4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/capability.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/gfp.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/backing-dev.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/cpuset.h>
32 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
33 #include <linux/hugetlb.h>
34 #include <linux/memcontrol.h>
35 #include <linux/cleancache.h>
36 #include <linux/rmap.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/filemap.h>
43 * FIXME: remove all knowledge of the buffer layer from the core VM
45 #include <linux/buffer_head.h> /* for try_to_free_buffers */
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * Shared mappings now work. 15.8.1995 Bruno.
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
64 * ->i_mmap_rwsem (truncate_pagecache)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
70 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
74 * ->page_table_lock or pte_lock (various, mainly in memory.c)
75 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
78 * ->lock_page (access_process_vm)
80 * ->i_mutex (generic_perform_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
84 * sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
103 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
112 static void page_cache_tree_delete(struct address_space *mapping,
113 struct page *page, void *shadow)
115 struct radix_tree_node *node;
121 VM_BUG_ON(!PageLocked(page));
123 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
126 mapping->nrshadows++;
128 * Make sure the nrshadows update is committed before
129 * the nrpages update so that final truncate racing
130 * with reclaim does not see both counters 0 at the
131 * same time and miss a shadow entry.
138 /* Clear direct pointer tags in root node */
139 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
140 radix_tree_replace_slot(slot, shadow);
144 /* Clear tree tags for the removed page */
146 offset = index & RADIX_TREE_MAP_MASK;
147 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
148 if (test_bit(offset, node->tags[tag]))
149 radix_tree_tag_clear(&mapping->page_tree, index, tag);
152 /* Delete page, swap shadow entry */
153 radix_tree_replace_slot(slot, shadow);
154 workingset_node_pages_dec(node);
156 workingset_node_shadows_inc(node);
158 if (__radix_tree_delete_node(&mapping->page_tree, node))
162 * Track node that only contains shadow entries.
164 * Avoid acquiring the list_lru lock if already tracked. The
165 * list_empty() test is safe as node->private_list is
166 * protected by mapping->tree_lock.
168 if (!workingset_node_pages(node) &&
169 list_empty(&node->private_list)) {
170 node->private_data = mapping;
171 list_lru_add(&workingset_shadow_nodes, &node->private_list);
176 * Delete a page from the page cache and free it. Caller has to make
177 * sure the page is locked and that nobody else uses it - or that usage
178 * is safe. The caller must hold the mapping's tree_lock and
179 * mem_cgroup_begin_page_stat().
181 void __delete_from_page_cache(struct page *page, void *shadow,
182 struct mem_cgroup *memcg)
184 struct address_space *mapping = page->mapping;
186 trace_mm_filemap_delete_from_page_cache(page);
188 * if we're uptodate, flush out into the cleancache, otherwise
189 * invalidate any existing cleancache entries. We can't leave
190 * stale data around in the cleancache once our page is gone
192 if (PageUptodate(page) && PageMappedToDisk(page))
193 cleancache_put_page(page);
195 cleancache_invalidate_page(mapping, page);
197 page_cache_tree_delete(mapping, page, shadow);
199 page->mapping = NULL;
200 /* Leave page->index set: truncation lookup relies upon it */
202 /* hugetlb pages do not participate in page cache accounting. */
204 __dec_zone_page_state(page, NR_FILE_PAGES);
205 if (PageSwapBacked(page))
206 __dec_zone_page_state(page, NR_SHMEM);
207 BUG_ON(page_mapped(page));
210 * At this point page must be either written or cleaned by truncate.
211 * Dirty page here signals a bug and loss of unwritten data.
213 * This fixes dirty accounting after removing the page entirely but
214 * leaves PageDirty set: it has no effect for truncated page and
215 * anyway will be cleared before returning page into buddy allocator.
217 if (WARN_ON_ONCE(PageDirty(page)))
218 account_page_cleaned(page, mapping, memcg,
219 inode_to_wb(mapping->host));
223 * delete_from_page_cache - delete page from page cache
224 * @page: the page which the kernel is trying to remove from page cache
226 * This must be called only on pages that have been verified to be in the page
227 * cache and locked. It will never put the page into the free list, the caller
228 * has a reference on the page.
230 void delete_from_page_cache(struct page *page)
232 struct address_space *mapping = page->mapping;
233 struct mem_cgroup *memcg;
236 void (*freepage)(struct page *);
238 BUG_ON(!PageLocked(page));
240 freepage = mapping->a_ops->freepage;
242 memcg = mem_cgroup_begin_page_stat(page);
243 spin_lock_irqsave(&mapping->tree_lock, flags);
244 __delete_from_page_cache(page, NULL, memcg);
245 spin_unlock_irqrestore(&mapping->tree_lock, flags);
246 mem_cgroup_end_page_stat(memcg);
250 page_cache_release(page);
252 EXPORT_SYMBOL(delete_from_page_cache);
254 static int filemap_check_errors(struct address_space *mapping)
257 /* Check for outstanding write errors */
258 if (test_bit(AS_ENOSPC, &mapping->flags) &&
259 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
261 if (test_bit(AS_EIO, &mapping->flags) &&
262 test_and_clear_bit(AS_EIO, &mapping->flags))
268 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
269 * @mapping: address space structure to write
270 * @start: offset in bytes where the range starts
271 * @end: offset in bytes where the range ends (inclusive)
272 * @sync_mode: enable synchronous operation
274 * Start writeback against all of a mapping's dirty pages that lie
275 * within the byte offsets <start, end> inclusive.
277 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
278 * opposed to a regular memory cleansing writeback. The difference between
279 * these two operations is that if a dirty page/buffer is encountered, it must
280 * be waited upon, and not just skipped over.
282 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
283 loff_t end, int sync_mode)
286 struct writeback_control wbc = {
287 .sync_mode = sync_mode,
288 .nr_to_write = LONG_MAX,
289 .range_start = start,
293 if (!mapping_cap_writeback_dirty(mapping))
296 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
297 ret = do_writepages(mapping, &wbc);
298 wbc_detach_inode(&wbc);
302 static inline int __filemap_fdatawrite(struct address_space *mapping,
305 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
308 int filemap_fdatawrite(struct address_space *mapping)
310 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
312 EXPORT_SYMBOL(filemap_fdatawrite);
314 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
317 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
319 EXPORT_SYMBOL(filemap_fdatawrite_range);
322 * filemap_flush - mostly a non-blocking flush
323 * @mapping: target address_space
325 * This is a mostly non-blocking flush. Not suitable for data-integrity
326 * purposes - I/O may not be started against all dirty pages.
328 int filemap_flush(struct address_space *mapping)
330 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
332 EXPORT_SYMBOL(filemap_flush);
335 * filemap_fdatawait_range - wait for writeback to complete
336 * @mapping: address space structure to wait for
337 * @start_byte: offset in bytes where the range starts
338 * @end_byte: offset in bytes where the range ends (inclusive)
340 * Walk the list of under-writeback pages of the given address space
341 * in the given range and wait for all of them.
