4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/capability.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/gfp.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/backing-dev.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/cpuset.h>
32 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
33 #include <linux/hugetlb.h>
34 #include <linux/memcontrol.h>
35 #include <linux/cleancache.h>
36 #include <linux/rmap.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/filemap.h>
43 * FIXME: remove all knowledge of the buffer layer from the core VM
45 #include <linux/buffer_head.h> /* for try_to_free_buffers */
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * Shared mappings now work. 15.8.1995 Bruno.
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
64 * ->i_mmap_rwsem (truncate_pagecache)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
70 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
74 * ->page_table_lock or pte_lock (various, mainly in memory.c)
75 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
78 * ->lock_page (access_process_vm)
80 * ->i_mutex (generic_perform_write)
81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
84 * sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
103 * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat)
104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
112 static int page_cache_tree_insert(struct address_space *mapping,
113 struct page *page, void **shadowp)
115 struct radix_tree_node *node;
119 error = __radix_tree_create(&mapping->page_tree, page->index,
126 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
127 if (!radix_tree_exceptional_entry(p))
131 mapping->nrshadows--;
133 workingset_node_shadows_dec(node);
135 radix_tree_replace_slot(slot, page);
138 workingset_node_pages_inc(node);
140 * Don't track node that contains actual pages.
142 * Avoid acquiring the list_lru lock if already
143 * untracked. The list_empty() test is safe as
144 * node->private_list is protected by
145 * mapping->tree_lock.
147 if (!list_empty(&node->private_list))
148 list_lru_del(&workingset_shadow_nodes,
149 &node->private_list);
154 static void page_cache_tree_delete(struct address_space *mapping,
155 struct page *page, void *shadow)
157 struct radix_tree_node *node;
163 VM_BUG_ON(!PageLocked(page));
165 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
169 * We need a node to properly account shadow
170 * entries. Don't plant any without. XXX
176 mapping->nrshadows++;
178 * Make sure the nrshadows update is committed before
179 * the nrpages update so that final truncate racing
180 * with reclaim does not see both counters 0 at the
181 * same time and miss a shadow entry.
188 /* Clear direct pointer tags in root node */
189 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
190 radix_tree_replace_slot(slot, shadow);
194 /* Clear tree tags for the removed page */
196 offset = index & RADIX_TREE_MAP_MASK;
197 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
198 if (test_bit(offset, node->tags[tag]))
199 radix_tree_tag_clear(&mapping->page_tree, index, tag);
202 /* Delete page, swap shadow entry */
203 radix_tree_replace_slot(slot, shadow);
204 workingset_node_pages_dec(node);
206 workingset_node_shadows_inc(node);
208 if (__radix_tree_delete_node(&mapping->page_tree, node))
212 * Track node that only contains shadow entries.
214 * Avoid acquiring the list_lru lock if already tracked. The
215 * list_empty() test is safe as node->private_list is
216 * protected by mapping->tree_lock.
218 if (!workingset_node_pages(node) &&
219 list_empty(&node->private_list)) {
220 node->private_data = mapping;
221 list_lru_add(&workingset_shadow_nodes, &node->private_list);
226 * Delete a page from the page cache and free it. Caller has to make
227 * sure the page is locked and that nobody else uses it - or that usage
228 * is safe. The caller must hold the mapping's tree_lock and
229 * mem_cgroup_begin_page_stat().
231 void __delete_from_page_cache(struct page *page, void *shadow,
232 struct mem_cgroup *memcg)
234 struct address_space *mapping = page->mapping;
236 trace_mm_filemap_delete_from_page_cache(page);
238 * if we're uptodate, flush out into the cleancache, otherwise
239 * invalidate any existing cleancache entries. We can't leave
240 * stale data around in the cleancache once our page is gone
242 if (PageUptodate(page) && PageMappedToDisk(page))
243 cleancache_put_page(page);
245 cleancache_invalidate_page(mapping, page);
247 page_cache_tree_delete(mapping, page, shadow);
249 page->mapping = NULL;
250 /* Leave page->index set: truncation lookup relies upon it */
252 /* hugetlb pages do not participate in page cache accounting. */
254 __dec_zone_page_state(page, NR_FILE_PAGES);
255 if (PageSwapBacked(page))
256 __dec_zone_page_state(page, NR_SHMEM);
257 BUG_ON(page_mapped(page));
260 * At this point page must be either written or cleaned by truncate.
261 * Dirty page here signals a bug and loss of unwritten data.
263 * This fixes dirty accounting after removing the page entirely but
264 * leaves PageDirty set: it has no effect for truncated page and
265 * anyway will be cleared before returning page into buddy allocator.
267 if (WARN_ON_ONCE(PageDirty(page)))
268 account_page_cleaned(page, mapping, memcg,
269 inode_to_wb(mapping->host));
273 * delete_from_page_cache - delete page from page cache
274 * @page: the page which the kernel is trying to remove from page cache
276 * This must be called only on pages that have been verified to be in the page
277 * cache and locked. It will never put the page into the free list, the caller
278 * has a reference on the page.
280 void delete_from_page_cache(struct page *page)
282 struct address_space *mapping = page->mapping;
283 struct mem_cgroup *memcg;
286 void (*freepage)(struct page *);
288 BUG_ON(!PageLocked(page));
290 freepage = mapping->a_ops->freepage;
292 memcg = mem_cgroup_begin_page_stat(page);
293 spin_lock_irqsave(&mapping->tree_lock, flags);
294 __delete_from_page_cache(page, NULL, memcg);
295 spin_unlock_irqrestore(&mapping->tree_lock, flags);
296 mem_cgroup_end_page_stat(memcg);
300 page_cache_release(page);
302 EXPORT_SYMBOL(delete_from_page_cache);
304 static int filemap_check_errors(struct address_space *mapping)
307 /* Check for outstanding write errors */
308 if (test_bit(AS_ENOSPC, &mapping->flags) &&
309 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
311 if (test_bit(AS_EIO, &mapping->flags) &&
312 test_and_clear_bit(AS_EIO, &mapping->flags))
318 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
319 * @mapping: address space structure to write
320 * @start: offset in bytes where the range starts
321 * @end: offset in bytes where the range ends (inclusive)
322 * @sync_mode: enable synchronous operation
324 * Start writeback against all of a mapping's dirty pages that lie
325 * within the byte offsets <start, end> inclusive.
327 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
328 * opposed to a regular memory cleansing writeback. The difference between
329 * these two operations is that if a dirty page/buffer is encountered, it must
330 * be waited upon, and not just skipped over.
332 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
333 loff_t end, int sync_mode)
336 struct writeback_control wbc = {
337 .sync_mode = sync_mode,
338 .nr_to_write = LONG_MAX,
339 .range_start = start,
343 if (!mapping_cap_writeback_dirty(mapping))
346 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
347 ret = do_writepages(mapping, &wbc);
348 wbc_detach_inode(&wbc);
352 static inline int __filemap_fdatawrite(struct address_space *mapping,
355 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
358 int filemap_fdatawrite(struct address_space *mapping)
360 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
362 EXPORT_SYMBOL(filemap_fdatawrite);
364 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
367 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
369 EXPORT_SYMBOL(filemap_fdatawrite_range);
372 * filemap_flush - mostly a non-blocking flush
373 * @mapping: target address_space
375 * This is a mostly non-blocking flush. Not suitable for data-integrity
376 * purposes - I/O may not be started against all dirty pages.
