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 * 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 local_lock(workingset_shadow_lock);
171 list_lru_add(&__workingset_shadow_nodes, &node->private_list);
172 local_unlock(workingset_shadow_lock);
177 * Delete a page from the page cache and free it. Caller has to make
178 * sure the page is locked and that nobody else uses it - or that usage
179 * is safe. The caller must hold the mapping's tree_lock.
181 void __delete_from_page_cache(struct page *page, void *shadow)
183 struct address_space *mapping = page->mapping;
185 trace_mm_filemap_delete_from_page_cache(page);
187 * if we're uptodate, flush out into the cleancache, otherwise
188 * invalidate any existing cleancache entries. We can't leave
189 * stale data around in the cleancache once our page is gone
191 if (PageUptodate(page) && PageMappedToDisk(page))
192 cleancache_put_page(page);
194 cleancache_invalidate_page(mapping, page);
196 page_cache_tree_delete(mapping, page, shadow);
198 page->mapping = NULL;
199 /* Leave page->index set: truncation lookup relies upon it */
201 __dec_zone_page_state(page, NR_FILE_PAGES);
202 if (PageSwapBacked(page))
203 __dec_zone_page_state(page, NR_SHMEM);
204 BUG_ON(page_mapped(page));
207 * At this point page must be either written or cleaned by truncate.
208 * Dirty page here signals a bug and loss of unwritten data.
210 * This fixes dirty accounting after removing the page entirely but
211 * leaves PageDirty set: it has no effect for truncated page and
212 * anyway will be cleared before returning page into buddy allocator.
214 if (WARN_ON_ONCE(PageDirty(page)))
215 account_page_cleaned(page, mapping);
219 * delete_from_page_cache - delete page from page cache
220 * @page: the page which the kernel is trying to remove from page cache
222 * This must be called only on pages that have been verified to be in the page
223 * cache and locked. It will never put the page into the free list, the caller
224 * has a reference on the page.
226 void delete_from_page_cache(struct page *page)
228 struct address_space *mapping = page->mapping;
229 void (*freepage)(struct page *);
231 BUG_ON(!PageLocked(page));
233 freepage = mapping->a_ops->freepage;
234 spin_lock_irq(&mapping->tree_lock);
235 __delete_from_page_cache(page, NULL);
236 spin_unlock_irq(&mapping->tree_lock);
240 page_cache_release(page);
242 EXPORT_SYMBOL(delete_from_page_cache);
244 static int filemap_check_errors(struct address_space *mapping)
247 /* Check for outstanding write errors */
248 if (test_bit(AS_ENOSPC, &mapping->flags) &&
249 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
251 if (test_bit(AS_EIO, &mapping->flags) &&
252 test_and_clear_bit(AS_EIO, &mapping->flags))
258 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
259 * @mapping: address space structure to write
260 * @start: offset in bytes where the range starts
261 * @end: offset in bytes where the range ends (inclusive)
262 * @sync_mode: enable synchronous operation
264 * Start writeback against all of a mapping's dirty pages that lie
265 * within the byte offsets <start, end> inclusive.
267 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
268 * opposed to a regular memory cleansing writeback. The difference between
269 * these two operations is that if a dirty page/buffer is encountered, it must
270 * be waited upon, and not just skipped over.
272 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
273 loff_t end, int sync_mode)
276 struct writeback_control wbc = {
277 .sync_mode = sync_mode,
278 .nr_to_write = LONG_MAX,
279 .range_start = start,
283 if (!mapping_cap_writeback_dirty(mapping))
286 ret = do_writepages(mapping, &wbc);
290 static inline int __filemap_fdatawrite(struct address_space *mapping,
293 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
296 int filemap_fdatawrite(struct address_space *mapping)
298 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
300 EXPORT_SYMBOL(filemap_fdatawrite);
302 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
305 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
307 EXPORT_SYMBOL(filemap_fdatawrite_range);
310 * filemap_flush - mostly a non-blocking flush
311 * @mapping: target address_space
313 * This is a mostly non-blocking flush. Not suitable for data-integrity
314 * purposes - I/O may not be started against all dirty pages.
316 int filemap_flush(struct address_space *mapping)
318 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
320 EXPORT_SYMBOL(filemap_flush);
323 * filemap_fdatawait_range - wait for writeback to complete
324 * @mapping: address space structure to wait for
325 * @start_byte: offset in bytes where the range starts
326 * @end_byte: offset in bytes where the range ends (inclusive)
328 * Walk the list of under-writeback pages of the given address space
329 * in the given range and wait for all of them.
331 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
334 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
335 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
340 if (end_byte < start_byte)
343 pagevec_init(&pvec, 0);
344 while ((index <= end) &&
345 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
346 PAGECACHE_TAG_WRITEBACK,
347 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
350 for (i = 0; i < nr_pages; i++) {
351 struct page *page = pvec.pages[i];
353 /* until radix tree lookup accepts end_index */
354 if (page->index > end)
357 wait_on_page_writeback(page);
358 if (TestClearPageError(page))
361 pagevec_release(&pvec);
365 ret2 = filemap_check_errors(mapping);
371 EXPORT_SYMBOL(filemap_fdatawait_range);
374 * filemap_fdatawait - wait for all under-writeback pages to complete
375 * @mapping: address space structure to wait for
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
380 int filemap_fdatawait(struct address_space *mapping)
382 loff_t i_size = i_size_read(mapping->host);
387 return filemap_fdatawait_range(mapping, 0, i_size - 1);
389 EXPORT_SYMBOL(filemap_fdatawait);
391 int filemap_write_and_wait(struct address_space *mapping)
395 if (mapping->nrpages) {
396 err = filemap_fdatawrite(mapping);
398 * Even if the above returned error, the pages may be
399 * written partially (e.g. -ENOSPC), so we wait for it.
400 * But the -EIO is special case, it may indicate the worst
401 * thing (e.g. bug) happened, so we avoid waiting for it.
404 int err2 = filemap_fdatawait(mapping);
409 err = filemap_check_errors(mapping);
413 EXPORT_SYMBOL(filemap_write_and_wait);
416 * filemap_write_and_wait_range - write out & wait on a file range
417 * @mapping: the address_space for the pages
418 * @lstart: offset in bytes where the range starts
419 * @lend: offset in bytes where the range ends (inclusive)
421 * Write out and wait upon file offsets lstart->lend, inclusive.
423 * Note that `lend' is inclusive (describes the last byte to be written) so
424 * that this function can be used to write to the very end-of-file (end = -1).
426 int filemap_write_and_wait_range(struct address_space *mapping,
427 loff_t lstart, loff_t lend)
431 if (mapping->nrpages) {
432 err = __filemap_fdatawrite_range(mapping, lstart, lend,
434 /* See comment of filemap_write_and_wait() */
436 int err2 = filemap_fdatawait_range(mapping,
442 err = filemap_check_errors(mapping);
446 EXPORT_SYMBOL(filemap_write_and_wait_range);
449 * replace_page_cache_page - replace a pagecache page with a new one
450 * @old: page to be replaced
451 * @new: page to replace with
452 * @gfp_mask: allocation mode
454 * This function replaces a page in the pagecache with a new one. On
455 * success it acquires the pagecache reference for the new page and
456 * drops it for the old page. Both the old and new pages must be
457 * locked. This function does not add the new page to the LRU, the
458 * caller must do that.