343 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
346 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
347 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
352 if (end_byte < start_byte)
355 pagevec_init(&pvec, 0);
356 while ((index <= end) &&
357 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
358 PAGECACHE_TAG_WRITEBACK,
359 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
362 for (i = 0; i < nr_pages; i++) {
363 struct page *page = pvec.pages[i];
365 /* until radix tree lookup accepts end_index */
366 if (page->index > end)
369 wait_on_page_writeback(page);
370 if (TestClearPageError(page))
373 pagevec_release(&pvec);
377 ret2 = filemap_check_errors(mapping);
383 EXPORT_SYMBOL(filemap_fdatawait_range);
386 * filemap_fdatawait - wait for all under-writeback pages to complete
387 * @mapping: address space structure to wait for
389 * Walk the list of under-writeback pages of the given address space
390 * and wait for all of them.
392 int filemap_fdatawait(struct address_space *mapping)
394 loff_t i_size = i_size_read(mapping->host);
399 return filemap_fdatawait_range(mapping, 0, i_size - 1);
401 EXPORT_SYMBOL(filemap_fdatawait);
403 int filemap_write_and_wait(struct address_space *mapping)
407 if (mapping->nrpages) {
408 err = filemap_fdatawrite(mapping);
410 * Even if the above returned error, the pages may be
411 * written partially (e.g. -ENOSPC), so we wait for it.
412 * But the -EIO is special case, it may indicate the worst
413 * thing (e.g. bug) happened, so we avoid waiting for it.
416 int err2 = filemap_fdatawait(mapping);
421 err = filemap_check_errors(mapping);
425 EXPORT_SYMBOL(filemap_write_and_wait);
428 * filemap_write_and_wait_range - write out & wait on a file range
429 * @mapping: the address_space for the pages
430 * @lstart: offset in bytes where the range starts
431 * @lend: offset in bytes where the range ends (inclusive)
433 * Write out and wait upon file offsets lstart->lend, inclusive.
435 * Note that `lend' is inclusive (describes the last byte to be written) so
436 * that this function can be used to write to the very end-of-file (end = -1).
438 int filemap_write_and_wait_range(struct address_space *mapping,
439 loff_t lstart, loff_t lend)
443 if (mapping->nrpages) {
444 err = __filemap_fdatawrite_range(mapping, lstart, lend,
446 /* See comment of filemap_write_and_wait() */
448 int err2 = filemap_fdatawait_range(mapping,
454 err = filemap_check_errors(mapping);
458 EXPORT_SYMBOL(filemap_write_and_wait_range);
461 * replace_page_cache_page - replace a pagecache page with a new one
462 * @old: page to be replaced
463 * @new: page to replace with
464 * @gfp_mask: allocation mode
466 * This function replaces a page in the pagecache with a new one. On
467 * success it acquires the pagecache reference for the new page and
468 * drops it for the old page. Both the old and new pages must be
469 * locked. This function does not add the new page to the LRU, the
470 * caller must do that.
472 * The remove + add is atomic. The only way this function can fail is
473 * memory allocation failure.
475 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
479 VM_BUG_ON_PAGE(!PageLocked(old), old);
480 VM_BUG_ON_PAGE(!PageLocked(new), new);
481 VM_BUG_ON_PAGE(new->mapping, new);
483 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
485 struct address_space *mapping = old->mapping;
486 void (*freepage)(struct page *);
487 struct mem_cgroup *memcg;
490 pgoff_t offset = old->index;
491 freepage = mapping->a_ops->freepage;
494 new->mapping = mapping;
497 memcg = mem_cgroup_begin_page_stat(old);
498 spin_lock_irqsave(&mapping->tree_lock, flags);
499 __delete_from_page_cache(old, NULL, memcg);
500 error = radix_tree_insert(&mapping->page_tree, offset, new);
505 * hugetlb pages do not participate in page cache accounting.
508 __inc_zone_page_state(new, NR_FILE_PAGES);
509 if (PageSwapBacked(new))
510 __inc_zone_page_state(new, NR_SHMEM);
511 spin_unlock_irqrestore(&mapping->tree_lock, flags);
512 mem_cgroup_end_page_stat(memcg);
513 mem_cgroup_migrate(old, new, true);
514 radix_tree_preload_end();
517 page_cache_release(old);
522 EXPORT_SYMBOL_GPL(replace_page_cache_page);
524 static int page_cache_tree_insert(struct address_space *mapping,
525 struct page *page, void **shadowp)
527 struct radix_tree_node *node;
531 error = __radix_tree_create(&mapping->page_tree, page->index,
538 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
539 if (!radix_tree_exceptional_entry(p))
543 mapping->nrshadows--;
545 workingset_node_shadows_dec(node);
547 radix_tree_replace_slot(slot, page);
550 workingset_node_pages_inc(node);
552 * Don't track node that contains actual pages.
554 * Avoid acquiring the list_lru lock if already
555 * untracked. The list_empty() test is safe as
556 * node->private_list is protected by
557 * mapping->tree_lock.
559 if (!list_empty(&node->private_list))
560 list_lru_del(&workingset_shadow_nodes,
561 &node->private_list);
566 static int __add_to_page_cache_locked(struct page *page,
567 struct address_space *mapping,
568 pgoff_t offset, gfp_t gfp_mask,
571 int huge = PageHuge(page);
572 struct mem_cgroup *memcg;
575 VM_BUG_ON_PAGE(!PageLocked(page), page);
576 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
579 error = mem_cgroup_try_charge(page, current->mm,
585 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
588 mem_cgroup_cancel_charge(page, memcg);
592 page_cache_get(page);
593 page->mapping = mapping;
594 page->index = offset;
596 spin_lock_irq(&mapping->tree_lock);
597 error = page_cache_tree_insert(mapping, page, shadowp);
598 radix_tree_preload_end();
602 /* hugetlb pages do not participate in page cache accounting. */
604 __inc_zone_page_state(page, NR_FILE_PAGES);
605 spin_unlock_irq(&mapping->tree_lock);
607 mem_cgroup_commit_charge(page, memcg, false);
608 trace_mm_filemap_add_to_page_cache(page);
611 page->mapping = NULL;
612 /* Leave page->index set: truncation relies upon it */
613 spin_unlock_irq(&mapping->tree_lock);
615 mem_cgroup_cancel_charge(page, memcg);
616 page_cache_release(page);
621 * add_to_page_cache_locked - add a locked page to the pagecache
623 * @mapping: the page's address_space
624 * @offset: page index
625 * @gfp_mask: page allocation mode
627 * This function is used to add a page to the pagecache. It must be locked.
628 * This function does not add the page to the LRU. The caller must do that.
