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/capability.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/gfp.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/backing-dev.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/cpuset.h>
32 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
33 #include <linux/hugetlb.h>
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_rwsem (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_rwsem (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 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
112 static void page_cache_tree_delete(struct address_space *mapping,
113 struct page *page, void *shadow)
115 struct radix_tree_node *node;
121 VM_BUG_ON(!PageLocked(page));
123 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
126 mapping->nrshadows++;
128 * Make sure the nrshadows update is committed before
129 * the nrpages update so that final truncate racing
130 * with reclaim does not see both counters 0 at the
131 * same time and miss a shadow entry.
138 /* Clear direct pointer tags in root node */
139 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
140 radix_tree_replace_slot(slot, shadow);
144 /* Clear tree tags for the removed page */
146 offset = index & RADIX_TREE_MAP_MASK;
147 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
148 if (test_bit(offset, node->tags[tag]))
149 radix_tree_tag_clear(&mapping->page_tree, index, tag);
152 /* Delete page, swap shadow entry */
153 radix_tree_replace_slot(slot, shadow);
154 workingset_node_pages_dec(node);
156 workingset_node_shadows_inc(node);
158 if (__radix_tree_delete_node(&mapping->page_tree, node))
162 * Track node that only contains shadow entries.
164 * Avoid acquiring the list_lru lock if already tracked. The
165 * list_empty() test is safe as node->private_list is
166 * protected by mapping->tree_lock.
168 if (!workingset_node_pages(node) &&
169 list_empty(&node->private_list)) {
170 node->private_data = mapping;
171 local_lock(workingset_shadow_lock);
172 list_lru_add(&__workingset_shadow_nodes, &node->private_list);
173 local_unlock(workingset_shadow_lock);
178 * Delete a page from the page cache and free it. Caller has to make
179 * sure the page is locked and that nobody else uses it - or that usage
180 * is safe. The caller must hold the mapping's tree_lock and
181 * mem_cgroup_begin_page_stat().
183 void __delete_from_page_cache(struct page *page, void *shadow,
184 struct mem_cgroup *memcg)
186 struct address_space *mapping = page->mapping;
188 trace_mm_filemap_delete_from_page_cache(page);
190 * if we're uptodate, flush out into the cleancache, otherwise
191 * invalidate any existing cleancache entries. We can't leave
192 * stale data around in the cleancache once our page is gone
194 if (PageUptodate(page) && PageMappedToDisk(page))
195 cleancache_put_page(page);
197 cleancache_invalidate_page(mapping, page);
199 page_cache_tree_delete(mapping, page, shadow);
201 page->mapping = NULL;
202 /* Leave page->index set: truncation lookup relies upon it */
204 /* hugetlb pages do not participate in page cache accounting. */
206 __dec_zone_page_state(page, NR_FILE_PAGES);
207 if (PageSwapBacked(page))
208 __dec_zone_page_state(page, NR_SHMEM);
209 BUG_ON(page_mapped(page));
212 * At this point page must be either written or cleaned by truncate.
213 * Dirty page here signals a bug and loss of unwritten data.
215 * This fixes dirty accounting after removing the page entirely but
216 * leaves PageDirty set: it has no effect for truncated page and
217 * anyway will be cleared before returning page into buddy allocator.
219 if (WARN_ON_ONCE(PageDirty(page)))
220 account_page_cleaned(page, mapping, memcg,
221 inode_to_wb(mapping->host));
225 * delete_from_page_cache - delete page from page cache
226 * @page: the page which the kernel is trying to remove from page cache
228 * This must be called only on pages that have been verified to be in the page
229 * cache and locked. It will never put the page into the free list, the caller
230 * has a reference on the page.
232 void delete_from_page_cache(struct page *page)
234 struct address_space *mapping = page->mapping;
235 struct mem_cgroup *memcg;
238 void (*freepage)(struct page *);
240 BUG_ON(!PageLocked(page));
242 freepage = mapping->a_ops->freepage;
244 memcg = mem_cgroup_begin_page_stat(page);
245 spin_lock_irqsave(&mapping->tree_lock, flags);
246 __delete_from_page_cache(page, NULL, memcg);
247 spin_unlock_irqrestore(&mapping->tree_lock, flags);
248 mem_cgroup_end_page_stat(memcg);
252 page_cache_release(page);
254 EXPORT_SYMBOL(delete_from_page_cache);
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 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
299 ret = do_writepages(mapping, &wbc);
300 wbc_detach_inode(&wbc);
304 static inline int __filemap_fdatawrite(struct address_space *mapping,
307 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
310 int filemap_fdatawrite(struct address_space *mapping)
312 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
314 EXPORT_SYMBOL(filemap_fdatawrite);
316 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
319 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
321 EXPORT_SYMBOL(filemap_fdatawrite_range);
324 * filemap_flush - mostly a non-blocking flush
325 * @mapping: target address_space
327 * This is a mostly non-blocking flush. Not suitable for data-integrity
328 * purposes - I/O may not be started against all dirty pages.
330 int filemap_flush(struct address_space *mapping)
332 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
334 EXPORT_SYMBOL(filemap_flush);
336 static int __filemap_fdatawait_range(struct address_space *mapping,
337 loff_t start_byte, loff_t end_byte)
339 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
340 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
345 if (end_byte < start_byte)
348 pagevec_init(&pvec, 0);
349 while ((index <= end) &&
350 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
351 PAGECACHE_TAG_WRITEBACK,
352 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
355 for (i = 0; i < nr_pages; i++) {
356 struct page *page = pvec.pages[i];
358 /* until radix tree lookup accepts end_index */
359 if (page->index > end)
362 wait_on_page_writeback(page);
363 if (TestClearPageError(page))
366 pagevec_release(&pvec);
374 * filemap_fdatawait_range - wait for writeback to complete
375 * @mapping: address space structure to wait for
376 * @start_byte: offset in bytes where the range starts
377 * @end_byte: offset in bytes where the range ends (inclusive)
379 * Walk the list of under-writeback pages of the given address space
380 * in the given range and wait for all of them. Check error status of
381 * the address space and return it.
383 * Since the error status of the address space is cleared by this function,
384 * callers are responsible for checking the return value and handling and/or
385 * reporting the error.
387 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
392 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
393 ret2 = filemap_check_errors(mapping);
399 EXPORT_SYMBOL(filemap_fdatawait_range);
402 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
403 * @mapping: address space structure to wait for
405 * Walk the list of under-writeback pages of the given address space
406 * and wait for all of them. Unlike filemap_fdatawait(), this function
407 * does not clear error status of the address space.
409 * Use this function if callers don't handle errors themselves. Expected
410 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
413 void filemap_fdatawait_keep_errors(struct address_space *mapping)
415 loff_t i_size = i_size_read(mapping->host);
420 __filemap_fdatawait_range(mapping, 0, i_size - 1);
424 * filemap_fdatawait - wait for all under-writeback pages to complete
425 * @mapping: address space structure to wait for
427 * Walk the list of under-writeback pages of the given address space
428 * and wait for all of them. Check error status of the address space
431 * Since the error status of the address space is cleared by this function,
432 * callers are responsible for checking the return value and handling and/or
433 * reporting the error.
435 int filemap_fdatawait(struct address_space *mapping)
437 loff_t i_size = i_size_read(mapping->host);
442 return filemap_fdatawait_range(mapping, 0, i_size - 1);
444 EXPORT_SYMBOL(filemap_fdatawait);
446 int filemap_write_and_wait(struct address_space *mapping)
450 if (mapping->nrpages) {
451 err = filemap_fdatawrite(mapping);
453 * Even if the above returned error, the pages may be
454 * written partially (e.g. -ENOSPC), so we wait for it.
