4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/aio.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/cpuset.h>
33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
34 #include <linux/memcontrol.h>
35 #include <linux/cleancache.h>
36 #include <linux/rmap.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/filemap.h>
43 * FIXME: remove all knowledge of the buffer layer from the core VM
45 #include <linux/buffer_head.h> /* for try_to_free_buffers */
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * Shared mappings now work. 15.8.1995 Bruno.
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
64 * ->i_mmap_mutex (truncate_pagecache)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
70 * ->i_mmap_mutex (truncate->unmap_mapping_range)
74 * ->page_table_lock or pte_lock (various, mainly in memory.c)
75 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
78 * ->lock_page (access_process_vm)
80 * ->i_mutex (generic_perform_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
84 * sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
104 * ->inode->i_lock (zap_pte_range->set_page_dirty)
105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 static void page_cache_tree_delete(struct address_space *mapping,
112 struct page *page, void *shadow)
114 struct radix_tree_node *node;
120 VM_BUG_ON(!PageLocked(page));
122 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
125 mapping->nrshadows++;
127 * Make sure the nrshadows update is committed before
128 * the nrpages update so that final truncate racing
129 * with reclaim does not see both counters 0 at the
130 * same time and miss a shadow entry.
137 /* Clear direct pointer tags in root node */
138 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
139 radix_tree_replace_slot(slot, shadow);
143 /* Clear tree tags for the removed page */
145 offset = index & RADIX_TREE_MAP_MASK;
146 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
147 if (test_bit(offset, node->tags[tag]))
148 radix_tree_tag_clear(&mapping->page_tree, index, tag);
151 /* Delete page, swap shadow entry */
152 radix_tree_replace_slot(slot, shadow);
153 workingset_node_pages_dec(node);
155 workingset_node_shadows_inc(node);
157 if (__radix_tree_delete_node(&mapping->page_tree, node))
161 * Track node that only contains shadow entries.
163 * Avoid acquiring the list_lru lock if already tracked. The
164 * list_empty() test is safe as node->private_list is
165 * protected by mapping->tree_lock.
167 if (!workingset_node_pages(node) &&
168 list_empty(&node->private_list)) {
169 node->private_data = mapping;
170 list_lru_add(&workingset_shadow_nodes, &node->private_list);
175 * Delete a page from the page cache and free it. Caller has to make
176 * sure the page is locked and that nobody else uses it - or that usage
177 * is safe. The caller must hold the mapping's tree_lock.
179 void __delete_from_page_cache(struct page *page, void *shadow)
181 struct address_space *mapping = page->mapping;
183 trace_mm_filemap_delete_from_page_cache(page);
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
189 if (PageUptodate(page) && PageMappedToDisk(page))
190 cleancache_put_page(page);
192 cleancache_invalidate_page(mapping, page);
194 page_cache_tree_delete(mapping, page, shadow);
196 page->mapping = NULL;
197 /* Leave page->index set: truncation lookup relies upon it */
199 __dec_zone_page_state(page, NR_FILE_PAGES);
200 if (PageSwapBacked(page))
201 __dec_zone_page_state(page, NR_SHMEM);
202 BUG_ON(page_mapped(page));
205 * Some filesystems seem to re-dirty the page even after
206 * the VM has canceled the dirty bit (eg ext3 journaling).
208 * Fix it up by doing a final dirty accounting check after
209 * having removed the page entirely.
211 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
212 dec_zone_page_state(page, NR_FILE_DIRTY);
213 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
218 * delete_from_page_cache - delete page from page cache
219 * @page: the page which the kernel is trying to remove from page cache
221 * This must be called only on pages that have been verified to be in the page
222 * cache and locked. It will never put the page into the free list, the caller
223 * has a reference on the page.
225 void delete_from_page_cache(struct page *page)
227 struct address_space *mapping = page->mapping;
228 void (*freepage)(struct page *);
230 BUG_ON(!PageLocked(page));
232 freepage = mapping->a_ops->freepage;
233 spin_lock_irq(&mapping->tree_lock);
234 __delete_from_page_cache(page, NULL);
235 spin_unlock_irq(&mapping->tree_lock);
236 mem_cgroup_uncharge_cache_page(page);
240 page_cache_release(page);
242 EXPORT_SYMBOL(delete_from_page_cache);
244 static int filemap_check_errors(struct address_space *mapping)
247 /* Check for outstanding write errors */
248 if (test_bit(AS_ENOSPC, &mapping->flags) &&
249 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
251 if (test_bit(AS_EIO, &mapping->flags) &&
252 test_and_clear_bit(AS_EIO, &mapping->flags))
258 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
259 * @mapping: address space structure to write
260 * @start: offset in bytes where the range starts
261 * @end: offset in bytes where the range ends (inclusive)
262 * @sync_mode: enable synchronous operation
264 * Start writeback against all of a mapping's dirty pages that lie
265 * within the byte offsets <start, end> inclusive.
267 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
268 * opposed to a regular memory cleansing writeback. The difference between
269 * these two operations is that if a dirty page/buffer is encountered, it must
270 * be waited upon, and not just skipped over.
272 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
273 loff_t end, int sync_mode)
276 struct writeback_control wbc = {
277 .sync_mode = sync_mode,
278 .nr_to_write = LONG_MAX,
279 .range_start = start,
283 if (!mapping_cap_writeback_dirty(mapping))
286 ret = do_writepages(mapping, &wbc);
290 static inline int __filemap_fdatawrite(struct address_space *mapping,
293 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
296 int filemap_fdatawrite(struct address_space *mapping)
298 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
300 EXPORT_SYMBOL(filemap_fdatawrite);
302 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
305 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
307 EXPORT_SYMBOL(filemap_fdatawrite_range);
310 * filemap_flush - mostly a non-blocking flush
311 * @mapping: target address_space
313 * This is a mostly non-blocking flush. Not suitable for data-integrity
314 * purposes - I/O may not be started against all dirty pages.
316 int filemap_flush(struct address_space *mapping)
318 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
320 EXPORT_SYMBOL(filemap_flush);
323 * filemap_fdatawait_range - wait for writeback to complete
324 * @mapping: address space structure to wait for
325 * @start_byte: offset in bytes where the range starts
326 * @end_byte: offset in bytes where the range ends (inclusive)
328 * Walk the list of under-writeback pages of the given address space
329 * in the given range and wait for all of them.
331 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
334 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
335 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
340 if (end_byte < start_byte)
343 pagevec_init(&pvec, 0);
344 while ((index <= end) &&
345 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
346 PAGECACHE_TAG_WRITEBACK,
347 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
350 for (i = 0; i < nr_pages; i++) {
351 struct page *page = pvec.pages[i];
353 /* until radix tree lookup accepts end_index */
354 if (page->index > end)
357 wait_on_page_writeback(page);
358 if (TestClearPageError(page))
361 pagevec_release(&pvec);
365 ret2 = filemap_check_errors(mapping);
371 EXPORT_SYMBOL(filemap_fdatawait_range);
374 * filemap_fdatawait - wait for all under-writeback pages to complete
375 * @mapping: address space structure to wait for
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
380 int filemap_fdatawait(struct address_space *mapping)
382 loff_t i_size = i_size_read(mapping->host);
387 return filemap_fdatawait_range(mapping, 0, i_size - 1);
389 EXPORT_SYMBOL(filemap_fdatawait);
391 int filemap_write_and_wait(struct address_space *mapping)
395 if (mapping->nrpages) {
396 err = filemap_fdatawrite(mapping);
398 * Even if the above returned error, the pages may be
399 * written partially (e.g. -ENOSPC), so we wait for it.
