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/module.h>
13 #include <linux/slab.h>
14 #include <linux/compiler.h>
16 #include <linux/uaccess.h>
17 #include <linux/aio.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.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/syscalls.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/memcontrol.h>
39 * FIXME: remove all knowledge of the buffer layer from the core VM
41 #include <linux/buffer_head.h> /* for generic_osync_inode */
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
50 * Shared mappings now work. 15.8.1995 Bruno.
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
61 * ->i_mmap_lock (vmtruncate)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
67 * ->i_mmap_lock (truncate->unmap_mapping_range)
71 * ->page_table_lock or pte_lock (various, mainly in memory.c)
72 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
75 * ->lock_page (access_process_vm)
77 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
81 * ->i_alloc_sem (various)
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 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
106 * ->dcache_lock (proc_pid_lookup)
110 * Remove a page from the page cache and free it. Caller has to make
111 * sure the page is locked and that nobody else uses it - or that usage
112 * is safe. The caller must hold the mapping's tree_lock.
114 void __remove_from_page_cache(struct page *page)
116 struct address_space *mapping = page->mapping;
118 mem_cgroup_uncharge_cache_page(page);
119 radix_tree_delete(&mapping->page_tree, page->index);
120 page->mapping = NULL;
122 __dec_zone_page_state(page, NR_FILE_PAGES);
123 BUG_ON(page_mapped(page));
126 * Some filesystems seem to re-dirty the page even after
127 * the VM has canceled the dirty bit (eg ext3 journaling).
129 * Fix it up by doing a final dirty accounting check after
130 * having removed the page entirely.
132 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
133 dec_zone_page_state(page, NR_FILE_DIRTY);
134 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
138 void remove_from_page_cache(struct page *page)
140 struct address_space *mapping = page->mapping;
142 BUG_ON(!PageLocked(page));
144 spin_lock_irq(&mapping->tree_lock);
145 __remove_from_page_cache(page);
146 spin_unlock_irq(&mapping->tree_lock);
149 static int sync_page(void *word)
151 struct address_space *mapping;
154 page = container_of((unsigned long *)word, struct page, flags);
157 * page_mapping() is being called without PG_locked held.
158 * Some knowledge of the state and use of the page is used to
159 * reduce the requirements down to a memory barrier.
160 * The danger here is of a stale page_mapping() return value
161 * indicating a struct address_space different from the one it's
162 * associated with when it is associated with one.
163 * After smp_mb(), it's either the correct page_mapping() for
164 * the page, or an old page_mapping() and the page's own
165 * page_mapping() has gone NULL.
166 * The ->sync_page() address_space operation must tolerate
167 * page_mapping() going NULL. By an amazing coincidence,
168 * this comes about because none of the users of the page
169 * in the ->sync_page() methods make essential use of the
170 * page_mapping(), merely passing the page down to the backing
171 * device's unplug functions when it's non-NULL, which in turn
172 * ignore it for all cases but swap, where only page_private(page) is
173 * of interest. When page_mapping() does go NULL, the entire
174 * call stack gracefully ignores the page and returns.
178 mapping = page_mapping(page);
179 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
180 mapping->a_ops->sync_page(page);
185 static int sync_page_killable(void *word)
188 return fatal_signal_pending(current) ? -EINTR : 0;
192 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
193 * @mapping: address space structure to write
194 * @start: offset in bytes where the range starts
195 * @end: offset in bytes where the range ends (inclusive)
196 * @sync_mode: enable synchronous operation
198 * Start writeback against all of a mapping's dirty pages that lie
199 * within the byte offsets <start, end> inclusive.
201 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
202 * opposed to a regular memory cleansing writeback. The difference between
203 * these two operations is that if a dirty page/buffer is encountered, it must
204 * be waited upon, and not just skipped over.
206 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
207 loff_t end, int sync_mode)
210 struct writeback_control wbc = {
211 .sync_mode = sync_mode,
212 .nr_to_write = mapping->nrpages * 2,
213 .range_start = start,
217 if (!mapping_cap_writeback_dirty(mapping))
220 ret = do_writepages(mapping, &wbc);
224 static inline int __filemap_fdatawrite(struct address_space *mapping,
227 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
230 int filemap_fdatawrite(struct address_space *mapping)
232 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
234 EXPORT_SYMBOL(filemap_fdatawrite);
236 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
239 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
241 EXPORT_SYMBOL(filemap_fdatawrite_range);
244 * filemap_flush - mostly a non-blocking flush
245 * @mapping: target address_space
247 * This is a mostly non-blocking flush. Not suitable for data-integrity
248 * purposes - I/O may not be started against all dirty pages.
250 int filemap_flush(struct address_space *mapping)
252 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
254 EXPORT_SYMBOL(filemap_flush);
257 * wait_on_page_writeback_range - wait for writeback to complete
258 * @mapping: target address_space
259 * @start: beginning page index
260 * @end: ending page index
262 * Wait for writeback to complete against pages indexed by start->end
265 int wait_on_page_writeback_range(struct address_space *mapping,
266 pgoff_t start, pgoff_t end)
276 pagevec_init(&pvec, 0);
278 while ((index <= end) &&
279 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
280 PAGECACHE_TAG_WRITEBACK,
281 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
284 for (i = 0; i < nr_pages; i++) {
285 struct page *page = pvec.pages[i];
287 /* until radix tree lookup accepts end_index */
288 if (page->index > end)
291 wait_on_page_writeback(page);
295 pagevec_release(&pvec);
299 /* Check for outstanding write errors */
300 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
302 if (test_and_clear_bit(AS_EIO, &mapping->flags))
309 * sync_page_range - write and wait on all pages in the passed range
310 * @inode: target inode
311 * @mapping: target address_space
312 * @pos: beginning offset in pages to write
313 * @count: number of bytes to write
315 * Write and wait upon all the pages in the passed range. This is a "data
316 * integrity" operation. It waits upon in-flight writeout before starting and
317 * waiting upon new writeout. If there was an IO error, return it.
319 * We need to re-take i_mutex during the generic_osync_inode list walk because
320 * it is otherwise livelockable.
322 int sync_page_range(struct inode *inode, struct address_space *mapping,
323 loff_t pos, loff_t count)
325 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
326 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
329 if (!mapping_cap_writeback_dirty(mapping) || !count)
331 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
333 mutex_lock(&inode->i_mutex);
334 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
335 mutex_unlock(&inode->i_mutex);
338 ret = wait_on_page_writeback_range(mapping, start, end);
341 EXPORT_SYMBOL(sync_page_range);
344 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
345 * @inode: target inode
346 * @mapping: target address_space
347 * @pos: beginning offset in pages to write
348 * @count: number of bytes to write
350 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
351 * as it forces O_SYNC writers to different parts of the same file
352 * to be serialised right until io completion.
354 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
355 loff_t pos, loff_t count)
357 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
358 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
361 if (!mapping_cap_writeback_dirty(mapping) || !count)
363 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
365 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
367 ret = wait_on_page_writeback_range(mapping, start, end);
370 EXPORT_SYMBOL(sync_page_range_nolock);
373 * filemap_fdatawait - wait for all under-writeback pages to complete
374 * @mapping: address space structure to wait for
376 * Walk the list of under-writeback pages of the given address space
377 * and wait for all of them.
