4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/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/cleancache.h>
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_mutex (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_mutex (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 * sb_lock (fs/fs-writeback.c)
82 * ->mapping->tree_lock (__sync_single_inode)
85 * ->anon_vma.lock (vma_adjust)
88 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
90 * ->page_table_lock or pte_lock
91 * ->swap_lock (try_to_unmap_one)
92 * ->private_lock (try_to_unmap_one)
93 * ->tree_lock (try_to_unmap_one)
94 * ->zone.lru_lock (follow_page->mark_page_accessed)
95 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
96 * ->private_lock (page_remove_rmap->set_page_dirty)
97 * ->tree_lock (page_remove_rmap->set_page_dirty)
98 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
99 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
100 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
101 * ->inode->i_lock (zap_pte_range->set_page_dirty)
102 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 * ->tasklist_lock (memory_failure, collect_procs_ao)
109 * Delete a page from the page cache and free it. Caller has to make
110 * sure the page is locked and that nobody else uses it - or that usage
111 * is safe. The caller must hold the mapping's tree_lock.
113 void __delete_from_page_cache(struct page *page)
115 struct address_space *mapping = page->mapping;
118 * if we're uptodate, flush out into the cleancache, otherwise
119 * invalidate any existing cleancache entries. We can't leave
120 * stale data around in the cleancache once our page is gone
122 if (PageUptodate(page) && PageMappedToDisk(page))
123 cleancache_put_page(page);
125 cleancache_invalidate_page(mapping, page);
127 radix_tree_delete(&mapping->page_tree, page->index);
128 page->mapping = NULL;
129 /* Leave page->index set: truncation lookup relies upon it */
131 __dec_zone_page_state(page, NR_FILE_PAGES);
132 if (PageSwapBacked(page))
133 __dec_zone_page_state(page, NR_SHMEM);
134 BUG_ON(page_mapped(page));
137 * Some filesystems seem to re-dirty the page even after
138 * the VM has canceled the dirty bit (eg ext3 journaling).
140 * Fix it up by doing a final dirty accounting check after
141 * having removed the page entirely.
143 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
144 dec_zone_page_state(page, NR_FILE_DIRTY);
145 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
150 * delete_from_page_cache - delete page from page cache
151 * @page: the page which the kernel is trying to remove from page cache
153 * This must be called only on pages that have been verified to be in the page
154 * cache and locked. It will never put the page into the free list, the caller
155 * has a reference on the page.
157 void delete_from_page_cache(struct page *page)
159 struct address_space *mapping = page->mapping;
160 void (*freepage)(struct page *);
162 BUG_ON(!PageLocked(page));
164 freepage = mapping->a_ops->freepage;
165 spin_lock_irq(&mapping->tree_lock);
166 __delete_from_page_cache(page);
167 spin_unlock_irq(&mapping->tree_lock);
168 mem_cgroup_uncharge_cache_page(page);
172 page_cache_release(page);
174 EXPORT_SYMBOL(delete_from_page_cache);
176 static int sleep_on_page(void *word)
182 static int sleep_on_page_killable(void *word)
185 return fatal_signal_pending(current) ? -EINTR : 0;
189 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
190 * @mapping: address space structure to write
191 * @start: offset in bytes where the range starts
192 * @end: offset in bytes where the range ends (inclusive)
193 * @sync_mode: enable synchronous operation
195 * Start writeback against all of a mapping's dirty pages that lie
196 * within the byte offsets <start, end> inclusive.
198 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
199 * opposed to a regular memory cleansing writeback. The difference between
200 * these two operations is that if a dirty page/buffer is encountered, it must
201 * be waited upon, and not just skipped over.
203 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
204 loff_t end, int sync_mode)
207 struct writeback_control wbc = {
208 .sync_mode = sync_mode,
209 .nr_to_write = LONG_MAX,
210 .range_start = start,
214 if (!mapping_cap_writeback_dirty(mapping))
217 ret = do_writepages(mapping, &wbc);
221 static inline int __filemap_fdatawrite(struct address_space *mapping,
224 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
227 int filemap_fdatawrite(struct address_space *mapping)
229 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
231 EXPORT_SYMBOL(filemap_fdatawrite);
233 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
236 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
238 EXPORT_SYMBOL(filemap_fdatawrite_range);
241 * filemap_flush - mostly a non-blocking flush
242 * @mapping: target address_space
244 * This is a mostly non-blocking flush. Not suitable for data-integrity
245 * purposes - I/O may not be started against all dirty pages.
247 int filemap_flush(struct address_space *mapping)
249 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
251 EXPORT_SYMBOL(filemap_flush);
254 * filemap_fdatawait_range - wait for writeback to complete
255 * @mapping: address space structure to wait for
256 * @start_byte: offset in bytes where the range starts
257 * @end_byte: offset in bytes where the range ends (inclusive)
259 * Walk the list of under-writeback pages of the given address space
260 * in the given range and wait for all of them.
262 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
265 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
266 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
271 if (end_byte < start_byte)
274 pagevec_init(&pvec, 0);
275 while ((index <= end) &&
276 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
277 PAGECACHE_TAG_WRITEBACK,
278 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
281 for (i = 0; i < nr_pages; i++) {
282 struct page *page = pvec.pages[i];
284 /* until radix tree lookup accepts end_index */
285 if (page->index > end)
288 wait_on_page_writeback(page);
289 if (TestClearPageError(page))
292 pagevec_release(&pvec);
296 /* Check for outstanding write errors */
297 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
299 if (test_and_clear_bit(AS_EIO, &mapping->flags))
304 EXPORT_SYMBOL(filemap_fdatawait_range);
307 * filemap_fdatawait - wait for all under-writeback pages to complete
308 * @mapping: address space structure to wait for
310 * Walk the list of under-writeback pages of the given address space
311 * and wait for all of them.
313 int filemap_fdatawait(struct address_space *mapping)
315 loff_t i_size = i_size_read(mapping->host);
320 return filemap_fdatawait_range(mapping, 0, i_size - 1);
322 EXPORT_SYMBOL(filemap_fdatawait);
324 int filemap_write_and_wait(struct address_space *mapping)
328 if (mapping->nrpages) {
329 err = filemap_fdatawrite(mapping);
331 * Even if the above returned error, the pages may be
332 * written partially (e.g. -ENOSPC), so we wait for it.
333 * But the -EIO is special case, it may indicate the worst
334 * thing (e.g. bug) happened, so we avoid waiting for it.
337 int err2 = filemap_fdatawait(mapping);
344 EXPORT_SYMBOL(filemap_write_and_wait);
347 * filemap_write_and_wait_range - write out & wait on a file range
348 * @mapping: the address_space for the pages
349 * @lstart: offset in bytes where the range starts
350 * @lend: offset in bytes where the range ends (inclusive)
352 * Write out and wait upon file offsets lstart->lend, inclusive.
354 * Note that `lend' is inclusive (describes the last byte to be written) so
355 * that this function can be used to write to the very end-of-file (end = -1).