455 * But the -EIO is special case, it may indicate the worst
456 * thing (e.g. bug) happened, so we avoid waiting for it.
459 int err2 = filemap_fdatawait(mapping);
464 err = filemap_check_errors(mapping);
468 EXPORT_SYMBOL(filemap_write_and_wait);
471 * filemap_write_and_wait_range - write out & wait on a file range
472 * @mapping: the address_space for the pages
473 * @lstart: offset in bytes where the range starts
474 * @lend: offset in bytes where the range ends (inclusive)
476 * Write out and wait upon file offsets lstart->lend, inclusive.
478 * Note that `lend' is inclusive (describes the last byte to be written) so
479 * that this function can be used to write to the very end-of-file (end = -1).
481 int filemap_write_and_wait_range(struct address_space *mapping,
482 loff_t lstart, loff_t lend)
486 if (mapping->nrpages) {
487 err = __filemap_fdatawrite_range(mapping, lstart, lend,
489 /* See comment of filemap_write_and_wait() */
491 int err2 = filemap_fdatawait_range(mapping,
497 err = filemap_check_errors(mapping);
501 EXPORT_SYMBOL(filemap_write_and_wait_range);
504 * replace_page_cache_page - replace a pagecache page with a new one
505 * @old: page to be replaced
506 * @new: page to replace with
507 * @gfp_mask: allocation mode
509 * This function replaces a page in the pagecache with a new one. On
510 * success it acquires the pagecache reference for the new page and
511 * drops it for the old page. Both the old and new pages must be
512 * locked. This function does not add the new page to the LRU, the
513 * caller must do that.
515 * The remove + add is atomic. The only way this function can fail is
516 * memory allocation failure.
518 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
522 VM_BUG_ON_PAGE(!PageLocked(old), old);
523 VM_BUG_ON_PAGE(!PageLocked(new), new);
524 VM_BUG_ON_PAGE(new->mapping, new);
526 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
528 struct address_space *mapping = old->mapping;
529 void (*freepage)(struct page *);
530 struct mem_cgroup *memcg;
533 pgoff_t offset = old->index;
534 freepage = mapping->a_ops->freepage;
537 new->mapping = mapping;
540 memcg = mem_cgroup_begin_page_stat(old);
541 spin_lock_irqsave(&mapping->tree_lock, flags);
542 __delete_from_page_cache(old, NULL, memcg);
543 error = radix_tree_insert(&mapping->page_tree, offset, new);
548 * hugetlb pages do not participate in page cache accounting.
551 __inc_zone_page_state(new, NR_FILE_PAGES);
552 if (PageSwapBacked(new))
553 __inc_zone_page_state(new, NR_SHMEM);
554 spin_unlock_irqrestore(&mapping->tree_lock, flags);
555 mem_cgroup_end_page_stat(memcg);
556 mem_cgroup_replace_page(old, new);
557 radix_tree_preload_end();
560 page_cache_release(old);
565 EXPORT_SYMBOL_GPL(replace_page_cache_page);
567 static int page_cache_tree_insert(struct address_space *mapping,
568 struct page *page, void **shadowp)
570 struct radix_tree_node *node;
574 error = __radix_tree_create(&mapping->page_tree, page->index,
581 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
582 if (!radix_tree_exceptional_entry(p))
586 mapping->nrshadows--;
588 workingset_node_shadows_dec(node);
590 radix_tree_replace_slot(slot, page);
593 workingset_node_pages_inc(node);
595 * Don't track node that contains actual pages.
597 * Avoid acquiring the list_lru lock if already
598 * untracked. The list_empty() test is safe as
599 * node->private_list is protected by
600 * mapping->tree_lock.
602 if (!list_empty(&node->private_list)) {
603 local_lock(workingset_shadow_lock);
604 list_lru_del(&__workingset_shadow_nodes,
605 &node->private_list);
606 local_unlock(workingset_shadow_lock);
612 static int __add_to_page_cache_locked(struct page *page,
613 struct address_space *mapping,
614 pgoff_t offset, gfp_t gfp_mask,
617 int huge = PageHuge(page);
618 struct mem_cgroup *memcg;
621 VM_BUG_ON_PAGE(!PageLocked(page), page);
622 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
625 error = mem_cgroup_try_charge(page, current->mm,
631 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
634 mem_cgroup_cancel_charge(page, memcg);
638 page_cache_get(page);
639 page->mapping = mapping;
640 page->index = offset;
642 spin_lock_irq(&mapping->tree_lock);
643 error = page_cache_tree_insert(mapping, page, shadowp);
644 radix_tree_preload_end();
648 /* hugetlb pages do not participate in page cache accounting. */
650 __inc_zone_page_state(page, NR_FILE_PAGES);
651 spin_unlock_irq(&mapping->tree_lock);
653 mem_cgroup_commit_charge(page, memcg, false);
654 trace_mm_filemap_add_to_page_cache(page);
657 page->mapping = NULL;
658 /* Leave page->index set: truncation relies upon it */
659 spin_unlock_irq(&mapping->tree_lock);
661 mem_cgroup_cancel_charge(page, memcg);
662 page_cache_release(page);
667 * add_to_page_cache_locked - add a locked page to the pagecache
669 * @mapping: the page's address_space
670 * @offset: page index
671 * @gfp_mask: page allocation mode
673 * This function is used to add a page to the pagecache. It must be locked.
674 * This function does not add the page to the LRU. The caller must do that.
676 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
677 pgoff_t offset, gfp_t gfp_mask)
679 return __add_to_page_cache_locked(page, mapping, offset,
682 EXPORT_SYMBOL(add_to_page_cache_locked);
684 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
685 pgoff_t offset, gfp_t gfp_mask)
690 __set_page_locked(page);
691 ret = __add_to_page_cache_locked(page, mapping, offset,
694 __clear_page_locked(page);
697 * The page might have been evicted from cache only
698 * recently, in which case it should be activated like
699 * any other repeatedly accessed page.
701 if (shadow && workingset_refault(shadow)) {
703 workingset_activation(page);
705 ClearPageActive(page);
710 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
713 struct page *__page_cache_alloc(gfp_t gfp)
718 if (cpuset_do_page_mem_spread()) {
719 unsigned int cpuset_mems_cookie;
721 cpuset_mems_cookie = read_mems_allowed_begin();
722 n = cpuset_mem_spread_node();
723 page = __alloc_pages_node(n, gfp, 0);
724 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
728 return alloc_pages(gfp, 0);
730 EXPORT_SYMBOL(__page_cache_alloc);
734 * In order to wait for pages to become available there must be
735 * waitqueues associated with pages. By using a hash table of
736 * waitqueues where the bucket discipline is to maintain all
737 * waiters on the same queue and wake all when any of the pages
738 * become available, and for the woken contexts to check to be
739 * sure the appropriate page became available, this saves space
740 * at a cost of "thundering herd" phenomena during rare hash
743 wait_queue_head_t *page_waitqueue(struct page *page)
745 const struct zone *zone = page_zone(page);
747 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
749 EXPORT_SYMBOL(page_waitqueue);
751 void wait_on_page_bit(struct page *page, int bit_nr)
753 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
755 if (test_bit(bit_nr, &page->flags))
756 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
757 TASK_UNINTERRUPTIBLE);
759 EXPORT_SYMBOL(wait_on_page_bit);
761 int wait_on_page_bit_killable(struct page *page, int bit_nr)
763 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
765 if (!test_bit(bit_nr, &page->flags))
768 return __wait_on_bit(page_waitqueue(page), &wait,
769 bit_wait_io, TASK_KILLABLE);
772 int wait_on_page_bit_killable_timeout(struct page *page,
773 int bit_nr, unsigned long timeout)
775 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
777 wait.key.timeout = jiffies + timeout;
778 if (!test_bit(bit_nr, &page->flags))
780 return __wait_on_bit(page_waitqueue(page), &wait,
781 bit_wait_io_timeout, TASK_KILLABLE);
783 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
786 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
787 * @page: Page defining the wait queue of interest
788 * @waiter: Waiter to add to the queue
790 * Add an arbitrary @waiter to the wait queue for the nominated @page.