379 int filemap_fdatawait(struct address_space *mapping)
381 loff_t i_size = i_size_read(mapping->host);
386 return wait_on_page_writeback_range(mapping, 0,
387 (i_size - 1) >> PAGE_CACHE_SHIFT);
389 EXPORT_SYMBOL(filemap_fdatawait);
391 int filemap_write_and_wait(struct address_space *mapping)
395 if (mapping->nrpages) {
396 err = filemap_fdatawrite(mapping);
398 * Even if the above returned error, the pages may be
399 * written partially (e.g. -ENOSPC), so we wait for it.
400 * But the -EIO is special case, it may indicate the worst
401 * thing (e.g. bug) happened, so we avoid waiting for it.
404 int err2 = filemap_fdatawait(mapping);
411 EXPORT_SYMBOL(filemap_write_and_wait);
414 * filemap_write_and_wait_range - write out & wait on a file range
415 * @mapping: the address_space for the pages
416 * @lstart: offset in bytes where the range starts
417 * @lend: offset in bytes where the range ends (inclusive)
419 * Write out and wait upon file offsets lstart->lend, inclusive.
421 * Note that `lend' is inclusive (describes the last byte to be written) so
422 * that this function can be used to write to the very end-of-file (end = -1).
424 int filemap_write_and_wait_range(struct address_space *mapping,
425 loff_t lstart, loff_t lend)
429 if (mapping->nrpages) {
430 err = __filemap_fdatawrite_range(mapping, lstart, lend,
432 /* See comment of filemap_write_and_wait() */
434 int err2 = wait_on_page_writeback_range(mapping,
435 lstart >> PAGE_CACHE_SHIFT,
436 lend >> PAGE_CACHE_SHIFT);
445 * add_to_page_cache_locked - add a locked page to the pagecache
447 * @mapping: the page's address_space
448 * @offset: page index
449 * @gfp_mask: page allocation mode
451 * This function is used to add a page to the pagecache. It must be locked.
452 * This function does not add the page to the LRU. The caller must do that.
454 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
455 pgoff_t offset, gfp_t gfp_mask)
459 VM_BUG_ON(!PageLocked(page));
461 error = mem_cgroup_cache_charge(page, current->mm,
462 gfp_mask & ~__GFP_HIGHMEM);
466 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
468 page_cache_get(page);
469 page->mapping = mapping;
470 page->index = offset;
472 spin_lock_irq(&mapping->tree_lock);
473 error = radix_tree_insert(&mapping->page_tree, offset, page);
474 if (likely(!error)) {
476 __inc_zone_page_state(page, NR_FILE_PAGES);
478 page->mapping = NULL;
479 mem_cgroup_uncharge_cache_page(page);
480 page_cache_release(page);
483 spin_unlock_irq(&mapping->tree_lock);
484 radix_tree_preload_end();
486 mem_cgroup_uncharge_cache_page(page);
490 EXPORT_SYMBOL(add_to_page_cache_locked);
492 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
493 pgoff_t offset, gfp_t gfp_mask)
495 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
502 struct page *__page_cache_alloc(gfp_t gfp)
504 if (cpuset_do_page_mem_spread()) {
505 int n = cpuset_mem_spread_node();
506 return alloc_pages_node(n, gfp, 0);
508 return alloc_pages(gfp, 0);
510 EXPORT_SYMBOL(__page_cache_alloc);
513 static int __sleep_on_page_lock(void *word)
520 * In order to wait for pages to become available there must be
521 * waitqueues associated with pages. By using a hash table of
522 * waitqueues where the bucket discipline is to maintain all
523 * waiters on the same queue and wake all when any of the pages
524 * become available, and for the woken contexts to check to be
525 * sure the appropriate page became available, this saves space
526 * at a cost of "thundering herd" phenomena during rare hash
529 static wait_queue_head_t *page_waitqueue(struct page *page)
531 const struct zone *zone = page_zone(page);
533 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
536 static inline void wake_up_page(struct page *page, int bit)
538 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
541 void wait_on_page_bit(struct page *page, int bit_nr)
543 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
545 if (test_bit(bit_nr, &page->flags))
546 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
547 TASK_UNINTERRUPTIBLE);
549 EXPORT_SYMBOL(wait_on_page_bit);
552 * unlock_page - unlock a locked page
555 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
556 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
557 * mechananism between PageLocked pages and PageWriteback pages is shared.
558 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
560 * The first mb is necessary to safely close the critical section opened by the
561 * test_and_set_bit() to lock the page; the second mb is necessary to enforce
562 * ordering between the clear_bit and the read of the waitqueue (to avoid SMP
563 * races with a parallel wait_on_page_locked()).
565 void unlock_page(struct page *page)
567 smp_mb__before_clear_bit();
568 if (!test_and_clear_bit(PG_locked, &page->flags))
570 smp_mb__after_clear_bit();
571 wake_up_page(page, PG_locked);
573 EXPORT_SYMBOL(unlock_page);
576 * end_page_writeback - end writeback against a page
579 void end_page_writeback(struct page *page)
581 if (TestClearPageReclaim(page))
582 rotate_reclaimable_page(page);
584 if (!test_clear_page_writeback(page))
587 smp_mb__after_clear_bit();
588 wake_up_page(page, PG_writeback);
590 EXPORT_SYMBOL(end_page_writeback);
593 * __lock_page - get a lock on the page, assuming we need to sleep to get it
594 * @page: the page to lock
596 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
597 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
598 * chances are that on the second loop, the block layer's plug list is empty,
599 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
601 void __lock_page(struct page *page)
603 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
605 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
606 TASK_UNINTERRUPTIBLE);
608 EXPORT_SYMBOL(__lock_page);
610 int __lock_page_killable(struct page *page)
612 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
614 return __wait_on_bit_lock(page_waitqueue(page), &wait,
615 sync_page_killable, TASK_KILLABLE);
619 * __lock_page_nosync - get a lock on the page, without calling sync_page()
620 * @page: the page to lock
622 * Variant of lock_page that does not require the caller to hold a reference
623 * on the page's mapping.
625 void __lock_page_nosync(struct page *page)
627 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
628 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
629 TASK_UNINTERRUPTIBLE);
633 * find_get_page - find and get a page reference
634 * @mapping: the address_space to search
635 * @offset: the page index
637 * Is there a pagecache struct page at the given (mapping, offset) tuple?
638 * If yes, increment its refcount and return it; if no, return NULL.
640 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
648 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
650 page = radix_tree_deref_slot(pagep);
651 if (unlikely(!page || page == RADIX_TREE_RETRY))
654 if (!page_cache_get_speculative(page))
658 * Has the page moved?
659 * This is part of the lockless pagecache protocol. See
660 * include/linux/pagemap.h for details.
662 if (unlikely(page != *pagep)) {
663 page_cache_release(page);
671 EXPORT_SYMBOL(find_get_page);
674 * find_lock_page - locate, pin and lock a pagecache page
675 * @mapping: the address_space to search
676 * @offset: the page index
678 * Locates the desired pagecache page, locks it, increments its reference
679 * count and returns its address.
681 * Returns zero if the page was not present. find_lock_page() may sleep.