400 * But the -EIO is special case, it may indicate the worst
401 * thing (e.g. bug) happened, so we avoid waiting for it.
404 int err2 = filemap_fdatawait(mapping);
409 err = filemap_check_errors(mapping);
413 EXPORT_SYMBOL(filemap_write_and_wait);
416 * filemap_write_and_wait_range - write out & wait on a file range
417 * @mapping: the address_space for the pages
418 * @lstart: offset in bytes where the range starts
419 * @lend: offset in bytes where the range ends (inclusive)
421 * Write out and wait upon file offsets lstart->lend, inclusive.
423 * Note that `lend' is inclusive (describes the last byte to be written) so
424 * that this function can be used to write to the very end-of-file (end = -1).
426 int filemap_write_and_wait_range(struct address_space *mapping,
427 loff_t lstart, loff_t lend)
431 if (mapping->nrpages) {
432 err = __filemap_fdatawrite_range(mapping, lstart, lend,
434 /* See comment of filemap_write_and_wait() */
436 int err2 = filemap_fdatawait_range(mapping,
442 err = filemap_check_errors(mapping);
446 EXPORT_SYMBOL(filemap_write_and_wait_range);
449 * replace_page_cache_page - replace a pagecache page with a new one
450 * @old: page to be replaced
451 * @new: page to replace with
452 * @gfp_mask: allocation mode
454 * This function replaces a page in the pagecache with a new one. On
455 * success it acquires the pagecache reference for the new page and
456 * drops it for the old page. Both the old and new pages must be
457 * locked. This function does not add the new page to the LRU, the
458 * caller must do that.
460 * The remove + add is atomic. The only way this function can fail is
461 * memory allocation failure.
463 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
467 VM_BUG_ON_PAGE(!PageLocked(old), old);
468 VM_BUG_ON_PAGE(!PageLocked(new), new);
469 VM_BUG_ON_PAGE(new->mapping, new);
471 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
473 struct address_space *mapping = old->mapping;
474 void (*freepage)(struct page *);
476 pgoff_t offset = old->index;
477 freepage = mapping->a_ops->freepage;
480 new->mapping = mapping;
483 spin_lock_irq(&mapping->tree_lock);
484 __delete_from_page_cache(old, NULL);
485 error = radix_tree_insert(&mapping->page_tree, offset, new);
488 __inc_zone_page_state(new, NR_FILE_PAGES);
489 if (PageSwapBacked(new))
490 __inc_zone_page_state(new, NR_SHMEM);
491 spin_unlock_irq(&mapping->tree_lock);
492 /* mem_cgroup codes must not be called under tree_lock */
493 mem_cgroup_replace_page_cache(old, new);
494 radix_tree_preload_end();
497 page_cache_release(old);
502 EXPORT_SYMBOL_GPL(replace_page_cache_page);
504 static int page_cache_tree_insert(struct address_space *mapping,
505 struct page *page, void **shadowp)
507 struct radix_tree_node *node;
511 error = __radix_tree_create(&mapping->page_tree, page->index,
518 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
519 if (!radix_tree_exceptional_entry(p))
523 mapping->nrshadows--;
525 workingset_node_shadows_dec(node);
527 radix_tree_replace_slot(slot, page);
530 workingset_node_pages_inc(node);
532 * Don't track node that contains actual pages.
534 * Avoid acquiring the list_lru lock if already
535 * untracked. The list_empty() test is safe as
536 * node->private_list is protected by
537 * mapping->tree_lock.
539 if (!list_empty(&node->private_list))
540 list_lru_del(&workingset_shadow_nodes,
541 &node->private_list);
546 static int __add_to_page_cache_locked(struct page *page,
547 struct address_space *mapping,
548 pgoff_t offset, gfp_t gfp_mask,
553 VM_BUG_ON_PAGE(!PageLocked(page), page);
554 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
556 error = mem_cgroup_charge_file(page, current->mm,
557 gfp_mask & GFP_RECLAIM_MASK);
561 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
563 mem_cgroup_uncharge_cache_page(page);
567 page_cache_get(page);
568 page->mapping = mapping;
569 page->index = offset;
571 spin_lock_irq(&mapping->tree_lock);
572 error = page_cache_tree_insert(mapping, page, shadowp);
573 radix_tree_preload_end();
576 __inc_zone_page_state(page, NR_FILE_PAGES);
577 spin_unlock_irq(&mapping->tree_lock);
578 trace_mm_filemap_add_to_page_cache(page);
581 page->mapping = NULL;
582 /* Leave page->index set: truncation relies upon it */
583 spin_unlock_irq(&mapping->tree_lock);
584 mem_cgroup_uncharge_cache_page(page);
585 page_cache_release(page);
590 * add_to_page_cache_locked - add a locked page to the pagecache
592 * @mapping: the page's address_space
593 * @offset: page index
594 * @gfp_mask: page allocation mode
596 * This function is used to add a page to the pagecache. It must be locked.
597 * This function does not add the page to the LRU. The caller must do that.
599 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
600 pgoff_t offset, gfp_t gfp_mask)
602 return __add_to_page_cache_locked(page, mapping, offset,
605 EXPORT_SYMBOL(add_to_page_cache_locked);
607 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
608 pgoff_t offset, gfp_t gfp_mask)
613 __set_page_locked(page);
614 ret = __add_to_page_cache_locked(page, mapping, offset,
617 __clear_page_locked(page);
620 * The page might have been evicted from cache only
621 * recently, in which case it should be activated like
622 * any other repeatedly accessed page.
624 if (shadow && workingset_refault(shadow)) {
626 workingset_activation(page);
628 ClearPageActive(page);
633 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
636 struct page *__page_cache_alloc(gfp_t gfp)
641 if (cpuset_do_page_mem_spread()) {
642 unsigned int cpuset_mems_cookie;
644 cpuset_mems_cookie = read_mems_allowed_begin();
645 n = cpuset_mem_spread_node();
646 page = alloc_pages_exact_node(n, gfp, 0);
647 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
651 return alloc_pages(gfp, 0);
653 EXPORT_SYMBOL(__page_cache_alloc);
657 * In order to wait for pages to become available there must be
658 * waitqueues associated with pages. By using a hash table of
659 * waitqueues where the bucket discipline is to maintain all
660 * waiters on the same queue and wake all when any of the pages
661 * become available, and for the woken contexts to check to be
662 * sure the appropriate page became available, this saves space
663 * at a cost of "thundering herd" phenomena during rare hash
666 static wait_queue_head_t *page_waitqueue(struct page *page)
668 const struct zone *zone = page_zone(page);
670 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
673 static inline void wake_up_page(struct page *page, int bit)
675 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
678 void wait_on_page_bit(struct page *page, int bit_nr)
680 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
682 if (test_bit(bit_nr, &page->flags))
683 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
684 TASK_UNINTERRUPTIBLE);
686 EXPORT_SYMBOL(wait_on_page_bit);
688 int wait_on_page_bit_killable(struct page *page, int bit_nr)
690 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
692 if (!test_bit(bit_nr, &page->flags))
695 return __wait_on_bit(page_waitqueue(page), &wait,
696 bit_wait_io, TASK_KILLABLE);
700 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
701 * @page: Page defining the wait queue of interest
702 * @waiter: Waiter to add to the queue
704 * Add an arbitrary @waiter to the wait queue for the nominated @page.