357 int filemap_write_and_wait_range(struct address_space *mapping,
358 loff_t lstart, loff_t lend)
362 if (mapping->nrpages) {
363 err = __filemap_fdatawrite_range(mapping, lstart, lend,
365 /* See comment of filemap_write_and_wait() */
367 int err2 = filemap_fdatawait_range(mapping,
375 EXPORT_SYMBOL(filemap_write_and_wait_range);
378 * replace_page_cache_page - replace a pagecache page with a new one
379 * @old: page to be replaced
380 * @new: page to replace with
381 * @gfp_mask: allocation mode
383 * This function replaces a page in the pagecache with a new one. On
384 * success it acquires the pagecache reference for the new page and
385 * drops it for the old page. Both the old and new pages must be
386 * locked. This function does not add the new page to the LRU, the
387 * caller must do that.
389 * The remove + add is atomic. The only way this function can fail is
390 * memory allocation failure.
392 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
396 VM_BUG_ON(!PageLocked(old));
397 VM_BUG_ON(!PageLocked(new));
398 VM_BUG_ON(new->mapping);
400 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
402 struct address_space *mapping = old->mapping;
403 void (*freepage)(struct page *);
405 pgoff_t offset = old->index;
406 freepage = mapping->a_ops->freepage;
409 new->mapping = mapping;
412 spin_lock_irq(&mapping->tree_lock);
413 __delete_from_page_cache(old);
414 error = radix_tree_insert(&mapping->page_tree, offset, new);
417 __inc_zone_page_state(new, NR_FILE_PAGES);
418 if (PageSwapBacked(new))
419 __inc_zone_page_state(new, NR_SHMEM);
420 spin_unlock_irq(&mapping->tree_lock);
421 /* mem_cgroup codes must not be called under tree_lock */
422 mem_cgroup_replace_page_cache(old, new);
423 radix_tree_preload_end();
426 page_cache_release(old);
431 EXPORT_SYMBOL_GPL(replace_page_cache_page);
434 * add_to_page_cache_locked - add a locked page to the pagecache
436 * @mapping: the page's address_space
437 * @offset: page index
438 * @gfp_mask: page allocation mode
440 * This function is used to add a page to the pagecache. It must be locked.
441 * This function does not add the page to the LRU. The caller must do that.
443 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
444 pgoff_t offset, gfp_t gfp_mask)
448 VM_BUG_ON(!PageLocked(page));
449 VM_BUG_ON(PageSwapBacked(page));
451 error = mem_cgroup_cache_charge(page, current->mm,
452 gfp_mask & GFP_RECLAIM_MASK);
456 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
458 page_cache_get(page);
459 page->mapping = mapping;
460 page->index = offset;
462 spin_lock_irq(&mapping->tree_lock);
463 error = radix_tree_insert(&mapping->page_tree, offset, page);
464 if (likely(!error)) {
466 __inc_zone_page_state(page, NR_FILE_PAGES);
467 spin_unlock_irq(&mapping->tree_lock);
469 page->mapping = NULL;
470 /* Leave page->index set: truncation relies upon it */
471 spin_unlock_irq(&mapping->tree_lock);
472 mem_cgroup_uncharge_cache_page(page);
473 page_cache_release(page);
475 radix_tree_preload_end();
477 mem_cgroup_uncharge_cache_page(page);
481 EXPORT_SYMBOL(add_to_page_cache_locked);
483 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
484 pgoff_t offset, gfp_t gfp_mask)
488 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
490 lru_cache_add_file(page);
493 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
496 struct page *__page_cache_alloc(gfp_t gfp)
501 if (cpuset_do_page_mem_spread()) {
502 unsigned int cpuset_mems_cookie;
504 cpuset_mems_cookie = get_mems_allowed();
505 n = cpuset_mem_spread_node();
506 page = alloc_pages_exact_node(n, gfp, 0);
507 } while (!put_mems_allowed(cpuset_mems_cookie) && !page);
511 return alloc_pages(gfp, 0);
513 EXPORT_SYMBOL(__page_cache_alloc);
517 * In order to wait for pages to become available there must be
518 * waitqueues associated with pages. By using a hash table of
519 * waitqueues where the bucket discipline is to maintain all
520 * waiters on the same queue and wake all when any of the pages
521 * become available, and for the woken contexts to check to be
522 * sure the appropriate page became available, this saves space
523 * at a cost of "thundering herd" phenomena during rare hash
526 static wait_queue_head_t *page_waitqueue(struct page *page)
528 const struct zone *zone = page_zone(page);
530 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
533 static inline void wake_up_page(struct page *page, int bit)
535 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
538 void wait_on_page_bit(struct page *page, int bit_nr)
540 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
542 if (test_bit(bit_nr, &page->flags))
543 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
544 TASK_UNINTERRUPTIBLE);
546 EXPORT_SYMBOL(wait_on_page_bit);
548 int wait_on_page_bit_killable(struct page *page, int bit_nr)
550 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
552 if (!test_bit(bit_nr, &page->flags))
555 return __wait_on_bit(page_waitqueue(page), &wait,
556 sleep_on_page_killable, TASK_KILLABLE);
560 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
561 * @page: Page defining the wait queue of interest
562 * @waiter: Waiter to add to the queue
564 * Add an arbitrary @waiter to the wait queue for the nominated @page.
566 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
568 wait_queue_head_t *q = page_waitqueue(page);
571 spin_lock_irqsave(&q->lock, flags);
572 __add_wait_queue(q, waiter);
573 spin_unlock_irqrestore(&q->lock, flags);
575 EXPORT_SYMBOL_GPL(add_page_wait_queue);
578 * unlock_page - unlock a locked page
581 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
582 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
583 * mechananism between PageLocked pages and PageWriteback pages is shared.
584 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
586 * The mb is necessary to enforce ordering between the clear_bit and the read
587 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
589 void unlock_page(struct page *page)
591 VM_BUG_ON(!PageLocked(page));
592 clear_bit_unlock(PG_locked, &page->flags);
593 smp_mb__after_clear_bit();
594 wake_up_page(page, PG_locked);
596 EXPORT_SYMBOL(unlock_page);
599 * end_page_writeback - end writeback against a page
602 void end_page_writeback(struct page *page)
604 if (TestClearPageReclaim(page))
605 rotate_reclaimable_page(page);
607 if (!test_clear_page_writeback(page))
610 smp_mb__after_clear_bit();
611 wake_up_page(page, PG_writeback);
613 EXPORT_SYMBOL(end_page_writeback);
616 * __lock_page - get a lock on the page, assuming we need to sleep to get it
617 * @page: the page to lock
619 void __lock_page(struct page *page)
621 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
623 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
624 TASK_UNINTERRUPTIBLE);
626 EXPORT_SYMBOL(__lock_page);
628 int __lock_page_killable(struct page *page)
630 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
632 return __wait_on_bit_lock(page_waitqueue(page), &wait,
633 sleep_on_page_killable, TASK_KILLABLE);
635 EXPORT_SYMBOL_GPL(__lock_page_killable);
637 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
640 if (flags & FAULT_FLAG_ALLOW_RETRY) {
642 * CAUTION! In this case, mmap_sem is not released
643 * even though return 0.