630 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
631 pgoff_t offset, gfp_t gfp_mask)
633 return __add_to_page_cache_locked(page, mapping, offset,
636 EXPORT_SYMBOL(add_to_page_cache_locked);
638 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
639 pgoff_t offset, gfp_t gfp_mask)
644 __set_page_locked(page);
645 ret = __add_to_page_cache_locked(page, mapping, offset,
648 __clear_page_locked(page);
651 * The page might have been evicted from cache only
652 * recently, in which case it should be activated like
653 * any other repeatedly accessed page.
655 if (shadow && workingset_refault(shadow)) {
657 workingset_activation(page);
659 ClearPageActive(page);
664 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
667 struct page *__page_cache_alloc(gfp_t gfp)
672 if (cpuset_do_page_mem_spread()) {
673 unsigned int cpuset_mems_cookie;
675 cpuset_mems_cookie = read_mems_allowed_begin();
676 n = cpuset_mem_spread_node();
677 page = __alloc_pages_node(n, gfp, 0);
678 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
682 return alloc_pages(gfp, 0);
684 EXPORT_SYMBOL(__page_cache_alloc);
688 * In order to wait for pages to become available there must be
689 * waitqueues associated with pages. By using a hash table of
690 * waitqueues where the bucket discipline is to maintain all
691 * waiters on the same queue and wake all when any of the pages
692 * become available, and for the woken contexts to check to be
693 * sure the appropriate page became available, this saves space
694 * at a cost of "thundering herd" phenomena during rare hash
697 wait_queue_head_t *page_waitqueue(struct page *page)
699 const struct zone *zone = page_zone(page);
701 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
703 EXPORT_SYMBOL(page_waitqueue);
705 void wait_on_page_bit(struct page *page, int bit_nr)
707 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
709 if (test_bit(bit_nr, &page->flags))
710 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
711 TASK_UNINTERRUPTIBLE);
713 EXPORT_SYMBOL(wait_on_page_bit);
715 int wait_on_page_bit_killable(struct page *page, int bit_nr)
717 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
719 if (!test_bit(bit_nr, &page->flags))
722 return __wait_on_bit(page_waitqueue(page), &wait,
723 bit_wait_io, TASK_KILLABLE);
726 int wait_on_page_bit_killable_timeout(struct page *page,
727 int bit_nr, unsigned long timeout)
729 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
731 wait.key.timeout = jiffies + timeout;
732 if (!test_bit(bit_nr, &page->flags))
734 return __wait_on_bit(page_waitqueue(page), &wait,
735 bit_wait_io_timeout, TASK_KILLABLE);
737 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
740 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
741 * @page: Page defining the wait queue of interest
742 * @waiter: Waiter to add to the queue
744 * Add an arbitrary @waiter to the wait queue for the nominated @page.
746 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
748 wait_queue_head_t *q = page_waitqueue(page);
751 spin_lock_irqsave(&q->lock, flags);
752 __add_wait_queue(q, waiter);
753 spin_unlock_irqrestore(&q->lock, flags);
755 EXPORT_SYMBOL_GPL(add_page_wait_queue);
758 * unlock_page - unlock a locked page
761 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
762 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
763 * mechanism between PageLocked pages and PageWriteback pages is shared.
764 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
766 * The mb is necessary to enforce ordering between the clear_bit and the read
767 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
769 void unlock_page(struct page *page)
771 VM_BUG_ON_PAGE(!PageLocked(page), page);
772 clear_bit_unlock(PG_locked, &page->flags);
773 smp_mb__after_atomic();
774 wake_up_page(page, PG_locked);
776 EXPORT_SYMBOL(unlock_page);
779 * end_page_writeback - end writeback against a page
782 void end_page_writeback(struct page *page)
785 * TestClearPageReclaim could be used here but it is an atomic
786 * operation and overkill in this particular case. Failing to
787 * shuffle a page marked for immediate reclaim is too mild to
788 * justify taking an atomic operation penalty at the end of
789 * ever page writeback.
791 if (PageReclaim(page)) {
792 ClearPageReclaim(page);
793 rotate_reclaimable_page(page);
796 if (!test_clear_page_writeback(page))
799 smp_mb__after_atomic();
800 wake_up_page(page, PG_writeback);
802 EXPORT_SYMBOL(end_page_writeback);
805 * After completing I/O on a page, call this routine to update the page
806 * flags appropriately
808 void page_endio(struct page *page, int rw, int err)
812 SetPageUptodate(page);
814 ClearPageUptodate(page);
818 } else { /* rw == WRITE */
822 mapping_set_error(page->mapping, err);
824 end_page_writeback(page);
827 EXPORT_SYMBOL_GPL(page_endio);
830 * __lock_page - get a lock on the page, assuming we need to sleep to get it
831 * @page: the page to lock
833 void __lock_page(struct page *page)
835 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
837 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
838 TASK_UNINTERRUPTIBLE);
840 EXPORT_SYMBOL(__lock_page);
842 int __lock_page_killable(struct page *page)
844 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
846 return __wait_on_bit_lock(page_waitqueue(page), &wait,
847 bit_wait_io, TASK_KILLABLE);
849 EXPORT_SYMBOL_GPL(__lock_page_killable);
853 * 1 - page is locked; mmap_sem is still held.
854 * 0 - page is not locked.
855 * mmap_sem has been released (up_read()), unless flags had both
856 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
857 * which case mmap_sem is still held.
859 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
860 * with the page locked and the mmap_sem unperturbed.
862 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
865 if (flags & FAULT_FLAG_ALLOW_RETRY) {
867 * CAUTION! In this case, mmap_sem is not released
868 * even though return 0.
870 if (flags & FAULT_FLAG_RETRY_NOWAIT)
873 up_read(&mm->mmap_sem);
874 if (flags & FAULT_FLAG_KILLABLE)
875 wait_on_page_locked_killable(page);
877 wait_on_page_locked(page);
880 if (flags & FAULT_FLAG_KILLABLE) {
883 ret = __lock_page_killable(page);
885 up_read(&mm->mmap_sem);
895 * page_cache_next_hole - find the next hole (not-present entry)
898 * @max_scan: maximum range to search
900 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
901 * lowest indexed hole.
903 * Returns: the index of the hole if found, otherwise returns an index
904 * outside of the set specified (in which case 'return - index >=
905 * max_scan' will be true). In rare cases of index wrap-around, 0 will
908 * page_cache_next_hole may be called under rcu_read_lock. However,
909 * like radix_tree_gang_lookup, this will not atomically search a
910 * snapshot of the tree at a single point in time. For example, if a
911 * hole is created at index 5, then subsequently a hole is created at
912 * index 10, page_cache_next_hole covering both indexes may return 10
913 * if called under rcu_read_lock.