460 * The remove + add is atomic. The only way this function can fail is
461 * memory allocation failure.
463 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
467 VM_BUG_ON_PAGE(!PageLocked(old), old);
468 VM_BUG_ON_PAGE(!PageLocked(new), new);
469 VM_BUG_ON_PAGE(new->mapping, new);
471 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
473 struct address_space *mapping = old->mapping;
474 void (*freepage)(struct page *);
476 pgoff_t offset = old->index;
477 freepage = mapping->a_ops->freepage;
480 new->mapping = mapping;
483 spin_lock_irq(&mapping->tree_lock);
484 __delete_from_page_cache(old, NULL);
485 error = radix_tree_insert(&mapping->page_tree, offset, new);
488 __inc_zone_page_state(new, NR_FILE_PAGES);
489 if (PageSwapBacked(new))
490 __inc_zone_page_state(new, NR_SHMEM);
491 spin_unlock_irq(&mapping->tree_lock);
492 mem_cgroup_migrate(old, new, true);
493 radix_tree_preload_end();
496 page_cache_release(old);
501 EXPORT_SYMBOL_GPL(replace_page_cache_page);
503 static int page_cache_tree_insert(struct address_space *mapping,
504 struct page *page, void **shadowp)
506 struct radix_tree_node *node;
510 error = __radix_tree_create(&mapping->page_tree, page->index,
517 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
518 if (!radix_tree_exceptional_entry(p))
522 mapping->nrshadows--;
524 workingset_node_shadows_dec(node);
526 radix_tree_replace_slot(slot, page);
529 workingset_node_pages_inc(node);
531 * Don't track node that contains actual pages.
533 * Avoid acquiring the list_lru lock if already
534 * untracked. The list_empty() test is safe as
535 * node->private_list is protected by
536 * mapping->tree_lock.
538 if (!list_empty(&node->private_list)) {
539 local_lock(workingset_shadow_lock);
540 list_lru_del(&__workingset_shadow_nodes,
541 &node->private_list);
542 local_unlock(workingset_shadow_lock);
548 static int __add_to_page_cache_locked(struct page *page,
549 struct address_space *mapping,
550 pgoff_t offset, gfp_t gfp_mask,
553 int huge = PageHuge(page);
554 struct mem_cgroup *memcg;
557 VM_BUG_ON_PAGE(!PageLocked(page), page);
558 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
561 error = mem_cgroup_try_charge(page, current->mm,
567 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
570 mem_cgroup_cancel_charge(page, memcg);
574 page_cache_get(page);
575 page->mapping = mapping;
576 page->index = offset;
578 spin_lock_irq(&mapping->tree_lock);
579 error = page_cache_tree_insert(mapping, page, shadowp);
580 radix_tree_preload_end();
583 __inc_zone_page_state(page, NR_FILE_PAGES);
584 spin_unlock_irq(&mapping->tree_lock);
586 mem_cgroup_commit_charge(page, memcg, false);
587 trace_mm_filemap_add_to_page_cache(page);
590 page->mapping = NULL;
591 /* Leave page->index set: truncation relies upon it */
592 spin_unlock_irq(&mapping->tree_lock);
594 mem_cgroup_cancel_charge(page, memcg);
595 page_cache_release(page);
600 * add_to_page_cache_locked - add a locked page to the pagecache
602 * @mapping: the page's address_space
603 * @offset: page index
604 * @gfp_mask: page allocation mode
606 * This function is used to add a page to the pagecache. It must be locked.
607 * This function does not add the page to the LRU. The caller must do that.
609 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
610 pgoff_t offset, gfp_t gfp_mask)
612 return __add_to_page_cache_locked(page, mapping, offset,
615 EXPORT_SYMBOL(add_to_page_cache_locked);
617 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
618 pgoff_t offset, gfp_t gfp_mask)
623 __set_page_locked(page);
624 ret = __add_to_page_cache_locked(page, mapping, offset,
627 __clear_page_locked(page);
630 * The page might have been evicted from cache only
631 * recently, in which case it should be activated like
632 * any other repeatedly accessed page.
634 if (shadow && workingset_refault(shadow)) {
636 workingset_activation(page);
638 ClearPageActive(page);
643 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
646 struct page *__page_cache_alloc(gfp_t gfp)
651 if (cpuset_do_page_mem_spread()) {
652 unsigned int cpuset_mems_cookie;
654 cpuset_mems_cookie = read_mems_allowed_begin();
655 n = cpuset_mem_spread_node();
656 page = alloc_pages_exact_node(n, gfp, 0);
657 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
661 return alloc_pages(gfp, 0);
663 EXPORT_SYMBOL(__page_cache_alloc);
667 * In order to wait for pages to become available there must be
668 * waitqueues associated with pages. By using a hash table of
669 * waitqueues where the bucket discipline is to maintain all
670 * waiters on the same queue and wake all when any of the pages
671 * become available, and for the woken contexts to check to be
672 * sure the appropriate page became available, this saves space
673 * at a cost of "thundering herd" phenomena during rare hash
676 wait_queue_head_t *page_waitqueue(struct page *page)
678 const struct zone *zone = page_zone(page);
680 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
682 EXPORT_SYMBOL(page_waitqueue);
684 void wait_on_page_bit(struct page *page, int bit_nr)
686 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
688 if (test_bit(bit_nr, &page->flags))
689 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
690 TASK_UNINTERRUPTIBLE);
692 EXPORT_SYMBOL(wait_on_page_bit);
694 int wait_on_page_bit_killable(struct page *page, int bit_nr)
696 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
698 if (!test_bit(bit_nr, &page->flags))
701 return __wait_on_bit(page_waitqueue(page), &wait,
702 bit_wait_io, TASK_KILLABLE);
705 int wait_on_page_bit_killable_timeout(struct page *page,
706 int bit_nr, unsigned long timeout)
708 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
710 wait.key.timeout = jiffies + timeout;
711 if (!test_bit(bit_nr, &page->flags))
713 return __wait_on_bit(page_waitqueue(page), &wait,
714 bit_wait_io_timeout, TASK_KILLABLE);
716 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
719 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
720 * @page: Page defining the wait queue of interest
721 * @waiter: Waiter to add to the queue
723 * Add an arbitrary @waiter to the wait queue for the nominated @page.
725 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
727 wait_queue_head_t *q = page_waitqueue(page);
730 spin_lock_irqsave(&q->lock, flags);
731 __add_wait_queue(q, waiter);
732 spin_unlock_irqrestore(&q->lock, flags);
734 EXPORT_SYMBOL_GPL(add_page_wait_queue);
737 * unlock_page - unlock a locked page
740 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
741 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
742 * mechanism between PageLocked pages and PageWriteback pages is shared.
743 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
745 * The mb is necessary to enforce ordering between the clear_bit and the read
746 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
748 void unlock_page(struct page *page)
750 VM_BUG_ON_PAGE(!PageLocked(page), page);
751 clear_bit_unlock(PG_locked, &page->flags);
752 smp_mb__after_atomic();
753 wake_up_page(page, PG_locked);
755 EXPORT_SYMBOL(unlock_page);
758 * end_page_writeback - end writeback against a page
761 void end_page_writeback(struct page *page)
764 * TestClearPageReclaim could be used here but it is an atomic
765 * operation and overkill in this particular case. Failing to
766 * shuffle a page marked for immediate reclaim is too mild to
767 * justify taking an atomic operation penalty at the end of
768 * ever page writeback.
