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 list_lru_add(&workingset_shadow_nodes, &node->private_list);
175 * Delete a page from the page cache and free it. Caller has to make
176 * sure the page is locked and that nobody else uses it - or that usage
177 * is safe. The caller must hold the mapping's tree_lock.
179 void __delete_from_page_cache(struct page *page, void *shadow)
181 struct address_space *mapping = page->mapping;
183 trace_mm_filemap_delete_from_page_cache(page);
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
189 if (PageUptodate(page) && PageMappedToDisk(page))
190 cleancache_put_page(page);
192 cleancache_invalidate_page(mapping, page);
194 page_cache_tree_delete(mapping, page, shadow);
196 page->mapping = NULL;
197 /* Leave page->index set: truncation lookup relies upon it */
199 __dec_zone_page_state(page, NR_FILE_PAGES);
200 if (PageSwapBacked(page))
201 __dec_zone_page_state(page, NR_SHMEM);
202 BUG_ON(page_mapped(page));
205 * Some filesystems seem to re-dirty the page even after
206 * the VM has canceled the dirty bit (eg ext3 journaling).
208 * Fix it up by doing a final dirty accounting check after
209 * having removed the page entirely.
211 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
212 dec_zone_page_state(page, NR_FILE_DIRTY);
213 dec_bdi_stat(inode_to_bdi(mapping->host), BDI_RECLAIMABLE);
218 * delete_from_page_cache - delete page from page cache
219 * @page: the page which the kernel is trying to remove from page cache
221 * This must be called only on pages that have been verified to be in the page
222 * cache and locked. It will never put the page into the free list, the caller
223 * has a reference on the page.
225 void delete_from_page_cache(struct page *page)
227 struct address_space *mapping = page->mapping;
228 void (*freepage)(struct page *);
230 BUG_ON(!PageLocked(page));
232 freepage = mapping->a_ops->freepage;
233 spin_lock_irq(&mapping->tree_lock);
234 __delete_from_page_cache(page, NULL);
235 spin_unlock_irq(&mapping->tree_lock);
239 page_cache_release(page);
241 EXPORT_SYMBOL(delete_from_page_cache);
243 static int filemap_check_errors(struct address_space *mapping)
246 /* Check for outstanding write errors */
247 if (test_bit(AS_ENOSPC, &mapping->flags) &&
248 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
250 if (test_bit(AS_EIO, &mapping->flags) &&
251 test_and_clear_bit(AS_EIO, &mapping->flags))
257 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
258 * @mapping: address space structure to write
259 * @start: offset in bytes where the range starts
260 * @end: offset in bytes where the range ends (inclusive)
261 * @sync_mode: enable synchronous operation
263 * Start writeback against all of a mapping's dirty pages that lie
264 * within the byte offsets <start, end> inclusive.
266 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
267 * opposed to a regular memory cleansing writeback. The difference between
268 * these two operations is that if a dirty page/buffer is encountered, it must
269 * be waited upon, and not just skipped over.
271 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
272 loff_t end, int sync_mode)
275 struct writeback_control wbc = {
276 .sync_mode = sync_mode,
277 .nr_to_write = LONG_MAX,
278 .range_start = start,
282 if (!mapping_cap_writeback_dirty(mapping))
285 ret = do_writepages(mapping, &wbc);
289 static inline int __filemap_fdatawrite(struct address_space *mapping,
292 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
295 int filemap_fdatawrite(struct address_space *mapping)
297 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
299 EXPORT_SYMBOL(filemap_fdatawrite);
301 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
304 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
306 EXPORT_SYMBOL(filemap_fdatawrite_range);
309 * filemap_flush - mostly a non-blocking flush
310 * @mapping: target address_space
312 * This is a mostly non-blocking flush. Not suitable for data-integrity
313 * purposes - I/O may not be started against all dirty pages.
315 int filemap_flush(struct address_space *mapping)
317 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
319 EXPORT_SYMBOL(filemap_flush);
322 * filemap_fdatawait_range - wait for writeback to complete
323 * @mapping: address space structure to wait for
324 * @start_byte: offset in bytes where the range starts
325 * @end_byte: offset in bytes where the range ends (inclusive)
327 * Walk the list of under-writeback pages of the given address space
328 * in the given range and wait for all of them.
330 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
333 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
334 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
339 if (end_byte < start_byte)
342 pagevec_init(&pvec, 0);
343 while ((index <= end) &&
344 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
345 PAGECACHE_TAG_WRITEBACK,
346 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
349 for (i = 0; i < nr_pages; i++) {
350 struct page *page = pvec.pages[i];
352 /* until radix tree lookup accepts end_index */
353 if (page->index > end)
356 wait_on_page_writeback(page);
357 if (TestClearPageError(page))
360 pagevec_release(&pvec);
364 ret2 = filemap_check_errors(mapping);
370 EXPORT_SYMBOL(filemap_fdatawait_range);
373 * filemap_fdatawait - wait for all under-writeback pages to complete
374 * @mapping: address space structure to wait for
376 * Walk the list of under-writeback pages of the given address space
377 * and wait for all of them.
379 int filemap_fdatawait(struct address_space *mapping)
381 loff_t i_size = i_size_read(mapping->host);
386 return filemap_fdatawait_range(mapping, 0, i_size - 1);
388 EXPORT_SYMBOL(filemap_fdatawait);
390 int filemap_write_and_wait(struct address_space *mapping)
394 if (mapping->nrpages) {
395 err = filemap_fdatawrite(mapping);
397 * Even if the above returned error, the pages may be
398 * written partially (e.g. -ENOSPC), so we wait for it.
399 * But the -EIO is special case, it may indicate the worst
400 * thing (e.g. bug) happened, so we avoid waiting for it.
403 int err2 = filemap_fdatawait(mapping);
408 err = filemap_check_errors(mapping);
412 EXPORT_SYMBOL(filemap_write_and_wait);
415 * filemap_write_and_wait_range - write out & wait on a file range
416 * @mapping: the address_space for the pages
417 * @lstart: offset in bytes where the range starts
418 * @lend: offset in bytes where the range ends (inclusive)
420 * Write out and wait upon file offsets lstart->lend, inclusive.
422 * Note that `lend' is inclusive (describes the last byte to be written) so
423 * that this function can be used to write to the very end-of-file (end = -1).
425 int filemap_write_and_wait_range(struct address_space *mapping,
426 loff_t lstart, loff_t lend)
430 if (mapping->nrpages) {
431 err = __filemap_fdatawrite_range(mapping, lstart, lend,
433 /* See comment of filemap_write_and_wait() */
435 int err2 = filemap_fdatawait_range(mapping,
441 err = filemap_check_errors(mapping);
445 EXPORT_SYMBOL(filemap_write_and_wait_range);
448 * replace_page_cache_page - replace a pagecache page with a new one
449 * @old: page to be replaced
450 * @new: page to replace with
451 * @gfp_mask: allocation mode
453 * This function replaces a page in the pagecache with a new one. On
454 * success it acquires the pagecache reference for the new page and
455 * drops it for the old page. Both the old and new pages must be
456 * locked. This function does not add the new page to the LRU, the
457 * caller must do that.
459 * The remove + add is atomic. The only way this function can fail is
460 * memory allocation failure.
