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 sleep_on_page(void *word)
250 static int sleep_on_page_killable(void *word)
253 return fatal_signal_pending(current) ? -EINTR : 0;
256 static int filemap_check_errors(struct address_space *mapping)
259 /* Check for outstanding write errors */
260 if (test_bit(AS_ENOSPC, &mapping->flags) &&
261 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
263 if (test_bit(AS_EIO, &mapping->flags) &&
264 test_and_clear_bit(AS_EIO, &mapping->flags))
270 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
271 * @mapping: address space structure to write
272 * @start: offset in bytes where the range starts
273 * @end: offset in bytes where the range ends (inclusive)
274 * @sync_mode: enable synchronous operation
276 * Start writeback against all of a mapping's dirty pages that lie
277 * within the byte offsets <start, end> inclusive.
279 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
280 * opposed to a regular memory cleansing writeback. The difference between
281 * these two operations is that if a dirty page/buffer is encountered, it must
282 * be waited upon, and not just skipped over.
284 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
285 loff_t end, int sync_mode)
288 struct writeback_control wbc = {
289 .sync_mode = sync_mode,
290 .nr_to_write = LONG_MAX,
291 .range_start = start,
295 if (!mapping_cap_writeback_dirty(mapping))
298 ret = do_writepages(mapping, &wbc);
302 static inline int __filemap_fdatawrite(struct address_space *mapping,
305 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
308 int filemap_fdatawrite(struct address_space *mapping)
310 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
312 EXPORT_SYMBOL(filemap_fdatawrite);
314 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
317 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
319 EXPORT_SYMBOL(filemap_fdatawrite_range);
322 * filemap_flush - mostly a non-blocking flush
323 * @mapping: target address_space
325 * This is a mostly non-blocking flush. Not suitable for data-integrity
326 * purposes - I/O may not be started against all dirty pages.
328 int filemap_flush(struct address_space *mapping)
330 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
332 EXPORT_SYMBOL(filemap_flush);
335 * filemap_fdatawait_range - wait for writeback to complete
336 * @mapping: address space structure to wait for
337 * @start_byte: offset in bytes where the range starts
338 * @end_byte: offset in bytes where the range ends (inclusive)
340 * Walk the list of under-writeback pages of the given address space
341 * in the given range and wait for all of them.
343 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
346 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
347 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
352 if (end_byte < start_byte)
355 pagevec_init(&pvec, 0);
356 while ((index <= end) &&
357 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
358 PAGECACHE_TAG_WRITEBACK,
359 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
362 for (i = 0; i < nr_pages; i++) {
363 struct page *page = pvec.pages[i];
365 /* until radix tree lookup accepts end_index */
366 if (page->index > end)
369 wait_on_page_writeback(page);
370 if (TestClearPageError(page))
373 pagevec_release(&pvec);
377 ret2 = filemap_check_errors(mapping);
383 EXPORT_SYMBOL(filemap_fdatawait_range);
386 * filemap_fdatawait - wait for all under-writeback pages to complete
387 * @mapping: address space structure to wait for
389 * Walk the list of under-writeback pages of the given address space
390 * and wait for all of them.
392 int filemap_fdatawait(struct address_space *mapping)
394 loff_t i_size = i_size_read(mapping->host);
399 return filemap_fdatawait_range(mapping, 0, i_size - 1);
401 EXPORT_SYMBOL(filemap_fdatawait);
403 int filemap_write_and_wait(struct address_space *mapping)
407 if (mapping->nrpages) {
408 err = filemap_fdatawrite(mapping);
410 * Even if the above returned error, the pages may be
411 * written partially (e.g. -ENOSPC), so we wait for it.
412 * But the -EIO is special case, it may indicate the worst
413 * thing (e.g. bug) happened, so we avoid waiting for it.
416 int err2 = filemap_fdatawait(mapping);
421 err = filemap_check_errors(mapping);
425 EXPORT_SYMBOL(filemap_write_and_wait);
428 * filemap_write_and_wait_range - write out & wait on a file range
429 * @mapping: the address_space for the pages
430 * @lstart: offset in bytes where the range starts
431 * @lend: offset in bytes where the range ends (inclusive)
433 * Write out and wait upon file offsets lstart->lend, inclusive.
435 * Note that `lend' is inclusive (describes the last byte to be written) so
436 * that this function can be used to write to the very end-of-file (end = -1).
438 int filemap_write_and_wait_range(struct address_space *mapping,
439 loff_t lstart, loff_t lend)
443 if (mapping->nrpages) {
444 err = __filemap_fdatawrite_range(mapping, lstart, lend,
446 /* See comment of filemap_write_and_wait() */
448 int err2 = filemap_fdatawait_range(mapping,
454 err = filemap_check_errors(mapping);
458 EXPORT_SYMBOL(filemap_write_and_wait_range);
461 * replace_page_cache_page - replace a pagecache page with a new one
462 * @old: page to be replaced
463 * @new: page to replace with
464 * @gfp_mask: allocation mode
466 * This function replaces a page in the pagecache with a new one. On
467 * success it acquires the pagecache reference for the new page and
468 * drops it for the old page. Both the old and new pages must be
469 * locked. This function does not add the new page to the LRU, the
470 * caller must do that.
472 * The remove + add is atomic. The only way this function can fail is
473 * memory allocation failure.
475 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
479 VM_BUG_ON_PAGE(!PageLocked(old), old);
480 VM_BUG_ON_PAGE(!PageLocked(new), new);
481 VM_BUG_ON_PAGE(new->mapping, new);
483 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
485 struct address_space *mapping = old->mapping;
486 void (*freepage)(struct page *);
488 pgoff_t offset = old->index;
489 freepage = mapping->a_ops->freepage;
492 new->mapping = mapping;
495 spin_lock_irq(&mapping->tree_lock);
496 __delete_from_page_cache(old, NULL);
497 error = radix_tree_insert(&mapping->page_tree, offset, new);
500 __inc_zone_page_state(new, NR_FILE_PAGES);
501 if (PageSwapBacked(new))
502 __inc_zone_page_state(new, NR_SHMEM);
503 spin_unlock_irq(&mapping->tree_lock);
504 /* mem_cgroup codes must not be called under tree_lock */
505 mem_cgroup_replace_page_cache(old, new);
506 radix_tree_preload_end();
509 page_cache_release(old);
514 EXPORT_SYMBOL_GPL(replace_page_cache_page);
516 static int page_cache_tree_insert(struct address_space *mapping,
517 struct page *page, void **shadowp)
519 struct radix_tree_node *node;
523 error = __radix_tree_create(&mapping->page_tree, page->index,
530 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
531 if (!radix_tree_exceptional_entry(p))
535 mapping->nrshadows--;
537 workingset_node_shadows_dec(node);
539 radix_tree_replace_slot(slot, page);
542 workingset_node_pages_inc(node);
544 * Don't track node that contains actual pages.
546 * Avoid acquiring the list_lru lock if already
547 * untracked. The list_empty() test is safe as
548 * node->private_list is protected by
549 * mapping->tree_lock.