915 pgoff_t page_cache_next_hole(struct address_space *mapping,
916 pgoff_t index, unsigned long max_scan)
920 for (i = 0; i < max_scan; i++) {
923 page = radix_tree_lookup(&mapping->page_tree, index);
924 if (!page || radix_tree_exceptional_entry(page))
933 EXPORT_SYMBOL(page_cache_next_hole);
936 * page_cache_prev_hole - find the prev hole (not-present entry)
939 * @max_scan: maximum range to search
941 * Search backwards in the range [max(index-max_scan+1, 0), index] for
944 * Returns: the index of the hole if found, otherwise returns an index
945 * outside of the set specified (in which case 'index - return >=
946 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
949 * page_cache_prev_hole may be called under rcu_read_lock. However,
950 * like radix_tree_gang_lookup, this will not atomically search a
951 * snapshot of the tree at a single point in time. For example, if a
952 * hole is created at index 10, then subsequently a hole is created at
953 * index 5, page_cache_prev_hole covering both indexes may return 5 if
954 * called under rcu_read_lock.
956 pgoff_t page_cache_prev_hole(struct address_space *mapping,
957 pgoff_t index, unsigned long max_scan)
961 for (i = 0; i < max_scan; i++) {
964 page = radix_tree_lookup(&mapping->page_tree, index);
965 if (!page || radix_tree_exceptional_entry(page))
968 if (index == ULONG_MAX)
974 EXPORT_SYMBOL(page_cache_prev_hole);
977 * find_get_entry - find and get a page cache entry
978 * @mapping: the address_space to search
979 * @offset: the page cache index
981 * Looks up the page cache slot at @mapping & @offset. If there is a
982 * page cache page, it is returned with an increased refcount.
984 * If the slot holds a shadow entry of a previously evicted page, or a
985 * swap entry from shmem/tmpfs, it is returned.
987 * Otherwise, %NULL is returned.
989 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
997 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
999 page = radix_tree_deref_slot(pagep);
1000 if (unlikely(!page))
1002 if (radix_tree_exception(page)) {
1003 if (radix_tree_deref_retry(page))
1006 * A shadow entry of a recently evicted page,
1007 * or a swap entry from shmem/tmpfs. Return
1008 * it without attempting to raise page count.
1012 if (!page_cache_get_speculative(page))
1016 * Has the page moved?
1017 * This is part of the lockless pagecache protocol. See
1018 * include/linux/pagemap.h for details.
1020 if (unlikely(page != *pagep)) {
1021 page_cache_release(page);
1030 EXPORT_SYMBOL(find_get_entry);
1033 * find_lock_entry - locate, pin and lock a page cache entry
1034 * @mapping: the address_space to search
1035 * @offset: the page cache index
1037 * Looks up the page cache slot at @mapping & @offset. If there is a
1038 * page cache page, it is returned locked and with an increased
1041 * If the slot holds a shadow entry of a previously evicted page, or a
1042 * swap entry from shmem/tmpfs, it is returned.
1044 * Otherwise, %NULL is returned.
1046 * find_lock_entry() may sleep.
1048 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1053 page = find_get_entry(mapping, offset);
1054 if (page && !radix_tree_exception(page)) {
1056 /* Has the page been truncated? */
1057 if (unlikely(page->mapping != mapping)) {
1059 page_cache_release(page);
1062 VM_BUG_ON_PAGE(page->index != offset, page);
1066 EXPORT_SYMBOL(find_lock_entry);
1069 * pagecache_get_page - find and get a page reference
1070 * @mapping: the address_space to search
1071 * @offset: the page index
1072 * @fgp_flags: PCG flags
1073 * @gfp_mask: gfp mask to use for the page cache data page allocation
1075 * Looks up the page cache slot at @mapping & @offset.
1077 * PCG flags modify how the page is returned.
1079 * FGP_ACCESSED: the page will be marked accessed
1080 * FGP_LOCK: Page is return locked
1081 * FGP_CREAT: If page is not present then a new page is allocated using
1082 * @gfp_mask and added to the page cache and the VM's LRU
1083 * list. The page is returned locked and with an increased
1084 * refcount. Otherwise, %NULL is returned.
1086 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1087 * if the GFP flags specified for FGP_CREAT are atomic.
1089 * If there is a page cache page, it is returned with an increased refcount.
1091 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1092 int fgp_flags, gfp_t gfp_mask)
1097 page = find_get_entry(mapping, offset);
1098 if (radix_tree_exceptional_entry(page))
1103 if (fgp_flags & FGP_LOCK) {
1104 if (fgp_flags & FGP_NOWAIT) {
1105 if (!trylock_page(page)) {
1106 page_cache_release(page);
1113 /* Has the page been truncated? */
1114 if (unlikely(page->mapping != mapping)) {
1116 page_cache_release(page);
1119 VM_BUG_ON_PAGE(page->index != offset, page);
1122 if (page && (fgp_flags & FGP_ACCESSED))
1123 mark_page_accessed(page);
1126 if (!page && (fgp_flags & FGP_CREAT)) {
1128 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1129 gfp_mask |= __GFP_WRITE;
1130 if (fgp_flags & FGP_NOFS)
1131 gfp_mask &= ~__GFP_FS;
1133 page = __page_cache_alloc(gfp_mask);
1137 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1138 fgp_flags |= FGP_LOCK;
1140 /* Init accessed so avoid atomic mark_page_accessed later */
1141 if (fgp_flags & FGP_ACCESSED)
1142 __SetPageReferenced(page);
1144 err = add_to_page_cache_lru(page, mapping, offset,
1145 gfp_mask & GFP_RECLAIM_MASK);
1146 if (unlikely(err)) {
1147 page_cache_release(page);
1156 EXPORT_SYMBOL(pagecache_get_page);
1159 * find_get_entries - gang pagecache lookup
1160 * @mapping: The address_space to search
1161 * @start: The starting page cache index
1162 * @nr_entries: The maximum number of entries
1163 * @entries: Where the resulting entries are placed
1164 * @indices: The cache indices corresponding to the entries in @entries
1166 * find_get_entries() will search for and return a group of up to
1167 * @nr_entries entries in the mapping. The entries are placed at
1168 * @entries. find_get_entries() takes a reference against any actual
1171 * The search returns a group of mapping-contiguous page cache entries
1172 * with ascending indexes. There may be holes in the indices due to
1173 * not-present pages.
1175 * Any shadow entries of evicted pages, or swap entries from
1176 * shmem/tmpfs, are included in the returned array.
1178 * find_get_entries() returns the number of pages and shadow entries
1181 unsigned find_get_entries(struct address_space *mapping,
1182 pgoff_t start, unsigned int nr_entries,
1183 struct page **entries, pgoff_t *indices)
1186 unsigned int ret = 0;
1187 struct radix_tree_iter iter;
1194 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1197 page = radix_tree_deref_slot(slot);
1198 if (unlikely(!page))
1200 if (radix_tree_exception(page)) {
1201 if (radix_tree_deref_retry(page))
1204 * A shadow entry of a recently evicted page,
1205 * or a swap entry from shmem/tmpfs. Return
1206 * it without attempting to raise page count.