645 if (flags & FAULT_FLAG_RETRY_NOWAIT)
648 up_read(&mm->mmap_sem);
649 if (flags & FAULT_FLAG_KILLABLE)
650 wait_on_page_locked_killable(page);
652 wait_on_page_locked(page);
655 if (flags & FAULT_FLAG_KILLABLE) {
658 ret = __lock_page_killable(page);
660 up_read(&mm->mmap_sem);
670 * find_get_page - find and get a page reference
671 * @mapping: the address_space to search
672 * @offset: the page index
674 * Is there a pagecache struct page at the given (mapping, offset) tuple?
675 * If yes, increment its refcount and return it; if no, return NULL.
677 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
685 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
687 page = radix_tree_deref_slot(pagep);
690 if (radix_tree_exception(page)) {
691 if (radix_tree_deref_retry(page))
694 * Otherwise, shmem/tmpfs must be storing a swap entry
695 * here as an exceptional entry: so return it without
696 * attempting to raise page count.
700 if (!page_cache_get_speculative(page))
704 * Has the page moved?
705 * This is part of the lockless pagecache protocol. See
706 * include/linux/pagemap.h for details.
708 if (unlikely(page != *pagep)) {
709 page_cache_release(page);
718 EXPORT_SYMBOL(find_get_page);
721 * find_lock_page - locate, pin and lock a pagecache page
722 * @mapping: the address_space to search
723 * @offset: the page index
725 * Locates the desired pagecache page, locks it, increments its reference
726 * count and returns its address.
728 * Returns zero if the page was not present. find_lock_page() may sleep.
730 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
735 page = find_get_page(mapping, offset);
736 if (page && !radix_tree_exception(page)) {
738 /* Has the page been truncated? */
739 if (unlikely(page->mapping != mapping)) {
741 page_cache_release(page);
744 VM_BUG_ON(page->index != offset);
748 EXPORT_SYMBOL(find_lock_page);
751 * find_or_create_page - locate or add a pagecache page
752 * @mapping: the page's address_space
753 * @index: the page's index into the mapping
754 * @gfp_mask: page allocation mode
756 * Locates a page in the pagecache. If the page is not present, a new page
757 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
758 * LRU list. The returned page is locked and has its reference count
761 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
764 * find_or_create_page() returns the desired page's address, or zero on
767 struct page *find_or_create_page(struct address_space *mapping,
768 pgoff_t index, gfp_t gfp_mask)
773 page = find_lock_page(mapping, index);
775 page = __page_cache_alloc(gfp_mask);
779 * We want a regular kernel memory (not highmem or DMA etc)
780 * allocation for the radix tree nodes, but we need to honour
781 * the context-specific requirements the caller has asked for.
782 * GFP_RECLAIM_MASK collects those requirements.
784 err = add_to_page_cache_lru(page, mapping, index,
785 (gfp_mask & GFP_RECLAIM_MASK));
787 page_cache_release(page);
795 EXPORT_SYMBOL(find_or_create_page);
798 * find_get_pages - gang pagecache lookup
799 * @mapping: The address_space to search
800 * @start: The starting page index
801 * @nr_pages: The maximum number of pages
802 * @pages: Where the resulting pages are placed
804 * find_get_pages() will search for and return a group of up to
805 * @nr_pages pages in the mapping. The pages are placed at @pages.
806 * find_get_pages() takes a reference against the returned pages.
808 * The search returns a group of mapping-contiguous pages with ascending
809 * indexes. There may be holes in the indices due to not-present pages.
811 * find_get_pages() returns the number of pages which were found.
813 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
814 unsigned int nr_pages, struct page **pages)
816 struct radix_tree_iter iter;
820 if (unlikely(!nr_pages))
825 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
828 page = radix_tree_deref_slot(slot);
832 if (radix_tree_exception(page)) {
833 if (radix_tree_deref_retry(page)) {
835 * Transient condition which can only trigger
836 * when entry at index 0 moves out of or back
837 * to root: none yet gotten, safe to restart.
843 * Otherwise, shmem/tmpfs must be storing a swap entry
844 * here as an exceptional entry: so skip over it -
845 * we only reach this from invalidate_mapping_pages().
850 if (!page_cache_get_speculative(page))
853 /* Has the page moved? */
854 if (unlikely(page != *slot)) {
855 page_cache_release(page);
860 if (++ret == nr_pages)
869 * find_get_pages_contig - gang contiguous pagecache lookup
870 * @mapping: The address_space to search
871 * @index: The starting page index
872 * @nr_pages: The maximum number of pages
873 * @pages: Where the resulting pages are placed
875 * find_get_pages_contig() works exactly like find_get_pages(), except
876 * that the returned number of pages are guaranteed to be contiguous.
878 * find_get_pages_contig() returns the number of pages which were found.
880 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
881 unsigned int nr_pages, struct page **pages)
883 struct radix_tree_iter iter;
885 unsigned int ret = 0;
887 if (unlikely(!nr_pages))
892 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
895 page = radix_tree_deref_slot(slot);
896 /* The hole, there no reason to continue */
900 if (radix_tree_exception(page)) {
901 if (radix_tree_deref_retry(page)) {
903 * Transient condition which can only trigger
904 * when entry at index 0 moves out of or back
905 * to root: none yet gotten, safe to restart.
910 * Otherwise, shmem/tmpfs must be storing a swap entry
911 * here as an exceptional entry: so stop looking for
917 if (!page_cache_get_speculative(page))
920 /* Has the page moved? */
921 if (unlikely(page != *slot)) {
922 page_cache_release(page);
927 * must check mapping and index after taking the ref.
928 * otherwise we can get both false positives and false
929 * negatives, which is just confusing to the caller.
931 if (page->mapping == NULL || page->index != iter.index) {
932 page_cache_release(page);
937 if (++ret == nr_pages)
943 EXPORT_SYMBOL(find_get_pages_contig);
946 * find_get_pages_tag - find and return pages that match @tag
947 * @mapping: the address_space to search
948 * @index: the starting page index
949 * @tag: the tag index
950 * @nr_pages: the maximum number of pages
951 * @pages: where the resulting pages are placed
953 * Like find_get_pages, except we only return pages which are tagged with
954 * @tag. We update @index to index the next page for the traversal.
956 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
957 int tag, unsigned int nr_pages, struct page **pages)
959 struct radix_tree_iter iter;
963 if (unlikely(!nr_pages))
968 radix_tree_for_each_tagged(slot, &mapping->page_tree,
969 &iter, *index, tag) {
972 page = radix_tree_deref_slot(slot);
976 if (radix_tree_exception(page)) {
977 if (radix_tree_deref_retry(page)) {
979 * Transient condition which can only trigger
980 * when entry at index 0 moves out of or back
981 * to root: none yet gotten, safe to restart.
986 * This function is never used on a shmem/tmpfs
987 * mapping, so a swap entry won't be found here.