551 if (!list_empty(&node->private_list))
552 list_lru_del(&workingset_shadow_nodes,
553 &node->private_list);
558 static int __add_to_page_cache_locked(struct page *page,
559 struct address_space *mapping,
560 pgoff_t offset, gfp_t gfp_mask,
565 VM_BUG_ON_PAGE(!PageLocked(page), page);
566 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
568 error = mem_cgroup_charge_file(page, current->mm,
569 gfp_mask & GFP_RECLAIM_MASK);
573 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
575 mem_cgroup_uncharge_cache_page(page);
579 page_cache_get(page);
580 page->mapping = mapping;
581 page->index = offset;
583 spin_lock_irq(&mapping->tree_lock);
584 error = page_cache_tree_insert(mapping, page, shadowp);
585 radix_tree_preload_end();
588 __inc_zone_page_state(page, NR_FILE_PAGES);
589 spin_unlock_irq(&mapping->tree_lock);
590 trace_mm_filemap_add_to_page_cache(page);
593 page->mapping = NULL;
594 /* Leave page->index set: truncation relies upon it */
595 spin_unlock_irq(&mapping->tree_lock);
596 mem_cgroup_uncharge_cache_page(page);
597 page_cache_release(page);
602 * add_to_page_cache_locked - add a locked page to the pagecache
604 * @mapping: the page's address_space
605 * @offset: page index
606 * @gfp_mask: page allocation mode
608 * This function is used to add a page to the pagecache. It must be locked.
609 * This function does not add the page to the LRU. The caller must do that.
611 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
612 pgoff_t offset, gfp_t gfp_mask)
614 return __add_to_page_cache_locked(page, mapping, offset,
617 EXPORT_SYMBOL(add_to_page_cache_locked);
619 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
620 pgoff_t offset, gfp_t gfp_mask)
625 __set_page_locked(page);
626 ret = __add_to_page_cache_locked(page, mapping, offset,
629 __clear_page_locked(page);
632 * The page might have been evicted from cache only
633 * recently, in which case it should be activated like
634 * any other repeatedly accessed page.
636 if (shadow && workingset_refault(shadow)) {
638 workingset_activation(page);
640 ClearPageActive(page);
645 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
648 struct page *__page_cache_alloc(gfp_t gfp)
653 if (cpuset_do_page_mem_spread()) {
654 unsigned int cpuset_mems_cookie;
656 cpuset_mems_cookie = read_mems_allowed_begin();
657 n = cpuset_mem_spread_node();
658 page = alloc_pages_exact_node(n, gfp, 0);
659 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
663 return alloc_pages(gfp, 0);
665 EXPORT_SYMBOL(__page_cache_alloc);
669 * In order to wait for pages to become available there must be
670 * waitqueues associated with pages. By using a hash table of
671 * waitqueues where the bucket discipline is to maintain all
672 * waiters on the same queue and wake all when any of the pages
673 * become available, and for the woken contexts to check to be
674 * sure the appropriate page became available, this saves space
675 * at a cost of "thundering herd" phenomena during rare hash
678 static wait_queue_head_t *page_waitqueue(struct page *page)
680 const struct zone *zone = page_zone(page);
682 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
685 static inline void wake_up_page(struct page *page, int bit)
687 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
690 void wait_on_page_bit(struct page *page, int bit_nr)
692 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
694 if (test_bit(bit_nr, &page->flags))
695 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
696 TASK_UNINTERRUPTIBLE);
698 EXPORT_SYMBOL(wait_on_page_bit);
700 int wait_on_page_bit_killable(struct page *page, int bit_nr)
702 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
704 if (!test_bit(bit_nr, &page->flags))
707 return __wait_on_bit(page_waitqueue(page), &wait,
708 sleep_on_page_killable, TASK_KILLABLE);
712 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
713 * @page: Page defining the wait queue of interest
714 * @waiter: Waiter to add to the queue
716 * Add an arbitrary @waiter to the wait queue for the nominated @page.
718 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
720 wait_queue_head_t *q = page_waitqueue(page);
723 spin_lock_irqsave(&q->lock, flags);
724 __add_wait_queue(q, waiter);
725 spin_unlock_irqrestore(&q->lock, flags);
727 EXPORT_SYMBOL_GPL(add_page_wait_queue);
730 * unlock_page - unlock a locked page
733 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
734 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
735 * mechananism between PageLocked pages and PageWriteback pages is shared.
736 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
738 * The mb is necessary to enforce ordering between the clear_bit and the read
739 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
741 void unlock_page(struct page *page)
743 VM_BUG_ON_PAGE(!PageLocked(page), page);
744 clear_bit_unlock(PG_locked, &page->flags);
745 smp_mb__after_atomic();
746 wake_up_page(page, PG_locked);
748 EXPORT_SYMBOL(unlock_page);
751 * end_page_writeback - end writeback against a page
754 void end_page_writeback(struct page *page)
756 if (TestClearPageReclaim(page))
757 rotate_reclaimable_page(page);
759 if (!test_clear_page_writeback(page))
762 smp_mb__after_atomic();
763 wake_up_page(page, PG_writeback);
765 EXPORT_SYMBOL(end_page_writeback);
768 * After completing I/O on a page, call this routine to update the page
769 * flags appropriately
771 void page_endio(struct page *page, int rw, int err)
775 SetPageUptodate(page);
777 ClearPageUptodate(page);
781 } else { /* rw == WRITE */
785 mapping_set_error(page->mapping, err);
787 end_page_writeback(page);
790 EXPORT_SYMBOL_GPL(page_endio);
793 * __lock_page - get a lock on the page, assuming we need to sleep to get it
794 * @page: the page to lock
796 void __lock_page(struct page *page)
798 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
800 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
801 TASK_UNINTERRUPTIBLE);
803 EXPORT_SYMBOL(__lock_page);
805 int __lock_page_killable(struct page *page)
807 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
809 return __wait_on_bit_lock(page_waitqueue(page), &wait,
810 sleep_on_page_killable, TASK_KILLABLE);
812 EXPORT_SYMBOL_GPL(__lock_page_killable);
814 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
817 if (flags & FAULT_FLAG_ALLOW_RETRY) {
819 * CAUTION! In this case, mmap_sem is not released
820 * even though return 0.
822 if (flags & FAULT_FLAG_RETRY_NOWAIT)
825 up_read(&mm->mmap_sem);
826 if (flags & FAULT_FLAG_KILLABLE)
827 wait_on_page_locked_killable(page);
829 wait_on_page_locked(page);
832 if (flags & FAULT_FLAG_KILLABLE) {
835 ret = __lock_page_killable(page);
837 up_read(&mm->mmap_sem);
847 * page_cache_next_hole - find the next hole (not-present entry)
850 * @max_scan: maximum range to search
852 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
853 * lowest indexed hole.
855 * Returns: the index of the hole if found, otherwise returns an index
856 * outside of the set specified (in which case 'return - index >=
857 * max_scan' will be true). In rare cases of index wrap-around, 0 will
860 * page_cache_next_hole may be called under rcu_read_lock. However,
861 * like radix_tree_gang_lookup, this will not atomically search a
862 * snapshot of the tree at a single point in time. For example, if a
863 * hole is created at index 5, then subsequently a hole is created at
864 * index 10, page_cache_next_hole covering both indexes may return 10
865 * if called under rcu_read_lock.