706 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
708 wait_queue_head_t *q = page_waitqueue(page);
711 spin_lock_irqsave(&q->lock, flags);
712 __add_wait_queue(q, waiter);
713 spin_unlock_irqrestore(&q->lock, flags);
715 EXPORT_SYMBOL_GPL(add_page_wait_queue);
718 * unlock_page - unlock a locked page
721 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
722 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
723 * mechananism between PageLocked pages and PageWriteback pages is shared.
724 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
726 * The mb is necessary to enforce ordering between the clear_bit and the read
727 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
729 void unlock_page(struct page *page)
731 VM_BUG_ON_PAGE(!PageLocked(page), page);
732 clear_bit_unlock(PG_locked, &page->flags);
733 smp_mb__after_atomic();
734 wake_up_page(page, PG_locked);
736 EXPORT_SYMBOL(unlock_page);
739 * end_page_writeback - end writeback against a page
742 void end_page_writeback(struct page *page)
745 * TestClearPageReclaim could be used here but it is an atomic
746 * operation and overkill in this particular case. Failing to
747 * shuffle a page marked for immediate reclaim is too mild to
748 * justify taking an atomic operation penalty at the end of
749 * ever page writeback.
751 if (PageReclaim(page)) {
752 ClearPageReclaim(page);
753 rotate_reclaimable_page(page);
756 if (!test_clear_page_writeback(page))
759 smp_mb__after_atomic();
760 wake_up_page(page, PG_writeback);
762 EXPORT_SYMBOL(end_page_writeback);
765 * After completing I/O on a page, call this routine to update the page
766 * flags appropriately
768 void page_endio(struct page *page, int rw, int err)
772 SetPageUptodate(page);
774 ClearPageUptodate(page);
778 } else { /* rw == WRITE */
782 mapping_set_error(page->mapping, err);
784 end_page_writeback(page);
787 EXPORT_SYMBOL_GPL(page_endio);
790 * __lock_page - get a lock on the page, assuming we need to sleep to get it
791 * @page: the page to lock
793 void __lock_page(struct page *page)
795 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
797 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
798 TASK_UNINTERRUPTIBLE);
800 EXPORT_SYMBOL(__lock_page);
802 int __lock_page_killable(struct page *page)
804 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
806 return __wait_on_bit_lock(page_waitqueue(page), &wait,
807 bit_wait_io, TASK_KILLABLE);
809 EXPORT_SYMBOL_GPL(__lock_page_killable);
813 * 1 - page is locked; mmap_sem is still held.
814 * 0 - page is not locked.
815 * mmap_sem has been released (up_read()), unless flags had both
816 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
817 * which case mmap_sem is still held.
819 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
820 * with the page locked and the mmap_sem unperturbed.
822 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
825 if (flags & FAULT_FLAG_ALLOW_RETRY) {
827 * CAUTION! In this case, mmap_sem is not released
828 * even though return 0.
830 if (flags & FAULT_FLAG_RETRY_NOWAIT)
833 up_read(&mm->mmap_sem);
834 if (flags & FAULT_FLAG_KILLABLE)
835 wait_on_page_locked_killable(page);
837 wait_on_page_locked(page);
840 if (flags & FAULT_FLAG_KILLABLE) {
843 ret = __lock_page_killable(page);
845 up_read(&mm->mmap_sem);
855 * page_cache_next_hole - find the next hole (not-present entry)
858 * @max_scan: maximum range to search
860 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
861 * lowest indexed hole.
863 * Returns: the index of the hole if found, otherwise returns an index
864 * outside of the set specified (in which case 'return - index >=
865 * max_scan' will be true). In rare cases of index wrap-around, 0 will
868 * page_cache_next_hole may be called under rcu_read_lock. However,
869 * like radix_tree_gang_lookup, this will not atomically search a
870 * snapshot of the tree at a single point in time. For example, if a
871 * hole is created at index 5, then subsequently a hole is created at
872 * index 10, page_cache_next_hole covering both indexes may return 10
873 * if called under rcu_read_lock.
875 pgoff_t page_cache_next_hole(struct address_space *mapping,
876 pgoff_t index, unsigned long max_scan)
880 for (i = 0; i < max_scan; i++) {
883 page = radix_tree_lookup(&mapping->page_tree, index);
884 if (!page || radix_tree_exceptional_entry(page))
893 EXPORT_SYMBOL(page_cache_next_hole);
896 * page_cache_prev_hole - find the prev hole (not-present entry)
899 * @max_scan: maximum range to search
901 * Search backwards in the range [max(index-max_scan+1, 0), index] for
904 * Returns: the index of the hole if found, otherwise returns an index
905 * outside of the set specified (in which case 'index - return >=
906 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
909 * page_cache_prev_hole may be called under rcu_read_lock. However,
910 * like radix_tree_gang_lookup, this will not atomically search a
911 * snapshot of the tree at a single point in time. For example, if a
912 * hole is created at index 10, then subsequently a hole is created at
913 * index 5, page_cache_prev_hole covering both indexes may return 5 if
914 * called under rcu_read_lock.
916 pgoff_t page_cache_prev_hole(struct address_space *mapping,
917 pgoff_t index, unsigned long max_scan)
921 for (i = 0; i < max_scan; i++) {
924 page = radix_tree_lookup(&mapping->page_tree, index);
925 if (!page || radix_tree_exceptional_entry(page))
928 if (index == ULONG_MAX)
934 EXPORT_SYMBOL(page_cache_prev_hole);
937 * find_get_entry - find and get a page cache entry
938 * @mapping: the address_space to search
939 * @offset: the page cache index
941 * Looks up the page cache slot at @mapping & @offset. If there is a
942 * page cache page, it is returned with an increased refcount.
944 * If the slot holds a shadow entry of a previously evicted page, or a
945 * swap entry from shmem/tmpfs, it is returned.
947 * Otherwise, %NULL is returned.
949 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
957 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
959 page = radix_tree_deref_slot(pagep);
962 if (radix_tree_exception(page)) {
963 if (radix_tree_deref_retry(page))
966 * A shadow entry of a recently evicted page,
967 * or a swap entry from shmem/tmpfs. Return
968 * it without attempting to raise page count.
972 if (!page_cache_get_speculative(page))
976 * Has the page moved?
977 * This is part of the lockless pagecache protocol. See
978 * include/linux/pagemap.h for details.
980 if (unlikely(page != *pagep)) {
981 page_cache_release(page);
990 EXPORT_SYMBOL(find_get_entry);
993 * find_lock_entry - locate, pin and lock a page cache entry
994 * @mapping: the address_space to search
995 * @offset: the page cache index
997 * Looks up the page cache slot at @mapping & @offset. If there is a
998 * page cache page, it is returned locked and with an increased
1001 * If the slot holds a shadow entry of a previously evicted page, or a
1002 * swap entry from shmem/tmpfs, it is returned.
1004 * Otherwise, %NULL is returned.
1006 * find_lock_entry() may sleep.