992 if (!page_cache_get_speculative(page))
995 /* Has the page moved? */
996 if (unlikely(page != *slot)) {
997 page_cache_release(page);
1002 if (++ret == nr_pages)
1009 *index = pages[ret - 1]->index + 1;
1013 EXPORT_SYMBOL(find_get_pages_tag);
1016 * grab_cache_page_nowait - returns locked page at given index in given cache
1017 * @mapping: target address_space
1018 * @index: the page index
1020 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1021 * This is intended for speculative data generators, where the data can
1022 * be regenerated if the page couldn't be grabbed. This routine should
1023 * be safe to call while holding the lock for another page.
1025 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1026 * and deadlock against the caller's locked page.
1029 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1031 struct page *page = find_get_page(mapping, index);
1034 if (trylock_page(page))
1036 page_cache_release(page);
1039 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1040 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1041 page_cache_release(page);
1046 EXPORT_SYMBOL(grab_cache_page_nowait);
1049 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1050 * a _large_ part of the i/o request. Imagine the worst scenario:
1052 * ---R__________________________________________B__________
1053 * ^ reading here ^ bad block(assume 4k)
1055 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1056 * => failing the whole request => read(R) => read(R+1) =>
1057 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1058 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1059 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1061 * It is going insane. Fix it by quickly scaling down the readahead size.
1063 static void shrink_readahead_size_eio(struct file *filp,
1064 struct file_ra_state *ra)
1070 * do_generic_file_read - generic file read routine
1071 * @filp: the file to read
1072 * @ppos: current file position
1073 * @desc: read_descriptor
1074 * @actor: read method
1076 * This is a generic file read routine, and uses the
1077 * mapping->a_ops->readpage() function for the actual low-level stuff.
1079 * This is really ugly. But the goto's actually try to clarify some
1080 * of the logic when it comes to error handling etc.
1082 static void do_generic_file_read(struct file *filp, loff_t *ppos,
1083 read_descriptor_t *desc, read_actor_t actor)
1085 struct address_space *mapping = filp->f_mapping;
1086 struct inode *inode = mapping->host;
1087 struct file_ra_state *ra = &filp->f_ra;
1091 unsigned long offset; /* offset into pagecache page */
1092 unsigned int prev_offset;
1095 index = *ppos >> PAGE_CACHE_SHIFT;
1096 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1097 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1098 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1099 offset = *ppos & ~PAGE_CACHE_MASK;
1105 unsigned long nr, ret;
1109 page = find_get_page(mapping, index);
1111 page_cache_sync_readahead(mapping,
1113 index, last_index - index);
1114 page = find_get_page(mapping, index);
1115 if (unlikely(page == NULL))
1116 goto no_cached_page;
1118 if (PageReadahead(page)) {
1119 page_cache_async_readahead(mapping,
1121 index, last_index - index);
1123 if (!PageUptodate(page)) {
1124 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1125 !mapping->a_ops->is_partially_uptodate)
1126 goto page_not_up_to_date;
1127 if (!trylock_page(page))
1128 goto page_not_up_to_date;
1129 /* Did it get truncated before we got the lock? */
1131 goto page_not_up_to_date_locked;
1132 if (!mapping->a_ops->is_partially_uptodate(page,
1134 goto page_not_up_to_date_locked;
1139 * i_size must be checked after we know the page is Uptodate.
1141 * Checking i_size after the check allows us to calculate
1142 * the correct value for "nr", which means the zero-filled
1143 * part of the page is not copied back to userspace (unless
1144 * another truncate extends the file - this is desired though).
1147 isize = i_size_read(inode);
1148 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1149 if (unlikely(!isize || index > end_index)) {
1150 page_cache_release(page);
1154 /* nr is the maximum number of bytes to copy from this page */
1155 nr = PAGE_CACHE_SIZE;
1156 if (index == end_index) {
1157 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1159 page_cache_release(page);
1165 /* If users can be writing to this page using arbitrary
1166 * virtual addresses, take care about potential aliasing
1167 * before reading the page on the kernel side.
1169 if (mapping_writably_mapped(mapping))
1170 flush_dcache_page(page);
1173 * When a sequential read accesses a page several times,
1174 * only mark it as accessed the first time.
1176 if (prev_index != index || offset != prev_offset)
1177 mark_page_accessed(page);
1181 * Ok, we have the page, and it's up-to-date, so
1182 * now we can copy it to user space...
1184 * The actor routine returns how many bytes were actually used..
1185 * NOTE! This may not be the same as how much of a user buffer
1186 * we filled up (we may be padding etc), so we can only update
1187 * "pos" here (the actor routine has to update the user buffer
1188 * pointers and the remaining count).
1190 ret = actor(desc, page, offset, nr);
1192 index += offset >> PAGE_CACHE_SHIFT;
1193 offset &= ~PAGE_CACHE_MASK;
1194 prev_offset = offset;
1196 page_cache_release(page);
1197 if (ret == nr && desc->count)
1201 page_not_up_to_date:
1202 /* Get exclusive access to the page ... */
1203 error = lock_page_killable(page);
1204 if (unlikely(error))
1205 goto readpage_error;
1207 page_not_up_to_date_locked:
1208 /* Did it get truncated before we got the lock? */
1209 if (!page->mapping) {
1211 page_cache_release(page);
1215 /* Did somebody else fill it already? */
1216 if (PageUptodate(page)) {
1223 * A previous I/O error may have been due to temporary
1224 * failures, eg. multipath errors.
1225 * PG_error will be set again if readpage fails.
1227 ClearPageError(page);
1228 /* Start the actual read. The read will unlock the page. */
1229 error = mapping->a_ops->readpage(filp, page);
1231 if (unlikely(error)) {
1232 if (error == AOP_TRUNCATED_PAGE) {
1233 page_cache_release(page);
1236 goto readpage_error;
1239 if (!PageUptodate(page)) {
1240 error = lock_page_killable(page);
1241 if (unlikely(error))
1242 goto readpage_error;
1243 if (!PageUptodate(page)) {
1244 if (page->mapping == NULL) {
1246 * invalidate_mapping_pages got it
1249 page_cache_release(page);
1253 shrink_readahead_size_eio(filp, ra);
1255 goto readpage_error;
1263 /* UHHUH! A synchronous read error occurred. Report it */
1264 desc->error = error;
1265 page_cache_release(page);
1270 * Ok, it wasn't cached, so we need to create a new
1273 page = page_cache_alloc_cold(mapping);
1275 desc->error = -ENOMEM;
1278 error = add_to_page_cache_lru(page, mapping,
1281 page_cache_release(page);
1282 if (error == -EEXIST)
1284 desc->error = error;
1291 ra->prev_pos = prev_index;
1292 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1293 ra->prev_pos |= prev_offset;
1295 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1296 file_accessed(filp);
1299 int file_read_actor(read_descriptor_t *desc, struct page *page,
1300 unsigned long offset, unsigned long size)
1303 unsigned long left, count = desc->count;
1309 * Faults on the destination of a read are common, so do it before
1312 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1313 kaddr = kmap_atomic(page);
1314 left = __copy_to_user_inatomic(desc->arg.buf,
1315 kaddr + offset, size);
1316 kunmap_atomic(kaddr);
1321 /* Do it the slow way */
1323 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1328 desc->error = -EFAULT;
1331 desc->count = count - size;
1332 desc->written += size;
1333 desc->arg.buf += size;
1338 * Performs necessary checks before doing a write
1339 * @iov: io vector request
1340 * @nr_segs: number of segments in the iovec
1341 * @count: number of bytes to write
1342 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1344 * Adjust number of segments and amount of bytes to write (nr_segs should be
1345 * properly initialized first). Returns appropriate error code that caller
1346 * should return or zero in case that write should be allowed.