867 pgoff_t page_cache_next_hole(struct address_space *mapping,
868 pgoff_t index, unsigned long max_scan)
872 for (i = 0; i < max_scan; i++) {
875 page = radix_tree_lookup(&mapping->page_tree, index);
876 if (!page || radix_tree_exceptional_entry(page))
885 EXPORT_SYMBOL(page_cache_next_hole);
888 * page_cache_prev_hole - find the prev hole (not-present entry)
891 * @max_scan: maximum range to search
893 * Search backwards in the range [max(index-max_scan+1, 0), index] for
896 * Returns: the index of the hole if found, otherwise returns an index
897 * outside of the set specified (in which case 'index - return >=
898 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
901 * page_cache_prev_hole may be called under rcu_read_lock. However,
902 * like radix_tree_gang_lookup, this will not atomically search a
903 * snapshot of the tree at a single point in time. For example, if a
904 * hole is created at index 10, then subsequently a hole is created at
905 * index 5, page_cache_prev_hole covering both indexes may return 5 if
906 * called under rcu_read_lock.
908 pgoff_t page_cache_prev_hole(struct address_space *mapping,
909 pgoff_t index, unsigned long max_scan)
913 for (i = 0; i < max_scan; i++) {
916 page = radix_tree_lookup(&mapping->page_tree, index);
917 if (!page || radix_tree_exceptional_entry(page))
920 if (index == ULONG_MAX)
926 EXPORT_SYMBOL(page_cache_prev_hole);
929 * find_get_entry - find and get a page cache entry
930 * @mapping: the address_space to search
931 * @offset: the page cache index
933 * Looks up the page cache slot at @mapping & @offset. If there is a
934 * page cache page, it is returned with an increased refcount.
936 * If the slot holds a shadow entry of a previously evicted page, or a
937 * swap entry from shmem/tmpfs, it is returned.
939 * Otherwise, %NULL is returned.
941 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
949 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
951 page = radix_tree_deref_slot(pagep);
954 if (radix_tree_exception(page)) {
955 if (radix_tree_deref_retry(page))
958 * A shadow entry of a recently evicted page,
959 * or a swap entry from shmem/tmpfs. Return
960 * it without attempting to raise page count.
964 if (!page_cache_get_speculative(page))
968 * Has the page moved?
969 * This is part of the lockless pagecache protocol. See
970 * include/linux/pagemap.h for details.
972 if (unlikely(page != *pagep)) {
973 page_cache_release(page);
982 EXPORT_SYMBOL(find_get_entry);
985 * find_get_page - find and get a page reference
986 * @mapping: the address_space to search
987 * @offset: the page index
989 * Looks up the page cache slot at @mapping & @offset. If there is a
990 * page cache page, it is returned with an increased refcount.
992 * Otherwise, %NULL is returned.
994 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
996 struct page *page = find_get_entry(mapping, offset);
998 if (radix_tree_exceptional_entry(page))
1002 EXPORT_SYMBOL(find_get_page);
1005 * find_lock_entry - locate, pin and lock a page cache entry
1006 * @mapping: the address_space to search
1007 * @offset: the page cache index
1009 * Looks up the page cache slot at @mapping & @offset. If there is a
1010 * page cache page, it is returned locked and with an increased
1013 * If the slot holds a shadow entry of a previously evicted page, or a
1014 * swap entry from shmem/tmpfs, it is returned.
1016 * Otherwise, %NULL is returned.
1018 * find_lock_entry() may sleep.
1020 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1025 page = find_get_entry(mapping, offset);
1026 if (page && !radix_tree_exception(page)) {
1028 /* Has the page been truncated? */
1029 if (unlikely(page->mapping != mapping)) {
1031 page_cache_release(page);
1034 VM_BUG_ON_PAGE(page->index != offset, page);
1038 EXPORT_SYMBOL(find_lock_entry);
1041 * find_lock_page - locate, pin and lock a pagecache page
1042 * @mapping: the address_space to search
1043 * @offset: the page index
1045 * Looks up the page cache slot at @mapping & @offset. If there is a
1046 * page cache page, it is returned locked and with an increased
1049 * Otherwise, %NULL is returned.
1051 * find_lock_page() may sleep.
1053 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
1055 struct page *page = find_lock_entry(mapping, offset);
1057 if (radix_tree_exceptional_entry(page))
1061 EXPORT_SYMBOL(find_lock_page);
1064 * find_or_create_page - locate or add a pagecache page
1065 * @mapping: the page's address_space
1066 * @index: the page's index into the mapping
1067 * @gfp_mask: page allocation mode
1069 * Looks up the page cache slot at @mapping & @offset. If there is a
1070 * page cache page, it is returned locked and with an increased
1073 * If the page is not present, a new page is allocated using @gfp_mask
1074 * and added to the page cache and the VM's LRU list. The page is
1075 * returned locked and with an increased refcount.
1077 * On memory exhaustion, %NULL is returned.
1079 * find_or_create_page() may sleep, even if @gfp_flags specifies an
1080 * atomic allocation!
1082 struct page *find_or_create_page(struct address_space *mapping,
1083 pgoff_t index, gfp_t gfp_mask)
1088 page = find_lock_page(mapping, index);
1090 page = __page_cache_alloc(gfp_mask);
1094 * We want a regular kernel memory (not highmem or DMA etc)
1095 * allocation for the radix tree nodes, but we need to honour
1096 * the context-specific requirements the caller has asked for.
1097 * GFP_RECLAIM_MASK collects those requirements.
1099 err = add_to_page_cache_lru(page, mapping, index,
1100 (gfp_mask & GFP_RECLAIM_MASK));
1101 if (unlikely(err)) {
1102 page_cache_release(page);
1110 EXPORT_SYMBOL(find_or_create_page);
1113 * find_get_entries - gang pagecache lookup
1114 * @mapping: The address_space to search
1115 * @start: The starting page cache index
1116 * @nr_entries: The maximum number of entries
1117 * @entries: Where the resulting entries are placed
1118 * @indices: The cache indices corresponding to the entries in @entries
1120 * find_get_entries() will search for and return a group of up to
1121 * @nr_entries entries in the mapping. The entries are placed at
1122 * @entries. find_get_entries() takes a reference against any actual
1125 * The search returns a group of mapping-contiguous page cache entries
1126 * with ascending indexes. There may be holes in the indices due to
1127 * not-present pages.
1129 * Any shadow entries of evicted pages, or swap entries from
1130 * shmem/tmpfs, are included in the returned array.
1132 * find_get_entries() returns the number of pages and shadow entries
1135 unsigned find_get_entries(struct address_space *mapping,
1136 pgoff_t start, unsigned int nr_entries,
1137 struct page **entries, pgoff_t *indices)
1140 unsigned int ret = 0;
1141 struct radix_tree_iter iter;
1148 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1151 page = radix_tree_deref_slot(slot);
1152 if (unlikely(!page))
1154 if (radix_tree_exception(page)) {
1155 if (radix_tree_deref_retry(page))
1158 * A shadow entry of a recently evicted page,
1159 * or a swap entry from shmem/tmpfs. Return
1160 * it without attempting to raise page count.