1008 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1013 page = find_get_entry(mapping, offset);
1014 if (page && !radix_tree_exception(page)) {
1016 /* Has the page been truncated? */
1017 if (unlikely(page->mapping != mapping)) {
1019 page_cache_release(page);
1022 VM_BUG_ON_PAGE(page->index != offset, page);
1026 EXPORT_SYMBOL(find_lock_entry);
1029 * pagecache_get_page - find and get a page reference
1030 * @mapping: the address_space to search
1031 * @offset: the page index
1032 * @fgp_flags: PCG flags
1033 * @cache_gfp_mask: gfp mask to use for the page cache data page allocation
1034 * @radix_gfp_mask: gfp mask to use for radix tree node allocation
1036 * Looks up the page cache slot at @mapping & @offset.
1038 * PCG flags modify how the page is returned.
1040 * FGP_ACCESSED: the page will be marked accessed
1041 * FGP_LOCK: Page is return locked
1042 * FGP_CREAT: If page is not present then a new page is allocated using
1043 * @cache_gfp_mask and added to the page cache and the VM's LRU
1044 * list. If radix tree nodes are allocated during page cache
1045 * insertion then @radix_gfp_mask is used. The page is returned
1046 * locked and with an increased refcount. Otherwise, %NULL is
1049 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1050 * if the GFP flags specified for FGP_CREAT are atomic.
1052 * If there is a page cache page, it is returned with an increased refcount.
1054 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1055 int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask)
1060 page = find_get_entry(mapping, offset);
1061 if (radix_tree_exceptional_entry(page))
1066 if (fgp_flags & FGP_LOCK) {
1067 if (fgp_flags & FGP_NOWAIT) {
1068 if (!trylock_page(page)) {
1069 page_cache_release(page);
1076 /* Has the page been truncated? */
1077 if (unlikely(page->mapping != mapping)) {
1079 page_cache_release(page);
1082 VM_BUG_ON_PAGE(page->index != offset, page);
1085 if (page && (fgp_flags & FGP_ACCESSED))
1086 mark_page_accessed(page);
1089 if (!page && (fgp_flags & FGP_CREAT)) {
1091 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1092 cache_gfp_mask |= __GFP_WRITE;
1093 if (fgp_flags & FGP_NOFS) {
1094 cache_gfp_mask &= ~__GFP_FS;
1095 radix_gfp_mask &= ~__GFP_FS;
1098 page = __page_cache_alloc(cache_gfp_mask);
1102 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1103 fgp_flags |= FGP_LOCK;
1105 /* Init accessed so avoid atomic mark_page_accessed later */
1106 if (fgp_flags & FGP_ACCESSED)
1107 __SetPageReferenced(page);
1109 err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask);
1110 if (unlikely(err)) {
1111 page_cache_release(page);
1120 EXPORT_SYMBOL(pagecache_get_page);
1123 * find_get_entries - gang pagecache lookup
1124 * @mapping: The address_space to search
1125 * @start: The starting page cache index
1126 * @nr_entries: The maximum number of entries
1127 * @entries: Where the resulting entries are placed
1128 * @indices: The cache indices corresponding to the entries in @entries
1130 * find_get_entries() will search for and return a group of up to
1131 * @nr_entries entries in the mapping. The entries are placed at
1132 * @entries. find_get_entries() takes a reference against any actual
1135 * The search returns a group of mapping-contiguous page cache entries
1136 * with ascending indexes. There may be holes in the indices due to
1137 * not-present pages.
1139 * Any shadow entries of evicted pages, or swap entries from
1140 * shmem/tmpfs, are included in the returned array.
1142 * find_get_entries() returns the number of pages and shadow entries
1145 unsigned find_get_entries(struct address_space *mapping,
1146 pgoff_t start, unsigned int nr_entries,
1147 struct page **entries, pgoff_t *indices)
1150 unsigned int ret = 0;
1151 struct radix_tree_iter iter;
1158 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1161 page = radix_tree_deref_slot(slot);
1162 if (unlikely(!page))
1164 if (radix_tree_exception(page)) {
1165 if (radix_tree_deref_retry(page))
1168 * A shadow entry of a recently evicted page,
1169 * or a swap entry from shmem/tmpfs. Return
1170 * it without attempting to raise page count.
1174 if (!page_cache_get_speculative(page))
1177 /* Has the page moved? */
1178 if (unlikely(page != *slot)) {
1179 page_cache_release(page);
1183 indices[ret] = iter.index;
1184 entries[ret] = page;
1185 if (++ret == nr_entries)
1193 * find_get_pages - gang pagecache lookup
1194 * @mapping: The address_space to search
1195 * @start: The starting page index
1196 * @nr_pages: The maximum number of pages
1197 * @pages: Where the resulting pages are placed
1199 * find_get_pages() will search for and return a group of up to
1200 * @nr_pages pages in the mapping. The pages are placed at @pages.
1201 * find_get_pages() takes a reference against the returned pages.
1203 * The search returns a group of mapping-contiguous pages with ascending
1204 * indexes. There may be holes in the indices due to not-present pages.
1206 * find_get_pages() returns the number of pages which were found.
1208 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1209 unsigned int nr_pages, struct page **pages)
1211 struct radix_tree_iter iter;
1215 if (unlikely(!nr_pages))
1220 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1223 page = radix_tree_deref_slot(slot);
1224 if (unlikely(!page))
1227 if (radix_tree_exception(page)) {
1228 if (radix_tree_deref_retry(page)) {
1230 * Transient condition which can only trigger
1231 * when entry at index 0 moves out of or back
1232 * to root: none yet gotten, safe to restart.
1234 WARN_ON(iter.index);
1238 * A shadow entry of a recently evicted page,
1239 * or a swap entry from shmem/tmpfs. Skip
1245 if (!page_cache_get_speculative(page))
1248 /* Has the page moved? */
1249 if (unlikely(page != *slot)) {
1250 page_cache_release(page);
1255 if (++ret == nr_pages)
1264 * find_get_pages_contig - gang contiguous pagecache lookup
1265 * @mapping: The address_space to search
1266 * @index: The starting page index
1267 * @nr_pages: The maximum number of pages
1268 * @pages: Where the resulting pages are placed
1270 * find_get_pages_contig() works exactly like find_get_pages(), except
1271 * that the returned number of pages are guaranteed to be contiguous.
1273 * find_get_pages_contig() returns the number of pages which were found.
1275 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1276 unsigned int nr_pages, struct page **pages)
1278 struct radix_tree_iter iter;
1280 unsigned int ret = 0;
1282 if (unlikely(!nr_pages))
1287 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1290 page = radix_tree_deref_slot(slot);
1291 /* The hole, there no reason to continue */
1292 if (unlikely(!page))
1295 if (radix_tree_exception(page)) {
1296 if (radix_tree_deref_retry(page)) {
1298 * Transient condition which can only trigger
1299 * when entry at index 0 moves out of or back
1300 * to root: none yet gotten, safe to restart.
1305 * A shadow entry of a recently evicted page,
1306 * or a swap entry from shmem/tmpfs. Stop
1307 * looking for contiguous pages.
1312 if (!page_cache_get_speculative(page))
1315 /* Has the page moved? */
1316 if (unlikely(page != *slot)) {
1317 page_cache_release(page);
1322 * must check mapping and index after taking the ref.
1323 * otherwise we can get both false positives and false
1324 * negatives, which is just confusing to the caller.
