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>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/filemap.h>
42 * FIXME: remove all knowledge of the buffer layer from the core VM
44 #include <linux/buffer_head.h> /* for try_to_free_buffers */
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * Shared mappings now work. 15.8.1995 Bruno.
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
63 * ->i_mmap_mutex (truncate_pagecache)
64 * ->private_lock (__free_pte->__set_page_dirty_buffers)
65 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
69 * ->i_mmap_mutex (truncate->unmap_mapping_range)
73 * ->page_table_lock or pte_lock (various, mainly in memory.c)
74 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 * ->lock_page (access_process_vm)
79 * ->i_mutex (generic_file_buffered_write)
80 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
83 * sb_lock (fs/fs-writeback.c)
84 * ->mapping->tree_lock (__sync_single_inode)
87 * ->anon_vma.lock (vma_adjust)
90 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
92 * ->page_table_lock or pte_lock
93 * ->swap_lock (try_to_unmap_one)
94 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
97 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
98 * ->private_lock (page_remove_rmap->set_page_dirty)
99 * ->tree_lock (page_remove_rmap->set_page_dirty)
100 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
101 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
103 * ->inode->i_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 * Delete a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold the mapping's tree_lock.
115 void __delete_from_page_cache(struct page *page)
117 struct address_space *mapping = page->mapping;
119 trace_mm_filemap_delete_from_page_cache(page);
121 * if we're uptodate, flush out into the cleancache, otherwise
122 * invalidate any existing cleancache entries. We can't leave
123 * stale data around in the cleancache once our page is gone
125 if (PageUptodate(page) && PageMappedToDisk(page))
126 cleancache_put_page(page);
128 cleancache_invalidate_page(mapping, page);
130 radix_tree_delete(&mapping->page_tree, page->index);
131 page->mapping = NULL;
132 /* Leave page->index set: truncation lookup relies upon it */
134 __dec_zone_page_state(page, NR_FILE_PAGES);
135 if (PageSwapBacked(page))
136 __dec_zone_page_state(page, NR_SHMEM);
137 BUG_ON(page_mapped(page));
140 * Some filesystems seem to re-dirty the page even after
141 * the VM has canceled the dirty bit (eg ext3 journaling).
143 * Fix it up by doing a final dirty accounting check after
144 * having removed the page entirely.
146 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
147 dec_zone_page_state(page, NR_FILE_DIRTY);
148 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
153 * delete_from_page_cache - delete page from page cache
154 * @page: the page which the kernel is trying to remove from page cache
156 * This must be called only on pages that have been verified to be in the page
157 * cache and locked. It will never put the page into the free list, the caller
158 * has a reference on the page.
160 void delete_from_page_cache(struct page *page)
162 struct address_space *mapping = page->mapping;
163 void (*freepage)(struct page *);
165 BUG_ON(!PageLocked(page));
167 freepage = mapping->a_ops->freepage;
168 spin_lock_irq(&mapping->tree_lock);
169 __delete_from_page_cache(page);
170 spin_unlock_irq(&mapping->tree_lock);
171 mem_cgroup_uncharge_cache_page(page);
175 page_cache_release(page);
177 EXPORT_SYMBOL(delete_from_page_cache);
179 static int sleep_on_page(void *word)
185 static int sleep_on_page_killable(void *word)
188 return fatal_signal_pending(current) ? -EINTR : 0;
191 static int filemap_check_errors(struct address_space *mapping)
194 /* Check for outstanding write errors */
195 if (test_bit(AS_ENOSPC, &mapping->flags) &&
196 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
198 if (test_bit(AS_EIO, &mapping->flags) &&
199 test_and_clear_bit(AS_EIO, &mapping->flags))
205 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
206 * @mapping: address space structure to write
207 * @start: offset in bytes where the range starts
208 * @end: offset in bytes where the range ends (inclusive)
209 * @sync_mode: enable synchronous operation
211 * Start writeback against all of a mapping's dirty pages that lie
212 * within the byte offsets <start, end> inclusive.
214 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
215 * opposed to a regular memory cleansing writeback. The difference between
216 * these two operations is that if a dirty page/buffer is encountered, it must
217 * be waited upon, and not just skipped over.
219 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
220 loff_t end, int sync_mode)
223 struct writeback_control wbc = {
224 .sync_mode = sync_mode,
225 .nr_to_write = LONG_MAX,
226 .range_start = start,
230 if (!mapping_cap_writeback_dirty(mapping))
233 ret = do_writepages(mapping, &wbc);
237 static inline int __filemap_fdatawrite(struct address_space *mapping,
240 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
243 int filemap_fdatawrite(struct address_space *mapping)
245 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
247 EXPORT_SYMBOL(filemap_fdatawrite);
249 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
252 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
254 EXPORT_SYMBOL(filemap_fdatawrite_range);
257 * filemap_flush - mostly a non-blocking flush
258 * @mapping: target address_space
260 * This is a mostly non-blocking flush. Not suitable for data-integrity
261 * purposes - I/O may not be started against all dirty pages.
263 int filemap_flush(struct address_space *mapping)
265 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
267 EXPORT_SYMBOL(filemap_flush);
270 * filemap_fdatawait_range - wait for writeback to complete
271 * @mapping: address space structure to wait for
272 * @start_byte: offset in bytes where the range starts
273 * @end_byte: offset in bytes where the range ends (inclusive)
275 * Walk the list of under-writeback pages of the given address space
276 * in the given range and wait for all of them.
278 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
281 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
282 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
287 if (end_byte < start_byte)
290 pagevec_init(&pvec, 0);
291 while ((index <= end) &&
292 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
293 PAGECACHE_TAG_WRITEBACK,
294 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
297 for (i = 0; i < nr_pages; i++) {
298 struct page *page = pvec.pages[i];
300 /* until radix tree lookup accepts end_index */
301 if (page->index > end)
304 wait_on_page_writeback(page);
305 if (TestClearPageError(page))
308 pagevec_release(&pvec);
312 ret2 = filemap_check_errors(mapping);
318 EXPORT_SYMBOL(filemap_fdatawait_range);
321 * filemap_fdatawait - wait for all under-writeback pages to complete
322 * @mapping: address space structure to wait for
324 * Walk the list of under-writeback pages of the given address space
325 * and wait for all of them.
327 int filemap_fdatawait(struct address_space *mapping)
329 loff_t i_size = i_size_read(mapping->host);
334 return filemap_fdatawait_range(mapping, 0, i_size - 1);
336 EXPORT_SYMBOL(filemap_fdatawait);
338 int filemap_write_and_wait(struct address_space *mapping)
342 if (mapping->nrpages) {
343 err = filemap_fdatawrite(mapping);
345 * Even if the above returned error, the pages may be
346 * written partially (e.g. -ENOSPC), so we wait for it.
347 * But the -EIO is special case, it may indicate the worst
348 * thing (e.g. bug) happened, so we avoid waiting for it.
351 int err2 = filemap_fdatawait(mapping);
356 err = filemap_check_errors(mapping);
360 EXPORT_SYMBOL(filemap_write_and_wait);
363 * filemap_write_and_wait_range - write out & wait on a file range
364 * @mapping: the address_space for the pages
365 * @lstart: offset in bytes where the range starts
366 * @lend: offset in bytes where the range ends (inclusive)
368 * Write out and wait upon file offsets lstart->lend, inclusive.
370 * Note that `lend' is inclusive (describes the last byte to be written) so
371 * that this function can be used to write to the very end-of-file (end = -1).
