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 if (!(flags & FAULT_FLAG_RETRY_NOWAIT)) {
625 up_read(&mm->mmap_sem);
626 wait_on_page_locked(page);
633 * find_get_page - find and get a page reference
634 * @mapping: the address_space to search
635 * @offset: the page index
637 * Is there a pagecache struct page at the given (mapping, offset) tuple?
638 * If yes, increment its refcount and return it; if no, return NULL.
640 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
648 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
650 page = radix_tree_deref_slot(pagep);
653 if (radix_tree_deref_retry(page))
656 if (!page_cache_get_speculative(page))
660 * Has the page moved?
661 * This is part of the lockless pagecache protocol. See
662 * include/linux/pagemap.h for details.
664 if (unlikely(page != *pagep)) {
665 page_cache_release(page);
674 EXPORT_SYMBOL(find_get_page);
677 * find_lock_page - locate, pin and lock a pagecache page
678 * @mapping: the address_space to search
679 * @offset: the page index
681 * Locates the desired pagecache page, locks it, increments its reference
682 * count and returns its address.
684 * Returns zero if the page was not present. find_lock_page() may sleep.
686 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
691 page = find_get_page(mapping, offset);
694 /* Has the page been truncated? */
695 if (unlikely(page->mapping != mapping)) {
697 page_cache_release(page);
700 VM_BUG_ON(page->index != offset);
704 EXPORT_SYMBOL(find_lock_page);
707 * find_or_create_page - locate or add a pagecache page
708 * @mapping: the page's address_space
709 * @index: the page's index into the mapping
710 * @gfp_mask: page allocation mode
712 * Locates a page in the pagecache. If the page is not present, a new page
713 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
714 * LRU list. The returned page is locked and has its reference count
717 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
720 * find_or_create_page() returns the desired page's address, or zero on
723 struct page *find_or_create_page(struct address_space *mapping,
724 pgoff_t index, gfp_t gfp_mask)
729 page = find_lock_page(mapping, index);
731 page = __page_cache_alloc(gfp_mask);
735 * We want a regular kernel memory (not highmem or DMA etc)
736 * allocation for the radix tree nodes, but we need to honour
737 * the context-specific requirements the caller has asked for.
738 * GFP_RECLAIM_MASK collects those requirements.
740 err = add_to_page_cache_lru(page, mapping, index,
741 (gfp_mask & GFP_RECLAIM_MASK));
743 page_cache_release(page);
751 EXPORT_SYMBOL(find_or_create_page);
754 * find_get_pages - gang pagecache lookup
755 * @mapping: The address_space to search
756 * @start: The starting page index
757 * @nr_pages: The maximum number of pages
758 * @pages: Where the resulting pages are placed
760 * find_get_pages() will search for and return a group of up to
761 * @nr_pages pages in the mapping. The pages are placed at @pages.
762 * find_get_pages() takes a reference against the returned pages.
764 * The search returns a group of mapping-contiguous pages with ascending
765 * indexes. There may be holes in the indices due to not-present pages.
767 * find_get_pages() returns the number of pages which were found.
769 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
770 unsigned int nr_pages, struct page **pages)
774 unsigned int nr_found;
778 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
779 (void ***)pages, start, nr_pages);
781 for (i = 0; i < nr_found; i++) {
784 page = radix_tree_deref_slot((void **)pages[i]);
787 if (radix_tree_deref_retry(page)) {
789 start = pages[ret-1]->index;
793 if (!page_cache_get_speculative(page))
796 /* Has the page moved? */
797 if (unlikely(page != *((void **)pages[i]))) {
798 page_cache_release(page);
810 * find_get_pages_contig - gang contiguous pagecache lookup
811 * @mapping: The address_space to search
812 * @index: The starting page index
813 * @nr_pages: The maximum number of pages
814 * @pages: Where the resulting pages are placed
816 * find_get_pages_contig() works exactly like find_get_pages(), except
817 * that the returned number of pages are guaranteed to be contiguous.
819 * find_get_pages_contig() returns the number of pages which were found.
821 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
822 unsigned int nr_pages, struct page **pages)
826 unsigned int nr_found;
830 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
831 (void ***)pages, index, nr_pages);
833 for (i = 0; i < nr_found; i++) {
836 page = radix_tree_deref_slot((void **)pages[i]);
839 if (radix_tree_deref_retry(page))
842 if (!page_cache_get_speculative(page))
845 /* Has the page moved? */
846 if (unlikely(page != *((void **)pages[i]))) {
847 page_cache_release(page);
852 * must check mapping and index after taking the ref.
853 * otherwise we can get both false positives and false
854 * negatives, which is just confusing to the caller.
856 if (page->mapping == NULL || page->index != index) {
857 page_cache_release(page);
868 EXPORT_SYMBOL(find_get_pages_contig);
871 * find_get_pages_tag - find and return pages that match @tag
872 * @mapping: the address_space to search
873 * @index: the starting page index
874 * @tag: the tag index
875 * @nr_pages: the maximum number of pages
876 * @pages: where the resulting pages are placed
878 * Like find_get_pages, except we only return pages which are tagged with
879 * @tag. We update @index to index the next page for the traversal.
