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>
36 #include <linux/mm_inline.h> /* for page_is_file_cache() */
40 * FIXME: remove all knowledge of the buffer layer from the core VM
42 #include <linux/buffer_head.h> /* for generic_osync_inode */
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * Shared mappings now work. 15.8.1995 Bruno.
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
64 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
68 * ->i_mmap_lock (truncate->unmap_mapping_range)
72 * ->page_table_lock or pte_lock (various, mainly in memory.c)
73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 * ->lock_page (access_process_vm)
78 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
82 * ->i_alloc_sem (various)
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->dcache_lock (proc_pid_lookup)
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold the mapping's tree_lock.
115 void __remove_from_page_cache(struct page *page)
117 struct address_space *mapping = page->mapping;
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);
147 mem_cgroup_uncharge_cache_page(page);
150 static int sync_page(void *word)
152 struct address_space *mapping;
155 page = container_of((unsigned long *)word, struct page, flags);
158 * page_mapping() is being called without PG_locked held.
159 * Some knowledge of the state and use of the page is used to
160 * reduce the requirements down to a memory barrier.
161 * The danger here is of a stale page_mapping() return value
162 * indicating a struct address_space different from the one it's
163 * associated with when it is associated with one.
164 * After smp_mb(), it's either the correct page_mapping() for
165 * the page, or an old page_mapping() and the page's own
166 * page_mapping() has gone NULL.
167 * The ->sync_page() address_space operation must tolerate
168 * page_mapping() going NULL. By an amazing coincidence,
169 * this comes about because none of the users of the page
170 * in the ->sync_page() methods make essential use of the
171 * page_mapping(), merely passing the page down to the backing
172 * device's unplug functions when it's non-NULL, which in turn
173 * ignore it for all cases but swap, where only page_private(page) is
174 * of interest. When page_mapping() does go NULL, the entire
175 * call stack gracefully ignores the page and returns.
179 mapping = page_mapping(page);
180 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
181 mapping->a_ops->sync_page(page);
186 static int sync_page_killable(void *word)
189 return fatal_signal_pending(current) ? -EINTR : 0;
193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
194 * @mapping: address space structure to write
195 * @start: offset in bytes where the range starts
196 * @end: offset in bytes where the range ends (inclusive)
197 * @sync_mode: enable synchronous operation
199 * Start writeback against all of a mapping's dirty pages that lie
200 * within the byte offsets <start, end> inclusive.
202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
203 * opposed to a regular memory cleansing writeback. The difference between
204 * these two operations is that if a dirty page/buffer is encountered, it must
205 * be waited upon, and not just skipped over.
207 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
208 loff_t end, int sync_mode)
211 struct writeback_control wbc = {
212 .sync_mode = sync_mode,
213 .nr_to_write = LONG_MAX,
214 .range_start = start,
218 if (!mapping_cap_writeback_dirty(mapping))
221 ret = do_writepages(mapping, &wbc);
225 static inline int __filemap_fdatawrite(struct address_space *mapping,
228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
231 int filemap_fdatawrite(struct address_space *mapping)
233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
235 EXPORT_SYMBOL(filemap_fdatawrite);
237 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
242 EXPORT_SYMBOL(filemap_fdatawrite_range);
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
251 int filemap_flush(struct address_space *mapping)
253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
255 EXPORT_SYMBOL(filemap_flush);
258 * wait_on_page_writeback_range - wait for writeback to complete
259 * @mapping: target address_space
260 * @start: beginning page index
261 * @end: ending page index
263 * Wait for writeback to complete against pages indexed by start->end
266 int wait_on_page_writeback_range(struct address_space *mapping,
267 pgoff_t start, pgoff_t end)
277 pagevec_init(&pvec, 0);
279 while ((index <= end) &&
280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
281 PAGECACHE_TAG_WRITEBACK,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
285 for (i = 0; i < nr_pages; i++) {
286 struct page *page = pvec.pages[i];
288 /* until radix tree lookup accepts end_index */
289 if (page->index > end)
292 wait_on_page_writeback(page);
296 pagevec_release(&pvec);
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
303 if (test_and_clear_bit(AS_EIO, &mapping->flags))
310 * filemap_fdatawait_range - wait for all under-writeback pages to complete in a given range
311 * @mapping: address space structure to wait for
312 * @start: offset in bytes where the range starts
313 * @end: offset in bytes where the range ends (inclusive)
315 * Walk the list of under-writeback pages of the given address space
316 * in the given range and wait for all of them.
318 * This is just a simple wrapper so that callers don't have to convert offsets
319 * to page indexes themselves
321 int filemap_fdatawait_range(struct address_space *mapping, loff_t start,
324 return wait_on_page_writeback_range(mapping, start >> PAGE_CACHE_SHIFT,
325 end >> PAGE_CACHE_SHIFT);
327 EXPORT_SYMBOL(filemap_fdatawait_range);
330 * sync_page_range - write and wait on all pages in the passed range
331 * @inode: target inode
332 * @mapping: target address_space
333 * @pos: beginning offset in pages to write
334 * @count: number of bytes to write
336 * Write and wait upon all the pages in the passed range. This is a "data
337 * integrity" operation. It waits upon in-flight writeout before starting and
338 * waiting upon new writeout. If there was an IO error, return it.
340 * We need to re-take i_mutex during the generic_osync_inode list walk because
341 * it is otherwise livelockable.
343 int sync_page_range(struct inode *inode, struct address_space *mapping,
344 loff_t pos, loff_t count)
346 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
347 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
350 if (!mapping_cap_writeback_dirty(mapping) || !count)
352 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
354 mutex_lock(&inode->i_mutex);
355 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
356 mutex_unlock(&inode->i_mutex);
359 ret = wait_on_page_writeback_range(mapping, start, end);
362 EXPORT_SYMBOL(sync_page_range);
365 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
366 * @inode: target inode
367 * @mapping: target address_space
368 * @pos: beginning offset in pages to write
369 * @count: number of bytes to write
371 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
372 * as it forces O_SYNC writers to different parts of the same file
373 * to be serialised right until io completion.
375 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
376 loff_t pos, loff_t count)
378 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
379 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
382 if (!mapping_cap_writeback_dirty(mapping) || !count)
384 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
386 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
388 ret = wait_on_page_writeback_range(mapping, start, end);
391 EXPORT_SYMBOL(sync_page_range_nolock);
394 * filemap_fdatawait - wait for all under-writeback pages to complete
395 * @mapping: address space structure to wait for
397 * Walk the list of under-writeback pages of the given address space
398 * and wait for all of them.
