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/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/aio.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/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 try_to_free_buffers */
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 (truncate_pagecache)
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)
105 * (code doesn't rely on that order, so you could switch it around)
106 * ->tasklist_lock (memory_failure, collect_procs_ao)
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 if (PageSwapBacked(page))
124 __dec_zone_page_state(page, NR_SHMEM);
125 BUG_ON(page_mapped(page));
128 * Some filesystems seem to re-dirty the page even after
129 * the VM has canceled the dirty bit (eg ext3 journaling).
131 * Fix it up by doing a final dirty accounting check after
132 * having removed the page entirely.
134 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
135 dec_zone_page_state(page, NR_FILE_DIRTY);
136 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
140 void remove_from_page_cache(struct page *page)
142 struct address_space *mapping = page->mapping;
143 void (*freepage)(struct page *);
145 BUG_ON(!PageLocked(page));
147 freepage = mapping->a_ops->freepage;
148 spin_lock_irq(&mapping->tree_lock);
149 __remove_from_page_cache(page);
150 spin_unlock_irq(&mapping->tree_lock);
151 mem_cgroup_uncharge_cache_page(page);
156 EXPORT_SYMBOL(remove_from_page_cache);
158 static int sync_page(void *word)
160 struct address_space *mapping;
163 page = container_of((unsigned long *)word, struct page, flags);
166 * page_mapping() is being called without PG_locked held.
167 * Some knowledge of the state and use of the page is used to
168 * reduce the requirements down to a memory barrier.
169 * The danger here is of a stale page_mapping() return value
170 * indicating a struct address_space different from the one it's
171 * associated with when it is associated with one.
172 * After smp_mb(), it's either the correct page_mapping() for
173 * the page, or an old page_mapping() and the page's own
174 * page_mapping() has gone NULL.
175 * The ->sync_page() address_space operation must tolerate
176 * page_mapping() going NULL. By an amazing coincidence,
177 * this comes about because none of the users of the page
178 * in the ->sync_page() methods make essential use of the
179 * page_mapping(), merely passing the page down to the backing
180 * device's unplug functions when it's non-NULL, which in turn
181 * ignore it for all cases but swap, where only page_private(page) is
182 * of interest. When page_mapping() does go NULL, the entire
183 * call stack gracefully ignores the page and returns.
187 mapping = page_mapping(page);
188 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
189 mapping->a_ops->sync_page(page);
194 static int sync_page_killable(void *word)
197 return fatal_signal_pending(current) ? -EINTR : 0;
201 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
202 * @mapping: address space structure to write
203 * @start: offset in bytes where the range starts
204 * @end: offset in bytes where the range ends (inclusive)
205 * @sync_mode: enable synchronous operation
207 * Start writeback against all of a mapping's dirty pages that lie
208 * within the byte offsets <start, end> inclusive.
210 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
211 * opposed to a regular memory cleansing writeback. The difference between
212 * these two operations is that if a dirty page/buffer is encountered, it must
213 * be waited upon, and not just skipped over.
215 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
216 loff_t end, int sync_mode)
219 struct writeback_control wbc = {
220 .sync_mode = sync_mode,
221 .nr_to_write = LONG_MAX,
222 .range_start = start,
226 if (!mapping_cap_writeback_dirty(mapping))
229 ret = do_writepages(mapping, &wbc);
233 static inline int __filemap_fdatawrite(struct address_space *mapping,
236 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
239 int filemap_fdatawrite(struct address_space *mapping)
241 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
243 EXPORT_SYMBOL(filemap_fdatawrite);
245 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
248 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
250 EXPORT_SYMBOL(filemap_fdatawrite_range);
253 * filemap_flush - mostly a non-blocking flush
254 * @mapping: target address_space
256 * This is a mostly non-blocking flush. Not suitable for data-integrity
257 * purposes - I/O may not be started against all dirty pages.
259 int filemap_flush(struct address_space *mapping)
261 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
263 EXPORT_SYMBOL(filemap_flush);
266 * filemap_fdatawait_range - wait for writeback to complete
267 * @mapping: address space structure to wait for
268 * @start_byte: offset in bytes where the range starts
269 * @end_byte: offset in bytes where the range ends (inclusive)
271 * Walk the list of under-writeback pages of the given address space
272 * in the given range and wait for all of them.
274 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
277 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
278 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
283 if (end_byte < start_byte)
286 pagevec_init(&pvec, 0);
287 while ((index <= end) &&
288 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
289 PAGECACHE_TAG_WRITEBACK,
290 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
293 for (i = 0; i < nr_pages; i++) {
294 struct page *page = pvec.pages[i];
296 /* until radix tree lookup accepts end_index */
297 if (page->index > end)
300 wait_on_page_writeback(page);
301 if (TestClearPageError(page))
304 pagevec_release(&pvec);
308 /* Check for outstanding write errors */
309 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
311 if (test_and_clear_bit(AS_EIO, &mapping->flags))
316 EXPORT_SYMBOL(filemap_fdatawait_range);
319 * filemap_fdatawait - wait for all under-writeback pages to complete
320 * @mapping: address space structure to wait for
322 * Walk the list of under-writeback pages of the given address space
323 * and wait for all of them.
325 int filemap_fdatawait(struct address_space *mapping)
327 loff_t i_size = i_size_read(mapping->host);
332 return filemap_fdatawait_range(mapping, 0, i_size - 1);
334 EXPORT_SYMBOL(filemap_fdatawait);
336 int filemap_write_and_wait(struct address_space *mapping)
340 if (mapping->nrpages) {
341 err = filemap_fdatawrite(mapping);
343 * Even if the above returned error, the pages may be
344 * written partially (e.g. -ENOSPC), so we wait for it.
345 * But the -EIO is special case, it may indicate the worst
346 * thing (e.g. bug) happened, so we avoid waiting for it.
349 int err2 = filemap_fdatawait(mapping);
356 EXPORT_SYMBOL(filemap_write_and_wait);
359 * filemap_write_and_wait_range - write out & wait on a file range
360 * @mapping: the address_space for the pages
361 * @lstart: offset in bytes where the range starts
362 * @lend: offset in bytes where the range ends (inclusive)
364 * Write out and wait upon file offsets lstart->lend, inclusive.
366 * Note that `lend' is inclusive (describes the last byte to be written) so
367 * that this function can be used to write to the very end-of-file (end = -1).
369 int filemap_write_and_wait_range(struct address_space *mapping,
370 loff_t lstart, loff_t lend)
374 if (mapping->nrpages) {
375 err = __filemap_fdatawrite_range(mapping, lstart, lend,
377 /* See comment of filemap_write_and_wait() */
379 int err2 = filemap_fdatawait_range(mapping,
387 EXPORT_SYMBOL(filemap_write_and_wait_range);
390 * add_to_page_cache_locked - add a locked page to the pagecache
392 * @mapping: the page's address_space
393 * @offset: page index
394 * @gfp_mask: page allocation mode
396 * This function is used to add a page to the pagecache. It must be locked.