770 if (PageReclaim(page)) {
771 ClearPageReclaim(page);
772 rotate_reclaimable_page(page);
775 if (!test_clear_page_writeback(page))
778 smp_mb__after_atomic();
779 wake_up_page(page, PG_writeback);
781 EXPORT_SYMBOL(end_page_writeback);
784 * After completing I/O on a page, call this routine to update the page
785 * flags appropriately
787 void page_endio(struct page *page, int rw, int err)
791 SetPageUptodate(page);
793 ClearPageUptodate(page);
797 } else { /* rw == WRITE */
801 mapping_set_error(page->mapping, err);
803 end_page_writeback(page);
806 EXPORT_SYMBOL_GPL(page_endio);
809 * __lock_page - get a lock on the page, assuming we need to sleep to get it
810 * @page: the page to lock
812 void __lock_page(struct page *page)
814 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
816 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
817 TASK_UNINTERRUPTIBLE);
819 EXPORT_SYMBOL(__lock_page);
821 int __lock_page_killable(struct page *page)
823 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
825 return __wait_on_bit_lock(page_waitqueue(page), &wait,
826 bit_wait_io, TASK_KILLABLE);
828 EXPORT_SYMBOL_GPL(__lock_page_killable);
832 * 1 - page is locked; mmap_sem is still held.
833 * 0 - page is not locked.
834 * mmap_sem has been released (up_read()), unless flags had both
835 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
836 * which case mmap_sem is still held.
838 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
839 * with the page locked and the mmap_sem unperturbed.
841 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
844 if (flags & FAULT_FLAG_ALLOW_RETRY) {
846 * CAUTION! In this case, mmap_sem is not released
847 * even though return 0.
849 if (flags & FAULT_FLAG_RETRY_NOWAIT)
852 up_read(&mm->mmap_sem);
853 if (flags & FAULT_FLAG_KILLABLE)
854 wait_on_page_locked_killable(page);
856 wait_on_page_locked(page);
859 if (flags & FAULT_FLAG_KILLABLE) {
862 ret = __lock_page_killable(page);
864 up_read(&mm->mmap_sem);
874 * page_cache_next_hole - find the next hole (not-present entry)
877 * @max_scan: maximum range to search
879 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
880 * lowest indexed hole.
882 * Returns: the index of the hole if found, otherwise returns an index
883 * outside of the set specified (in which case 'return - index >=
884 * max_scan' will be true). In rare cases of index wrap-around, 0 will
887 * page_cache_next_hole may be called under rcu_read_lock. However,
888 * like radix_tree_gang_lookup, this will not atomically search a
889 * snapshot of the tree at a single point in time. For example, if a
890 * hole is created at index 5, then subsequently a hole is created at
891 * index 10, page_cache_next_hole covering both indexes may return 10
892 * if called under rcu_read_lock.
894 pgoff_t page_cache_next_hole(struct address_space *mapping,
895 pgoff_t index, unsigned long max_scan)
899 for (i = 0; i < max_scan; i++) {
902 page = radix_tree_lookup(&mapping->page_tree, index);
903 if (!page || radix_tree_exceptional_entry(page))
912 EXPORT_SYMBOL(page_cache_next_hole);
915 * page_cache_prev_hole - find the prev hole (not-present entry)
918 * @max_scan: maximum range to search
920 * Search backwards in the range [max(index-max_scan+1, 0), index] for
923 * Returns: the index of the hole if found, otherwise returns an index
924 * outside of the set specified (in which case 'index - return >=
925 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
928 * page_cache_prev_hole may be called under rcu_read_lock. However,
929 * like radix_tree_gang_lookup, this will not atomically search a
930 * snapshot of the tree at a single point in time. For example, if a
931 * hole is created at index 10, then subsequently a hole is created at
932 * index 5, page_cache_prev_hole covering both indexes may return 5 if
933 * called under rcu_read_lock.
935 pgoff_t page_cache_prev_hole(struct address_space *mapping,
936 pgoff_t index, unsigned long max_scan)
940 for (i = 0; i < max_scan; i++) {
943 page = radix_tree_lookup(&mapping->page_tree, index);
944 if (!page || radix_tree_exceptional_entry(page))
947 if (index == ULONG_MAX)
953 EXPORT_SYMBOL(page_cache_prev_hole);
956 * find_get_entry - find and get a page cache entry
957 * @mapping: the address_space to search
958 * @offset: the page cache index
960 * Looks up the page cache slot at @mapping & @offset. If there is a
961 * page cache page, it is returned with an increased refcount.
963 * If the slot holds a shadow entry of a previously evicted page, or a
964 * swap entry from shmem/tmpfs, it is returned.
966 * Otherwise, %NULL is returned.
968 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
976 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
978 page = radix_tree_deref_slot(pagep);
981 if (radix_tree_exception(page)) {
982 if (radix_tree_deref_retry(page))
985 * A shadow entry of a recently evicted page,
986 * or a swap entry from shmem/tmpfs. Return
987 * it without attempting to raise page count.
991 if (!page_cache_get_speculative(page))
995 * Has the page moved?
996 * This is part of the lockless pagecache protocol. See
997 * include/linux/pagemap.h for details.
999 if (unlikely(page != *pagep)) {
1000 page_cache_release(page);
1009 EXPORT_SYMBOL(find_get_entry);
1012 * find_lock_entry - locate, pin and lock a page cache entry
1013 * @mapping: the address_space to search
1014 * @offset: the page cache index
1016 * Looks up the page cache slot at @mapping & @offset. If there is a
1017 * page cache page, it is returned locked and with an increased
1020 * If the slot holds a shadow entry of a previously evicted page, or a
1021 * swap entry from shmem/tmpfs, it is returned.
1023 * Otherwise, %NULL is returned.
1025 * find_lock_entry() may sleep.
1027 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1032 page = find_get_entry(mapping, offset);
1033 if (page && !radix_tree_exception(page)) {
1035 /* Has the page been truncated? */
1036 if (unlikely(page->mapping != mapping)) {
1038 page_cache_release(page);
1041 VM_BUG_ON_PAGE(page->index != offset, page);
1045 EXPORT_SYMBOL(find_lock_entry);
1048 * pagecache_get_page - find and get a page reference
1049 * @mapping: the address_space to search
1050 * @offset: the page index
1051 * @fgp_flags: PCG flags
1052 * @gfp_mask: gfp mask to use for the page cache data page allocation
1054 * Looks up the page cache slot at @mapping & @offset.
1056 * PCG flags modify how the page is returned.
1058 * FGP_ACCESSED: the page will be marked accessed
1059 * FGP_LOCK: Page is return locked
1060 * FGP_CREAT: If page is not present then a new page is allocated using
1061 * @gfp_mask and added to the page cache and the VM's LRU
1062 * list. The page is returned locked and with an increased
1063 * refcount. Otherwise, %NULL is returned.
1065 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1066 * if the GFP flags specified for FGP_CREAT are atomic.