683 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
688 page = find_get_page(mapping, offset);
691 /* Has the page been truncated? */
692 if (unlikely(page->mapping != mapping)) {
694 page_cache_release(page);
697 VM_BUG_ON(page->index != offset);
701 EXPORT_SYMBOL(find_lock_page);
704 * find_or_create_page - locate or add a pagecache page
705 * @mapping: the page's address_space
706 * @index: the page's index into the mapping
707 * @gfp_mask: page allocation mode
709 * Locates a page in the pagecache. If the page is not present, a new page
710 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
711 * LRU list. The returned page is locked and has its reference count
714 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
717 * find_or_create_page() returns the desired page's address, or zero on
720 struct page *find_or_create_page(struct address_space *mapping,
721 pgoff_t index, gfp_t gfp_mask)
726 page = find_lock_page(mapping, index);
728 page = __page_cache_alloc(gfp_mask);
731 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
733 page_cache_release(page);
741 EXPORT_SYMBOL(find_or_create_page);
744 * find_get_pages - gang pagecache lookup
745 * @mapping: The address_space to search
746 * @start: The starting page index
747 * @nr_pages: The maximum number of pages
748 * @pages: Where the resulting pages are placed
750 * find_get_pages() will search for and return a group of up to
751 * @nr_pages pages in the mapping. The pages are placed at @pages.
752 * find_get_pages() takes a reference against the returned pages.
754 * The search returns a group of mapping-contiguous pages with ascending
755 * indexes. There may be holes in the indices due to not-present pages.
757 * find_get_pages() returns the number of pages which were found.
759 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
760 unsigned int nr_pages, struct page **pages)
764 unsigned int nr_found;
768 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
769 (void ***)pages, start, nr_pages);
771 for (i = 0; i < nr_found; i++) {
774 page = radix_tree_deref_slot((void **)pages[i]);
778 * this can only trigger if nr_found == 1, making livelock
781 if (unlikely(page == RADIX_TREE_RETRY))
784 if (!page_cache_get_speculative(page))
787 /* Has the page moved? */
788 if (unlikely(page != *((void **)pages[i]))) {
789 page_cache_release(page);
801 * find_get_pages_contig - gang contiguous pagecache lookup
802 * @mapping: The address_space to search
803 * @index: The starting page index
804 * @nr_pages: The maximum number of pages
805 * @pages: Where the resulting pages are placed
807 * find_get_pages_contig() works exactly like find_get_pages(), except
808 * that the returned number of pages are guaranteed to be contiguous.
810 * find_get_pages_contig() returns the number of pages which were found.
812 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
813 unsigned int nr_pages, struct page **pages)
817 unsigned int nr_found;
821 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
822 (void ***)pages, index, nr_pages);
824 for (i = 0; i < nr_found; i++) {
827 page = radix_tree_deref_slot((void **)pages[i]);
831 * this can only trigger if nr_found == 1, making livelock
834 if (unlikely(page == RADIX_TREE_RETRY))
837 if (page->mapping == NULL || page->index != index)
840 if (!page_cache_get_speculative(page))
843 /* Has the page moved? */
844 if (unlikely(page != *((void **)pages[i]))) {
845 page_cache_release(page);
856 EXPORT_SYMBOL(find_get_pages_contig);
859 * find_get_pages_tag - find and return pages that match @tag
860 * @mapping: the address_space to search
861 * @index: the starting page index
862 * @tag: the tag index
863 * @nr_pages: the maximum number of pages
864 * @pages: where the resulting pages are placed
866 * Like find_get_pages, except we only return pages which are tagged with
867 * @tag. We update @index to index the next page for the traversal.
869 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
870 int tag, unsigned int nr_pages, struct page **pages)
874 unsigned int nr_found;
878 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
879 (void ***)pages, *index, nr_pages, tag);
881 for (i = 0; i < nr_found; i++) {
884 page = radix_tree_deref_slot((void **)pages[i]);
888 * this can only trigger if nr_found == 1, making livelock
891 if (unlikely(page == RADIX_TREE_RETRY))
894 if (!page_cache_get_speculative(page))
897 /* Has the page moved? */
898 if (unlikely(page != *((void **)pages[i]))) {
899 page_cache_release(page);
909 *index = pages[ret - 1]->index + 1;
913 EXPORT_SYMBOL(find_get_pages_tag);
916 * grab_cache_page_nowait - returns locked page at given index in given cache
917 * @mapping: target address_space
918 * @index: the page index
920 * Same as grab_cache_page(), but do not wait if the page is unavailable.
921 * This is intended for speculative data generators, where the data can
922 * be regenerated if the page couldn't be grabbed. This routine should
923 * be safe to call while holding the lock for another page.
925 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
926 * and deadlock against the caller's locked page.
929 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
931 struct page *page = find_get_page(mapping, index);
934 if (trylock_page(page))
936 page_cache_release(page);
939 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
940 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
941 page_cache_release(page);
946 EXPORT_SYMBOL(grab_cache_page_nowait);
949 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
950 * a _large_ part of the i/o request. Imagine the worst scenario:
952 * ---R__________________________________________B__________
953 * ^ reading here ^ bad block(assume 4k)
955 * read(R) => miss => readahead(R...B) => media error => frustrating retries
956 * => failing the whole request => read(R) => read(R+1) =>
957 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
958 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
959 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
961 * It is going insane. Fix it by quickly scaling down the readahead size.
963 static void shrink_readahead_size_eio(struct file *filp,
964 struct file_ra_state *ra)
973 * do_generic_file_read - generic file read routine
974 * @filp: the file to read
975 * @ppos: current file position
976 * @desc: read_descriptor
977 * @actor: read method
979 * This is a generic file read routine, and uses the
980 * mapping->a_ops->readpage() function for the actual low-level stuff.
982 * This is really ugly. But the goto's actually try to clarify some
983 * of the logic when it comes to error handling etc.
985 static void do_generic_file_read(struct file *filp, loff_t *ppos,
986 read_descriptor_t *desc, read_actor_t actor)
988 struct address_space *mapping = filp->f_mapping;
989 struct inode *inode = mapping->host;
990 struct file_ra_state *ra = &filp->f_ra;
994 unsigned long offset; /* offset into pagecache page */
995 unsigned int prev_offset;
998 index = *ppos >> PAGE_CACHE_SHIFT;
999 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1000 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1001 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1002 offset = *ppos & ~PAGE_CACHE_MASK;
1008 unsigned long nr, ret;
1012 page = find_get_page(mapping, index);
1014 page_cache_sync_readahead(mapping,
1016 index, last_index - index);
1017 page = find_get_page(mapping, index);
1018 if (unlikely(page == NULL))
1019 goto no_cached_page;
1021 if (PageReadahead(page)) {
1022 page_cache_async_readahead(mapping,
1024 index, last_index - index);
1026 if (!PageUptodate(page)) {
1027 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1028 !mapping->a_ops->is_partially_uptodate)
1029 goto page_not_up_to_date;
1030 if (!trylock_page(page))
1031 goto page_not_up_to_date;
1032 if (!mapping->a_ops->is_partially_uptodate(page,
1034 goto page_not_up_to_date_locked;
1039 * i_size must be checked after we know the page is Uptodate.
1041 * Checking i_size after the check allows us to calculate
1042 * the correct value for "nr", which means the zero-filled
1043 * part of the page is not copied back to userspace (unless
1044 * another truncate extends the file - this is desired though).