373 int filemap_write_and_wait_range(struct address_space *mapping,
374 loff_t lstart, loff_t lend)
378 if (mapping->nrpages) {
379 err = __filemap_fdatawrite_range(mapping, lstart, lend,
381 /* See comment of filemap_write_and_wait() */
383 int err2 = filemap_fdatawait_range(mapping,
389 err = filemap_check_errors(mapping);
393 EXPORT_SYMBOL(filemap_write_and_wait_range);
396 * replace_page_cache_page - replace a pagecache page with a new one
397 * @old: page to be replaced
398 * @new: page to replace with
399 * @gfp_mask: allocation mode
401 * This function replaces a page in the pagecache with a new one. On
402 * success it acquires the pagecache reference for the new page and
403 * drops it for the old page. Both the old and new pages must be
404 * locked. This function does not add the new page to the LRU, the
405 * caller must do that.
407 * The remove + add is atomic. The only way this function can fail is
408 * memory allocation failure.
410 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
414 VM_BUG_ON_PAGE(!PageLocked(old), old);
415 VM_BUG_ON_PAGE(!PageLocked(new), new);
416 VM_BUG_ON_PAGE(new->mapping, new);
418 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
420 struct address_space *mapping = old->mapping;
421 void (*freepage)(struct page *);
423 pgoff_t offset = old->index;
424 freepage = mapping->a_ops->freepage;
427 new->mapping = mapping;
430 spin_lock_irq(&mapping->tree_lock);
431 __delete_from_page_cache(old);
432 error = radix_tree_insert(&mapping->page_tree, offset, new);
435 __inc_zone_page_state(new, NR_FILE_PAGES);
436 if (PageSwapBacked(new))
437 __inc_zone_page_state(new, NR_SHMEM);
438 spin_unlock_irq(&mapping->tree_lock);
439 /* mem_cgroup codes must not be called under tree_lock */
440 mem_cgroup_replace_page_cache(old, new);
441 radix_tree_preload_end();
444 page_cache_release(old);
449 EXPORT_SYMBOL_GPL(replace_page_cache_page);
451 static int page_cache_tree_insert(struct address_space *mapping,
457 slot = radix_tree_lookup_slot(&mapping->page_tree, page->index);
461 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
462 if (!radix_tree_exceptional_entry(p))
464 radix_tree_replace_slot(slot, page);
468 error = radix_tree_insert(&mapping->page_tree, page->index, page);
475 * add_to_page_cache_locked - add a locked page to the pagecache
477 * @mapping: the page's address_space
478 * @offset: page index
479 * @gfp_mask: page allocation mode
481 * This function is used to add a page to the pagecache. It must be locked.
482 * This function does not add the page to the LRU. The caller must do that.
484 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
485 pgoff_t offset, gfp_t gfp_mask)
489 VM_BUG_ON_PAGE(!PageLocked(page), page);
490 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
492 error = mem_cgroup_cache_charge(page, current->mm,
493 gfp_mask & GFP_RECLAIM_MASK);
497 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
499 mem_cgroup_uncharge_cache_page(page);
503 page_cache_get(page);
504 page->mapping = mapping;
505 page->index = offset;
507 spin_lock_irq(&mapping->tree_lock);
508 error = page_cache_tree_insert(mapping, page);
509 radix_tree_preload_end();
512 __inc_zone_page_state(page, NR_FILE_PAGES);
513 spin_unlock_irq(&mapping->tree_lock);
514 trace_mm_filemap_add_to_page_cache(page);
517 page->mapping = NULL;
518 /* Leave page->index set: truncation relies upon it */
519 spin_unlock_irq(&mapping->tree_lock);
520 mem_cgroup_uncharge_cache_page(page);
521 page_cache_release(page);
524 EXPORT_SYMBOL(add_to_page_cache_locked);
526 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
527 pgoff_t offset, gfp_t gfp_mask)
531 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
533 lru_cache_add_file(page);
536 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
539 struct page *__page_cache_alloc(gfp_t gfp)
544 if (cpuset_do_page_mem_spread()) {
545 unsigned int cpuset_mems_cookie;
547 cpuset_mems_cookie = read_mems_allowed_begin();
548 n = cpuset_mem_spread_node();
549 page = alloc_pages_exact_node(n, gfp, 0);
550 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
554 return alloc_pages(gfp, 0);
556 EXPORT_SYMBOL(__page_cache_alloc);
560 * In order to wait for pages to become available there must be
561 * waitqueues associated with pages. By using a hash table of
562 * waitqueues where the bucket discipline is to maintain all
563 * waiters on the same queue and wake all when any of the pages
564 * become available, and for the woken contexts to check to be
565 * sure the appropriate page became available, this saves space
566 * at a cost of "thundering herd" phenomena during rare hash
569 static wait_queue_head_t *page_waitqueue(struct page *page)
571 const struct zone *zone = page_zone(page);
573 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
576 static inline void wake_up_page(struct page *page, int bit)
578 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
581 void wait_on_page_bit(struct page *page, int bit_nr)
583 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
585 if (test_bit(bit_nr, &page->flags))
586 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
587 TASK_UNINTERRUPTIBLE);
589 EXPORT_SYMBOL(wait_on_page_bit);
591 int wait_on_page_bit_killable(struct page *page, int bit_nr)
593 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
595 if (!test_bit(bit_nr, &page->flags))
598 return __wait_on_bit(page_waitqueue(page), &wait,
599 sleep_on_page_killable, TASK_KILLABLE);
603 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
604 * @page: Page defining the wait queue of interest
605 * @waiter: Waiter to add to the queue
607 * Add an arbitrary @waiter to the wait queue for the nominated @page.
609 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
611 wait_queue_head_t *q = page_waitqueue(page);
614 spin_lock_irqsave(&q->lock, flags);
615 __add_wait_queue(q, waiter);
616 spin_unlock_irqrestore(&q->lock, flags);
618 EXPORT_SYMBOL_GPL(add_page_wait_queue);
621 * unlock_page - unlock a locked page
624 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
625 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
626 * mechananism between PageLocked pages and PageWriteback pages is shared.
627 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
629 * The mb is necessary to enforce ordering between the clear_bit and the read
630 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
632 void unlock_page(struct page *page)
634 VM_BUG_ON_PAGE(!PageLocked(page), page);
635 clear_bit_unlock(PG_locked, &page->flags);
636 smp_mb__after_clear_bit();
637 wake_up_page(page, PG_locked);
639 EXPORT_SYMBOL(unlock_page);
642 * end_page_writeback - end writeback against a page
645 void end_page_writeback(struct page *page)
647 if (TestClearPageReclaim(page))
648 rotate_reclaimable_page(page);
650 if (!test_clear_page_writeback(page))
653 smp_mb__after_clear_bit();
654 wake_up_page(page, PG_writeback);
656 EXPORT_SYMBOL(end_page_writeback);
659 * __lock_page - get a lock on the page, assuming we need to sleep to get it
660 * @page: the page to lock
662 void __lock_page(struct page *page)
664 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
666 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
667 TASK_UNINTERRUPTIBLE);
669 EXPORT_SYMBOL(__lock_page);
671 int __lock_page_killable(struct page *page)
673 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
675 return __wait_on_bit_lock(page_waitqueue(page), &wait,
676 sleep_on_page_killable, TASK_KILLABLE);
678 EXPORT_SYMBOL_GPL(__lock_page_killable);
680 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
683 if (flags & FAULT_FLAG_ALLOW_RETRY) {
685 * CAUTION! In this case, mmap_sem is not released
686 * even though return 0.
688 if (flags & FAULT_FLAG_RETRY_NOWAIT)
691 up_read(&mm->mmap_sem);
692 if (flags & FAULT_FLAG_KILLABLE)
693 wait_on_page_locked_killable(page);
695 wait_on_page_locked(page);
698 if (flags & FAULT_FLAG_KILLABLE) {
701 ret = __lock_page_killable(page);
703 up_read(&mm->mmap_sem);
713 * page_cache_next_hole - find the next hole (not-present entry)
716 * @max_scan: maximum range to search
718 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
719 * lowest indexed hole.