1068 * If there is a page cache page, it is returned with an increased refcount.
1070 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1071 int fgp_flags, gfp_t gfp_mask)
1076 page = find_get_entry(mapping, offset);
1077 if (radix_tree_exceptional_entry(page))
1082 if (fgp_flags & FGP_LOCK) {
1083 if (fgp_flags & FGP_NOWAIT) {
1084 if (!trylock_page(page)) {
1085 page_cache_release(page);
1092 /* Has the page been truncated? */
1093 if (unlikely(page->mapping != mapping)) {
1095 page_cache_release(page);
1098 VM_BUG_ON_PAGE(page->index != offset, page);
1101 if (page && (fgp_flags & FGP_ACCESSED))
1102 mark_page_accessed(page);
1105 if (!page && (fgp_flags & FGP_CREAT)) {
1107 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1108 gfp_mask |= __GFP_WRITE;
1109 if (fgp_flags & FGP_NOFS)
1110 gfp_mask &= ~__GFP_FS;
1112 page = __page_cache_alloc(gfp_mask);
1116 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1117 fgp_flags |= FGP_LOCK;
1119 /* Init accessed so avoid atomic mark_page_accessed later */
1120 if (fgp_flags & FGP_ACCESSED)
1121 __SetPageReferenced(page);
1123 err = add_to_page_cache_lru(page, mapping, offset,
1124 gfp_mask & GFP_RECLAIM_MASK);
1125 if (unlikely(err)) {
1126 page_cache_release(page);
1135 EXPORT_SYMBOL(pagecache_get_page);
1138 * find_get_entries - gang pagecache lookup
1139 * @mapping: The address_space to search
1140 * @start: The starting page cache index
1141 * @nr_entries: The maximum number of entries
1142 * @entries: Where the resulting entries are placed
1143 * @indices: The cache indices corresponding to the entries in @entries
1145 * find_get_entries() will search for and return a group of up to
1146 * @nr_entries entries in the mapping. The entries are placed at
1147 * @entries. find_get_entries() takes a reference against any actual
1150 * The search returns a group of mapping-contiguous page cache entries
1151 * with ascending indexes. There may be holes in the indices due to
1152 * not-present pages.
1154 * Any shadow entries of evicted pages, or swap entries from
1155 * shmem/tmpfs, are included in the returned array.
1157 * find_get_entries() returns the number of pages and shadow entries
1160 unsigned find_get_entries(struct address_space *mapping,
1161 pgoff_t start, unsigned int nr_entries,
1162 struct page **entries, pgoff_t *indices)
1165 unsigned int ret = 0;
1166 struct radix_tree_iter iter;
1173 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1176 page = radix_tree_deref_slot(slot);
1177 if (unlikely(!page))
1179 if (radix_tree_exception(page)) {
1180 if (radix_tree_deref_retry(page))
1183 * A shadow entry of a recently evicted page,
1184 * or a swap entry from shmem/tmpfs. Return
1185 * it without attempting to raise page count.
1189 if (!page_cache_get_speculative(page))
1192 /* Has the page moved? */
1193 if (unlikely(page != *slot)) {
1194 page_cache_release(page);
1198 indices[ret] = iter.index;
1199 entries[ret] = page;
1200 if (++ret == nr_entries)
1208 * find_get_pages - gang pagecache lookup
1209 * @mapping: The address_space to search
1210 * @start: The starting page index
1211 * @nr_pages: The maximum number of pages
1212 * @pages: Where the resulting pages are placed
1214 * find_get_pages() will search for and return a group of up to
1215 * @nr_pages pages in the mapping. The pages are placed at @pages.
1216 * find_get_pages() takes a reference against the returned pages.
1218 * The search returns a group of mapping-contiguous pages with ascending
1219 * indexes. There may be holes in the indices due to not-present pages.
1221 * find_get_pages() returns the number of pages which were found.
1223 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1224 unsigned int nr_pages, struct page **pages)
1226 struct radix_tree_iter iter;
1230 if (unlikely(!nr_pages))
1235 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1238 page = radix_tree_deref_slot(slot);
1239 if (unlikely(!page))
1242 if (radix_tree_exception(page)) {
1243 if (radix_tree_deref_retry(page)) {
1245 * Transient condition which can only trigger
1246 * when entry at index 0 moves out of or back
1247 * to root: none yet gotten, safe to restart.
1249 WARN_ON(iter.index);
1253 * A shadow entry of a recently evicted page,
1254 * or a swap entry from shmem/tmpfs. Skip
1260 if (!page_cache_get_speculative(page))
1263 /* Has the page moved? */
1264 if (unlikely(page != *slot)) {
1265 page_cache_release(page);
1270 if (++ret == nr_pages)
1279 * find_get_pages_contig - gang contiguous pagecache lookup
1280 * @mapping: The address_space to search
1281 * @index: The starting page index
1282 * @nr_pages: The maximum number of pages
1283 * @pages: Where the resulting pages are placed
1285 * find_get_pages_contig() works exactly like find_get_pages(), except
1286 * that the returned number of pages are guaranteed to be contiguous.
1288 * find_get_pages_contig() returns the number of pages which were found.
1290 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1291 unsigned int nr_pages, struct page **pages)
1293 struct radix_tree_iter iter;
1295 unsigned int ret = 0;
1297 if (unlikely(!nr_pages))
1302 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1305 page = radix_tree_deref_slot(slot);
1306 /* The hole, there no reason to continue */
1307 if (unlikely(!page))
1310 if (radix_tree_exception(page)) {
1311 if (radix_tree_deref_retry(page)) {
1313 * Transient condition which can only trigger
1314 * when entry at index 0 moves out of or back
1315 * to root: none yet gotten, safe to restart.
1320 * A shadow entry of a recently evicted page,
1321 * or a swap entry from shmem/tmpfs. Stop
1322 * looking for contiguous pages.
1327 if (!page_cache_get_speculative(page))
1330 /* Has the page moved? */
1331 if (unlikely(page != *slot)) {
1332 page_cache_release(page);
1337 * must check mapping and index after taking the ref.
1338 * otherwise we can get both false positives and false
1339 * negatives, which is just confusing to the caller.
1341 if (page->mapping == NULL || page->index != iter.index) {
1342 page_cache_release(page);
1347 if (++ret == nr_pages)
1353 EXPORT_SYMBOL(find_get_pages_contig);
1356 * find_get_pages_tag - find and return pages that match @tag
1357 * @mapping: the address_space to search
1358 * @index: the starting page index
1359 * @tag: the tag index
1360 * @nr_pages: the maximum number of pages
1361 * @pages: where the resulting pages are placed
1363 * Like find_get_pages, except we only return pages which are tagged with
1364 * @tag. We update @index to index the next page for the traversal.
1366 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1367 int tag, unsigned int nr_pages, struct page **pages)
1369 struct radix_tree_iter iter;
1373 if (unlikely(!nr_pages))
1378 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1379 &iter, *index, tag) {
1382 page = radix_tree_deref_slot(slot);
1383 if (unlikely(!page))
1386 if (radix_tree_exception(page)) {
1387 if (radix_tree_deref_retry(page)) {
1389 * Transient condition which can only trigger
1390 * when entry at index 0 moves out of or back
1391 * to root: none yet gotten, safe to restart.