881 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
882 int tag, unsigned int nr_pages, struct page **pages)
886 unsigned int nr_found;
890 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
891 (void ***)pages, *index, nr_pages, tag);
893 for (i = 0; i < nr_found; i++) {
896 page = radix_tree_deref_slot((void **)pages[i]);
899 if (radix_tree_deref_retry(page))
902 if (!page_cache_get_speculative(page))
905 /* Has the page moved? */
906 if (unlikely(page != *((void **)pages[i]))) {
907 page_cache_release(page);
917 *index = pages[ret - 1]->index + 1;
921 EXPORT_SYMBOL(find_get_pages_tag);
924 * grab_cache_page_nowait - returns locked page at given index in given cache
925 * @mapping: target address_space
926 * @index: the page index
928 * Same as grab_cache_page(), but do not wait if the page is unavailable.
929 * This is intended for speculative data generators, where the data can
930 * be regenerated if the page couldn't be grabbed. This routine should
931 * be safe to call while holding the lock for another page.
933 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
934 * and deadlock against the caller's locked page.
937 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
939 struct page *page = find_get_page(mapping, index);
942 if (trylock_page(page))
944 page_cache_release(page);
947 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
948 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
949 page_cache_release(page);
954 EXPORT_SYMBOL(grab_cache_page_nowait);
957 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
958 * a _large_ part of the i/o request. Imagine the worst scenario:
960 * ---R__________________________________________B__________
961 * ^ reading here ^ bad block(assume 4k)
963 * read(R) => miss => readahead(R...B) => media error => frustrating retries
964 * => failing the whole request => read(R) => read(R+1) =>
965 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
966 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
967 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
969 * It is going insane. Fix it by quickly scaling down the readahead size.
971 static void shrink_readahead_size_eio(struct file *filp,
972 struct file_ra_state *ra)
978 * do_generic_file_read - generic file read routine
979 * @filp: the file to read
980 * @ppos: current file position
981 * @desc: read_descriptor
982 * @actor: read method
984 * This is a generic file read routine, and uses the
985 * mapping->a_ops->readpage() function for the actual low-level stuff.
987 * This is really ugly. But the goto's actually try to clarify some
988 * of the logic when it comes to error handling etc.
990 static void do_generic_file_read(struct file *filp, loff_t *ppos,
991 read_descriptor_t *desc, read_actor_t actor)
993 struct address_space *mapping = filp->f_mapping;
994 struct inode *inode = mapping->host;
995 struct file_ra_state *ra = &filp->f_ra;
999 unsigned long offset; /* offset into pagecache page */
1000 unsigned int prev_offset;
1003 index = *ppos >> PAGE_CACHE_SHIFT;
1004 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1005 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1006 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1007 offset = *ppos & ~PAGE_CACHE_MASK;
1013 unsigned long nr, ret;
1017 page = find_get_page(mapping, index);
1019 page_cache_sync_readahead(mapping,
1021 index, last_index - index);
1022 page = find_get_page(mapping, index);
1023 if (unlikely(page == NULL))
1024 goto no_cached_page;
1026 if (PageReadahead(page)) {
1027 page_cache_async_readahead(mapping,
1029 index, last_index - index);
1031 if (!PageUptodate(page)) {
1032 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1033 !mapping->a_ops->is_partially_uptodate)
1034 goto page_not_up_to_date;
1035 if (!trylock_page(page))
1036 goto page_not_up_to_date;
1037 /* Did it get truncated before we got the lock? */
1039 goto page_not_up_to_date_locked;
1040 if (!mapping->a_ops->is_partially_uptodate(page,
1042 goto page_not_up_to_date_locked;
1047 * i_size must be checked after we know the page is Uptodate.
1049 * Checking i_size after the check allows us to calculate
1050 * the correct value for "nr", which means the zero-filled
1051 * part of the page is not copied back to userspace (unless
1052 * another truncate extends the file - this is desired though).
1055 isize = i_size_read(inode);
1056 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1057 if (unlikely(!isize || index > end_index)) {
1058 page_cache_release(page);
1062 /* nr is the maximum number of bytes to copy from this page */
1063 nr = PAGE_CACHE_SIZE;
1064 if (index == end_index) {
1065 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1067 page_cache_release(page);
1073 /* If users can be writing to this page using arbitrary
1074 * virtual addresses, take care about potential aliasing
1075 * before reading the page on the kernel side.
1077 if (mapping_writably_mapped(mapping))
1078 flush_dcache_page(page);
1081 * When a sequential read accesses a page several times,
1082 * only mark it as accessed the first time.
1084 if (prev_index != index || offset != prev_offset)
1085 mark_page_accessed(page);
1089 * Ok, we have the page, and it's up-to-date, so
1090 * now we can copy it to user space...
1092 * The actor routine returns how many bytes were actually used..
1093 * NOTE! This may not be the same as how much of a user buffer
1094 * we filled up (we may be padding etc), so we can only update
1095 * "pos" here (the actor routine has to update the user buffer
1096 * pointers and the remaining count).