792 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
794 wait_queue_head_t *q = page_waitqueue(page);
797 spin_lock_irqsave(&q->lock, flags);
798 __add_wait_queue(q, waiter);
799 spin_unlock_irqrestore(&q->lock, flags);
801 EXPORT_SYMBOL_GPL(add_page_wait_queue);
804 * unlock_page - unlock a locked page
807 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
808 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
809 * mechanism between PageLocked pages and PageWriteback pages is shared.
810 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
812 * The mb is necessary to enforce ordering between the clear_bit and the read
813 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
815 void unlock_page(struct page *page)
817 VM_BUG_ON_PAGE(!PageLocked(page), page);
818 clear_bit_unlock(PG_locked, &page->flags);
819 smp_mb__after_atomic();
820 wake_up_page(page, PG_locked);
822 EXPORT_SYMBOL(unlock_page);
825 * end_page_writeback - end writeback against a page
828 void end_page_writeback(struct page *page)
831 * TestClearPageReclaim could be used here but it is an atomic
832 * operation and overkill in this particular case. Failing to
833 * shuffle a page marked for immediate reclaim is too mild to
834 * justify taking an atomic operation penalty at the end of
835 * ever page writeback.
837 if (PageReclaim(page)) {
838 ClearPageReclaim(page);
839 rotate_reclaimable_page(page);
842 if (!test_clear_page_writeback(page))
845 smp_mb__after_atomic();
846 wake_up_page(page, PG_writeback);
848 EXPORT_SYMBOL(end_page_writeback);
851 * After completing I/O on a page, call this routine to update the page
852 * flags appropriately
854 void page_endio(struct page *page, int rw, int err)
858 SetPageUptodate(page);
860 ClearPageUptodate(page);
864 } else { /* rw == WRITE */
868 mapping_set_error(page->mapping, err);
870 end_page_writeback(page);
873 EXPORT_SYMBOL_GPL(page_endio);
876 * __lock_page - get a lock on the page, assuming we need to sleep to get it
877 * @page: the page to lock
879 void __lock_page(struct page *page)
881 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
883 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
884 TASK_UNINTERRUPTIBLE);
886 EXPORT_SYMBOL(__lock_page);
888 int __lock_page_killable(struct page *page)
890 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
892 return __wait_on_bit_lock(page_waitqueue(page), &wait,
893 bit_wait_io, TASK_KILLABLE);
895 EXPORT_SYMBOL_GPL(__lock_page_killable);
899 * 1 - page is locked; mmap_sem is still held.
900 * 0 - page is not locked.
901 * mmap_sem has been released (up_read()), unless flags had both
902 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
903 * which case mmap_sem is still held.
905 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
906 * with the page locked and the mmap_sem unperturbed.
908 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
911 if (flags & FAULT_FLAG_ALLOW_RETRY) {
913 * CAUTION! In this case, mmap_sem is not released
914 * even though return 0.
916 if (flags & FAULT_FLAG_RETRY_NOWAIT)
919 up_read(&mm->mmap_sem);
920 if (flags & FAULT_FLAG_KILLABLE)
921 wait_on_page_locked_killable(page);
923 wait_on_page_locked(page);
926 if (flags & FAULT_FLAG_KILLABLE) {
929 ret = __lock_page_killable(page);
931 up_read(&mm->mmap_sem);
941 * page_cache_next_hole - find the next hole (not-present entry)
944 * @max_scan: maximum range to search
946 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
947 * lowest indexed hole.
949 * Returns: the index of the hole if found, otherwise returns an index
950 * outside of the set specified (in which case 'return - index >=
951 * max_scan' will be true). In rare cases of index wrap-around, 0 will
954 * page_cache_next_hole may be called under rcu_read_lock. However,
955 * like radix_tree_gang_lookup, this will not atomically search a
956 * snapshot of the tree at a single point in time. For example, if a
957 * hole is created at index 5, then subsequently a hole is created at
958 * index 10, page_cache_next_hole covering both indexes may return 10
959 * if called under rcu_read_lock.
961 pgoff_t page_cache_next_hole(struct address_space *mapping,
962 pgoff_t index, unsigned long max_scan)
966 for (i = 0; i < max_scan; i++) {
969 page = radix_tree_lookup(&mapping->page_tree, index);
970 if (!page || radix_tree_exceptional_entry(page))
979 EXPORT_SYMBOL(page_cache_next_hole);
982 * page_cache_prev_hole - find the prev hole (not-present entry)
985 * @max_scan: maximum range to search
987 * Search backwards in the range [max(index-max_scan+1, 0), index] for
990 * Returns: the index of the hole if found, otherwise returns an index
991 * outside of the set specified (in which case 'index - return >=
992 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
995 * page_cache_prev_hole may be called under rcu_read_lock. However,
996 * like radix_tree_gang_lookup, this will not atomically search a
997 * snapshot of the tree at a single point in time. For example, if a
998 * hole is created at index 10, then subsequently a hole is created at
999 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1000 * called under rcu_read_lock.
1002 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1003 pgoff_t index, unsigned long max_scan)
1007 for (i = 0; i < max_scan; i++) {
1010 page = radix_tree_lookup(&mapping->page_tree, index);
1011 if (!page || radix_tree_exceptional_entry(page))
1014 if (index == ULONG_MAX)
1020 EXPORT_SYMBOL(page_cache_prev_hole);
1023 * find_get_entry - find and get a page cache entry
1024 * @mapping: the address_space to search
1025 * @offset: the page cache index
1027 * Looks up the page cache slot at @mapping & @offset. If there is a
1028 * page cache page, it is returned with an increased refcount.
1030 * If the slot holds a shadow entry of a previously evicted page, or a
1031 * swap entry from shmem/tmpfs, it is returned.
1033 * Otherwise, %NULL is returned.
1035 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1043 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1045 page = radix_tree_deref_slot(pagep);
1046 if (unlikely(!page))
1048 if (radix_tree_exception(page)) {
1049 if (radix_tree_deref_retry(page))
1052 * A shadow entry of a recently evicted page,
1053 * or a swap entry from shmem/tmpfs. Return
1054 * it without attempting to raise page count.
1058 if (!page_cache_get_speculative(page))
1062 * Has the page moved?
1063 * This is part of the lockless pagecache protocol. See
1064 * include/linux/pagemap.h for details.
1066 if (unlikely(page != *pagep)) {
1067 page_cache_release(page);
1076 EXPORT_SYMBOL(find_get_entry);
1079 * find_lock_entry - locate, pin and lock a page cache entry
1080 * @mapping: the address_space to search
1081 * @offset: the page cache index
1083 * Looks up the page cache slot at @mapping & @offset. If there is a
1084 * page cache page, it is returned locked and with an increased
1087 * If the slot holds a shadow entry of a previously evicted page, or a
1088 * swap entry from shmem/tmpfs, it is returned.