400 int filemap_fdatawait(struct address_space *mapping)
402 loff_t i_size = i_size_read(mapping->host);
407 return wait_on_page_writeback_range(mapping, 0,
408 (i_size - 1) >> PAGE_CACHE_SHIFT);
410 EXPORT_SYMBOL(filemap_fdatawait);
412 int filemap_write_and_wait(struct address_space *mapping)
416 if (mapping->nrpages) {
417 err = filemap_fdatawrite(mapping);
419 * Even if the above returned error, the pages may be
420 * written partially (e.g. -ENOSPC), so we wait for it.
421 * But the -EIO is special case, it may indicate the worst
422 * thing (e.g. bug) happened, so we avoid waiting for it.
425 int err2 = filemap_fdatawait(mapping);
432 EXPORT_SYMBOL(filemap_write_and_wait);
435 * filemap_write_and_wait_range - write out & wait on a file range
436 * @mapping: the address_space for the pages
437 * @lstart: offset in bytes where the range starts
438 * @lend: offset in bytes where the range ends (inclusive)
440 * Write out and wait upon file offsets lstart->lend, inclusive.
442 * Note that `lend' is inclusive (describes the last byte to be written) so
443 * that this function can be used to write to the very end-of-file (end = -1).
445 int filemap_write_and_wait_range(struct address_space *mapping,
446 loff_t lstart, loff_t lend)
450 if (mapping->nrpages) {
451 err = __filemap_fdatawrite_range(mapping, lstart, lend,
453 /* See comment of filemap_write_and_wait() */
455 int err2 = wait_on_page_writeback_range(mapping,
456 lstart >> PAGE_CACHE_SHIFT,
457 lend >> PAGE_CACHE_SHIFT);
464 EXPORT_SYMBOL(filemap_write_and_wait_range);
467 * add_to_page_cache_locked - add a locked page to the pagecache
469 * @mapping: the page's address_space
470 * @offset: page index
471 * @gfp_mask: page allocation mode
473 * This function is used to add a page to the pagecache. It must be locked.
474 * This function does not add the page to the LRU. The caller must do that.
476 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
477 pgoff_t offset, gfp_t gfp_mask)
481 VM_BUG_ON(!PageLocked(page));
483 error = mem_cgroup_cache_charge(page, current->mm,
484 gfp_mask & GFP_RECLAIM_MASK);
488 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
490 page_cache_get(page);
491 page->mapping = mapping;
492 page->index = offset;
494 spin_lock_irq(&mapping->tree_lock);
495 error = radix_tree_insert(&mapping->page_tree, offset, page);
496 if (likely(!error)) {
498 __inc_zone_page_state(page, NR_FILE_PAGES);
499 spin_unlock_irq(&mapping->tree_lock);
501 page->mapping = NULL;
502 spin_unlock_irq(&mapping->tree_lock);
503 mem_cgroup_uncharge_cache_page(page);
504 page_cache_release(page);
506 radix_tree_preload_end();
508 mem_cgroup_uncharge_cache_page(page);
512 EXPORT_SYMBOL(add_to_page_cache_locked);
514 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
515 pgoff_t offset, gfp_t gfp_mask)
520 * Splice_read and readahead add shmem/tmpfs pages into the page cache
521 * before shmem_readpage has a chance to mark them as SwapBacked: they
522 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
523 * (called in add_to_page_cache) needs to know where they're going too.
525 if (mapping_cap_swap_backed(mapping))
526 SetPageSwapBacked(page);
528 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
530 if (page_is_file_cache(page))
531 lru_cache_add_file(page);
533 lru_cache_add_active_anon(page);
537 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
540 struct page *__page_cache_alloc(gfp_t gfp)
542 if (cpuset_do_page_mem_spread()) {
543 int n = cpuset_mem_spread_node();
544 return alloc_pages_exact_node(n, gfp, 0);
546 return alloc_pages(gfp, 0);
548 EXPORT_SYMBOL(__page_cache_alloc);
551 static int __sleep_on_page_lock(void *word)
558 * In order to wait for pages to become available there must be
559 * waitqueues associated with pages. By using a hash table of
560 * waitqueues where the bucket discipline is to maintain all
561 * waiters on the same queue and wake all when any of the pages
562 * become available, and for the woken contexts to check to be
563 * sure the appropriate page became available, this saves space
564 * at a cost of "thundering herd" phenomena during rare hash
567 static wait_queue_head_t *page_waitqueue(struct page *page)
569 const struct zone *zone = page_zone(page);
571 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
574 static inline void wake_up_page(struct page *page, int bit)
576 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
579 void wait_on_page_bit(struct page *page, int bit_nr)
581 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
583 if (test_bit(bit_nr, &page->flags))
584 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
585 TASK_UNINTERRUPTIBLE);
587 EXPORT_SYMBOL(wait_on_page_bit);
590 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
591 * @page: Page defining the wait queue of interest
592 * @waiter: Waiter to add to the queue
594 * Add an arbitrary @waiter to the wait queue for the nominated @page.
596 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
598 wait_queue_head_t *q = page_waitqueue(page);
601 spin_lock_irqsave(&q->lock, flags);
602 __add_wait_queue(q, waiter);
603 spin_unlock_irqrestore(&q->lock, flags);
605 EXPORT_SYMBOL_GPL(add_page_wait_queue);
608 * unlock_page - unlock a locked page
611 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
612 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
613 * mechananism between PageLocked pages and PageWriteback pages is shared.
614 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
616 * The mb is necessary to enforce ordering between the clear_bit and the read
617 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
619 void unlock_page(struct page *page)
621 VM_BUG_ON(!PageLocked(page));
622 clear_bit_unlock(PG_locked, &page->flags);
623 smp_mb__after_clear_bit();
624 wake_up_page(page, PG_locked);
626 EXPORT_SYMBOL(unlock_page);
629 * end_page_writeback - end writeback against a page
632 void end_page_writeback(struct page *page)
634 if (TestClearPageReclaim(page))
635 rotate_reclaimable_page(page);
637 if (!test_clear_page_writeback(page))
640 smp_mb__after_clear_bit();
641 wake_up_page(page, PG_writeback);
643 EXPORT_SYMBOL(end_page_writeback);
646 * __lock_page - get a lock on the page, assuming we need to sleep to get it
647 * @page: the page to lock
649 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
650 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
651 * chances are that on the second loop, the block layer's plug list is empty,
652 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
654 void __lock_page(struct page *page)
656 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
658 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
659 TASK_UNINTERRUPTIBLE);
661 EXPORT_SYMBOL(__lock_page);
663 int __lock_page_killable(struct page *page)
665 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
667 return __wait_on_bit_lock(page_waitqueue(page), &wait,
668 sync_page_killable, TASK_KILLABLE);
670 EXPORT_SYMBOL_GPL(__lock_page_killable);
673 * __lock_page_nosync - get a lock on the page, without calling sync_page()
674 * @page: the page to lock
676 * Variant of lock_page that does not require the caller to hold a reference
677 * on the page's mapping.