397 * This function does not add the page to the LRU. The caller must do that.
399 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
400 pgoff_t offset, gfp_t gfp_mask)
404 VM_BUG_ON(!PageLocked(page));
406 error = mem_cgroup_cache_charge(page, current->mm,
407 gfp_mask & GFP_RECLAIM_MASK);
411 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
413 page_cache_get(page);
414 page->mapping = mapping;
415 page->index = offset;
417 spin_lock_irq(&mapping->tree_lock);
418 error = radix_tree_insert(&mapping->page_tree, offset, page);
419 if (likely(!error)) {
421 __inc_zone_page_state(page, NR_FILE_PAGES);
422 if (PageSwapBacked(page))
423 __inc_zone_page_state(page, NR_SHMEM);
424 spin_unlock_irq(&mapping->tree_lock);
426 page->mapping = NULL;
427 spin_unlock_irq(&mapping->tree_lock);
428 mem_cgroup_uncharge_cache_page(page);
429 page_cache_release(page);
431 radix_tree_preload_end();
433 mem_cgroup_uncharge_cache_page(page);
437 EXPORT_SYMBOL(add_to_page_cache_locked);
439 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
440 pgoff_t offset, gfp_t gfp_mask)
445 * Splice_read and readahead add shmem/tmpfs pages into the page cache
446 * before shmem_readpage has a chance to mark them as SwapBacked: they
447 * need to go on the anon lru below, and mem_cgroup_cache_charge
448 * (called in add_to_page_cache) needs to know where they're going too.
450 if (mapping_cap_swap_backed(mapping))
451 SetPageSwapBacked(page);
453 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
455 if (page_is_file_cache(page))
456 lru_cache_add_file(page);
458 lru_cache_add_anon(page);
462 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
465 struct page *__page_cache_alloc(gfp_t gfp)
470 if (cpuset_do_page_mem_spread()) {
472 n = cpuset_mem_spread_node();
473 page = alloc_pages_exact_node(n, gfp, 0);
477 return alloc_pages(gfp, 0);
479 EXPORT_SYMBOL(__page_cache_alloc);
482 static int __sleep_on_page_lock(void *word)
489 * In order to wait for pages to become available there must be
490 * waitqueues associated with pages. By using a hash table of
491 * waitqueues where the bucket discipline is to maintain all
492 * waiters on the same queue and wake all when any of the pages
493 * become available, and for the woken contexts to check to be
494 * sure the appropriate page became available, this saves space
495 * at a cost of "thundering herd" phenomena during rare hash
498 static wait_queue_head_t *page_waitqueue(struct page *page)
500 const struct zone *zone = page_zone(page);
502 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
505 static inline void wake_up_page(struct page *page, int bit)
507 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
510 void wait_on_page_bit(struct page *page, int bit_nr)
512 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
514 if (test_bit(bit_nr, &page->flags))
515 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
516 TASK_UNINTERRUPTIBLE);
518 EXPORT_SYMBOL(wait_on_page_bit);
521 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
522 * @page: Page defining the wait queue of interest
523 * @waiter: Waiter to add to the queue
525 * Add an arbitrary @waiter to the wait queue for the nominated @page.
527 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
529 wait_queue_head_t *q = page_waitqueue(page);
532 spin_lock_irqsave(&q->lock, flags);
533 __add_wait_queue(q, waiter);
534 spin_unlock_irqrestore(&q->lock, flags);
536 EXPORT_SYMBOL_GPL(add_page_wait_queue);
539 * unlock_page - unlock a locked page
542 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
543 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
544 * mechananism between PageLocked pages and PageWriteback pages is shared.
545 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
547 * The mb is necessary to enforce ordering between the clear_bit and the read
548 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
550 void unlock_page(struct page *page)
552 VM_BUG_ON(!PageLocked(page));
553 clear_bit_unlock(PG_locked, &page->flags);
554 smp_mb__after_clear_bit();
555 wake_up_page(page, PG_locked);
557 EXPORT_SYMBOL(unlock_page);
560 * end_page_writeback - end writeback against a page
563 void end_page_writeback(struct page *page)
565 if (TestClearPageReclaim(page))
566 rotate_reclaimable_page(page);
568 if (!test_clear_page_writeback(page))
571 smp_mb__after_clear_bit();
572 wake_up_page(page, PG_writeback);
574 EXPORT_SYMBOL(end_page_writeback);
577 * __lock_page - get a lock on the page, assuming we need to sleep to get it
578 * @page: the page to lock
580 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
581 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
582 * chances are that on the second loop, the block layer's plug list is empty,
583 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
585 void __lock_page(struct page *page)
587 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
589 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
590 TASK_UNINTERRUPTIBLE);
592 EXPORT_SYMBOL(__lock_page);
594 int __lock_page_killable(struct page *page)
596 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
598 return __wait_on_bit_lock(page_waitqueue(page), &wait,
599 sync_page_killable, TASK_KILLABLE);
601 EXPORT_SYMBOL_GPL(__lock_page_killable);
604 * __lock_page_nosync - get a lock on the page, without calling sync_page()
605 * @page: the page to lock
607 * Variant of lock_page that does not require the caller to hold a reference
608 * on the page's mapping.
610 void __lock_page_nosync(struct page *page)
612 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
613 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
614 TASK_UNINTERRUPTIBLE);
617 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
620 if (!(flags & FAULT_FLAG_ALLOW_RETRY)) {
624 up_read(&mm->mmap_sem);
625 wait_on_page_locked(page);
631 * find_get_page - find and get a page reference
632 * @mapping: the address_space to search
633 * @offset: the page index
635 * Is there a pagecache struct page at the given (mapping, offset) tuple?
636 * If yes, increment its refcount and return it; if no, return NULL.
638 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
646 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
648 page = radix_tree_deref_slot(pagep);
651 if (radix_tree_deref_retry(page))
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);
672 EXPORT_SYMBOL(find_get_page);
675 * find_lock_page - locate, pin and lock a pagecache page
676 * @mapping: the address_space to search
677 * @offset: the page index
679 * Locates the desired pagecache page, locks it, increments its reference
680 * count and returns its address.
682 * Returns zero if the page was not present. find_lock_page() may sleep.