378 int filemap_flush(struct address_space *mapping)
380 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
382 EXPORT_SYMBOL(filemap_flush);
384 static int __filemap_fdatawait_range(struct address_space *mapping,
385 loff_t start_byte, loff_t end_byte)
387 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
388 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
393 if (end_byte < start_byte)
396 pagevec_init(&pvec, 0);
397 while ((index <= end) &&
398 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
399 PAGECACHE_TAG_WRITEBACK,
400 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
403 for (i = 0; i < nr_pages; i++) {
404 struct page *page = pvec.pages[i];
406 /* until radix tree lookup accepts end_index */
407 if (page->index > end)
410 wait_on_page_writeback(page);
411 if (TestClearPageError(page))
414 pagevec_release(&pvec);
422 * filemap_fdatawait_range - wait for writeback to complete
423 * @mapping: address space structure to wait for
424 * @start_byte: offset in bytes where the range starts
425 * @end_byte: offset in bytes where the range ends (inclusive)
427 * Walk the list of under-writeback pages of the given address space
428 * in the given range and wait for all of them. Check error status of
429 * the address space and return it.
431 * Since the error status of the address space is cleared by this function,
432 * callers are responsible for checking the return value and handling and/or
433 * reporting the error.
435 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
440 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
441 ret2 = filemap_check_errors(mapping);
447 EXPORT_SYMBOL(filemap_fdatawait_range);
450 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
451 * @mapping: address space structure to wait for
453 * Walk the list of under-writeback pages of the given address space
454 * and wait for all of them. Unlike filemap_fdatawait(), this function
455 * does not clear error status of the address space.
457 * Use this function if callers don't handle errors themselves. Expected
458 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
461 void filemap_fdatawait_keep_errors(struct address_space *mapping)
463 loff_t i_size = i_size_read(mapping->host);
468 __filemap_fdatawait_range(mapping, 0, i_size - 1);
472 * filemap_fdatawait - wait for all under-writeback pages to complete
473 * @mapping: address space structure to wait for
475 * Walk the list of under-writeback pages of the given address space
476 * and wait for all of them. Check error status of the address space
479 * Since the error status of the address space is cleared by this function,
480 * callers are responsible for checking the return value and handling and/or
481 * reporting the error.
483 int filemap_fdatawait(struct address_space *mapping)
485 loff_t i_size = i_size_read(mapping->host);
490 return filemap_fdatawait_range(mapping, 0, i_size - 1);
492 EXPORT_SYMBOL(filemap_fdatawait);
494 int filemap_write_and_wait(struct address_space *mapping)
498 if (mapping->nrpages) {
499 err = filemap_fdatawrite(mapping);
501 * Even if the above returned error, the pages may be
502 * written partially (e.g. -ENOSPC), so we wait for it.
503 * But the -EIO is special case, it may indicate the worst
504 * thing (e.g. bug) happened, so we avoid waiting for it.
507 int err2 = filemap_fdatawait(mapping);
512 err = filemap_check_errors(mapping);
516 EXPORT_SYMBOL(filemap_write_and_wait);
519 * filemap_write_and_wait_range - write out & wait on a file range
520 * @mapping: the address_space for the pages
521 * @lstart: offset in bytes where the range starts
522 * @lend: offset in bytes where the range ends (inclusive)
524 * Write out and wait upon file offsets lstart->lend, inclusive.
526 * Note that `lend' is inclusive (describes the last byte to be written) so
527 * that this function can be used to write to the very end-of-file (end = -1).
529 int filemap_write_and_wait_range(struct address_space *mapping,
530 loff_t lstart, loff_t lend)
534 if (mapping->nrpages) {
535 err = __filemap_fdatawrite_range(mapping, lstart, lend,
537 /* See comment of filemap_write_and_wait() */
539 int err2 = filemap_fdatawait_range(mapping,
545 err = filemap_check_errors(mapping);
549 EXPORT_SYMBOL(filemap_write_and_wait_range);
552 * replace_page_cache_page - replace a pagecache page with a new one
553 * @old: page to be replaced
554 * @new: page to replace with
555 * @gfp_mask: allocation mode
557 * This function replaces a page in the pagecache with a new one. On
558 * success it acquires the pagecache reference for the new page and
559 * drops it for the old page. Both the old and new pages must be
560 * locked. This function does not add the new page to the LRU, the
561 * caller must do that.
563 * The remove + add is atomic. The only way this function can fail is
564 * memory allocation failure.
566 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
570 VM_BUG_ON_PAGE(!PageLocked(old), old);
571 VM_BUG_ON_PAGE(!PageLocked(new), new);
572 VM_BUG_ON_PAGE(new->mapping, new);
574 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
576 struct address_space *mapping = old->mapping;
577 void (*freepage)(struct page *);
578 struct mem_cgroup *memcg;
581 pgoff_t offset = old->index;
582 freepage = mapping->a_ops->freepage;
585 new->mapping = mapping;
588 memcg = mem_cgroup_begin_page_stat(old);
589 spin_lock_irqsave(&mapping->tree_lock, flags);
590 __delete_from_page_cache(old, NULL, memcg);
591 error = page_cache_tree_insert(mapping, new, NULL);
596 * hugetlb pages do not participate in page cache accounting.
599 __inc_zone_page_state(new, NR_FILE_PAGES);
600 if (PageSwapBacked(new))
601 __inc_zone_page_state(new, NR_SHMEM);
602 spin_unlock_irqrestore(&mapping->tree_lock, flags);
603 mem_cgroup_end_page_stat(memcg);
604 mem_cgroup_replace_page(old, new);
605 radix_tree_preload_end();
608 page_cache_release(old);
613 EXPORT_SYMBOL_GPL(replace_page_cache_page);
615 static int __add_to_page_cache_locked(struct page *page,
616 struct address_space *mapping,
617 pgoff_t offset, gfp_t gfp_mask,
620 int huge = PageHuge(page);
621 struct mem_cgroup *memcg;
624 VM_BUG_ON_PAGE(!PageLocked(page), page);
625 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
628 error = mem_cgroup_try_charge(page, current->mm,
634 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
637 mem_cgroup_cancel_charge(page, memcg);
641 page_cache_get(page);
642 page->mapping = mapping;
643 page->index = offset;
645 spin_lock_irq(&mapping->tree_lock);
646 error = page_cache_tree_insert(mapping, page, shadowp);
647 radix_tree_preload_end();
651 /* hugetlb pages do not participate in page cache accounting. */
653 __inc_zone_page_state(page, NR_FILE_PAGES);
654 spin_unlock_irq(&mapping->tree_lock);
656 mem_cgroup_commit_charge(page, memcg, false);
657 trace_mm_filemap_add_to_page_cache(page);
660 page->mapping = NULL;
661 /* Leave page->index set: truncation relies upon it */
662 spin_unlock_irq(&mapping->tree_lock);
664 mem_cgroup_cancel_charge(page, memcg);
665 page_cache_release(page);
670 * add_to_page_cache_locked - add a locked page to the pagecache
672 * @mapping: the page's address_space
673 * @offset: page index
674 * @gfp_mask: page allocation mode
676 * This function is used to add a page to the pagecache. It must be locked.
677 * This function does not add the page to the LRU. The caller must do that.
679 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
680 pgoff_t offset, gfp_t gfp_mask)
682 return __add_to_page_cache_locked(page, mapping, offset,
685 EXPORT_SYMBOL(add_to_page_cache_locked);
687 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
688 pgoff_t offset, gfp_t gfp_mask)
693 __set_page_locked(page);
694 ret = __add_to_page_cache_locked(page, mapping, offset,
697 __clear_page_locked(page);
700 * The page might have been evicted from cache only
701 * recently, in which case it should be activated like
702 * any other repeatedly accessed page.