1348 int generic_segment_checks(const struct iovec *iov,
1349 unsigned long *nr_segs, size_t *count, int access_flags)
1353 for (seg = 0; seg < *nr_segs; seg++) {
1354 const struct iovec *iv = &iov[seg];
1357 * If any segment has a negative length, or the cumulative
1358 * length ever wraps negative then return -EINVAL.
1361 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1363 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1368 cnt -= iv->iov_len; /* This segment is no good */
1374 EXPORT_SYMBOL(generic_segment_checks);
1377 * generic_file_aio_read - generic filesystem read routine
1378 * @iocb: kernel I/O control block
1379 * @iov: io vector request
1380 * @nr_segs: number of segments in the iovec
1381 * @pos: current file position
1383 * This is the "read()" routine for all filesystems
1384 * that can use the page cache directly.
1387 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1388 unsigned long nr_segs, loff_t pos)
1390 struct file *filp = iocb->ki_filp;
1392 unsigned long seg = 0;
1394 loff_t *ppos = &iocb->ki_pos;
1397 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1401 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1402 if (filp->f_flags & O_DIRECT) {
1404 struct address_space *mapping;
1405 struct inode *inode;
1407 mapping = filp->f_mapping;
1408 inode = mapping->host;
1410 goto out; /* skip atime */
1411 size = i_size_read(inode);
1413 retval = filemap_write_and_wait_range(mapping, pos,
1414 pos + iov_length(iov, nr_segs) - 1);
1416 struct blk_plug plug;
1418 blk_start_plug(&plug);
1419 retval = mapping->a_ops->direct_IO(READ, iocb,
1421 blk_finish_plug(&plug);
1424 *ppos = pos + retval;
1429 * Btrfs can have a short DIO read if we encounter
1430 * compressed extents, so if there was an error, or if
1431 * we've already read everything we wanted to, or if
1432 * there was a short read because we hit EOF, go ahead
1433 * and return. Otherwise fallthrough to buffered io for
1434 * the rest of the read.
1436 if (retval < 0 || !count || *ppos >= size) {
1437 file_accessed(filp);
1444 for (seg = 0; seg < nr_segs; seg++) {
1445 read_descriptor_t desc;
1449 * If we did a short DIO read we need to skip the section of the
1450 * iov that we've already read data into.
1453 if (count > iov[seg].iov_len) {
1454 count -= iov[seg].iov_len;
1462 desc.arg.buf = iov[seg].iov_base + offset;
1463 desc.count = iov[seg].iov_len - offset;
1464 if (desc.count == 0)
1467 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1468 retval += desc.written;
1470 retval = retval ?: desc.error;
1479 EXPORT_SYMBOL(generic_file_aio_read);
1482 do_readahead(struct address_space *mapping, struct file *filp,
1483 pgoff_t index, unsigned long nr)
1485 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1488 force_page_cache_readahead(mapping, filp, index, nr);
1492 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1500 if (file->f_mode & FMODE_READ) {
1501 struct address_space *mapping = file->f_mapping;
1502 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1503 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1504 unsigned long len = end - start + 1;
1505 ret = do_readahead(mapping, file, start, len);
1511 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1512 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1514 return SYSC_readahead((int) fd, offset, (size_t) count);
1516 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1521 * page_cache_read - adds requested page to the page cache if not already there
1522 * @file: file to read
1523 * @offset: page index
1525 * This adds the requested page to the page cache if it isn't already there,
1526 * and schedules an I/O to read in its contents from disk.
1528 static int page_cache_read(struct file *file, pgoff_t offset)
1530 struct address_space *mapping = file->f_mapping;
1535 page = page_cache_alloc_cold(mapping);
1539 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1541 ret = mapping->a_ops->readpage(file, page);
1542 else if (ret == -EEXIST)
1543 ret = 0; /* losing race to add is OK */
1545 page_cache_release(page);
1547 } while (ret == AOP_TRUNCATED_PAGE);
1552 #define MMAP_LOTSAMISS (100)
1555 * Synchronous readahead happens when we don't even find
1556 * a page in the page cache at all.
1558 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1559 struct file_ra_state *ra,
1563 unsigned long ra_pages;
1564 struct address_space *mapping = file->f_mapping;
1566 /* If we don't want any read-ahead, don't bother */
1567 if (VM_RandomReadHint(vma))
1572 if (VM_SequentialReadHint(vma)) {
1573 page_cache_sync_readahead(mapping, ra, file, offset,
1578 /* Avoid banging the cache line if not needed */
1579 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1583 * Do we miss much more than hit in this file? If so,
1584 * stop bothering with read-ahead. It will only hurt.
1586 if (ra->mmap_miss > MMAP_LOTSAMISS)
1592 ra_pages = max_sane_readahead(ra->ra_pages);
1593 ra->start = max_t(long, 0, offset - ra_pages / 2);
1594 ra->size = ra_pages;
1595 ra->async_size = ra_pages / 4;
1596 ra_submit(ra, mapping, file);
1600 * Asynchronous readahead happens when we find the page and PG_readahead,
1601 * so we want to possibly extend the readahead further..
1603 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1604 struct file_ra_state *ra,
1609 struct address_space *mapping = file->f_mapping;
1611 /* If we don't want any read-ahead, don't bother */
1612 if (VM_RandomReadHint(vma))
1614 if (ra->mmap_miss > 0)
1616 if (PageReadahead(page))
1617 page_cache_async_readahead(mapping, ra, file,
1618 page, offset, ra->ra_pages);
1622 * filemap_fault - read in file data for page fault handling
1623 * @vma: vma in which the fault was taken
1624 * @vmf: struct vm_fault containing details of the fault
1626 * filemap_fault() is invoked via the vma operations vector for a
1627 * mapped memory region to read in file data during a page fault.
1629 * The goto's are kind of ugly, but this streamlines the normal case of having
1630 * it in the page cache, and handles the special cases reasonably without
1631 * having a lot of duplicated code.
1633 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1636 struct file *file = vma->vm_file;
1637 struct address_space *mapping = file->f_mapping;
1638 struct file_ra_state *ra = &file->f_ra;
1639 struct inode *inode = mapping->host;
1640 pgoff_t offset = vmf->pgoff;
1645 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1647 return VM_FAULT_SIGBUS;
1650 * Do we have something in the page cache already?
1652 page = find_get_page(mapping, offset);
1655 * We found the page, so try async readahead before
1656 * waiting for the lock.