1326 if (page->mapping == NULL || page->index != iter.index) {
1327 page_cache_release(page);
1332 if (++ret == nr_pages)
1338 EXPORT_SYMBOL(find_get_pages_contig);
1341 * find_get_pages_tag - find and return pages that match @tag
1342 * @mapping: the address_space to search
1343 * @index: the starting page index
1344 * @tag: the tag index
1345 * @nr_pages: the maximum number of pages
1346 * @pages: where the resulting pages are placed
1348 * Like find_get_pages, except we only return pages which are tagged with
1349 * @tag. We update @index to index the next page for the traversal.
1351 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1352 int tag, unsigned int nr_pages, struct page **pages)
1354 struct radix_tree_iter iter;
1358 if (unlikely(!nr_pages))
1363 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1364 &iter, *index, tag) {
1367 page = radix_tree_deref_slot(slot);
1368 if (unlikely(!page))
1371 if (radix_tree_exception(page)) {
1372 if (radix_tree_deref_retry(page)) {
1374 * Transient condition which can only trigger
1375 * when entry at index 0 moves out of or back
1376 * to root: none yet gotten, safe to restart.
1381 * A shadow entry of a recently evicted page.
1383 * Those entries should never be tagged, but
1384 * this tree walk is lockless and the tags are
1385 * looked up in bulk, one radix tree node at a
1386 * time, so there is a sizable window for page
1387 * reclaim to evict a page we saw tagged.
1394 if (!page_cache_get_speculative(page))
1397 /* Has the page moved? */
1398 if (unlikely(page != *slot)) {
1399 page_cache_release(page);
1404 if (++ret == nr_pages)
1411 *index = pages[ret - 1]->index + 1;
1415 EXPORT_SYMBOL(find_get_pages_tag);
1418 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1419 * a _large_ part of the i/o request. Imagine the worst scenario:
1421 * ---R__________________________________________B__________
1422 * ^ reading here ^ bad block(assume 4k)
1424 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1425 * => failing the whole request => read(R) => read(R+1) =>
1426 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1427 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1428 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1430 * It is going insane. Fix it by quickly scaling down the readahead size.
1432 static void shrink_readahead_size_eio(struct file *filp,
1433 struct file_ra_state *ra)
1439 * do_generic_file_read - generic file read routine
1440 * @filp: the file to read
1441 * @ppos: current file position
1442 * @iter: data destination
1443 * @written: already copied
1445 * This is a generic file read routine, and uses the
1446 * mapping->a_ops->readpage() function for the actual low-level stuff.
1448 * This is really ugly. But the goto's actually try to clarify some
1449 * of the logic when it comes to error handling etc.
1451 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1452 struct iov_iter *iter, ssize_t written)
1454 struct address_space *mapping = filp->f_mapping;
1455 struct inode *inode = mapping->host;
1456 struct file_ra_state *ra = &filp->f_ra;
1460 unsigned long offset; /* offset into pagecache page */
1461 unsigned int prev_offset;
1464 index = *ppos >> PAGE_CACHE_SHIFT;
1465 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1466 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1467 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1468 offset = *ppos & ~PAGE_CACHE_MASK;
1474 unsigned long nr, ret;
1478 page = find_get_page(mapping, index);
1480 page_cache_sync_readahead(mapping,
1482 index, last_index - index);
1483 page = find_get_page(mapping, index);
1484 if (unlikely(page == NULL))
1485 goto no_cached_page;
1487 if (PageReadahead(page)) {
1488 page_cache_async_readahead(mapping,
1490 index, last_index - index);
1492 if (!PageUptodate(page)) {
1493 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1494 !mapping->a_ops->is_partially_uptodate)
1495 goto page_not_up_to_date;
1496 if (!trylock_page(page))
1497 goto page_not_up_to_date;
1498 /* Did it get truncated before we got the lock? */
1500 goto page_not_up_to_date_locked;
1501 if (!mapping->a_ops->is_partially_uptodate(page,
1502 offset, iter->count))
1503 goto page_not_up_to_date_locked;
1508 * i_size must be checked after we know the page is Uptodate.
1510 * Checking i_size after the check allows us to calculate
1511 * the correct value for "nr", which means the zero-filled
1512 * part of the page is not copied back to userspace (unless
1513 * another truncate extends the file - this is desired though).
1516 isize = i_size_read(inode);
1517 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1518 if (unlikely(!isize || index > end_index)) {
1519 page_cache_release(page);
1523 /* nr is the maximum number of bytes to copy from this page */
1524 nr = PAGE_CACHE_SIZE;
1525 if (index == end_index) {
1526 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1528 page_cache_release(page);
1534 /* If users can be writing to this page using arbitrary
1535 * virtual addresses, take care about potential aliasing
1536 * before reading the page on the kernel side.
1538 if (mapping_writably_mapped(mapping))
1539 flush_dcache_page(page);
1542 * When a sequential read accesses a page several times,
1543 * only mark it as accessed the first time.
1545 if (prev_index != index || offset != prev_offset)
1546 mark_page_accessed(page);
1550 * Ok, we have the page, and it's up-to-date, so
1551 * now we can copy it to user space...
1554 ret = copy_page_to_iter(page, offset, nr, iter);
1556 index += offset >> PAGE_CACHE_SHIFT;
1557 offset &= ~PAGE_CACHE_MASK;
1558 prev_offset = offset;
1560 page_cache_release(page);
1562 if (!iov_iter_count(iter))
1570 page_not_up_to_date:
1571 /* Get exclusive access to the page ... */
1572 error = lock_page_killable(page);
1573 if (unlikely(error))
1574 goto readpage_error;
1576 page_not_up_to_date_locked:
1577 /* Did it get truncated before we got the lock? */
1578 if (!page->mapping) {
1580 page_cache_release(page);
1584 /* Did somebody else fill it already? */
1585 if (PageUptodate(page)) {
1592 * A previous I/O error may have been due to temporary
1593 * failures, eg. multipath errors.
1594 * PG_error will be set again if readpage fails.
1596 ClearPageError(page);
1597 /* Start the actual read. The read will unlock the page. */
1598 error = mapping->a_ops->readpage(filp, page);
1600 if (unlikely(error)) {
1601 if (error == AOP_TRUNCATED_PAGE) {
1602 page_cache_release(page);
1606 goto readpage_error;
1609 if (!PageUptodate(page)) {
1610 error = lock_page_killable(page);
1611 if (unlikely(error))
1612 goto readpage_error;
1613 if (!PageUptodate(page)) {
1614 if (page->mapping == NULL) {
1616 * invalidate_mapping_pages got it
1619 page_cache_release(page);
1623 shrink_readahead_size_eio(filp, ra);
1625 goto readpage_error;
1633 /* UHHUH! A synchronous read error occurred. Report it */
1634 page_cache_release(page);
1639 * Ok, it wasn't cached, so we need to create a new
1642 page = page_cache_alloc_cold(mapping);
1647 error = add_to_page_cache_lru(page, mapping,
1650 page_cache_release(page);
1651 if (error == -EEXIST) {
1661 ra->prev_pos = prev_index;
1662 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1663 ra->prev_pos |= prev_offset;
1665 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1666 file_accessed(filp);
1667 return written ? written : error;
1671 * generic_file_read_iter - generic filesystem read routine
1672 * @iocb: kernel I/O control block
1673 * @iter: destination for the data read
1675 * This is the "read_iter()" routine for all filesystems
1676 * that can use the page cache directly.