721 * Returns: the index of the hole if found, otherwise returns an index
722 * outside of the set specified (in which case 'return - index >=
723 * max_scan' will be true). In rare cases of index wrap-around, 0 will
726 * page_cache_next_hole may be called under rcu_read_lock. However,
727 * like radix_tree_gang_lookup, this will not atomically search a
728 * snapshot of the tree at a single point in time. For example, if a
729 * hole is created at index 5, then subsequently a hole is created at
730 * index 10, page_cache_next_hole covering both indexes may return 10
731 * if called under rcu_read_lock.
733 pgoff_t page_cache_next_hole(struct address_space *mapping,
734 pgoff_t index, unsigned long max_scan)
738 for (i = 0; i < max_scan; i++) {
741 page = radix_tree_lookup(&mapping->page_tree, index);
742 if (!page || radix_tree_exceptional_entry(page))
751 EXPORT_SYMBOL(page_cache_next_hole);
754 * page_cache_prev_hole - find the prev hole (not-present entry)
757 * @max_scan: maximum range to search
759 * Search backwards in the range [max(index-max_scan+1, 0), index] for
762 * Returns: the index of the hole if found, otherwise returns an index
763 * outside of the set specified (in which case 'index - return >=
764 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
767 * page_cache_prev_hole may be called under rcu_read_lock. However,
768 * like radix_tree_gang_lookup, this will not atomically search a
769 * snapshot of the tree at a single point in time. For example, if a
770 * hole is created at index 10, then subsequently a hole is created at
771 * index 5, page_cache_prev_hole covering both indexes may return 5 if
772 * called under rcu_read_lock.
774 pgoff_t page_cache_prev_hole(struct address_space *mapping,
775 pgoff_t index, unsigned long max_scan)
779 for (i = 0; i < max_scan; i++) {
782 page = radix_tree_lookup(&mapping->page_tree, index);
783 if (!page || radix_tree_exceptional_entry(page))
786 if (index == ULONG_MAX)
792 EXPORT_SYMBOL(page_cache_prev_hole);
795 * find_get_entry - find and get a page cache entry
796 * @mapping: the address_space to search
797 * @offset: the page cache index
799 * Looks up the page cache slot at @mapping & @offset. If there is a
800 * page cache page, it is returned with an increased refcount.
802 * If the slot holds a shadow entry of a previously evicted page, it
805 * Otherwise, %NULL is returned.
807 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
815 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
817 page = radix_tree_deref_slot(pagep);
820 if (radix_tree_exception(page)) {
821 if (radix_tree_deref_retry(page))
824 * Otherwise, shmem/tmpfs must be storing a swap entry
825 * here as an exceptional entry: so return it without
826 * attempting to raise page count.
830 if (!page_cache_get_speculative(page))
834 * Has the page moved?
835 * This is part of the lockless pagecache protocol. See
836 * include/linux/pagemap.h for details.
838 if (unlikely(page != *pagep)) {
839 page_cache_release(page);
848 EXPORT_SYMBOL(find_get_entry);
851 * find_get_page - find and get a page reference
852 * @mapping: the address_space to search
853 * @offset: the page index
855 * Looks up the page cache slot at @mapping & @offset. If there is a
856 * page cache page, it is returned with an increased refcount.
858 * Otherwise, %NULL is returned.
860 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
862 struct page *page = find_get_entry(mapping, offset);
864 if (radix_tree_exceptional_entry(page))
868 EXPORT_SYMBOL(find_get_page);
871 * find_lock_entry - locate, pin and lock a page cache entry
872 * @mapping: the address_space to search
873 * @offset: the page cache index
875 * Looks up the page cache slot at @mapping & @offset. If there is a
876 * page cache page, it is returned locked and with an increased
879 * If the slot holds a shadow entry of a previously evicted page, it
882 * Otherwise, %NULL is returned.
884 * find_lock_entry() may sleep.
886 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
891 page = find_get_entry(mapping, offset);
892 if (page && !radix_tree_exception(page)) {
894 /* Has the page been truncated? */
895 if (unlikely(page->mapping != mapping)) {
897 page_cache_release(page);
900 VM_BUG_ON_PAGE(page->index != offset, page);
904 EXPORT_SYMBOL(find_lock_entry);
907 * find_lock_page - locate, pin and lock a pagecache page
908 * @mapping: the address_space to search
909 * @offset: the page index
911 * Looks up the page cache slot at @mapping & @offset. If there is a
912 * page cache page, it is returned locked and with an increased
915 * Otherwise, %NULL is returned.
917 * find_lock_page() may sleep.
919 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
921 struct page *page = find_lock_entry(mapping, offset);
923 if (radix_tree_exceptional_entry(page))
927 EXPORT_SYMBOL(find_lock_page);
930 * find_or_create_page - locate or add a pagecache page
931 * @mapping: the page's address_space
932 * @index: the page's index into the mapping
933 * @gfp_mask: page allocation mode
935 * Looks up the page cache slot at @mapping & @offset. If there is a
936 * page cache page, it is returned locked and with an increased
939 * If the page is not present, a new page is allocated using @gfp_mask
940 * and added to the page cache and the VM's LRU list. The page is
941 * returned locked and with an increased refcount.
943 * On memory exhaustion, %NULL is returned.
945 * find_or_create_page() may sleep, even if @gfp_flags specifies an
948 struct page *find_or_create_page(struct address_space *mapping,
949 pgoff_t index, gfp_t gfp_mask)
954 page = find_lock_page(mapping, index);
956 page = __page_cache_alloc(gfp_mask);
960 * We want a regular kernel memory (not highmem or DMA etc)
961 * allocation for the radix tree nodes, but we need to honour
962 * the context-specific requirements the caller has asked for.
963 * GFP_RECLAIM_MASK collects those requirements.
965 err = add_to_page_cache_lru(page, mapping, index,
966 (gfp_mask & GFP_RECLAIM_MASK));
968 page_cache_release(page);
976 EXPORT_SYMBOL(find_or_create_page);
979 * find_get_entries - gang pagecache lookup
980 * @mapping: The address_space to search
981 * @start: The starting page cache index
982 * @nr_entries: The maximum number of entries
983 * @entries: Where the resulting entries are placed
984 * @indices: The cache indices corresponding to the entries in @entries
986 * find_get_entries() will search for and return a group of up to
987 * @nr_entries entries in the mapping. The entries are placed at
988 * @entries. find_get_entries() takes a reference against any actual
991 * The search returns a group of mapping-contiguous page cache entries
992 * with ascending indexes. There may be holes in the indices due to
995 * Any shadow entries of evicted pages are included in the returned
998 * find_get_entries() returns the number of pages and shadow entries
1001 unsigned find_get_entries(struct address_space *mapping,
1002 pgoff_t start, unsigned int nr_entries,
1003 struct page **entries, pgoff_t *indices)
1006 unsigned int ret = 0;
1007 struct radix_tree_iter iter;
1014 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1017 page = radix_tree_deref_slot(slot);
1018 if (unlikely(!page))
1020 if (radix_tree_exception(page)) {
1021 if (radix_tree_deref_retry(page))
1024 * Otherwise, we must be storing a swap entry
1025 * here as an exceptional entry: so return it
1026 * without attempting to raise page count.
1030 if (!page_cache_get_speculative(page))
1033 /* Has the page moved? */
1034 if (unlikely(page != *slot)) {
1035 page_cache_release(page);
1039 indices[ret] = iter.index;
1040 entries[ret] = page;
1041 if (++ret == nr_entries)
1049 * find_get_pages - gang pagecache lookup
1050 * @mapping: The address_space to search
1051 * @start: The starting page index
1052 * @nr_pages: The maximum number of pages
1053 * @pages: Where the resulting pages are placed
1055 * find_get_pages() will search for and return a group of up to
1056 * @nr_pages pages in the mapping. The pages are placed at @pages.