1098 ret = actor(desc, page, offset, nr);
1100 index += offset >> PAGE_CACHE_SHIFT;
1101 offset &= ~PAGE_CACHE_MASK;
1102 prev_offset = offset;
1104 page_cache_release(page);
1105 if (ret == nr && desc->count)
1109 page_not_up_to_date:
1110 /* Get exclusive access to the page ... */
1111 error = lock_page_killable(page);
1112 if (unlikely(error))
1113 goto readpage_error;
1115 page_not_up_to_date_locked:
1116 /* Did it get truncated before we got the lock? */
1117 if (!page->mapping) {
1119 page_cache_release(page);
1123 /* Did somebody else fill it already? */
1124 if (PageUptodate(page)) {
1131 * A previous I/O error may have been due to temporary
1132 * failures, eg. multipath errors.
1133 * PG_error will be set again if readpage fails.
1135 ClearPageError(page);
1136 /* Start the actual read. The read will unlock the page. */
1137 error = mapping->a_ops->readpage(filp, page);
1139 if (unlikely(error)) {
1140 if (error == AOP_TRUNCATED_PAGE) {
1141 page_cache_release(page);
1144 goto readpage_error;
1147 if (!PageUptodate(page)) {
1148 error = lock_page_killable(page);
1149 if (unlikely(error))
1150 goto readpage_error;
1151 if (!PageUptodate(page)) {
1152 if (page->mapping == NULL) {
1154 * invalidate_mapping_pages got it
1157 page_cache_release(page);
1161 shrink_readahead_size_eio(filp, ra);
1163 goto readpage_error;
1171 /* UHHUH! A synchronous read error occurred. Report it */
1172 desc->error = error;
1173 page_cache_release(page);
1178 * Ok, it wasn't cached, so we need to create a new
1181 page = page_cache_alloc_cold(mapping);
1183 desc->error = -ENOMEM;
1186 error = add_to_page_cache_lru(page, mapping,
1189 page_cache_release(page);
1190 if (error == -EEXIST)
1192 desc->error = error;
1199 ra->prev_pos = prev_index;
1200 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1201 ra->prev_pos |= prev_offset;
1203 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1204 file_accessed(filp);
1207 int file_read_actor(read_descriptor_t *desc, struct page *page,
1208 unsigned long offset, unsigned long size)
1211 unsigned long left, count = desc->count;
1217 * Faults on the destination of a read are common, so do it before
1220 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1221 kaddr = kmap_atomic(page, KM_USER0);
1222 left = __copy_to_user_inatomic(desc->arg.buf,
1223 kaddr + offset, size);
1224 kunmap_atomic(kaddr, KM_USER0);
1229 /* Do it the slow way */
1231 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1236 desc->error = -EFAULT;
1239 desc->count = count - size;
1240 desc->written += size;
1241 desc->arg.buf += size;
1246 * Performs necessary checks before doing a write
1247 * @iov: io vector request
1248 * @nr_segs: number of segments in the iovec
1249 * @count: number of bytes to write
1250 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1252 * Adjust number of segments and amount of bytes to write (nr_segs should be
1253 * properly initialized first). Returns appropriate error code that caller
1254 * should return or zero in case that write should be allowed.
1256 int generic_segment_checks(const struct iovec *iov,
1257 unsigned long *nr_segs, size_t *count, int access_flags)
1261 for (seg = 0; seg < *nr_segs; seg++) {
1262 const struct iovec *iv = &iov[seg];
1265 * If any segment has a negative length, or the cumulative
1266 * length ever wraps negative then return -EINVAL.
1269 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1271 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1276 cnt -= iv->iov_len; /* This segment is no good */
1282 EXPORT_SYMBOL(generic_segment_checks);
1285 * generic_file_aio_read - generic filesystem read routine
1286 * @iocb: kernel I/O control block
1287 * @iov: io vector request
1288 * @nr_segs: number of segments in the iovec
1289 * @pos: current file position
1291 * This is the "read()" routine for all filesystems
1292 * that can use the page cache directly.
1295 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1296 unsigned long nr_segs, loff_t pos)
1298 struct file *filp = iocb->ki_filp;
1300 unsigned long seg = 0;
1302 loff_t *ppos = &iocb->ki_pos;
1305 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1309 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1310 if (filp->f_flags & O_DIRECT) {
1312 struct address_space *mapping;
1313 struct inode *inode;
1315 mapping = filp->f_mapping;
1316 inode = mapping->host;
1318 goto out; /* skip atime */
1319 size = i_size_read(inode);
1321 retval = filemap_write_and_wait_range(mapping, pos,
1322 pos + iov_length(iov, nr_segs) - 1);
1324 retval = mapping->a_ops->direct_IO(READ, iocb,
1328 *ppos = pos + retval;
1333 * Btrfs can have a short DIO read if we encounter
1334 * compressed extents, so if there was an error, or if
1335 * we've already read everything we wanted to, or if
1336 * there was a short read because we hit EOF, go ahead
1337 * and return. Otherwise fallthrough to buffered io for
1338 * the rest of the read.
1340 if (retval < 0 || !count || *ppos >= size) {
1341 file_accessed(filp);
1348 for (seg = 0; seg < nr_segs; seg++) {
1349 read_descriptor_t desc;
1353 * If we did a short DIO read we need to skip the section of the
1354 * iov that we've already read data into.