704 if (shadow && workingset_refault(shadow)) {
706 workingset_activation(page);
708 ClearPageActive(page);
713 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
716 struct page *__page_cache_alloc(gfp_t gfp)
721 if (cpuset_do_page_mem_spread()) {
722 unsigned int cpuset_mems_cookie;
724 cpuset_mems_cookie = read_mems_allowed_begin();
725 n = cpuset_mem_spread_node();
726 page = __alloc_pages_node(n, gfp, 0);
727 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
731 return alloc_pages(gfp, 0);
733 EXPORT_SYMBOL(__page_cache_alloc);
737 * In order to wait for pages to become available there must be
738 * waitqueues associated with pages. By using a hash table of
739 * waitqueues where the bucket discipline is to maintain all
740 * waiters on the same queue and wake all when any of the pages
741 * become available, and for the woken contexts to check to be
742 * sure the appropriate page became available, this saves space
743 * at a cost of "thundering herd" phenomena during rare hash
746 wait_queue_head_t *page_waitqueue(struct page *page)
748 const struct zone *zone = page_zone(page);
750 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
752 EXPORT_SYMBOL(page_waitqueue);
754 void wait_on_page_bit(struct page *page, int bit_nr)
756 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
758 if (test_bit(bit_nr, &page->flags))
759 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
760 TASK_UNINTERRUPTIBLE);
762 EXPORT_SYMBOL(wait_on_page_bit);
764 int wait_on_page_bit_killable(struct page *page, int bit_nr)
766 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
768 if (!test_bit(bit_nr, &page->flags))
771 return __wait_on_bit(page_waitqueue(page), &wait,
772 bit_wait_io, TASK_KILLABLE);
775 int wait_on_page_bit_killable_timeout(struct page *page,
776 int bit_nr, unsigned long timeout)
778 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
780 wait.key.timeout = jiffies + timeout;
781 if (!test_bit(bit_nr, &page->flags))
783 return __wait_on_bit(page_waitqueue(page), &wait,
784 bit_wait_io_timeout, TASK_KILLABLE);
786 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
789 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
790 * @page: Page defining the wait queue of interest
791 * @waiter: Waiter to add to the queue
793 * Add an arbitrary @waiter to the wait queue for the nominated @page.
795 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
797 wait_queue_head_t *q = page_waitqueue(page);
800 spin_lock_irqsave(&q->lock, flags);
801 __add_wait_queue(q, waiter);
802 spin_unlock_irqrestore(&q->lock, flags);
804 EXPORT_SYMBOL_GPL(add_page_wait_queue);
807 * unlock_page - unlock a locked page
810 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
811 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
812 * mechanism between PageLocked pages and PageWriteback pages is shared.
813 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
815 * The mb is necessary to enforce ordering between the clear_bit and the read
816 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
818 void unlock_page(struct page *page)
820 VM_BUG_ON_PAGE(!PageLocked(page), page);
821 clear_bit_unlock(PG_locked, &page->flags);
822 smp_mb__after_atomic();
823 wake_up_page(page, PG_locked);
825 EXPORT_SYMBOL(unlock_page);
828 * end_page_writeback - end writeback against a page
831 void end_page_writeback(struct page *page)
834 * TestClearPageReclaim could be used here but it is an atomic
835 * operation and overkill in this particular case. Failing to
836 * shuffle a page marked for immediate reclaim is too mild to
837 * justify taking an atomic operation penalty at the end of
838 * ever page writeback.
840 if (PageReclaim(page)) {
841 ClearPageReclaim(page);
842 rotate_reclaimable_page(page);
845 if (!test_clear_page_writeback(page))
848 smp_mb__after_atomic();
849 wake_up_page(page, PG_writeback);
851 EXPORT_SYMBOL(end_page_writeback);
854 * After completing I/O on a page, call this routine to update the page
855 * flags appropriately
857 void page_endio(struct page *page, int rw, int err)
861 SetPageUptodate(page);
863 ClearPageUptodate(page);
867 } else { /* rw == WRITE */
871 mapping_set_error(page->mapping, err);
873 end_page_writeback(page);
876 EXPORT_SYMBOL_GPL(page_endio);
879 * __lock_page - get a lock on the page, assuming we need to sleep to get it
880 * @page: the page to lock
882 void __lock_page(struct page *page)
884 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
886 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
887 TASK_UNINTERRUPTIBLE);
889 EXPORT_SYMBOL(__lock_page);
891 int __lock_page_killable(struct page *page)
893 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
895 return __wait_on_bit_lock(page_waitqueue(page), &wait,
896 bit_wait_io, TASK_KILLABLE);
898 EXPORT_SYMBOL_GPL(__lock_page_killable);
902 * 1 - page is locked; mmap_sem is still held.
903 * 0 - page is not locked.
904 * mmap_sem has been released (up_read()), unless flags had both
905 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
906 * which case mmap_sem is still held.
908 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
909 * with the page locked and the mmap_sem unperturbed.
911 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
914 if (flags & FAULT_FLAG_ALLOW_RETRY) {
916 * CAUTION! In this case, mmap_sem is not released
917 * even though return 0.
919 if (flags & FAULT_FLAG_RETRY_NOWAIT)
922 up_read(&mm->mmap_sem);
923 if (flags & FAULT_FLAG_KILLABLE)
924 wait_on_page_locked_killable(page);
926 wait_on_page_locked(page);
929 if (flags & FAULT_FLAG_KILLABLE) {
932 ret = __lock_page_killable(page);
934 up_read(&mm->mmap_sem);
944 * page_cache_next_hole - find the next hole (not-present entry)
947 * @max_scan: maximum range to search
949 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
950 * lowest indexed hole.
952 * Returns: the index of the hole if found, otherwise returns an index
953 * outside of the set specified (in which case 'return - index >=
954 * max_scan' will be true). In rare cases of index wrap-around, 0 will
957 * page_cache_next_hole may be called under rcu_read_lock. However,
958 * like radix_tree_gang_lookup, this will not atomically search a
959 * snapshot of the tree at a single point in time. For example, if a
960 * hole is created at index 5, then subsequently a hole is created at
961 * index 10, page_cache_next_hole covering both indexes may return 10
962 * if called under rcu_read_lock.
964 pgoff_t page_cache_next_hole(struct address_space *mapping,
965 pgoff_t index, unsigned long max_scan)
969 for (i = 0; i < max_scan; i++) {
972 page = radix_tree_lookup(&mapping->page_tree, index);
973 if (!page || radix_tree_exceptional_entry(page))
982 EXPORT_SYMBOL(page_cache_next_hole);
985 * page_cache_prev_hole - find the prev hole (not-present entry)
988 * @max_scan: maximum range to search
990 * Search backwards in the range [max(index-max_scan+1, 0), index] for
993 * Returns: the index of the hole if found, otherwise returns an index
994 * outside of the set specified (in which case 'index - return >=
995 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
998 * page_cache_prev_hole may be called under rcu_read_lock. However,
999 * like radix_tree_gang_lookup, this will not atomically search a
1000 * snapshot of the tree at a single point in time. For example, if a
1001 * hole is created at index 10, then subsequently a hole is created at
1002 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1003 * called under rcu_read_lock.
1005 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1006 pgoff_t index, unsigned long max_scan)
1010 for (i = 0; i < max_scan; i++) {
1013 page = radix_tree_lookup(&mapping->page_tree, index);
1014 if (!page || radix_tree_exceptional_entry(page))
1017 if (index == ULONG_MAX)
1023 EXPORT_SYMBOL(page_cache_prev_hole);
1026 * find_get_entry - find and get a page cache entry
1027 * @mapping: the address_space to search
1028 * @offset: the page cache index
1030 * Looks up the page cache slot at @mapping & @offset. If there is a
1031 * page cache page, it is returned with an increased refcount.
1033 * If the slot holds a shadow entry of a previously evicted page, or a
1034 * swap entry from shmem/tmpfs, it is returned.
1036 * Otherwise, %NULL is returned.