684 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
689 page = find_get_page(mapping, offset);
692 /* Has the page been truncated? */
693 if (unlikely(page->mapping != mapping)) {
695 page_cache_release(page);
698 VM_BUG_ON(page->index != offset);
702 EXPORT_SYMBOL(find_lock_page);
705 * find_or_create_page - locate or add a pagecache page
706 * @mapping: the page's address_space
707 * @index: the page's index into the mapping
708 * @gfp_mask: page allocation mode
710 * Locates a page in the pagecache. If the page is not present, a new page
711 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
712 * LRU list. The returned page is locked and has its reference count
715 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
718 * find_or_create_page() returns the desired page's address, or zero on
721 struct page *find_or_create_page(struct address_space *mapping,
722 pgoff_t index, gfp_t gfp_mask)
727 page = find_lock_page(mapping, index);
729 page = __page_cache_alloc(gfp_mask);
733 * We want a regular kernel memory (not highmem or DMA etc)
734 * allocation for the radix tree nodes, but we need to honour
735 * the context-specific requirements the caller has asked for.
736 * GFP_RECLAIM_MASK collects those requirements.
738 err = add_to_page_cache_lru(page, mapping, index,
739 (gfp_mask & GFP_RECLAIM_MASK));
741 page_cache_release(page);
749 EXPORT_SYMBOL(find_or_create_page);
752 * find_get_pages - gang pagecache lookup
753 * @mapping: The address_space to search
754 * @start: The starting page index
755 * @nr_pages: The maximum number of pages
756 * @pages: Where the resulting pages are placed
758 * find_get_pages() will search for and return a group of up to
759 * @nr_pages pages in the mapping. The pages are placed at @pages.
760 * find_get_pages() takes a reference against the returned pages.
762 * The search returns a group of mapping-contiguous pages with ascending
763 * indexes. There may be holes in the indices due to not-present pages.
765 * find_get_pages() returns the number of pages which were found.
767 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
768 unsigned int nr_pages, struct page **pages)
772 unsigned int nr_found;
776 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
777 (void ***)pages, start, nr_pages);
779 for (i = 0; i < nr_found; i++) {
782 page = radix_tree_deref_slot((void **)pages[i]);
785 if (radix_tree_deref_retry(page)) {
787 start = pages[ret-1]->index;
791 if (!page_cache_get_speculative(page))
794 /* Has the page moved? */
795 if (unlikely(page != *((void **)pages[i]))) {
796 page_cache_release(page);
808 * find_get_pages_contig - gang contiguous pagecache lookup
809 * @mapping: The address_space to search
810 * @index: The starting page index
811 * @nr_pages: The maximum number of pages
812 * @pages: Where the resulting pages are placed
814 * find_get_pages_contig() works exactly like find_get_pages(), except
815 * that the returned number of pages are guaranteed to be contiguous.
817 * find_get_pages_contig() returns the number of pages which were found.
819 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
820 unsigned int nr_pages, struct page **pages)
824 unsigned int nr_found;
828 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
829 (void ***)pages, index, nr_pages);
831 for (i = 0; i < nr_found; i++) {
834 page = radix_tree_deref_slot((void **)pages[i]);
837 if (radix_tree_deref_retry(page))
840 if (!page_cache_get_speculative(page))
843 /* Has the page moved? */
844 if (unlikely(page != *((void **)pages[i]))) {
845 page_cache_release(page);
850 * must check mapping and index after taking the ref.
851 * otherwise we can get both false positives and false
852 * negatives, which is just confusing to the caller.
854 if (page->mapping == NULL || page->index != index) {
855 page_cache_release(page);
866 EXPORT_SYMBOL(find_get_pages_contig);
869 * find_get_pages_tag - find and return pages that match @tag
870 * @mapping: the address_space to search
871 * @index: the starting page index
872 * @tag: the tag index
873 * @nr_pages: the maximum number of pages
874 * @pages: where the resulting pages are placed
876 * Like find_get_pages, except we only return pages which are tagged with
877 * @tag. We update @index to index the next page for the traversal.
879 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
880 int tag, unsigned int nr_pages, struct page **pages)
884 unsigned int nr_found;
888 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
889 (void ***)pages, *index, nr_pages, tag);
891 for (i = 0; i < nr_found; i++) {
894 page = radix_tree_deref_slot((void **)pages[i]);
897 if (radix_tree_deref_retry(page))
900 if (!page_cache_get_speculative(page))
903 /* Has the page moved? */
904 if (unlikely(page != *((void **)pages[i]))) {
905 page_cache_release(page);
915 *index = pages[ret - 1]->index + 1;
919 EXPORT_SYMBOL(find_get_pages_tag);
922 * grab_cache_page_nowait - returns locked page at given index in given cache
923 * @mapping: target address_space
924 * @index: the page index
926 * Same as grab_cache_page(), but do not wait if the page is unavailable.
927 * This is intended for speculative data generators, where the data can
928 * be regenerated if the page couldn't be grabbed. This routine should
929 * be safe to call while holding the lock for another page.
931 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
932 * and deadlock against the caller's locked page.
935 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
937 struct page *page = find_get_page(mapping, index);
940 if (trylock_page(page))
942 page_cache_release(page);
945 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
946 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
947 page_cache_release(page);
952 EXPORT_SYMBOL(grab_cache_page_nowait);
955 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
956 * a _large_ part of the i/o request. Imagine the worst scenario:
958 * ---R__________________________________________B__________
959 * ^ reading here ^ bad block(assume 4k)
961 * read(R) => miss => readahead(R...B) => media error => frustrating retries
962 * => failing the whole request => read(R) => read(R+1) =>
963 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
964 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
965 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
967 * It is going insane. Fix it by quickly scaling down the readahead size.
969 static void shrink_readahead_size_eio(struct file *filp,
970 struct file_ra_state *ra)
976 * do_generic_file_read - generic file read routine
977 * @filp: the file to read
978 * @ppos: current file position
979 * @desc: read_descriptor
980 * @actor: read method
982 * This is a generic file read routine, and uses the
983 * mapping->a_ops->readpage() function for the actual low-level stuff.
985 * This is really ugly. But the goto's actually try to clarify some
986 * of the logic when it comes to error handling etc.
988 static void do_generic_file_read(struct file *filp, loff_t *ppos,
989 read_descriptor_t *desc, read_actor_t actor)
991 struct address_space *mapping = filp->f_mapping;
992 struct inode *inode = mapping->host;
993 struct file_ra_state *ra = &filp->f_ra;
997 unsigned long offset; /* offset into pagecache page */
998 unsigned int prev_offset;
1001 index = *ppos >> PAGE_CACHE_SHIFT;
1002 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1003 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1004 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1005 offset = *ppos & ~PAGE_CACHE_MASK;
1011 unsigned long nr, ret;
1015 page = find_get_page(mapping, index);
1017 page_cache_sync_readahead(mapping,
1019 index, last_index - index);
1020 page = find_get_page(mapping, index);
1021 if (unlikely(page == NULL))
1022 goto no_cached_page;
1024 if (PageReadahead(page)) {
1025 page_cache_async_readahead(mapping,
1027 index, last_index - index);
1029 if (!PageUptodate(page)) {
1030 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1031 !mapping->a_ops->is_partially_uptodate)
1032 goto page_not_up_to_date;
1033 if (!trylock_page(page))
1034 goto page_not_up_to_date;
1035 /* Did it get truncated before we got the lock? */
1037 goto page_not_up_to_date_locked;
1038 if (!mapping->a_ops->is_partially_uptodate(page,
1040 goto page_not_up_to_date_locked;
1045 * i_size must be checked after we know the page is Uptodate.
