *mapping,
2105 int (*filler)(void *, struct page *),
2112 page = find_get_page(mapping, index);
2114 page = __page_cache_alloc(gfp | __GFP_COLD);
2116 return ERR_PTR(-ENOMEM);
2117 err = add_to_page_cache_lru(page, mapping, index, gfp);
2118 if (unlikely(err)) {
2119 page_cache_release(page);
2122 /* Presumably ENOMEM for radix tree node */
2123 return ERR_PTR(err);
2125 err = filler(data, page);
2127 page_cache_release(page);
2128 page = ERR_PTR(err);
2130 page = wait_on_page_read(page);
2136 static struct page *do_read_cache_page(struct address_space *mapping,
2138 int (*filler)(void *, struct page *),
2147 page = __read_cache_page(mapping, index, filler, data, gfp);
2150 if (PageUptodate(page))
2154 if (!page->mapping) {
2156 page_cache_release(page);
2159 if (PageUptodate(page)) {
2163 err = filler(data, page);
2165 page_cache_release(page);
2166 return ERR_PTR(err);
2168 page = wait_on_page_read(page);
2173 mark_page_accessed(page);
2178 * read_cache_page - read into page cache, fill it if needed
2179 * @mapping: the page's address_space
2180 * @index: the page index
2181 * @filler: function to perform the read
2182 * @data: first arg to filler(data, page) function, often left as NULL
2184 * Read into the page cache. If a page already exists, and PageUptodate() is
2185 * not set, try to fill the page and wait for it to become unlocked.
2187 * If the page does not get brought uptodate, return -EIO.
2189 struct page *read_cache_page(struct address_space *mapping,
2191 int (*filler)(void *, struct page *),
2194 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2196 EXPORT_SYMBOL(read_cache_page);
2199 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2200 * @mapping: the page's address_space
2201 * @index: the page index
2202 * @gfp: the page allocator flags to use if allocating
2204 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2205 * any new page allocations done using the specified allocation flags.
2207 * If the page does not get brought uptodate, return -EIO.
2209 struct page *read_cache_page_gfp(struct address_space *mapping,
2213 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2215 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2217 EXPORT_SYMBOL(read_cache_page_gfp);
2220 * Performs necessary checks before doing a write
2222 * Can adjust writing position or amount of bytes to write.
2223 * Returns appropriate error code that caller should return or
2224 * zero in case that write should be allowed.
2226 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2228 struct inode *inode = file->f_mapping->host;
2229 unsigned long limit = rlimit(RLIMIT_FSIZE);
2231 if (unlikely(*pos < 0))
2235 /* FIXME: this is for backwards compatibility with 2.4 */
2236 if (file->f_flags & O_APPEND)
2237 *pos = i_size_read(inode);
2239 if (limit != RLIM_INFINITY) {
2240 if (*pos >= limit) {
2241 send_sig(SIGXFSZ, current, 0);
2244 if (*count > limit - (typeof(limit))*pos) {
2245 *count = limit - (typeof(limit))*pos;
2253 if (unlikely(*pos + *count > MAX_NON_LFS &&
2254 !(file->f_flags & O_LARGEFILE))) {
2255 if (*pos >= MAX_NON_LFS) {
2258 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2259 *count = MAX_NON_LFS - (unsigned long)*pos;
2264 * Are we about to exceed the fs block limit ?
2266 * If we have written data it becomes a short write. If we have
2267 * exceeded without writing data we send a signal and return EFBIG.
2268 * Linus frestrict idea will clean these up nicely..