1057 * find_get_pages() takes a reference against the returned pages.
1059 * The search returns a group of mapping-contiguous pages with ascending
1060 * indexes. There may be holes in the indices due to not-present pages.
1062 * find_get_pages() returns the number of pages which were found.
1064 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1065 unsigned int nr_pages, struct page **pages)
1067 struct radix_tree_iter iter;
1071 if (unlikely(!nr_pages))
1076 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1079 page = radix_tree_deref_slot(slot);
1080 if (unlikely(!page))
1083 if (radix_tree_exception(page)) {
1084 if (radix_tree_deref_retry(page)) {
1086 * Transient condition which can only trigger
1087 * when entry at index 0 moves out of or back
1088 * to root: none yet gotten, safe to restart.
1090 WARN_ON(iter.index);
1094 * Otherwise, shmem/tmpfs must be storing a swap entry
1095 * here as an exceptional entry: so skip over it -
1096 * we only reach this from invalidate_mapping_pages().
1101 if (!page_cache_get_speculative(page))
1104 /* Has the page moved? */
1105 if (unlikely(page != *slot)) {
1106 page_cache_release(page);
1111 if (++ret == nr_pages)
1120 * find_get_pages_contig - gang contiguous pagecache lookup
1121 * @mapping: The address_space to search
1122 * @index: The starting page index
1123 * @nr_pages: The maximum number of pages
1124 * @pages: Where the resulting pages are placed
1126 * find_get_pages_contig() works exactly like find_get_pages(), except
1127 * that the returned number of pages are guaranteed to be contiguous.
1129 * find_get_pages_contig() returns the number of pages which were found.
1131 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1132 unsigned int nr_pages, struct page **pages)
1134 struct radix_tree_iter iter;
1136 unsigned int ret = 0;
1138 if (unlikely(!nr_pages))
1143 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1146 page = radix_tree_deref_slot(slot);
1147 /* The hole, there no reason to continue */
1148 if (unlikely(!page))
1151 if (radix_tree_exception(page)) {
1152 if (radix_tree_deref_retry(page)) {
1154 * Transient condition which can only trigger
1155 * when entry at index 0 moves out of or back
1156 * to root: none yet gotten, safe to restart.
1161 * Otherwise, shmem/tmpfs must be storing a swap entry
1162 * here as an exceptional entry: so stop looking for
1168 if (!page_cache_get_speculative(page))
1171 /* Has the page moved? */
1172 if (unlikely(page != *slot)) {
1173 page_cache_release(page);
1178 * must check mapping and index after taking the ref.
1179 * otherwise we can get both false positives and false
1180 * negatives, which is just confusing to the caller.
1182 if (page->mapping == NULL || page->index != iter.index) {
1183 page_cache_release(page);
1188 if (++ret == nr_pages)
1194 EXPORT_SYMBOL(find_get_pages_contig);
1197 * find_get_pages_tag - find and return pages that match @tag
1198 * @mapping: the address_space to search
1199 * @index: the starting page index
1200 * @tag: the tag index
1201 * @nr_pages: the maximum number of pages
1202 * @pages: where the resulting pages are placed
1204 * Like find_get_pages, except we only return pages which are tagged with
1205 * @tag. We update @index to index the next page for the traversal.
1207 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1208 int tag, unsigned int nr_pages, struct page **pages)
1210 struct radix_tree_iter iter;
1214 if (unlikely(!nr_pages))
1219 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1220 &iter, *index, tag) {
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.
1237 * This function is never used on a shmem/tmpfs
1238 * mapping, so a swap entry won't be found here.
1243 if (!page_cache_get_speculative(page))
1246 /* Has the page moved? */
1247 if (unlikely(page != *slot)) {
1248 page_cache_release(page);
1253 if (++ret == nr_pages)
1260 *index = pages[ret - 1]->index + 1;
1264 EXPORT_SYMBOL(find_get_pages_tag);
1267 * grab_cache_page_nowait - returns locked page at given index in given cache
1268 * @mapping: target address_space
1269 * @index: the page index
1271 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1272 * This is intended for speculative data generators, where the data can
1273 * be regenerated if the page couldn't be grabbed. This routine should
1274 * be safe to call while holding the lock for another page.
1276 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1277 * and deadlock against the caller's locked page.
1280 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1282 struct page *page = find_get_page(mapping, index);
1285 if (trylock_page(page))
1287 page_cache_release(page);
1290 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1291 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1292 page_cache_release(page);
1297 EXPORT_SYMBOL(grab_cache_page_nowait);
1300 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1301 * a _large_ part of the i/o request. Imagine the worst scenario:
1303 * ---R__________________________________________B__________
1304 * ^ reading here ^ bad block(assume 4k)
1306 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1307 * => failing the whole request => read(R) => read(R+1) =>
1308 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1309 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1310 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1312 * It is going insane. Fix it by quickly scaling down the readahead size.
1314 static void shrink_readahead_size_eio(struct file *filp,
1315 struct file_ra_state *ra)
1321 * do_generic_file_read - generic file read routine
1322 * @filp: the file to read
1323 * @ppos: current file position
1324 * @desc: read_descriptor
1326 * This is a generic file read routine, and uses the
1327 * mapping->a_ops->readpage() function for the actual low-level stuff.
1329 * This is really ugly. But the goto's actually try to clarify some
1330 * of the logic when it comes to error handling etc.
1332 static void do_generic_file_read(struct file *filp, loff_t *ppos,
1333 read_descriptor_t *desc)
1335 struct address_space *mapping = filp->f_mapping;
1336 struct inode *inode = mapping->host;
1337 struct file_ra_state *ra = &filp->f_ra;
1341 unsigned long offset; /* offset into pagecache page */
1342 unsigned int prev_offset;
1345 index = *ppos >> PAGE_CACHE_SHIFT;
1346 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1347 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1348 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1349 offset = *ppos & ~PAGE_CACHE_MASK;
1355 unsigned long nr, ret;
1359 page = find_get_page(mapping, index);
1361 page_cache_sync_readahead(mapping,
1363 index, last_index - index);
1364 page = find_get_page(mapping, index);
1365 if (unlikely(page == NULL))
1366 goto no_cached_page;
1368 if (PageReadahead(page)) {
1369 page_cache_async_readahead(mapping,
1371 index, last_index - index);
1373 if (!PageUptodate(page)) {
1374 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1375 !mapping->a_ops->is_partially_uptodate)
1376 goto page_not_up_to_date;
1377 if (!trylock_page(page))
1378 goto page_not_up_to_date;
1379 /* Did it get truncated before we got the lock? */
1381 goto page_not_up_to_date_locked;
1382 if (!mapping->a_ops->is_partially_uptodate(page,
1384 goto page_not_up_to_date_locked;
1389 * i_size must be checked after we know the page is Uptodate.
1391 * Checking i_size after the check allows us to calculate
1392 * the correct value for "nr", which means the zero-filled
1393 * part of the page is not copied back to userspace (unless
1394 * another truncate extends the file - this is desired though).
1397 isize = i_size_read(inode);
1398 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1399 if (unlikely(!isize || index > end_index)) {
1400 page_cache_release(page);
1404 /* nr is the maximum number of bytes to copy from this page */
1405 nr = PAGE_CACHE_SIZE;
1406 if (index == end_index) {
1407 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1409 page_cache_release(page);
1415 /* If users can be writing to this page using arbitrary
1416 * virtual addresses, take care about potential aliasing
1417 * before reading the page on the kernel side.
1419 if (mapping_writably_mapped(mapping))
1420 flush_dcache_page(page);
1423 * When a sequential read accesses a page several times,
1424 * only mark it as accessed the first time.