1047 isize = i_size_read(inode);
1048 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1049 if (unlikely(!isize || index > end_index)) {
1050 page_cache_release(page);
1054 /* nr is the maximum number of bytes to copy from this page */
1055 nr = PAGE_CACHE_SIZE;
1056 if (index == end_index) {
1057 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1059 page_cache_release(page);
1065 /* If users can be writing to this page using arbitrary
1066 * virtual addresses, take care about potential aliasing
1067 * before reading the page on the kernel side.
1069 if (mapping_writably_mapped(mapping))
1070 flush_dcache_page(page);
1073 * When a sequential read accesses a page several times,
1074 * only mark it as accessed the first time.
1076 if (prev_index != index || offset != prev_offset)
1077 mark_page_accessed(page);
1081 * Ok, we have the page, and it's up-to-date, so
1082 * now we can copy it to user space...
1084 * The actor routine returns how many bytes were actually used..
1085 * NOTE! This may not be the same as how much of a user buffer
1086 * we filled up (we may be padding etc), so we can only update
1087 * "pos" here (the actor routine has to update the user buffer
1088 * pointers and the remaining count).
1090 ret = actor(desc, page, offset, nr);
1092 index += offset >> PAGE_CACHE_SHIFT;
1093 offset &= ~PAGE_CACHE_MASK;
1094 prev_offset = offset;
1096 page_cache_release(page);
1097 if (ret == nr && desc->count)
1101 page_not_up_to_date:
1102 /* Get exclusive access to the page ... */
1103 if (lock_page_killable(page))
1106 page_not_up_to_date_locked:
1107 /* Did it get truncated before we got the lock? */
1108 if (!page->mapping) {
1110 page_cache_release(page);
1114 /* Did somebody else fill it already? */
1115 if (PageUptodate(page)) {
1121 /* Start the actual read. The read will unlock the page. */
1122 error = mapping->a_ops->readpage(filp, page);
1124 if (unlikely(error)) {
1125 if (error == AOP_TRUNCATED_PAGE) {
1126 page_cache_release(page);
1129 goto readpage_error;
1132 if (!PageUptodate(page)) {
1133 if (lock_page_killable(page))
1135 if (!PageUptodate(page)) {
1136 if (page->mapping == NULL) {
1138 * invalidate_inode_pages got it
1141 page_cache_release(page);
1145 shrink_readahead_size_eio(filp, ra);
1156 /* UHHUH! A synchronous read error occurred. Report it */
1157 desc->error = error;
1158 page_cache_release(page);
1163 * Ok, it wasn't cached, so we need to create a new
1166 page = page_cache_alloc_cold(mapping);
1168 desc->error = -ENOMEM;
1171 error = add_to_page_cache_lru(page, mapping,
1174 page_cache_release(page);
1175 if (error == -EEXIST)
1177 desc->error = error;
1184 ra->prev_pos = prev_index;
1185 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1186 ra->prev_pos |= prev_offset;
1188 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1190 file_accessed(filp);
1193 int file_read_actor(read_descriptor_t *desc, struct page *page,
1194 unsigned long offset, unsigned long size)
1197 unsigned long left, count = desc->count;
1203 * Faults on the destination of a read are common, so do it before
1206 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1207 kaddr = kmap_atomic(page, KM_USER0);
1208 left = __copy_to_user_inatomic(desc->arg.buf,
1209 kaddr + offset, size);
1210 kunmap_atomic(kaddr, KM_USER0);
1215 /* Do it the slow way */
1217 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1222 desc->error = -EFAULT;
1225 desc->count = count - size;
1226 desc->written += size;
1227 desc->arg.buf += size;
1232 * Performs necessary checks before doing a write
1233 * @iov: io vector request
1234 * @nr_segs: number of segments in the iovec
1235 * @count: number of bytes to write
1236 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1238 * Adjust number of segments and amount of bytes to write (nr_segs should be
1239 * properly initialized first). Returns appropriate error code that caller
1240 * should return or zero in case that write should be allowed.
1242 int generic_segment_checks(const struct iovec *iov,
1243 unsigned long *nr_segs, size_t *count, int access_flags)
1247 for (seg = 0; seg < *nr_segs; seg++) {
1248 const struct iovec *iv = &iov[seg];
1251 * If any segment has a negative length, or the cumulative
1252 * length ever wraps negative then return -EINVAL.
1255 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1257 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1262 cnt -= iv->iov_len; /* This segment is no good */
1268 EXPORT_SYMBOL(generic_segment_checks);
1271 * generic_file_aio_read - generic filesystem read routine
1272 * @iocb: kernel I/O control block
1273 * @iov: io vector request
1274 * @nr_segs: number of segments in the iovec
1275 * @pos: current file position
1277 * This is the "read()" routine for all filesystems
1278 * that can use the page cache directly.
1281 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1282 unsigned long nr_segs, loff_t pos)
1284 struct file *filp = iocb->ki_filp;
1288 loff_t *ppos = &iocb->ki_pos;
1291 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1295 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1296 if (filp->f_flags & O_DIRECT) {
1298 struct address_space *mapping;
1299 struct inode *inode;
1301 mapping = filp->f_mapping;
1302 inode = mapping->host;
1304 goto out; /* skip atime */
1305 size = i_size_read(inode);
1307 retval = filemap_write_and_wait(mapping);
1309 retval = mapping->a_ops->direct_IO(READ, iocb,
1313 *ppos = pos + retval;
1315 file_accessed(filp);
1321 for (seg = 0; seg < nr_segs; seg++) {
1322 read_descriptor_t desc;
1325 desc.arg.buf = iov[seg].iov_base;
1326 desc.count = iov[seg].iov_len;
1327 if (desc.count == 0)
1330 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1331 retval += desc.written;
1333 retval = retval ?: desc.error;
1342 EXPORT_SYMBOL(generic_file_aio_read);
1345 do_readahead(struct address_space *mapping, struct file *filp,
1346 pgoff_t index, unsigned long nr)
1348 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1351 force_page_cache_readahead(mapping, filp, index,
1352 max_sane_readahead(nr));
1356 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1364 if (file->f_mode & FMODE_READ) {
1365 struct address_space *mapping = file->f_mapping;
1366 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1367 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1368 unsigned long len = end - start + 1;
1369 ret = do_readahead(mapping, file, start, len);
1375 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1376 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1378 return SYSC_readahead((int) fd, offset, (size_t) count);
1380 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1385 * page_cache_read - adds requested page to the page cache if not already there
1386 * @file: file to read
1387 * @offset: page index
1389 * This adds the requested page to the page cache if it isn't already there,
1390 * and schedules an I/O to read in its contents from disk.
1392 static int page_cache_read(struct file *file, pgoff_t offset)
1394 struct address_space *mapping = file->f_mapping;
1399 page = page_cache_alloc_cold(mapping);
1403 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1405 ret = mapping->a_ops->readpage(file, page);
1406 else if (ret == -EEXIST)
1407 ret = 0; /* losing race to add is OK */
1409 page_cache_release(page);
1411 } while (ret == AOP_TRUNCATED_PAGE);
1416 #define MMAP_LOTSAMISS (100)
1419 * filemap_fault - read in file data for page fault handling
1420 * @vma: vma in which the fault was taken
1421 * @vmf: struct vm_fault containing details of the fault
1423 * filemap_fault() is invoked via the vma operations vector for a
1424 * mapped memory region to read in file data during a page fault.