2270 if (likely(!isblk)) {
2271 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2272 if (*count || *pos > inode->i_sb->s_maxbytes) {
2275 /* zero-length writes at ->s_maxbytes are OK */
2278 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2279 *count = inode->i_sb->s_maxbytes - *pos;
2283 if (bdev_read_only(I_BDEV(inode)))
2285 isize = i_size_read(inode);
2286 if (*pos >= isize) {
2287 if (*count || *pos > isize)
2291 if (*pos + *count > isize)
2292 *count = isize - *pos;
2299 EXPORT_SYMBOL(generic_write_checks);
2301 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2302 loff_t pos, unsigned len, unsigned flags,
2303 struct page **pagep, void **fsdata)
2305 const struct address_space_operations *aops = mapping->a_ops;
2307 return aops->write_begin(file, mapping, pos, len, flags,
2310 EXPORT_SYMBOL(pagecache_write_begin);
2312 int pagecache_write_end(struct file *file, struct address_space *mapping,
2313 loff_t pos, unsigned len, unsigned copied,
2314 struct page *page, void *fsdata)
2316 const struct address_space_operations *aops = mapping->a_ops;
2318 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2320 EXPORT_SYMBOL(pagecache_write_end);
2323 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2325 struct file *file = iocb->ki_filp;
2326 struct address_space *mapping = file->f_mapping;
2327 struct inode *inode = mapping->host;
2331 struct iov_iter data;
2333 write_len = iov_iter_count(from);
2334 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2336 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2341 * After a write we want buffered reads to be sure to go to disk to get
2342 * the new data. We invalidate clean cached page from the region we're
2343 * about to write. We do this *before* the write so that we can return
2344 * without clobbering -EIOCBQUEUED from ->direct_IO().
2346 if (mapping->nrpages) {
2347 written = invalidate_inode_pages2_range(mapping,
2348 pos >> PAGE_CACHE_SHIFT, end);
2350 * If a page can not be invalidated, return 0 to fall back
2351 * to buffered write.
2354 if (written == -EBUSY)
2361 written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos);
2364 * Finally, try again to invalidate clean pages which might have been
2365 * cached by non-direct readahead, or faulted in by get_user_pages()
2366 * if the source of the write was an mmap'ed region of the file
2367 * we're writing. Either one is a pretty crazy thing to do,
2368 * so we don't support it 100%. If this invalidation
2369 * fails, tough, the write still worked...
2371 if (mapping->nrpages) {
2372 invalidate_inode_pages2_range(mapping,
2373 pos >> PAGE_CACHE_SHIFT, end);
2378 iov_iter_advance(from, written);
2379 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2380 i_size_write(inode, pos);
2381 mark_inode_dirty(inode);
2388 EXPORT_SYMBOL(generic_file_direct_write);
2391 * Find or create a page at the given pagecache position. Return the locked
2392 * page. This function is specifically for buffered writes.
2394 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2395 pgoff_t index, unsigned flags)
2398 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2400 if (flags & AOP_FLAG_NOFS)
2401 fgp_flags |= FGP_NOFS;
2403 page = pagecache_get_page(mapping, index, fgp_flags,
2404 mapping_gfp_mask(mapping),
2407 wait_for_stable_page(page);
2411 EXPORT_SYMBOL(grab_cache_page_write_begin);
2413 ssize_t generic_perform_write(struct file *file,
2414 struct iov_iter *i, loff_t pos)
2416 struct address_space *mapping = file->f_mapping;
2417 const struct address_space_operations *a_ops = mapping->a_ops;
2419 ssize_t written = 0;
2420 unsigned int flags = 0;
2423 * Copies from kernel address space cannot fail (NFSD is a big user).
2425 if (segment_eq(get_fs(), KERNEL_DS))
2426 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2430 unsigned long offset; /* Offset into pagecache page */
2431 unsigned long bytes; /* Bytes to write to page */
2432 size_t copied; /* Bytes copied from user */
2435 offset = (pos & (PAGE_CACHE_SIZE - 1));
2436 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2441 * Bring in the user page that we will copy from _first_.
2442 * Otherwise there's a nasty deadlock on copying from the
2443 * same page as we're writing to, without it being marked
2446 * Not only is this an optimisation, but it is also required
2447 * to check that the address is actually valid, when atomic
2448 * usercopies are used, below.
2450 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2455 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2457 if (unlikely(status < 0))
2460 if (mapping_writably_mapped(mapping))
2461 flush_dcache_page(page);
2463 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2464 flush_dcache_page(page);
2466 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2468 if (unlikely(status < 0))
2474 iov_iter_advance(i, copied);
2475 if (unlikely(copied == 0)) {
2477 * If we were unable to copy any data at all, we must
2478 * fall back to a single segment length write.
2480 * If we didn't fallback here, we could livelock
2481 * because not all segments in the iov can be copied at
2482 * once without a pagefault.
2484 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2485 iov_iter_single_seg_count(i));
2491 balance_dirty_pages_ratelimited(mapping);
2492 if (fatal_signal_pending(current)) {
2496 } while (iov_iter_count(i));
2498 return written ? written : status;
2500 EXPORT_SYMBOL(generic_perform_write);
2503 * __generic_file_write_iter - write data to a file
2504 * @iocb: IO state structure (file, offset, etc.)