1658 do_async_mmap_readahead(vma, ra, file, page, offset);
1660 /* No page in the page cache at all */
1661 do_sync_mmap_readahead(vma, ra, file, offset);
1662 count_vm_event(PGMAJFAULT);
1663 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1664 ret = VM_FAULT_MAJOR;
1666 page = find_get_page(mapping, offset);
1668 goto no_cached_page;
1671 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1672 page_cache_release(page);
1673 return ret | VM_FAULT_RETRY;
1676 /* Did it get truncated? */
1677 if (unlikely(page->mapping != mapping)) {
1682 VM_BUG_ON(page->index != offset);
1685 * We have a locked page in the page cache, now we need to check
1686 * that it's up-to-date. If not, it is going to be due to an error.
1688 if (unlikely(!PageUptodate(page)))
1689 goto page_not_uptodate;
1692 * Found the page and have a reference on it.
1693 * We must recheck i_size under page lock.
1695 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1696 if (unlikely(offset >= size)) {
1698 page_cache_release(page);
1699 return VM_FAULT_SIGBUS;
1703 return ret | VM_FAULT_LOCKED;
1707 * We're only likely to ever get here if MADV_RANDOM is in
1710 error = page_cache_read(file, offset);
1713 * The page we want has now been added to the page cache.
1714 * In the unlikely event that someone removed it in the
1715 * meantime, we'll just come back here and read it again.
1721 * An error return from page_cache_read can result if the
1722 * system is low on memory, or a problem occurs while trying
1725 if (error == -ENOMEM)
1726 return VM_FAULT_OOM;
1727 return VM_FAULT_SIGBUS;
1731 * Umm, take care of errors if the page isn't up-to-date.
1732 * Try to re-read it _once_. We do this synchronously,
1733 * because there really aren't any performance issues here
1734 * and we need to check for errors.
1736 ClearPageError(page);
1737 error = mapping->a_ops->readpage(file, page);
1739 wait_on_page_locked(page);
1740 if (!PageUptodate(page))
1743 page_cache_release(page);
1745 if (!error || error == AOP_TRUNCATED_PAGE)
1748 /* Things didn't work out. Return zero to tell the mm layer so. */
1749 shrink_readahead_size_eio(file, ra);
1750 return VM_FAULT_SIGBUS;
1752 EXPORT_SYMBOL(filemap_fault);
1754 const struct vm_operations_struct generic_file_vm_ops = {
1755 .fault = filemap_fault,
1758 /* This is used for a general mmap of a disk file */
1760 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1762 struct address_space *mapping = file->f_mapping;
1764 if (!mapping->a_ops->readpage)
1766 file_accessed(file);
1767 vma->vm_ops = &generic_file_vm_ops;
1768 vma->vm_flags |= VM_CAN_NONLINEAR;
1773 * This is for filesystems which do not implement ->writepage.
1775 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1777 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1779 return generic_file_mmap(file, vma);
1782 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1786 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1790 #endif /* CONFIG_MMU */
1792 EXPORT_SYMBOL(generic_file_mmap);
1793 EXPORT_SYMBOL(generic_file_readonly_mmap);
1795 static struct page *__read_cache_page(struct address_space *mapping,
1797 int (*filler)(void *, struct page *),
1804 page = find_get_page(mapping, index);
1806 page = __page_cache_alloc(gfp | __GFP_COLD);
1808 return ERR_PTR(-ENOMEM);
1809 err = add_to_page_cache_lru(page, mapping, index, gfp);
1810 if (unlikely(err)) {
1811 page_cache_release(page);
1814 /* Presumably ENOMEM for radix tree node */
1815 return ERR_PTR(err);
1817 err = filler(data, page);
1819 page_cache_release(page);
1820 page = ERR_PTR(err);
1826 static struct page *do_read_cache_page(struct address_space *mapping,
1828 int (*filler)(void *, struct page *),
1837 page = __read_cache_page(mapping, index, filler, data, gfp);
1840 if (PageUptodate(page))
1844 if (!page->mapping) {
1846 page_cache_release(page);
1849 if (PageUptodate(page)) {
1853 err = filler(data, page);
1855 page_cache_release(page);
1856 return ERR_PTR(err);
1859 mark_page_accessed(page);
1864 * read_cache_page_async - read into page cache, fill it if needed
1865 * @mapping: the page's address_space
1866 * @index: the page index
1867 * @filler: function to perform the read
1868 * @data: first arg to filler(data, page) function, often left as NULL
1870 * Same as read_cache_page, but don't wait for page to become unlocked
1871 * after submitting it to the filler.
1873 * Read into the page cache. If a page already exists, and PageUptodate() is
1874 * not set, try to fill the page but don't wait for it to become unlocked.
1876 * If the page does not get brought uptodate, return -EIO.
1878 struct page *read_cache_page_async(struct address_space *mapping,
1880 int (*filler)(void *, struct page *),
1883 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1885 EXPORT_SYMBOL(read_cache_page_async);
1887 static struct page *wait_on_page_read(struct page *page)
1889 if (!IS_ERR(page)) {
1890 wait_on_page_locked(page);
1891 if (!PageUptodate(page)) {
1892 page_cache_release(page);
1893 page = ERR_PTR(-EIO);
1900 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1901 * @mapping: the page's address_space
1902 * @index: the page index
1903 * @gfp: the page allocator flags to use if allocating
1905 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1906 * any new page allocations done using the specified allocation flags.
1908 * If the page does not get brought uptodate, return -EIO.
1910 struct page *read_cache_page_gfp(struct address_space *mapping,
1914 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1916 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1918 EXPORT_SYMBOL(read_cache_page_gfp);
1921 * read_cache_page - read into page cache, fill it if needed
1922 * @mapping: the page's address_space
1923 * @index: the page index
1924 * @filler: function to perform the read
1925 * @data: first arg to filler(data, page) function, often left as NULL
1927 * Read into the page cache. If a page already exists, and PageUptodate() is
1928 * not set, try to fill the page then wait for it to become unlocked.
1930 * If the page does not get brought uptodate, return -EIO.
1932 struct page *read_cache_page(struct address_space *mapping,
1934 int (*filler)(void *, struct page *),
1937 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1939 EXPORT_SYMBOL(read_cache_page);
1941 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1942 const struct iovec *iov, size_t base, size_t bytes)
1944 size_t copied = 0, left = 0;
1947 char __user *buf = iov->iov_base + base;
1948 int copy = min(bytes, iov->iov_len - base);
1951 left = __copy_from_user_inatomic(vaddr, buf, copy);
1960 return copied - left;
1964 * Copy as much as we can into the page and return the number of bytes which
1965 * were successfully copied. If a fault is encountered then return the number of
1966 * bytes which were copied.
1968 size_t iov_iter_copy_from_user_atomic(struct page *page,
1969 struct iov_iter *i, unsigned long offset, size_t bytes)
1974 BUG_ON(!in_atomic());
1975 kaddr = kmap_atomic(page);
1976 if (likely(i->nr_segs == 1)) {
1978 char __user *buf = i->iov->iov_base + i->iov_offset;
1979 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1980 copied = bytes - left;
1982 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1983 i->iov, i->iov_offset, bytes);
1985 kunmap_atomic(kaddr);
1989 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1992 * This has the same sideeffects and return value as
1993 * iov_iter_copy_from_user_atomic().
1994 * The difference is that it attempts to resolve faults.