1038 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1046 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1048 page = radix_tree_deref_slot(pagep);
1049 if (unlikely(!page))
1051 if (radix_tree_exception(page)) {
1052 if (radix_tree_deref_retry(page))
1055 * A shadow entry of a recently evicted page,
1056 * or a swap entry from shmem/tmpfs. Return
1057 * it without attempting to raise page count.
1061 if (!page_cache_get_speculative(page))
1065 * Has the page moved?
1066 * This is part of the lockless pagecache protocol. See
1067 * include/linux/pagemap.h for details.
1069 if (unlikely(page != *pagep)) {
1070 page_cache_release(page);
1079 EXPORT_SYMBOL(find_get_entry);
1082 * find_lock_entry - locate, pin and lock a page cache entry
1083 * @mapping: the address_space to search
1084 * @offset: the page cache index
1086 * Looks up the page cache slot at @mapping & @offset. If there is a
1087 * page cache page, it is returned locked and with an increased
1090 * If the slot holds a shadow entry of a previously evicted page, or a
1091 * swap entry from shmem/tmpfs, it is returned.
1093 * Otherwise, %NULL is returned.
1095 * find_lock_entry() may sleep.
1097 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1102 page = find_get_entry(mapping, offset);
1103 if (page && !radix_tree_exception(page)) {
1105 /* Has the page been truncated? */
1106 if (unlikely(page->mapping != mapping)) {
1108 page_cache_release(page);
1111 VM_BUG_ON_PAGE(page->index != offset, page);
1115 EXPORT_SYMBOL(find_lock_entry);
1118 * pagecache_get_page - find and get a page reference
1119 * @mapping: the address_space to search
1120 * @offset: the page index
1121 * @fgp_flags: PCG flags
1122 * @gfp_mask: gfp mask to use for the page cache data page allocation
1124 * Looks up the page cache slot at @mapping & @offset.
1126 * PCG flags modify how the page is returned.
1128 * FGP_ACCESSED: the page will be marked accessed
1129 * FGP_LOCK: Page is return locked
1130 * FGP_CREAT: If page is not present then a new page is allocated using
1131 * @gfp_mask and added to the page cache and the VM's LRU
1132 * list. The page is returned locked and with an increased
1133 * refcount. Otherwise, %NULL is returned.
1135 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1136 * if the GFP flags specified for FGP_CREAT are atomic.
1138 * If there is a page cache page, it is returned with an increased refcount.
1140 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1141 int fgp_flags, gfp_t gfp_mask)
1146 page = find_get_entry(mapping, offset);
1147 if (radix_tree_exceptional_entry(page))
1152 if (fgp_flags & FGP_LOCK) {
1153 if (fgp_flags & FGP_NOWAIT) {
1154 if (!trylock_page(page)) {
1155 page_cache_release(page);
1162 /* Has the page been truncated? */
1163 if (unlikely(page->mapping != mapping)) {
1165 page_cache_release(page);
1168 VM_BUG_ON_PAGE(page->index != offset, page);
1171 if (page && (fgp_flags & FGP_ACCESSED))
1172 mark_page_accessed(page);
1175 if (!page && (fgp_flags & FGP_CREAT)) {
1177 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1178 gfp_mask |= __GFP_WRITE;
1179 if (fgp_flags & FGP_NOFS)
1180 gfp_mask &= ~__GFP_FS;
1182 page = __page_cache_alloc(gfp_mask);
1186 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1187 fgp_flags |= FGP_LOCK;
1189 /* Init accessed so avoid atomic mark_page_accessed later */
1190 if (fgp_flags & FGP_ACCESSED)
1191 __SetPageReferenced(page);
1193 err = add_to_page_cache_lru(page, mapping, offset,
1194 gfp_mask & GFP_RECLAIM_MASK);
1195 if (unlikely(err)) {
1196 page_cache_release(page);
1205 EXPORT_SYMBOL(pagecache_get_page);
1208 * find_get_entries - gang pagecache lookup
1209 * @mapping: The address_space to search
1210 * @start: The starting page cache index
1211 * @nr_entries: The maximum number of entries
1212 * @entries: Where the resulting entries are placed
1213 * @indices: The cache indices corresponding to the entries in @entries
1215 * find_get_entries() will search for and return a group of up to
1216 * @nr_entries entries in the mapping. The entries are placed at
1217 * @entries. find_get_entries() takes a reference against any actual
1220 * The search returns a group of mapping-contiguous page cache entries
1221 * with ascending indexes. There may be holes in the indices due to
1222 * not-present pages.
1224 * Any shadow entries of evicted pages, or swap entries from
1225 * shmem/tmpfs, are included in the returned array.
1227 * find_get_entries() returns the number of pages and shadow entries
1230 unsigned find_get_entries(struct address_space *mapping,
1231 pgoff_t start, unsigned int nr_entries,
1232 struct page **entries, pgoff_t *indices)
1235 unsigned int ret = 0;
1236 struct radix_tree_iter iter;
1243 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1246 page = radix_tree_deref_slot(slot);
1247 if (unlikely(!page))
1249 if (radix_tree_exception(page)) {
1250 if (radix_tree_deref_retry(page))
1253 * A shadow entry of a recently evicted page,
1254 * or a swap entry from shmem/tmpfs. Return
1255 * it without attempting to raise page count.
1259 if (!page_cache_get_speculative(page))
1262 /* Has the page moved? */
1263 if (unlikely(page != *slot)) {
1264 page_cache_release(page);
1268 indices[ret] = iter.index;
1269 entries[ret] = page;
1270 if (++ret == nr_entries)
1278 * find_get_pages - gang pagecache lookup
1279 * @mapping: The address_space to search
1280 * @start: The starting page index
1281 * @nr_pages: The maximum number of pages
1282 * @pages: Where the resulting pages are placed
1284 * find_get_pages() will search for and return a group of up to
1285 * @nr_pages pages in the mapping. The pages are placed at @pages.
1286 * find_get_pages() takes a reference against the returned pages.
1288 * The search returns a group of mapping-contiguous pages with ascending
1289 * indexes. There may be holes in the indices due to not-present pages.
1291 * find_get_pages() returns the number of pages which were found.
1293 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1294 unsigned int nr_pages, struct page **pages)
1296 struct radix_tree_iter iter;
1300 if (unlikely(!nr_pages))
1305 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1308 page = radix_tree_deref_slot(slot);
1309 if (unlikely(!page))
1312 if (radix_tree_exception(page)) {
1313 if (radix_tree_deref_retry(page)) {
1315 * Transient condition which can only trigger
1316 * when entry at index 0 moves out of or back
1317 * to root: none yet gotten, safe to restart.
1319 WARN_ON(iter.index);
1323 * A shadow entry of a recently evicted page,
1324 * or a swap entry from shmem/tmpfs. Skip
1330 if (!page_cache_get_speculative(page))
1333 /* Has the page moved? */
1334 if (unlikely(page != *slot)) {
1335 page_cache_release(page);
1340 if (++ret == nr_pages)
1349 * find_get_pages_contig - gang contiguous pagecache lookup
1350 * @mapping: The address_space to search
1351 * @index: The starting page index
1352 * @nr_pages: The maximum number of pages
1353 * @pages: Where the resulting pages are placed
1355 * find_get_pages_contig() works exactly like find_get_pages(), except
1356 * that the returned number of pages are guaranteed to be contiguous.
1358 * find_get_pages_contig() returns the number of pages which were found.