679 void __lock_page_nosync(struct page *page)
681 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
682 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
683 TASK_UNINTERRUPTIBLE);
687 * find_get_page - find and get a page reference
688 * @mapping: the address_space to search
689 * @offset: the page index
691 * Is there a pagecache struct page at the given (mapping, offset) tuple?
692 * If yes, increment its refcount and return it; if no, return NULL.
694 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
702 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
704 page = radix_tree_deref_slot(pagep);
705 if (unlikely(!page || page == RADIX_TREE_RETRY))
708 if (!page_cache_get_speculative(page))
712 * Has the page moved?
713 * This is part of the lockless pagecache protocol. See
714 * include/linux/pagemap.h for details.
716 if (unlikely(page != *pagep)) {
717 page_cache_release(page);
725 EXPORT_SYMBOL(find_get_page);
728 * find_lock_page - locate, pin and lock a pagecache page
729 * @mapping: the address_space to search
730 * @offset: the page index
732 * Locates the desired pagecache page, locks it, increments its reference
733 * count and returns its address.
735 * Returns zero if the page was not present. find_lock_page() may sleep.
737 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
742 page = find_get_page(mapping, offset);
745 /* Has the page been truncated? */
746 if (unlikely(page->mapping != mapping)) {
748 page_cache_release(page);
751 VM_BUG_ON(page->index != offset);
755 EXPORT_SYMBOL(find_lock_page);
758 * find_or_create_page - locate or add a pagecache page
759 * @mapping: the page's address_space
760 * @index: the page's index into the mapping
761 * @gfp_mask: page allocation mode
763 * Locates a page in the pagecache. If the page is not present, a new page
764 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
765 * LRU list. The returned page is locked and has its reference count
768 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
771 * find_or_create_page() returns the desired page's address, or zero on
774 struct page *find_or_create_page(struct address_space *mapping,
775 pgoff_t index, gfp_t gfp_mask)
780 page = find_lock_page(mapping, index);
782 page = __page_cache_alloc(gfp_mask);
786 * We want a regular kernel memory (not highmem or DMA etc)
787 * allocation for the radix tree nodes, but we need to honour
788 * the context-specific requirements the caller has asked for.
789 * GFP_RECLAIM_MASK collects those requirements.
791 err = add_to_page_cache_lru(page, mapping, index,
792 (gfp_mask & GFP_RECLAIM_MASK));
794 page_cache_release(page);
802 EXPORT_SYMBOL(find_or_create_page);
805 * find_get_pages - gang pagecache lookup
806 * @mapping: The address_space to search
807 * @start: The starting page index
808 * @nr_pages: The maximum number of pages
809 * @pages: Where the resulting pages are placed
811 * find_get_pages() will search for and return a group of up to
812 * @nr_pages pages in the mapping. The pages are placed at @pages.
813 * find_get_pages() takes a reference against the returned pages.
815 * The search returns a group of mapping-contiguous pages with ascending
816 * indexes. There may be holes in the indices due to not-present pages.
818 * find_get_pages() returns the number of pages which were found.
820 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
821 unsigned int nr_pages, struct page **pages)
825 unsigned int nr_found;
829 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
830 (void ***)pages, start, nr_pages);
832 for (i = 0; i < nr_found; i++) {
835 page = radix_tree_deref_slot((void **)pages[i]);
839 * this can only trigger if nr_found == 1, making livelock
842 if (unlikely(page == RADIX_TREE_RETRY))
845 if (!page_cache_get_speculative(page))
848 /* Has the page moved? */
849 if (unlikely(page != *((void **)pages[i]))) {
850 page_cache_release(page);
862 * find_get_pages_contig - gang contiguous pagecache lookup
863 * @mapping: The address_space to search
864 * @index: The starting page index
865 * @nr_pages: The maximum number of pages
866 * @pages: Where the resulting pages are placed
868 * find_get_pages_contig() works exactly like find_get_pages(), except
869 * that the returned number of pages are guaranteed to be contiguous.
871 * find_get_pages_contig() returns the number of pages which were found.
873 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
874 unsigned int nr_pages, struct page **pages)
878 unsigned int nr_found;
882 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
883 (void ***)pages, index, nr_pages);
885 for (i = 0; i < nr_found; i++) {
888 page = radix_tree_deref_slot((void **)pages[i]);
892 * this can only trigger if nr_found == 1, making livelock
895 if (unlikely(page == RADIX_TREE_RETRY))
898 if (page->mapping == NULL || page->index != index)
901 if (!page_cache_get_speculative(page))
904 /* Has the page moved? */
905 if (unlikely(page != *((void **)pages[i]))) {
906 page_cache_release(page);
917 EXPORT_SYMBOL(find_get_pages_contig);
920 * find_get_pages_tag - find and return pages that match @tag
921 * @mapping: the address_space to search
922 * @index: the starting page index
923 * @tag: the tag index
924 * @nr_pages: the maximum number of pages
925 * @pages: where the resulting pages are placed
927 * Like find_get_pages, except we only return pages which are tagged with
928 * @tag. We update @index to index the next page for the traversal.
930 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
931 int tag, unsigned int nr_pages, struct page **pages)
935 unsigned int nr_found;
939 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
940 (void ***)pages, *index, nr_pages, tag);
942 for (i = 0; i < nr_found; i++) {
945 page = radix_tree_deref_slot((void **)pages[i]);
949 * this can only trigger if nr_found == 1, making livelock
952 if (unlikely(page == RADIX_TREE_RETRY))
955 if (!page_cache_get_speculative(page))
958 /* Has the page moved? */
959 if (unlikely(page != *((void **)pages[i]))) {
960 page_cache_release(page);
970 *index = pages[ret - 1]->index + 1;
974 EXPORT_SYMBOL(find_get_pages_tag);
977 * grab_cache_page_nowait - returns locked page at given index in given cache
978 * @mapping: target address_space
979 * @index: the page index
981 * Same as grab_cache_page(), but do not wait if the page is unavailable.
982 * This is intended for speculative data generators, where the data can
983 * be regenerated if the page couldn't be grabbed. This routine should
984 * be safe to call while holding the lock for another page.
986 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
987 * and deadlock against the caller's locked page.
990 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
992 struct page *page = find_get_page(mapping, index);
995 if (trylock_page(page))
997 page_cache_release(page);
1000 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1001 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1002 page_cache_release(page);
1007 EXPORT_SYMBOL(grab_cache_page_nowait);
1010 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1011 * a _large_ part of the i/o request. Imagine the worst scenario:
1013 * ---R__________________________________________B__________
1014 * ^ reading here ^ bad block(assume 4k)
1016 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1017 * => failing the whole request => read(R) => read(R+1) =>
1018 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1019 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1020 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1022 * It is going insane. Fix it by quickly scaling down the readahead size.