1426 if (prev_index != index || offset != prev_offset)
1427 mark_page_accessed(page);
1431 * Ok, we have the page, and it's up-to-date, so
1432 * now we can copy it to user space...
1434 * The file_read_actor routine returns how many bytes were
1436 * NOTE! This may not be the same as how much of a user buffer
1437 * we filled up (we may be padding etc), so we can only update
1438 * "pos" here (the actor routine has to update the user buffer
1439 * pointers and the remaining count).
1441 ret = file_read_actor(desc, page, offset, nr);
1443 index += offset >> PAGE_CACHE_SHIFT;
1444 offset &= ~PAGE_CACHE_MASK;
1445 prev_offset = offset;
1447 page_cache_release(page);
1448 if (ret == nr && desc->count)
1452 page_not_up_to_date:
1453 /* Get exclusive access to the page ... */
1454 error = lock_page_killable(page);
1455 if (unlikely(error))
1456 goto readpage_error;
1458 page_not_up_to_date_locked:
1459 /* Did it get truncated before we got the lock? */
1460 if (!page->mapping) {
1462 page_cache_release(page);
1466 /* Did somebody else fill it already? */
1467 if (PageUptodate(page)) {
1474 * A previous I/O error may have been due to temporary
1475 * failures, eg. multipath errors.
1476 * PG_error will be set again if readpage fails.
1478 ClearPageError(page);
1479 /* Start the actual read. The read will unlock the page. */
1480 error = mapping->a_ops->readpage(filp, page);
1482 if (unlikely(error)) {
1483 if (error == AOP_TRUNCATED_PAGE) {
1484 page_cache_release(page);
1487 goto readpage_error;
1490 if (!PageUptodate(page)) {
1491 error = lock_page_killable(page);
1492 if (unlikely(error))
1493 goto readpage_error;
1494 if (!PageUptodate(page)) {
1495 if (page->mapping == NULL) {
1497 * invalidate_mapping_pages got it
1500 page_cache_release(page);
1504 shrink_readahead_size_eio(filp, ra);
1506 goto readpage_error;
1514 /* UHHUH! A synchronous read error occurred. Report it */
1515 desc->error = error;
1516 page_cache_release(page);
1521 * Ok, it wasn't cached, so we need to create a new
1524 page = page_cache_alloc_cold(mapping);
1526 desc->error = -ENOMEM;
1529 error = add_to_page_cache_lru(page, mapping,
1532 page_cache_release(page);
1533 if (error == -EEXIST)
1535 desc->error = error;
1542 ra->prev_pos = prev_index;
1543 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1544 ra->prev_pos |= prev_offset;
1546 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1547 file_accessed(filp);
1550 int file_read_actor(read_descriptor_t *desc, struct page *page,
1551 unsigned long offset, unsigned long size)
1554 unsigned long left, count = desc->count;
1560 * Faults on the destination of a read are common, so do it before
1563 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1564 kaddr = kmap_atomic(page);
1565 left = __copy_to_user_inatomic(desc->arg.buf,
1566 kaddr + offset, size);
1567 kunmap_atomic(kaddr);
1572 /* Do it the slow way */
1574 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1579 desc->error = -EFAULT;
1582 desc->count = count - size;
1583 desc->written += size;
1584 desc->arg.buf += size;
1589 * Performs necessary checks before doing a write
1590 * @iov: io vector request
1591 * @nr_segs: number of segments in the iovec
1592 * @count: number of bytes to write
1593 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1595 * Adjust number of segments and amount of bytes to write (nr_segs should be
1596 * properly initialized first). Returns appropriate error code that caller
1597 * should return or zero in case that write should be allowed.
1599 int generic_segment_checks(const struct iovec *iov,
1600 unsigned long *nr_segs, size_t *count, int access_flags)
1604 for (seg = 0; seg < *nr_segs; seg++) {
1605 const struct iovec *iv = &iov[seg];
1608 * If any segment has a negative length, or the cumulative
1609 * length ever wraps negative then return -EINVAL.
1612 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1614 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1619 cnt -= iv->iov_len; /* This segment is no good */
1625 EXPORT_SYMBOL(generic_segment_checks);
1628 * generic_file_aio_read - generic filesystem read routine
1629 * @iocb: kernel I/O control block
1630 * @iov: io vector request
1631 * @nr_segs: number of segments in the iovec
1632 * @pos: current file position
1634 * This is the "read()" routine for all filesystems
1635 * that can use the page cache directly.
1638 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1639 unsigned long nr_segs, loff_t pos)
1641 struct file *filp = iocb->ki_filp;
1643 unsigned long seg = 0;
1645 loff_t *ppos = &iocb->ki_pos;
1648 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1652 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1653 if (filp->f_flags & O_DIRECT) {
1655 struct address_space *mapping;
1656 struct inode *inode;
1658 mapping = filp->f_mapping;
1659 inode = mapping->host;
1661 goto out; /* skip atime */
1662 size = i_size_read(inode);
1663 retval = filemap_write_and_wait_range(mapping, pos,
1664 pos + iov_length(iov, nr_segs) - 1);
1666 retval = mapping->a_ops->direct_IO(READ, iocb,
1670 *ppos = pos + retval;
1675 * Btrfs can have a short DIO read if we encounter
1676 * compressed extents, so if there was an error, or if
1677 * we've already read everything we wanted to, or if
1678 * there was a short read because we hit EOF, go ahead
1679 * and return. Otherwise fallthrough to buffered io for
1680 * the rest of the read.
1682 if (retval < 0 || !count || *ppos >= size) {
1683 file_accessed(filp);
1689 for (seg = 0; seg < nr_segs; seg++) {
1690 read_descriptor_t desc;
1694 * If we did a short DIO read we need to skip the section of the
1695 * iov that we've already read data into.
1698 if (count > iov[seg].iov_len) {
1699 count -= iov[seg].iov_len;
1707 desc.arg.buf = iov[seg].iov_base + offset;
1708 desc.count = iov[seg].iov_len - offset;
1709 if (desc.count == 0)
1712 do_generic_file_read(filp, ppos, &desc);
1713 retval += desc.written;
1715 retval = retval ?: desc.error;
1724 EXPORT_SYMBOL(generic_file_aio_read);
1728 * page_cache_read - adds requested page to the page cache if not already there
1729 * @file: file to read
1730 * @offset: page index
1732 * This adds the requested page to the page cache if it isn't already there,
1733 * and schedules an I/O to read in its contents from disk.
1735 static int page_cache_read(struct file *file, pgoff_t offset)
1737 struct address_space *mapping = file->f_mapping;
1742 page = page_cache_alloc_cold(mapping);
1746 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1748 ret = mapping->a_ops->readpage(file, page);
1749 else if (ret == -EEXIST)
1750 ret = 0; /* losing race to add is OK */
1752 page_cache_release(page);
1754 } while (ret == AOP_TRUNCATED_PAGE);
1759 #define MMAP_LOTSAMISS (100)
1762 * Synchronous readahead happens when we don't even find
1763 * a page in the page cache at all.
1765 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1766 struct file_ra_state *ra,
1770 unsigned long ra_pages;
1771 struct address_space *mapping = file->f_mapping;
1773 /* If we don't want any read-ahead, don't bother */
1774 if (vma->vm_flags & VM_RAND_READ)
1779 if (vma->vm_flags & VM_SEQ_READ) {
1780 page_cache_sync_readahead(mapping, ra, file, offset,
1785 /* Avoid banging the cache line if not needed */
1786 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1790 * Do we miss much more than hit in this file? If so,
1791 * stop bothering with read-ahead. It will only hurt.