1396 * A shadow entry of a recently evicted page.
1398 * Those entries should never be tagged, but
1399 * this tree walk is lockless and the tags are
1400 * looked up in bulk, one radix tree node at a
1401 * time, so there is a sizable window for page
1402 * reclaim to evict a page we saw tagged.
1409 if (!page_cache_get_speculative(page))
1412 /* Has the page moved? */
1413 if (unlikely(page != *slot)) {
1414 page_cache_release(page);
1419 if (++ret == nr_pages)
1426 *index = pages[ret - 1]->index + 1;
1430 EXPORT_SYMBOL(find_get_pages_tag);
1433 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1434 * a _large_ part of the i/o request. Imagine the worst scenario:
1436 * ---R__________________________________________B__________
1437 * ^ reading here ^ bad block(assume 4k)
1439 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1440 * => failing the whole request => read(R) => read(R+1) =>
1441 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1442 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1443 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1445 * It is going insane. Fix it by quickly scaling down the readahead size.
1447 static void shrink_readahead_size_eio(struct file *filp,
1448 struct file_ra_state *ra)
1454 * do_generic_file_read - generic file read routine
1455 * @filp: the file to read
1456 * @ppos: current file position
1457 * @iter: data destination
1458 * @written: already copied
1460 * This is a generic file read routine, and uses the
1461 * mapping->a_ops->readpage() function for the actual low-level stuff.
1463 * This is really ugly. But the goto's actually try to clarify some
1464 * of the logic when it comes to error handling etc.
1466 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1467 struct iov_iter *iter, ssize_t written)
1469 struct address_space *mapping = filp->f_mapping;
1470 struct inode *inode = mapping->host;
1471 struct file_ra_state *ra = &filp->f_ra;
1475 unsigned long offset; /* offset into pagecache page */
1476 unsigned int prev_offset;
1479 index = *ppos >> PAGE_CACHE_SHIFT;
1480 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1481 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1482 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1483 offset = *ppos & ~PAGE_CACHE_MASK;
1489 unsigned long nr, ret;
1493 page = find_get_page(mapping, index);
1495 page_cache_sync_readahead(mapping,
1497 index, last_index - index);
1498 page = find_get_page(mapping, index);
1499 if (unlikely(page == NULL))
1500 goto no_cached_page;
1502 if (PageReadahead(page)) {
1503 page_cache_async_readahead(mapping,
1505 index, last_index - index);
1507 if (!PageUptodate(page)) {
1508 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1509 !mapping->a_ops->is_partially_uptodate)
1510 goto page_not_up_to_date;
1511 if (!trylock_page(page))
1512 goto page_not_up_to_date;
1513 /* Did it get truncated before we got the lock? */
1515 goto page_not_up_to_date_locked;
1516 if (!mapping->a_ops->is_partially_uptodate(page,
1517 offset, iter->count))
1518 goto page_not_up_to_date_locked;
1523 * i_size must be checked after we know the page is Uptodate.
1525 * Checking i_size after the check allows us to calculate
1526 * the correct value for "nr", which means the zero-filled
1527 * part of the page is not copied back to userspace (unless
1528 * another truncate extends the file - this is desired though).
1531 isize = i_size_read(inode);
1532 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1533 if (unlikely(!isize || index > end_index)) {
1534 page_cache_release(page);
1538 /* nr is the maximum number of bytes to copy from this page */
1539 nr = PAGE_CACHE_SIZE;
1540 if (index == end_index) {
1541 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1543 page_cache_release(page);
1549 /* If users can be writing to this page using arbitrary
1550 * virtual addresses, take care about potential aliasing
1551 * before reading the page on the kernel side.
1553 if (mapping_writably_mapped(mapping))
1554 flush_dcache_page(page);
1557 * When a sequential read accesses a page several times,
1558 * only mark it as accessed the first time.
1560 if (prev_index != index || offset != prev_offset)
1561 mark_page_accessed(page);
1565 * Ok, we have the page, and it's up-to-date, so
1566 * now we can copy it to user space...
1569 ret = copy_page_to_iter(page, offset, nr, iter);
1571 index += offset >> PAGE_CACHE_SHIFT;
1572 offset &= ~PAGE_CACHE_MASK;
1573 prev_offset = offset;
1575 page_cache_release(page);
1577 if (!iov_iter_count(iter))
1585 page_not_up_to_date:
1586 /* Get exclusive access to the page ... */
1587 error = lock_page_killable(page);
1588 if (unlikely(error))
1589 goto readpage_error;
1591 page_not_up_to_date_locked:
1592 /* Did it get truncated before we got the lock? */
1593 if (!page->mapping) {
1595 page_cache_release(page);
1599 /* Did somebody else fill it already? */
1600 if (PageUptodate(page)) {
1607 * A previous I/O error may have been due to temporary
1608 * failures, eg. multipath errors.
1609 * PG_error will be set again if readpage fails.
1611 ClearPageError(page);
1612 /* Start the actual read. The read will unlock the page. */
1613 error = mapping->a_ops->readpage(filp, page);
1615 if (unlikely(error)) {
1616 if (error == AOP_TRUNCATED_PAGE) {
1617 page_cache_release(page);
1621 goto readpage_error;
1624 if (!PageUptodate(page)) {
1625 error = lock_page_killable(page);
1626 if (unlikely(error))
1627 goto readpage_error;
1628 if (!PageUptodate(page)) {
1629 if (page->mapping == NULL) {
1631 * invalidate_mapping_pages got it
1634 page_cache_release(page);
1638 shrink_readahead_size_eio(filp, ra);
1640 goto readpage_error;
1648 /* UHHUH! A synchronous read error occurred. Report it */
1649 page_cache_release(page);
1654 * Ok, it wasn't cached, so we need to create a new
1657 page = page_cache_alloc_cold(mapping);
1662 error = add_to_page_cache_lru(page, mapping,
1665 page_cache_release(page);
1666 if (error == -EEXIST) {
1676 ra->prev_pos = prev_index;
1677 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1678 ra->prev_pos |= prev_offset;
1680 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1681 file_accessed(filp);
1682 return written ? written : error;
1686 * generic_file_read_iter - generic filesystem read routine
1687 * @iocb: kernel I/O control block
1688 * @iter: destination for the data read
1690 * This is the "read_iter()" routine for all filesystems
1691 * that can use the page cache directly.
1694 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1696 struct file *file = iocb->ki_filp;
1698 loff_t *ppos = &iocb->ki_pos;
1701 if (iocb->ki_flags & IOCB_DIRECT) {
1702 struct address_space *mapping = file->f_mapping;
1703 struct inode *inode = mapping->host;
1704 size_t count = iov_iter_count(iter);
1708 goto out; /* skip atime */
1709 size = i_size_read(inode);
1710 retval = filemap_write_and_wait_range(mapping, pos,
1713 struct iov_iter data = *iter;
1714 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
1718 *ppos = pos + retval;
1719 iov_iter_advance(iter, retval);
1723 * Btrfs can have a short DIO read if we encounter
1724 * compressed extents, so if there was an error, or if
1725 * we've already read everything we wanted to, or if
1726 * there was a short read because we hit EOF, go ahead
1727 * and return. Otherwise fallthrough to buffered io for
1728 * the rest of the read. Buffered reads will not work for
1729 * DAX files, so don't bother trying.