1164 if (!page_cache_get_speculative(page))
1167 /* Has the page moved? */
1168 if (unlikely(page != *slot)) {
1169 page_cache_release(page);
1173 indices[ret] = iter.index;
1174 entries[ret] = page;
1175 if (++ret == nr_entries)
1183 * find_get_pages - gang pagecache lookup
1184 * @mapping: The address_space to search
1185 * @start: The starting page index
1186 * @nr_pages: The maximum number of pages
1187 * @pages: Where the resulting pages are placed
1189 * find_get_pages() will search for and return a group of up to
1190 * @nr_pages pages in the mapping. The pages are placed at @pages.
1191 * find_get_pages() takes a reference against the returned pages.
1193 * The search returns a group of mapping-contiguous pages with ascending
1194 * indexes. There may be holes in the indices due to not-present pages.
1196 * find_get_pages() returns the number of pages which were found.
1198 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1199 unsigned int nr_pages, struct page **pages)
1201 struct radix_tree_iter iter;
1205 if (unlikely(!nr_pages))
1210 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1213 page = radix_tree_deref_slot(slot);
1214 if (unlikely(!page))
1217 if (radix_tree_exception(page)) {
1218 if (radix_tree_deref_retry(page)) {
1220 * Transient condition which can only trigger
1221 * when entry at index 0 moves out of or back
1222 * to root: none yet gotten, safe to restart.
1224 WARN_ON(iter.index);
1228 * A shadow entry of a recently evicted page,
1229 * or a swap entry from shmem/tmpfs. Skip
1235 if (!page_cache_get_speculative(page))
1238 /* Has the page moved? */
1239 if (unlikely(page != *slot)) {
1240 page_cache_release(page);
1245 if (++ret == nr_pages)
1254 * find_get_pages_contig - gang contiguous pagecache lookup
1255 * @mapping: The address_space to search
1256 * @index: The starting page index
1257 * @nr_pages: The maximum number of pages
1258 * @pages: Where the resulting pages are placed
1260 * find_get_pages_contig() works exactly like find_get_pages(), except
1261 * that the returned number of pages are guaranteed to be contiguous.
1263 * find_get_pages_contig() returns the number of pages which were found.
1265 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1266 unsigned int nr_pages, struct page **pages)
1268 struct radix_tree_iter iter;
1270 unsigned int ret = 0;
1272 if (unlikely(!nr_pages))
1277 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1280 page = radix_tree_deref_slot(slot);
1281 /* The hole, there no reason to continue */
1282 if (unlikely(!page))
1285 if (radix_tree_exception(page)) {
1286 if (radix_tree_deref_retry(page)) {
1288 * Transient condition which can only trigger
1289 * when entry at index 0 moves out of or back
1290 * to root: none yet gotten, safe to restart.
1295 * A shadow entry of a recently evicted page,
1296 * or a swap entry from shmem/tmpfs. Stop
1297 * looking for contiguous pages.
1302 if (!page_cache_get_speculative(page))
1305 /* Has the page moved? */
1306 if (unlikely(page != *slot)) {
1307 page_cache_release(page);
1312 * must check mapping and index after taking the ref.
1313 * otherwise we can get both false positives and false
1314 * negatives, which is just confusing to the caller.
1316 if (page->mapping == NULL || page->index != iter.index) {
1317 page_cache_release(page);
1322 if (++ret == nr_pages)
1328 EXPORT_SYMBOL(find_get_pages_contig);
1331 * find_get_pages_tag - find and return pages that match @tag
1332 * @mapping: the address_space to search
1333 * @index: the starting page index
1334 * @tag: the tag index
1335 * @nr_pages: the maximum number of pages
1336 * @pages: where the resulting pages are placed
1338 * Like find_get_pages, except we only return pages which are tagged with
1339 * @tag. We update @index to index the next page for the traversal.
1341 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1342 int tag, unsigned int nr_pages, struct page **pages)
1344 struct radix_tree_iter iter;
1348 if (unlikely(!nr_pages))
1353 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1354 &iter, *index, tag) {
1357 page = radix_tree_deref_slot(slot);
1358 if (unlikely(!page))
1361 if (radix_tree_exception(page)) {
1362 if (radix_tree_deref_retry(page)) {
1364 * Transient condition which can only trigger
1365 * when entry at index 0 moves out of or back
1366 * to root: none yet gotten, safe to restart.
1371 * A shadow entry of a recently evicted page.
1373 * Those entries should never be tagged, but
1374 * this tree walk is lockless and the tags are
1375 * looked up in bulk, one radix tree node at a
1376 * time, so there is a sizable window for page
1377 * reclaim to evict a page we saw tagged.
1384 if (!page_cache_get_speculative(page))
1387 /* Has the page moved? */
1388 if (unlikely(page != *slot)) {
1389 page_cache_release(page);
1394 if (++ret == nr_pages)
1401 *index = pages[ret - 1]->index + 1;
1405 EXPORT_SYMBOL(find_get_pages_tag);
1408 * grab_cache_page_nowait - returns locked page at given index in given cache
1409 * @mapping: target address_space
1410 * @index: the page index
1412 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1413 * This is intended for speculative data generators, where the data can
1414 * be regenerated if the page couldn't be grabbed. This routine should
1415 * be safe to call while holding the lock for another page.
1417 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1418 * and deadlock against the caller's locked page.
1421 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1423 struct page *page = find_get_page(mapping, index);
1426 if (trylock_page(page))
1428 page_cache_release(page);
1431 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1432 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1433 page_cache_release(page);
1438 EXPORT_SYMBOL(grab_cache_page_nowait);
1441 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1442 * a _large_ part of the i/o request. Imagine the worst scenario:
1444 * ---R__________________________________________B__________
1445 * ^ reading here ^ bad block(assume 4k)
1447 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1448 * => failing the whole request => read(R) => read(R+1) =>
1449 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1450 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1451 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1453 * It is going insane. Fix it by quickly scaling down the readahead size.
1455 static void shrink_readahead_size_eio(struct file *filp,
1456 struct file_ra_state *ra)
1462 * do_generic_file_read - generic file read routine
1463 * @filp: the file to read
1464 * @ppos: current file position
1465 * @iter: data destination
1466 * @written: already copied
1468 * This is a generic file read routine, and uses the
1469 * mapping->a_ops->readpage() function for the actual low-level stuff.
1471 * This is really ugly. But the goto's actually try to clarify some
1472 * of the logic when it comes to error handling etc.
1474 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1475 struct iov_iter *iter, ssize_t written)
1477 struct address_space *mapping = filp->f_mapping;
1478 struct inode *inode = mapping->host;
1479 struct file_ra_state *ra = &filp->f_ra;
1483 unsigned long offset; /* offset into pagecache page */
1484 unsigned int prev_offset;
1487 index = *ppos >> PAGE_CACHE_SHIFT;
1488 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1489 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1490 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1491 offset = *ppos & ~PAGE_CACHE_MASK;
1497 unsigned long nr, ret;
1501 page = find_get_page(mapping, index);
1503 page_cache_sync_readahead(mapping,
1505 index, last_index - index);
1506 page = find_get_page(mapping, index);
1507 if (unlikely(page == NULL))
1508 goto no_cached_page;
1510 if (PageReadahead(page)) {
1511 page_cache_async_readahead(mapping,
1513 index, last_index - index);
1515 if (!PageUptodate(page)) {
1516 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1517 !mapping->a_ops->is_partially_uptodate)
1518 goto page_not_up_to_date;
1519 if (!trylock_page(page))
1520 goto page_not_up_to_date;
1521 /* Did it get truncated before we got the lock? */
1523 goto page_not_up_to_date_locked;
1524 if (!mapping->a_ops->is_partially_uptodate(page,
1525 offset, iter->count))
1526 goto page_not_up_to_date_locked;
1531 * i_size must be checked after we know the page is Uptodate.