1047 * Checking i_size after the check allows us to calculate
1048 * the correct value for "nr", which means the zero-filled
1049 * part of the page is not copied back to userspace (unless
1050 * another truncate extends the file - this is desired though).
1053 isize = i_size_read(inode);
1054 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1055 if (unlikely(!isize || index > end_index)) {
1056 page_cache_release(page);
1060 /* nr is the maximum number of bytes to copy from this page */
1061 nr = PAGE_CACHE_SIZE;
1062 if (index == end_index) {
1063 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1065 page_cache_release(page);
1071 /* If users can be writing to this page using arbitrary
1072 * virtual addresses, take care about potential aliasing
1073 * before reading the page on the kernel side.
1075 if (mapping_writably_mapped(mapping))
1076 flush_dcache_page(page);
1079 * When a sequential read accesses a page several times,
1080 * only mark it as accessed the first time.
1082 if (prev_index != index || offset != prev_offset)
1083 mark_page_accessed(page);
1087 * Ok, we have the page, and it's up-to-date, so
1088 * now we can copy it to user space...
1090 * The actor routine returns how many bytes were actually used..
1091 * NOTE! This may not be the same as how much of a user buffer
1092 * we filled up (we may be padding etc), so we can only update
1093 * "pos" here (the actor routine has to update the user buffer
1094 * pointers and the remaining count).
1096 ret = actor(desc, page, offset, nr);
1098 index += offset >> PAGE_CACHE_SHIFT;
1099 offset &= ~PAGE_CACHE_MASK;
1100 prev_offset = offset;
1102 page_cache_release(page);
1103 if (ret == nr && desc->count)
1107 page_not_up_to_date:
1108 /* Get exclusive access to the page ... */
1109 error = lock_page_killable(page);
1110 if (unlikely(error))
1111 goto readpage_error;
1113 page_not_up_to_date_locked:
1114 /* Did it get truncated before we got the lock? */
1115 if (!page->mapping) {
1117 page_cache_release(page);
1121 /* Did somebody else fill it already? */
1122 if (PageUptodate(page)) {
1129 * A previous I/O error may have been due to temporary
1130 * failures, eg. multipath errors.
1131 * PG_error will be set again if readpage fails.
1133 ClearPageError(page);
1134 /* Start the actual read. The read will unlock the page. */
1135 error = mapping->a_ops->readpage(filp, page);
1137 if (unlikely(error)) {
1138 if (error == AOP_TRUNCATED_PAGE) {
1139 page_cache_release(page);
1142 goto readpage_error;
1145 if (!PageUptodate(page)) {
1146 error = lock_page_killable(page);
1147 if (unlikely(error))
1148 goto readpage_error;
1149 if (!PageUptodate(page)) {
1150 if (page->mapping == NULL) {
1152 * invalidate_mapping_pages got it
1155 page_cache_release(page);
1159 shrink_readahead_size_eio(filp, ra);
1161 goto readpage_error;
1169 /* UHHUH! A synchronous read error occurred. Report it */
1170 desc->error = error;
1171 page_cache_release(page);
1176 * Ok, it wasn't cached, so we need to create a new
1179 page = page_cache_alloc_cold(mapping);
1181 desc->error = -ENOMEM;
1184 error = add_to_page_cache_lru(page, mapping,
1187 page_cache_release(page);
1188 if (error == -EEXIST)
1190 desc->error = error;
1197 ra->prev_pos = prev_index;
1198 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1199 ra->prev_pos |= prev_offset;
1201 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1202 file_accessed(filp);
1205 int file_read_actor(read_descriptor_t *desc, struct page *page,
1206 unsigned long offset, unsigned long size)
1209 unsigned long left, count = desc->count;
1214 if (PageReadaheadUnused(page))
1215 ClearPageReadaheadUnused(page);
1218 * Faults on the destination of a read are common, so do it before
1221 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1222 kaddr = kmap_atomic(page, KM_USER0);
1223 left = __copy_to_user_inatomic(desc->arg.buf,
1224 kaddr + offset, size);
1225 kunmap_atomic(kaddr, KM_USER0);
1230 /* Do it the slow way */
1232 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1237 desc->error = -EFAULT;
1240 desc->count = count - size;
1241 desc->written += size;
1242 desc->arg.buf += size;
1247 * Performs necessary checks before doing a write
1248 * @iov: io vector request
1249 * @nr_segs: number of segments in the iovec
1250 * @count: number of bytes to write
1251 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1253 * Adjust number of segments and amount of bytes to write (nr_segs should be
1254 * properly initialized first). Returns appropriate error code that caller
1255 * should return or zero in case that write should be allowed.
1257 int generic_segment_checks(const struct iovec *iov,
1258 unsigned long *nr_segs, size_t *count, int access_flags)
1262 for (seg = 0; seg < *nr_segs; seg++) {
1263 const struct iovec *iv = &iov[seg];
1266 * If any segment has a negative length, or the cumulative
1267 * length ever wraps negative then return -EINVAL.
1270 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1272 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1277 cnt -= iv->iov_len; /* This segment is no good */
1283 EXPORT_SYMBOL(generic_segment_checks);
1286 * generic_file_aio_read - generic filesystem read routine
1287 * @iocb: kernel I/O control block
1288 * @iov: io vector request
1289 * @nr_segs: number of segments in the iovec
1290 * @pos: current file position
1292 * This is the "read()" routine for all filesystems
1293 * that can use the page cache directly.
1296 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1297 unsigned long nr_segs, loff_t pos)
1299 struct file *filp = iocb->ki_filp;
1301 unsigned long seg = 0;
1303 loff_t *ppos = &iocb->ki_pos;
1306 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1310 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1311 if (filp->f_flags & O_DIRECT) {
1313 struct address_space *mapping;
1314 struct inode *inode;
1316 mapping = filp->f_mapping;
1317 inode = mapping->host;
1319 goto out; /* skip atime */
1320 size = i_size_read(inode);
1322 retval = filemap_write_and_wait_range(mapping, pos,
1323 pos + iov_length(iov, nr_segs) - 1);
1325 retval = mapping->a_ops->direct_IO(READ, iocb,
1329 *ppos = pos + retval;
1334 * Btrfs can have a short DIO read if we encounter
1335 * compressed extents, so if there was an error, or if
1336 * we've already read everything we wanted to, or if
1337 * there was a short read because we hit EOF, go ahead
1338 * and return. Otherwise fallthrough to buffered io for
1339 * the rest of the read.