1360 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1361 unsigned int nr_pages, struct page **pages)
1363 struct radix_tree_iter iter;
1365 unsigned int ret = 0;
1367 if (unlikely(!nr_pages))
1372 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1375 page = radix_tree_deref_slot(slot);
1376 /* The hole, there no reason to continue */
1377 if (unlikely(!page))
1380 if (radix_tree_exception(page)) {
1381 if (radix_tree_deref_retry(page)) {
1383 * Transient condition which can only trigger
1384 * when entry at index 0 moves out of or back
1385 * to root: none yet gotten, safe to restart.
1390 * A shadow entry of a recently evicted page,
1391 * or a swap entry from shmem/tmpfs. Stop
1392 * looking for contiguous pages.
1397 if (!page_cache_get_speculative(page))
1400 /* Has the page moved? */
1401 if (unlikely(page != *slot)) {
1402 page_cache_release(page);
1407 * must check mapping and index after taking the ref.
1408 * otherwise we can get both false positives and false
1409 * negatives, which is just confusing to the caller.
1411 if (page->mapping == NULL || page->index != iter.index) {
1412 page_cache_release(page);
1417 if (++ret == nr_pages)
1423 EXPORT_SYMBOL(find_get_pages_contig);
1426 * find_get_pages_tag - find and return pages that match @tag
1427 * @mapping: the address_space to search
1428 * @index: the starting page index
1429 * @tag: the tag index
1430 * @nr_pages: the maximum number of pages
1431 * @pages: where the resulting pages are placed
1433 * Like find_get_pages, except we only return pages which are tagged with
1434 * @tag. We update @index to index the next page for the traversal.
1436 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1437 int tag, unsigned int nr_pages, struct page **pages)
1439 struct radix_tree_iter iter;
1443 if (unlikely(!nr_pages))
1448 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1449 &iter, *index, tag) {
1452 page = radix_tree_deref_slot(slot);
1453 if (unlikely(!page))
1456 if (radix_tree_exception(page)) {
1457 if (radix_tree_deref_retry(page)) {
1459 * Transient condition which can only trigger
1460 * when entry at index 0 moves out of or back
1461 * to root: none yet gotten, safe to restart.
1466 * A shadow entry of a recently evicted page.
1468 * Those entries should never be tagged, but
1469 * this tree walk is lockless and the tags are
1470 * looked up in bulk, one radix tree node at a
1471 * time, so there is a sizable window for page
1472 * reclaim to evict a page we saw tagged.
1479 if (!page_cache_get_speculative(page))
1482 /* Has the page moved? */
1483 if (unlikely(page != *slot)) {
1484 page_cache_release(page);
1489 if (++ret == nr_pages)
1496 *index = pages[ret - 1]->index + 1;
1500 EXPORT_SYMBOL(find_get_pages_tag);
1503 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1504 * a _large_ part of the i/o request. Imagine the worst scenario:
1506 * ---R__________________________________________B__________
1507 * ^ reading here ^ bad block(assume 4k)
1509 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1510 * => failing the whole request => read(R) => read(R+1) =>
1511 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1512 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1513 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1515 * It is going insane. Fix it by quickly scaling down the readahead size.
1517 static void shrink_readahead_size_eio(struct file *filp,
1518 struct file_ra_state *ra)
1524 * do_generic_file_read - generic file read routine
1525 * @filp: the file to read
1526 * @ppos: current file position
1527 * @iter: data destination
1528 * @written: already copied
1530 * This is a generic file read routine, and uses the
1531 * mapping->a_ops->readpage() function for the actual low-level stuff.
1533 * This is really ugly. But the goto's actually try to clarify some
1534 * of the logic when it comes to error handling etc.
1536 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1537 struct iov_iter *iter, ssize_t written)
1539 struct address_space *mapping = filp->f_mapping;
1540 struct inode *inode = mapping->host;
1541 struct file_ra_state *ra = &filp->f_ra;
1545 unsigned long offset; /* offset into pagecache page */
1546 unsigned int prev_offset;
1549 index = *ppos >> PAGE_CACHE_SHIFT;
1550 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1551 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1552 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1553 offset = *ppos & ~PAGE_CACHE_MASK;
1559 unsigned long nr, ret;
1563 page = find_get_page(mapping, index);
1565 page_cache_sync_readahead(mapping,
1567 index, last_index - index);
1568 page = find_get_page(mapping, index);
1569 if (unlikely(page == NULL))
1570 goto no_cached_page;
1572 if (PageReadahead(page)) {
1573 page_cache_async_readahead(mapping,
1575 index, last_index - index);
1577 if (!PageUptodate(page)) {
1578 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1579 !mapping->a_ops->is_partially_uptodate)
1580 goto page_not_up_to_date;
1581 if (!trylock_page(page))
1582 goto page_not_up_to_date;
1583 /* Did it get truncated before we got the lock? */
1585 goto page_not_up_to_date_locked;
1586 if (!mapping->a_ops->is_partially_uptodate(page,
1587 offset, iter->count))
1588 goto page_not_up_to_date_locked;
1593 * i_size must be checked after we know the page is Uptodate.
1595 * Checking i_size after the check allows us to calculate
1596 * the correct value for "nr", which means the zero-filled
1597 * part of the page is not copied back to userspace (unless
1598 * another truncate extends the file - this is desired though).
1601 isize = i_size_read(inode);
1602 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1603 if (unlikely(!isize || index > end_index)) {
1604 page_cache_release(page);
1608 /* nr is the maximum number of bytes to copy from this page */
1609 nr = PAGE_CACHE_SIZE;
1610 if (index == end_index) {
1611 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1613 page_cache_release(page);
1619 /* If users can be writing to this page using arbitrary
1620 * virtual addresses, take care about potential aliasing
1621 * before reading the page on the kernel side.
1623 if (mapping_writably_mapped(mapping))
1624 flush_dcache_page(page);
1627 * When a sequential read accesses a page several times,
1628 * only mark it as accessed the first time.
1630 if (prev_index != index || offset != prev_offset)
1631 mark_page_accessed(page);
1635 * Ok, we have the page, and it's up-to-date, so
1636 * now we can copy it to user space...
1639 ret = copy_page_to_iter(page, offset, nr, iter);
1641 index += offset >> PAGE_CACHE_SHIFT;
1642 offset &= ~PAGE_CACHE_MASK;
1643 prev_offset = offset;
1645 page_cache_release(page);
1647 if (!iov_iter_count(iter))
1655 page_not_up_to_date:
1656 /* Get exclusive access to the page ... */
1657 error = lock_page_killable(page);
1658 if (unlikely(error))
1659 goto readpage_error;
1661 page_not_up_to_date_locked:
1662 /* Did it get truncated before we got the lock? */
1663 if (!page->mapping) {
1665 page_cache_release(page);
1669 /* Did somebody else fill it already? */
1670 if (PageUptodate(page)) {
1677 * A previous I/O error may have been due to temporary
1678 * failures, eg. multipath errors.
1679 * PG_error will be set again if readpage fails.
1681 ClearPageError(page);
1682 /* Start the actual read. The read will unlock the page. */
1683 error = mapping->a_ops->readpage(filp, page);
1685 if (unlikely(error)) {
1686 if (error == AOP_TRUNCATED_PAGE) {
1687 page_cache_release(page);
1691 goto readpage_error;
1694 if (!PageUptodate(page)) {
1695 error = lock_page_killable(page);
1696 if (unlikely(error))
1697 goto readpage_error;
1698 if (!PageUptodate(page)) {
1699 if (page->mapping == NULL) {
1701 * invalidate_mapping_pages got it
1704 page_cache_release(page);
1708 shrink_readahead_size_eio(filp, ra);
1710 goto readpage_error;
1718 /* UHHUH! A synchronous read error occurred. Report it */
1719 page_cache_release(page);
1724 * Ok, it wasn't cached, so we need to create a new
1727 page = page_cache_alloc_cold(mapping);
1732 error = add_to_page_cache_lru(page, mapping, index,
1733 mapping_gfp_constraint(mapping, GFP_KERNEL));
1735 page_cache_release(page);
1736 if (error == -EEXIST) {
1746 ra->prev_pos = prev_index;
1747 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1748 ra->prev_pos |= prev_offset;
1750 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1751 file_accessed(filp);
1752 return written ? written : error;
1756 * generic_file_read_iter - generic filesystem read routine
1757 * @iocb: kernel I/O control block
1758 * @iter: destination for the data read
1760 * This is the "read_iter()" routine for all filesystems
1761 * that can use the page cache directly.