1533 * Checking i_size after the check allows us to calculate
1534 * the correct value for "nr", which means the zero-filled
1535 * part of the page is not copied back to userspace (unless
1536 * another truncate extends the file - this is desired though).
1539 isize = i_size_read(inode);
1540 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1541 if (unlikely(!isize || index > end_index)) {
1542 page_cache_release(page);
1546 /* nr is the maximum number of bytes to copy from this page */
1547 nr = PAGE_CACHE_SIZE;
1548 if (index == end_index) {
1549 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1551 page_cache_release(page);
1557 /* If users can be writing to this page using arbitrary
1558 * virtual addresses, take care about potential aliasing
1559 * before reading the page on the kernel side.
1561 if (mapping_writably_mapped(mapping))
1562 flush_dcache_page(page);
1565 * When a sequential read accesses a page several times,
1566 * only mark it as accessed the first time.
1568 if (prev_index != index || offset != prev_offset)
1569 mark_page_accessed(page);
1573 * Ok, we have the page, and it's up-to-date, so
1574 * now we can copy it to user space...
1577 ret = copy_page_to_iter(page, offset, nr, iter);
1579 index += offset >> PAGE_CACHE_SHIFT;
1580 offset &= ~PAGE_CACHE_MASK;
1581 prev_offset = offset;
1583 page_cache_release(page);
1585 if (!iov_iter_count(iter))
1593 page_not_up_to_date:
1594 /* Get exclusive access to the page ... */
1595 error = lock_page_killable(page);
1596 if (unlikely(error))
1597 goto readpage_error;
1599 page_not_up_to_date_locked:
1600 /* Did it get truncated before we got the lock? */
1601 if (!page->mapping) {
1603 page_cache_release(page);
1607 /* Did somebody else fill it already? */
1608 if (PageUptodate(page)) {
1615 * A previous I/O error may have been due to temporary
1616 * failures, eg. multipath errors.
1617 * PG_error will be set again if readpage fails.
1619 ClearPageError(page);
1620 /* Start the actual read. The read will unlock the page. */
1621 error = mapping->a_ops->readpage(filp, page);
1623 if (unlikely(error)) {
1624 if (error == AOP_TRUNCATED_PAGE) {
1625 page_cache_release(page);
1629 goto readpage_error;
1632 if (!PageUptodate(page)) {
1633 error = lock_page_killable(page);
1634 if (unlikely(error))
1635 goto readpage_error;
1636 if (!PageUptodate(page)) {
1637 if (page->mapping == NULL) {
1639 * invalidate_mapping_pages got it
1642 page_cache_release(page);
1646 shrink_readahead_size_eio(filp, ra);
1648 goto readpage_error;
1656 /* UHHUH! A synchronous read error occurred. Report it */
1657 page_cache_release(page);
1662 * Ok, it wasn't cached, so we need to create a new
1665 page = page_cache_alloc_cold(mapping);
1670 error = add_to_page_cache_lru(page, mapping,
1673 page_cache_release(page);
1674 if (error == -EEXIST) {
1684 ra->prev_pos = prev_index;
1685 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1686 ra->prev_pos |= prev_offset;
1688 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1689 file_accessed(filp);
1690 return written ? written : error;
1694 * Performs necessary checks before doing a write
1695 * @iov: io vector request
1696 * @nr_segs: number of segments in the iovec
1697 * @count: number of bytes to write
1698 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1700 * Adjust number of segments and amount of bytes to write (nr_segs should be
1701 * properly initialized first). Returns appropriate error code that caller
1702 * should return or zero in case that write should be allowed.
1704 int generic_segment_checks(const struct iovec *iov,
1705 unsigned long *nr_segs, size_t *count, int access_flags)
1709 for (seg = 0; seg < *nr_segs; seg++) {
1710 const struct iovec *iv = &iov[seg];
1713 * If any segment has a negative length, or the cumulative
1714 * length ever wraps negative then return -EINVAL.
1717 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1719 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1724 cnt -= iv->iov_len; /* This segment is no good */
1730 EXPORT_SYMBOL(generic_segment_checks);
1733 * generic_file_aio_read - generic filesystem read routine
1734 * @iocb: kernel I/O control block
1735 * @iov: io vector request
1736 * @nr_segs: number of segments in the iovec
1737 * @pos: current file position
1739 * This is the "read()" routine for all filesystems
1740 * that can use the page cache directly.
1743 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1744 unsigned long nr_segs, loff_t pos)
1746 struct file *filp = iocb->ki_filp;
1749 loff_t *ppos = &iocb->ki_pos;
1753 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1756 iov_iter_init(&i, iov, nr_segs, count, 0);
1758 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1759 if (filp->f_flags & O_DIRECT) {
1761 struct address_space *mapping;
1762 struct inode *inode;
1764 mapping = filp->f_mapping;
1765 inode = mapping->host;
1767 goto out; /* skip atime */
1768 size = i_size_read(inode);
1769 retval = filemap_write_and_wait_range(mapping, pos,
1770 pos + iov_length(iov, nr_segs) - 1);
1772 retval = mapping->a_ops->direct_IO(READ, iocb,
1776 *ppos = pos + retval;
1779 * If we did a short DIO read we need to skip the
1780 * section of the iov that we've already read data into.
1782 iov_iter_advance(&i, retval);
1786 * Btrfs can have a short DIO read if we encounter
1787 * compressed extents, so if there was an error, or if
1788 * we've already read everything we wanted to, or if
1789 * there was a short read because we hit EOF, go ahead
1790 * and return. Otherwise fallthrough to buffered io for
1791 * the rest of the read.
1793 if (retval < 0 || !count || *ppos >= size) {
1794 file_accessed(filp);
1799 retval = do_generic_file_read(filp, ppos, &i, retval);
1803 EXPORT_SYMBOL(generic_file_aio_read);
1807 * page_cache_read - adds requested page to the page cache if not already there
1808 * @file: file to read
1809 * @offset: page index
1811 * This adds the requested page to the page cache if it isn't already there,
1812 * and schedules an I/O to read in its contents from disk.
1814 static int page_cache_read(struct file *file, pgoff_t offset)
1816 struct address_space *mapping = file->f_mapping;
1821 page = page_cache_alloc_cold(mapping);
1825 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1827 ret = mapping->a_ops->readpage(file, page);
1828 else if (ret == -EEXIST)
1829 ret = 0; /* losing race to add is OK */
1831 page_cache_release(page);
1833 } while (ret == AOP_TRUNCATED_PAGE);
1838 #define MMAP_LOTSAMISS (100)
1841 * Synchronous readahead happens when we don't even find
1842 * a page in the page cache at all.