1679 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1681 struct file *file = iocb->ki_filp;
1683 loff_t *ppos = &iocb->ki_pos;
1686 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1687 if (file->f_flags & O_DIRECT) {
1688 struct address_space *mapping = file->f_mapping;
1689 struct inode *inode = mapping->host;
1690 size_t count = iov_iter_count(iter);
1694 goto out; /* skip atime */
1695 size = i_size_read(inode);
1696 retval = filemap_write_and_wait_range(mapping, pos,
1699 struct iov_iter data = *iter;
1700 retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos);
1704 *ppos = pos + retval;
1705 iov_iter_advance(iter, retval);
1709 * Btrfs can have a short DIO read if we encounter
1710 * compressed extents, so if there was an error, or if
1711 * we've already read everything we wanted to, or if
1712 * there was a short read because we hit EOF, go ahead
1713 * and return. Otherwise fallthrough to buffered io for
1714 * the rest of the read.
1716 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) {
1717 file_accessed(file);
1722 retval = do_generic_file_read(file, ppos, iter, retval);
1726 EXPORT_SYMBOL(generic_file_read_iter);
1730 * page_cache_read - adds requested page to the page cache if not already there
1731 * @file: file to read
1732 * @offset: page index
1734 * This adds the requested page to the page cache if it isn't already there,
1735 * and schedules an I/O to read in its contents from disk.
1737 static int page_cache_read(struct file *file, pgoff_t offset)
1739 struct address_space *mapping = file->f_mapping;
1744 page = page_cache_alloc_cold(mapping);
1748 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1750 ret = mapping->a_ops->readpage(file, page);
1751 else if (ret == -EEXIST)
1752 ret = 0; /* losing race to add is OK */
1754 page_cache_release(page);
1756 } while (ret == AOP_TRUNCATED_PAGE);
1761 #define MMAP_LOTSAMISS (100)
1764 * Synchronous readahead happens when we don't even find
1765 * a page in the page cache at all.
1767 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1768 struct file_ra_state *ra,
1772 unsigned long ra_pages;
1773 struct address_space *mapping = file->f_mapping;
1775 /* If we don't want any read-ahead, don't bother */
1776 if (vma->vm_flags & VM_RAND_READ)
1781 if (vma->vm_flags & VM_SEQ_READ) {
1782 page_cache_sync_readahead(mapping, ra, file, offset,
1787 /* Avoid banging the cache line if not needed */
1788 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1792 * Do we miss much more than hit in this file? If so,
1793 * stop bothering with read-ahead. It will only hurt.
1795 if (ra->mmap_miss > MMAP_LOTSAMISS)
1801 ra_pages = max_sane_readahead(ra->ra_pages);
1802 ra->start = max_t(long, 0, offset - ra_pages / 2);
1803 ra->size = ra_pages;
1804 ra->async_size = ra_pages / 4;
1805 ra_submit(ra, mapping, file);
1809 * Asynchronous readahead happens when we find the page and PG_readahead,
1810 * so we want to possibly extend the readahead further..
1812 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1813 struct file_ra_state *ra,
1818 struct address_space *mapping = file->f_mapping;
1820 /* If we don't want any read-ahead, don't bother */
1821 if (vma->vm_flags & VM_RAND_READ)
1823 if (ra->mmap_miss > 0)
1825 if (PageReadahead(page))
1826 page_cache_async_readahead(mapping, ra, file,
1827 page, offset, ra->ra_pages);
1831 * filemap_fault - read in file data for page fault handling
1832 * @vma: vma in which the fault was taken
1833 * @vmf: struct vm_fault containing details of the fault
1835 * filemap_fault() is invoked via the vma operations vector for a
1836 * mapped memory region to read in file data during a page fault.
1838 * The goto's are kind of ugly, but this streamlines the normal case of having
1839 * it in the page cache, and handles the special cases reasonably without
1840 * having a lot of duplicated code.
1842 * vma->vm_mm->mmap_sem must be held on entry.
1844 * If our return value has VM_FAULT_RETRY set, it's because
1845 * lock_page_or_retry() returned 0.
1846 * The mmap_sem has usually been released in this case.
1847 * See __lock_page_or_retry() for the exception.
1849 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1850 * has not been released.
1852 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1854 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1857 struct file *file = vma->vm_file;
1858 struct address_space *mapping = file->f_mapping;
1859 struct file_ra_state *ra = &file->f_ra;
1860 struct inode *inode = mapping->host;
1861 pgoff_t offset = vmf->pgoff;
1866 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1867 if (offset >= size >> PAGE_CACHE_SHIFT)
1868 return VM_FAULT_SIGBUS;
1871 * Do we have something in the page cache already?
1873 page = find_get_page(mapping, offset);
1874 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1876 * We found the page, so try async readahead before
1877 * waiting for the lock.
1879 do_async_mmap_readahead(vma, ra, file, page, offset);
1881 /* No page in the page cache at all */
1882 do_sync_mmap_readahead(vma, ra, file, offset);
1883 count_vm_event(PGMAJFAULT);
1884 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1885 ret = VM_FAULT_MAJOR;
1887 page = find_get_page(mapping, offset);
1889 goto no_cached_page;
1892 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1893 page_cache_release(page);
1894 return ret | VM_FAULT_RETRY;
1897 /* Did it get truncated? */
1898 if (unlikely(page->mapping != mapping)) {
1903 VM_BUG_ON_PAGE(page->index != offset, page);
1906 * We have a locked page in the page cache, now we need to check
1907 * that it's up-to-date. If not, it is going to be due to an error.
1909 if (unlikely(!PageUptodate(page)))
1910 goto page_not_uptodate;
1913 * Found the page and have a reference on it.
1914 * We must recheck i_size under page lock.
1916 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1917 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1919 page_cache_release(page);
1920 return VM_FAULT_SIGBUS;
1924 return ret | VM_FAULT_LOCKED;
1928 * We're only likely to ever get here if MADV_RANDOM is in
1931 error = page_cache_read(file, offset);
1934 * The page we want has now been added to the page cache.
1935 * In the unlikely event that someone removed it in the
1936 * meantime, we'll just come back here and read it again.
1942 * An error return from page_cache_read can result if the
1943 * system is low on memory, or a problem occurs while trying
1946 if (error == -ENOMEM)
1947 return VM_FAULT_OOM;
1948 return VM_FAULT_SIGBUS;
1952 * Umm, take care of errors if the page isn't up-to-date.
1953 * Try to re-read it _once_. We do this synchronously,
1954 * because there really aren't any performance issues here
1955 * and we need to check for errors.