1764 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1766 struct file *file = iocb->ki_filp;
1768 loff_t *ppos = &iocb->ki_pos;
1771 if (iocb->ki_flags & IOCB_DIRECT) {
1772 struct address_space *mapping = file->f_mapping;
1773 struct inode *inode = mapping->host;
1774 size_t count = iov_iter_count(iter);
1778 goto out; /* skip atime */
1779 size = i_size_read(inode);
1780 retval = filemap_write_and_wait_range(mapping, pos,
1783 struct iov_iter data = *iter;
1784 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
1788 *ppos = pos + retval;
1789 iov_iter_advance(iter, retval);
1793 * Btrfs can have a short DIO read if we encounter
1794 * compressed extents, so if there was an error, or if
1795 * we've already read everything we wanted to, or if
1796 * there was a short read because we hit EOF, go ahead
1797 * and return. Otherwise fallthrough to buffered io for
1798 * the rest of the read. Buffered reads will not work for
1799 * DAX files, so don't bother trying.
1801 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1803 file_accessed(file);
1808 retval = do_generic_file_read(file, ppos, iter, retval);
1812 EXPORT_SYMBOL(generic_file_read_iter);
1816 * page_cache_read - adds requested page to the page cache if not already there
1817 * @file: file to read
1818 * @offset: page index
1820 * This adds the requested page to the page cache if it isn't already there,
1821 * and schedules an I/O to read in its contents from disk.
1823 static int page_cache_read(struct file *file, pgoff_t offset)
1825 struct address_space *mapping = file->f_mapping;
1830 page = page_cache_alloc_cold(mapping);
1834 ret = add_to_page_cache_lru(page, mapping, offset,
1835 mapping_gfp_constraint(mapping, GFP_KERNEL));
1837 ret = mapping->a_ops->readpage(file, page);
1838 else if (ret == -EEXIST)
1839 ret = 0; /* losing race to add is OK */
1841 page_cache_release(page);
1843 } while (ret == AOP_TRUNCATED_PAGE);
1848 #define MMAP_LOTSAMISS (100)
1851 * Synchronous readahead happens when we don't even find
1852 * a page in the page cache at all.
1854 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1855 struct file_ra_state *ra,
1859 struct address_space *mapping = file->f_mapping;
1861 /* If we don't want any read-ahead, don't bother */
1862 if (vma->vm_flags & VM_RAND_READ)
1867 if (vma->vm_flags & VM_SEQ_READ) {
1868 page_cache_sync_readahead(mapping, ra, file, offset,
1873 /* Avoid banging the cache line if not needed */
1874 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1878 * Do we miss much more than hit in this file? If so,
1879 * stop bothering with read-ahead. It will only hurt.
1881 if (ra->mmap_miss > MMAP_LOTSAMISS)
1887 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1888 ra->size = ra->ra_pages;
1889 ra->async_size = ra->ra_pages / 4;
1890 ra_submit(ra, mapping, file);
1894 * Asynchronous readahead happens when we find the page and PG_readahead,
1895 * so we want to possibly extend the readahead further..
1897 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1898 struct file_ra_state *ra,
1903 struct address_space *mapping = file->f_mapping;
1905 /* If we don't want any read-ahead, don't bother */
1906 if (vma->vm_flags & VM_RAND_READ)
1908 if (ra->mmap_miss > 0)
1910 if (PageReadahead(page))
1911 page_cache_async_readahead(mapping, ra, file,
1912 page, offset, ra->ra_pages);
1916 * filemap_fault - read in file data for page fault handling
1917 * @vma: vma in which the fault was taken
1918 * @vmf: struct vm_fault containing details of the fault
1920 * filemap_fault() is invoked via the vma operations vector for a
1921 * mapped memory region to read in file data during a page fault.
1923 * The goto's are kind of ugly, but this streamlines the normal case of having
1924 * it in the page cache, and handles the special cases reasonably without
1925 * having a lot of duplicated code.
1927 * vma->vm_mm->mmap_sem must be held on entry.
1929 * If our return value has VM_FAULT_RETRY set, it's because
1930 * lock_page_or_retry() returned 0.
1931 * The mmap_sem has usually been released in this case.
1932 * See __lock_page_or_retry() for the exception.
1934 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1935 * has not been released.
1937 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1939 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1942 struct file *file = vma->vm_file;
1943 struct address_space *mapping = file->f_mapping;
1944 struct file_ra_state *ra = &file->f_ra;
1945 struct inode *inode = mapping->host;
1946 pgoff_t offset = vmf->pgoff;
1951 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1952 if (offset >= size >> PAGE_CACHE_SHIFT)
1953 return VM_FAULT_SIGBUS;
1956 * Do we have something in the page cache already?
1958 page = find_get_page(mapping, offset);
1959 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1961 * We found the page, so try async readahead before
1962 * waiting for the lock.
1964 do_async_mmap_readahead(vma, ra, file, page, offset);
1966 /* No page in the page cache at all */
1967 do_sync_mmap_readahead(vma, ra, file, offset);
1968 count_vm_event(PGMAJFAULT);
1969 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1970 ret = VM_FAULT_MAJOR;
1972 page = find_get_page(mapping, offset);
1974 goto no_cached_page;
1977 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1978 page_cache_release(page);
1979 return ret | VM_FAULT_RETRY;
1982 /* Did it get truncated? */
1983 if (unlikely(page->mapping != mapping)) {
1988 VM_BUG_ON_PAGE(page->index != offset, page);
1991 * We have a locked page in the page cache, now we need to check
1992 * that it's up-to-date. If not, it is going to be due to an error.
1994 if (unlikely(!PageUptodate(page)))
1995 goto page_not_uptodate;
1998 * Found the page and have a reference on it.
1999 * We must recheck i_size under page lock.
2001 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2002 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
2004 page_cache_release(page);
2005 return VM_FAULT_SIGBUS;
2009 return ret | VM_FAULT_LOCKED;
2013 * We're only likely to ever get here if MADV_RANDOM is in
2016 error = page_cache_read(file, offset);
2019 * The page we want has now been added to the page cache.
2020 * In the unlikely event that someone removed it in the
2021 * meantime, we'll just come back here and read it again.
2027 * An error return from page_cache_read can result if the
2028 * system is low on memory, or a problem occurs while trying
2031 if (error == -ENOMEM)
2032 return VM_FAULT_OOM;
2033 return VM_FAULT_SIGBUS;
2037 * Umm, take care of errors if the page isn't up-to-date.
2038 * Try to re-read it _once_. We do this synchronously,
2039 * because there really aren't any performance issues here
2040 * and we need to check for errors.