462 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
466 VM_BUG_ON_PAGE(!PageLocked(old), old);
467 VM_BUG_ON_PAGE(!PageLocked(new), new);
468 VM_BUG_ON_PAGE(new->mapping, new);
470 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
472 struct address_space *mapping = old->mapping;
473 void (*freepage)(struct page *);
475 pgoff_t offset = old->index;
476 freepage = mapping->a_ops->freepage;
479 new->mapping = mapping;
482 spin_lock_irq(&mapping->tree_lock);
483 __delete_from_page_cache(old, NULL);
484 error = radix_tree_insert(&mapping->page_tree, offset, new);
487 __inc_zone_page_state(new, NR_FILE_PAGES);
488 if (PageSwapBacked(new))
489 __inc_zone_page_state(new, NR_SHMEM);
490 spin_unlock_irq(&mapping->tree_lock);
491 mem_cgroup_migrate(old, new, true);
492 radix_tree_preload_end();
495 page_cache_release(old);
500 EXPORT_SYMBOL_GPL(replace_page_cache_page);
502 static int page_cache_tree_insert(struct address_space *mapping,
503 struct page *page, void **shadowp)
505 struct radix_tree_node *node;
509 error = __radix_tree_create(&mapping->page_tree, page->index,
516 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
517 if (!radix_tree_exceptional_entry(p))
521 mapping->nrshadows--;
523 workingset_node_shadows_dec(node);
525 radix_tree_replace_slot(slot, page);
528 workingset_node_pages_inc(node);
530 * Don't track node that contains actual pages.
532 * Avoid acquiring the list_lru lock if already
533 * untracked. The list_empty() test is safe as
534 * node->private_list is protected by
535 * mapping->tree_lock.
537 if (!list_empty(&node->private_list))
538 list_lru_del(&workingset_shadow_nodes,
539 &node->private_list);
544 static int __add_to_page_cache_locked(struct page *page,
545 struct address_space *mapping,
546 pgoff_t offset, gfp_t gfp_mask,
549 int huge = PageHuge(page);
550 struct mem_cgroup *memcg;
553 VM_BUG_ON_PAGE(!PageLocked(page), page);
554 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
557 error = mem_cgroup_try_charge(page, current->mm,
563 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
566 mem_cgroup_cancel_charge(page, memcg);
570 page_cache_get(page);
571 page->mapping = mapping;
572 page->index = offset;
574 spin_lock_irq(&mapping->tree_lock);
575 error = page_cache_tree_insert(mapping, page, shadowp);
576 radix_tree_preload_end();
579 __inc_zone_page_state(page, NR_FILE_PAGES);
580 spin_unlock_irq(&mapping->tree_lock);
582 mem_cgroup_commit_charge(page, memcg, false);
583 trace_mm_filemap_add_to_page_cache(page);
586 page->mapping = NULL;
587 /* Leave page->index set: truncation relies upon it */
588 spin_unlock_irq(&mapping->tree_lock);
590 mem_cgroup_cancel_charge(page, memcg);
591 page_cache_release(page);
596 * add_to_page_cache_locked - add a locked page to the pagecache
598 * @mapping: the page's address_space
599 * @offset: page index
600 * @gfp_mask: page allocation mode
602 * This function is used to add a page to the pagecache. It must be locked.
603 * This function does not add the page to the LRU. The caller must do that.
605 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
606 pgoff_t offset, gfp_t gfp_mask)
608 return __add_to_page_cache_locked(page, mapping, offset,
611 EXPORT_SYMBOL(add_to_page_cache_locked);
613 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
614 pgoff_t offset, gfp_t gfp_mask)
619 __set_page_locked(page);
620 ret = __add_to_page_cache_locked(page, mapping, offset,
623 __clear_page_locked(page);
626 * The page might have been evicted from cache only
627 * recently, in which case it should be activated like
628 * any other repeatedly accessed page.
630 if (shadow && workingset_refault(shadow)) {
632 workingset_activation(page);
634 ClearPageActive(page);
639 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
642 struct page *__page_cache_alloc(gfp_t gfp)
647 if (cpuset_do_page_mem_spread()) {
648 unsigned int cpuset_mems_cookie;
650 cpuset_mems_cookie = read_mems_allowed_begin();
651 n = cpuset_mem_spread_node();
652 page = alloc_pages_exact_node(n, gfp, 0);
653 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
657 return alloc_pages(gfp, 0);
659 EXPORT_SYMBOL(__page_cache_alloc);
663 * In order to wait for pages to become available there must be
664 * waitqueues associated with pages. By using a hash table of
665 * waitqueues where the bucket discipline is to maintain all
666 * waiters on the same queue and wake all when any of the pages
667 * become available, and for the woken contexts to check to be
668 * sure the appropriate page became available, this saves space
669 * at a cost of "thundering herd" phenomena during rare hash
672 wait_queue_head_t *page_waitqueue(struct page *page)
674 const struct zone *zone = page_zone(page);
676 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
678 EXPORT_SYMBOL(page_waitqueue);
680 void wait_on_page_bit(struct page *page, int bit_nr)
682 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
684 if (test_bit(bit_nr, &page->flags))
685 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
686 TASK_UNINTERRUPTIBLE);
688 EXPORT_SYMBOL(wait_on_page_bit);
690 int wait_on_page_bit_killable(struct page *page, int bit_nr)
692 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
694 if (!test_bit(bit_nr, &page->flags))
697 return __wait_on_bit(page_waitqueue(page), &wait,
698 bit_wait_io, TASK_KILLABLE);
701 int wait_on_page_bit_killable_timeout(struct page *page,
702 int bit_nr, unsigned long timeout)
704 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
706 wait.key.timeout = jiffies + timeout;
707 if (!test_bit(bit_nr, &page->flags))
709 return __wait_on_bit(page_waitqueue(page), &wait,
710 bit_wait_io_timeout, TASK_KILLABLE);
712 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
715 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
716 * @page: Page defining the wait queue of interest
717 * @waiter: Waiter to add to the queue
719 * Add an arbitrary @waiter to the wait queue for the nominated @page.
721 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
723 wait_queue_head_t *q = page_waitqueue(page);
726 spin_lock_irqsave(&q->lock, flags);
727 __add_wait_queue(q, waiter);
728 spin_unlock_irqrestore(&q->lock, flags);
730 EXPORT_SYMBOL_GPL(add_page_wait_queue);
733 * unlock_page - unlock a locked page
736 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
737 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
738 * mechanism between PageLocked pages and PageWriteback pages is shared.
739 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
741 * The mb is necessary to enforce ordering between the clear_bit and the read
742 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
744 void unlock_page(struct page *page)
746 VM_BUG_ON_PAGE(!PageLocked(page), page);
747 clear_bit_unlock(PG_locked, &page->flags);
748 smp_mb__after_atomic();
749 wake_up_page(page, PG_locked);
751 EXPORT_SYMBOL(unlock_page);
754 * end_page_writeback - end writeback against a page
757 void end_page_writeback(struct page *page)
760 * TestClearPageReclaim could be used here but it is an atomic
761 * operation and overkill in this particular case. Failing to
762 * shuffle a page marked for immediate reclaim is too mild to
763 * justify taking an atomic operation penalty at the end of
764 * ever page writeback.