1341 if (retval < 0 || !count || *ppos >= size) {
1342 file_accessed(filp);
1349 for (seg = 0; seg < nr_segs; seg++) {
1350 read_descriptor_t desc;
1354 * If we did a short DIO read we need to skip the section of the
1355 * iov that we've already read data into.
1358 if (count > iov[seg].iov_len) {
1359 count -= iov[seg].iov_len;
1367 desc.arg.buf = iov[seg].iov_base + offset;
1368 desc.count = iov[seg].iov_len - offset;
1369 if (desc.count == 0)
1372 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1373 retval += desc.written;
1375 retval = retval ?: desc.error;
1384 EXPORT_SYMBOL(generic_file_aio_read);
1387 do_readahead(struct address_space *mapping, struct file *filp,
1388 pgoff_t index, unsigned long nr)
1390 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1393 force_page_cache_readahead(mapping, filp, index, nr);
1397 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1405 if (file->f_mode & FMODE_READ) {
1406 struct address_space *mapping = file->f_mapping;
1407 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1408 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1409 unsigned long len = end - start + 1;
1410 ret = do_readahead(mapping, file, start, len);
1416 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1417 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1419 return SYSC_readahead((int) fd, offset, (size_t) count);
1421 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1426 * page_cache_read - adds requested page to the page cache if not already there
1427 * @file: file to read
1428 * @offset: page index
1430 * This adds the requested page to the page cache if it isn't already there,
1431 * and schedules an I/O to read in its contents from disk.
1433 static int page_cache_read(struct file *file, pgoff_t offset)
1435 struct address_space *mapping = file->f_mapping;
1440 page = page_cache_alloc_cold(mapping);
1444 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1446 ret = mapping->a_ops->readpage(file, page);
1447 else if (ret == -EEXIST)
1448 ret = 0; /* losing race to add is OK */
1450 page_cache_release(page);
1452 } while (ret == AOP_TRUNCATED_PAGE);
1457 #define MMAP_LOTSAMISS (100)
1460 * Synchronous readahead happens when we don't even find
1461 * a page in the page cache at all.
1463 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1464 struct file_ra_state *ra,
1468 unsigned long ra_pages;
1469 struct address_space *mapping = file->f_mapping;
1471 /* If we don't want any read-ahead, don't bother */
1472 if (VM_RandomReadHint(vma))
1475 if (VM_SequentialReadHint(vma) ||
1476 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1477 page_cache_sync_readahead(mapping, ra, file, offset,
1482 if (ra->mmap_miss < INT_MAX)
1486 * Do we miss much more than hit in this file? If so,
1487 * stop bothering with read-ahead. It will only hurt.
1489 if (ra->mmap_miss > MMAP_LOTSAMISS)
1495 ra_pages = max_sane_readahead(ra->ra_pages);
1497 ra->start = max_t(long, 0, offset - ra_pages/2);
1498 ra->size = ra_pages;
1500 ra_submit(ra, mapping, file);
1505 * Asynchronous readahead happens when we find the page and PG_readahead,
1506 * so we want to possibly extend the readahead further..
1508 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1509 struct file_ra_state *ra,
1514 struct address_space *mapping = file->f_mapping;
1516 /* If we don't want any read-ahead, don't bother */
1517 if (VM_RandomReadHint(vma))
1519 if (ra->mmap_miss > 0)
1521 if (PageReadahead(page))
1522 page_cache_async_readahead(mapping, ra, file,
1523 page, offset, ra->ra_pages);
1527 * filemap_fault - read in file data for page fault handling
1528 * @vma: vma in which the fault was taken
1529 * @vmf: struct vm_fault containing details of the fault
1531 * filemap_fault() is invoked via the vma operations vector for a
1532 * mapped memory region to read in file data during a page fault.
1534 * The goto's are kind of ugly, but this streamlines the normal case of having
1535 * it in the page cache, and handles the special cases reasonably without
1536 * having a lot of duplicated code.
1538 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1541 struct file *file = vma->vm_file;
1542 struct address_space *mapping = file->f_mapping;
1543 struct file_ra_state *ra = &file->f_ra;
1544 struct inode *inode = mapping->host;
1545 pgoff_t offset = vmf->pgoff;
1550 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1552 return VM_FAULT_SIGBUS;
1555 * Do we have something in the page cache already?
1557 page = find_get_page(mapping, offset);
1560 * We found the page, so try async readahead before
1561 * waiting for the lock.
1563 do_async_mmap_readahead(vma, ra, file, page, offset);
1565 /* No page in the page cache at all */
1566 do_sync_mmap_readahead(vma, ra, file, offset);
1567 count_vm_event(PGMAJFAULT);
1568 ret = VM_FAULT_MAJOR;
1570 page = find_get_page(mapping, offset);
1572 goto no_cached_page;
1575 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1576 page_cache_release(page);
1577 return ret | VM_FAULT_RETRY;
1580 /* Did it get truncated? */
1581 if (unlikely(page->mapping != mapping)) {
1586 VM_BUG_ON(page->index != offset);
1589 * We have a locked page in the page cache, now we need to check
1590 * that it's up-to-date. If not, it is going to be due to an error.
1592 if (unlikely(!PageUptodate(page)))
1593 goto page_not_uptodate;
1596 * Found the page and have a reference on it.
1597 * We must recheck i_size under page lock.
1599 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1600 if (unlikely(offset >= size)) {
1602 page_cache_release(page);
1603 return VM_FAULT_SIGBUS;
1606 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1608 return ret | VM_FAULT_LOCKED;
1612 * We're only likely to ever get here if MADV_RANDOM is in
1615 error = page_cache_read(file, offset);
1618 * The page we want has now been added to the page cache.
1619 * In the unlikely event that someone removed it in the
1620 * meantime, we'll just come back here and read it again.
1626 * An error return from page_cache_read can result if the
1627 * system is low on memory, or a problem occurs while trying
1630 if (error == -ENOMEM)
1631 return VM_FAULT_OOM;
1632 return VM_FAULT_SIGBUS;
1636 * Umm, take care of errors if the page isn't up-to-date.
1637 * Try to re-read it _once_. We do this synchronously,
1638 * because there really aren't any performance issues here
1639 * and we need to check for errors.