1426 * The goto's are kind of ugly, but this streamlines the normal case of having
1427 * it in the page cache, and handles the special cases reasonably without
1428 * having a lot of duplicated code.
1430 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1433 struct file *file = vma->vm_file;
1434 struct address_space *mapping = file->f_mapping;
1435 struct file_ra_state *ra = &file->f_ra;
1436 struct inode *inode = mapping->host;
1439 int did_readaround = 0;
1442 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1443 if (vmf->pgoff >= size)
1444 return VM_FAULT_SIGBUS;
1446 /* If we don't want any read-ahead, don't bother */
1447 if (VM_RandomReadHint(vma))
1448 goto no_cached_page;
1451 * Do we have something in the page cache already?
1454 page = find_lock_page(mapping, vmf->pgoff);
1456 * For sequential accesses, we use the generic readahead logic.
1458 if (VM_SequentialReadHint(vma)) {
1460 page_cache_sync_readahead(mapping, ra, file,
1462 page = find_lock_page(mapping, vmf->pgoff);
1464 goto no_cached_page;
1466 if (PageReadahead(page)) {
1467 page_cache_async_readahead(mapping, ra, file, page,
1473 unsigned long ra_pages;
1478 * Do we miss much more than hit in this file? If so,
1479 * stop bothering with read-ahead. It will only hurt.
1481 if (ra->mmap_miss > MMAP_LOTSAMISS)
1482 goto no_cached_page;
1485 * To keep the pgmajfault counter straight, we need to
1486 * check did_readaround, as this is an inner loop.
1488 if (!did_readaround) {
1489 ret = VM_FAULT_MAJOR;
1490 count_vm_event(PGMAJFAULT);
1493 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1497 if (vmf->pgoff > ra_pages / 2)
1498 start = vmf->pgoff - ra_pages / 2;
1499 do_page_cache_readahead(mapping, file, start, ra_pages);
1501 page = find_lock_page(mapping, vmf->pgoff);
1503 goto no_cached_page;
1506 if (!did_readaround)
1510 * We have a locked page in the page cache, now we need to check
1511 * that it's up-to-date. If not, it is going to be due to an error.
1513 if (unlikely(!PageUptodate(page)))
1514 goto page_not_uptodate;
1516 /* Must recheck i_size under page lock */
1517 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1518 if (unlikely(vmf->pgoff >= size)) {
1520 page_cache_release(page);
1521 return VM_FAULT_SIGBUS;
1525 * Found the page and have a reference on it.
1527 mark_page_accessed(page);
1528 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
1530 return ret | VM_FAULT_LOCKED;
1534 * We're only likely to ever get here if MADV_RANDOM is in
1537 error = page_cache_read(file, vmf->pgoff);
1540 * The page we want has now been added to the page cache.
1541 * In the unlikely event that someone removed it in the
1542 * meantime, we'll just come back here and read it again.
1548 * An error return from page_cache_read can result if the
1549 * system is low on memory, or a problem occurs while trying
1552 if (error == -ENOMEM)
1553 return VM_FAULT_OOM;
1554 return VM_FAULT_SIGBUS;
1558 if (!did_readaround) {
1559 ret = VM_FAULT_MAJOR;
1560 count_vm_event(PGMAJFAULT);
1564 * Umm, take care of errors if the page isn't up-to-date.
1565 * Try to re-read it _once_. We do this synchronously,
1566 * because there really aren't any performance issues here
1567 * and we need to check for errors.
1569 ClearPageError(page);
1570 error = mapping->a_ops->readpage(file, page);
1572 wait_on_page_locked(page);
1573 if (!PageUptodate(page))
1576 page_cache_release(page);
1578 if (!error || error == AOP_TRUNCATED_PAGE)
1581 /* Things didn't work out. Return zero to tell the mm layer so. */
1582 shrink_readahead_size_eio(file, ra);
1583 return VM_FAULT_SIGBUS;
1585 EXPORT_SYMBOL(filemap_fault);
1587 struct vm_operations_struct generic_file_vm_ops = {
1588 .fault = filemap_fault,
1591 /* This is used for a general mmap of a disk file */
1593 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1595 struct address_space *mapping = file->f_mapping;
1597 if (!mapping->a_ops->readpage)
1599 file_accessed(file);
1600 vma->vm_ops = &generic_file_vm_ops;
1601 vma->vm_flags |= VM_CAN_NONLINEAR;
1606 * This is for filesystems which do not implement ->writepage.
1608 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1610 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1612 return generic_file_mmap(file, vma);
1615 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1619 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1623 #endif /* CONFIG_MMU */
1625 EXPORT_SYMBOL(generic_file_mmap);
1626 EXPORT_SYMBOL(generic_file_readonly_mmap);
1628 static struct page *__read_cache_page(struct address_space *mapping,
1630 int (*filler)(void *,struct page*),
1636 page = find_get_page(mapping, index);
1638 page = page_cache_alloc_cold(mapping);
1640 return ERR_PTR(-ENOMEM);
1641 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1642 if (unlikely(err)) {
1643 page_cache_release(page);
1646 /* Presumably ENOMEM for radix tree node */
1647 return ERR_PTR(err);
1649 err = filler(data, page);
1651 page_cache_release(page);
1652 page = ERR_PTR(err);
1659 * read_cache_page_async - read into page cache, fill it if needed
1660 * @mapping: the page's address_space
1661 * @index: the page index
1662 * @filler: function to perform the read
1663 * @data: destination for read data
1665 * Same as read_cache_page, but don't wait for page to become unlocked
1666 * after submitting it to the filler.
1668 * Read into the page cache. If a page already exists, and PageUptodate() is
1669 * not set, try to fill the page but don't wait for it to become unlocked.
1671 * If the page does not get brought uptodate, return -EIO.
1673 struct page *read_cache_page_async(struct address_space *mapping,
1675 int (*filler)(void *,struct page*),
1682 page = __read_cache_page(mapping, index, filler, data);
1685 if (PageUptodate(page))
1689 if (!page->mapping) {
1691 page_cache_release(page);
1694 if (PageUptodate(page)) {
1698 err = filler(data, page);
1700 page_cache_release(page);
1701 return ERR_PTR(err);
1704 mark_page_accessed(page);
1707 EXPORT_SYMBOL(read_cache_page_async);
1710 * read_cache_page - read into page cache, fill it if needed
1711 * @mapping: the page's address_space
1712 * @index: the page index
1713 * @filler: function to perform the read
1714 * @data: destination for read data
1716 * Read into the page cache. If a page already exists, and PageUptodate() is
1717 * not set, try to fill the page then wait for it to become unlocked.
1719 * If the page does not get brought uptodate, return -EIO.
1721 struct page *read_cache_page(struct address_space *mapping,
1723 int (*filler)(void *,struct page*),
1728 page = read_cache_page_async(mapping, index, filler, data);
1731 wait_on_page_locked(page);
1732 if (!PageUptodate(page)) {
1733 page_cache_release(page);
1734 page = ERR_PTR(-EIO);
1739 EXPORT_SYMBOL(read_cache_page);
1742 * The logic we want is
1744 * if suid or (sgid and xgrp)
1747 int should_remove_suid(struct dentry *dentry)
1749 mode_t mode = dentry->d_inode->i_mode;
1752 /* suid always must be killed */
1753 if (unlikely(mode & S_ISUID))
1754 kill = ATTR_KILL_SUID;
1757 * sgid without any exec bits is just a mandatory locking mark; leave
1758 * it alone. If some exec bits are set, it's a real sgid; kill it.