766 if (PageReclaim(page)) {
767 ClearPageReclaim(page);
768 rotate_reclaimable_page(page);
771 if (!test_clear_page_writeback(page))
774 smp_mb__after_atomic();
775 wake_up_page(page, PG_writeback);
777 EXPORT_SYMBOL(end_page_writeback);
780 * After completing I/O on a page, call this routine to update the page
781 * flags appropriately
783 void page_endio(struct page *page, int rw, int err)
787 SetPageUptodate(page);
789 ClearPageUptodate(page);
793 } else { /* rw == WRITE */
797 mapping_set_error(page->mapping, err);
799 end_page_writeback(page);
802 EXPORT_SYMBOL_GPL(page_endio);
805 * __lock_page - get a lock on the page, assuming we need to sleep to get it
806 * @page: the page to lock
808 void __lock_page(struct page *page)
810 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
812 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
813 TASK_UNINTERRUPTIBLE);
815 EXPORT_SYMBOL(__lock_page);
817 int __lock_page_killable(struct page *page)
819 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
821 return __wait_on_bit_lock(page_waitqueue(page), &wait,
822 bit_wait_io, TASK_KILLABLE);
824 EXPORT_SYMBOL_GPL(__lock_page_killable);
828 * 1 - page is locked; mmap_sem is still held.
829 * 0 - page is not locked.
830 * mmap_sem has been released (up_read()), unless flags had both
831 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
832 * which case mmap_sem is still held.
834 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
835 * with the page locked and the mmap_sem unperturbed.
837 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
840 if (flags & FAULT_FLAG_ALLOW_RETRY) {
842 * CAUTION! In this case, mmap_sem is not released
843 * even though return 0.
845 if (flags & FAULT_FLAG_RETRY_NOWAIT)
848 up_read(&mm->mmap_sem);
849 if (flags & FAULT_FLAG_KILLABLE)
850 wait_on_page_locked_killable(page);
852 wait_on_page_locked(page);
855 if (flags & FAULT_FLAG_KILLABLE) {
858 ret = __lock_page_killable(page);
860 up_read(&mm->mmap_sem);
870 * page_cache_next_hole - find the next hole (not-present entry)
873 * @max_scan: maximum range to search
875 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
876 * lowest indexed hole.
878 * Returns: the index of the hole if found, otherwise returns an index
879 * outside of the set specified (in which case 'return - index >=
880 * max_scan' will be true). In rare cases of index wrap-around, 0 will
883 * page_cache_next_hole may be called under rcu_read_lock. However,
884 * like radix_tree_gang_lookup, this will not atomically search a
885 * snapshot of the tree at a single point in time. For example, if a
886 * hole is created at index 5, then subsequently a hole is created at
887 * index 10, page_cache_next_hole covering both indexes may return 10
888 * if called under rcu_read_lock.
890 pgoff_t page_cache_next_hole(struct address_space *mapping,
891 pgoff_t index, unsigned long max_scan)
895 for (i = 0; i < max_scan; i++) {
898 page = radix_tree_lookup(&mapping->page_tree, index);
899 if (!page || radix_tree_exceptional_entry(page))
908 EXPORT_SYMBOL(page_cache_next_hole);
911 * page_cache_prev_hole - find the prev hole (not-present entry)
914 * @max_scan: maximum range to search
916 * Search backwards in the range [max(index-max_scan+1, 0), index] for
919 * Returns: the index of the hole if found, otherwise returns an index
920 * outside of the set specified (in which case 'index - return >=
921 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
924 * page_cache_prev_hole may be called under rcu_read_lock. However,
925 * like radix_tree_gang_lookup, this will not atomically search a
926 * snapshot of the tree at a single point in time. For example, if a
927 * hole is created at index 10, then subsequently a hole is created at
928 * index 5, page_cache_prev_hole covering both indexes may return 5 if
929 * called under rcu_read_lock.
931 pgoff_t page_cache_prev_hole(struct address_space *mapping,
932 pgoff_t index, unsigned long max_scan)
936 for (i = 0; i < max_scan; i++) {
939 page = radix_tree_lookup(&mapping->page_tree, index);
940 if (!page || radix_tree_exceptional_entry(page))
943 if (index == ULONG_MAX)
949 EXPORT_SYMBOL(page_cache_prev_hole);
952 * find_get_entry - find and get a page cache entry
953 * @mapping: the address_space to search
954 * @offset: the page cache index
956 * Looks up the page cache slot at @mapping & @offset. If there is a
957 * page cache page, it is returned with an increased refcount.
959 * If the slot holds a shadow entry of a previously evicted page, or a
960 * swap entry from shmem/tmpfs, it is returned.
962 * Otherwise, %NULL is returned.
964 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
972 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
974 page = radix_tree_deref_slot(pagep);
977 if (radix_tree_exception(page)) {
978 if (radix_tree_deref_retry(page))
981 * A shadow entry of a recently evicted page,
982 * or a swap entry from shmem/tmpfs. Return
983 * it without attempting to raise page count.
987 if (!page_cache_get_speculative(page))
991 * Has the page moved?
992 * This is part of the lockless pagecache protocol. See
993 * include/linux/pagemap.h for details.
995 if (unlikely(page != *pagep)) {
996 page_cache_release(page);
1005 EXPORT_SYMBOL(find_get_entry);
1008 * find_lock_entry - locate, pin and lock a page cache entry
1009 * @mapping: the address_space to search
1010 * @offset: the page cache index
1012 * Looks up the page cache slot at @mapping & @offset. If there is a
1013 * page cache page, it is returned locked and with an increased
1016 * If the slot holds a shadow entry of a previously evicted page, or a
1017 * swap entry from shmem/tmpfs, it is returned.
1019 * Otherwise, %NULL is returned.
1021 * find_lock_entry() may sleep.
1023 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1028 page = find_get_entry(mapping, offset);
1029 if (page && !radix_tree_exception(page)) {
1031 /* Has the page been truncated? */
1032 if (unlikely(page->mapping != mapping)) {
1034 page_cache_release(page);
1037 VM_BUG_ON_PAGE(page->index != offset, page);
1041 EXPORT_SYMBOL(find_lock_entry);
1044 * pagecache_get_page - find and get a page reference
1045 * @mapping: the address_space to search
1046 * @offset: the page index
1047 * @fgp_flags: PCG flags
1048 * @gfp_mask: gfp mask to use for the page cache data page allocation
1050 * Looks up the page cache slot at @mapping & @offset.
1052 * PCG flags modify how the page is returned.
1054 * FGP_ACCESSED: the page will be marked accessed
1055 * FGP_LOCK: Page is return locked
1056 * FGP_CREAT: If page is not present then a new page is allocated using
1057 * @gfp_mask and added to the page cache and the VM's LRU
1058 * list. The page is returned locked and with an increased
1059 * refcount. Otherwise, %NULL is returned.
1061 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1062 * if the GFP flags specified for FGP_CREAT are atomic.
1064 * If there is a page cache page, it is returned with an increased refcount.