1210 if (!page_cache_get_speculative(page))
1213 /* Has the page moved? */
1214 if (unlikely(page != *slot)) {
1215 page_cache_release(page);
1219 indices[ret] = iter.index;
1220 entries[ret] = page;
1221 if (++ret == nr_entries)
1229 * find_get_pages - gang pagecache lookup
1230 * @mapping: The address_space to search
1231 * @start: The starting page index
1232 * @nr_pages: The maximum number of pages
1233 * @pages: Where the resulting pages are placed
1235 * find_get_pages() will search for and return a group of up to
1236 * @nr_pages pages in the mapping. The pages are placed at @pages.
1237 * find_get_pages() takes a reference against the returned pages.
1239 * The search returns a group of mapping-contiguous pages with ascending
1240 * indexes. There may be holes in the indices due to not-present pages.
1242 * find_get_pages() returns the number of pages which were found.
1244 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1245 unsigned int nr_pages, struct page **pages)
1247 struct radix_tree_iter iter;
1251 if (unlikely(!nr_pages))
1256 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1259 page = radix_tree_deref_slot(slot);
1260 if (unlikely(!page))
1263 if (radix_tree_exception(page)) {
1264 if (radix_tree_deref_retry(page)) {
1266 * Transient condition which can only trigger
1267 * when entry at index 0 moves out of or back
1268 * to root: none yet gotten, safe to restart.
1270 WARN_ON(iter.index);
1274 * A shadow entry of a recently evicted page,
1275 * or a swap entry from shmem/tmpfs. Skip
1281 if (!page_cache_get_speculative(page))
1284 /* Has the page moved? */
1285 if (unlikely(page != *slot)) {
1286 page_cache_release(page);
1291 if (++ret == nr_pages)
1300 * find_get_pages_contig - gang contiguous pagecache lookup
1301 * @mapping: The address_space to search
1302 * @index: The starting page index
1303 * @nr_pages: The maximum number of pages
1304 * @pages: Where the resulting pages are placed
1306 * find_get_pages_contig() works exactly like find_get_pages(), except
1307 * that the returned number of pages are guaranteed to be contiguous.
1309 * find_get_pages_contig() returns the number of pages which were found.
1311 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1312 unsigned int nr_pages, struct page **pages)
1314 struct radix_tree_iter iter;
1316 unsigned int ret = 0;
1318 if (unlikely(!nr_pages))
1323 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1326 page = radix_tree_deref_slot(slot);
1327 /* The hole, there no reason to continue */
1328 if (unlikely(!page))
1331 if (radix_tree_exception(page)) {
1332 if (radix_tree_deref_retry(page)) {
1334 * Transient condition which can only trigger
1335 * when entry at index 0 moves out of or back
1336 * to root: none yet gotten, safe to restart.
1341 * A shadow entry of a recently evicted page,
1342 * or a swap entry from shmem/tmpfs. Stop
1343 * looking for contiguous pages.
1348 if (!page_cache_get_speculative(page))
1351 /* Has the page moved? */
1352 if (unlikely(page != *slot)) {
1353 page_cache_release(page);
1358 * must check mapping and index after taking the ref.
1359 * otherwise we can get both false positives and false
1360 * negatives, which is just confusing to the caller.
1362 if (page->mapping == NULL || page->index != iter.index) {
1363 page_cache_release(page);
1368 if (++ret == nr_pages)
1374 EXPORT_SYMBOL(find_get_pages_contig);
1377 * find_get_pages_tag - find and return pages that match @tag
1378 * @mapping: the address_space to search
1379 * @index: the starting page index
1380 * @tag: the tag index
1381 * @nr_pages: the maximum number of pages
1382 * @pages: where the resulting pages are placed
1384 * Like find_get_pages, except we only return pages which are tagged with
1385 * @tag. We update @index to index the next page for the traversal.
1387 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1388 int tag, unsigned int nr_pages, struct page **pages)
1390 struct radix_tree_iter iter;
1394 if (unlikely(!nr_pages))
1399 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1400 &iter, *index, tag) {
1403 page = radix_tree_deref_slot(slot);
1404 if (unlikely(!page))
1407 if (radix_tree_exception(page)) {
1408 if (radix_tree_deref_retry(page)) {
1410 * Transient condition which can only trigger
1411 * when entry at index 0 moves out of or back
1412 * to root: none yet gotten, safe to restart.
1417 * A shadow entry of a recently evicted page.
1419 * Those entries should never be tagged, but
1420 * this tree walk is lockless and the tags are
1421 * looked up in bulk, one radix tree node at a
1422 * time, so there is a sizable window for page
1423 * reclaim to evict a page we saw tagged.
1430 if (!page_cache_get_speculative(page))
1433 /* Has the page moved? */
1434 if (unlikely(page != *slot)) {
1435 page_cache_release(page);
1440 if (++ret == nr_pages)
1447 *index = pages[ret - 1]->index + 1;
1451 EXPORT_SYMBOL(find_get_pages_tag);
1454 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1455 * a _large_ part of the i/o request. Imagine the worst scenario:
1457 * ---R__________________________________________B__________
1458 * ^ reading here ^ bad block(assume 4k)
1460 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1461 * => failing the whole request => read(R) => read(R+1) =>
1462 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1463 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1464 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1466 * It is going insane. Fix it by quickly scaling down the readahead size.
1468 static void shrink_readahead_size_eio(struct file *filp,
1469 struct file_ra_state *ra)
1475 * do_generic_file_read - generic file read routine
1476 * @filp: the file to read
1477 * @ppos: current file position
1478 * @iter: data destination
1479 * @written: already copied
1481 * This is a generic file read routine, and uses the
1482 * mapping->a_ops->readpage() function for the actual low-level stuff.
1484 * This is really ugly. But the goto's actually try to clarify some
1485 * of the logic when it comes to error handling etc.
1487 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1488 struct iov_iter *iter, ssize_t written)
1490 struct address_space *mapping = filp->f_mapping;
1491 struct inode *inode = mapping->host;
1492 struct file_ra_state *ra = &filp->f_ra;
1496 unsigned long offset; /* offset into pagecache page */
1497 unsigned int prev_offset;
1500 index = *ppos >> PAGE_CACHE_SHIFT;
1501 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1502 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1503 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1504 offset = *ppos & ~PAGE_CACHE_MASK;
1510 unsigned long nr, ret;
1514 page = find_get_page(mapping, index);
1516 page_cache_sync_readahead(mapping,
1518 index, last_index - index);
1519 page = find_get_page(mapping, index);
1520 if (unlikely(page == NULL))
1521 goto no_cached_page;
1523 if (PageReadahead(page)) {
1524 page_cache_async_readahead(mapping,
1526 index, last_index - index);
1528 if (!PageUptodate(page)) {
1529 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1530 !mapping->a_ops->is_partially_uptodate)
1531 goto page_not_up_to_date;
1532 if (!trylock_page(page))
1533 goto page_not_up_to_date;
1534 /* Did it get truncated before we got the lock? */
1536 goto page_not_up_to_date_locked;
1537 if (!mapping->a_ops->is_partially_uptodate(page,
1538 offset, iter->count))
1539 goto page_not_up_to_date_locked;
1544 * i_size must be checked after we know the page is Uptodate.