1641 ClearPageError(page);
1642 error = mapping->a_ops->readpage(file, page);
1644 wait_on_page_locked(page);
1645 if (!PageUptodate(page))
1648 page_cache_release(page);
1650 if (!error || error == AOP_TRUNCATED_PAGE)
1653 /* Things didn't work out. Return zero to tell the mm layer so. */
1654 shrink_readahead_size_eio(file, ra);
1655 return VM_FAULT_SIGBUS;
1657 EXPORT_SYMBOL(filemap_fault);
1659 const struct vm_operations_struct generic_file_vm_ops = {
1660 .fault = filemap_fault,
1663 /* This is used for a general mmap of a disk file */
1665 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1667 struct address_space *mapping = file->f_mapping;
1669 if (!mapping->a_ops->readpage)
1671 file_accessed(file);
1672 vma->vm_ops = &generic_file_vm_ops;
1673 vma->vm_flags |= VM_CAN_NONLINEAR;
1678 * This is for filesystems which do not implement ->writepage.
1680 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1682 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1684 return generic_file_mmap(file, vma);
1687 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1691 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1695 #endif /* CONFIG_MMU */
1697 EXPORT_SYMBOL(generic_file_mmap);
1698 EXPORT_SYMBOL(generic_file_readonly_mmap);
1700 static struct page *__read_cache_page(struct address_space *mapping,
1702 int (*filler)(void *,struct page*),
1709 page = find_get_page(mapping, index);
1711 page = __page_cache_alloc(gfp | __GFP_COLD);
1713 return ERR_PTR(-ENOMEM);
1714 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1715 if (unlikely(err)) {
1716 page_cache_release(page);
1719 /* Presumably ENOMEM for radix tree node */
1720 return ERR_PTR(err);
1722 err = filler(data, page);
1724 page_cache_release(page);
1725 page = ERR_PTR(err);
1731 static struct page *do_read_cache_page(struct address_space *mapping,
1733 int (*filler)(void *,struct page*),
1742 page = __read_cache_page(mapping, index, filler, data, gfp);
1745 if (PageUptodate(page))
1749 if (!page->mapping) {
1751 page_cache_release(page);
1754 if (PageUptodate(page)) {
1758 err = filler(data, page);
1760 page_cache_release(page);
1761 return ERR_PTR(err);
1764 mark_page_accessed(page);
1769 * read_cache_page_async - read into page cache, fill it if needed
1770 * @mapping: the page's address_space
1771 * @index: the page index
1772 * @filler: function to perform the read
1773 * @data: destination for read data
1775 * Same as read_cache_page, but don't wait for page to become unlocked
1776 * after submitting it to the filler.
1778 * Read into the page cache. If a page already exists, and PageUptodate() is
1779 * not set, try to fill the page but don't wait for it to become unlocked.
1781 * If the page does not get brought uptodate, return -EIO.
1783 struct page *read_cache_page_async(struct address_space *mapping,
1785 int (*filler)(void *,struct page*),
1788 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1790 EXPORT_SYMBOL(read_cache_page_async);
1792 static struct page *wait_on_page_read(struct page *page)
1794 if (!IS_ERR(page)) {
1795 wait_on_page_locked(page);
1796 if (!PageUptodate(page)) {
1797 page_cache_release(page);
1798 page = ERR_PTR(-EIO);
1805 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1806 * @mapping: the page's address_space
1807 * @index: the page index
1808 * @gfp: the page allocator flags to use if allocating
1810 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1811 * any new page allocations done using the specified allocation flags. Note
1812 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1813 * expect to do this atomically or anything like that - but you can pass in
1814 * other page requirements.
1816 * If the page does not get brought uptodate, return -EIO.
1818 struct page *read_cache_page_gfp(struct address_space *mapping,
1822 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1824 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1826 EXPORT_SYMBOL(read_cache_page_gfp);
1829 * read_cache_page - read into page cache, fill it if needed
1830 * @mapping: the page's address_space
1831 * @index: the page index
1832 * @filler: function to perform the read
1833 * @data: destination for read data
1835 * Read into the page cache. If a page already exists, and PageUptodate() is
1836 * not set, try to fill the page then wait for it to become unlocked.
1838 * If the page does not get brought uptodate, return -EIO.
1840 struct page *read_cache_page(struct address_space *mapping,
1842 int (*filler)(void *,struct page*),
1845 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1847 EXPORT_SYMBOL(read_cache_page);
1850 * The logic we want is
1852 * if suid or (sgid and xgrp)
1855 int should_remove_suid(struct dentry *dentry)
1857 mode_t mode = dentry->d_inode->i_mode;
1860 /* suid always must be killed */
1861 if (unlikely(mode & S_ISUID))
1862 kill = ATTR_KILL_SUID;
1865 * sgid without any exec bits is just a mandatory locking mark; leave
1866 * it alone. If some exec bits are set, it's a real sgid; kill it.
1868 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1869 kill |= ATTR_KILL_SGID;
1871 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1876 EXPORT_SYMBOL(should_remove_suid);
1878 static int __remove_suid(struct dentry *dentry, int kill)
1880 struct iattr newattrs;
1882 newattrs.ia_valid = ATTR_FORCE | kill;
1883 return notify_change(dentry, &newattrs);
1886 int file_remove_suid(struct file *file)
1888 struct dentry *dentry = file->f_path.dentry;
1889 int killsuid = should_remove_suid(dentry);
1890 int killpriv = security_inode_need_killpriv(dentry);
1896 error = security_inode_killpriv(dentry);
1897 if (!error && killsuid)
1898 error = __remove_suid(dentry, killsuid);
1902 EXPORT_SYMBOL(file_remove_suid);
1904 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1905 const struct iovec *iov, size_t base, size_t bytes)
1907 size_t copied = 0, left = 0;
1910 char __user *buf = iov->iov_base + base;
1911 int copy = min(bytes, iov->iov_len - base);
1914 left = __copy_from_user_inatomic(vaddr, buf, copy);
1923 return copied - left;
1927 * Copy as much as we can into the page and return the number of bytes which
1928 * were successfully copied. If a fault is encountered then return the number of
1929 * bytes which were copied.
1931 size_t iov_iter_copy_from_user_atomic(struct page *page,
1932 struct iov_iter *i, unsigned long offset, size_t bytes)
1937 BUG_ON(!in_atomic());
1938 kaddr = kmap_atomic(page, KM_USER0);
1939 if (likely(i->nr_segs == 1)) {
1941 char __user *buf = i->iov->iov_base + i->iov_offset;
1942 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1943 copied = bytes - left;
1945 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1946 i->iov, i->iov_offset, bytes);
1948 kunmap_atomic(kaddr, KM_USER0);
1952 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1955 * This has the same sideeffects and return value as
1956 * iov_iter_copy_from_user_atomic().
1957 * The difference is that it attempts to resolve faults.
1958 * Page must not be locked.