1066 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1067 int fgp_flags, gfp_t gfp_mask)
1072 page = find_get_entry(mapping, offset);
1073 if (radix_tree_exceptional_entry(page))
1078 if (fgp_flags & FGP_LOCK) {
1079 if (fgp_flags & FGP_NOWAIT) {
1080 if (!trylock_page(page)) {
1081 page_cache_release(page);
1088 /* Has the page been truncated? */
1089 if (unlikely(page->mapping != mapping)) {
1091 page_cache_release(page);
1094 VM_BUG_ON_PAGE(page->index != offset, page);
1097 if (page && (fgp_flags & FGP_ACCESSED))
1098 mark_page_accessed(page);
1101 if (!page && (fgp_flags & FGP_CREAT)) {
1103 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1104 gfp_mask |= __GFP_WRITE;
1105 if (fgp_flags & FGP_NOFS)
1106 gfp_mask &= ~__GFP_FS;
1108 page = __page_cache_alloc(gfp_mask);
1112 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1113 fgp_flags |= FGP_LOCK;
1115 /* Init accessed so avoid atomic mark_page_accessed later */
1116 if (fgp_flags & FGP_ACCESSED)
1117 __SetPageReferenced(page);
1119 err = add_to_page_cache_lru(page, mapping, offset,
1120 gfp_mask & GFP_RECLAIM_MASK);
1121 if (unlikely(err)) {
1122 page_cache_release(page);
1131 EXPORT_SYMBOL(pagecache_get_page);
1134 * find_get_entries - gang pagecache lookup
1135 * @mapping: The address_space to search
1136 * @start: The starting page cache index
1137 * @nr_entries: The maximum number of entries
1138 * @entries: Where the resulting entries are placed
1139 * @indices: The cache indices corresponding to the entries in @entries
1141 * find_get_entries() will search for and return a group of up to
1142 * @nr_entries entries in the mapping. The entries are placed at
1143 * @entries. find_get_entries() takes a reference against any actual
1146 * The search returns a group of mapping-contiguous page cache entries
1147 * with ascending indexes. There may be holes in the indices due to
1148 * not-present pages.
1150 * Any shadow entries of evicted pages, or swap entries from
1151 * shmem/tmpfs, are included in the returned array.
1153 * find_get_entries() returns the number of pages and shadow entries
1156 unsigned find_get_entries(struct address_space *mapping,
1157 pgoff_t start, unsigned int nr_entries,
1158 struct page **entries, pgoff_t *indices)
1161 unsigned int ret = 0;
1162 struct radix_tree_iter iter;
1169 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1172 page = radix_tree_deref_slot(slot);
1173 if (unlikely(!page))
1175 if (radix_tree_exception(page)) {
1176 if (radix_tree_deref_retry(page))
1179 * A shadow entry of a recently evicted page,
1180 * or a swap entry from shmem/tmpfs. Return
1181 * it without attempting to raise page count.
1185 if (!page_cache_get_speculative(page))
1188 /* Has the page moved? */
1189 if (unlikely(page != *slot)) {
1190 page_cache_release(page);
1194 indices[ret] = iter.index;
1195 entries[ret] = page;
1196 if (++ret == nr_entries)
1204 * find_get_pages - gang pagecache lookup
1205 * @mapping: The address_space to search
1206 * @start: The starting page index
1207 * @nr_pages: The maximum number of pages
1208 * @pages: Where the resulting pages are placed
1210 * find_get_pages() will search for and return a group of up to
1211 * @nr_pages pages in the mapping. The pages are placed at @pages.
1212 * find_get_pages() takes a reference against the returned pages.
1214 * The search returns a group of mapping-contiguous pages with ascending
1215 * indexes. There may be holes in the indices due to not-present pages.
1217 * find_get_pages() returns the number of pages which were found.
1219 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1220 unsigned int nr_pages, struct page **pages)
1222 struct radix_tree_iter iter;
1226 if (unlikely(!nr_pages))
1231 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1234 page = radix_tree_deref_slot(slot);
1235 if (unlikely(!page))
1238 if (radix_tree_exception(page)) {
1239 if (radix_tree_deref_retry(page)) {
1241 * Transient condition which can only trigger
1242 * when entry at index 0 moves out of or back
1243 * to root: none yet gotten, safe to restart.
1245 WARN_ON(iter.index);
1249 * A shadow entry of a recently evicted page,
1250 * or a swap entry from shmem/tmpfs. Skip
1256 if (!page_cache_get_speculative(page))
1259 /* Has the page moved? */
1260 if (unlikely(page != *slot)) {
1261 page_cache_release(page);
1266 if (++ret == nr_pages)
1275 * find_get_pages_contig - gang contiguous pagecache lookup
1276 * @mapping: The address_space to search
1277 * @index: The starting page index
1278 * @nr_pages: The maximum number of pages
1279 * @pages: Where the resulting pages are placed
1281 * find_get_pages_contig() works exactly like find_get_pages(), except
1282 * that the returned number of pages are guaranteed to be contiguous.
1284 * find_get_pages_contig() returns the number of pages which were found.
1286 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1287 unsigned int nr_pages, struct page **pages)
1289 struct radix_tree_iter iter;
1291 unsigned int ret = 0;
1293 if (unlikely(!nr_pages))
1298 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1301 page = radix_tree_deref_slot(slot);
1302 /* The hole, there no reason to continue */
1303 if (unlikely(!page))
1306 if (radix_tree_exception(page)) {
1307 if (radix_tree_deref_retry(page)) {
1309 * Transient condition which can only trigger
1310 * when entry at index 0 moves out of or back
1311 * to root: none yet gotten, safe to restart.
1316 * A shadow entry of a recently evicted page,
1317 * or a swap entry from shmem/tmpfs. Stop
1318 * looking for contiguous pages.
1323 if (!page_cache_get_speculative(page))
1326 /* Has the page moved? */
1327 if (unlikely(page != *slot)) {
1328 page_cache_release(page);
1333 * must check mapping and index after taking the ref.
1334 * otherwise we can get both false positives and false
1335 * negatives, which is just confusing to the caller.
1337 if (page->mapping == NULL || page->index != iter.index) {
1338 page_cache_release(page);
1343 if (++ret == nr_pages)
1349 EXPORT_SYMBOL(find_get_pages_contig);
1352 * find_get_pages_tag - find and return pages that match @tag
1353 * @mapping: the address_space to search
1354 * @index: the starting page index
1355 * @tag: the tag index
1356 * @nr_pages: the maximum number of pages
1357 * @pages: where the resulting pages are placed
1359 * Like find_get_pages, except we only return pages which are tagged with
1360 * @tag. We update @index to index the next page for the traversal.
1362 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1363 int tag, unsigned int nr_pages, struct page **pages)
1365 struct radix_tree_iter iter;
1369 if (unlikely(!nr_pages))
1374 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1375 &iter, *index, tag) {
1378 page = radix_tree_deref_slot(slot);
1379 if (unlikely(!page))
1382 if (radix_tree_exception(page)) {
1383 if (radix_tree_deref_retry(page)) {
1385 * Transient condition which can only trigger
1386 * when entry at index 0 moves out of or back
1387 * to root: none yet gotten, safe to restart.
1392 * A shadow entry of a recently evicted page.
1394 * Those entries should never be tagged, but
1395 * this tree walk is lockless and the tags are
1396 * looked up in bulk, one radix tree node at a
1397 * time, so there is a sizable window for page
1398 * reclaim to evict a page we saw tagged.