2505 * @from: iov_iter with data to write
2507 * This function does all the work needed for actually writing data to a
2508 * file. It does all basic checks, removes SUID from the file, updates
2509 * modification times and calls proper subroutines depending on whether we
2510 * do direct IO or a standard buffered write.
2512 * It expects i_mutex to be grabbed unless we work on a block device or similar
2513 * object which does not need locking at all.
2515 * This function does *not* take care of syncing data in case of O_SYNC write.
2516 * A caller has to handle it. This is mainly due to the fact that we want to
2517 * avoid syncing under i_mutex.
2519 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2521 struct file *file = iocb->ki_filp;
2522 struct address_space * mapping = file->f_mapping;
2523 struct inode *inode = mapping->host;
2524 loff_t pos = iocb->ki_pos;
2525 ssize_t written = 0;
2528 size_t count = iov_iter_count(from);
2530 /* We can write back this queue in page reclaim */
2531 current->backing_dev_info = mapping->backing_dev_info;
2532 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2539 iov_iter_truncate(from, count);
2541 err = file_remove_suid(file);
2545 err = file_update_time(file);
2549 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2550 if (unlikely(file->f_flags & O_DIRECT)) {
2553 written = generic_file_direct_write(iocb, from, pos);
2554 if (written < 0 || written == count)
2558 * direct-io write to a hole: fall through to buffered I/O
2559 * for completing the rest of the request.
2564 status = generic_perform_write(file, from, pos);
2566 * If generic_perform_write() returned a synchronous error
2567 * then we want to return the number of bytes which were
2568 * direct-written, or the error code if that was zero. Note
2569 * that this differs from normal direct-io semantics, which
2570 * will return -EFOO even if some bytes were written.
2572 if (unlikely(status < 0) && !written) {
2576 iocb->ki_pos = pos + status;
2578 * We need to ensure that the page cache pages are written to
2579 * disk and invalidated to preserve the expected O_DIRECT
2582 endbyte = pos + status - 1;
2583 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2586 invalidate_mapping_pages(mapping,
2587 pos >> PAGE_CACHE_SHIFT,
2588 endbyte >> PAGE_CACHE_SHIFT);
2591 * We don't know how much we wrote, so just return
2592 * the number of bytes which were direct-written
2596 written = generic_perform_write(file, from, pos);
2597 if (likely(written >= 0))
2598 iocb->ki_pos = pos + written;
2601 current->backing_dev_info = NULL;
2602 return written ? written : err;
2604 EXPORT_SYMBOL(__generic_file_write_iter);
2607 * generic_file_write_iter - write data to a file
2608 * @iocb: IO state structure
2609 * @from: iov_iter with data to write
2611 * This is a wrapper around __generic_file_write_iter() to be used by most
2612 * filesystems. It takes care of syncing the file in case of O_SYNC file
2613 * and acquires i_mutex as needed.
2615 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2617 struct file *file = iocb->ki_filp;
2618 struct inode *inode = file->f_mapping->host;
2621 mutex_lock(&inode->i_mutex);
2622 ret = __generic_file_write_iter(iocb, from);
2623 mutex_unlock(&inode->i_mutex);
2628 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2634 EXPORT_SYMBOL(generic_file_write_iter);
2637 * try_to_release_page() - release old fs-specific metadata on a page
2639 * @page: the page which the kernel is trying to free
2640 * @gfp_mask: memory allocation flags (and I/O mode)
2642 * The address_space is to try to release any data against the page
2643 * (presumably at page->private). If the release was successful, return `1'.
2644 * Otherwise return zero.
2646 * This may also be called if PG_fscache is set on a page, indicating that the
2647 * page is known to the local caching routines.
2649 * The @gfp_mask argument specifies whether I/O may be performed to release
2650 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2653 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2655 struct address_space * const mapping = page->mapping;
2657 BUG_ON(!PageLocked(page));
2658 if (PageWriteback(page))
2661 if (mapping && mapping->a_ops->releasepage)
2662 return mapping->a_ops->releasepage(page, gfp_mask);
2663 return try_to_free_buffers(page);
2666 EXPORT_SYMBOL(try_to_release_page);