1090 * Otherwise, %NULL is returned.
1092 * find_lock_entry() may sleep.
1094 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1099 page = find_get_entry(mapping, offset);
1100 if (page && !radix_tree_exception(page)) {
1102 /* Has the page been truncated? */
1103 if (unlikely(page->mapping != mapping)) {
1105 page_cache_release(page);
1108 VM_BUG_ON_PAGE(page->index != offset, page);
1112 EXPORT_SYMBOL(find_lock_entry);
1115 * pagecache_get_page - find and get a page reference
1116 * @mapping: the address_space to search
1117 * @offset: the page index
1118 * @fgp_flags: PCG flags
1119 * @gfp_mask: gfp mask to use for the page cache data page allocation
1121 * Looks up the page cache slot at @mapping & @offset.
1123 * PCG flags modify how the page is returned.
1125 * FGP_ACCESSED: the page will be marked accessed
1126 * FGP_LOCK: Page is return locked
1127 * FGP_CREAT: If page is not present then a new page is allocated using
1128 * @gfp_mask and added to the page cache and the VM's LRU
1129 * list. The page is returned locked and with an increased
1130 * refcount. Otherwise, %NULL is returned.
1132 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1133 * if the GFP flags specified for FGP_CREAT are atomic.
1135 * If there is a page cache page, it is returned with an increased refcount.
1137 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1138 int fgp_flags, gfp_t gfp_mask)
1143 page = find_get_entry(mapping, offset);
1144 if (radix_tree_exceptional_entry(page))
1149 if (fgp_flags & FGP_LOCK) {
1150 if (fgp_flags & FGP_NOWAIT) {
1151 if (!trylock_page(page)) {
1152 page_cache_release(page);
1159 /* Has the page been truncated? */
1160 if (unlikely(page->mapping != mapping)) {
1162 page_cache_release(page);
1165 VM_BUG_ON_PAGE(page->index != offset, page);
1168 if (page && (fgp_flags & FGP_ACCESSED))
1169 mark_page_accessed(page);
1172 if (!page && (fgp_flags & FGP_CREAT)) {
1174 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1175 gfp_mask |= __GFP_WRITE;
1176 if (fgp_flags & FGP_NOFS)
1177 gfp_mask &= ~__GFP_FS;
1179 page = __page_cache_alloc(gfp_mask);
1183 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1184 fgp_flags |= FGP_LOCK;
1186 /* Init accessed so avoid atomic mark_page_accessed later */
1187 if (fgp_flags & FGP_ACCESSED)
1188 __SetPageReferenced(page);
1190 err = add_to_page_cache_lru(page, mapping, offset,
1191 gfp_mask & GFP_RECLAIM_MASK);
1192 if (unlikely(err)) {
1193 page_cache_release(page);
1202 EXPORT_SYMBOL(pagecache_get_page);
1205 * find_get_entries - gang pagecache lookup
1206 * @mapping: The address_space to search
1207 * @start: The starting page cache index
1208 * @nr_entries: The maximum number of entries
1209 * @entries: Where the resulting entries are placed
1210 * @indices: The cache indices corresponding to the entries in @entries
1212 * find_get_entries() will search for and return a group of up to
1213 * @nr_entries entries in the mapping. The entries are placed at
1214 * @entries. find_get_entries() takes a reference against any actual
1217 * The search returns a group of mapping-contiguous page cache entries
1218 * with ascending indexes. There may be holes in the indices due to
1219 * not-present pages.
1221 * Any shadow entries of evicted pages, or swap entries from
1222 * shmem/tmpfs, are included in the returned array.
1224 * find_get_entries() returns the number of pages and shadow entries
1227 unsigned find_get_entries(struct address_space *mapping,
1228 pgoff_t start, unsigned int nr_entries,
1229 struct page **entries, pgoff_t *indices)
1232 unsigned int ret = 0;
1233 struct radix_tree_iter iter;
1240 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1243 page = radix_tree_deref_slot(slot);
1244 if (unlikely(!page))
1246 if (radix_tree_exception(page)) {
1247 if (radix_tree_deref_retry(page))
1250 * A shadow entry of a recently evicted page,
1251 * or a swap entry from shmem/tmpfs. Return
1252 * it without attempting to raise page count.
1256 if (!page_cache_get_speculative(page))
1259 /* Has the page moved? */
1260 if (unlikely(page != *slot)) {
1261 page_cache_release(page);
1265 indices[ret] = iter.index;
1266 entries[ret] = page;
1267 if (++ret == nr_entries)
1275 * find_get_pages - gang pagecache lookup
1276 * @mapping: The address_space to search
1277 * @start: The starting page index
1278 * @nr_pages: The maximum number of pages
1279 * @pages: Where the resulting pages are placed
1281 * find_get_pages() will search for and return a group of up to
1282 * @nr_pages pages in the mapping. The pages are placed at @pages.
1283 * find_get_pages() takes a reference against the returned pages.
1285 * The search returns a group of mapping-contiguous pages with ascending
1286 * indexes. There may be holes in the indices due to not-present pages.
1288 * find_get_pages() returns the number of pages which were found.
1290 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1291 unsigned int nr_pages, struct page **pages)
1293 struct radix_tree_iter iter;
1297 if (unlikely(!nr_pages))
1302 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1305 page = radix_tree_deref_slot(slot);
1306 if (unlikely(!page))
1309 if (radix_tree_exception(page)) {
1310 if (radix_tree_deref_retry(page)) {
1312 * Transient condition which can only trigger
1313 * when entry at index 0 moves out of or back
1314 * to root: none yet gotten, safe to restart.
1316 WARN_ON(iter.index);
1320 * A shadow entry of a recently evicted page,
1321 * or a swap entry from shmem/tmpfs. Skip
1327 if (!page_cache_get_speculative(page))
1330 /* Has the page moved? */
1331 if (unlikely(page != *slot)) {
1332 page_cache_release(page);
1337 if (++ret == nr_pages)
1346 * find_get_pages_contig - gang contiguous pagecache lookup
1347 * @mapping: The address_space to search
1348 * @index: The starting page index
1349 * @nr_pages: The maximum number of pages
1350 * @pages: Where the resulting pages are placed
1352 * find_get_pages_contig() works exactly like find_get_pages(), except
1353 * that the returned number of pages are guaranteed to be contiguous.
1355 * find_get_pages_contig() returns the number of pages which were found.
1357 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1358 unsigned int nr_pages, struct page **pages)
1360 struct radix_tree_iter iter;
1362 unsigned int ret = 0;
1364 if (unlikely(!nr_pages))
1369 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1372 page = radix_tree_deref_slot(slot);
1373 /* The hole, there no reason to continue */
1374 if (unlikely(!page))
1377 if (radix_tree_exception(page)) {
1378 if (radix_tree_deref_retry(page)) {
1380 * Transient condition which can only trigger
1381 * when entry at index 0 moves out of or back
1382 * to root: none yet gotten, safe to restart.
1387 * A shadow entry of a recently evicted page,
1388 * or a swap entry from shmem/tmpfs. Stop
1389 * looking for contiguous pages.
1394 if (!page_cache_get_speculative(page))
1397 /* Has the page moved? */
1398 if (unlikely(page != *slot)) {
1399 page_cache_release(page);
1404 * must check mapping and index after taking the ref.
1405 * otherwise we can get both false positives and false
1406 * negatives, which is just confusing to the caller.