1760 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1761 kill |= ATTR_KILL_SGID;
1763 if (unlikely(kill && !capable(CAP_FSETID)))
1768 EXPORT_SYMBOL(should_remove_suid);
1770 static int __remove_suid(struct dentry *dentry, int kill)
1772 struct iattr newattrs;
1774 newattrs.ia_valid = ATTR_FORCE | kill;
1775 return notify_change(dentry, &newattrs);
1778 int file_remove_suid(struct file *file)
1780 struct dentry *dentry = file->f_path.dentry;
1781 int killsuid = should_remove_suid(dentry);
1782 int killpriv = security_inode_need_killpriv(dentry);
1788 error = security_inode_killpriv(dentry);
1789 if (!error && killsuid)
1790 error = __remove_suid(dentry, killsuid);
1794 EXPORT_SYMBOL(file_remove_suid);
1796 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1797 const struct iovec *iov, size_t base, size_t bytes)
1799 size_t copied = 0, left = 0;
1802 char __user *buf = iov->iov_base + base;
1803 int copy = min(bytes, iov->iov_len - base);
1806 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1815 return copied - left;
1819 * Copy as much as we can into the page and return the number of bytes which
1820 * were sucessfully copied. If a fault is encountered then return the number of
1821 * bytes which were copied.
1823 size_t iov_iter_copy_from_user_atomic(struct page *page,
1824 struct iov_iter *i, unsigned long offset, size_t bytes)
1829 BUG_ON(!in_atomic());
1830 kaddr = kmap_atomic(page, KM_USER0);
1831 if (likely(i->nr_segs == 1)) {
1833 char __user *buf = i->iov->iov_base + i->iov_offset;
1834 left = __copy_from_user_inatomic_nocache(kaddr + offset,
1836 copied = bytes - left;
1838 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1839 i->iov, i->iov_offset, bytes);
1841 kunmap_atomic(kaddr, KM_USER0);
1845 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1848 * This has the same sideeffects and return value as
1849 * iov_iter_copy_from_user_atomic().
1850 * The difference is that it attempts to resolve faults.
1851 * Page must not be locked.
1853 size_t iov_iter_copy_from_user(struct page *page,
1854 struct iov_iter *i, unsigned long offset, size_t bytes)
1860 if (likely(i->nr_segs == 1)) {
1862 char __user *buf = i->iov->iov_base + i->iov_offset;
1863 left = __copy_from_user_nocache(kaddr + offset, buf, bytes);
1864 copied = bytes - left;
1866 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1867 i->iov, i->iov_offset, bytes);
1872 EXPORT_SYMBOL(iov_iter_copy_from_user);
1874 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1876 BUG_ON(i->count < bytes);
1878 if (likely(i->nr_segs == 1)) {
1879 i->iov_offset += bytes;
1882 const struct iovec *iov = i->iov;
1883 size_t base = i->iov_offset;
1886 * The !iov->iov_len check ensures we skip over unlikely
1887 * zero-length segments (without overruning the iovec).
1889 while (bytes || unlikely(i->count && !iov->iov_len)) {
1892 copy = min(bytes, iov->iov_len - base);
1893 BUG_ON(!i->count || i->count < copy);
1897 if (iov->iov_len == base) {
1903 i->iov_offset = base;
1906 EXPORT_SYMBOL(iov_iter_advance);
1909 * Fault in the first iovec of the given iov_iter, to a maximum length
1910 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1911 * accessed (ie. because it is an invalid address).
1913 * writev-intensive code may want this to prefault several iovecs -- that
1914 * would be possible (callers must not rely on the fact that _only_ the
1915 * first iovec will be faulted with the current implementation).
1917 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
1919 char __user *buf = i->iov->iov_base + i->iov_offset;
1920 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1921 return fault_in_pages_readable(buf, bytes);
1923 EXPORT_SYMBOL(iov_iter_fault_in_readable);
1926 * Return the count of just the current iov_iter segment.
1928 size_t iov_iter_single_seg_count(struct iov_iter *i)
1930 const struct iovec *iov = i->iov;
1931 if (i->nr_segs == 1)
1934 return min(i->count, iov->iov_len - i->iov_offset);
1936 EXPORT_SYMBOL(iov_iter_single_seg_count);
1939 * Performs necessary checks before doing a write
1941 * Can adjust writing position or amount of bytes to write.
1942 * Returns appropriate error code that caller should return or
1943 * zero in case that write should be allowed.
1945 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1947 struct inode *inode = file->f_mapping->host;
1948 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1950 if (unlikely(*pos < 0))
1954 /* FIXME: this is for backwards compatibility with 2.4 */
1955 if (file->f_flags & O_APPEND)
1956 *pos = i_size_read(inode);
1958 if (limit != RLIM_INFINITY) {
1959 if (*pos >= limit) {
1960 send_sig(SIGXFSZ, current, 0);
1963 if (*count > limit - (typeof(limit))*pos) {
1964 *count = limit - (typeof(limit))*pos;
1972 if (unlikely(*pos + *count > MAX_NON_LFS &&
1973 !(file->f_flags & O_LARGEFILE))) {
1974 if (*pos >= MAX_NON_LFS) {
1977 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1978 *count = MAX_NON_LFS - (unsigned long)*pos;
1983 * Are we about to exceed the fs block limit ?
1985 * If we have written data it becomes a short write. If we have
1986 * exceeded without writing data we send a signal and return EFBIG.
1987 * Linus frestrict idea will clean these up nicely..
1989 if (likely(!isblk)) {
1990 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1991 if (*count || *pos > inode->i_sb->s_maxbytes) {
1994 /* zero-length writes at ->s_maxbytes are OK */
1997 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1998 *count = inode->i_sb->s_maxbytes - *pos;
2002 if (bdev_read_only(I_BDEV(inode)))
2004 isize = i_size_read(inode);
2005 if (*pos >= isize) {
2006 if (*count || *pos > isize)
2010 if (*pos + *count > isize)
2011 *count = isize - *pos;
2018 EXPORT_SYMBOL(generic_write_checks);
2020 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2021 loff_t pos, unsigned len, unsigned flags,
2022 struct page **pagep, void **fsdata)
2024 const struct address_space_operations *aops = mapping->a_ops;
2026 if (aops->write_begin) {
2027 return aops->write_begin(file, mapping, pos, len, flags,
2031 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2032 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2033 struct inode *inode = mapping->host;
2036 page = grab_cache_page_write_begin(mapping, index, flags);
2041 if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) {
2043 * There is no way to resolve a short write situation
2044 * for a !Uptodate page (except by double copying in
2045 * the caller done by generic_perform_write_2copy).
2047 * Instead, we have to bring it uptodate here.