1793 if (ra->mmap_miss > MMAP_LOTSAMISS)
1799 ra_pages = max_sane_readahead(ra->ra_pages);
1800 ra->start = max_t(long, 0, offset - ra_pages / 2);
1801 ra->size = ra_pages;
1802 ra->async_size = ra_pages / 4;
1803 ra_submit(ra, mapping, file);
1807 * Asynchronous readahead happens when we find the page and PG_readahead,
1808 * so we want to possibly extend the readahead further..
1810 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1811 struct file_ra_state *ra,
1816 struct address_space *mapping = file->f_mapping;
1818 /* If we don't want any read-ahead, don't bother */
1819 if (vma->vm_flags & VM_RAND_READ)
1821 if (ra->mmap_miss > 0)
1823 if (PageReadahead(page))
1824 page_cache_async_readahead(mapping, ra, file,
1825 page, offset, ra->ra_pages);
1829 * filemap_fault - read in file data for page fault handling
1830 * @vma: vma in which the fault was taken
1831 * @vmf: struct vm_fault containing details of the fault
1833 * filemap_fault() is invoked via the vma operations vector for a
1834 * mapped memory region to read in file data during a page fault.
1836 * The goto's are kind of ugly, but this streamlines the normal case of having
1837 * it in the page cache, and handles the special cases reasonably without
1838 * having a lot of duplicated code.
1840 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1843 struct file *file = vma->vm_file;
1844 struct address_space *mapping = file->f_mapping;
1845 struct file_ra_state *ra = &file->f_ra;
1846 struct inode *inode = mapping->host;
1847 pgoff_t offset = vmf->pgoff;
1852 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1854 return VM_FAULT_SIGBUS;
1857 * Do we have something in the page cache already?
1859 page = find_get_page(mapping, offset);
1860 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1862 * We found the page, so try async readahead before
1863 * waiting for the lock.
1865 do_async_mmap_readahead(vma, ra, file, page, offset);
1867 /* No page in the page cache at all */
1868 do_sync_mmap_readahead(vma, ra, file, offset);
1869 count_vm_event(PGMAJFAULT);
1870 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1871 ret = VM_FAULT_MAJOR;
1873 page = find_get_page(mapping, offset);
1875 goto no_cached_page;
1878 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1879 page_cache_release(page);
1880 return ret | VM_FAULT_RETRY;
1883 /* Did it get truncated? */
1884 if (unlikely(page->mapping != mapping)) {
1889 VM_BUG_ON_PAGE(page->index != offset, page);
1892 * We have a locked page in the page cache, now we need to check
1893 * that it's up-to-date. If not, it is going to be due to an error.
1895 if (unlikely(!PageUptodate(page)))
1896 goto page_not_uptodate;
1899 * Found the page and have a reference on it.
1900 * We must recheck i_size under page lock.
1902 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1903 if (unlikely(offset >= size)) {
1905 page_cache_release(page);
1906 return VM_FAULT_SIGBUS;
1910 return ret | VM_FAULT_LOCKED;
1914 * We're only likely to ever get here if MADV_RANDOM is in
1917 error = page_cache_read(file, offset);
1920 * The page we want has now been added to the page cache.
1921 * In the unlikely event that someone removed it in the
1922 * meantime, we'll just come back here and read it again.
1928 * An error return from page_cache_read can result if the
1929 * system is low on memory, or a problem occurs while trying
1932 if (error == -ENOMEM)
1933 return VM_FAULT_OOM;
1934 return VM_FAULT_SIGBUS;
1938 * Umm, take care of errors if the page isn't up-to-date.
1939 * Try to re-read it _once_. We do this synchronously,
1940 * because there really aren't any performance issues here
1941 * and we need to check for errors.
1943 ClearPageError(page);
1944 error = mapping->a_ops->readpage(file, page);
1946 wait_on_page_locked(page);
1947 if (!PageUptodate(page))
1950 page_cache_release(page);
1952 if (!error || error == AOP_TRUNCATED_PAGE)
1955 /* Things didn't work out. Return zero to tell the mm layer so. */
1956 shrink_readahead_size_eio(file, ra);
1957 return VM_FAULT_SIGBUS;
1959 EXPORT_SYMBOL(filemap_fault);
1961 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1963 struct page *page = vmf->page;
1964 struct inode *inode = file_inode(vma->vm_file);
1965 int ret = VM_FAULT_LOCKED;
1967 sb_start_pagefault(inode->i_sb);
1968 file_update_time(vma->vm_file);
1970 if (page->mapping != inode->i_mapping) {
1972 ret = VM_FAULT_NOPAGE;
1976 * We mark the page dirty already here so that when freeze is in
1977 * progress, we are guaranteed that writeback during freezing will
1978 * see the dirty page and writeprotect it again.
1980 set_page_dirty(page);
1981 wait_for_stable_page(page);
1983 sb_end_pagefault(inode->i_sb);
1986 EXPORT_SYMBOL(filemap_page_mkwrite);
1988 const struct vm_operations_struct generic_file_vm_ops = {
1989 .fault = filemap_fault,
1990 .page_mkwrite = filemap_page_mkwrite,
1991 .remap_pages = generic_file_remap_pages,
1994 /* This is used for a general mmap of a disk file */
1996 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1998 struct address_space *mapping = file->f_mapping;
2000 if (!mapping->a_ops->readpage)
2002 file_accessed(file);
2003 vma->vm_ops = &generic_file_vm_ops;
2008 * This is for filesystems which do not implement ->writepage.
2010 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2012 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2014 return generic_file_mmap(file, vma);
2017 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2021 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2025 #endif /* CONFIG_MMU */
2027 EXPORT_SYMBOL(generic_file_mmap);
2028 EXPORT_SYMBOL(generic_file_readonly_mmap);
2030 static struct page *wait_on_page_read(struct page *page)
2032 if (!IS_ERR(page)) {
2033 wait_on_page_locked(page);
2034 if (!PageUptodate(page)) {
2035 page_cache_release(page);
2036 page = ERR_PTR(-EIO);
2042 static struct page *__read_cache_page(struct address_space *mapping,
2044 int (*filler)(void *, struct page *),
2051 page = find_get_page(mapping, index);
2053 page = __page_cache_alloc(gfp | __GFP_COLD);
2055 return ERR_PTR(-ENOMEM);
2056 err = add_to_page_cache_lru(page, mapping, index, gfp);
2057 if (unlikely(err)) {
2058 page_cache_release(page);
2061 /* Presumably ENOMEM for radix tree node */
2062 return ERR_PTR(err);
2064 err = filler(data, page);
2066 page_cache_release(page);
2067 page = ERR_PTR(err);
2069 page = wait_on_page_read(page);
2075 static struct page *do_read_cache_page(struct address_space *mapping,
2077 int (*filler)(void *, struct page *),
2086 page = __read_cache_page(mapping, index, filler, data, gfp);
2089 if (PageUptodate(page))
2093 if (!page->mapping) {
2095 page_cache_release(page);
2098 if (PageUptodate(page)) {
2102 err = filler(data, page);
2104 page_cache_release(page);
2105 return ERR_PTR(err);
2107 page = wait_on_page_read(page);
2112 mark_page_accessed(page);
2117 * read_cache_page - read into page cache, fill it if needed
2118 * @mapping: the page's address_space
2119 * @index: the page index
2120 * @filler: function to perform the read
2121 * @data: first arg to filler(data, page) function, often left as NULL
2123 * Read into the page cache. If a page already exists, and PageUptodate() is
2124 * not set, try to fill the page and wait for it to become unlocked.
2126 * If the page does not get brought uptodate, return -EIO.
2128 struct page *read_cache_page(struct address_space *mapping,
2130 int (*filler)(void *, struct page *),
2133 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2135 EXPORT_SYMBOL(read_cache_page);
2138 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2139 * @mapping: the page's address_space
2140 * @index: the page index
2141 * @gfp: the page allocator flags to use if allocating
2143 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2144 * any new page allocations done using the specified allocation flags.