1408 if (page->mapping == NULL || page->index != iter.index) {
1409 page_cache_release(page);
1414 if (++ret == nr_pages)
1420 EXPORT_SYMBOL(find_get_pages_contig);
1423 * find_get_pages_tag - find and return pages that match @tag
1424 * @mapping: the address_space to search
1425 * @index: the starting page index
1426 * @tag: the tag index
1427 * @nr_pages: the maximum number of pages
1428 * @pages: where the resulting pages are placed
1430 * Like find_get_pages, except we only return pages which are tagged with
1431 * @tag. We update @index to index the next page for the traversal.
1433 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1434 int tag, unsigned int nr_pages, struct page **pages)
1436 struct radix_tree_iter iter;
1440 if (unlikely(!nr_pages))
1445 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1446 &iter, *index, tag) {
1449 page = radix_tree_deref_slot(slot);
1450 if (unlikely(!page))
1453 if (radix_tree_exception(page)) {
1454 if (radix_tree_deref_retry(page)) {
1456 * Transient condition which can only trigger
1457 * when entry at index 0 moves out of or back
1458 * to root: none yet gotten, safe to restart.
1463 * A shadow entry of a recently evicted page.
1465 * Those entries should never be tagged, but
1466 * this tree walk is lockless and the tags are
1467 * looked up in bulk, one radix tree node at a
1468 * time, so there is a sizable window for page
1469 * reclaim to evict a page we saw tagged.
1476 if (!page_cache_get_speculative(page))
1479 /* Has the page moved? */
1480 if (unlikely(page != *slot)) {
1481 page_cache_release(page);
1486 if (++ret == nr_pages)
1493 *index = pages[ret - 1]->index + 1;
1497 EXPORT_SYMBOL(find_get_pages_tag);
1500 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1501 * a _large_ part of the i/o request. Imagine the worst scenario:
1503 * ---R__________________________________________B__________
1504 * ^ reading here ^ bad block(assume 4k)
1506 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1507 * => failing the whole request => read(R) => read(R+1) =>
1508 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1509 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1510 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1512 * It is going insane. Fix it by quickly scaling down the readahead size.
1514 static void shrink_readahead_size_eio(struct file *filp,
1515 struct file_ra_state *ra)
1521 * do_generic_file_read - generic file read routine
1522 * @filp: the file to read
1523 * @ppos: current file position
1524 * @iter: data destination
1525 * @written: already copied
1527 * This is a generic file read routine, and uses the
1528 * mapping->a_ops->readpage() function for the actual low-level stuff.
1530 * This is really ugly. But the goto's actually try to clarify some
1531 * of the logic when it comes to error handling etc.
1533 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1534 struct iov_iter *iter, ssize_t written)
1536 struct address_space *mapping = filp->f_mapping;
1537 struct inode *inode = mapping->host;
1538 struct file_ra_state *ra = &filp->f_ra;
1542 unsigned long offset; /* offset into pagecache page */
1543 unsigned int prev_offset;
1546 index = *ppos >> PAGE_CACHE_SHIFT;
1547 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1548 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1549 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1550 offset = *ppos & ~PAGE_CACHE_MASK;
1556 unsigned long nr, ret;
1560 page = find_get_page(mapping, index);
1562 page_cache_sync_readahead(mapping,
1564 index, last_index - index);
1565 page = find_get_page(mapping, index);
1566 if (unlikely(page == NULL))
1567 goto no_cached_page;
1569 if (PageReadahead(page)) {
1570 page_cache_async_readahead(mapping,
1572 index, last_index - index);
1574 if (!PageUptodate(page)) {
1575 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1576 !mapping->a_ops->is_partially_uptodate)
1577 goto page_not_up_to_date;
1578 if (!trylock_page(page))
1579 goto page_not_up_to_date;
1580 /* Did it get truncated before we got the lock? */
1582 goto page_not_up_to_date_locked;
1583 if (!mapping->a_ops->is_partially_uptodate(page,
1584 offset, iter->count))
1585 goto page_not_up_to_date_locked;
1590 * i_size must be checked after we know the page is Uptodate.
1592 * Checking i_size after the check allows us to calculate
1593 * the correct value for "nr", which means the zero-filled
1594 * part of the page is not copied back to userspace (unless
1595 * another truncate extends the file - this is desired though).
1598 isize = i_size_read(inode);
1599 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1600 if (unlikely(!isize || index > end_index)) {
1601 page_cache_release(page);
1605 /* nr is the maximum number of bytes to copy from this page */
1606 nr = PAGE_CACHE_SIZE;
1607 if (index == end_index) {
1608 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1610 page_cache_release(page);
1616 /* If users can be writing to this page using arbitrary
1617 * virtual addresses, take care about potential aliasing
1618 * before reading the page on the kernel side.
1620 if (mapping_writably_mapped(mapping))
1621 flush_dcache_page(page);
1624 * When a sequential read accesses a page several times,
1625 * only mark it as accessed the first time.
1627 if (prev_index != index || offset != prev_offset)
1628 mark_page_accessed(page);
1632 * Ok, we have the page, and it's up-to-date, so
1633 * now we can copy it to user space...
1636 ret = copy_page_to_iter(page, offset, nr, iter);
1638 index += offset >> PAGE_CACHE_SHIFT;
1639 offset &= ~PAGE_CACHE_MASK;
1640 prev_offset = offset;
1642 page_cache_release(page);
1644 if (!iov_iter_count(iter))
1652 page_not_up_to_date:
1653 /* Get exclusive access to the page ... */
1654 error = lock_page_killable(page);
1655 if (unlikely(error))
1656 goto readpage_error;
1658 page_not_up_to_date_locked:
1659 /* Did it get truncated before we got the lock? */
1660 if (!page->mapping) {
1662 page_cache_release(page);
1666 /* Did somebody else fill it already? */
1667 if (PageUptodate(page)) {
1674 * A previous I/O error may have been due to temporary
1675 * failures, eg. multipath errors.
1676 * PG_error will be set again if readpage fails.
1678 ClearPageError(page);
1679 /* Start the actual read. The read will unlock the page. */
1680 error = mapping->a_ops->readpage(filp, page);
1682 if (unlikely(error)) {
1683 if (error == AOP_TRUNCATED_PAGE) {
1684 page_cache_release(page);
1688 goto readpage_error;
1691 if (!PageUptodate(page)) {
1692 error = lock_page_killable(page);
1693 if (unlikely(error))
1694 goto readpage_error;
1695 if (!PageUptodate(page)) {
1696 if (page->mapping == NULL) {
1698 * invalidate_mapping_pages got it
1701 page_cache_release(page);
1705 shrink_readahead_size_eio(filp, ra);
1707 goto readpage_error;
1715 /* UHHUH! A synchronous read error occurred. Report it */
1716 page_cache_release(page);
1721 * Ok, it wasn't cached, so we need to create a new
1724 page = page_cache_alloc_cold(mapping);
1729 error = add_to_page_cache_lru(page, mapping, index,
1730 mapping_gfp_constraint(mapping, GFP_KERNEL));
1732 page_cache_release(page);
1733 if (error == -EEXIST) {
1743 ra->prev_pos = prev_index;
1744 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1745 ra->prev_pos |= prev_offset;
1747 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1748 file_accessed(filp);
1749 return written ? written : error;
1753 * generic_file_read_iter - generic filesystem read routine
1754 * @iocb: kernel I/O control block
1755 * @iter: destination for the data read
1757 * This is the "read_iter()" routine for all filesystems
1758 * that can use the page cache directly.