1024 static void shrink_readahead_size_eio(struct file *filp,
1025 struct file_ra_state *ra)
1031 * do_generic_file_read - generic file read routine
1032 * @filp: the file to read
1033 * @ppos: current file position
1034 * @desc: read_descriptor
1035 * @actor: read method
1037 * This is a generic file read routine, and uses the
1038 * mapping->a_ops->readpage() function for the actual low-level stuff.
1040 * This is really ugly. But the goto's actually try to clarify some
1041 * of the logic when it comes to error handling etc.
1043 static void do_generic_file_read(struct file *filp, loff_t *ppos,
1044 read_descriptor_t *desc, read_actor_t actor)
1046 struct address_space *mapping = filp->f_mapping;
1047 struct inode *inode = mapping->host;
1048 struct file_ra_state *ra = &filp->f_ra;
1052 unsigned long offset; /* offset into pagecache page */
1053 unsigned int prev_offset;
1056 index = *ppos >> PAGE_CACHE_SHIFT;
1057 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1058 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1059 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1060 offset = *ppos & ~PAGE_CACHE_MASK;
1066 unsigned long nr, ret;
1070 page = find_get_page(mapping, index);
1072 page_cache_sync_readahead(mapping,
1074 index, last_index - index);
1075 page = find_get_page(mapping, index);
1076 if (unlikely(page == NULL))
1077 goto no_cached_page;
1079 if (PageReadahead(page)) {
1080 page_cache_async_readahead(mapping,
1082 index, last_index - index);
1084 if (!PageUptodate(page)) {
1085 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1086 !mapping->a_ops->is_partially_uptodate)
1087 goto page_not_up_to_date;
1088 if (!trylock_page(page))
1089 goto page_not_up_to_date;
1090 if (!mapping->a_ops->is_partially_uptodate(page,
1092 goto page_not_up_to_date_locked;
1097 * i_size must be checked after we know the page is Uptodate.
1099 * Checking i_size after the check allows us to calculate
1100 * the correct value for "nr", which means the zero-filled
1101 * part of the page is not copied back to userspace (unless
1102 * another truncate extends the file - this is desired though).
1105 isize = i_size_read(inode);
1106 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1107 if (unlikely(!isize || index > end_index)) {
1108 page_cache_release(page);
1112 /* nr is the maximum number of bytes to copy from this page */
1113 nr = PAGE_CACHE_SIZE;
1114 if (index == end_index) {
1115 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1117 page_cache_release(page);
1123 /* If users can be writing to this page using arbitrary
1124 * virtual addresses, take care about potential aliasing
1125 * before reading the page on the kernel side.
1127 if (mapping_writably_mapped(mapping))
1128 flush_dcache_page(page);
1131 * When a sequential read accesses a page several times,
1132 * only mark it as accessed the first time.
1134 if (prev_index != index || offset != prev_offset)
1135 mark_page_accessed(page);
1139 * Ok, we have the page, and it's up-to-date, so
1140 * now we can copy it to user space...
1142 * The actor routine returns how many bytes were actually used..
1143 * NOTE! This may not be the same as how much of a user buffer
1144 * we filled up (we may be padding etc), so we can only update
1145 * "pos" here (the actor routine has to update the user buffer
1146 * pointers and the remaining count).
1148 ret = actor(desc, page, offset, nr);
1150 index += offset >> PAGE_CACHE_SHIFT;
1151 offset &= ~PAGE_CACHE_MASK;
1152 prev_offset = offset;
1154 page_cache_release(page);
1155 if (ret == nr && desc->count)
1159 page_not_up_to_date:
1160 /* Get exclusive access to the page ... */
1161 error = lock_page_killable(page);
1162 if (unlikely(error))
1163 goto readpage_error;
1165 page_not_up_to_date_locked:
1166 /* Did it get truncated before we got the lock? */
1167 if (!page->mapping) {
1169 page_cache_release(page);
1173 /* Did somebody else fill it already? */
1174 if (PageUptodate(page)) {
1180 /* Start the actual read. The read will unlock the page. */
1181 error = mapping->a_ops->readpage(filp, page);
1183 if (unlikely(error)) {
1184 if (error == AOP_TRUNCATED_PAGE) {
1185 page_cache_release(page);
1188 goto readpage_error;
1191 if (!PageUptodate(page)) {
1192 error = lock_page_killable(page);
1193 if (unlikely(error))
1194 goto readpage_error;
1195 if (!PageUptodate(page)) {
1196 if (page->mapping == NULL) {
1198 * invalidate_inode_pages got it
1201 page_cache_release(page);
1205 shrink_readahead_size_eio(filp, ra);
1207 goto readpage_error;
1215 /* UHHUH! A synchronous read error occurred. Report it */
1216 desc->error = error;
1217 page_cache_release(page);
1222 * Ok, it wasn't cached, so we need to create a new
1225 page = page_cache_alloc_cold(mapping);
1227 desc->error = -ENOMEM;
1230 error = add_to_page_cache_lru(page, mapping,
1233 page_cache_release(page);
1234 if (error == -EEXIST)
1236 desc->error = error;
1243 ra->prev_pos = prev_index;
1244 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1245 ra->prev_pos |= prev_offset;
1247 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1248 file_accessed(filp);
1251 int file_read_actor(read_descriptor_t *desc, struct page *page,
1252 unsigned long offset, unsigned long size)
1255 unsigned long left, count = desc->count;
1261 * Faults on the destination of a read are common, so do it before
1264 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1265 kaddr = kmap_atomic(page, KM_USER0);
1266 left = __copy_to_user_inatomic(desc->arg.buf,
1267 kaddr + offset, size);
1268 kunmap_atomic(kaddr, KM_USER0);
1273 /* Do it the slow way */
1275 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1280 desc->error = -EFAULT;
1283 desc->count = count - size;
1284 desc->written += size;
1285 desc->arg.buf += size;
1290 * Performs necessary checks before doing a write
1291 * @iov: io vector request
1292 * @nr_segs: number of segments in the iovec
1293 * @count: number of bytes to write
1294 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1296 * Adjust number of segments and amount of bytes to write (nr_segs should be
1297 * properly initialized first). Returns appropriate error code that caller
1298 * should return or zero in case that write should be allowed.
1300 int generic_segment_checks(const struct iovec *iov,
1301 unsigned long *nr_segs, size_t *count, int access_flags)
1305 for (seg = 0; seg < *nr_segs; seg++) {
1306 const struct iovec *iv = &iov[seg];
1309 * If any segment has a negative length, or the cumulative
1310 * length ever wraps negative then return -EINVAL.