1957 ClearPageError(page);
1958 error = mapping->a_ops->readpage(file, page);
1960 wait_on_page_locked(page);
1961 if (!PageUptodate(page))
1964 page_cache_release(page);
1966 if (!error || error == AOP_TRUNCATED_PAGE)
1969 /* Things didn't work out. Return zero to tell the mm layer so. */
1970 shrink_readahead_size_eio(file, ra);
1971 return VM_FAULT_SIGBUS;
1973 EXPORT_SYMBOL(filemap_fault);
1975 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1977 struct radix_tree_iter iter;
1979 struct file *file = vma->vm_file;
1980 struct address_space *mapping = file->f_mapping;
1983 unsigned long address = (unsigned long) vmf->virtual_address;
1988 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
1989 if (iter.index > vmf->max_pgoff)
1992 page = radix_tree_deref_slot(slot);
1993 if (unlikely(!page))
1995 if (radix_tree_exception(page)) {
1996 if (radix_tree_deref_retry(page))
2002 if (!page_cache_get_speculative(page))
2005 /* Has the page moved? */
2006 if (unlikely(page != *slot)) {
2007 page_cache_release(page);
2011 if (!PageUptodate(page) ||
2012 PageReadahead(page) ||
2015 if (!trylock_page(page))
2018 if (page->mapping != mapping || !PageUptodate(page))
2021 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2022 if (page->index >= size >> PAGE_CACHE_SHIFT)
2025 pte = vmf->pte + page->index - vmf->pgoff;
2026 if (!pte_none(*pte))
2029 if (file->f_ra.mmap_miss > 0)
2030 file->f_ra.mmap_miss--;
2031 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2032 do_set_pte(vma, addr, page, pte, false, false);
2038 page_cache_release(page);
2040 if (iter.index == vmf->max_pgoff)
2045 EXPORT_SYMBOL(filemap_map_pages);
2047 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2049 struct page *page = vmf->page;
2050 struct inode *inode = file_inode(vma->vm_file);
2051 int ret = VM_FAULT_LOCKED;
2053 sb_start_pagefault(inode->i_sb);
2054 file_update_time(vma->vm_file);
2056 if (page->mapping != inode->i_mapping) {
2058 ret = VM_FAULT_NOPAGE;
2062 * We mark the page dirty already here so that when freeze is in
2063 * progress, we are guaranteed that writeback during freezing will
2064 * see the dirty page and writeprotect it again.
2066 set_page_dirty(page);
2067 wait_for_stable_page(page);
2069 sb_end_pagefault(inode->i_sb);
2072 EXPORT_SYMBOL(filemap_page_mkwrite);
2074 const struct vm_operations_struct generic_file_vm_ops = {
2075 .fault = filemap_fault,
2076 .map_pages = filemap_map_pages,
2077 .page_mkwrite = filemap_page_mkwrite,
2078 .remap_pages = generic_file_remap_pages,
2081 /* This is used for a general mmap of a disk file */
2083 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2085 struct address_space *mapping = file->f_mapping;
2087 if (!mapping->a_ops->readpage)
2089 file_accessed(file);
2090 vma->vm_ops = &generic_file_vm_ops;
2095 * This is for filesystems which do not implement ->writepage.
2097 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2099 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2101 return generic_file_mmap(file, vma);
2104 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2108 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2112 #endif /* CONFIG_MMU */
2114 EXPORT_SYMBOL(generic_file_mmap);
2115 EXPORT_SYMBOL(generic_file_readonly_mmap);
2117 static struct page *wait_on_page_read(struct page *page)
2119 if (!IS_ERR(page)) {
2120 wait_on_page_locked(page);
2121 if (!PageUptodate(page)) {
2122 page_cache_release(page);
2123 page = ERR_PTR(-EIO);
2129 static struct page *__read_cache_page(struct address_space *mapping,
2131 int (*filler)(void *, struct page *),
2138 page = find_get_page(mapping, index);
2140 page = __page_cache_alloc(gfp | __GFP_COLD);
2142 return ERR_PTR(-ENOMEM);
2143 err = add_to_page_cache_lru(page, mapping, index, gfp);
2144 if (unlikely(err)) {
2145 page_cache_release(page);
2148 /* Presumably ENOMEM for radix tree node */
2149 return ERR_PTR(err);
2151 err = filler(data, page);
2153 page_cache_release(page);
2154 page = ERR_PTR(err);
2156 page = wait_on_page_read(page);
2162 static struct page *do_read_cache_page(struct address_space *mapping,
2164 int (*filler)(void *, struct page *),
2173 page = __read_cache_page(mapping, index, filler, data, gfp);
2176 if (PageUptodate(page))
2180 if (!page->mapping) {
2182 page_cache_release(page);
2185 if (PageUptodate(page)) {
2189 err = filler(data, page);
2191 page_cache_release(page);
2192 return ERR_PTR(err);
2194 page = wait_on_page_read(page);
2199 mark_page_accessed(page);
2204 * read_cache_page - read into page cache, fill it if needed
2205 * @mapping: the page's address_space
2206 * @index: the page index
2207 * @filler: function to perform the read
2208 * @data: first arg to filler(data, page) function, often left as NULL
2210 * Read into the page cache. If a page already exists, and PageUptodate() is
2211 * not set, try to fill the page and wait for it to become unlocked.
2213 * If the page does not get brought uptodate, return -EIO.
2215 struct page *read_cache_page(struct address_space *mapping,
2217 int (*filler)(void *, struct page *),
2220 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2222 EXPORT_SYMBOL(read_cache_page);
2225 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2226 * @mapping: the page's address_space
2227 * @index: the page index
2228 * @gfp: the page allocator flags to use if allocating
2230 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2231 * any new page allocations done using the specified allocation flags.
2233 * If the page does not get brought uptodate, return -EIO.
2235 struct page *read_cache_page_gfp(struct address_space *mapping,
2239 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2241 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2243 EXPORT_SYMBOL(read_cache_page_gfp);
2246 * Performs necessary checks before doing a write
2248 * Can adjust writing position or amount of bytes to write.
2249 * Returns appropriate error code that caller should return or
2250 * zero in case that write should be allowed.
2252 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2254 struct inode *inode = file->f_mapping->host;
2255 unsigned long limit = rlimit(RLIMIT_FSIZE);
2257 if (unlikely(*pos < 0))
2261 /* FIXME: this is for backwards compatibility with 2.4 */
2262 if (file->f_flags & O_APPEND)
2263 *pos = i_size_read(inode);
2265 if (limit != RLIM_INFINITY) {
2266 if (*pos >= limit) {
2267 send_sig(SIGXFSZ, current, 0);
2270 if (*count > limit - (typeof(limit))*pos) {
2271 *count = limit - (typeof(limit))*pos;
2279 if (unlikely(*pos + *count > MAX_NON_LFS &&
2280 !(file->f_flags & O_LARGEFILE))) {
2281 if (*pos >= MAX_NON_LFS) {
2284 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2285 *count = MAX_NON_LFS - (unsigned long)*pos;
2290 * Are we about to exceed the fs block limit ?
2292 * If we have written data it becomes a short write. If we have
2293 * exceeded without writing data we send a signal and return EFBIG.
2294 * Linus frestrict idea will clean these up nicely..
2296 if (likely(!isblk)) {
2297 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2298 if (*count || *pos > inode->i_sb->s_maxbytes) {
2301 /* zero-length writes at ->s_maxbytes are OK */
2304 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2305 *count = inode->i_sb->s_maxbytes - *pos;
2309 if (bdev_read_only(I_BDEV(inode)))
2311 isize = i_size_read(inode);
2312 if (*pos >= isize) {
2313 if (*count || *pos > isize)
2317 if (*pos + *count > isize)
2318 *count = isize - *pos;
2325 EXPORT_SYMBOL(generic_write_checks);
2327 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2328 loff_t pos, unsigned len, unsigned flags,
2329 struct page **pagep, void **fsdata)
2331 const struct address_space_operations *aops = mapping->a_ops;
2333 return aops->write_begin(file, mapping, pos, len, flags,
2336 EXPORT_SYMBOL(pagecache_write_begin);
2338 int pagecache_write_end(struct file *file, struct address_space *mapping,
2339 loff_t pos, unsigned len, unsigned copied,
2340 struct page *page, void *fsdata)
2342 const struct address_space_operations *aops = mapping->a_ops;
2344 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2346 EXPORT_SYMBOL(pagecache_write_end);
2349 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2351 struct file *file = iocb->ki_filp;
2352 struct address_space *mapping = file->f_mapping;
2353 struct inode *inode = mapping->host;
2357 struct iov_iter data;
2359 write_len = iov_iter_count(from);
2360 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2362 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2367 * After a write we want buffered reads to be sure to go to disk to get
2368 * the new data. We invalidate clean cached page from the region we're
2369 * about to write. We do this *before* the write so that we can return
2370 * without clobbering -EIOCBQUEUED from ->direct_IO().