1761 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1763 struct file *file = iocb->ki_filp;
1765 loff_t *ppos = &iocb->ki_pos;
1768 if (iocb->ki_flags & IOCB_DIRECT) {
1769 struct address_space *mapping = file->f_mapping;
1770 struct inode *inode = mapping->host;
1771 size_t count = iov_iter_count(iter);
1775 goto out; /* skip atime */
1776 size = i_size_read(inode);
1777 retval = filemap_write_and_wait_range(mapping, pos,
1780 struct iov_iter data = *iter;
1781 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
1785 *ppos = pos + retval;
1786 iov_iter_advance(iter, retval);
1790 * Btrfs can have a short DIO read if we encounter
1791 * compressed extents, so if there was an error, or if
1792 * we've already read everything we wanted to, or if
1793 * there was a short read because we hit EOF, go ahead
1794 * and return. Otherwise fallthrough to buffered io for
1795 * the rest of the read. Buffered reads will not work for
1796 * DAX files, so don't bother trying.
1798 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1800 file_accessed(file);
1805 retval = do_generic_file_read(file, ppos, iter, retval);
1809 EXPORT_SYMBOL(generic_file_read_iter);
1813 * page_cache_read - adds requested page to the page cache if not already there
1814 * @file: file to read
1815 * @offset: page index
1817 * This adds the requested page to the page cache if it isn't already there,
1818 * and schedules an I/O to read in its contents from disk.
1820 static int page_cache_read(struct file *file, pgoff_t offset)
1822 struct address_space *mapping = file->f_mapping;
1827 page = page_cache_alloc_cold(mapping);
1831 ret = add_to_page_cache_lru(page, mapping, offset,
1832 mapping_gfp_constraint(mapping, GFP_KERNEL));
1834 ret = mapping->a_ops->readpage(file, page);
1835 else if (ret == -EEXIST)
1836 ret = 0; /* losing race to add is OK */
1838 page_cache_release(page);
1840 } while (ret == AOP_TRUNCATED_PAGE);
1845 #define MMAP_LOTSAMISS (100)
1848 * Synchronous readahead happens when we don't even find
1849 * a page in the page cache at all.
1851 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1852 struct file_ra_state *ra,
1856 struct address_space *mapping = file->f_mapping;
1858 /* If we don't want any read-ahead, don't bother */
1859 if (vma->vm_flags & VM_RAND_READ)
1864 if (vma->vm_flags & VM_SEQ_READ) {
1865 page_cache_sync_readahead(mapping, ra, file, offset,
1870 /* Avoid banging the cache line if not needed */
1871 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1875 * Do we miss much more than hit in this file? If so,
1876 * stop bothering with read-ahead. It will only hurt.
1878 if (ra->mmap_miss > MMAP_LOTSAMISS)
1884 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1885 ra->size = ra->ra_pages;
1886 ra->async_size = ra->ra_pages / 4;
1887 ra_submit(ra, mapping, file);
1891 * Asynchronous readahead happens when we find the page and PG_readahead,
1892 * so we want to possibly extend the readahead further..
1894 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1895 struct file_ra_state *ra,
1900 struct address_space *mapping = file->f_mapping;
1902 /* If we don't want any read-ahead, don't bother */
1903 if (vma->vm_flags & VM_RAND_READ)
1905 if (ra->mmap_miss > 0)
1907 if (PageReadahead(page))
1908 page_cache_async_readahead(mapping, ra, file,
1909 page, offset, ra->ra_pages);
1913 * filemap_fault - read in file data for page fault handling
1914 * @vma: vma in which the fault was taken
1915 * @vmf: struct vm_fault containing details of the fault
1917 * filemap_fault() is invoked via the vma operations vector for a
1918 * mapped memory region to read in file data during a page fault.
1920 * The goto's are kind of ugly, but this streamlines the normal case of having
1921 * it in the page cache, and handles the special cases reasonably without
1922 * having a lot of duplicated code.
1924 * vma->vm_mm->mmap_sem must be held on entry.
1926 * If our return value has VM_FAULT_RETRY set, it's because
1927 * lock_page_or_retry() returned 0.
1928 * The mmap_sem has usually been released in this case.
1929 * See __lock_page_or_retry() for the exception.
1931 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1932 * has not been released.
1934 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1936 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1939 struct file *file = vma->vm_file;
1940 struct address_space *mapping = file->f_mapping;
1941 struct file_ra_state *ra = &file->f_ra;
1942 struct inode *inode = mapping->host;
1943 pgoff_t offset = vmf->pgoff;
1948 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1949 if (offset >= size >> PAGE_CACHE_SHIFT)
1950 return VM_FAULT_SIGBUS;
1953 * Do we have something in the page cache already?
1955 page = find_get_page(mapping, offset);
1956 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1958 * We found the page, so try async readahead before
1959 * waiting for the lock.
1961 do_async_mmap_readahead(vma, ra, file, page, offset);
1963 /* No page in the page cache at all */
1964 do_sync_mmap_readahead(vma, ra, file, offset);
1965 count_vm_event(PGMAJFAULT);
1966 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1967 ret = VM_FAULT_MAJOR;
1969 page = find_get_page(mapping, offset);
1971 goto no_cached_page;
1974 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1975 page_cache_release(page);
1976 return ret | VM_FAULT_RETRY;
1979 /* Did it get truncated? */
1980 if (unlikely(page->mapping != mapping)) {
1985 VM_BUG_ON_PAGE(page->index != offset, page);
1988 * We have a locked page in the page cache, now we need to check
1989 * that it's up-to-date. If not, it is going to be due to an error.
1991 if (unlikely(!PageUptodate(page)))
1992 goto page_not_uptodate;
1995 * Found the page and have a reference on it.
1996 * We must recheck i_size under page lock.
1998 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1999 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
2001 page_cache_release(page);
2002 return VM_FAULT_SIGBUS;
2006 return ret | VM_FAULT_LOCKED;
2010 * We're only likely to ever get here if MADV_RANDOM is in
2013 error = page_cache_read(file, offset);
2016 * The page we want has now been added to the page cache.
2017 * In the unlikely event that someone removed it in the
2018 * meantime, we'll just come back here and read it again.
2024 * An error return from page_cache_read can result if the
2025 * system is low on memory, or a problem occurs while trying
2028 if (error == -ENOMEM)
2029 return VM_FAULT_OOM;
2030 return VM_FAULT_SIGBUS;
2034 * Umm, take care of errors if the page isn't up-to-date.
2035 * Try to re-read it _once_. We do this synchronously,
2036 * because there really aren't any performance issues here
2037 * and we need to check for errors.