1995 * Page must not be locked.
1997 size_t iov_iter_copy_from_user(struct page *page,
1998 struct iov_iter *i, unsigned long offset, size_t bytes)
2004 if (likely(i->nr_segs == 1)) {
2006 char __user *buf = i->iov->iov_base + i->iov_offset;
2007 left = __copy_from_user(kaddr + offset, buf, bytes);
2008 copied = bytes - left;
2010 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2011 i->iov, i->iov_offset, bytes);
2016 EXPORT_SYMBOL(iov_iter_copy_from_user);
2018 void iov_iter_advance(struct iov_iter *i, size_t bytes)
2020 BUG_ON(i->count < bytes);
2022 if (likely(i->nr_segs == 1)) {
2023 i->iov_offset += bytes;
2026 const struct iovec *iov = i->iov;
2027 size_t base = i->iov_offset;
2028 unsigned long nr_segs = i->nr_segs;
2031 * The !iov->iov_len check ensures we skip over unlikely
2032 * zero-length segments (without overruning the iovec).
2034 while (bytes || unlikely(i->count && !iov->iov_len)) {
2037 copy = min(bytes, iov->iov_len - base);
2038 BUG_ON(!i->count || i->count < copy);
2042 if (iov->iov_len == base) {
2049 i->iov_offset = base;
2050 i->nr_segs = nr_segs;
2053 EXPORT_SYMBOL(iov_iter_advance);
2056 * Fault in the first iovec of the given iov_iter, to a maximum length
2057 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2058 * accessed (ie. because it is an invalid address).
2060 * writev-intensive code may want this to prefault several iovecs -- that
2061 * would be possible (callers must not rely on the fact that _only_ the
2062 * first iovec will be faulted with the current implementation).
2064 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2066 char __user *buf = i->iov->iov_base + i->iov_offset;
2067 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2068 return fault_in_pages_readable(buf, bytes);
2070 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2073 * Return the count of just the current iov_iter segment.
2075 size_t iov_iter_single_seg_count(struct iov_iter *i)
2077 const struct iovec *iov = i->iov;
2078 if (i->nr_segs == 1)
2081 return min(i->count, iov->iov_len - i->iov_offset);
2083 EXPORT_SYMBOL(iov_iter_single_seg_count);
2086 * Performs necessary checks before doing a write
2088 * Can adjust writing position or amount of bytes to write.
2089 * Returns appropriate error code that caller should return or
2090 * zero in case that write should be allowed.
2092 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2094 struct inode *inode = file->f_mapping->host;
2095 unsigned long limit = rlimit(RLIMIT_FSIZE);
2097 if (unlikely(*pos < 0))
2101 /* FIXME: this is for backwards compatibility with 2.4 */
2102 if (file->f_flags & O_APPEND)
2103 *pos = i_size_read(inode);
2105 if (limit != RLIM_INFINITY) {
2106 if (*pos >= limit) {
2107 send_sig(SIGXFSZ, current, 0);
2110 if (*count > limit - (typeof(limit))*pos) {
2111 *count = limit - (typeof(limit))*pos;
2119 if (unlikely(*pos + *count > MAX_NON_LFS &&
2120 !(file->f_flags & O_LARGEFILE))) {
2121 if (*pos >= MAX_NON_LFS) {
2124 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2125 *count = MAX_NON_LFS - (unsigned long)*pos;
2130 * Are we about to exceed the fs block limit ?
2132 * If we have written data it becomes a short write. If we have
2133 * exceeded without writing data we send a signal and return EFBIG.
2134 * Linus frestrict idea will clean these up nicely..
2136 if (likely(!isblk)) {
2137 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2138 if (*count || *pos > inode->i_sb->s_maxbytes) {
2141 /* zero-length writes at ->s_maxbytes are OK */
2144 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2145 *count = inode->i_sb->s_maxbytes - *pos;
2149 if (bdev_read_only(I_BDEV(inode)))
2151 isize = i_size_read(inode);
2152 if (*pos >= isize) {
2153 if (*count || *pos > isize)
2157 if (*pos + *count > isize)
2158 *count = isize - *pos;
2165 EXPORT_SYMBOL(generic_write_checks);
2167 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2168 loff_t pos, unsigned len, unsigned flags,
2169 struct page **pagep, void **fsdata)
2171 const struct address_space_operations *aops = mapping->a_ops;
2173 return aops->write_begin(file, mapping, pos, len, flags,
2176 EXPORT_SYMBOL(pagecache_write_begin);
2178 int pagecache_write_end(struct file *file, struct address_space *mapping,
2179 loff_t pos, unsigned len, unsigned copied,
2180 struct page *page, void *fsdata)
2182 const struct address_space_operations *aops = mapping->a_ops;
2184 mark_page_accessed(page);
2185 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2187 EXPORT_SYMBOL(pagecache_write_end);
2190 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2191 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2192 size_t count, size_t ocount)
2194 struct file *file = iocb->ki_filp;
2195 struct address_space *mapping = file->f_mapping;
2196 struct inode *inode = mapping->host;
2201 if (count != ocount)
2202 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2204 write_len = iov_length(iov, *nr_segs);
2205 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2207 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2212 * After a write we want buffered reads to be sure to go to disk to get
2213 * the new data. We invalidate clean cached page from the region we're
2214 * about to write. We do this *before* the write so that we can return
2215 * without clobbering -EIOCBQUEUED from ->direct_IO().
2217 if (mapping->nrpages) {
2218 written = invalidate_inode_pages2_range(mapping,
2219 pos >> PAGE_CACHE_SHIFT, end);
2221 * If a page can not be invalidated, return 0 to fall back
2222 * to buffered write.
2225 if (written == -EBUSY)
2231 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2234 * Finally, try again to invalidate clean pages which might have been
2235 * cached by non-direct readahead, or faulted in by get_user_pages()
2236 * if the source of the write was an mmap'ed region of the file
2237 * we're writing. Either one is a pretty crazy thing to do,
2238 * so we don't support it 100%. If this invalidation
2239 * fails, tough, the write still worked...
2241 if (mapping->nrpages) {
2242 invalidate_inode_pages2_range(mapping,
2243 pos >> PAGE_CACHE_SHIFT, end);
2248 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2249 i_size_write(inode, pos);
2250 mark_inode_dirty(inode);
2257 EXPORT_SYMBOL(generic_file_direct_write);
2260 * Find or create a page at the given pagecache position. Return the locked
2261 * page. This function is specifically for buffered writes.
2263 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2264 pgoff_t index, unsigned flags)
2269 gfp_t gfp_notmask = 0;
2271 gfp_mask = mapping_gfp_mask(mapping);
2272 if (mapping_cap_account_dirty(mapping))
2273 gfp_mask |= __GFP_WRITE;
2274 if (flags & AOP_FLAG_NOFS)
2275 gfp_notmask = __GFP_FS;
2277 page = find_lock_page(mapping, index);
2281 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2284 status = add_to_page_cache_lru(page, mapping, index,
2285 GFP_KERNEL & ~gfp_notmask);
2286 if (unlikely(status)) {
2287 page_cache_release(page);
2288 if (status == -EEXIST)
2293 wait_on_page_writeback(page);
2296 EXPORT_SYMBOL(grab_cache_page_write_begin);
2298 static ssize_t generic_perform_write(struct file *file,
2299 struct iov_iter *i, loff_t pos)
2301 struct address_space *mapping = file->f_mapping;
2302 const struct address_space_operations *a_ops = mapping->a_ops;
2304 ssize_t written = 0;
2305 unsigned int flags = 0;
2308 * Copies from kernel address space cannot fail (NFSD is a big user).