1405 if (!page_cache_get_speculative(page))
1408 /* Has the page moved? */
1409 if (unlikely(page != *slot)) {
1410 page_cache_release(page);
1415 if (++ret == nr_pages)
1422 *index = pages[ret - 1]->index + 1;
1426 EXPORT_SYMBOL(find_get_pages_tag);
1429 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1430 * a _large_ part of the i/o request. Imagine the worst scenario:
1432 * ---R__________________________________________B__________
1433 * ^ reading here ^ bad block(assume 4k)
1435 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1436 * => failing the whole request => read(R) => read(R+1) =>
1437 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1438 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1439 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1441 * It is going insane. Fix it by quickly scaling down the readahead size.
1443 static void shrink_readahead_size_eio(struct file *filp,
1444 struct file_ra_state *ra)
1450 * do_generic_file_read - generic file read routine
1451 * @filp: the file to read
1452 * @ppos: current file position
1453 * @iter: data destination
1454 * @written: already copied
1456 * This is a generic file read routine, and uses the
1457 * mapping->a_ops->readpage() function for the actual low-level stuff.
1459 * This is really ugly. But the goto's actually try to clarify some
1460 * of the logic when it comes to error handling etc.
1462 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1463 struct iov_iter *iter, ssize_t written)
1465 struct address_space *mapping = filp->f_mapping;
1466 struct inode *inode = mapping->host;
1467 struct file_ra_state *ra = &filp->f_ra;
1471 unsigned long offset; /* offset into pagecache page */
1472 unsigned int prev_offset;
1475 index = *ppos >> PAGE_CACHE_SHIFT;
1476 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1477 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1478 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1479 offset = *ppos & ~PAGE_CACHE_MASK;
1485 unsigned long nr, ret;
1489 page = find_get_page(mapping, index);
1491 page_cache_sync_readahead(mapping,
1493 index, last_index - index);
1494 page = find_get_page(mapping, index);
1495 if (unlikely(page == NULL))
1496 goto no_cached_page;
1498 if (PageReadahead(page)) {
1499 page_cache_async_readahead(mapping,
1501 index, last_index - index);
1503 if (!PageUptodate(page)) {
1504 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1505 !mapping->a_ops->is_partially_uptodate)
1506 goto page_not_up_to_date;
1507 if (!trylock_page(page))
1508 goto page_not_up_to_date;
1509 /* Did it get truncated before we got the lock? */
1511 goto page_not_up_to_date_locked;
1512 if (!mapping->a_ops->is_partially_uptodate(page,
1513 offset, iter->count))
1514 goto page_not_up_to_date_locked;
1519 * i_size must be checked after we know the page is Uptodate.
1521 * Checking i_size after the check allows us to calculate
1522 * the correct value for "nr", which means the zero-filled
1523 * part of the page is not copied back to userspace (unless
1524 * another truncate extends the file - this is desired though).
1527 isize = i_size_read(inode);
1528 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1529 if (unlikely(!isize || index > end_index)) {
1530 page_cache_release(page);
1534 /* nr is the maximum number of bytes to copy from this page */
1535 nr = PAGE_CACHE_SIZE;
1536 if (index == end_index) {
1537 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1539 page_cache_release(page);
1545 /* If users can be writing to this page using arbitrary
1546 * virtual addresses, take care about potential aliasing
1547 * before reading the page on the kernel side.
1549 if (mapping_writably_mapped(mapping))
1550 flush_dcache_page(page);
1553 * When a sequential read accesses a page several times,
1554 * only mark it as accessed the first time.
1556 if (prev_index != index || offset != prev_offset)
1557 mark_page_accessed(page);
1561 * Ok, we have the page, and it's up-to-date, so
1562 * now we can copy it to user space...
1565 ret = copy_page_to_iter(page, offset, nr, iter);
1567 index += offset >> PAGE_CACHE_SHIFT;
1568 offset &= ~PAGE_CACHE_MASK;
1569 prev_offset = offset;
1571 page_cache_release(page);
1573 if (!iov_iter_count(iter))
1581 page_not_up_to_date:
1582 /* Get exclusive access to the page ... */
1583 error = lock_page_killable(page);
1584 if (unlikely(error))
1585 goto readpage_error;
1587 page_not_up_to_date_locked:
1588 /* Did it get truncated before we got the lock? */
1589 if (!page->mapping) {
1591 page_cache_release(page);
1595 /* Did somebody else fill it already? */
1596 if (PageUptodate(page)) {
1603 * A previous I/O error may have been due to temporary
1604 * failures, eg. multipath errors.
1605 * PG_error will be set again if readpage fails.
1607 ClearPageError(page);
1608 /* Start the actual read. The read will unlock the page. */
1609 error = mapping->a_ops->readpage(filp, page);
1611 if (unlikely(error)) {
1612 if (error == AOP_TRUNCATED_PAGE) {
1613 page_cache_release(page);
1617 goto readpage_error;
1620 if (!PageUptodate(page)) {
1621 error = lock_page_killable(page);
1622 if (unlikely(error))
1623 goto readpage_error;
1624 if (!PageUptodate(page)) {
1625 if (page->mapping == NULL) {
1627 * invalidate_mapping_pages got it
1630 page_cache_release(page);
1634 shrink_readahead_size_eio(filp, ra);
1636 goto readpage_error;
1644 /* UHHUH! A synchronous read error occurred. Report it */
1645 page_cache_release(page);
1650 * Ok, it wasn't cached, so we need to create a new
1653 page = page_cache_alloc_cold(mapping);
1658 error = add_to_page_cache_lru(page, mapping,
1661 page_cache_release(page);
1662 if (error == -EEXIST) {
1672 ra->prev_pos = prev_index;
1673 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1674 ra->prev_pos |= prev_offset;
1676 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1677 file_accessed(filp);
1678 return written ? written : error;
1682 * generic_file_read_iter - generic filesystem read routine
1683 * @iocb: kernel I/O control block
1684 * @iter: destination for the data read
1686 * This is the "read_iter()" routine for all filesystems
1687 * that can use the page cache directly.
1690 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1692 struct file *file = iocb->ki_filp;
1694 loff_t *ppos = &iocb->ki_pos;
1697 if (io_is_direct(file)) {
1698 struct address_space *mapping = file->f_mapping;
1699 struct inode *inode = mapping->host;
1700 size_t count = iov_iter_count(iter);
1704 goto out; /* skip atime */
1705 size = i_size_read(inode);
1706 retval = filemap_write_and_wait_range(mapping, pos,
1709 struct iov_iter data = *iter;
1710 retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos);
1714 *ppos = pos + retval;
1715 iov_iter_advance(iter, retval);
1719 * Btrfs can have a short DIO read if we encounter
1720 * compressed extents, so if there was an error, or if
1721 * we've already read everything we wanted to, or if
1722 * there was a short read because we hit EOF, go ahead
1723 * and return. Otherwise fallthrough to buffered io for
1724 * the rest of the read. Buffered reads will not work for
1725 * DAX files, so don't bother trying.
1727 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1729 file_accessed(file);
1734 retval = do_generic_file_read(file, ppos, iter, retval);
1738 EXPORT_SYMBOL(generic_file_read_iter);
1742 * page_cache_read - adds requested page to the page cache if not already there
1743 * @file: file to read
1744 * @offset: page index
1746 * This adds the requested page to the page cache if it isn't already there,
1747 * and schedules an I/O to read in its contents from disk.