1960 size_t iov_iter_copy_from_user(struct page *page,
1961 struct iov_iter *i, unsigned long offset, size_t bytes)
1967 if (likely(i->nr_segs == 1)) {
1969 char __user *buf = i->iov->iov_base + i->iov_offset;
1970 left = __copy_from_user(kaddr + offset, buf, bytes);
1971 copied = bytes - left;
1973 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1974 i->iov, i->iov_offset, bytes);
1979 EXPORT_SYMBOL(iov_iter_copy_from_user);
1981 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1983 BUG_ON(i->count < bytes);
1985 if (likely(i->nr_segs == 1)) {
1986 i->iov_offset += bytes;
1989 const struct iovec *iov = i->iov;
1990 size_t base = i->iov_offset;
1993 * The !iov->iov_len check ensures we skip over unlikely
1994 * zero-length segments (without overruning the iovec).
1996 while (bytes || unlikely(i->count && !iov->iov_len)) {
1999 copy = min(bytes, iov->iov_len - base);
2000 BUG_ON(!i->count || i->count < copy);
2004 if (iov->iov_len == base) {
2010 i->iov_offset = base;
2013 EXPORT_SYMBOL(iov_iter_advance);
2016 * Fault in the first iovec of the given iov_iter, to a maximum length
2017 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2018 * accessed (ie. because it is an invalid address).
2020 * writev-intensive code may want this to prefault several iovecs -- that
2021 * would be possible (callers must not rely on the fact that _only_ the
2022 * first iovec will be faulted with the current implementation).
2024 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2026 char __user *buf = i->iov->iov_base + i->iov_offset;
2027 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2028 return fault_in_pages_readable(buf, bytes);
2030 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2033 * Return the count of just the current iov_iter segment.
2035 size_t iov_iter_single_seg_count(struct iov_iter *i)
2037 const struct iovec *iov = i->iov;
2038 if (i->nr_segs == 1)
2041 return min(i->count, iov->iov_len - i->iov_offset);
2043 EXPORT_SYMBOL(iov_iter_single_seg_count);
2046 * Performs necessary checks before doing a write
2048 * Can adjust writing position or amount of bytes to write.
2049 * Returns appropriate error code that caller should return or
2050 * zero in case that write should be allowed.
2052 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2054 struct inode *inode = file->f_mapping->host;
2055 unsigned long limit = rlimit(RLIMIT_FSIZE);
2057 if (unlikely(*pos < 0))
2061 /* FIXME: this is for backwards compatibility with 2.4 */
2062 if (file->f_flags & O_APPEND)
2063 *pos = i_size_read(inode);
2065 if (limit != RLIM_INFINITY) {
2066 if (*pos >= limit) {
2067 send_sig(SIGXFSZ, current, 0);
2070 if (*count > limit - (typeof(limit))*pos) {
2071 *count = limit - (typeof(limit))*pos;
2079 if (unlikely(*pos + *count > MAX_NON_LFS &&
2080 !(file->f_flags & O_LARGEFILE))) {
2081 if (*pos >= MAX_NON_LFS) {
2084 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2085 *count = MAX_NON_LFS - (unsigned long)*pos;
2090 * Are we about to exceed the fs block limit ?
2092 * If we have written data it becomes a short write. If we have
2093 * exceeded without writing data we send a signal and return EFBIG.
2094 * Linus frestrict idea will clean these up nicely..
2096 if (likely(!isblk)) {
2097 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2098 if (*count || *pos > inode->i_sb->s_maxbytes) {
2101 /* zero-length writes at ->s_maxbytes are OK */
2104 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2105 *count = inode->i_sb->s_maxbytes - *pos;
2109 if (bdev_read_only(I_BDEV(inode)))
2111 isize = i_size_read(inode);
2112 if (*pos >= isize) {
2113 if (*count || *pos > isize)
2117 if (*pos + *count > isize)
2118 *count = isize - *pos;
2125 EXPORT_SYMBOL(generic_write_checks);
2127 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2128 loff_t pos, unsigned len, unsigned flags,
2129 struct page **pagep, void **fsdata)
2131 const struct address_space_operations *aops = mapping->a_ops;
2133 return aops->write_begin(file, mapping, pos, len, flags,
2136 EXPORT_SYMBOL(pagecache_write_begin);
2138 int pagecache_write_end(struct file *file, struct address_space *mapping,
2139 loff_t pos, unsigned len, unsigned copied,
2140 struct page *page, void *fsdata)
2142 const struct address_space_operations *aops = mapping->a_ops;
2144 mark_page_accessed(page);
2145 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2147 EXPORT_SYMBOL(pagecache_write_end);
2150 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2151 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2152 size_t count, size_t ocount)
2154 struct file *file = iocb->ki_filp;
2155 struct address_space *mapping = file->f_mapping;
2156 struct inode *inode = mapping->host;
2161 if (count != ocount)
2162 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2164 write_len = iov_length(iov, *nr_segs);
2165 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2167 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2172 * After a write we want buffered reads to be sure to go to disk to get
2173 * the new data. We invalidate clean cached page from the region we're
2174 * about to write. We do this *before* the write so that we can return
2175 * without clobbering -EIOCBQUEUED from ->direct_IO().
2177 if (mapping->nrpages) {
2178 written = invalidate_inode_pages2_range(mapping,
2179 pos >> PAGE_CACHE_SHIFT, end);
2181 * If a page can not be invalidated, return 0 to fall back
2182 * to buffered write.
2185 if (written == -EBUSY)
2191 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2194 * Finally, try again to invalidate clean pages which might have been
2195 * cached by non-direct readahead, or faulted in by get_user_pages()
2196 * if the source of the write was an mmap'ed region of the file
2197 * we're writing. Either one is a pretty crazy thing to do,
2198 * so we don't support it 100%. If this invalidation
2199 * fails, tough, the write still worked...
2201 if (mapping->nrpages) {
2202 invalidate_inode_pages2_range(mapping,
2203 pos >> PAGE_CACHE_SHIFT, end);
2208 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2209 i_size_write(inode, pos);
2210 mark_inode_dirty(inode);
2217 EXPORT_SYMBOL(generic_file_direct_write);
2220 * Find or create a page at the given pagecache position. Return the locked
2221 * page. This function is specifically for buffered writes.
2223 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2224 pgoff_t index, unsigned flags)
2228 gfp_t gfp_notmask = 0;
2229 if (flags & AOP_FLAG_NOFS)
2230 gfp_notmask = __GFP_FS;
2232 page = find_lock_page(mapping, index);
2236 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2239 status = add_to_page_cache_lru(page, mapping, index,
2240 GFP_KERNEL & ~gfp_notmask);
2241 if (unlikely(status)) {
2242 page_cache_release(page);
2243 if (status == -EEXIST)
2249 EXPORT_SYMBOL(grab_cache_page_write_begin);
2251 static ssize_t generic_perform_write(struct file *file,
2252 struct iov_iter *i, loff_t pos)
2254 struct address_space *mapping = file->f_mapping;
2255 const struct address_space_operations *a_ops = mapping->a_ops;
2257 ssize_t written = 0;
2258 unsigned int flags = 0;
2261 * Copies from kernel address space cannot fail (NFSD is a big user).