1844 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1845 struct file_ra_state *ra,
1849 unsigned long ra_pages;
1850 struct address_space *mapping = file->f_mapping;
1852 /* If we don't want any read-ahead, don't bother */
1853 if (vma->vm_flags & VM_RAND_READ)
1858 if (vma->vm_flags & VM_SEQ_READ) {
1859 page_cache_sync_readahead(mapping, ra, file, offset,
1864 /* Avoid banging the cache line if not needed */
1865 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1869 * Do we miss much more than hit in this file? If so,
1870 * stop bothering with read-ahead. It will only hurt.
1872 if (ra->mmap_miss > MMAP_LOTSAMISS)
1878 ra_pages = max_sane_readahead(ra->ra_pages);
1879 ra->start = max_t(long, 0, offset - ra_pages / 2);
1880 ra->size = ra_pages;
1881 ra->async_size = ra_pages / 4;
1882 ra_submit(ra, mapping, file);
1886 * Asynchronous readahead happens when we find the page and PG_readahead,
1887 * so we want to possibly extend the readahead further..
1889 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1890 struct file_ra_state *ra,
1895 struct address_space *mapping = file->f_mapping;
1897 /* If we don't want any read-ahead, don't bother */
1898 if (vma->vm_flags & VM_RAND_READ)
1900 if (ra->mmap_miss > 0)
1902 if (PageReadahead(page))
1903 page_cache_async_readahead(mapping, ra, file,
1904 page, offset, ra->ra_pages);
1908 * filemap_fault - read in file data for page fault handling
1909 * @vma: vma in which the fault was taken
1910 * @vmf: struct vm_fault containing details of the fault
1912 * filemap_fault() is invoked via the vma operations vector for a
1913 * mapped memory region to read in file data during a page fault.
1915 * The goto's are kind of ugly, but this streamlines the normal case of having
1916 * it in the page cache, and handles the special cases reasonably without
1917 * having a lot of duplicated code.
1919 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1922 struct file *file = vma->vm_file;
1923 struct address_space *mapping = file->f_mapping;
1924 struct file_ra_state *ra = &file->f_ra;
1925 struct inode *inode = mapping->host;
1926 pgoff_t offset = vmf->pgoff;
1931 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1932 if (offset >= size >> PAGE_CACHE_SHIFT)
1933 return VM_FAULT_SIGBUS;
1936 * Do we have something in the page cache already?
1938 page = find_get_page(mapping, offset);
1939 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1941 * We found the page, so try async readahead before
1942 * waiting for the lock.
1944 do_async_mmap_readahead(vma, ra, file, page, offset);
1946 /* No page in the page cache at all */
1947 do_sync_mmap_readahead(vma, ra, file, offset);
1948 count_vm_event(PGMAJFAULT);
1949 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1950 ret = VM_FAULT_MAJOR;
1952 page = find_get_page(mapping, offset);
1954 goto no_cached_page;
1957 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1958 page_cache_release(page);
1959 return ret | VM_FAULT_RETRY;
1962 /* Did it get truncated? */
1963 if (unlikely(page->mapping != mapping)) {
1968 VM_BUG_ON_PAGE(page->index != offset, page);
1971 * We have a locked page in the page cache, now we need to check
1972 * that it's up-to-date. If not, it is going to be due to an error.
1974 if (unlikely(!PageUptodate(page)))
1975 goto page_not_uptodate;
1978 * Found the page and have a reference on it.
1979 * We must recheck i_size under page lock.
1981 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1982 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1984 page_cache_release(page);
1985 return VM_FAULT_SIGBUS;
1989 return ret | VM_FAULT_LOCKED;
1993 * We're only likely to ever get here if MADV_RANDOM is in
1996 error = page_cache_read(file, offset);
1999 * The page we want has now been added to the page cache.
2000 * In the unlikely event that someone removed it in the
2001 * meantime, we'll just come back here and read it again.
2007 * An error return from page_cache_read can result if the
2008 * system is low on memory, or a problem occurs while trying
2011 if (error == -ENOMEM)
2012 return VM_FAULT_OOM;
2013 return VM_FAULT_SIGBUS;
2017 * Umm, take care of errors if the page isn't up-to-date.
2018 * Try to re-read it _once_. We do this synchronously,
2019 * because there really aren't any performance issues here
2020 * and we need to check for errors.
2022 ClearPageError(page);
2023 error = mapping->a_ops->readpage(file, page);
2025 wait_on_page_locked(page);
2026 if (!PageUptodate(page))
2029 page_cache_release(page);
2031 if (!error || error == AOP_TRUNCATED_PAGE)
2034 /* Things didn't work out. Return zero to tell the mm layer so. */
2035 shrink_readahead_size_eio(file, ra);
2036 return VM_FAULT_SIGBUS;
2038 EXPORT_SYMBOL(filemap_fault);
2040 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2042 struct radix_tree_iter iter;
2044 struct file *file = vma->vm_file;
2045 struct address_space *mapping = file->f_mapping;
2048 unsigned long address = (unsigned long) vmf->virtual_address;
2053 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2054 if (iter.index > vmf->max_pgoff)
2057 page = radix_tree_deref_slot(slot);
2058 if (unlikely(!page))
2060 if (radix_tree_exception(page)) {
2061 if (radix_tree_deref_retry(page))
2067 if (!page_cache_get_speculative(page))
2070 /* Has the page moved? */
2071 if (unlikely(page != *slot)) {
2072 page_cache_release(page);
2076 if (!PageUptodate(page) ||
2077 PageReadahead(page) ||
2080 if (!trylock_page(page))
2083 if (page->mapping != mapping || !PageUptodate(page))
2086 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2087 if (page->index >= size >> PAGE_CACHE_SHIFT)
2090 pte = vmf->pte + page->index - vmf->pgoff;
2091 if (!pte_none(*pte))
2094 if (file->f_ra.mmap_miss > 0)
2095 file->f_ra.mmap_miss--;
2096 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2097 do_set_pte(vma, addr, page, pte, false, false);
2103 page_cache_release(page);
2105 if (iter.index == vmf->max_pgoff)
2110 EXPORT_SYMBOL(filemap_map_pages);
2112 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2114 struct page *page = vmf->page;
2115 struct inode *inode = file_inode(vma->vm_file);
2116 int ret = VM_FAULT_LOCKED;
2118 sb_start_pagefault(inode->i_sb);
2119 file_update_time(vma->vm_file);
2121 if (page->mapping != inode->i_mapping) {
2123 ret = VM_FAULT_NOPAGE;
2127 * We mark the page dirty already here so that when freeze is in
2128 * progress, we are guaranteed that writeback during freezing will
2129 * see the dirty page and writeprotect it again.
2131 set_page_dirty(page);
2132 wait_for_stable_page(page);
2134 sb_end_pagefault(inode->i_sb);
2137 EXPORT_SYMBOL(filemap_page_mkwrite);
2139 const struct vm_operations_struct generic_file_vm_ops = {
2140 .fault = filemap_fault,
2141 .map_pages = filemap_map_pages,
2142 .page_mkwrite = filemap_page_mkwrite,
2143 .remap_pages = generic_file_remap_pages,
2146 /* This is used for a general mmap of a disk file */
2148 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2150 struct address_space *mapping = file->f_mapping;
2152 if (!mapping->a_ops->readpage)
2154 file_accessed(file);
2155 vma->vm_ops = &generic_file_vm_ops;
2160 * This is for filesystems which do not implement ->writepage.