1749 static int page_cache_read(struct file *file, pgoff_t offset)
1751 struct address_space *mapping = file->f_mapping;
1756 page = page_cache_alloc_cold(mapping);
1760 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1762 ret = mapping->a_ops->readpage(file, page);
1763 else if (ret == -EEXIST)
1764 ret = 0; /* losing race to add is OK */
1766 page_cache_release(page);
1768 } while (ret == AOP_TRUNCATED_PAGE);
1773 #define MMAP_LOTSAMISS (100)
1776 * Synchronous readahead happens when we don't even find
1777 * a page in the page cache at all.
1779 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1780 struct file_ra_state *ra,
1784 unsigned long ra_pages;
1785 struct address_space *mapping = file->f_mapping;
1787 /* If we don't want any read-ahead, don't bother */
1788 if (vma->vm_flags & VM_RAND_READ)
1793 if (vma->vm_flags & VM_SEQ_READ) {
1794 page_cache_sync_readahead(mapping, ra, file, offset,
1799 /* Avoid banging the cache line if not needed */
1800 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1804 * Do we miss much more than hit in this file? If so,
1805 * stop bothering with read-ahead. It will only hurt.
1807 if (ra->mmap_miss > MMAP_LOTSAMISS)
1813 ra_pages = max_sane_readahead(ra->ra_pages);
1814 ra->start = max_t(long, 0, offset - ra_pages / 2);
1815 ra->size = ra_pages;
1816 ra->async_size = ra_pages / 4;
1817 ra_submit(ra, mapping, file);
1821 * Asynchronous readahead happens when we find the page and PG_readahead,
1822 * so we want to possibly extend the readahead further..
1824 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1825 struct file_ra_state *ra,
1830 struct address_space *mapping = file->f_mapping;
1832 /* If we don't want any read-ahead, don't bother */
1833 if (vma->vm_flags & VM_RAND_READ)
1835 if (ra->mmap_miss > 0)
1837 if (PageReadahead(page))
1838 page_cache_async_readahead(mapping, ra, file,
1839 page, offset, ra->ra_pages);
1843 * filemap_fault - read in file data for page fault handling
1844 * @vma: vma in which the fault was taken
1845 * @vmf: struct vm_fault containing details of the fault
1847 * filemap_fault() is invoked via the vma operations vector for a
1848 * mapped memory region to read in file data during a page fault.
1850 * The goto's are kind of ugly, but this streamlines the normal case of having
1851 * it in the page cache, and handles the special cases reasonably without
1852 * having a lot of duplicated code.
1854 * vma->vm_mm->mmap_sem must be held on entry.
1856 * If our return value has VM_FAULT_RETRY set, it's because
1857 * lock_page_or_retry() returned 0.
1858 * The mmap_sem has usually been released in this case.
1859 * See __lock_page_or_retry() for the exception.
1861 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1862 * has not been released.
1864 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1866 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1869 struct file *file = vma->vm_file;
1870 struct address_space *mapping = file->f_mapping;
1871 struct file_ra_state *ra = &file->f_ra;
1872 struct inode *inode = mapping->host;
1873 pgoff_t offset = vmf->pgoff;
1878 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1879 if (offset >= size >> PAGE_CACHE_SHIFT)
1880 return VM_FAULT_SIGBUS;
1883 * Do we have something in the page cache already?
1885 page = find_get_page(mapping, offset);
1886 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1888 * We found the page, so try async readahead before
1889 * waiting for the lock.
1891 do_async_mmap_readahead(vma, ra, file, page, offset);
1893 /* No page in the page cache at all */
1894 do_sync_mmap_readahead(vma, ra, file, offset);
1895 count_vm_event(PGMAJFAULT);
1896 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1897 ret = VM_FAULT_MAJOR;
1899 page = find_get_page(mapping, offset);
1901 goto no_cached_page;
1904 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1905 page_cache_release(page);
1906 return ret | VM_FAULT_RETRY;
1909 /* Did it get truncated? */
1910 if (unlikely(page->mapping != mapping)) {
1915 VM_BUG_ON_PAGE(page->index != offset, page);
1918 * We have a locked page in the page cache, now we need to check
1919 * that it's up-to-date. If not, it is going to be due to an error.
1921 if (unlikely(!PageUptodate(page)))
1922 goto page_not_uptodate;
1925 * Found the page and have a reference on it.
1926 * We must recheck i_size under page lock.
1928 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1929 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1931 page_cache_release(page);
1932 return VM_FAULT_SIGBUS;
1936 return ret | VM_FAULT_LOCKED;
1940 * We're only likely to ever get here if MADV_RANDOM is in
1943 error = page_cache_read(file, offset);
1946 * The page we want has now been added to the page cache.
1947 * In the unlikely event that someone removed it in the
1948 * meantime, we'll just come back here and read it again.
1954 * An error return from page_cache_read can result if the
1955 * system is low on memory, or a problem occurs while trying
1958 if (error == -ENOMEM)
1959 return VM_FAULT_OOM;
1960 return VM_FAULT_SIGBUS;
1964 * Umm, take care of errors if the page isn't up-to-date.
1965 * Try to re-read it _once_. We do this synchronously,
1966 * because there really aren't any performance issues here
1967 * and we need to check for errors.
1969 ClearPageError(page);
1970 error = mapping->a_ops->readpage(file, page);
1972 wait_on_page_locked(page);
1973 if (!PageUptodate(page))
1976 page_cache_release(page);
1978 if (!error || error == AOP_TRUNCATED_PAGE)
1981 /* Things didn't work out. Return zero to tell the mm layer so. */
1982 shrink_readahead_size_eio(file, ra);
1983 return VM_FAULT_SIGBUS;
1985 EXPORT_SYMBOL(filemap_fault);
1987 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1989 struct radix_tree_iter iter;
1991 struct file *file = vma->vm_file;
1992 struct address_space *mapping = file->f_mapping;
1995 unsigned long address = (unsigned long) vmf->virtual_address;
2000 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2001 if (iter.index > vmf->max_pgoff)
2004 page = radix_tree_deref_slot(slot);
2005 if (unlikely(!page))
2007 if (radix_tree_exception(page)) {
2008 if (radix_tree_deref_retry(page))
2014 if (!page_cache_get_speculative(page))
2017 /* Has the page moved? */
2018 if (unlikely(page != *slot)) {
2019 page_cache_release(page);
2023 if (!PageUptodate(page) ||
2024 PageReadahead(page) ||
2027 if (!trylock_page(page))
2030 if (page->mapping != mapping || !PageUptodate(page))
2033 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2034 if (page->index >= size >> PAGE_CACHE_SHIFT)
2037 pte = vmf->pte + page->index - vmf->pgoff;
2038 if (!pte_none(*pte))
2041 if (file->f_ra.mmap_miss > 0)
2042 file->f_ra.mmap_miss--;
2043 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2044 do_set_pte(vma, addr, page, pte, false, false);
2050 page_cache_release(page);
2052 if (iter.index == vmf->max_pgoff)
2057 EXPORT_SYMBOL(filemap_map_pages);
2059 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2061 struct page *page = vmf->page;
2062 struct inode *inode = file_inode(vma->vm_file);
2063 int ret = VM_FAULT_LOCKED;
2065 sb_start_pagefault(inode->i_sb);
2066 file_update_time(vma->vm_file);
2068 if (page->mapping != inode->i_mapping) {
2070 ret = VM_FAULT_NOPAGE;
2074 * We mark the page dirty already here so that when freeze is in
2075 * progress, we are guaranteed that writeback during freezing will
2076 * see the dirty page and writeprotect it again.