2042 ClearPageError(page);
2043 error = mapping->a_ops->readpage(file, page);
2045 wait_on_page_locked(page);
2046 if (!PageUptodate(page))
2049 page_cache_release(page);
2051 if (!error || error == AOP_TRUNCATED_PAGE)
2054 /* Things didn't work out. Return zero to tell the mm layer so. */
2055 shrink_readahead_size_eio(file, ra);
2056 return VM_FAULT_SIGBUS;
2058 EXPORT_SYMBOL(filemap_fault);
2060 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2062 struct radix_tree_iter iter;
2064 struct file *file = vma->vm_file;
2065 struct address_space *mapping = file->f_mapping;
2068 unsigned long address = (unsigned long) vmf->virtual_address;
2073 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2074 if (iter.index > vmf->max_pgoff)
2077 page = radix_tree_deref_slot(slot);
2078 if (unlikely(!page))
2080 if (radix_tree_exception(page)) {
2081 if (radix_tree_deref_retry(page))
2087 if (!page_cache_get_speculative(page))
2090 /* Has the page moved? */
2091 if (unlikely(page != *slot)) {
2092 page_cache_release(page);
2096 if (!PageUptodate(page) ||
2097 PageReadahead(page) ||
2100 if (!trylock_page(page))
2103 if (page->mapping != mapping || !PageUptodate(page))
2106 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2107 if (page->index >= size >> PAGE_CACHE_SHIFT)
2110 pte = vmf->pte + page->index - vmf->pgoff;
2111 if (!pte_none(*pte))
2114 if (file->f_ra.mmap_miss > 0)
2115 file->f_ra.mmap_miss--;
2116 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2117 do_set_pte(vma, addr, page, pte, false, false);
2123 page_cache_release(page);
2125 if (iter.index == vmf->max_pgoff)
2130 EXPORT_SYMBOL(filemap_map_pages);
2132 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2134 struct page *page = vmf->page;
2135 struct inode *inode = file_inode(vma->vm_file);
2136 int ret = VM_FAULT_LOCKED;
2138 sb_start_pagefault(inode->i_sb);
2139 file_update_time(vma->vm_file);
2141 if (page->mapping != inode->i_mapping) {
2143 ret = VM_FAULT_NOPAGE;
2147 * We mark the page dirty already here so that when freeze is in
2148 * progress, we are guaranteed that writeback during freezing will
2149 * see the dirty page and writeprotect it again.
2151 set_page_dirty(page);
2152 wait_for_stable_page(page);
2154 sb_end_pagefault(inode->i_sb);
2157 EXPORT_SYMBOL(filemap_page_mkwrite);
2159 const struct vm_operations_struct generic_file_vm_ops = {
2160 .fault = filemap_fault,
2161 .map_pages = filemap_map_pages,
2162 .page_mkwrite = filemap_page_mkwrite,
2165 /* This is used for a general mmap of a disk file */
2167 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2169 struct address_space *mapping = file->f_mapping;
2171 if (!mapping->a_ops->readpage)
2173 file_accessed(file);
2174 vma->vm_ops = &generic_file_vm_ops;
2179 * This is for filesystems which do not implement ->writepage.
2181 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2183 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2185 return generic_file_mmap(file, vma);
2188 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2192 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2196 #endif /* CONFIG_MMU */
2198 EXPORT_SYMBOL(generic_file_mmap);
2199 EXPORT_SYMBOL(generic_file_readonly_mmap);
2201 static struct page *wait_on_page_read(struct page *page)
2203 if (!IS_ERR(page)) {
2204 wait_on_page_locked(page);
2205 if (!PageUptodate(page)) {
2206 page_cache_release(page);
2207 page = ERR_PTR(-EIO);
2213 static struct page *__read_cache_page(struct address_space *mapping,
2215 int (*filler)(void *, struct page *),
2222 page = find_get_page(mapping, index);
2224 page = __page_cache_alloc(gfp | __GFP_COLD);
2226 return ERR_PTR(-ENOMEM);
2227 err = add_to_page_cache_lru(page, mapping, index, gfp);
2228 if (unlikely(err)) {
2229 page_cache_release(page);
2232 /* Presumably ENOMEM for radix tree node */
2233 return ERR_PTR(err);
2235 err = filler(data, page);
2237 page_cache_release(page);
2238 page = ERR_PTR(err);
2240 page = wait_on_page_read(page);
2246 static struct page *do_read_cache_page(struct address_space *mapping,
2248 int (*filler)(void *, struct page *),
2257 page = __read_cache_page(mapping, index, filler, data, gfp);
2260 if (PageUptodate(page))
2264 if (!page->mapping) {
2266 page_cache_release(page);
2269 if (PageUptodate(page)) {
2273 err = filler(data, page);
2275 page_cache_release(page);
2276 return ERR_PTR(err);
2278 page = wait_on_page_read(page);
2283 mark_page_accessed(page);
2288 * read_cache_page - read into page cache, fill it if needed
2289 * @mapping: the page's address_space
2290 * @index: the page index
2291 * @filler: function to perform the read
2292 * @data: first arg to filler(data, page) function, often left as NULL
2294 * Read into the page cache. If a page already exists, and PageUptodate() is
2295 * not set, try to fill the page and wait for it to become unlocked.
2297 * If the page does not get brought uptodate, return -EIO.
2299 struct page *read_cache_page(struct address_space *mapping,
2301 int (*filler)(void *, struct page *),
2304 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2306 EXPORT_SYMBOL(read_cache_page);
2309 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2310 * @mapping: the page's address_space
2311 * @index: the page index
2312 * @gfp: the page allocator flags to use if allocating
2314 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2315 * any new page allocations done using the specified allocation flags.
2317 * If the page does not get brought uptodate, return -EIO.
2319 struct page *read_cache_page_gfp(struct address_space *mapping,
2323 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2325 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2327 EXPORT_SYMBOL(read_cache_page_gfp);
2330 * Performs necessary checks before doing a write
2332 * Can adjust writing position or amount of bytes to write.
2333 * Returns appropriate error code that caller should return or
2334 * zero in case that write should be allowed.
2336 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2338 struct file *file = iocb->ki_filp;
2339 struct inode *inode = file->f_mapping->host;
2340 unsigned long limit = rlimit(RLIMIT_FSIZE);
2343 if (!iov_iter_count(from))
2346 /* FIXME: this is for backwards compatibility with 2.4 */
2347 if (iocb->ki_flags & IOCB_APPEND)
2348 iocb->ki_pos = i_size_read(inode);
2352 if (limit != RLIM_INFINITY) {
2353 if (iocb->ki_pos >= limit) {
2354 send_sig(SIGXFSZ, current, 0);
2357 iov_iter_truncate(from, limit - (unsigned long)pos);
2363 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2364 !(file->f_flags & O_LARGEFILE))) {
2365 if (pos >= MAX_NON_LFS)
2367 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2371 * Are we about to exceed the fs block limit ?
2373 * If we have written data it becomes a short write. If we have
2374 * exceeded without writing data we send a signal and return EFBIG.
2375 * Linus frestrict idea will clean these up nicely..
2377 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2380 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2381 return iov_iter_count(from);
2383 EXPORT_SYMBOL(generic_write_checks);
2385 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2386 loff_t pos, unsigned len, unsigned flags,
2387 struct page **pagep, void **fsdata)
2389 const struct address_space_operations *aops = mapping->a_ops;
2391 return aops->write_begin(file, mapping, pos, len, flags,
2394 EXPORT_SYMBOL(pagecache_write_begin);
2396 int pagecache_write_end(struct file *file, struct address_space *mapping,
2397 loff_t pos, unsigned len, unsigned copied,
2398 struct page *page, void *fsdata)
2400 const struct address_space_operations *aops = mapping->a_ops;
2402 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2404 EXPORT_SYMBOL(pagecache_write_end);
2407 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2409 struct file *file = iocb->ki_filp;
2410 struct address_space *mapping = file->f_mapping;
2411 struct inode *inode = mapping->host;
2415 struct iov_iter data;
2417 write_len = iov_iter_count(from);
2418 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2420 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2425 * After a write we want buffered reads to be sure to go to disk to get
2426 * the new data. We invalidate clean cached page from the region we're
2427 * about to write. We do this *before* the write so that we can return
2428 * without clobbering -EIOCBQUEUED from ->direct_IO().