1546 * Checking i_size after the check allows us to calculate
1547 * the correct value for "nr", which means the zero-filled
1548 * part of the page is not copied back to userspace (unless
1549 * another truncate extends the file - this is desired though).
1552 isize = i_size_read(inode);
1553 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1554 if (unlikely(!isize || index > end_index)) {
1555 page_cache_release(page);
1559 /* nr is the maximum number of bytes to copy from this page */
1560 nr = PAGE_CACHE_SIZE;
1561 if (index == end_index) {
1562 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1564 page_cache_release(page);
1570 /* If users can be writing to this page using arbitrary
1571 * virtual addresses, take care about potential aliasing
1572 * before reading the page on the kernel side.
1574 if (mapping_writably_mapped(mapping))
1575 flush_dcache_page(page);
1578 * When a sequential read accesses a page several times,
1579 * only mark it as accessed the first time.
1581 if (prev_index != index || offset != prev_offset)
1582 mark_page_accessed(page);
1586 * Ok, we have the page, and it's up-to-date, so
1587 * now we can copy it to user space...
1590 ret = copy_page_to_iter(page, offset, nr, iter);
1592 index += offset >> PAGE_CACHE_SHIFT;
1593 offset &= ~PAGE_CACHE_MASK;
1594 prev_offset = offset;
1596 page_cache_release(page);
1598 if (!iov_iter_count(iter))
1606 page_not_up_to_date:
1607 /* Get exclusive access to the page ... */
1608 error = lock_page_killable(page);
1609 if (unlikely(error))
1610 goto readpage_error;
1612 page_not_up_to_date_locked:
1613 /* Did it get truncated before we got the lock? */
1614 if (!page->mapping) {
1616 page_cache_release(page);
1620 /* Did somebody else fill it already? */
1621 if (PageUptodate(page)) {
1628 * A previous I/O error may have been due to temporary
1629 * failures, eg. multipath errors.
1630 * PG_error will be set again if readpage fails.
1632 ClearPageError(page);
1633 /* Start the actual read. The read will unlock the page. */
1634 error = mapping->a_ops->readpage(filp, page);
1636 if (unlikely(error)) {
1637 if (error == AOP_TRUNCATED_PAGE) {
1638 page_cache_release(page);
1642 goto readpage_error;
1645 if (!PageUptodate(page)) {
1646 error = lock_page_killable(page);
1647 if (unlikely(error))
1648 goto readpage_error;
1649 if (!PageUptodate(page)) {
1650 if (page->mapping == NULL) {
1652 * invalidate_mapping_pages got it
1655 page_cache_release(page);
1659 shrink_readahead_size_eio(filp, ra);
1661 goto readpage_error;
1669 /* UHHUH! A synchronous read error occurred. Report it */
1670 page_cache_release(page);
1675 * Ok, it wasn't cached, so we need to create a new
1678 page = page_cache_alloc_cold(mapping);
1683 error = add_to_page_cache_lru(page, mapping, index,
1684 GFP_KERNEL & mapping_gfp_mask(mapping));
1686 page_cache_release(page);
1687 if (error == -EEXIST) {
1697 ra->prev_pos = prev_index;
1698 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1699 ra->prev_pos |= prev_offset;
1701 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1702 file_accessed(filp);
1703 return written ? written : error;
1707 * generic_file_read_iter - generic filesystem read routine
1708 * @iocb: kernel I/O control block
1709 * @iter: destination for the data read
1711 * This is the "read_iter()" routine for all filesystems
1712 * that can use the page cache directly.
1715 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1717 struct file *file = iocb->ki_filp;
1719 loff_t *ppos = &iocb->ki_pos;
1722 if (iocb->ki_flags & IOCB_DIRECT) {
1723 struct address_space *mapping = file->f_mapping;
1724 struct inode *inode = mapping->host;
1725 size_t count = iov_iter_count(iter);
1729 goto out; /* skip atime */
1730 size = i_size_read(inode);
1731 retval = filemap_write_and_wait_range(mapping, pos,
1734 struct iov_iter data = *iter;
1735 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
1739 *ppos = pos + retval;
1740 iov_iter_advance(iter, retval);
1744 * Btrfs can have a short DIO read if we encounter
1745 * compressed extents, so if there was an error, or if
1746 * we've already read everything we wanted to, or if
1747 * there was a short read because we hit EOF, go ahead
1748 * and return. Otherwise fallthrough to buffered io for
1749 * the rest of the read. Buffered reads will not work for
1750 * DAX files, so don't bother trying.
1752 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1754 file_accessed(file);
1759 retval = do_generic_file_read(file, ppos, iter, retval);
1763 EXPORT_SYMBOL(generic_file_read_iter);
1767 * page_cache_read - adds requested page to the page cache if not already there
1768 * @file: file to read
1769 * @offset: page index
1771 * This adds the requested page to the page cache if it isn't already there,
1772 * and schedules an I/O to read in its contents from disk.
1774 static int page_cache_read(struct file *file, pgoff_t offset)
1776 struct address_space *mapping = file->f_mapping;
1781 page = page_cache_alloc_cold(mapping);
1785 ret = add_to_page_cache_lru(page, mapping, offset,
1786 GFP_KERNEL & mapping_gfp_mask(mapping));
1788 ret = mapping->a_ops->readpage(file, page);
1789 else if (ret == -EEXIST)
1790 ret = 0; /* losing race to add is OK */
1792 page_cache_release(page);
1794 } while (ret == AOP_TRUNCATED_PAGE);
1799 #define MMAP_LOTSAMISS (100)
1802 * Synchronous readahead happens when we don't even find
1803 * a page in the page cache at all.
1805 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1806 struct file_ra_state *ra,
1810 unsigned long ra_pages;
1811 struct address_space *mapping = file->f_mapping;
1813 /* If we don't want any read-ahead, don't bother */
1814 if (vma->vm_flags & VM_RAND_READ)
1819 if (vma->vm_flags & VM_SEQ_READ) {
1820 page_cache_sync_readahead(mapping, ra, file, offset,
1825 /* Avoid banging the cache line if not needed */
1826 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1830 * Do we miss much more than hit in this file? If so,
1831 * stop bothering with read-ahead. It will only hurt.
1833 if (ra->mmap_miss > MMAP_LOTSAMISS)
1839 ra_pages = max_sane_readahead(ra->ra_pages);
1840 ra->start = max_t(long, 0, offset - ra_pages / 2);
1841 ra->size = ra_pages;
1842 ra->async_size = ra_pages / 4;
1843 ra_submit(ra, mapping, file);
1847 * Asynchronous readahead happens when we find the page and PG_readahead,
1848 * so we want to possibly extend the readahead further..