1313 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1315 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1320 cnt -= iv->iov_len; /* This segment is no good */
1326 EXPORT_SYMBOL(generic_segment_checks);
1329 * generic_file_aio_read - generic filesystem read routine
1330 * @iocb: kernel I/O control block
1331 * @iov: io vector request
1332 * @nr_segs: number of segments in the iovec
1333 * @pos: current file position
1335 * This is the "read()" routine for all filesystems
1336 * that can use the page cache directly.
1339 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1340 unsigned long nr_segs, loff_t pos)
1342 struct file *filp = iocb->ki_filp;
1346 loff_t *ppos = &iocb->ki_pos;
1349 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1353 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1354 if (filp->f_flags & O_DIRECT) {
1356 struct address_space *mapping;
1357 struct inode *inode;
1359 mapping = filp->f_mapping;
1360 inode = mapping->host;
1362 goto out; /* skip atime */
1363 size = i_size_read(inode);
1365 retval = filemap_write_and_wait_range(mapping, pos,
1366 pos + iov_length(iov, nr_segs) - 1);
1368 retval = mapping->a_ops->direct_IO(READ, iocb,
1372 *ppos = pos + retval;
1374 file_accessed(filp);
1380 for (seg = 0; seg < nr_segs; seg++) {
1381 read_descriptor_t desc;
1384 desc.arg.buf = iov[seg].iov_base;
1385 desc.count = iov[seg].iov_len;
1386 if (desc.count == 0)
1389 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1390 retval += desc.written;
1392 retval = retval ?: desc.error;
1401 EXPORT_SYMBOL(generic_file_aio_read);
1404 do_readahead(struct address_space *mapping, struct file *filp,
1405 pgoff_t index, unsigned long nr)
1407 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1410 force_page_cache_readahead(mapping, filp, index, nr);
1414 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1422 if (file->f_mode & FMODE_READ) {
1423 struct address_space *mapping = file->f_mapping;
1424 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1425 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1426 unsigned long len = end - start + 1;
1427 ret = do_readahead(mapping, file, start, len);
1433 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1434 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1436 return SYSC_readahead((int) fd, offset, (size_t) count);
1438 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1443 * page_cache_read - adds requested page to the page cache if not already there
1444 * @file: file to read
1445 * @offset: page index
1447 * This adds the requested page to the page cache if it isn't already there,
1448 * and schedules an I/O to read in its contents from disk.
1450 static int page_cache_read(struct file *file, pgoff_t offset)
1452 struct address_space *mapping = file->f_mapping;
1457 page = page_cache_alloc_cold(mapping);
1461 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1463 ret = mapping->a_ops->readpage(file, page);
1464 else if (ret == -EEXIST)
1465 ret = 0; /* losing race to add is OK */
1467 page_cache_release(page);
1469 } while (ret == AOP_TRUNCATED_PAGE);
1474 #define MMAP_LOTSAMISS (100)
1477 * Synchronous readahead happens when we don't even find
1478 * a page in the page cache at all.
1480 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1481 struct file_ra_state *ra,
1485 unsigned long ra_pages;
1486 struct address_space *mapping = file->f_mapping;
1488 /* If we don't want any read-ahead, don't bother */
1489 if (VM_RandomReadHint(vma))
1492 if (VM_SequentialReadHint(vma) ||
1493 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1494 page_cache_sync_readahead(mapping, ra, file, offset,
1499 if (ra->mmap_miss < INT_MAX)
1503 * Do we miss much more than hit in this file? If so,
1504 * stop bothering with read-ahead. It will only hurt.
1506 if (ra->mmap_miss > MMAP_LOTSAMISS)
1512 ra_pages = max_sane_readahead(ra->ra_pages);
1514 ra->start = max_t(long, 0, offset - ra_pages/2);
1515 ra->size = ra_pages;
1517 ra_submit(ra, mapping, file);
1522 * Asynchronous readahead happens when we find the page and PG_readahead,
1523 * so we want to possibly extend the readahead further..
1525 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1526 struct file_ra_state *ra,
1531 struct address_space *mapping = file->f_mapping;
1533 /* If we don't want any read-ahead, don't bother */
1534 if (VM_RandomReadHint(vma))
1536 if (ra->mmap_miss > 0)
1538 if (PageReadahead(page))
1539 page_cache_async_readahead(mapping, ra, file,
1540 page, offset, ra->ra_pages);
1544 * filemap_fault - read in file data for page fault handling
1545 * @vma: vma in which the fault was taken
1546 * @vmf: struct vm_fault containing details of the fault
1548 * filemap_fault() is invoked via the vma operations vector for a
1549 * mapped memory region to read in file data during a page fault.
1551 * The goto's are kind of ugly, but this streamlines the normal case of having
1552 * it in the page cache, and handles the special cases reasonably without
1553 * having a lot of duplicated code.
1555 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1558 struct file *file = vma->vm_file;
1559 struct address_space *mapping = file->f_mapping;
1560 struct file_ra_state *ra = &file->f_ra;
1561 struct inode *inode = mapping->host;
1562 pgoff_t offset = vmf->pgoff;
1567 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1569 return VM_FAULT_SIGBUS;
1572 * Do we have something in the page cache already?
1574 page = find_get_page(mapping, offset);
1577 * We found the page, so try async readahead before
1578 * waiting for the lock.
1580 do_async_mmap_readahead(vma, ra, file, page, offset);
1583 /* Did it get truncated? */
1584 if (unlikely(page->mapping != mapping)) {
1587 goto no_cached_page;
1590 /* No page in the page cache at all */
1591 do_sync_mmap_readahead(vma, ra, file, offset);
1592 count_vm_event(PGMAJFAULT);
1593 ret = VM_FAULT_MAJOR;
1595 page = find_lock_page(mapping, offset);
1597 goto no_cached_page;
1601 * We have a locked page in the page cache, now we need to check
1602 * that it's up-to-date. If not, it is going to be due to an error.
1604 if (unlikely(!PageUptodate(page)))
1605 goto page_not_uptodate;
1608 * Found the page and have a reference on it.
1609 * We must recheck i_size under page lock.
1611 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1612 if (unlikely(offset >= size)) {
1614 page_cache_release(page);
1615 return VM_FAULT_SIGBUS;
1618 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1620 return ret | VM_FAULT_LOCKED;
1624 * We're only likely to ever get here if MADV_RANDOM is in
1627 error = page_cache_read(file, offset);
1630 * The page we want has now been added to the page cache.
1631 * In the unlikely event that someone removed it in the
1632 * meantime, we'll just come back here and read it again.
1638 * An error return from page_cache_read can result if the
1639 * system is low on memory, or a problem occurs while trying
1642 if (error == -ENOMEM)
1643 return VM_FAULT_OOM;
1644 return VM_FAULT_SIGBUS;
1648 * Umm, take care of errors if the page isn't up-to-date.