1731 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1733 file_accessed(file);
1738 retval = do_generic_file_read(file, ppos, iter, retval);
1742 EXPORT_SYMBOL(generic_file_read_iter);
1746 * page_cache_read - adds requested page to the page cache if not already there
1747 * @file: file to read
1748 * @offset: page index
1750 * This adds the requested page to the page cache if it isn't already there,
1751 * and schedules an I/O to read in its contents from disk.
1753 static int page_cache_read(struct file *file, pgoff_t offset)
1755 struct address_space *mapping = file->f_mapping;
1760 page = page_cache_alloc_cold(mapping);
1764 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1766 ret = mapping->a_ops->readpage(file, page);
1767 else if (ret == -EEXIST)
1768 ret = 0; /* losing race to add is OK */
1770 page_cache_release(page);
1772 } while (ret == AOP_TRUNCATED_PAGE);
1777 #define MMAP_LOTSAMISS (100)
1780 * Synchronous readahead happens when we don't even find
1781 * a page in the page cache at all.
1783 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1784 struct file_ra_state *ra,
1788 unsigned long ra_pages;
1789 struct address_space *mapping = file->f_mapping;
1791 /* If we don't want any read-ahead, don't bother */
1792 if (vma->vm_flags & VM_RAND_READ)
1797 if (vma->vm_flags & VM_SEQ_READ) {
1798 page_cache_sync_readahead(mapping, ra, file, offset,
1803 /* Avoid banging the cache line if not needed */
1804 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1808 * Do we miss much more than hit in this file? If so,
1809 * stop bothering with read-ahead. It will only hurt.
1811 if (ra->mmap_miss > MMAP_LOTSAMISS)
1817 ra_pages = max_sane_readahead(ra->ra_pages);
1818 ra->start = max_t(long, 0, offset - ra_pages / 2);
1819 ra->size = ra_pages;
1820 ra->async_size = ra_pages / 4;
1821 ra_submit(ra, mapping, file);
1825 * Asynchronous readahead happens when we find the page and PG_readahead,
1826 * so we want to possibly extend the readahead further..
1828 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1829 struct file_ra_state *ra,
1834 struct address_space *mapping = file->f_mapping;
1836 /* If we don't want any read-ahead, don't bother */
1837 if (vma->vm_flags & VM_RAND_READ)
1839 if (ra->mmap_miss > 0)
1841 if (PageReadahead(page))
1842 page_cache_async_readahead(mapping, ra, file,
1843 page, offset, ra->ra_pages);
1847 * filemap_fault - read in file data for page fault handling
1848 * @vma: vma in which the fault was taken
1849 * @vmf: struct vm_fault containing details of the fault
1851 * filemap_fault() is invoked via the vma operations vector for a
1852 * mapped memory region to read in file data during a page fault.
1854 * The goto's are kind of ugly, but this streamlines the normal case of having
1855 * it in the page cache, and handles the special cases reasonably without
1856 * having a lot of duplicated code.
1858 * vma->vm_mm->mmap_sem must be held on entry.
1860 * If our return value has VM_FAULT_RETRY set, it's because
1861 * lock_page_or_retry() returned 0.
1862 * The mmap_sem has usually been released in this case.
1863 * See __lock_page_or_retry() for the exception.
1865 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1866 * has not been released.
1868 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1870 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1873 struct file *file = vma->vm_file;
1874 struct address_space *mapping = file->f_mapping;
1875 struct file_ra_state *ra = &file->f_ra;
1876 struct inode *inode = mapping->host;
1877 pgoff_t offset = vmf->pgoff;
1882 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1883 if (offset >= size >> PAGE_CACHE_SHIFT)
1884 return VM_FAULT_SIGBUS;
1887 * Do we have something in the page cache already?
1889 page = find_get_page(mapping, offset);
1890 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1892 * We found the page, so try async readahead before
1893 * waiting for the lock.
1895 do_async_mmap_readahead(vma, ra, file, page, offset);
1897 /* No page in the page cache at all */
1898 do_sync_mmap_readahead(vma, ra, file, offset);
1899 count_vm_event(PGMAJFAULT);
1900 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1901 ret = VM_FAULT_MAJOR;
1903 page = find_get_page(mapping, offset);
1905 goto no_cached_page;
1908 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1909 page_cache_release(page);
1910 return ret | VM_FAULT_RETRY;
1913 /* Did it get truncated? */
1914 if (unlikely(page->mapping != mapping)) {
1919 VM_BUG_ON_PAGE(page->index != offset, page);
1922 * We have a locked page in the page cache, now we need to check
1923 * that it's up-to-date. If not, it is going to be due to an error.
1925 if (unlikely(!PageUptodate(page)))
1926 goto page_not_uptodate;
1929 * Found the page and have a reference on it.
1930 * We must recheck i_size under page lock.
1932 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1933 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1935 page_cache_release(page);
1936 return VM_FAULT_SIGBUS;
1940 return ret | VM_FAULT_LOCKED;
1944 * We're only likely to ever get here if MADV_RANDOM is in
1947 error = page_cache_read(file, offset);
1950 * The page we want has now been added to the page cache.
1951 * In the unlikely event that someone removed it in the
1952 * meantime, we'll just come back here and read it again.
1958 * An error return from page_cache_read can result if the
1959 * system is low on memory, or a problem occurs while trying
1962 if (error == -ENOMEM)
1963 return VM_FAULT_OOM;
1964 return VM_FAULT_SIGBUS;
1968 * Umm, take care of errors if the page isn't up-to-date.
1969 * Try to re-read it _once_. We do this synchronously,
1970 * because there really aren't any performance issues here
1971 * and we need to check for errors.
1973 ClearPageError(page);
1974 error = mapping->a_ops->readpage(file, page);
1976 wait_on_page_locked(page);
1977 if (!PageUptodate(page))
1980 page_cache_release(page);
1982 if (!error || error == AOP_TRUNCATED_PAGE)
1985 /* Things didn't work out. Return zero to tell the mm layer so. */
1986 shrink_readahead_size_eio(file, ra);
1987 return VM_FAULT_SIGBUS;
1989 EXPORT_SYMBOL(filemap_fault);
1991 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1993 struct radix_tree_iter iter;
1995 struct file *file = vma->vm_file;
1996 struct address_space *mapping = file->f_mapping;
1999 unsigned long address = (unsigned long) vmf->virtual_address;
2004 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2005 if (iter.index > vmf->max_pgoff)
2008 page = radix_tree_deref_slot(slot);
2009 if (unlikely(!page))
2011 if (radix_tree_exception(page)) {
2012 if (radix_tree_deref_retry(page))
2018 if (!page_cache_get_speculative(page))
2021 /* Has the page moved? */
2022 if (unlikely(page != *slot)) {
2023 page_cache_release(page);
2027 if (!PageUptodate(page) ||
2028 PageReadahead(page) ||
2031 if (!trylock_page(page))
2034 if (page->mapping != mapping || !PageUptodate(page))
2037 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2038 if (page->index >= size >> PAGE_CACHE_SHIFT)
2041 pte = vmf->pte + page->index - vmf->pgoff;
2042 if (!pte_none(*pte))
2045 if (file->f_ra.mmap_miss > 0)
2046 file->f_ra.mmap_miss--;
2047 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2048 do_set_pte(vma, addr, page, pte, false, false);
2054 page_cache_release(page);
2056 if (iter.index == vmf->max_pgoff)
2061 EXPORT_SYMBOL(filemap_map_pages);
2063 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2065 struct page *page = vmf->page;
2066 struct inode *inode = file_inode(vma->vm_file);
2067 int ret = VM_FAULT_LOCKED;
2069 sb_start_pagefault(inode->i_sb);
2070 file_update_time(vma->vm_file);
2072 if (page->mapping != inode->i_mapping) {
2074 ret = VM_FAULT_NOPAGE;
2078 * We mark the page dirty already here so that when freeze is in
2079 * progress, we are guaranteed that writeback during freezing will
2080 * see the dirty page and writeprotect it again.