2430 if (mapping->nrpages) {
2431 written = invalidate_inode_pages2_range(mapping,
2432 pos >> PAGE_CACHE_SHIFT, end);
2434 * If a page can not be invalidated, return 0 to fall back
2435 * to buffered write.
2438 if (written == -EBUSY)
2445 written = mapping->a_ops->direct_IO(iocb, &data, pos);
2448 * Finally, try again to invalidate clean pages which might have been
2449 * cached by non-direct readahead, or faulted in by get_user_pages()
2450 * if the source of the write was an mmap'ed region of the file
2451 * we're writing. Either one is a pretty crazy thing to do,
2452 * so we don't support it 100%. If this invalidation
2453 * fails, tough, the write still worked...
2455 if (mapping->nrpages) {
2456 invalidate_inode_pages2_range(mapping,
2457 pos >> PAGE_CACHE_SHIFT, end);
2462 iov_iter_advance(from, written);
2463 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2464 i_size_write(inode, pos);
2465 mark_inode_dirty(inode);
2472 EXPORT_SYMBOL(generic_file_direct_write);
2475 * Find or create a page at the given pagecache position. Return the locked
2476 * page. This function is specifically for buffered writes.
2478 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2479 pgoff_t index, unsigned flags)
2482 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2484 if (flags & AOP_FLAG_NOFS)
2485 fgp_flags |= FGP_NOFS;
2487 page = pagecache_get_page(mapping, index, fgp_flags,
2488 mapping_gfp_mask(mapping));
2490 wait_for_stable_page(page);
2494 EXPORT_SYMBOL(grab_cache_page_write_begin);
2496 ssize_t generic_perform_write(struct file *file,
2497 struct iov_iter *i, loff_t pos)
2499 struct address_space *mapping = file->f_mapping;
2500 const struct address_space_operations *a_ops = mapping->a_ops;
2502 ssize_t written = 0;
2503 unsigned int flags = 0;
2506 * Copies from kernel address space cannot fail (NFSD is a big user).
2508 if (!iter_is_iovec(i))
2509 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2513 unsigned long offset; /* Offset into pagecache page */
2514 unsigned long bytes; /* Bytes to write to page */
2515 size_t copied; /* Bytes copied from user */
2518 offset = (pos & (PAGE_CACHE_SIZE - 1));
2519 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2524 * Bring in the user page that we will copy from _first_.
2525 * Otherwise there's a nasty deadlock on copying from the
2526 * same page as we're writing to, without it being marked
2529 * Not only is this an optimisation, but it is also required
2530 * to check that the address is actually valid, when atomic
2531 * usercopies are used, below.
2533 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2538 if (fatal_signal_pending(current)) {
2543 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2545 if (unlikely(status < 0))
2548 if (mapping_writably_mapped(mapping))
2549 flush_dcache_page(page);
2551 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2552 flush_dcache_page(page);
2554 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2556 if (unlikely(status < 0))
2562 iov_iter_advance(i, copied);
2563 if (unlikely(copied == 0)) {
2565 * If we were unable to copy any data at all, we must
2566 * fall back to a single segment length write.
2568 * If we didn't fallback here, we could livelock
2569 * because not all segments in the iov can be copied at
2570 * once without a pagefault.
2572 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2573 iov_iter_single_seg_count(i));
2579 balance_dirty_pages_ratelimited(mapping);
2580 } while (iov_iter_count(i));
2582 return written ? written : status;
2584 EXPORT_SYMBOL(generic_perform_write);
2587 * __generic_file_write_iter - write data to a file
2588 * @iocb: IO state structure (file, offset, etc.)
2589 * @from: iov_iter with data to write
2591 * This function does all the work needed for actually writing data to a
2592 * file. It does all basic checks, removes SUID from the file, updates
2593 * modification times and calls proper subroutines depending on whether we
2594 * do direct IO or a standard buffered write.
2596 * It expects i_mutex to be grabbed unless we work on a block device or similar
2597 * object which does not need locking at all.
2599 * This function does *not* take care of syncing data in case of O_SYNC write.
2600 * A caller has to handle it. This is mainly due to the fact that we want to
2601 * avoid syncing under i_mutex.
2603 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2605 struct file *file = iocb->ki_filp;
2606 struct address_space * mapping = file->f_mapping;
2607 struct inode *inode = mapping->host;
2608 ssize_t written = 0;
2612 /* We can write back this queue in page reclaim */
2613 current->backing_dev_info = inode_to_bdi(inode);
2614 err = file_remove_privs(file);
2618 err = file_update_time(file);
2622 if (iocb->ki_flags & IOCB_DIRECT) {
2623 loff_t pos, endbyte;
2625 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
2627 * If the write stopped short of completing, fall back to
2628 * buffered writes. Some filesystems do this for writes to
2629 * holes, for example. For DAX files, a buffered write will
2630 * not succeed (even if it did, DAX does not handle dirty
2631 * page-cache pages correctly).
2633 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2636 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2638 * If generic_perform_write() returned a synchronous error
2639 * then we want to return the number of bytes which were
2640 * direct-written, or the error code if that was zero. Note
2641 * that this differs from normal direct-io semantics, which
2642 * will return -EFOO even if some bytes were written.
2644 if (unlikely(status < 0)) {
2649 * We need to ensure that the page cache pages are written to
2650 * disk and invalidated to preserve the expected O_DIRECT
2653 endbyte = pos + status - 1;
2654 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2656 iocb->ki_pos = endbyte + 1;
2658 invalidate_mapping_pages(mapping,
2659 pos >> PAGE_CACHE_SHIFT,
2660 endbyte >> PAGE_CACHE_SHIFT);
2663 * We don't know how much we wrote, so just return
2664 * the number of bytes which were direct-written
2668 written = generic_perform_write(file, from, iocb->ki_pos);
2669 if (likely(written > 0))
2670 iocb->ki_pos += written;
2673 current->backing_dev_info = NULL;
2674 return written ? written : err;
2676 EXPORT_SYMBOL(__generic_file_write_iter);
2679 * generic_file_write_iter - write data to a file
2680 * @iocb: IO state structure
2681 * @from: iov_iter with data to write
2683 * This is a wrapper around __generic_file_write_iter() to be used by most
2684 * filesystems. It takes care of syncing the file in case of O_SYNC file
2685 * and acquires i_mutex as needed.
2687 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2689 struct file *file = iocb->ki_filp;
2690 struct inode *inode = file->f_mapping->host;
2693 mutex_lock(&inode->i_mutex);
2694 ret = generic_write_checks(iocb, from);
2696 ret = __generic_file_write_iter(iocb, from);
2697 mutex_unlock(&inode->i_mutex);
2702 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2708 EXPORT_SYMBOL(generic_file_write_iter);
2711 * try_to_release_page() - release old fs-specific metadata on a page
2713 * @page: the page which the kernel is trying to free
2714 * @gfp_mask: memory allocation flags (and I/O mode)
2716 * The address_space is to try to release any data against the page
2717 * (presumably at page->private). If the release was successful, return `1'.
2718 * Otherwise return zero.
2720 * This may also be called if PG_fscache is set on a page, indicating that the
2721 * page is known to the local caching routines.
2723 * The @gfp_mask argument specifies whether I/O may be performed to release
2724 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2727 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2729 struct address_space * const mapping = page->mapping;
2731 BUG_ON(!PageLocked(page));
2732 if (PageWriteback(page))
2735 if (mapping && mapping->a_ops->releasepage)
2736 return mapping->a_ops->releasepage(page, gfp_mask);
2737 return try_to_free_buffers(page);
2740 EXPORT_SYMBOL(try_to_release_page);