1649 * Try to re-read it _once_. We do this synchronously,
1650 * because there really aren't any performance issues here
1651 * and we need to check for errors.
1653 ClearPageError(page);
1654 error = mapping->a_ops->readpage(file, page);
1656 wait_on_page_locked(page);
1657 if (!PageUptodate(page))
1660 page_cache_release(page);
1662 if (!error || error == AOP_TRUNCATED_PAGE)
1665 /* Things didn't work out. Return zero to tell the mm layer so. */
1666 shrink_readahead_size_eio(file, ra);
1667 return VM_FAULT_SIGBUS;
1669 EXPORT_SYMBOL(filemap_fault);
1671 struct vm_operations_struct generic_file_vm_ops = {
1672 .fault = filemap_fault,
1675 /* This is used for a general mmap of a disk file */
1677 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1679 struct address_space *mapping = file->f_mapping;
1681 if (!mapping->a_ops->readpage)
1683 file_accessed(file);
1684 vma->vm_ops = &generic_file_vm_ops;
1685 vma->vm_flags |= VM_CAN_NONLINEAR;
1690 * This is for filesystems which do not implement ->writepage.
1692 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1694 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1696 return generic_file_mmap(file, vma);
1699 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1703 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1707 #endif /* CONFIG_MMU */
1709 EXPORT_SYMBOL(generic_file_mmap);
1710 EXPORT_SYMBOL(generic_file_readonly_mmap);
1712 static struct page *__read_cache_page(struct address_space *mapping,
1714 int (*filler)(void *,struct page*),
1720 page = find_get_page(mapping, index);
1722 page = page_cache_alloc_cold(mapping);
1724 return ERR_PTR(-ENOMEM);
1725 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1726 if (unlikely(err)) {
1727 page_cache_release(page);
1730 /* Presumably ENOMEM for radix tree node */
1731 return ERR_PTR(err);
1733 err = filler(data, page);
1735 page_cache_release(page);
1736 page = ERR_PTR(err);
1743 * read_cache_page_async - read into page cache, fill it if needed
1744 * @mapping: the page's address_space
1745 * @index: the page index
1746 * @filler: function to perform the read
1747 * @data: destination for read data
1749 * Same as read_cache_page, but don't wait for page to become unlocked
1750 * after submitting it to the filler.
1752 * Read into the page cache. If a page already exists, and PageUptodate() is
1753 * not set, try to fill the page but don't wait for it to become unlocked.
1755 * If the page does not get brought uptodate, return -EIO.
1757 struct page *read_cache_page_async(struct address_space *mapping,
1759 int (*filler)(void *,struct page*),
1766 page = __read_cache_page(mapping, index, filler, data);
1769 if (PageUptodate(page))
1773 if (!page->mapping) {
1775 page_cache_release(page);
1778 if (PageUptodate(page)) {
1782 err = filler(data, page);
1784 page_cache_release(page);
1785 return ERR_PTR(err);
1788 mark_page_accessed(page);
1791 EXPORT_SYMBOL(read_cache_page_async);
1794 * read_cache_page - read into page cache, fill it if needed
1795 * @mapping: the page's address_space
1796 * @index: the page index
1797 * @filler: function to perform the read
1798 * @data: destination for read data
1800 * Read into the page cache. If a page already exists, and PageUptodate() is
1801 * not set, try to fill the page then wait for it to become unlocked.
1803 * If the page does not get brought uptodate, return -EIO.
1805 struct page *read_cache_page(struct address_space *mapping,
1807 int (*filler)(void *,struct page*),
1812 page = read_cache_page_async(mapping, index, filler, data);
1815 wait_on_page_locked(page);
1816 if (!PageUptodate(page)) {
1817 page_cache_release(page);
1818 page = ERR_PTR(-EIO);
1823 EXPORT_SYMBOL(read_cache_page);
1826 * The logic we want is
1828 * if suid or (sgid and xgrp)
1831 int should_remove_suid(struct dentry *dentry)
1833 mode_t mode = dentry->d_inode->i_mode;
1836 /* suid always must be killed */
1837 if (unlikely(mode & S_ISUID))
1838 kill = ATTR_KILL_SUID;
1841 * sgid without any exec bits is just a mandatory locking mark; leave
1842 * it alone. If some exec bits are set, it's a real sgid; kill it.
1844 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1845 kill |= ATTR_KILL_SGID;
1847 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1852 EXPORT_SYMBOL(should_remove_suid);
1854 static int __remove_suid(struct dentry *dentry, int kill)
1856 struct iattr newattrs;
1858 newattrs.ia_valid = ATTR_FORCE | kill;
1859 return notify_change(dentry, &newattrs);
1862 int file_remove_suid(struct file *file)
1864 struct dentry *dentry = file->f_path.dentry;
1865 int killsuid = should_remove_suid(dentry);
1866 int killpriv = security_inode_need_killpriv(dentry);
1872 error = security_inode_killpriv(dentry);
1873 if (!error && killsuid)
1874 error = __remove_suid(dentry, killsuid);
1878 EXPORT_SYMBOL(file_remove_suid);
1880 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1881 const struct iovec *iov, size_t base, size_t bytes)
1883 size_t copied = 0, left = 0;
1886 char __user *buf = iov->iov_base + base;
1887 int copy = min(bytes, iov->iov_len - base);
1890 left = __copy_from_user_inatomic(vaddr, buf, copy);
1899 return copied - left;
1903 * Copy as much as we can into the page and return the number of bytes which
1904 * were sucessfully copied. If a fault is encountered then return the number of
1905 * bytes which were copied.
1907 size_t iov_iter_copy_from_user_atomic(struct page *page,
1908 struct iov_iter *i, unsigned long offset, size_t bytes)
1913 BUG_ON(!in_atomic());
1914 kaddr = kmap_atomic(page, KM_USER0);
1915 if (likely(i->nr_segs == 1)) {
1917 char __user *buf = i->iov->iov_base + i->iov_offset;
1918 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1919 copied = bytes - left;
1921 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1922 i->iov, i->iov_offset, bytes);
1924 kunmap_atomic(kaddr, KM_USER0);
1928 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1931 * This has the same sideeffects and return value as
1932 * iov_iter_copy_from_user_atomic().
1933 * The difference is that it attempts to resolve faults.
1934 * Page must not be locked.
1936 size_t iov_iter_copy_from_user(struct page *page,
1937 struct iov_iter *i, unsigned long offset, size_t bytes)
1943 if (likely(i->nr_segs == 1)) {
1945 char __user *buf = i->iov->iov_base + i->iov_offset;
1946 left = __copy_from_user(kaddr + offset, buf, bytes);
1947 copied = bytes - left;
1949 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1950 i->iov, i->iov_offset, bytes);
1955 EXPORT_SYMBOL(iov_iter_copy_from_user);
1957 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1959 BUG_ON(i->count < bytes);
1961 if (likely(i->nr_segs == 1)) {
1962 i->iov_offset += bytes;
1965 const struct iovec *iov = i->iov;
1966 size_t base = i->iov_offset;
1969 * The !iov->iov_len check ensures we skip over unlikely
1970 * zero-length segments (without overruning the iovec).