2162 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2164 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2166 return generic_file_mmap(file, vma);
2169 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2173 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2177 #endif /* CONFIG_MMU */
2179 EXPORT_SYMBOL(generic_file_mmap);
2180 EXPORT_SYMBOL(generic_file_readonly_mmap);
2182 static struct page *wait_on_page_read(struct page *page)
2184 if (!IS_ERR(page)) {
2185 wait_on_page_locked(page);
2186 if (!PageUptodate(page)) {
2187 page_cache_release(page);
2188 page = ERR_PTR(-EIO);
2194 static struct page *__read_cache_page(struct address_space *mapping,
2196 int (*filler)(void *, struct page *),
2203 page = find_get_page(mapping, index);
2205 page = __page_cache_alloc(gfp | __GFP_COLD);
2207 return ERR_PTR(-ENOMEM);
2208 err = add_to_page_cache_lru(page, mapping, index, gfp);
2209 if (unlikely(err)) {
2210 page_cache_release(page);
2213 /* Presumably ENOMEM for radix tree node */
2214 return ERR_PTR(err);
2216 err = filler(data, page);
2218 page_cache_release(page);
2219 page = ERR_PTR(err);
2221 page = wait_on_page_read(page);
2227 static struct page *do_read_cache_page(struct address_space *mapping,
2229 int (*filler)(void *, struct page *),
2238 page = __read_cache_page(mapping, index, filler, data, gfp);
2241 if (PageUptodate(page))
2245 if (!page->mapping) {
2247 page_cache_release(page);
2250 if (PageUptodate(page)) {
2254 err = filler(data, page);
2256 page_cache_release(page);
2257 return ERR_PTR(err);
2259 page = wait_on_page_read(page);
2264 mark_page_accessed(page);
2269 * read_cache_page - read into page cache, fill it if needed
2270 * @mapping: the page's address_space
2271 * @index: the page index
2272 * @filler: function to perform the read
2273 * @data: first arg to filler(data, page) function, often left as NULL
2275 * Read into the page cache. If a page already exists, and PageUptodate() is
2276 * not set, try to fill the page and wait for it to become unlocked.
2278 * If the page does not get brought uptodate, return -EIO.
2280 struct page *read_cache_page(struct address_space *mapping,
2282 int (*filler)(void *, struct page *),
2285 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2287 EXPORT_SYMBOL(read_cache_page);
2290 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2291 * @mapping: the page's address_space
2292 * @index: the page index
2293 * @gfp: the page allocator flags to use if allocating
2295 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2296 * any new page allocations done using the specified allocation flags.
2298 * If the page does not get brought uptodate, return -EIO.
2300 struct page *read_cache_page_gfp(struct address_space *mapping,
2304 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2306 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2308 EXPORT_SYMBOL(read_cache_page_gfp);
2311 * Performs necessary checks before doing a write
2313 * Can adjust writing position or amount of bytes to write.
2314 * Returns appropriate error code that caller should return or
2315 * zero in case that write should be allowed.
2317 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2319 struct inode *inode = file->f_mapping->host;
2320 unsigned long limit = rlimit(RLIMIT_FSIZE);
2322 if (unlikely(*pos < 0))
2326 /* FIXME: this is for backwards compatibility with 2.4 */
2327 if (file->f_flags & O_APPEND)
2328 *pos = i_size_read(inode);
2330 if (limit != RLIM_INFINITY) {
2331 if (*pos >= limit) {
2332 send_sig(SIGXFSZ, current, 0);
2335 if (*count > limit - (typeof(limit))*pos) {
2336 *count = limit - (typeof(limit))*pos;
2344 if (unlikely(*pos + *count > MAX_NON_LFS &&
2345 !(file->f_flags & O_LARGEFILE))) {
2346 if (*pos >= MAX_NON_LFS) {
2349 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2350 *count = MAX_NON_LFS - (unsigned long)*pos;
2355 * Are we about to exceed the fs block limit ?
2357 * If we have written data it becomes a short write. If we have
2358 * exceeded without writing data we send a signal and return EFBIG.
2359 * Linus frestrict idea will clean these up nicely..
2361 if (likely(!isblk)) {
2362 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2363 if (*count || *pos > inode->i_sb->s_maxbytes) {
2366 /* zero-length writes at ->s_maxbytes are OK */
2369 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2370 *count = inode->i_sb->s_maxbytes - *pos;
2374 if (bdev_read_only(I_BDEV(inode)))
2376 isize = i_size_read(inode);
2377 if (*pos >= isize) {
2378 if (*count || *pos > isize)
2382 if (*pos + *count > isize)
2383 *count = isize - *pos;
2390 EXPORT_SYMBOL(generic_write_checks);
2392 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2393 loff_t pos, unsigned len, unsigned flags,
2394 struct page **pagep, void **fsdata)
2396 const struct address_space_operations *aops = mapping->a_ops;
2398 return aops->write_begin(file, mapping, pos, len, flags,
2401 EXPORT_SYMBOL(pagecache_write_begin);
2403 int pagecache_write_end(struct file *file, struct address_space *mapping,
2404 loff_t pos, unsigned len, unsigned copied,
2405 struct page *page, void *fsdata)
2407 const struct address_space_operations *aops = mapping->a_ops;
2409 mark_page_accessed(page);
2410 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2412 EXPORT_SYMBOL(pagecache_write_end);
2415 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2416 unsigned long *nr_segs, loff_t pos,
2417 size_t count, size_t ocount)
2419 struct file *file = iocb->ki_filp;
2420 struct address_space *mapping = file->f_mapping;
2421 struct inode *inode = mapping->host;
2426 if (count != ocount)
2427 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2429 write_len = iov_length(iov, *nr_segs);
2430 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2432 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2437 * After a write we want buffered reads to be sure to go to disk to get
2438 * the new data. We invalidate clean cached page from the region we're
2439 * about to write. We do this *before* the write so that we can return
2440 * without clobbering -EIOCBQUEUED from ->direct_IO().
2442 if (mapping->nrpages) {
2443 written = invalidate_inode_pages2_range(mapping,
2444 pos >> PAGE_CACHE_SHIFT, end);
2446 * If a page can not be invalidated, return 0 to fall back
2447 * to buffered write.
2450 if (written == -EBUSY)
2456 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2459 * Finally, try again to invalidate clean pages which might have been
2460 * cached by non-direct readahead, or faulted in by get_user_pages()
2461 * if the source of the write was an mmap'ed region of the file
2462 * we're writing. Either one is a pretty crazy thing to do,
2463 * so we don't support it 100%. If this invalidation
2464 * fails, tough, the write still worked...
2466 if (mapping->nrpages) {
2467 invalidate_inode_pages2_range(mapping,
2468 pos >> PAGE_CACHE_SHIFT, end);
2473 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2474 i_size_write(inode, pos);
2475 mark_inode_dirty(inode);
2482 EXPORT_SYMBOL(generic_file_direct_write);
2485 * Find or create a page at the given pagecache position. Return the locked
2486 * page. This function is specifically for buffered writes.