2049 ret = aops->readpage(file, page);
2050 page_cache_release(page);
2052 if (ret == AOP_TRUNCATED_PAGE)
2059 ret = aops->prepare_write(file, page, offset, offset+len);
2062 page_cache_release(page);
2063 if (pos + len > inode->i_size)
2064 vmtruncate(inode, inode->i_size);
2069 EXPORT_SYMBOL(pagecache_write_begin);
2071 int pagecache_write_end(struct file *file, struct address_space *mapping,
2072 loff_t pos, unsigned len, unsigned copied,
2073 struct page *page, void *fsdata)
2075 const struct address_space_operations *aops = mapping->a_ops;
2078 if (aops->write_end) {
2079 mark_page_accessed(page);
2080 ret = aops->write_end(file, mapping, pos, len, copied,
2083 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2084 struct inode *inode = mapping->host;
2086 flush_dcache_page(page);
2087 ret = aops->commit_write(file, page, offset, offset+len);
2089 mark_page_accessed(page);
2090 page_cache_release(page);
2093 if (pos + len > inode->i_size)
2094 vmtruncate(inode, inode->i_size);
2096 ret = min_t(size_t, copied, ret);
2103 EXPORT_SYMBOL(pagecache_write_end);
2106 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2107 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2108 size_t count, size_t ocount)
2110 struct file *file = iocb->ki_filp;
2111 struct address_space *mapping = file->f_mapping;
2112 struct inode *inode = mapping->host;
2117 if (count != ocount)
2118 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2121 * Unmap all mmappings of the file up-front.
2123 * This will cause any pte dirty bits to be propagated into the
2124 * pageframes for the subsequent filemap_write_and_wait().
2126 write_len = iov_length(iov, *nr_segs);
2127 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2128 if (mapping_mapped(mapping))
2129 unmap_mapping_range(mapping, pos, write_len, 0);
2131 written = filemap_write_and_wait(mapping);
2136 * After a write we want buffered reads to be sure to go to disk to get
2137 * the new data. We invalidate clean cached page from the region we're
2138 * about to write. We do this *before* the write so that we can return
2139 * without clobbering -EIOCBQUEUED from ->direct_IO().
2141 if (mapping->nrpages) {
2142 written = invalidate_inode_pages2_range(mapping,
2143 pos >> PAGE_CACHE_SHIFT, end);
2145 * If a page can not be invalidated, return 0 to fall back
2146 * to buffered write.
2149 if (written == -EBUSY)
2155 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2158 * Finally, try again to invalidate clean pages which might have been
2159 * cached by non-direct readahead, or faulted in by get_user_pages()
2160 * if the source of the write was an mmap'ed region of the file
2161 * we're writing. Either one is a pretty crazy thing to do,
2162 * so we don't support it 100%. If this invalidation
2163 * fails, tough, the write still worked...
2165 if (mapping->nrpages) {
2166 invalidate_inode_pages2_range(mapping,
2167 pos >> PAGE_CACHE_SHIFT, end);
2171 loff_t end = pos + written;
2172 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2173 i_size_write(inode, end);
2174 mark_inode_dirty(inode);
2180 * Sync the fs metadata but not the minor inode changes and
2181 * of course not the data as we did direct DMA for the IO.
2182 * i_mutex is held, which protects generic_osync_inode() from
2183 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
2186 if ((written >= 0 || written == -EIOCBQUEUED) &&
2187 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2188 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2194 EXPORT_SYMBOL(generic_file_direct_write);
2197 * Find or create a page at the given pagecache position. Return the locked
2198 * page. This function is specifically for buffered writes.
2200 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2201 pgoff_t index, unsigned flags)
2205 gfp_t gfp_notmask = 0;
2206 if (flags & AOP_FLAG_NOFS)
2207 gfp_notmask = __GFP_FS;
2209 page = find_lock_page(mapping, index);
2213 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2216 status = add_to_page_cache_lru(page, mapping, index,
2217 GFP_KERNEL & ~gfp_notmask);
2218 if (unlikely(status)) {
2219 page_cache_release(page);
2220 if (status == -EEXIST)
2226 EXPORT_SYMBOL(grab_cache_page_write_begin);
2228 static ssize_t generic_perform_write_2copy(struct file *file,
2229 struct iov_iter *i, loff_t pos)
2231 struct address_space *mapping = file->f_mapping;
2232 const struct address_space_operations *a_ops = mapping->a_ops;
2233 struct inode *inode = mapping->host;
2235 ssize_t written = 0;
2238 struct page *src_page;
2240 pgoff_t index; /* Pagecache index for current page */
2241 unsigned long offset; /* Offset into pagecache page */
2242 unsigned long bytes; /* Bytes to write to page */
2243 size_t copied; /* Bytes copied from user */
2245 offset = (pos & (PAGE_CACHE_SIZE - 1));
2246 index = pos >> PAGE_CACHE_SHIFT;
2247 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2251 * a non-NULL src_page indicates that we're doing the
2252 * copy via get_user_pages and kmap.
2257 * Bring in the user page that we will copy from _first_.
2258 * Otherwise there's a nasty deadlock on copying from the
2259 * same page as we're writing to, without it being marked
2262 * Not only is this an optimisation, but it is also required
2263 * to check that the address is actually valid, when atomic
2264 * usercopies are used, below.
2266 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2271 page = grab_cache_page_write_begin(mapping, index, 0);
2278 * non-uptodate pages cannot cope with short copies, and we
2279 * cannot take a pagefault with the destination page locked.
2280 * So pin the source page to copy it.
2282 if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) {
2285 src_page = alloc_page(GFP_KERNEL);
2287 page_cache_release(page);
2293 * Cannot get_user_pages with a page locked for the
2294 * same reason as we can't take a page fault with a
2295 * page locked (as explained below).
2297 copied = iov_iter_copy_from_user(src_page, i,
2299 if (unlikely(copied == 0)) {
2301 page_cache_release(page);
2302 page_cache_release(src_page);
2309 * Can't handle the page going uptodate here, because
2310 * that means we would use non-atomic usercopies, which
2311 * zero out the tail of the page, which can cause
2312 * zeroes to become transiently visible. We could just
2313 * use a non-zeroing copy, but the APIs aren't too
2316 if (unlikely(!page->mapping || PageUptodate(page))) {
2318 page_cache_release(page);
2319 page_cache_release(src_page);
2324 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2325 if (unlikely(status))
2326 goto fs_write_aop_error;
2330 * Must not enter the pagefault handler here, because
2331 * we hold the page lock, so we might recursively
2332 * deadlock on the same lock, or get an ABBA deadlock
2333 * against a different lock, or against the mmap_sem
2334 * (which nests outside the page lock). So increment
2335 * preempt count, and use _atomic usercopies.
2337 * The page is uptodate so we are OK to encounter a
2338 * short copy: if unmodified parts of the page are
2339 * marked dirty and written out to disk, it doesn't
2342 pagefault_disable();
2343 copied = iov_iter_copy_from_user_atomic(page, i,
2348 src = kmap_atomic(src_page, KM_USER0);
2349 dst = kmap_atomic(page, KM_USER1);
2350 memcpy(dst + offset, src + offset, bytes);
2351 kunmap_atomic(dst, KM_USER1);
2352 kunmap_atomic(src, KM_USER0);
2355 flush_dcache_page(page);
2357 status = a_ops->commit_write(file, page, offset, offset+bytes);
2358 if (unlikely(status < 0))
2359 goto fs_write_aop_error;
2360 if (unlikely(status > 0)) /* filesystem did partial write */
2361 copied = min_t(size_t, copied, status);
2364 mark_page_accessed(page);
2365 page_cache_release(page);
2367 page_cache_release(src_page);
2369 iov_iter_advance(i, copied);
2373 balance_dirty_pages_ratelimited(mapping);
2379 page_cache_release(page);
2381 page_cache_release(src_page);
2384 * prepare_write() may have instantiated a few blocks
2385 * outside i_size. Trim these off again. Don't need
2386 * i_size_read because we hold i_mutex.