2082 set_page_dirty(page);
2083 wait_for_stable_page(page);
2085 sb_end_pagefault(inode->i_sb);
2088 EXPORT_SYMBOL(filemap_page_mkwrite);
2090 const struct vm_operations_struct generic_file_vm_ops = {
2091 .fault = filemap_fault,
2092 .map_pages = filemap_map_pages,
2093 .page_mkwrite = filemap_page_mkwrite,
2096 /* This is used for a general mmap of a disk file */
2098 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2100 struct address_space *mapping = file->f_mapping;
2102 if (!mapping->a_ops->readpage)
2104 file_accessed(file);
2105 vma->vm_ops = &generic_file_vm_ops;
2110 * This is for filesystems which do not implement ->writepage.
2112 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2114 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2116 return generic_file_mmap(file, vma);
2119 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2123 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2127 #endif /* CONFIG_MMU */
2129 EXPORT_SYMBOL(generic_file_mmap);
2130 EXPORT_SYMBOL(generic_file_readonly_mmap);
2132 static struct page *wait_on_page_read(struct page *page)
2134 if (!IS_ERR(page)) {
2135 wait_on_page_locked(page);
2136 if (!PageUptodate(page)) {
2137 page_cache_release(page);
2138 page = ERR_PTR(-EIO);
2144 static struct page *__read_cache_page(struct address_space *mapping,
2146 int (*filler)(void *, struct page *),
2153 page = find_get_page(mapping, index);
2155 page = __page_cache_alloc(gfp | __GFP_COLD);
2157 return ERR_PTR(-ENOMEM);
2158 err = add_to_page_cache_lru(page, mapping, index, gfp);
2159 if (unlikely(err)) {
2160 page_cache_release(page);
2163 /* Presumably ENOMEM for radix tree node */
2164 return ERR_PTR(err);
2166 err = filler(data, page);
2168 page_cache_release(page);
2169 page = ERR_PTR(err);
2171 page = wait_on_page_read(page);
2177 static struct page *do_read_cache_page(struct address_space *mapping,
2179 int (*filler)(void *, struct page *),
2188 page = __read_cache_page(mapping, index, filler, data, gfp);
2191 if (PageUptodate(page))
2195 if (!page->mapping) {
2197 page_cache_release(page);
2200 if (PageUptodate(page)) {
2204 err = filler(data, page);
2206 page_cache_release(page);
2207 return ERR_PTR(err);
2209 page = wait_on_page_read(page);
2214 mark_page_accessed(page);
2219 * read_cache_page - read into page cache, fill it if needed
2220 * @mapping: the page's address_space
2221 * @index: the page index
2222 * @filler: function to perform the read
2223 * @data: first arg to filler(data, page) function, often left as NULL
2225 * Read into the page cache. If a page already exists, and PageUptodate() is
2226 * not set, try to fill the page and wait for it to become unlocked.
2228 * If the page does not get brought uptodate, return -EIO.
2230 struct page *read_cache_page(struct address_space *mapping,
2232 int (*filler)(void *, struct page *),
2235 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2237 EXPORT_SYMBOL(read_cache_page);
2240 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2241 * @mapping: the page's address_space
2242 * @index: the page index
2243 * @gfp: the page allocator flags to use if allocating
2245 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2246 * any new page allocations done using the specified allocation flags.
2248 * If the page does not get brought uptodate, return -EIO.
2250 struct page *read_cache_page_gfp(struct address_space *mapping,
2254 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2256 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2258 EXPORT_SYMBOL(read_cache_page_gfp);
2261 * Performs necessary checks before doing a write
2263 * Can adjust writing position or amount of bytes to write.
2264 * Returns appropriate error code that caller should return or
2265 * zero in case that write should be allowed.
2267 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2269 struct file *file = iocb->ki_filp;
2270 struct inode *inode = file->f_mapping->host;
2271 unsigned long limit = rlimit(RLIMIT_FSIZE);
2274 if (!iov_iter_count(from))
2277 /* FIXME: this is for backwards compatibility with 2.4 */
2278 if (iocb->ki_flags & IOCB_APPEND)
2279 iocb->ki_pos = i_size_read(inode);
2283 if (limit != RLIM_INFINITY) {
2284 if (iocb->ki_pos >= limit) {
2285 send_sig(SIGXFSZ, current, 0);
2288 iov_iter_truncate(from, limit - (unsigned long)pos);
2294 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2295 !(file->f_flags & O_LARGEFILE))) {
2296 if (pos >= MAX_NON_LFS)
2298 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2302 * Are we about to exceed the fs block limit ?
2304 * If we have written data it becomes a short write. If we have
2305 * exceeded without writing data we send a signal and return EFBIG.
2306 * Linus frestrict idea will clean these up nicely..
2308 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2311 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2312 return iov_iter_count(from);
2314 EXPORT_SYMBOL(generic_write_checks);
2316 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2317 loff_t pos, unsigned len, unsigned flags,
2318 struct page **pagep, void **fsdata)
2320 const struct address_space_operations *aops = mapping->a_ops;
2322 return aops->write_begin(file, mapping, pos, len, flags,
2325 EXPORT_SYMBOL(pagecache_write_begin);
2327 int pagecache_write_end(struct file *file, struct address_space *mapping,
2328 loff_t pos, unsigned len, unsigned copied,
2329 struct page *page, void *fsdata)
2331 const struct address_space_operations *aops = mapping->a_ops;
2333 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2335 EXPORT_SYMBOL(pagecache_write_end);
2338 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2340 struct file *file = iocb->ki_filp;
2341 struct address_space *mapping = file->f_mapping;
2342 struct inode *inode = mapping->host;
2346 struct iov_iter data;
2348 write_len = iov_iter_count(from);
2349 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2351 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2356 * After a write we want buffered reads to be sure to go to disk to get
2357 * the new data. We invalidate clean cached page from the region we're
2358 * about to write. We do this *before* the write so that we can return
2359 * without clobbering -EIOCBQUEUED from ->direct_IO().
2361 if (mapping->nrpages) {
2362 written = invalidate_inode_pages2_range(mapping,
2363 pos >> PAGE_CACHE_SHIFT, end);
2365 * If a page can not be invalidated, return 0 to fall back
2366 * to buffered write.