2078 set_page_dirty(page);
2079 wait_for_stable_page(page);
2081 sb_end_pagefault(inode->i_sb);
2084 EXPORT_SYMBOL(filemap_page_mkwrite);
2086 const struct vm_operations_struct generic_file_vm_ops = {
2087 .fault = filemap_fault,
2088 .map_pages = filemap_map_pages,
2089 .page_mkwrite = filemap_page_mkwrite,
2092 /* This is used for a general mmap of a disk file */
2094 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2096 struct address_space *mapping = file->f_mapping;
2098 if (!mapping->a_ops->readpage)
2100 file_accessed(file);
2101 vma->vm_ops = &generic_file_vm_ops;
2106 * This is for filesystems which do not implement ->writepage.
2108 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2110 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2112 return generic_file_mmap(file, vma);
2115 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2119 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2123 #endif /* CONFIG_MMU */
2125 EXPORT_SYMBOL(generic_file_mmap);
2126 EXPORT_SYMBOL(generic_file_readonly_mmap);
2128 static struct page *wait_on_page_read(struct page *page)
2130 if (!IS_ERR(page)) {
2131 wait_on_page_locked(page);
2132 if (!PageUptodate(page)) {
2133 page_cache_release(page);
2134 page = ERR_PTR(-EIO);
2140 static struct page *__read_cache_page(struct address_space *mapping,
2142 int (*filler)(void *, struct page *),
2149 page = find_get_page(mapping, index);
2151 page = __page_cache_alloc(gfp | __GFP_COLD);
2153 return ERR_PTR(-ENOMEM);
2154 err = add_to_page_cache_lru(page, mapping, index, gfp);
2155 if (unlikely(err)) {
2156 page_cache_release(page);
2159 /* Presumably ENOMEM for radix tree node */
2160 return ERR_PTR(err);
2162 err = filler(data, page);
2164 page_cache_release(page);
2165 page = ERR_PTR(err);
2167 page = wait_on_page_read(page);
2173 static struct page *do_read_cache_page(struct address_space *mapping,
2175 int (*filler)(void *, struct page *),
2184 page = __read_cache_page(mapping, index, filler, data, gfp);
2187 if (PageUptodate(page))
2191 if (!page->mapping) {
2193 page_cache_release(page);
2196 if (PageUptodate(page)) {
2200 err = filler(data, page);
2202 page_cache_release(page);
2203 return ERR_PTR(err);
2205 page = wait_on_page_read(page);
2210 mark_page_accessed(page);
2215 * read_cache_page - read into page cache, fill it if needed
2216 * @mapping: the page's address_space
2217 * @index: the page index
2218 * @filler: function to perform the read
2219 * @data: first arg to filler(data, page) function, often left as NULL
2221 * Read into the page cache. If a page already exists, and PageUptodate() is
2222 * not set, try to fill the page and wait for it to become unlocked.
2224 * If the page does not get brought uptodate, return -EIO.
2226 struct page *read_cache_page(struct address_space *mapping,
2228 int (*filler)(void *, struct page *),
2231 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2233 EXPORT_SYMBOL(read_cache_page);
2236 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2237 * @mapping: the page's address_space
2238 * @index: the page index
2239 * @gfp: the page allocator flags to use if allocating
2241 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2242 * any new page allocations done using the specified allocation flags.
2244 * If the page does not get brought uptodate, return -EIO.
2246 struct page *read_cache_page_gfp(struct address_space *mapping,
2250 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2252 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2254 EXPORT_SYMBOL(read_cache_page_gfp);
2257 * Performs necessary checks before doing a write
2259 * Can adjust writing position or amount of bytes to write.
2260 * Returns appropriate error code that caller should return or
2261 * zero in case that write should be allowed.
2263 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2265 struct inode *inode = file->f_mapping->host;
2266 unsigned long limit = rlimit(RLIMIT_FSIZE);
2268 if (unlikely(*pos < 0))
2272 /* FIXME: this is for backwards compatibility with 2.4 */
2273 if (file->f_flags & O_APPEND)
2274 *pos = i_size_read(inode);
2276 if (limit != RLIM_INFINITY) {
2277 if (*pos >= limit) {
2278 send_sig(SIGXFSZ, current, 0);
2281 if (*count > limit - (typeof(limit))*pos) {
2282 *count = limit - (typeof(limit))*pos;
2290 if (unlikely(*pos + *count > MAX_NON_LFS &&
2291 !(file->f_flags & O_LARGEFILE))) {
2292 if (*pos >= MAX_NON_LFS) {
2295 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2296 *count = MAX_NON_LFS - (unsigned long)*pos;
2301 * Are we about to exceed the fs block limit ?
2303 * If we have written data it becomes a short write. If we have
2304 * exceeded without writing data we send a signal and return EFBIG.
2305 * Linus frestrict idea will clean these up nicely..
2307 if (likely(!isblk)) {
2308 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2309 if (*count || *pos > inode->i_sb->s_maxbytes) {
2312 /* zero-length writes at ->s_maxbytes are OK */
2315 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2316 *count = inode->i_sb->s_maxbytes - *pos;
2320 if (bdev_read_only(I_BDEV(inode)))
2322 isize = i_size_read(inode);
2323 if (*pos >= isize) {
2324 if (*count || *pos > isize)
2328 if (*pos + *count > isize)
2329 *count = isize - *pos;
2336 EXPORT_SYMBOL(generic_write_checks);
2338 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2339 loff_t pos, unsigned len, unsigned flags,
2340 struct page **pagep, void **fsdata)
2342 const struct address_space_operations *aops = mapping->a_ops;
2344 return aops->write_begin(file, mapping, pos, len, flags,
2347 EXPORT_SYMBOL(pagecache_write_begin);
2349 int pagecache_write_end(struct file *file, struct address_space *mapping,
2350 loff_t pos, unsigned len, unsigned copied,
2351 struct page *page, void *fsdata)
2353 const struct address_space_operations *aops = mapping->a_ops;
2355 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2357 EXPORT_SYMBOL(pagecache_write_end);
2360 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2362 struct file *file = iocb->ki_filp;
2363 struct address_space *mapping = file->f_mapping;
2364 struct inode *inode = mapping->host;
2368 struct iov_iter data;
2370 write_len = iov_iter_count(from);
2371 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2373 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2378 * After a write we want buffered reads to be sure to go to disk to get
2379 * the new data. We invalidate clean cached page from the region we're
2380 * about to write. We do this *before* the write so that we can return
2381 * without clobbering -EIOCBQUEUED from ->direct_IO().
2383 if (mapping->nrpages) {
2384 written = invalidate_inode_pages2_range(mapping,
2385 pos >> PAGE_CACHE_SHIFT, end);
2387 * If a page can not be invalidated, return 0 to fall back
2388 * to buffered write.
2391 if (written == -EBUSY)
2398 written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos);
2401 * Finally, try again to invalidate clean pages which might have been
2402 * cached by non-direct readahead, or faulted in by get_user_pages()
2403 * if the source of the write was an mmap'ed region of the file
2404 * we're writing. Either one is a pretty crazy thing to do,
2405 * so we don't support it 100%. If this invalidation
2406 * fails, tough, the write still worked...