1850 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1851 struct file_ra_state *ra,
1856 struct address_space *mapping = file->f_mapping;
1858 /* If we don't want any read-ahead, don't bother */
1859 if (vma->vm_flags & VM_RAND_READ)
1861 if (ra->mmap_miss > 0)
1863 if (PageReadahead(page))
1864 page_cache_async_readahead(mapping, ra, file,
1865 page, offset, ra->ra_pages);
1869 * filemap_fault - read in file data for page fault handling
1870 * @vma: vma in which the fault was taken
1871 * @vmf: struct vm_fault containing details of the fault
1873 * filemap_fault() is invoked via the vma operations vector for a
1874 * mapped memory region to read in file data during a page fault.
1876 * The goto's are kind of ugly, but this streamlines the normal case of having
1877 * it in the page cache, and handles the special cases reasonably without
1878 * having a lot of duplicated code.
1880 * vma->vm_mm->mmap_sem must be held on entry.
1882 * If our return value has VM_FAULT_RETRY set, it's because
1883 * lock_page_or_retry() returned 0.
1884 * The mmap_sem has usually been released in this case.
1885 * See __lock_page_or_retry() for the exception.
1887 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1888 * has not been released.
1890 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1892 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1895 struct file *file = vma->vm_file;
1896 struct address_space *mapping = file->f_mapping;
1897 struct file_ra_state *ra = &file->f_ra;
1898 struct inode *inode = mapping->host;
1899 pgoff_t offset = vmf->pgoff;
1904 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1905 if (offset >= size >> PAGE_CACHE_SHIFT)
1906 return VM_FAULT_SIGBUS;
1909 * Do we have something in the page cache already?
1911 page = find_get_page(mapping, offset);
1912 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1914 * We found the page, so try async readahead before
1915 * waiting for the lock.
1917 do_async_mmap_readahead(vma, ra, file, page, offset);
1919 /* No page in the page cache at all */
1920 do_sync_mmap_readahead(vma, ra, file, offset);
1921 count_vm_event(PGMAJFAULT);
1922 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1923 ret = VM_FAULT_MAJOR;
1925 page = find_get_page(mapping, offset);
1927 goto no_cached_page;
1930 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1931 page_cache_release(page);
1932 return ret | VM_FAULT_RETRY;
1935 /* Did it get truncated? */
1936 if (unlikely(page->mapping != mapping)) {
1941 VM_BUG_ON_PAGE(page->index != offset, page);
1944 * We have a locked page in the page cache, now we need to check
1945 * that it's up-to-date. If not, it is going to be due to an error.
1947 if (unlikely(!PageUptodate(page)))
1948 goto page_not_uptodate;
1951 * Found the page and have a reference on it.
1952 * We must recheck i_size under page lock.
1954 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1955 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1957 page_cache_release(page);
1958 return VM_FAULT_SIGBUS;
1962 return ret | VM_FAULT_LOCKED;
1966 * We're only likely to ever get here if MADV_RANDOM is in
1969 error = page_cache_read(file, offset);
1972 * The page we want has now been added to the page cache.
1973 * In the unlikely event that someone removed it in the
1974 * meantime, we'll just come back here and read it again.
1980 * An error return from page_cache_read can result if the
1981 * system is low on memory, or a problem occurs while trying
1984 if (error == -ENOMEM)
1985 return VM_FAULT_OOM;
1986 return VM_FAULT_SIGBUS;
1990 * Umm, take care of errors if the page isn't up-to-date.
1991 * Try to re-read it _once_. We do this synchronously,
1992 * because there really aren't any performance issues here
1993 * and we need to check for errors.
1995 ClearPageError(page);
1996 error = mapping->a_ops->readpage(file, page);
1998 wait_on_page_locked(page);
1999 if (!PageUptodate(page))
2002 page_cache_release(page);
2004 if (!error || error == AOP_TRUNCATED_PAGE)
2007 /* Things didn't work out. Return zero to tell the mm layer so. */
2008 shrink_readahead_size_eio(file, ra);
2009 return VM_FAULT_SIGBUS;
2011 EXPORT_SYMBOL(filemap_fault);
2013 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2015 struct radix_tree_iter iter;
2017 struct file *file = vma->vm_file;
2018 struct address_space *mapping = file->f_mapping;
2021 unsigned long address = (unsigned long) vmf->virtual_address;
2026 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2027 if (iter.index > vmf->max_pgoff)
2030 page = radix_tree_deref_slot(slot);
2031 if (unlikely(!page))
2033 if (radix_tree_exception(page)) {
2034 if (radix_tree_deref_retry(page))
2040 if (!page_cache_get_speculative(page))
2043 /* Has the page moved? */
2044 if (unlikely(page != *slot)) {
2045 page_cache_release(page);
2049 if (!PageUptodate(page) ||
2050 PageReadahead(page) ||
2053 if (!trylock_page(page))
2056 if (page->mapping != mapping || !PageUptodate(page))
2059 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2060 if (page->index >= size >> PAGE_CACHE_SHIFT)
2063 pte = vmf->pte + page->index - vmf->pgoff;
2064 if (!pte_none(*pte))
2067 if (file->f_ra.mmap_miss > 0)
2068 file->f_ra.mmap_miss--;
2069 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2070 do_set_pte(vma, addr, page, pte, false, false);
2076 page_cache_release(page);
2078 if (iter.index == vmf->max_pgoff)
2083 EXPORT_SYMBOL(filemap_map_pages);
2085 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2087 struct page *page = vmf->page;
2088 struct inode *inode = file_inode(vma->vm_file);
2089 int ret = VM_FAULT_LOCKED;
2091 sb_start_pagefault(inode->i_sb);
2092 file_update_time(vma->vm_file);
2094 if (page->mapping != inode->i_mapping) {
2096 ret = VM_FAULT_NOPAGE;
2100 * We mark the page dirty already here so that when freeze is in
2101 * progress, we are guaranteed that writeback during freezing will
2102 * see the dirty page and writeprotect it again.
2104 set_page_dirty(page);
2105 wait_for_stable_page(page);
2107 sb_end_pagefault(inode->i_sb);
2110 EXPORT_SYMBOL(filemap_page_mkwrite);
2112 const struct vm_operations_struct generic_file_vm_ops = {
2113 .fault = filemap_fault,
2114 .map_pages = filemap_map_pages,
2115 .page_mkwrite = filemap_page_mkwrite,
2118 /* This is used for a general mmap of a disk file */
2120 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2122 struct address_space *mapping = file->f_mapping;
2124 if (!mapping->a_ops->readpage)
2126 file_accessed(file);
2127 vma->vm_ops = &generic_file_vm_ops;
2132 * This is for filesystems which do not implement ->writepage.