2488 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2489 pgoff_t index, unsigned flags)
2494 gfp_t gfp_notmask = 0;
2496 gfp_mask = mapping_gfp_mask(mapping);
2497 if (mapping_cap_account_dirty(mapping))
2498 gfp_mask |= __GFP_WRITE;
2499 if (flags & AOP_FLAG_NOFS)
2500 gfp_notmask = __GFP_FS;
2502 page = find_lock_page(mapping, index);
2506 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2509 status = add_to_page_cache_lru(page, mapping, index,
2510 GFP_KERNEL & ~gfp_notmask);
2511 if (unlikely(status)) {
2512 page_cache_release(page);
2513 if (status == -EEXIST)
2518 wait_for_stable_page(page);
2521 EXPORT_SYMBOL(grab_cache_page_write_begin);
2523 ssize_t generic_perform_write(struct file *file,
2524 struct iov_iter *i, loff_t pos)
2526 struct address_space *mapping = file->f_mapping;
2527 const struct address_space_operations *a_ops = mapping->a_ops;
2529 ssize_t written = 0;
2530 unsigned int flags = 0;
2533 * Copies from kernel address space cannot fail (NFSD is a big user).
2535 if (segment_eq(get_fs(), KERNEL_DS))
2536 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2540 unsigned long offset; /* Offset into pagecache page */
2541 unsigned long bytes; /* Bytes to write to page */
2542 size_t copied; /* Bytes copied from user */
2545 offset = (pos & (PAGE_CACHE_SIZE - 1));
2546 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2551 * Bring in the user page that we will copy from _first_.
2552 * Otherwise there's a nasty deadlock on copying from the
2553 * same page as we're writing to, without it being marked
2556 * Not only is this an optimisation, but it is also required
2557 * to check that the address is actually valid, when atomic
2558 * usercopies are used, below.
2560 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2565 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2567 if (unlikely(status))
2570 if (mapping_writably_mapped(mapping))
2571 flush_dcache_page(page);
2573 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2574 flush_dcache_page(page);
2576 mark_page_accessed(page);
2577 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2579 if (unlikely(status < 0))
2585 iov_iter_advance(i, copied);
2586 if (unlikely(copied == 0)) {
2588 * If we were unable to copy any data at all, we must
2589 * fall back to a single segment length write.
2591 * If we didn't fallback here, we could livelock
2592 * because not all segments in the iov can be copied at
2593 * once without a pagefault.
2595 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2596 iov_iter_single_seg_count(i));
2602 balance_dirty_pages_ratelimited(mapping);
2603 if (fatal_signal_pending(current)) {
2607 } while (iov_iter_count(i));
2609 return written ? written : status;
2611 EXPORT_SYMBOL(generic_perform_write);
2614 * __generic_file_aio_write - write data to a file
2615 * @iocb: IO state structure (file, offset, etc.)
2616 * @iov: vector with data to write
2617 * @nr_segs: number of segments in the vector
2619 * This function does all the work needed for actually writing data to a
2620 * file. It does all basic checks, removes SUID from the file, updates
2621 * modification times and calls proper subroutines depending on whether we
2622 * do direct IO or a standard buffered write.
2624 * It expects i_mutex to be grabbed unless we work on a block device or similar
2625 * object which does not need locking at all.
2627 * This function does *not* take care of syncing data in case of O_SYNC write.
2628 * A caller has to handle it. This is mainly due to the fact that we want to
2629 * avoid syncing under i_mutex.
2631 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2632 unsigned long nr_segs)
2634 struct file *file = iocb->ki_filp;
2635 struct address_space * mapping = file->f_mapping;
2636 size_t ocount; /* original count */
2637 size_t count; /* after file limit checks */
2638 struct inode *inode = mapping->host;
2639 loff_t pos = iocb->ki_pos;
2640 ssize_t written = 0;
2643 struct iov_iter from;
2646 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2652 /* We can write back this queue in page reclaim */
2653 current->backing_dev_info = mapping->backing_dev_info;
2654 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2661 err = file_remove_suid(file);
2665 err = file_update_time(file);
2669 iov_iter_init(&from, iov, nr_segs, count, 0);
2671 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2672 if (unlikely(file->f_flags & O_DIRECT)) {
2675 written = generic_file_direct_write(iocb, iov, &from.nr_segs, pos,
2677 if (written < 0 || written == count)
2679 iov_iter_advance(&from, written);
2682 * direct-io write to a hole: fall through to buffered I/O
2683 * for completing the rest of the request.
2688 status = generic_perform_write(file, &from, pos);
2690 * If generic_perform_write() returned a synchronous error
2691 * then we want to return the number of bytes which were
2692 * direct-written, or the error code if that was zero. Note
2693 * that this differs from normal direct-io semantics, which
2694 * will return -EFOO even if some bytes were written.
2696 if (unlikely(status < 0) && !written) {
2700 iocb->ki_pos = pos + status;
2702 * We need to ensure that the page cache pages are written to
2703 * disk and invalidated to preserve the expected O_DIRECT
2706 endbyte = pos + status - 1;
2707 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2710 invalidate_mapping_pages(mapping,
2711 pos >> PAGE_CACHE_SHIFT,
2712 endbyte >> PAGE_CACHE_SHIFT);
2715 * We don't know how much we wrote, so just return
2716 * the number of bytes which were direct-written
2720 written = generic_perform_write(file, &from, pos);
2721 if (likely(written >= 0))
2722 iocb->ki_pos = pos + written;
2725 current->backing_dev_info = NULL;
2726 return written ? written : err;
2728 EXPORT_SYMBOL(__generic_file_aio_write);
2731 * generic_file_aio_write - write data to a file
2732 * @iocb: IO state structure
2733 * @iov: vector with data to write
2734 * @nr_segs: number of segments in the vector
2735 * @pos: position in file where to write
2737 * This is a wrapper around __generic_file_aio_write() to be used by most
2738 * filesystems. It takes care of syncing the file in case of O_SYNC file
2739 * and acquires i_mutex as needed.
2741 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2742 unsigned long nr_segs, loff_t pos)
2744 struct file *file = iocb->ki_filp;
2745 struct inode *inode = file->f_mapping->host;
2748 BUG_ON(iocb->ki_pos != pos);
2750 mutex_lock(&inode->i_mutex);
2751 ret = __generic_file_aio_write(iocb, iov, nr_segs);
2752 mutex_unlock(&inode->i_mutex);
2757 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2763 EXPORT_SYMBOL(generic_file_aio_write);
2766 * try_to_release_page() - release old fs-specific metadata on a page
2768 * @page: the page which the kernel is trying to free
2769 * @gfp_mask: memory allocation flags (and I/O mode)
2771 * The address_space is to try to release any data against the page
2772 * (presumably at page->private). If the release was successful, return `1'.
2773 * Otherwise return zero.
2775 * This may also be called if PG_fscache is set on a page, indicating that the
2776 * page is known to the local caching routines.
2778 * The @gfp_mask argument specifies whether I/O may be performed to release
2779 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2782 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2784 struct address_space * const mapping = page->mapping;
2786 BUG_ON(!PageLocked(page));
2787 if (PageWriteback(page))
2790 if (mapping && mapping->a_ops->releasepage)
2791 return mapping->a_ops->releasepage(page, gfp_mask);
2792 return try_to_free_buffers(page);
2795 EXPORT_SYMBOL(try_to_release_page);