2039 ClearPageError(page);
2040 error = mapping->a_ops->readpage(file, page);
2042 wait_on_page_locked(page);
2043 if (!PageUptodate(page))
2046 page_cache_release(page);
2048 if (!error || error == AOP_TRUNCATED_PAGE)
2051 /* Things didn't work out. Return zero to tell the mm layer so. */
2052 shrink_readahead_size_eio(file, ra);
2053 return VM_FAULT_SIGBUS;
2055 EXPORT_SYMBOL(filemap_fault);
2057 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2059 struct radix_tree_iter iter;
2061 struct file *file = vma->vm_file;
2062 struct address_space *mapping = file->f_mapping;
2065 unsigned long address = (unsigned long) vmf->virtual_address;
2070 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2071 if (iter.index > vmf->max_pgoff)
2074 page = radix_tree_deref_slot(slot);
2075 if (unlikely(!page))
2077 if (radix_tree_exception(page)) {
2078 if (radix_tree_deref_retry(page))
2084 if (!page_cache_get_speculative(page))
2087 /* Has the page moved? */
2088 if (unlikely(page != *slot)) {
2089 page_cache_release(page);
2093 if (!PageUptodate(page) ||
2094 PageReadahead(page) ||
2097 if (!trylock_page(page))
2100 if (page->mapping != mapping || !PageUptodate(page))
2103 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2104 if (page->index >= size >> PAGE_CACHE_SHIFT)
2107 pte = vmf->pte + page->index - vmf->pgoff;
2108 if (!pte_none(*pte))
2111 if (file->f_ra.mmap_miss > 0)
2112 file->f_ra.mmap_miss--;
2113 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2114 do_set_pte(vma, addr, page, pte, false, false);
2120 page_cache_release(page);
2122 if (iter.index == vmf->max_pgoff)
2127 EXPORT_SYMBOL(filemap_map_pages);
2129 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2131 struct page *page = vmf->page;
2132 struct inode *inode = file_inode(vma->vm_file);
2133 int ret = VM_FAULT_LOCKED;
2135 sb_start_pagefault(inode->i_sb);
2136 file_update_time(vma->vm_file);
2138 if (page->mapping != inode->i_mapping) {
2140 ret = VM_FAULT_NOPAGE;
2144 * We mark the page dirty already here so that when freeze is in
2145 * progress, we are guaranteed that writeback during freezing will
2146 * see the dirty page and writeprotect it again.
2148 set_page_dirty(page);
2149 wait_for_stable_page(page);
2151 sb_end_pagefault(inode->i_sb);
2154 EXPORT_SYMBOL(filemap_page_mkwrite);
2156 const struct vm_operations_struct generic_file_vm_ops = {
2157 .fault = filemap_fault,
2158 .map_pages = filemap_map_pages,
2159 .page_mkwrite = filemap_page_mkwrite,
2162 /* This is used for a general mmap of a disk file */
2164 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2166 struct address_space *mapping = file->f_mapping;
2168 if (!mapping->a_ops->readpage)
2170 file_accessed(file);
2171 vma->vm_ops = &generic_file_vm_ops;
2176 * This is for filesystems which do not implement ->writepage.
2178 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2180 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2182 return generic_file_mmap(file, vma);
2185 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2189 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2193 #endif /* CONFIG_MMU */
2195 EXPORT_SYMBOL(generic_file_mmap);
2196 EXPORT_SYMBOL(generic_file_readonly_mmap);
2198 static struct page *wait_on_page_read(struct page *page)
2200 if (!IS_ERR(page)) {
2201 wait_on_page_locked(page);
2202 if (!PageUptodate(page)) {
2203 page_cache_release(page);
2204 page = ERR_PTR(-EIO);
2210 static struct page *__read_cache_page(struct address_space *mapping,
2212 int (*filler)(void *, struct page *),
2219 page = find_get_page(mapping, index);
2221 page = __page_cache_alloc(gfp | __GFP_COLD);
2223 return ERR_PTR(-ENOMEM);
2224 err = add_to_page_cache_lru(page, mapping, index, gfp);
2225 if (unlikely(err)) {
2226 page_cache_release(page);
2229 /* Presumably ENOMEM for radix tree node */
2230 return ERR_PTR(err);
2232 err = filler(data, page);
2234 page_cache_release(page);
2235 page = ERR_PTR(err);
2237 page = wait_on_page_read(page);
2243 static struct page *do_read_cache_page(struct address_space *mapping,
2245 int (*filler)(void *, struct page *),
2254 page = __read_cache_page(mapping, index, filler, data, gfp);
2257 if (PageUptodate(page))
2261 if (!page->mapping) {
2263 page_cache_release(page);
2266 if (PageUptodate(page)) {
2270 err = filler(data, page);
2272 page_cache_release(page);
2273 return ERR_PTR(err);
2275 page = wait_on_page_read(page);
2280 mark_page_accessed(page);
2285 * read_cache_page - read into page cache, fill it if needed
2286 * @mapping: the page's address_space
2287 * @index: the page index
2288 * @filler: function to perform the read
2289 * @data: first arg to filler(data, page) function, often left as NULL
2291 * Read into the page cache. If a page already exists, and PageUptodate() is
2292 * not set, try to fill the page and wait for it to become unlocked.
2294 * If the page does not get brought uptodate, return -EIO.
2296 struct page *read_cache_page(struct address_space *mapping,
2298 int (*filler)(void *, struct page *),
2301 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2303 EXPORT_SYMBOL(read_cache_page);
2306 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2307 * @mapping: the page's address_space
2308 * @index: the page index
2309 * @gfp: the page allocator flags to use if allocating
2311 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2312 * any new page allocations done using the specified allocation flags.
2314 * If the page does not get brought uptodate, return -EIO.
2316 struct page *read_cache_page_gfp(struct address_space *mapping,
2320 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2322 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2324 EXPORT_SYMBOL(read_cache_page_gfp);
2327 * Performs necessary checks before doing a write
2329 * Can adjust writing position or amount of bytes to write.
2330 * Returns appropriate error code that caller should return or
2331 * zero in case that write should be allowed.
2333 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2335 struct file *file = iocb->ki_filp;
2336 struct inode *inode = file->f_mapping->host;
2337 unsigned long limit = rlimit(RLIMIT_FSIZE);
2340 if (!iov_iter_count(from))
2343 /* FIXME: this is for backwards compatibility with 2.4 */
2344 if (iocb->ki_flags & IOCB_APPEND)
2345 iocb->ki_pos = i_size_read(inode);
2349 if (limit != RLIM_INFINITY) {
2350 if (iocb->ki_pos >= limit) {
2351 send_sig(SIGXFSZ, current, 0);
2354 iov_iter_truncate(from, limit - (unsigned long)pos);
2360 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2361 !(file->f_flags & O_LARGEFILE))) {
2362 if (pos >= MAX_NON_LFS)
2364 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2368 * Are we about to exceed the fs block limit ?
2370 * If we have written data it becomes a short write. If we have
2371 * exceeded without writing data we send a signal and return EFBIG.
2372 * Linus frestrict idea will clean these up nicely..
2374 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2377 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2378 return iov_iter_count(from);
2380 EXPORT_SYMBOL(generic_write_checks);
2382 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2383 loff_t pos, unsigned len, unsigned flags,
2384 struct page **pagep, void **fsdata)
2386 const struct address_space_operations *aops = mapping->a_ops;
2388 return aops->write_begin(file, mapping, pos, len, flags,
2391 EXPORT_SYMBOL(pagecache_write_begin);
2393 int pagecache_write_end(struct file *file, struct address_space *mapping,
2394 loff_t pos, unsigned len, unsigned copied,
2395 struct page *page, void *fsdata)
2397 const struct address_space_operations *aops = mapping->a_ops;
2399 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2401 EXPORT_SYMBOL(pagecache_write_end);
2404 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2406 struct file *file = iocb->ki_filp;
2407 struct address_space *mapping = file->f_mapping;
2408 struct inode *inode = mapping->host;
2412 struct iov_iter data;
2414 write_len = iov_iter_count(from);
2415 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2417 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2422 * After a write we want buffered reads to be sure to go to disk to get
2423 * the new data. We invalidate clean cached page from the region we're
2424 * about to write. We do this *before* the write so that we can return
2425 * without clobbering -EIOCBQUEUED from ->direct_IO().