2134 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2136 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2138 return generic_file_mmap(file, vma);
2141 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2145 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2149 #endif /* CONFIG_MMU */
2151 EXPORT_SYMBOL(generic_file_mmap);
2152 EXPORT_SYMBOL(generic_file_readonly_mmap);
2154 static struct page *wait_on_page_read(struct page *page)
2156 if (!IS_ERR(page)) {
2157 wait_on_page_locked(page);
2158 if (!PageUptodate(page)) {
2159 page_cache_release(page);
2160 page = ERR_PTR(-EIO);
2166 static struct page *__read_cache_page(struct address_space *mapping,
2168 int (*filler)(void *, struct page *),
2175 page = find_get_page(mapping, index);
2177 page = __page_cache_alloc(gfp | __GFP_COLD);
2179 return ERR_PTR(-ENOMEM);
2180 err = add_to_page_cache_lru(page, mapping, index, gfp);
2181 if (unlikely(err)) {
2182 page_cache_release(page);
2185 /* Presumably ENOMEM for radix tree node */
2186 return ERR_PTR(err);
2188 err = filler(data, page);
2190 page_cache_release(page);
2191 page = ERR_PTR(err);
2193 page = wait_on_page_read(page);
2199 static struct page *do_read_cache_page(struct address_space *mapping,
2201 int (*filler)(void *, struct page *),
2210 page = __read_cache_page(mapping, index, filler, data, gfp);
2213 if (PageUptodate(page))
2217 if (!page->mapping) {
2219 page_cache_release(page);
2222 if (PageUptodate(page)) {
2226 err = filler(data, page);
2228 page_cache_release(page);
2229 return ERR_PTR(err);
2231 page = wait_on_page_read(page);
2236 mark_page_accessed(page);
2241 * read_cache_page - read into page cache, fill it if needed
2242 * @mapping: the page's address_space
2243 * @index: the page index
2244 * @filler: function to perform the read
2245 * @data: first arg to filler(data, page) function, often left as NULL
2247 * Read into the page cache. If a page already exists, and PageUptodate() is
2248 * not set, try to fill the page and wait for it to become unlocked.
2250 * If the page does not get brought uptodate, return -EIO.
2252 struct page *read_cache_page(struct address_space *mapping,
2254 int (*filler)(void *, struct page *),
2257 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2259 EXPORT_SYMBOL(read_cache_page);
2262 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2263 * @mapping: the page's address_space
2264 * @index: the page index
2265 * @gfp: the page allocator flags to use if allocating
2267 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2268 * any new page allocations done using the specified allocation flags.
2270 * If the page does not get brought uptodate, return -EIO.
2272 struct page *read_cache_page_gfp(struct address_space *mapping,
2276 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2278 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2280 EXPORT_SYMBOL(read_cache_page_gfp);
2283 * Performs necessary checks before doing a write
2285 * Can adjust writing position or amount of bytes to write.
2286 * Returns appropriate error code that caller should return or
2287 * zero in case that write should be allowed.
2289 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2291 struct file *file = iocb->ki_filp;
2292 struct inode *inode = file->f_mapping->host;
2293 unsigned long limit = rlimit(RLIMIT_FSIZE);
2296 if (!iov_iter_count(from))
2299 /* FIXME: this is for backwards compatibility with 2.4 */
2300 if (iocb->ki_flags & IOCB_APPEND)
2301 iocb->ki_pos = i_size_read(inode);
2305 if (limit != RLIM_INFINITY) {
2306 if (iocb->ki_pos >= limit) {
2307 send_sig(SIGXFSZ, current, 0);
2310 iov_iter_truncate(from, limit - (unsigned long)pos);
2316 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2317 !(file->f_flags & O_LARGEFILE))) {
2318 if (pos >= MAX_NON_LFS)
2320 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2324 * Are we about to exceed the fs block limit ?
2326 * If we have written data it becomes a short write. If we have
2327 * exceeded without writing data we send a signal and return EFBIG.
2328 * Linus frestrict idea will clean these up nicely..
2330 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2333 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2334 return iov_iter_count(from);
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(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 ssize_t written = 0;
2564 /* We can write back this queue in page reclaim */
2565 current->backing_dev_info = inode_to_bdi(inode);
2566 err = file_remove_privs(file);
2570 err = file_update_time(file);
2574 if (iocb->ki_flags & IOCB_DIRECT) {
2575 loff_t pos, endbyte;
2577 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
2579 * If the write stopped short of completing, fall back to
2580 * buffered writes. Some filesystems do this for writes to
2581 * holes, for example. For DAX files, a buffered write will
2582 * not succeed (even if it did, DAX does not handle dirty
2583 * page-cache pages correctly).
2585 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2588 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2590 * If generic_perform_write() returned a synchronous error
2591 * then we want to return the number of bytes which were
2592 * direct-written, or the error code if that was zero. Note
2593 * that this differs from normal direct-io semantics, which
2594 * will return -EFOO even if some bytes were written.
2596 if (unlikely(status < 0)) {
2601 * We need to ensure that the page cache pages are written to
2602 * disk and invalidated to preserve the expected O_DIRECT
2605 endbyte = pos + status - 1;
2606 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2608 iocb->ki_pos = endbyte + 1;
2610 invalidate_mapping_pages(mapping,
2611 pos >> PAGE_CACHE_SHIFT,
2612 endbyte >> PAGE_CACHE_SHIFT);
2615 * We don't know how much we wrote, so just return
2616 * the number of bytes which were direct-written
2620 written = generic_perform_write(file, from, iocb->ki_pos);
2621 if (likely(written > 0))
2622 iocb->ki_pos += written;
2625 current->backing_dev_info = NULL;
2626 return written ? written : err;
2628 EXPORT_SYMBOL(__generic_file_write_iter);
2631 * generic_file_write_iter - write data to a file
2632 * @iocb: IO state structure
2633 * @from: iov_iter with data to write
2635 * This is a wrapper around __generic_file_write_iter() to be used by most
2636 * filesystems. It takes care of syncing the file in case of O_SYNC file
2637 * and acquires i_mutex as needed.
2639 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2641 struct file *file = iocb->ki_filp;
2642 struct inode *inode = file->f_mapping->host;
2645 mutex_lock(&inode->i_mutex);
2646 ret = generic_write_checks(iocb, from);
2648 ret = __generic_file_write_iter(iocb, from);
2649 mutex_unlock(&inode->i_mutex);
2654 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2660 EXPORT_SYMBOL(generic_file_write_iter);
2663 * try_to_release_page() - release old fs-specific metadata on a page
2665 * @page: the page which the kernel is trying to free
2666 * @gfp_mask: memory allocation flags (and I/O mode)
2668 * The address_space is to try to release any data against the page
2669 * (presumably at page->private). If the release was successful, return `1'.
2670 * Otherwise return zero.
2672 * This may also be called if PG_fscache is set on a page, indicating that the
2673 * page is known to the local caching routines.
2675 * The @gfp_mask argument specifies whether I/O may be performed to release
2676 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2679 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2681 struct address_space * const mapping = page->mapping;
2683 BUG_ON(!PageLocked(page));
2684 if (PageWriteback(page))
2687 if (mapping && mapping->a_ops->releasepage)
2688 return mapping->a_ops->releasepage(page, gfp_mask);
2689 return try_to_free_buffers(page);
2692 EXPORT_SYMBOL(try_to_release_page);