2408 if (mapping->nrpages) {
2409 invalidate_inode_pages2_range(mapping,
2410 pos >> PAGE_CACHE_SHIFT, end);
2415 iov_iter_advance(from, written);
2416 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2417 i_size_write(inode, pos);
2418 mark_inode_dirty(inode);
2425 EXPORT_SYMBOL(generic_file_direct_write);
2428 * Find or create a page at the given pagecache position. Return the locked
2429 * page. This function is specifically for buffered writes.
2431 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2432 pgoff_t index, unsigned flags)
2435 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2437 if (flags & AOP_FLAG_NOFS)
2438 fgp_flags |= FGP_NOFS;
2440 page = pagecache_get_page(mapping, index, fgp_flags,
2441 mapping_gfp_mask(mapping));
2443 wait_for_stable_page(page);
2447 EXPORT_SYMBOL(grab_cache_page_write_begin);
2449 ssize_t generic_perform_write(struct file *file,
2450 struct iov_iter *i, loff_t pos)
2452 struct address_space *mapping = file->f_mapping;
2453 const struct address_space_operations *a_ops = mapping->a_ops;
2455 ssize_t written = 0;
2456 unsigned int flags = 0;
2459 * Copies from kernel address space cannot fail (NFSD is a big user).
2461 if (!iter_is_iovec(i))
2462 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2466 unsigned long offset; /* Offset into pagecache page */
2467 unsigned long bytes; /* Bytes to write to page */
2468 size_t copied; /* Bytes copied from user */
2471 offset = (pos & (PAGE_CACHE_SIZE - 1));
2472 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2477 * Bring in the user page that we will copy from _first_.
2478 * Otherwise there's a nasty deadlock on copying from the
2479 * same page as we're writing to, without it being marked
2482 * Not only is this an optimisation, but it is also required
2483 * to check that the address is actually valid, when atomic
2484 * usercopies are used, below.
2486 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2491 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2493 if (unlikely(status < 0))
2496 if (mapping_writably_mapped(mapping))
2497 flush_dcache_page(page);
2499 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2500 flush_dcache_page(page);
2502 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2504 if (unlikely(status < 0))
2510 iov_iter_advance(i, copied);
2511 if (unlikely(copied == 0)) {
2513 * If we were unable to copy any data at all, we must
2514 * fall back to a single segment length write.
2516 * If we didn't fallback here, we could livelock
2517 * because not all segments in the iov can be copied at
2518 * once without a pagefault.
2520 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2521 iov_iter_single_seg_count(i));
2527 balance_dirty_pages_ratelimited(mapping);
2528 if (fatal_signal_pending(current)) {
2532 } while (iov_iter_count(i));
2534 return written ? written : status;
2536 EXPORT_SYMBOL(generic_perform_write);
2539 * __generic_file_write_iter - write data to a file
2540 * @iocb: IO state structure (file, offset, etc.)
2541 * @from: iov_iter with data to write
2543 * This function does all the work needed for actually writing data to a
2544 * file. It does all basic checks, removes SUID from the file, updates
2545 * modification times and calls proper subroutines depending on whether we
2546 * do direct IO or a standard buffered write.
2548 * It expects i_mutex to be grabbed unless we work on a block device or similar
2549 * object which does not need locking at all.
2551 * This function does *not* take care of syncing data in case of O_SYNC write.
2552 * A caller has to handle it. This is mainly due to the fact that we want to
2553 * avoid syncing under i_mutex.
2555 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2557 struct file *file = iocb->ki_filp;
2558 struct address_space * mapping = file->f_mapping;
2559 struct inode *inode = mapping->host;
2560 loff_t pos = iocb->ki_pos;
2561 ssize_t written = 0;
2564 size_t count = iov_iter_count(from);
2566 /* We can write back this queue in page reclaim */
2567 current->backing_dev_info = inode_to_bdi(inode);
2568 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2575 iov_iter_truncate(from, count);
2577 err = file_remove_suid(file);
2581 err = file_update_time(file);
2585 if (io_is_direct(file)) {
2588 written = generic_file_direct_write(iocb, from, pos);
2590 * If the write stopped short of completing, fall back to
2591 * buffered writes. Some filesystems do this for writes to
2592 * holes, for example. For DAX files, a buffered write will
2593 * not succeed (even if it did, DAX does not handle dirty
2594 * page-cache pages correctly).
2596 if (written < 0 || written == count || IS_DAX(inode))
2602 status = generic_perform_write(file, from, pos);
2604 * If generic_perform_write() returned a synchronous error
2605 * then we want to return the number of bytes which were
2606 * direct-written, or the error code if that was zero. Note
2607 * that this differs from normal direct-io semantics, which
2608 * will return -EFOO even if some bytes were written.
2610 if (unlikely(status < 0)) {
2614 iocb->ki_pos = pos + status;
2616 * We need to ensure that the page cache pages are written to
2617 * disk and invalidated to preserve the expected O_DIRECT
2620 endbyte = pos + status - 1;
2621 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2624 invalidate_mapping_pages(mapping,
2625 pos >> PAGE_CACHE_SHIFT,
2626 endbyte >> PAGE_CACHE_SHIFT);
2629 * We don't know how much we wrote, so just return
2630 * the number of bytes which were direct-written
2634 written = generic_perform_write(file, from, pos);
2635 if (likely(written >= 0))
2636 iocb->ki_pos = pos + written;
2639 current->backing_dev_info = NULL;
2640 return written ? written : err;
2642 EXPORT_SYMBOL(__generic_file_write_iter);
2645 * generic_file_write_iter - write data to a file
2646 * @iocb: IO state structure
2647 * @from: iov_iter with data to write
2649 * This is a wrapper around __generic_file_write_iter() to be used by most
2650 * filesystems. It takes care of syncing the file in case of O_SYNC file
2651 * and acquires i_mutex as needed.
2653 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2655 struct file *file = iocb->ki_filp;
2656 struct inode *inode = file->f_mapping->host;
2659 mutex_lock(&inode->i_mutex);
2660 ret = __generic_file_write_iter(iocb, from);
2661 mutex_unlock(&inode->i_mutex);
2666 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2672 EXPORT_SYMBOL(generic_file_write_iter);
2675 * try_to_release_page() - release old fs-specific metadata on a page
2677 * @page: the page which the kernel is trying to free
2678 * @gfp_mask: memory allocation flags (and I/O mode)
2680 * The address_space is to try to release any data against the page
2681 * (presumably at page->private). If the release was successful, return `1'.
2682 * Otherwise return zero.
2684 * This may also be called if PG_fscache is set on a page, indicating that the
2685 * page is known to the local caching routines.
2687 * The @gfp_mask argument specifies whether I/O may be performed to release
2688 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2691 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2693 struct address_space * const mapping = page->mapping;
2695 BUG_ON(!PageLocked(page));
2696 if (PageWriteback(page))
2699 if (mapping && mapping->a_ops->releasepage)
2700 return mapping->a_ops->releasepage(page, gfp_mask);
2701 return try_to_free_buffers(page);
2704 EXPORT_SYMBOL(try_to_release_page);