2146 * If the page does not get brought uptodate, return -EIO.
2148 struct page *read_cache_page_gfp(struct address_space *mapping,
2152 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2154 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2156 EXPORT_SYMBOL(read_cache_page_gfp);
2158 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
2159 const struct iovec *iov, size_t base, size_t bytes)
2161 size_t copied = 0, left = 0;
2164 char __user *buf = iov->iov_base + base;
2165 int copy = min(bytes, iov->iov_len - base);
2168 left = __copy_from_user_inatomic(vaddr, buf, copy);
2177 return copied - left;
2181 * Copy as much as we can into the page and return the number of bytes which
2182 * were successfully copied. If a fault is encountered then return the number of
2183 * bytes which were copied.
2185 size_t iov_iter_copy_from_user_atomic(struct page *page,
2186 struct iov_iter *i, unsigned long offset, size_t bytes)
2191 kaddr = kmap_atomic(page);
2192 if (likely(i->nr_segs == 1)) {
2194 char __user *buf = i->iov->iov_base + i->iov_offset;
2195 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
2196 copied = bytes - left;
2198 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2199 i->iov, i->iov_offset, bytes);
2201 kunmap_atomic(kaddr);
2205 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
2208 * This has the same sideeffects and return value as
2209 * iov_iter_copy_from_user_atomic().
2210 * The difference is that it attempts to resolve faults.
2211 * Page must not be locked.
2213 size_t iov_iter_copy_from_user(struct page *page,
2214 struct iov_iter *i, unsigned long offset, size_t bytes)
2220 if (likely(i->nr_segs == 1)) {
2222 char __user *buf = i->iov->iov_base + i->iov_offset;
2223 left = __copy_from_user(kaddr + offset, buf, bytes);
2224 copied = bytes - left;
2226 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2227 i->iov, i->iov_offset, bytes);
2232 EXPORT_SYMBOL(iov_iter_copy_from_user);
2234 void iov_iter_advance(struct iov_iter *i, size_t bytes)
2236 BUG_ON(i->count < bytes);
2238 if (likely(i->nr_segs == 1)) {
2239 i->iov_offset += bytes;
2242 const struct iovec *iov = i->iov;
2243 size_t base = i->iov_offset;
2244 unsigned long nr_segs = i->nr_segs;
2247 * The !iov->iov_len check ensures we skip over unlikely
2248 * zero-length segments (without overruning the iovec).
2250 while (bytes || unlikely(i->count && !iov->iov_len)) {
2253 copy = min(bytes, iov->iov_len - base);
2254 BUG_ON(!i->count || i->count < copy);
2258 if (iov->iov_len == base) {
2265 i->iov_offset = base;
2266 i->nr_segs = nr_segs;
2269 EXPORT_SYMBOL(iov_iter_advance);
2272 * Fault in the first iovec of the given iov_iter, to a maximum length
2273 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2274 * accessed (ie. because it is an invalid address).
2276 * writev-intensive code may want this to prefault several iovecs -- that
2277 * would be possible (callers must not rely on the fact that _only_ the
2278 * first iovec will be faulted with the current implementation).
2280 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2282 char __user *buf = i->iov->iov_base + i->iov_offset;
2283 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2284 return fault_in_pages_readable(buf, bytes);
2286 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2289 * Return the count of just the current iov_iter segment.
2291 size_t iov_iter_single_seg_count(const struct iov_iter *i)
2293 const struct iovec *iov = i->iov;
2294 if (i->nr_segs == 1)
2297 return min(i->count, iov->iov_len - i->iov_offset);
2299 EXPORT_SYMBOL(iov_iter_single_seg_count);
2302 * Performs necessary checks before doing a write
2304 * Can adjust writing position or amount of bytes to write.
2305 * Returns appropriate error code that caller should return or
2306 * zero in case that write should be allowed.
2308 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2310 struct inode *inode = file->f_mapping->host;
2311 unsigned long limit = rlimit(RLIMIT_FSIZE);
2313 if (unlikely(*pos < 0))
2317 /* FIXME: this is for backwards compatibility with 2.4 */
2318 if (file->f_flags & O_APPEND)
2319 *pos = i_size_read(inode);
2321 if (limit != RLIM_INFINITY) {
2322 if (*pos >= limit) {
2323 send_sig(SIGXFSZ, current, 0);
2326 if (*count > limit - (typeof(limit))*pos) {
2327 *count = limit - (typeof(limit))*pos;
2335 if (unlikely(*pos + *count > MAX_NON_LFS &&
2336 !(file->f_flags & O_LARGEFILE))) {
2337 if (*pos >= MAX_NON_LFS) {
2340 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2341 *count = MAX_NON_LFS - (unsigned long)*pos;
2346 * Are we about to exceed the fs block limit ?
2348 * If we have written data it becomes a short write. If we have
2349 * exceeded without writing data we send a signal and return EFBIG.
2350 * Linus frestrict idea will clean these up nicely..
2352 if (likely(!isblk)) {
2353 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2354 if (*count || *pos > inode->i_sb->s_maxbytes) {
2357 /* zero-length writes at ->s_maxbytes are OK */
2360 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2361 *count = inode->i_sb->s_maxbytes - *pos;
2365 if (bdev_read_only(I_BDEV(inode)))
2367 isize = i_size_read(inode);
2368 if (*pos >= isize) {
2369 if (*count || *pos > isize)
2373 if (*pos + *count > isize)
2374 *count = isize - *pos;
2381 EXPORT_SYMBOL(generic_write_checks);
2383 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2384 loff_t pos, unsigned len, unsigned flags,
2385 struct page **pagep, void **fsdata)
2387 const struct address_space_operations *aops = mapping->a_ops;
2389 return aops->write_begin(file, mapping, pos, len, flags,
2392 EXPORT_SYMBOL(pagecache_write_begin);
2394 int pagecache_write_end(struct file *file, struct address_space *mapping,
2395 loff_t pos, unsigned len, unsigned copied,
2396 struct page *page, void *fsdata)
2398 const struct address_space_operations *aops = mapping->a_ops;
2400 mark_page_accessed(page);
2401 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2403 EXPORT_SYMBOL(pagecache_write_end);
2406 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2407 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2408 size_t count, size_t ocount)
2410 struct file *file = iocb->ki_filp;
2411 struct address_space *mapping = file->f_mapping;
2412 struct inode *inode = mapping->host;
2417 if (count != ocount)
2418 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2420 write_len = iov_length(iov, *nr_segs);
2421 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2423 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2428 * After a write we want buffered reads to be sure to go to disk to get
2429 * the new data. We invalidate clean cached page from the region we're
2430 * about to write. We do this *before* the write so that we can return
2431 * without clobbering -EIOCBQUEUED from ->direct_IO().
2433 if (mapping->nrpages) {
2434 written = invalidate_inode_pages2_range(mapping,
2435 pos >> PAGE_CACHE_SHIFT, end);
2437 * If a page can not be invalidated, return 0 to fall back
2438 * to buffered write.
2441 if (written == -EBUSY)
2447 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2450 * Finally, try again to invalidate clean pages which might have been
2451 * cached by non-direct readahead, or faulted in by get_user_pages()
2452 * if the source of the write was an mmap'ed region of the file
2453 * we're writing. Either one is a pretty crazy thing to do,
2454 * so we don't support it 100%. If this invalidation
2455 * fails, tough, the write still worked...