1357 if (count > iov[seg].iov_len) {
1358 count -= iov[seg].iov_len;
1366 desc.arg.buf = iov[seg].iov_base + offset;
1367 desc.count = iov[seg].iov_len - offset;
1368 if (desc.count == 0)
1371 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1372 retval += desc.written;
1374 retval = retval ?: desc.error;
1383 EXPORT_SYMBOL(generic_file_aio_read);
1386 do_readahead(struct address_space *mapping, struct file *filp,
1387 pgoff_t index, unsigned long nr)
1389 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1392 force_page_cache_readahead(mapping, filp, index, nr);
1396 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1404 if (file->f_mode & FMODE_READ) {
1405 struct address_space *mapping = file->f_mapping;
1406 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1407 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1408 unsigned long len = end - start + 1;
1409 ret = do_readahead(mapping, file, start, len);
1415 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1416 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1418 return SYSC_readahead((int) fd, offset, (size_t) count);
1420 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1425 * page_cache_read - adds requested page to the page cache if not already there
1426 * @file: file to read
1427 * @offset: page index
1429 * This adds the requested page to the page cache if it isn't already there,
1430 * and schedules an I/O to read in its contents from disk.
1432 static int page_cache_read(struct file *file, pgoff_t offset)
1434 struct address_space *mapping = file->f_mapping;
1439 page = page_cache_alloc_cold(mapping);
1443 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1445 ret = mapping->a_ops->readpage(file, page);
1446 else if (ret == -EEXIST)
1447 ret = 0; /* losing race to add is OK */
1449 page_cache_release(page);
1451 } while (ret == AOP_TRUNCATED_PAGE);
1456 #define MMAP_LOTSAMISS (100)
1459 * Synchronous readahead happens when we don't even find
1460 * a page in the page cache at all.
1462 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1463 struct file_ra_state *ra,
1467 unsigned long ra_pages;
1468 struct address_space *mapping = file->f_mapping;
1470 /* If we don't want any read-ahead, don't bother */
1471 if (VM_RandomReadHint(vma))
1474 if (VM_SequentialReadHint(vma) ||
1475 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1476 page_cache_sync_readahead(mapping, ra, file, offset,
1481 if (ra->mmap_miss < INT_MAX)
1485 * Do we miss much more than hit in this file? If so,
1486 * stop bothering with read-ahead. It will only hurt.
1488 if (ra->mmap_miss > MMAP_LOTSAMISS)
1494 ra_pages = max_sane_readahead(ra->ra_pages);
1496 ra->start = max_t(long, 0, offset - ra_pages/2);
1497 ra->size = ra_pages;
1499 ra_submit(ra, mapping, file);
1504 * Asynchronous readahead happens when we find the page and PG_readahead,
1505 * so we want to possibly extend the readahead further..
1507 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1508 struct file_ra_state *ra,
1513 struct address_space *mapping = file->f_mapping;
1515 /* If we don't want any read-ahead, don't bother */
1516 if (VM_RandomReadHint(vma))
1518 if (ra->mmap_miss > 0)
1520 if (PageReadahead(page))
1521 page_cache_async_readahead(mapping, ra, file,
1522 page, offset, ra->ra_pages);
1526 * filemap_fault - read in file data for page fault handling
1527 * @vma: vma in which the fault was taken
1528 * @vmf: struct vm_fault containing details of the fault
1530 * filemap_fault() is invoked via the vma operations vector for a
1531 * mapped memory region to read in file data during a page fault.
1533 * The goto's are kind of ugly, but this streamlines the normal case of having
1534 * it in the page cache, and handles the special cases reasonably without
1535 * having a lot of duplicated code.
1537 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1540 struct file *file = vma->vm_file;
1541 struct address_space *mapping = file->f_mapping;
1542 struct file_ra_state *ra = &file->f_ra;
1543 struct inode *inode = mapping->host;
1544 pgoff_t offset = vmf->pgoff;
1549 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1551 return VM_FAULT_SIGBUS;
1554 * Do we have something in the page cache already?
1556 page = find_get_page(mapping, offset);
1559 * We found the page, so try async readahead before
1560 * waiting for the lock.
1562 do_async_mmap_readahead(vma, ra, file, page, offset);
1564 /* No page in the page cache at all */
1565 do_sync_mmap_readahead(vma, ra, file, offset);
1566 count_vm_event(PGMAJFAULT);
1567 ret = VM_FAULT_MAJOR;
1569 page = find_get_page(mapping, offset);
1571 goto no_cached_page;
1574 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1575 page_cache_release(page);
1576 return ret | VM_FAULT_RETRY;
1579 /* Did it get truncated? */
1580 if (unlikely(page->mapping != mapping)) {
1585 VM_BUG_ON(page->index != offset);
1588 * We have a locked page in the page cache, now we need to check
1589 * that it's up-to-date. If not, it is going to be due to an error.
1591 if (unlikely(!PageUptodate(page)))
1592 goto page_not_uptodate;
1595 * Found the page and have a reference on it.
1596 * We must recheck i_size under page lock.
1598 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1599 if (unlikely(offset >= size)) {
1601 page_cache_release(page);
1602 return VM_FAULT_SIGBUS;
1605 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1607 return ret | VM_FAULT_LOCKED;
1611 * We're only likely to ever get here if MADV_RANDOM is in
1614 error = page_cache_read(file, offset);
1617 * The page we want has now been added to the page cache.
1618 * In the unlikely event that someone removed it in the
1619 * meantime, we'll just come back here and read it again.