2369 if (written == -EBUSY)
2376 written = mapping->a_ops->direct_IO(iocb, &data, pos);
2379 * Finally, try again to invalidate clean pages which might have been
2380 * cached by non-direct readahead, or faulted in by get_user_pages()
2381 * if the source of the write was an mmap'ed region of the file
2382 * we're writing. Either one is a pretty crazy thing to do,
2383 * so we don't support it 100%. If this invalidation
2384 * fails, tough, the write still worked...
2386 if (mapping->nrpages) {
2387 invalidate_inode_pages2_range(mapping,
2388 pos >> PAGE_CACHE_SHIFT, end);
2393 iov_iter_advance(from, written);
2394 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2395 i_size_write(inode, pos);
2396 mark_inode_dirty(inode);
2403 EXPORT_SYMBOL(generic_file_direct_write);
2406 * Find or create a page at the given pagecache position. Return the locked
2407 * page. This function is specifically for buffered writes.
2409 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2410 pgoff_t index, unsigned flags)
2413 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2415 if (flags & AOP_FLAG_NOFS)
2416 fgp_flags |= FGP_NOFS;
2418 page = pagecache_get_page(mapping, index, fgp_flags,
2419 mapping_gfp_mask(mapping));
2421 wait_for_stable_page(page);
2425 EXPORT_SYMBOL(grab_cache_page_write_begin);
2427 ssize_t generic_perform_write(struct file *file,
2428 struct iov_iter *i, loff_t pos)
2430 struct address_space *mapping = file->f_mapping;
2431 const struct address_space_operations *a_ops = mapping->a_ops;
2433 ssize_t written = 0;
2434 unsigned int flags = 0;
2437 * Copies from kernel address space cannot fail (NFSD is a big user).
2439 if (!iter_is_iovec(i))
2440 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2444 unsigned long offset; /* Offset into pagecache page */
2445 unsigned long bytes; /* Bytes to write to page */
2446 size_t copied; /* Bytes copied from user */
2449 offset = (pos & (PAGE_CACHE_SIZE - 1));
2450 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2455 * Bring in the user page that we will copy from _first_.
2456 * Otherwise there's a nasty deadlock on copying from the
2457 * same page as we're writing to, without it being marked
2460 * Not only is this an optimisation, but it is also required
2461 * to check that the address is actually valid, when atomic
2462 * usercopies are used, below.
2464 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2469 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2471 if (unlikely(status < 0))
2474 if (mapping_writably_mapped(mapping))
2475 flush_dcache_page(page);
2477 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2478 flush_dcache_page(page);
2480 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2482 if (unlikely(status < 0))
2488 iov_iter_advance(i, copied);
2489 if (unlikely(copied == 0)) {
2491 * If we were unable to copy any data at all, we must
2492 * fall back to a single segment length write.
2494 * If we didn't fallback here, we could livelock
2495 * because not all segments in the iov can be copied at
2496 * once without a pagefault.
2498 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2499 iov_iter_single_seg_count(i));
2505 balance_dirty_pages_ratelimited(mapping);
2506 if (fatal_signal_pending(current)) {
2510 } while (iov_iter_count(i));
2512 return written ? written : status;
2514 EXPORT_SYMBOL(generic_perform_write);
2517 * __generic_file_write_iter - write data to a file
2518 * @iocb: IO state structure (file, offset, etc.)
2519 * @from: iov_iter with data to write
2521 * This function does all the work needed for actually writing data to a
2522 * file. It does all basic checks, removes SUID from the file, updates
2523 * modification times and calls proper subroutines depending on whether we
2524 * do direct IO or a standard buffered write.
2526 * It expects i_mutex to be grabbed unless we work on a block device or similar
2527 * object which does not need locking at all.
2529 * This function does *not* take care of syncing data in case of O_SYNC write.
2530 * A caller has to handle it. This is mainly due to the fact that we want to
2531 * avoid syncing under i_mutex.
2533 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2535 struct file *file = iocb->ki_filp;
2536 struct address_space * mapping = file->f_mapping;
2537 struct inode *inode = mapping->host;
2538 ssize_t written = 0;
2542 /* We can write back this queue in page reclaim */
2543 current->backing_dev_info = inode_to_bdi(inode);
2544 err = file_remove_suid(file);
2548 err = file_update_time(file);
2552 if (iocb->ki_flags & IOCB_DIRECT) {
2553 loff_t pos, endbyte;
2555 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
2557 * If the write stopped short of completing, fall back to
2558 * buffered writes. Some filesystems do this for writes to
2559 * holes, for example. For DAX files, a buffered write will
2560 * not succeed (even if it did, DAX does not handle dirty
2561 * page-cache pages correctly).
2563 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2566 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2568 * If generic_perform_write() returned a synchronous error
2569 * then we want to return the number of bytes which were
2570 * direct-written, or the error code if that was zero. Note
2571 * that this differs from normal direct-io semantics, which
2572 * will return -EFOO even if some bytes were written.
2574 if (unlikely(status < 0)) {
2579 * We need to ensure that the page cache pages are written to
2580 * disk and invalidated to preserve the expected O_DIRECT
2583 endbyte = pos + status - 1;
2584 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2586 iocb->ki_pos = endbyte + 1;
2588 invalidate_mapping_pages(mapping,
2589 pos >> PAGE_CACHE_SHIFT,
2590 endbyte >> PAGE_CACHE_SHIFT);
2593 * We don't know how much we wrote, so just return
2594 * the number of bytes which were direct-written
2598 written = generic_perform_write(file, from, iocb->ki_pos);
2599 if (likely(written > 0))
2600 iocb->ki_pos += written;
2603 current->backing_dev_info = NULL;
2604 return written ? written : err;
2606 EXPORT_SYMBOL(__generic_file_write_iter);
2609 * generic_file_write_iter - write data to a file
2610 * @iocb: IO state structure
2611 * @from: iov_iter with data to write
2613 * This is a wrapper around __generic_file_write_iter() to be used by most
2614 * filesystems. It takes care of syncing the file in case of O_SYNC file
2615 * and acquires i_mutex as needed.
2617 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2619 struct file *file = iocb->ki_filp;
2620 struct inode *inode = file->f_mapping->host;
2623 mutex_lock(&inode->i_mutex);
2624 ret = generic_write_checks(iocb, from);
2626 ret = __generic_file_write_iter(iocb, from);
2627 mutex_unlock(&inode->i_mutex);
2632 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2638 EXPORT_SYMBOL(generic_file_write_iter);
2641 * try_to_release_page() - release old fs-specific metadata on a page
2643 * @page: the page which the kernel is trying to free
2644 * @gfp_mask: memory allocation flags (and I/O mode)
2646 * The address_space is to try to release any data against the page
2647 * (presumably at page->private). If the release was successful, return `1'.
2648 * Otherwise return zero.
2650 * This may also be called if PG_fscache is set on a page, indicating that the
2651 * page is known to the local caching routines.
2653 * The @gfp_mask argument specifies whether I/O may be performed to release
2654 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2657 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2659 struct address_space * const mapping = page->mapping;
2661 BUG_ON(!PageLocked(page));
2662 if (PageWriteback(page))
2665 if (mapping && mapping->a_ops->releasepage)
2666 return mapping->a_ops->releasepage(page, gfp_mask);
2667 return try_to_free_buffers(page);
2670 EXPORT_SYMBOL(try_to_release_page);