1972 while (bytes || unlikely(i->count && !iov->iov_len)) {
1975 copy = min(bytes, iov->iov_len - base);
1976 BUG_ON(!i->count || i->count < copy);
1980 if (iov->iov_len == base) {
1986 i->iov_offset = base;
1989 EXPORT_SYMBOL(iov_iter_advance);
1992 * Fault in the first iovec of the given iov_iter, to a maximum length
1993 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1994 * accessed (ie. because it is an invalid address).
1996 * writev-intensive code may want this to prefault several iovecs -- that
1997 * would be possible (callers must not rely on the fact that _only_ the
1998 * first iovec will be faulted with the current implementation).
2000 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2002 char __user *buf = i->iov->iov_base + i->iov_offset;
2003 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2004 return fault_in_pages_readable(buf, bytes);
2006 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2009 * Return the count of just the current iov_iter segment.
2011 size_t iov_iter_single_seg_count(struct iov_iter *i)
2013 const struct iovec *iov = i->iov;
2014 if (i->nr_segs == 1)
2017 return min(i->count, iov->iov_len - i->iov_offset);
2019 EXPORT_SYMBOL(iov_iter_single_seg_count);
2022 * Performs necessary checks before doing a write
2024 * Can adjust writing position or amount of bytes to write.
2025 * Returns appropriate error code that caller should return or
2026 * zero in case that write should be allowed.
2028 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2030 struct inode *inode = file->f_mapping->host;
2031 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2033 if (unlikely(*pos < 0))
2037 /* FIXME: this is for backwards compatibility with 2.4 */
2038 if (file->f_flags & O_APPEND)
2039 *pos = i_size_read(inode);
2041 if (limit != RLIM_INFINITY) {
2042 if (*pos >= limit) {
2043 send_sig(SIGXFSZ, current, 0);
2046 if (*count > limit - (typeof(limit))*pos) {
2047 *count = limit - (typeof(limit))*pos;
2055 if (unlikely(*pos + *count > MAX_NON_LFS &&
2056 !(file->f_flags & O_LARGEFILE))) {
2057 if (*pos >= MAX_NON_LFS) {
2060 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2061 *count = MAX_NON_LFS - (unsigned long)*pos;
2066 * Are we about to exceed the fs block limit ?
2068 * If we have written data it becomes a short write. If we have
2069 * exceeded without writing data we send a signal and return EFBIG.
2070 * Linus frestrict idea will clean these up nicely..
2072 if (likely(!isblk)) {
2073 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2074 if (*count || *pos > inode->i_sb->s_maxbytes) {
2077 /* zero-length writes at ->s_maxbytes are OK */
2080 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2081 *count = inode->i_sb->s_maxbytes - *pos;
2085 if (bdev_read_only(I_BDEV(inode)))
2087 isize = i_size_read(inode);
2088 if (*pos >= isize) {
2089 if (*count || *pos > isize)
2093 if (*pos + *count > isize)
2094 *count = isize - *pos;
2101 EXPORT_SYMBOL(generic_write_checks);
2103 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2104 loff_t pos, unsigned len, unsigned flags,
2105 struct page **pagep, void **fsdata)
2107 const struct address_space_operations *aops = mapping->a_ops;
2109 return aops->write_begin(file, mapping, pos, len, flags,
2112 EXPORT_SYMBOL(pagecache_write_begin);
2114 int pagecache_write_end(struct file *file, struct address_space *mapping,
2115 loff_t pos, unsigned len, unsigned copied,
2116 struct page *page, void *fsdata)
2118 const struct address_space_operations *aops = mapping->a_ops;
2120 mark_page_accessed(page);
2121 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2123 EXPORT_SYMBOL(pagecache_write_end);
2126 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2127 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2128 size_t count, size_t ocount)
2130 struct file *file = iocb->ki_filp;
2131 struct address_space *mapping = file->f_mapping;
2132 struct inode *inode = mapping->host;
2137 if (count != ocount)
2138 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2140 write_len = iov_length(iov, *nr_segs);
2141 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2143 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2148 * After a write we want buffered reads to be sure to go to disk to get
2149 * the new data. We invalidate clean cached page from the region we're
2150 * about to write. We do this *before* the write so that we can return
2151 * without clobbering -EIOCBQUEUED from ->direct_IO().
2153 if (mapping->nrpages) {
2154 written = invalidate_inode_pages2_range(mapping,
2155 pos >> PAGE_CACHE_SHIFT, end);
2157 * If a page can not be invalidated, return 0 to fall back
2158 * to buffered write.
2161 if (written == -EBUSY)
2167 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2170 * Finally, try again to invalidate clean pages which might have been
2171 * cached by non-direct readahead, or faulted in by get_user_pages()
2172 * if the source of the write was an mmap'ed region of the file
2173 * we're writing. Either one is a pretty crazy thing to do,
2174 * so we don't support it 100%. If this invalidation
2175 * fails, tough, the write still worked...
2177 if (mapping->nrpages) {
2178 invalidate_inode_pages2_range(mapping,
2179 pos >> PAGE_CACHE_SHIFT, end);
2183 loff_t end = pos + written;
2184 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2185 i_size_write(inode, end);
2186 mark_inode_dirty(inode);
2192 * Sync the fs metadata but not the minor inode changes and
2193 * of course not the data as we did direct DMA for the IO.
2194 * i_mutex is held, which protects generic_osync_inode() from
2195 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
2198 if ((written >= 0 || written == -EIOCBQUEUED) &&
2199 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2200 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2206 EXPORT_SYMBOL(generic_file_direct_write);
2209 * Find or create a page at the given pagecache position. Return the locked
2210 * page. This function is specifically for buffered writes.
2212 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2213 pgoff_t index, unsigned flags)
2217 gfp_t gfp_notmask = 0;
2218 if (flags & AOP_FLAG_NOFS)
2219 gfp_notmask = __GFP_FS;
2221 page = find_lock_page(mapping, index);
2225 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2228 status = add_to_page_cache_lru(page, mapping, index,
2229 GFP_KERNEL & ~gfp_notmask);
2230 if (unlikely(status)) {
2231 page_cache_release(page);
2232 if (status == -EEXIST)
2238 EXPORT_SYMBOL(grab_cache_page_write_begin);
2240 static ssize_t generic_perform_write(struct file *file,
2241 struct iov_iter *i, loff_t pos)
2243 struct address_space *mapping = file->f_mapping;
2244 const struct address_space_operations *a_ops = mapping->a_ops;
2246 ssize_t written = 0;
2247 unsigned int flags = 0;
2250 * Copies from kernel address space cannot fail (NFSD is a big user).