1625 * An error return from page_cache_read can result if the
1626 * system is low on memory, or a problem occurs while trying
1629 if (error == -ENOMEM)
1630 return VM_FAULT_OOM;
1631 return VM_FAULT_SIGBUS;
1635 * Umm, take care of errors if the page isn't up-to-date.
1636 * Try to re-read it _once_. We do this synchronously,
1637 * because there really aren't any performance issues here
1638 * and we need to check for errors.
1640 ClearPageError(page);
1641 error = mapping->a_ops->readpage(file, page);
1643 wait_on_page_locked(page);
1644 if (!PageUptodate(page))
1647 page_cache_release(page);
1649 if (!error || error == AOP_TRUNCATED_PAGE)
1652 /* Things didn't work out. Return zero to tell the mm layer so. */
1653 shrink_readahead_size_eio(file, ra);
1654 return VM_FAULT_SIGBUS;
1656 EXPORT_SYMBOL(filemap_fault);
1658 const struct vm_operations_struct generic_file_vm_ops = {
1659 .fault = filemap_fault,
1662 /* This is used for a general mmap of a disk file */
1664 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1666 struct address_space *mapping = file->f_mapping;
1668 if (!mapping->a_ops->readpage)
1670 file_accessed(file);
1671 vma->vm_ops = &generic_file_vm_ops;
1672 vma->vm_flags |= VM_CAN_NONLINEAR;
1677 * This is for filesystems which do not implement ->writepage.
1679 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1681 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1683 return generic_file_mmap(file, vma);
1686 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1690 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1694 #endif /* CONFIG_MMU */
1696 EXPORT_SYMBOL(generic_file_mmap);
1697 EXPORT_SYMBOL(generic_file_readonly_mmap);
1699 static struct page *__read_cache_page(struct address_space *mapping,
1701 int (*filler)(void *,struct page*),
1708 page = find_get_page(mapping, index);
1710 page = __page_cache_alloc(gfp | __GFP_COLD);
1712 return ERR_PTR(-ENOMEM);
1713 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1714 if (unlikely(err)) {
1715 page_cache_release(page);
1718 /* Presumably ENOMEM for radix tree node */
1719 return ERR_PTR(err);
1721 err = filler(data, page);
1723 page_cache_release(page);
1724 page = ERR_PTR(err);
1730 static struct page *do_read_cache_page(struct address_space *mapping,
1732 int (*filler)(void *,struct page*),
1741 page = __read_cache_page(mapping, index, filler, data, gfp);
1744 if (PageUptodate(page))
1748 if (!page->mapping) {
1750 page_cache_release(page);
1753 if (PageUptodate(page)) {
1757 err = filler(data, page);
1759 page_cache_release(page);
1760 return ERR_PTR(err);
1763 mark_page_accessed(page);
1768 * read_cache_page_async - read into page cache, fill it if needed
1769 * @mapping: the page's address_space
1770 * @index: the page index
1771 * @filler: function to perform the read
1772 * @data: destination for read data
1774 * Same as read_cache_page, but don't wait for page to become unlocked
1775 * after submitting it to the filler.
1777 * Read into the page cache. If a page already exists, and PageUptodate() is
1778 * not set, try to fill the page but don't wait for it to become unlocked.
1780 * If the page does not get brought uptodate, return -EIO.
1782 struct page *read_cache_page_async(struct address_space *mapping,
1784 int (*filler)(void *,struct page*),
1787 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1789 EXPORT_SYMBOL(read_cache_page_async);
1791 static struct page *wait_on_page_read(struct page *page)
1793 if (!IS_ERR(page)) {
1794 wait_on_page_locked(page);
1795 if (!PageUptodate(page)) {
1796 page_cache_release(page);
1797 page = ERR_PTR(-EIO);
1804 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1805 * @mapping: the page's address_space
1806 * @index: the page index
1807 * @gfp: the page allocator flags to use if allocating
1809 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1810 * any new page allocations done using the specified allocation flags. Note
1811 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1812 * expect to do this atomically or anything like that - but you can pass in
1813 * other page requirements.
1815 * If the page does not get brought uptodate, return -EIO.
1817 struct page *read_cache_page_gfp(struct address_space *mapping,
1821 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1823 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1825 EXPORT_SYMBOL(read_cache_page_gfp);
1828 * read_cache_page - read into page cache, fill it if needed
1829 * @mapping: the page's address_space
1830 * @index: the page index
1831 * @filler: function to perform the read
1832 * @data: destination for read data
1834 * Read into the page cache. If a page already exists, and PageUptodate() is
1835 * not set, try to fill the page then wait for it to become unlocked.
1837 * If the page does not get brought uptodate, return -EIO.
1839 struct page *read_cache_page(struct address_space *mapping,
1841 int (*filler)(void *,struct page*),
1844 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1846 EXPORT_SYMBOL(read_cache_page);
1849 * The logic we want is
1851 * if suid or (sgid and xgrp)
1854 int should_remove_suid(struct dentry *dentry)
1856 mode_t mode = dentry->d_inode->i_mode;
1859 /* suid always must be killed */
1860 if (unlikely(mode & S_ISUID))
1861 kill = ATTR_KILL_SUID;
1864 * sgid without any exec bits is just a mandatory locking mark; leave
1865 * it alone. If some exec bits are set, it's a real sgid; kill it.