2427 if (mapping->nrpages) {
2428 written = invalidate_inode_pages2_range(mapping,
2429 pos >> PAGE_CACHE_SHIFT, end);
2431 * If a page can not be invalidated, return 0 to fall back
2432 * to buffered write.
2435 if (written == -EBUSY)
2442 written = mapping->a_ops->direct_IO(iocb, &data, pos);
2445 * Finally, try again to invalidate clean pages which might have been
2446 * cached by non-direct readahead, or faulted in by get_user_pages()
2447 * if the source of the write was an mmap'ed region of the file
2448 * we're writing. Either one is a pretty crazy thing to do,
2449 * so we don't support it 100%. If this invalidation
2450 * fails, tough, the write still worked...
2452 if (mapping->nrpages) {
2453 invalidate_inode_pages2_range(mapping,
2454 pos >> PAGE_CACHE_SHIFT, end);
2459 iov_iter_advance(from, written);
2460 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2461 i_size_write(inode, pos);
2462 mark_inode_dirty(inode);
2469 EXPORT_SYMBOL(generic_file_direct_write);
2472 * Find or create a page at the given pagecache position. Return the locked
2473 * page. This function is specifically for buffered writes.
2475 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2476 pgoff_t index, unsigned flags)
2479 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2481 if (flags & AOP_FLAG_NOFS)
2482 fgp_flags |= FGP_NOFS;
2484 page = pagecache_get_page(mapping, index, fgp_flags,
2485 mapping_gfp_mask(mapping));
2487 wait_for_stable_page(page);
2491 EXPORT_SYMBOL(grab_cache_page_write_begin);
2493 ssize_t generic_perform_write(struct file *file,
2494 struct iov_iter *i, loff_t pos)
2496 struct address_space *mapping = file->f_mapping;
2497 const struct address_space_operations *a_ops = mapping->a_ops;
2499 ssize_t written = 0;
2500 unsigned int flags = 0;
2503 * Copies from kernel address space cannot fail (NFSD is a big user).
2505 if (!iter_is_iovec(i))
2506 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2510 unsigned long offset; /* Offset into pagecache page */
2511 unsigned long bytes; /* Bytes to write to page */
2512 size_t copied; /* Bytes copied from user */
2515 offset = (pos & (PAGE_CACHE_SIZE - 1));
2516 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2521 * Bring in the user page that we will copy from _first_.
2522 * Otherwise there's a nasty deadlock on copying from the
2523 * same page as we're writing to, without it being marked
2526 * Not only is this an optimisation, but it is also required
2527 * to check that the address is actually valid, when atomic
2528 * usercopies are used, below.
2530 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2535 if (fatal_signal_pending(current)) {
2540 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2542 if (unlikely(status < 0))
2545 if (mapping_writably_mapped(mapping))
2546 flush_dcache_page(page);
2548 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2549 flush_dcache_page(page);
2551 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2553 if (unlikely(status < 0))
2559 iov_iter_advance(i, copied);
2560 if (unlikely(copied == 0)) {
2562 * If we were unable to copy any data at all, we must
2563 * fall back to a single segment length write.
2565 * If we didn't fallback here, we could livelock
2566 * because not all segments in the iov can be copied at
2567 * once without a pagefault.
2569 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2570 iov_iter_single_seg_count(i));
2576 balance_dirty_pages_ratelimited(mapping);
2577 } while (iov_iter_count(i));
2579 return written ? written : status;
2581 EXPORT_SYMBOL(generic_perform_write);
2584 * __generic_file_write_iter - write data to a file
2585 * @iocb: IO state structure (file, offset, etc.)
2586 * @from: iov_iter with data to write
2588 * This function does all the work needed for actually writing data to a
2589 * file. It does all basic checks, removes SUID from the file, updates
2590 * modification times and calls proper subroutines depending on whether we
2591 * do direct IO or a standard buffered write.
2593 * It expects i_mutex to be grabbed unless we work on a block device or similar
2594 * object which does not need locking at all.
2596 * This function does *not* take care of syncing data in case of O_SYNC write.
2597 * A caller has to handle it. This is mainly due to the fact that we want to
2598 * avoid syncing under i_mutex.
2600 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2602 struct file *file = iocb->ki_filp;
2603 struct address_space * mapping = file->f_mapping;
2604 struct inode *inode = mapping->host;
2605 ssize_t written = 0;
2609 /* We can write back this queue in page reclaim */
2610 current->backing_dev_info = inode_to_bdi(inode);
2611 err = file_remove_privs(file);
2615 err = file_update_time(file);
2619 if (iocb->ki_flags & IOCB_DIRECT) {
2620 loff_t pos, endbyte;
2622 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
2624 * If the write stopped short of completing, fall back to
2625 * buffered writes. Some filesystems do this for writes to
2626 * holes, for example. For DAX files, a buffered write will
2627 * not succeed (even if it did, DAX does not handle dirty
2628 * page-cache pages correctly).
2630 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2633 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2635 * If generic_perform_write() returned a synchronous error
2636 * then we want to return the number of bytes which were
2637 * direct-written, or the error code if that was zero. Note
2638 * that this differs from normal direct-io semantics, which
2639 * will return -EFOO even if some bytes were written.
2641 if (unlikely(status < 0)) {
2646 * We need to ensure that the page cache pages are written to
2647 * disk and invalidated to preserve the expected O_DIRECT
2650 endbyte = pos + status - 1;
2651 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2653 iocb->ki_pos = endbyte + 1;
2655 invalidate_mapping_pages(mapping,
2656 pos >> PAGE_CACHE_SHIFT,
2657 endbyte >> PAGE_CACHE_SHIFT);
2660 * We don't know how much we wrote, so just return
2661 * the number of bytes which were direct-written
2665 written = generic_perform_write(file, from, iocb->ki_pos);
2666 if (likely(written > 0))
2667 iocb->ki_pos += written;
2670 current->backing_dev_info = NULL;
2671 return written ? written : err;
2673 EXPORT_SYMBOL(__generic_file_write_iter);
2676 * generic_file_write_iter - write data to a file
2677 * @iocb: IO state structure
2678 * @from: iov_iter with data to write
2680 * This is a wrapper around __generic_file_write_iter() to be used by most
2681 * filesystems. It takes care of syncing the file in case of O_SYNC file
2682 * and acquires i_mutex as needed.
2684 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2686 struct file *file = iocb->ki_filp;
2687 struct inode *inode = file->f_mapping->host;
2690 mutex_lock(&inode->i_mutex);
2691 ret = generic_write_checks(iocb, from);
2693 ret = __generic_file_write_iter(iocb, from);
2694 mutex_unlock(&inode->i_mutex);
2699 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2705 EXPORT_SYMBOL(generic_file_write_iter);
2708 * try_to_release_page() - release old fs-specific metadata on a page
2710 * @page: the page which the kernel is trying to free
2711 * @gfp_mask: memory allocation flags (and I/O mode)
2713 * The address_space is to try to release any data against the page
2714 * (presumably at page->private). If the release was successful, return `1'.
2715 * Otherwise return zero.
2717 * This may also be called if PG_fscache is set on a page, indicating that the
2718 * page is known to the local caching routines.
2720 * The @gfp_mask argument specifies whether I/O may be performed to release
2721 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2724 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2726 struct address_space * const mapping = page->mapping;
2728 BUG_ON(!PageLocked(page));
2729 if (PageWriteback(page))
2732 if (mapping && mapping->a_ops->releasepage)
2733 return mapping->a_ops->releasepage(page, gfp_mask);
2734 return try_to_free_buffers(page);
2737 EXPORT_SYMBOL(try_to_release_page);