2457 if (mapping->nrpages) {
2458 invalidate_inode_pages2_range(mapping,
2459 pos >> PAGE_CACHE_SHIFT, end);
2464 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2465 i_size_write(inode, pos);
2466 mark_inode_dirty(inode);
2473 EXPORT_SYMBOL(generic_file_direct_write);
2476 * Find or create a page at the given pagecache position. Return the locked
2477 * page. This function is specifically for buffered writes.
2479 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2480 pgoff_t index, unsigned flags)
2485 gfp_t gfp_notmask = 0;
2487 gfp_mask = mapping_gfp_mask(mapping);
2488 if (mapping_cap_account_dirty(mapping))
2489 gfp_mask |= __GFP_WRITE;
2490 if (flags & AOP_FLAG_NOFS)
2491 gfp_notmask = __GFP_FS;
2493 page = find_lock_page(mapping, index);
2497 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2500 status = add_to_page_cache_lru(page, mapping, index,
2501 GFP_KERNEL & ~gfp_notmask);
2502 if (unlikely(status)) {
2503 page_cache_release(page);
2504 if (status == -EEXIST)
2509 wait_for_stable_page(page);
2512 EXPORT_SYMBOL(grab_cache_page_write_begin);
2514 static ssize_t generic_perform_write(struct file *file,
2515 struct iov_iter *i, loff_t pos)
2517 struct address_space *mapping = file->f_mapping;
2518 const struct address_space_operations *a_ops = mapping->a_ops;
2520 ssize_t written = 0;
2521 unsigned int flags = 0;
2524 * Copies from kernel address space cannot fail (NFSD is a big user).
2526 if (segment_eq(get_fs(), KERNEL_DS))
2527 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2531 unsigned long offset; /* Offset into pagecache page */
2532 unsigned long bytes; /* Bytes to write to page */
2533 size_t copied; /* Bytes copied from user */
2536 offset = (pos & (PAGE_CACHE_SIZE - 1));
2537 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2542 * Bring in the user page that we will copy from _first_.
2543 * Otherwise there's a nasty deadlock on copying from the
2544 * same page as we're writing to, without it being marked
2547 * Not only is this an optimisation, but it is also required
2548 * to check that the address is actually valid, when atomic
2549 * usercopies are used, below.
2551 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2556 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2558 if (unlikely(status))
2561 if (mapping_writably_mapped(mapping))
2562 flush_dcache_page(page);
2564 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2565 flush_dcache_page(page);
2567 mark_page_accessed(page);
2568 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2570 if (unlikely(status < 0))
2576 iov_iter_advance(i, copied);
2577 if (unlikely(copied == 0)) {
2579 * If we were unable to copy any data at all, we must
2580 * fall back to a single segment length write.
2582 * If we didn't fallback here, we could livelock
2583 * because not all segments in the iov can be copied at
2584 * once without a pagefault.
2586 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2587 iov_iter_single_seg_count(i));
2593 balance_dirty_pages_ratelimited(mapping);
2594 if (fatal_signal_pending(current)) {
2598 } while (iov_iter_count(i));
2600 return written ? written : status;
2604 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2605 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2606 size_t count, ssize_t written)
2608 struct file *file = iocb->ki_filp;
2612 iov_iter_init(&i, iov, nr_segs, count, written);
2613 status = generic_perform_write(file, &i, pos);
2615 if (likely(status >= 0)) {
2617 *ppos = pos + status;
2620 return written ? written : status;
2622 EXPORT_SYMBOL(generic_file_buffered_write);
2625 * __generic_file_aio_write - write data to a file
2626 * @iocb: IO state structure (file, offset, etc.)
2627 * @iov: vector with data to write
2628 * @nr_segs: number of segments in the vector
2629 * @ppos: position where to write
2631 * This function does all the work needed for actually writing data to a
2632 * file. It does all basic checks, removes SUID from the file, updates
2633 * modification times and calls proper subroutines depending on whether we
2634 * do direct IO or a standard buffered write.
2636 * It expects i_mutex to be grabbed unless we work on a block device or similar
2637 * object which does not need locking at all.
2639 * This function does *not* take care of syncing data in case of O_SYNC write.
2640 * A caller has to handle it. This is mainly due to the fact that we want to
2641 * avoid syncing under i_mutex.
2643 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2644 unsigned long nr_segs, loff_t *ppos)
2646 struct file *file = iocb->ki_filp;
2647 struct address_space * mapping = file->f_mapping;
2648 size_t ocount; /* original count */
2649 size_t count; /* after file limit checks */
2650 struct inode *inode = mapping->host;
2656 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2663 /* We can write back this queue in page reclaim */
2664 current->backing_dev_info = mapping->backing_dev_info;
2667 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2674 err = file_remove_suid(file);
2678 err = file_update_time(file);
2682 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2683 if (unlikely(file->f_flags & O_DIRECT)) {
2685 ssize_t written_buffered;
2687 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2688 ppos, count, ocount);
2689 if (written < 0 || written == count)
2692 * direct-io write to a hole: fall through to buffered I/O
2693 * for completing the rest of the request.
2697 written_buffered = generic_file_buffered_write(iocb, iov,
2698 nr_segs, pos, ppos, count,
2701 * If generic_file_buffered_write() retuned a synchronous error
2702 * then we want to return the number of bytes which were
2703 * direct-written, or the error code if that was zero. Note
2704 * that this differs from normal direct-io semantics, which
2705 * will return -EFOO even if some bytes were written.
2707 if (written_buffered < 0) {
2708 err = written_buffered;
2713 * We need to ensure that the page cache pages are written to
2714 * disk and invalidated to preserve the expected O_DIRECT
2717 endbyte = pos + written_buffered - written - 1;
2718 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2720 written = written_buffered;
2721 invalidate_mapping_pages(mapping,
2722 pos >> PAGE_CACHE_SHIFT,
2723 endbyte >> PAGE_CACHE_SHIFT);
2726 * We don't know how much we wrote, so just return
2727 * the number of bytes which were direct-written
2731 written = generic_file_buffered_write(iocb, iov, nr_segs,
2732 pos, ppos, count, written);
2735 current->backing_dev_info = NULL;
2736 return written ? written : err;
2738 EXPORT_SYMBOL(__generic_file_aio_write);
2741 * generic_file_aio_write - write data to a file
2742 * @iocb: IO state structure
2743 * @iov: vector with data to write
2744 * @nr_segs: number of segments in the vector
2745 * @pos: position in file where to write
2747 * This is a wrapper around __generic_file_aio_write() to be used by most
2748 * filesystems. It takes care of syncing the file in case of O_SYNC file
2749 * and acquires i_mutex as needed.
2751 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2752 unsigned long nr_segs, loff_t pos)
2754 struct file *file = iocb->ki_filp;
2755 struct inode *inode = file->f_mapping->host;
2758 BUG_ON(iocb->ki_pos != pos);
2760 mutex_lock(&inode->i_mutex);
2761 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2762 mutex_unlock(&inode->i_mutex);
2767 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2773 EXPORT_SYMBOL(generic_file_aio_write);
2776 * try_to_release_page() - release old fs-specific metadata on a page
2778 * @page: the page which the kernel is trying to free
2779 * @gfp_mask: memory allocation flags (and I/O mode)
2781 * The address_space is to try to release any data against the page
2782 * (presumably at page->private). If the release was successful, return `1'.
2783 * Otherwise return zero.
2785 * This may also be called if PG_fscache is set on a page, indicating that the
2786 * page is known to the local caching routines.
2788 * The @gfp_mask argument specifies whether I/O may be performed to release
2789 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2792 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2794 struct address_space * const mapping = page->mapping;
2796 BUG_ON(!PageLocked(page));
2797 if (PageWriteback(page))
2800 if (mapping && mapping->a_ops->releasepage)
2801 return mapping->a_ops->releasepage(page, gfp_mask);
2802 return try_to_free_buffers(page);
2805 EXPORT_SYMBOL(try_to_release_page);