2263 if (segment_eq(get_fs(), KERNEL_DS))
2264 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2268 unsigned long offset; /* Offset into pagecache page */
2269 unsigned long bytes; /* Bytes to write to page */
2270 size_t copied; /* Bytes copied from user */
2273 offset = (pos & (PAGE_CACHE_SIZE - 1));
2274 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2280 * Bring in the user page that we will copy from _first_.
2281 * Otherwise there's a nasty deadlock on copying from the
2282 * same page as we're writing to, without it being marked
2285 * Not only is this an optimisation, but it is also required
2286 * to check that the address is actually valid, when atomic
2287 * usercopies are used, below.
2289 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2294 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2296 if (unlikely(status))
2299 if (mapping_writably_mapped(mapping))
2300 flush_dcache_page(page);
2302 pagefault_disable();
2303 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2305 flush_dcache_page(page);
2307 mark_page_accessed(page);
2308 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2310 if (unlikely(status < 0))
2316 iov_iter_advance(i, copied);
2317 if (unlikely(copied == 0)) {
2319 * If we were unable to copy any data at all, we must
2320 * fall back to a single segment length write.
2322 * If we didn't fallback here, we could livelock
2323 * because not all segments in the iov can be copied at
2324 * once without a pagefault.
2326 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2327 iov_iter_single_seg_count(i));
2333 balance_dirty_pages_ratelimited(mapping);
2335 } while (iov_iter_count(i));
2337 return written ? written : status;
2341 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2342 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2343 size_t count, ssize_t written)
2345 struct file *file = iocb->ki_filp;
2349 iov_iter_init(&i, iov, nr_segs, count, written);
2350 status = generic_perform_write(file, &i, pos);
2352 if (likely(status >= 0)) {
2354 *ppos = pos + status;
2357 return written ? written : status;
2359 EXPORT_SYMBOL(generic_file_buffered_write);
2362 * __generic_file_aio_write - write data to a file
2363 * @iocb: IO state structure (file, offset, etc.)
2364 * @iov: vector with data to write
2365 * @nr_segs: number of segments in the vector
2366 * @ppos: position where to write
2368 * This function does all the work needed for actually writing data to a
2369 * file. It does all basic checks, removes SUID from the file, updates
2370 * modification times and calls proper subroutines depending on whether we
2371 * do direct IO or a standard buffered write.
2373 * It expects i_mutex to be grabbed unless we work on a block device or similar
2374 * object which does not need locking at all.
2376 * This function does *not* take care of syncing data in case of O_SYNC write.
2377 * A caller has to handle it. This is mainly due to the fact that we want to
2378 * avoid syncing under i_mutex.
2380 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2381 unsigned long nr_segs, loff_t *ppos)
2383 struct file *file = iocb->ki_filp;
2384 struct address_space * mapping = file->f_mapping;
2385 size_t ocount; /* original count */
2386 size_t count; /* after file limit checks */
2387 struct inode *inode = mapping->host;
2393 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2400 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2402 /* We can write back this queue in page reclaim */
2403 current->backing_dev_info = mapping->backing_dev_info;
2406 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2413 err = file_remove_suid(file);
2417 file_update_time(file);
2419 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2420 if (unlikely(file->f_flags & O_DIRECT)) {
2422 ssize_t written_buffered;
2424 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2425 ppos, count, ocount);
2426 if (written < 0 || written == count)
2429 * direct-io write to a hole: fall through to buffered I/O
2430 * for completing the rest of the request.
2434 written_buffered = generic_file_buffered_write(iocb, iov,
2435 nr_segs, pos, ppos, count,
2438 * If generic_file_buffered_write() retuned a synchronous error
2439 * then we want to return the number of bytes which were
2440 * direct-written, or the error code if that was zero. Note
2441 * that this differs from normal direct-io semantics, which
2442 * will return -EFOO even if some bytes were written.
2444 if (written_buffered < 0) {
2445 err = written_buffered;
2450 * We need to ensure that the page cache pages are written to
2451 * disk and invalidated to preserve the expected O_DIRECT
2454 endbyte = pos + written_buffered - written - 1;
2455 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2457 written = written_buffered;
2458 invalidate_mapping_pages(mapping,
2459 pos >> PAGE_CACHE_SHIFT,
2460 endbyte >> PAGE_CACHE_SHIFT);
2463 * We don't know how much we wrote, so just return
2464 * the number of bytes which were direct-written
2468 written = generic_file_buffered_write(iocb, iov, nr_segs,
2469 pos, ppos, count, written);
2472 current->backing_dev_info = NULL;
2473 return written ? written : err;
2475 EXPORT_SYMBOL(__generic_file_aio_write);
2478 * generic_file_aio_write - write data to a file
2479 * @iocb: IO state structure
2480 * @iov: vector with data to write
2481 * @nr_segs: number of segments in the vector
2482 * @pos: position in file where to write
2484 * This is a wrapper around __generic_file_aio_write() to be used by most
2485 * filesystems. It takes care of syncing the file in case of O_SYNC file
2486 * and acquires i_mutex as needed.
2488 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2489 unsigned long nr_segs, loff_t pos)
2491 struct file *file = iocb->ki_filp;
2492 struct inode *inode = file->f_mapping->host;
2495 BUG_ON(iocb->ki_pos != pos);
2497 mutex_lock(&inode->i_mutex);
2498 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2499 mutex_unlock(&inode->i_mutex);
2501 if (ret > 0 || ret == -EIOCBQUEUED) {
2504 err = generic_write_sync(file, pos, ret);
2505 if (err < 0 && ret > 0)
2510 EXPORT_SYMBOL(generic_file_aio_write);
2513 * try_to_release_page() - release old fs-specific metadata on a page
2515 * @page: the page which the kernel is trying to free
2516 * @gfp_mask: memory allocation flags (and I/O mode)
2518 * The address_space is to try to release any data against the page
2519 * (presumably at page->private). If the release was successful, return `1'.
2520 * Otherwise return zero.
2522 * This may also be called if PG_fscache is set on a page, indicating that the
2523 * page is known to the local caching routines.
2525 * The @gfp_mask argument specifies whether I/O may be performed to release
2526 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2529 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2531 struct address_space * const mapping = page->mapping;
2533 BUG_ON(!PageLocked(page));
2534 if (PageWriteback(page))
2537 if (mapping && mapping->a_ops->releasepage)
2538 return mapping->a_ops->releasepage(page, gfp_mask);
2539 return try_to_free_buffers(page);
2542 EXPORT_SYMBOL(try_to_release_page);