2372 if (mapping->nrpages) {
2373 written = invalidate_inode_pages2_range(mapping,
2374 pos >> PAGE_CACHE_SHIFT, end);
2376 * If a page can not be invalidated, return 0 to fall back
2377 * to buffered write.
2380 if (written == -EBUSY)
2387 written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos);
2390 * Finally, try again to invalidate clean pages which might have been
2391 * cached by non-direct readahead, or faulted in by get_user_pages()
2392 * if the source of the write was an mmap'ed region of the file
2393 * we're writing. Either one is a pretty crazy thing to do,
2394 * so we don't support it 100%. If this invalidation
2395 * fails, tough, the write still worked...
2397 if (mapping->nrpages) {
2398 invalidate_inode_pages2_range(mapping,
2399 pos >> PAGE_CACHE_SHIFT, end);
2404 iov_iter_advance(from, written);
2405 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2406 i_size_write(inode, pos);
2407 mark_inode_dirty(inode);
2414 EXPORT_SYMBOL(generic_file_direct_write);
2417 * Find or create a page at the given pagecache position. Return the locked
2418 * page. This function is specifically for buffered writes.
2420 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2421 pgoff_t index, unsigned flags)
2424 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2426 if (flags & AOP_FLAG_NOFS)
2427 fgp_flags |= FGP_NOFS;
2429 page = pagecache_get_page(mapping, index, fgp_flags,
2430 mapping_gfp_mask(mapping),
2433 wait_for_stable_page(page);
2437 EXPORT_SYMBOL(grab_cache_page_write_begin);
2439 ssize_t generic_perform_write(struct file *file,
2440 struct iov_iter *i, loff_t pos)
2442 struct address_space *mapping = file->f_mapping;
2443 const struct address_space_operations *a_ops = mapping->a_ops;
2445 ssize_t written = 0;
2446 unsigned int flags = 0;
2449 * Copies from kernel address space cannot fail (NFSD is a big user).
2451 if (segment_eq(get_fs(), KERNEL_DS))
2452 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2456 unsigned long offset; /* Offset into pagecache page */
2457 unsigned long bytes; /* Bytes to write to page */
2458 size_t copied; /* Bytes copied from user */
2461 offset = (pos & (PAGE_CACHE_SIZE - 1));
2462 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2467 * Bring in the user page that we will copy from _first_.
2468 * Otherwise there's a nasty deadlock on copying from the
2469 * same page as we're writing to, without it being marked
2472 * Not only is this an optimisation, but it is also required
2473 * to check that the address is actually valid, when atomic
2474 * usercopies are used, below.
2476 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2481 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2483 if (unlikely(status < 0))
2486 if (mapping_writably_mapped(mapping))
2487 flush_dcache_page(page);
2489 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2490 flush_dcache_page(page);
2492 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2494 if (unlikely(status < 0))
2500 iov_iter_advance(i, copied);
2501 if (unlikely(copied == 0)) {
2503 * If we were unable to copy any data at all, we must
2504 * fall back to a single segment length write.
2506 * If we didn't fallback here, we could livelock
2507 * because not all segments in the iov can be copied at
2508 * once without a pagefault.
2510 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2511 iov_iter_single_seg_count(i));
2517 balance_dirty_pages_ratelimited(mapping);
2518 if (fatal_signal_pending(current)) {
2522 } while (iov_iter_count(i));
2524 return written ? written : status;
2526 EXPORT_SYMBOL(generic_perform_write);
2529 * __generic_file_write_iter - write data to a file
2530 * @iocb: IO state structure (file, offset, etc.)
2531 * @from: iov_iter with data to write
2533 * This function does all the work needed for actually writing data to a
2534 * file. It does all basic checks, removes SUID from the file, updates
2535 * modification times and calls proper subroutines depending on whether we
2536 * do direct IO or a standard buffered write.
2538 * It expects i_mutex to be grabbed unless we work on a block device or similar
2539 * object which does not need locking at all.
2541 * This function does *not* take care of syncing data in case of O_SYNC write.
2542 * A caller has to handle it. This is mainly due to the fact that we want to
2543 * avoid syncing under i_mutex.
2545 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2547 struct file *file = iocb->ki_filp;
2548 struct address_space * mapping = file->f_mapping;
2549 struct inode *inode = mapping->host;
2550 loff_t pos = iocb->ki_pos;
2551 ssize_t written = 0;
2554 size_t count = iov_iter_count(from);
2556 /* We can write back this queue in page reclaim */
2557 current->backing_dev_info = mapping->backing_dev_info;
2558 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2565 iov_iter_truncate(from, count);
2567 err = file_remove_suid(file);
2571 err = file_update_time(file);
2575 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2576 if (unlikely(file->f_flags & O_DIRECT)) {
2579 written = generic_file_direct_write(iocb, from, pos);
2580 if (written < 0 || written == count)
2584 * direct-io write to a hole: fall through to buffered I/O
2585 * for completing the rest of the request.
2590 status = generic_perform_write(file, from, pos);
2592 * If generic_perform_write() returned a synchronous error
2593 * then we want to return the number of bytes which were
2594 * direct-written, or the error code if that was zero. Note
2595 * that this differs from normal direct-io semantics, which
2596 * will return -EFOO even if some bytes were written.
2598 if (unlikely(status < 0) && !written) {
2602 iocb->ki_pos = pos + status;
2604 * We need to ensure that the page cache pages are written to
2605 * disk and invalidated to preserve the expected O_DIRECT
2608 endbyte = pos + status - 1;
2609 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2612 invalidate_mapping_pages(mapping,
2613 pos >> PAGE_CACHE_SHIFT,
2614 endbyte >> PAGE_CACHE_SHIFT);
2617 * We don't know how much we wrote, so just return
2618 * the number of bytes which were direct-written
2622 written = generic_perform_write(file, from, pos);
2623 if (likely(written >= 0))
2624 iocb->ki_pos = pos + written;
2627 current->backing_dev_info = NULL;
2628 return written ? written : err;
2630 EXPORT_SYMBOL(__generic_file_write_iter);
2633 * generic_file_write_iter - write data to a file
2634 * @iocb: IO state structure
2635 * @from: iov_iter with data to write
2637 * This is a wrapper around __generic_file_write_iter() to be used by most
2638 * filesystems. It takes care of syncing the file in case of O_SYNC file
2639 * and acquires i_mutex as needed.
2641 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2643 struct file *file = iocb->ki_filp;
2644 struct inode *inode = file->f_mapping->host;
2647 mutex_lock(&inode->i_mutex);
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);