2388 if (pos + bytes > inode->i_size)
2389 vmtruncate(inode, inode->i_size);
2391 } while (iov_iter_count(i));
2393 return written ? written : status;
2396 static ssize_t generic_perform_write(struct file *file,
2397 struct iov_iter *i, loff_t pos)
2399 struct address_space *mapping = file->f_mapping;
2400 const struct address_space_operations *a_ops = mapping->a_ops;
2402 ssize_t written = 0;
2403 unsigned int flags = 0;
2406 * Copies from kernel address space cannot fail (NFSD is a big user).
2408 if (segment_eq(get_fs(), KERNEL_DS))
2409 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2413 pgoff_t index; /* Pagecache index for current page */
2414 unsigned long offset; /* Offset into pagecache page */
2415 unsigned long bytes; /* Bytes to write to page */
2416 size_t copied; /* Bytes copied from user */
2419 offset = (pos & (PAGE_CACHE_SIZE - 1));
2420 index = pos >> PAGE_CACHE_SHIFT;
2421 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2427 * Bring in the user page that we will copy from _first_.
2428 * Otherwise there's a nasty deadlock on copying from the
2429 * same page as we're writing to, without it being marked
2432 * Not only is this an optimisation, but it is also required
2433 * to check that the address is actually valid, when atomic
2434 * usercopies are used, below.
2436 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2441 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2443 if (unlikely(status))
2446 pagefault_disable();
2447 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2449 flush_dcache_page(page);
2451 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2453 if (unlikely(status < 0))
2459 iov_iter_advance(i, copied);
2460 if (unlikely(copied == 0)) {
2462 * If we were unable to copy any data at all, we must
2463 * fall back to a single segment length write.
2465 * If we didn't fallback here, we could livelock
2466 * because not all segments in the iov can be copied at
2467 * once without a pagefault.
2469 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2470 iov_iter_single_seg_count(i));
2476 balance_dirty_pages_ratelimited(mapping);
2478 } while (iov_iter_count(i));
2480 return written ? written : status;
2484 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2485 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2486 size_t count, ssize_t written)
2488 struct file *file = iocb->ki_filp;
2489 struct address_space *mapping = file->f_mapping;
2490 const struct address_space_operations *a_ops = mapping->a_ops;
2491 struct inode *inode = mapping->host;
2495 iov_iter_init(&i, iov, nr_segs, count, written);
2496 if (a_ops->write_begin)
2497 status = generic_perform_write(file, &i, pos);
2499 status = generic_perform_write_2copy(file, &i, pos);
2501 if (likely(status >= 0)) {
2503 *ppos = pos + status;
2506 * For now, when the user asks for O_SYNC, we'll actually give
2509 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2510 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2511 status = generic_osync_inode(inode, mapping,
2512 OSYNC_METADATA|OSYNC_DATA);
2517 * If we get here for O_DIRECT writes then we must have fallen through
2518 * to buffered writes (block instantiation inside i_size). So we sync
2519 * the file data here, to try to honour O_DIRECT expectations.
2521 if (unlikely(file->f_flags & O_DIRECT) && written)
2522 status = filemap_write_and_wait(mapping);
2524 return written ? written : status;
2526 EXPORT_SYMBOL(generic_file_buffered_write);
2529 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2530 unsigned long nr_segs, loff_t *ppos)
2532 struct file *file = iocb->ki_filp;
2533 struct address_space * mapping = file->f_mapping;
2534 size_t ocount; /* original count */
2535 size_t count; /* after file limit checks */
2536 struct inode *inode = mapping->host;
2542 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2549 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2551 /* We can write back this queue in page reclaim */
2552 current->backing_dev_info = mapping->backing_dev_info;
2555 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2562 err = file_remove_suid(file);
2566 file_update_time(file);
2568 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2569 if (unlikely(file->f_flags & O_DIRECT)) {
2571 ssize_t written_buffered;
2573 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2574 ppos, count, ocount);
2575 if (written < 0 || written == count)
2578 * direct-io write to a hole: fall through to buffered I/O
2579 * for completing the rest of the request.
2583 written_buffered = generic_file_buffered_write(iocb, iov,
2584 nr_segs, pos, ppos, count,
2587 * If generic_file_buffered_write() retuned a synchronous error
2588 * then we want to return the number of bytes which were
2589 * direct-written, or the error code if that was zero. Note
2590 * that this differs from normal direct-io semantics, which
2591 * will return -EFOO even if some bytes were written.
2593 if (written_buffered < 0) {
2594 err = written_buffered;
2599 * We need to ensure that the page cache pages are written to
2600 * disk and invalidated to preserve the expected O_DIRECT
2603 endbyte = pos + written_buffered - written - 1;
2604 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2605 SYNC_FILE_RANGE_WAIT_BEFORE|
2606 SYNC_FILE_RANGE_WRITE|
2607 SYNC_FILE_RANGE_WAIT_AFTER);
2609 written = written_buffered;
2610 invalidate_mapping_pages(mapping,
2611 pos >> PAGE_CACHE_SHIFT,
2612 endbyte >> PAGE_CACHE_SHIFT);
2615 * We don't know how much we wrote, so just return
2616 * the number of bytes which were direct-written
2620 written = generic_file_buffered_write(iocb, iov, nr_segs,
2621 pos, ppos, count, written);
2624 current->backing_dev_info = NULL;
2625 return written ? written : err;
2628 ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2629 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
2631 struct file *file = iocb->ki_filp;
2632 struct address_space *mapping = file->f_mapping;
2633 struct inode *inode = mapping->host;
2636 BUG_ON(iocb->ki_pos != pos);
2638 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2641 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2644 err = sync_page_range_nolock(inode, mapping, pos, ret);
2650 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2652 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2653 unsigned long nr_segs, loff_t pos)
2655 struct file *file = iocb->ki_filp;
2656 struct address_space *mapping = file->f_mapping;
2657 struct inode *inode = mapping->host;
2660 BUG_ON(iocb->ki_pos != pos);
2662 mutex_lock(&inode->i_mutex);
2663 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2665 mutex_unlock(&inode->i_mutex);
2667 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2670 err = sync_page_range(inode, mapping, pos, ret);
2676 EXPORT_SYMBOL(generic_file_aio_write);
2679 * try_to_release_page() - release old fs-specific metadata on a page
2681 * @page: the page which the kernel is trying to free
2682 * @gfp_mask: memory allocation flags (and I/O mode)
2684 * The address_space is to try to release any data against the page
2685 * (presumably at page->private). If the release was successful, return `1'.
2686 * Otherwise return zero.
2688 * The @gfp_mask argument specifies whether I/O may be performed to release
2689 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2692 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2694 struct address_space * const mapping = page->mapping;
2696 BUG_ON(!PageLocked(page));
2697 if (PageWriteback(page))
2700 if (mapping && mapping->a_ops->releasepage)
2701 return mapping->a_ops->releasepage(page, gfp_mask);
2702 return try_to_free_buffers(page);
2705 EXPORT_SYMBOL(try_to_release_page);