2252 if (segment_eq(get_fs(), KERNEL_DS))
2253 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2257 pgoff_t index; /* Pagecache index for current page */
2258 unsigned long offset; /* Offset into pagecache page */
2259 unsigned long bytes; /* Bytes to write to page */
2260 size_t copied; /* Bytes copied from user */
2263 offset = (pos & (PAGE_CACHE_SIZE - 1));
2264 index = pos >> PAGE_CACHE_SHIFT;
2265 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2271 * Bring in the user page that we will copy from _first_.
2272 * Otherwise there's a nasty deadlock on copying from the
2273 * same page as we're writing to, without it being marked
2276 * Not only is this an optimisation, but it is also required
2277 * to check that the address is actually valid, when atomic
2278 * usercopies are used, below.
2280 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2285 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2287 if (unlikely(status))
2290 pagefault_disable();
2291 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2293 flush_dcache_page(page);
2295 mark_page_accessed(page);
2296 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2298 if (unlikely(status < 0))
2304 iov_iter_advance(i, copied);
2305 if (unlikely(copied == 0)) {
2307 * If we were unable to copy any data at all, we must
2308 * fall back to a single segment length write.
2310 * If we didn't fallback here, we could livelock
2311 * because not all segments in the iov can be copied at
2312 * once without a pagefault.
2314 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2315 iov_iter_single_seg_count(i));
2321 balance_dirty_pages_ratelimited(mapping);
2323 } while (iov_iter_count(i));
2325 return written ? written : status;
2329 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2330 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2331 size_t count, ssize_t written)
2333 struct file *file = iocb->ki_filp;
2334 struct address_space *mapping = file->f_mapping;
2335 const struct address_space_operations *a_ops = mapping->a_ops;
2336 struct inode *inode = mapping->host;
2340 iov_iter_init(&i, iov, nr_segs, count, written);
2341 status = generic_perform_write(file, &i, pos);
2343 if (likely(status >= 0)) {
2345 *ppos = pos + status;
2348 * For now, when the user asks for O_SYNC, we'll actually give
2351 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2352 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2353 status = generic_osync_inode(inode, mapping,
2354 OSYNC_METADATA|OSYNC_DATA);
2359 * If we get here for O_DIRECT writes then we must have fallen through
2360 * to buffered writes (block instantiation inside i_size). So we sync
2361 * the file data here, to try to honour O_DIRECT expectations.
2363 if (unlikely(file->f_flags & O_DIRECT) && written)
2364 status = filemap_write_and_wait_range(mapping,
2365 pos, pos + written - 1);
2367 return written ? written : status;
2369 EXPORT_SYMBOL(generic_file_buffered_write);
2372 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2373 unsigned long nr_segs, loff_t *ppos)
2375 struct file *file = iocb->ki_filp;
2376 struct address_space * mapping = file->f_mapping;
2377 size_t ocount; /* original count */
2378 size_t count; /* after file limit checks */
2379 struct inode *inode = mapping->host;
2385 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2392 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2394 /* We can write back this queue in page reclaim */
2395 current->backing_dev_info = mapping->backing_dev_info;
2398 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2405 err = file_remove_suid(file);
2409 file_update_time(file);
2411 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2412 if (unlikely(file->f_flags & O_DIRECT)) {
2414 ssize_t written_buffered;
2416 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2417 ppos, count, ocount);
2418 if (written < 0 || written == count)
2421 * direct-io write to a hole: fall through to buffered I/O
2422 * for completing the rest of the request.
2426 written_buffered = generic_file_buffered_write(iocb, iov,
2427 nr_segs, pos, ppos, count,
2430 * If generic_file_buffered_write() retuned a synchronous error
2431 * then we want to return the number of bytes which were
2432 * direct-written, or the error code if that was zero. Note
2433 * that this differs from normal direct-io semantics, which
2434 * will return -EFOO even if some bytes were written.
2436 if (written_buffered < 0) {
2437 err = written_buffered;
2442 * We need to ensure that the page cache pages are written to
2443 * disk and invalidated to preserve the expected O_DIRECT
2446 endbyte = pos + written_buffered - written - 1;
2447 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2448 SYNC_FILE_RANGE_WAIT_BEFORE|
2449 SYNC_FILE_RANGE_WRITE|
2450 SYNC_FILE_RANGE_WAIT_AFTER);
2452 written = written_buffered;
2453 invalidate_mapping_pages(mapping,
2454 pos >> PAGE_CACHE_SHIFT,
2455 endbyte >> PAGE_CACHE_SHIFT);
2458 * We don't know how much we wrote, so just return
2459 * the number of bytes which were direct-written
2463 written = generic_file_buffered_write(iocb, iov, nr_segs,
2464 pos, ppos, count, written);
2467 current->backing_dev_info = NULL;
2468 return written ? written : err;
2471 ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2472 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
2474 struct file *file = iocb->ki_filp;
2475 struct address_space *mapping = file->f_mapping;
2476 struct inode *inode = mapping->host;
2479 BUG_ON(iocb->ki_pos != pos);
2481 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2484 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2487 err = sync_page_range_nolock(inode, mapping, pos, ret);
2493 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2495 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2496 unsigned long nr_segs, loff_t pos)
2498 struct file *file = iocb->ki_filp;
2499 struct address_space *mapping = file->f_mapping;
2500 struct inode *inode = mapping->host;
2503 BUG_ON(iocb->ki_pos != pos);
2505 mutex_lock(&inode->i_mutex);
2506 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2508 mutex_unlock(&inode->i_mutex);
2510 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2513 err = sync_page_range(inode, mapping, pos, ret);
2519 EXPORT_SYMBOL(generic_file_aio_write);
2522 * try_to_release_page() - release old fs-specific metadata on a page
2524 * @page: the page which the kernel is trying to free
2525 * @gfp_mask: memory allocation flags (and I/O mode)
2527 * The address_space is to try to release any data against the page
2528 * (presumably at page->private). If the release was successful, return `1'.
2529 * Otherwise return zero.
2531 * This may also be called if PG_fscache is set on a page, indicating that the
2532 * page is known to the local caching routines.
2534 * The @gfp_mask argument specifies whether I/O may be performed to release
2535 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2538 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2540 struct address_space * const mapping = page->mapping;
2542 BUG_ON(!PageLocked(page));
2543 if (PageWriteback(page))
2546 if (mapping && mapping->a_ops->releasepage)
2547 return mapping->a_ops->releasepage(page, gfp_mask);
2548 return try_to_free_buffers(page);
2551 EXPORT_SYMBOL(try_to_release_page);