1867 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1868 kill |= ATTR_KILL_SGID;
1870 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1875 EXPORT_SYMBOL(should_remove_suid);
1877 static int __remove_suid(struct dentry *dentry, int kill)
1879 struct iattr newattrs;
1881 newattrs.ia_valid = ATTR_FORCE | kill;
1882 return notify_change(dentry, &newattrs);
1885 int file_remove_suid(struct file *file)
1887 struct dentry *dentry = file->f_path.dentry;
1888 int killsuid = should_remove_suid(dentry);
1889 int killpriv = security_inode_need_killpriv(dentry);
1895 error = security_inode_killpriv(dentry);
1896 if (!error && killsuid)
1897 error = __remove_suid(dentry, killsuid);
1901 EXPORT_SYMBOL(file_remove_suid);
1903 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1904 const struct iovec *iov, size_t base, size_t bytes)
1906 size_t copied = 0, left = 0;
1909 char __user *buf = iov->iov_base + base;
1910 int copy = min(bytes, iov->iov_len - base);
1913 left = __copy_from_user_inatomic(vaddr, buf, copy);
1922 return copied - left;
1926 * Copy as much as we can into the page and return the number of bytes which
1927 * were successfully copied. If a fault is encountered then return the number of
1928 * bytes which were copied.
1930 size_t iov_iter_copy_from_user_atomic(struct page *page,
1931 struct iov_iter *i, unsigned long offset, size_t bytes)
1936 BUG_ON(!in_atomic());
1937 kaddr = kmap_atomic(page, KM_USER0);
1938 if (likely(i->nr_segs == 1)) {
1940 char __user *buf = i->iov->iov_base + i->iov_offset;
1941 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1942 copied = bytes - left;
1944 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1945 i->iov, i->iov_offset, bytes);
1947 kunmap_atomic(kaddr, KM_USER0);
1951 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1954 * This has the same sideeffects and return value as
1955 * iov_iter_copy_from_user_atomic().
1956 * The difference is that it attempts to resolve faults.
1957 * Page must not be locked.
1959 size_t iov_iter_copy_from_user(struct page *page,
1960 struct iov_iter *i, unsigned long offset, size_t bytes)
1966 if (likely(i->nr_segs == 1)) {
1968 char __user *buf = i->iov->iov_base + i->iov_offset;
1969 left = __copy_from_user(kaddr + offset, buf, bytes);
1970 copied = bytes - left;
1972 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1973 i->iov, i->iov_offset, bytes);
1978 EXPORT_SYMBOL(iov_iter_copy_from_user);
1980 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1982 BUG_ON(i->count < bytes);
1984 if (likely(i->nr_segs == 1)) {
1985 i->iov_offset += bytes;
1988 const struct iovec *iov = i->iov;
1989 size_t base = i->iov_offset;
1992 * The !iov->iov_len check ensures we skip over unlikely
1993 * zero-length segments (without overruning the iovec).
1995 while (bytes || unlikely(i->count && !iov->iov_len)) {
1998 copy = min(bytes, iov->iov_len - base);
1999 BUG_ON(!i->count || i->count < copy);
2003 if (iov->iov_len == base) {
2009 i->iov_offset = base;
2012 EXPORT_SYMBOL(iov_iter_advance);
2015 * Fault in the first iovec of the given iov_iter, to a maximum length
2016 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2017 * accessed (ie. because it is an invalid address).
2019 * writev-intensive code may want this to prefault several iovecs -- that
2020 * would be possible (callers must not rely on the fact that _only_ the
2021 * first iovec will be faulted with the current implementation).
2023 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2025 char __user *buf = i->iov->iov_base + i->iov_offset;
2026 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2027 return fault_in_pages_readable(buf, bytes);
2029 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2032 * Return the count of just the current iov_iter segment.
2034 size_t iov_iter_single_seg_count(struct iov_iter *i)
2036 const struct iovec *iov = i->iov;
2037 if (i->nr_segs == 1)
2040 return min(i->count, iov->iov_len - i->iov_offset);
2042 EXPORT_SYMBOL(iov_iter_single_seg_count);
2045 * Performs necessary checks before doing a write
2047 * Can adjust writing position or amount of bytes to write.
2048 * Returns appropriate error code that caller should return or
2049 * zero in case that write should be allowed.
2051 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2053 struct inode *inode = file->f_mapping->host;
2054 unsigned long limit = rlimit(RLIMIT_FSIZE);
2056 if (unlikely(*pos < 0))
2060 /* FIXME: this is for backwards compatibility with 2.4 */
2061 if (file->f_flags & O_APPEND)
2062 *pos = i_size_read(inode);
2064 if (limit != RLIM_INFINITY) {
2065 if (*pos >= limit) {
2066 send_sig(SIGXFSZ, current, 0);
2069 if (*count > limit - (typeof(limit))*pos) {
2070 *count = limit - (typeof(limit))*pos;
2078 if (unlikely(*pos + *count > MAX_NON_LFS &&
2079 !(file->f_flags & O_LARGEFILE))) {
2080 if (*pos >= MAX_NON_LFS) {
2083 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2084 *count = MAX_NON_LFS - (unsigned long)*pos;
2089 * Are we about to exceed the fs block limit ?
2091 * If we have written data it becomes a short write. If we have
2092 * exceeded without writing data we send a signal and return EFBIG.
2093 * Linus frestrict idea will clean these up nicely..
2095 if (likely(!isblk)) {
2096 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2097 if (*count || *pos > inode->i_sb->s_maxbytes) {
2100 /* zero-length writes at ->s_maxbytes are OK */
2103 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2104 *count = inode->i_sb->s_maxbytes - *pos;
2108 if (bdev_read_only(I_BDEV(inode)))
2110 isize = i_size_read(inode);
2111 if (*pos >= isize) {
2112 if (*count || *pos > isize)
2116 if (*pos + *count > isize)
2117 *count = isize - *pos;
2124 EXPORT_SYMBOL(generic_write_checks);
2126 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2127 loff_t pos, unsigned len, unsigned flags,
2128 struct page **pagep, void **fsdata)
2130 const struct address_space_operations *aops = mapping->a_ops;
2132 return aops->write_begin(file, mapping, pos, len, flags,
2135 EXPORT_SYMBOL(pagecache_write_begin);
2137 int pagecache_write_end(struct file *file, struct address_space *mapping,
2138 loff_t pos, unsigned len, unsigned copied,
2139 struct page *page, void *fsdata)
2141 const struct address_space_operations *aops = mapping->a_ops;
2143 mark_page_accessed(page);
2144 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2146 EXPORT_SYMBOL(pagecache_write_end);
2149 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2150 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2151 size_t count, size_t ocount)
2153 struct file *file = iocb->ki_filp;
2154 struct address_space *mapping = file->f_mapping;
2155 struct inode *inode = mapping->host;
2160 if (count != ocount)
2161 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2163 write_len = iov_length(iov, *nr_segs);
2164 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2166 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2171 * After a write we want buffered reads to be sure to go to disk to get
2172 * the new data. We invalidate clean cached page from the region we're
2173 * about to write. We do this *before* the write so that we can return
2174 * without clobbering -EIOCBQUEUED from ->direct_IO().
2176 if (mapping->nrpages) {
2177 written = invalidate_inode_pages2_range(mapping,
2178 pos >> PAGE_CACHE_SHIFT, end);
2180 * If a page can not be invalidated, return 0 to fall back
2181 * to buffered write.
2184 if (written == -EBUSY)
2190 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2193 * Finally, try again to invalidate clean pages which might have been
2194 * cached by non-direct readahead, or faulted in by get_user_pages()
2195 * if the source of the write was an mmap'ed region of the file
2196 * we're writing. Either one is a pretty crazy thing to do,
2197 * so we don't support it 100%. If this invalidation
2198 * fails, tough, the write still worked...
2200 if (mapping->nrpages) {
2201 invalidate_inode_pages2_range(mapping,
2202 pos >> PAGE_CACHE_SHIFT, end);
2207 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2208 i_size_write(inode, pos);
2209 mark_inode_dirty(inode);
2216 EXPORT_SYMBOL(generic_file_direct_write);
2219 * Find or create a page at the given pagecache position. Return the locked
2220 * page. This function is specifically for buffered writes.
2222 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2223 pgoff_t index, unsigned flags)
2227 gfp_t gfp_notmask = 0;
2228 if (flags & AOP_FLAG_NOFS)
2229 gfp_notmask = __GFP_FS;
2231 page = find_lock_page(mapping, index);
2235 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2238 status = add_to_page_cache_lru(page, mapping, index,
2239 GFP_KERNEL & ~gfp_notmask);
2240 if (unlikely(status)) {
2241 page_cache_release(page);
2242 if (status == -EEXIST)
2248 EXPORT_SYMBOL(grab_cache_page_write_begin);
2250 static ssize_t generic_perform_write(struct file *file,
2251 struct iov_iter *i, loff_t pos)
2253 struct address_space *mapping = file->f_mapping;
2254 const struct address_space_operations *a_ops = mapping->a_ops;
2256 ssize_t written = 0;
2257 unsigned int flags = 0;
2260 * Copies from kernel address space cannot fail (NFSD is a big user).
2262 if (segment_eq(get_fs(), KERNEL_DS))
2263 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2267 unsigned long offset; /* Offset into pagecache page */
2268 unsigned long bytes; /* Bytes to write to page */
2269 size_t copied; /* Bytes copied from user */
2272 offset = (pos & (PAGE_CACHE_SIZE - 1));
2273 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2279 * Bring in the user page that we will copy from _first_.
2280 * Otherwise there's a nasty deadlock on copying from the
2281 * same page as we're writing to, without it being marked
2284 * Not only is this an optimisation, but it is also required
2285 * to check that the address is actually valid, when atomic
2286 * usercopies are used, below.
2288 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2293 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2295 if (unlikely(status))
2298 if (mapping_writably_mapped(mapping))
2299 flush_dcache_page(page);
2301 pagefault_disable();
2302 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2304 flush_dcache_page(page);
2306 mark_page_accessed(page);
2307 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2309 if (unlikely(status < 0))
2315 iov_iter_advance(i, copied);
2316 if (unlikely(copied == 0)) {
2318 * If we were unable to copy any data at all, we must
2319 * fall back to a single segment length write.
2321 * If we didn't fallback here, we could livelock
2322 * because not all segments in the iov can be copied at
2323 * once without a pagefault.
2325 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2326 iov_iter_single_seg_count(i));
2332 balance_dirty_pages_ratelimited(mapping);
2334 } while (iov_iter_count(i));
2336 return written ? written : status;
2340 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2341 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2342 size_t count, ssize_t written)
2344 struct file *file = iocb->ki_filp;
2348 iov_iter_init(&i, iov, nr_segs, count, written);
2349 status = generic_perform_write(file, &i, pos);
2351 if (likely(status >= 0)) {
2353 *ppos = pos + status;
2356 return written ? written : status;
2358 EXPORT_SYMBOL(generic_file_buffered_write);
2361 * __generic_file_aio_write - write data to a file
2362 * @iocb: IO state structure (file, offset, etc.)
2363 * @iov: vector with data to write
2364 * @nr_segs: number of segments in the vector
2365 * @ppos: position where to write
2367 * This function does all the work needed for actually writing data to a
2368 * file. It does all basic checks, removes SUID from the file, updates
2369 * modification times and calls proper subroutines depending on whether we
2370 * do direct IO or a standard buffered write.
2372 * It expects i_mutex to be grabbed unless we work on a block device or similar
2373 * object which does not need locking at all.
2375 * This function does *not* take care of syncing data in case of O_SYNC write.
2376 * A caller has to handle it. This is mainly due to the fact that we want to
2377 * avoid syncing under i_mutex.
2379 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2380 unsigned long nr_segs, loff_t *ppos)
2382 struct file *file = iocb->ki_filp;
2383 struct address_space * mapping = file->f_mapping;
2384 size_t ocount; /* original count */
2385 size_t count; /* after file limit checks */
2386 struct inode *inode = mapping->host;
2392 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2399 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2401 /* We can write back this queue in page reclaim */
2402 current->backing_dev_info = mapping->backing_dev_info;
2405 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2412 err = file_remove_suid(file);
2416 file_update_time(file);
2418 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2419 if (unlikely(file->f_flags & O_DIRECT)) {
2421 ssize_t written_buffered;
2423 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2424 ppos, count, ocount);
2425 if (written < 0 || written == count)
2428 * direct-io write to a hole: fall through to buffered I/O
2429 * for completing the rest of the request.
2433 written_buffered = generic_file_buffered_write(iocb, iov,
2434 nr_segs, pos, ppos, count,
2437 * If generic_file_buffered_write() retuned a synchronous error
2438 * then we want to return the number of bytes which were
2439 * direct-written, or the error code if that was zero. Note
2440 * that this differs from normal direct-io semantics, which
2441 * will return -EFOO even if some bytes were written.
2443 if (written_buffered < 0) {
2444 err = written_buffered;
2449 * We need to ensure that the page cache pages are written to
2450 * disk and invalidated to preserve the expected O_DIRECT
2453 endbyte = pos + written_buffered - written - 1;
2454 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2456 written = written_buffered;
2457 invalidate_mapping_pages(mapping,
2458 pos >> PAGE_CACHE_SHIFT,
2459 endbyte >> PAGE_CACHE_SHIFT);
2462 * We don't know how much we wrote, so just return
2463 * the number of bytes which were direct-written
2467 written = generic_file_buffered_write(iocb, iov, nr_segs,
2468 pos, ppos, count, written);
2471 current->backing_dev_info = NULL;
2472 return written ? written : err;
2474 EXPORT_SYMBOL(__generic_file_aio_write);
2477 * generic_file_aio_write - write data to a file
2478 * @iocb: IO state structure
2479 * @iov: vector with data to write
2480 * @nr_segs: number of segments in the vector
2481 * @pos: position in file where to write
2483 * This is a wrapper around __generic_file_aio_write() to be used by most
2484 * filesystems. It takes care of syncing the file in case of O_SYNC file
2485 * and acquires i_mutex as needed.
2487 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2488 unsigned long nr_segs, loff_t pos)
2490 struct file *file = iocb->ki_filp;
2491 struct inode *inode = file->f_mapping->host;
2494 BUG_ON(iocb->ki_pos != pos);
2496 mutex_lock(&inode->i_mutex);
2497 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2498 mutex_unlock(&inode->i_mutex);
2500 if (ret > 0 || ret == -EIOCBQUEUED) {
2503 err = generic_write_sync(file, pos, ret);
2504 if (err < 0 && ret > 0)
2509 EXPORT_SYMBOL(generic_file_aio_write);
2512 * try_to_release_page() - release old fs-specific metadata on a page
2514 * @page: the page which the kernel is trying to free
2515 * @gfp_mask: memory allocation flags (and I/O mode)
2517 * The address_space is to try to release any data against the page
2518 * (presumably at page->private). If the release was successful, return `1'.
2519 * Otherwise return zero.
2521 * This may also be called if PG_fscache is set on a page, indicating that the
2522 * page is known to the local caching routines.
2524 * The @gfp_mask argument specifies whether I/O may be performed to release
2525 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2528 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2530 struct address_space * const mapping = page->mapping;
2532 BUG_ON(!PageLocked(page));
2533 if (PageWriteback(page))
2536 if (mapping && mapping->a_ops->releasepage)
2537 return mapping->a_ops->releasepage(page, gfp_mask);
2538 return try_to_free_buffers(page);
2541 EXPORT_SYMBOL(try_to_release_page);