2310 if (segment_eq(get_fs(), KERNEL_DS))
2311 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2315 unsigned long offset; /* Offset into pagecache page */
2316 unsigned long bytes; /* Bytes to write to page */
2317 size_t copied; /* Bytes copied from user */
2320 offset = (pos & (PAGE_CACHE_SIZE - 1));
2321 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2326 * Bring in the user page that we will copy from _first_.
2327 * Otherwise there's a nasty deadlock on copying from the
2328 * same page as we're writing to, without it being marked
2331 * Not only is this an optimisation, but it is also required
2332 * to check that the address is actually valid, when atomic
2333 * usercopies are used, below.
2335 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2340 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2342 if (unlikely(status))
2345 if (mapping_writably_mapped(mapping))
2346 flush_dcache_page(page);
2348 pagefault_disable();
2349 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2351 flush_dcache_page(page);
2353 mark_page_accessed(page);
2354 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2356 if (unlikely(status < 0))
2362 iov_iter_advance(i, copied);
2363 if (unlikely(copied == 0)) {
2365 * If we were unable to copy any data at all, we must
2366 * fall back to a single segment length write.
2368 * If we didn't fallback here, we could livelock
2369 * because not all segments in the iov can be copied at
2370 * once without a pagefault.
2372 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2373 iov_iter_single_seg_count(i));
2379 balance_dirty_pages_ratelimited(mapping);
2380 if (fatal_signal_pending(current)) {
2384 } while (iov_iter_count(i));
2386 return written ? written : status;
2390 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2391 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2392 size_t count, ssize_t written)
2394 struct file *file = iocb->ki_filp;
2398 iov_iter_init(&i, iov, nr_segs, count, written);
2399 status = generic_perform_write(file, &i, pos);
2401 if (likely(status >= 0)) {
2403 *ppos = pos + status;
2406 return written ? written : status;
2408 EXPORT_SYMBOL(generic_file_buffered_write);
2411 * __generic_file_aio_write - write data to a file
2412 * @iocb: IO state structure (file, offset, etc.)
2413 * @iov: vector with data to write
2414 * @nr_segs: number of segments in the vector
2415 * @ppos: position where to write
2417 * This function does all the work needed for actually writing data to a
2418 * file. It does all basic checks, removes SUID from the file, updates
2419 * modification times and calls proper subroutines depending on whether we
2420 * do direct IO or a standard buffered write.
2422 * It expects i_mutex to be grabbed unless we work on a block device or similar
2423 * object which does not need locking at all.
2425 * This function does *not* take care of syncing data in case of O_SYNC write.
2426 * A caller has to handle it. This is mainly due to the fact that we want to
2427 * avoid syncing under i_mutex.
2429 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2430 unsigned long nr_segs, loff_t *ppos)
2432 struct file *file = iocb->ki_filp;
2433 struct address_space * mapping = file->f_mapping;
2434 size_t ocount; /* original count */
2435 size_t count; /* after file limit checks */
2436 struct inode *inode = mapping->host;
2442 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2449 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2451 /* We can write back this queue in page reclaim */
2452 current->backing_dev_info = mapping->backing_dev_info;
2455 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2462 err = file_remove_suid(file);
2466 file_update_time(file);
2468 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2469 if (unlikely(file->f_flags & O_DIRECT)) {
2471 ssize_t written_buffered;
2473 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2474 ppos, count, ocount);
2475 if (written < 0 || written == count)
2478 * direct-io write to a hole: fall through to buffered I/O
2479 * for completing the rest of the request.
2483 written_buffered = generic_file_buffered_write(iocb, iov,
2484 nr_segs, pos, ppos, count,
2487 * If generic_file_buffered_write() retuned a synchronous error
2488 * then we want to return the number of bytes which were
2489 * direct-written, or the error code if that was zero. Note
2490 * that this differs from normal direct-io semantics, which
2491 * will return -EFOO even if some bytes were written.
2493 if (written_buffered < 0) {
2494 err = written_buffered;
2499 * We need to ensure that the page cache pages are written to
2500 * disk and invalidated to preserve the expected O_DIRECT
2503 endbyte = pos + written_buffered - written - 1;
2504 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2506 written = written_buffered;
2507 invalidate_mapping_pages(mapping,
2508 pos >> PAGE_CACHE_SHIFT,
2509 endbyte >> PAGE_CACHE_SHIFT);
2512 * We don't know how much we wrote, so just return
2513 * the number of bytes which were direct-written
2517 written = generic_file_buffered_write(iocb, iov, nr_segs,
2518 pos, ppos, count, written);
2521 current->backing_dev_info = NULL;
2522 return written ? written : err;
2524 EXPORT_SYMBOL(__generic_file_aio_write);
2527 * generic_file_aio_write - write data to a file
2528 * @iocb: IO state structure
2529 * @iov: vector with data to write
2530 * @nr_segs: number of segments in the vector
2531 * @pos: position in file where to write
2533 * This is a wrapper around __generic_file_aio_write() to be used by most
2534 * filesystems. It takes care of syncing the file in case of O_SYNC file
2535 * and acquires i_mutex as needed.
2537 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2538 unsigned long nr_segs, loff_t pos)
2540 struct file *file = iocb->ki_filp;
2541 struct inode *inode = file->f_mapping->host;
2542 struct blk_plug plug;
2545 BUG_ON(iocb->ki_pos != pos);
2547 mutex_lock(&inode->i_mutex);
2548 blk_start_plug(&plug);
2549 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2550 mutex_unlock(&inode->i_mutex);
2552 if (ret > 0 || ret == -EIOCBQUEUED) {
2555 err = generic_write_sync(file, pos, ret);
2556 if (err < 0 && ret > 0)
2559 blk_finish_plug(&plug);
2562 EXPORT_SYMBOL(generic_file_aio_write);
2565 * try_to_release_page() - release old fs-specific metadata on a page
2567 * @page: the page which the kernel is trying to free
2568 * @gfp_mask: memory allocation flags (and I/O mode)
2570 * The address_space is to try to release any data against the page
2571 * (presumably at page->private). If the release was successful, return `1'.
2572 * Otherwise return zero.
2574 * This may also be called if PG_fscache is set on a page, indicating that the
2575 * page is known to the local caching routines.
2577 * The @gfp_mask argument specifies whether I/O may be performed to release
2578 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2581 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2583 struct address_space * const mapping = page->mapping;
2585 BUG_ON(!PageLocked(page));
2586 if (PageWriteback(page))
2589 if (mapping && mapping->a_ops->releasepage)
2590 return mapping->a_ops->releasepage(page, gfp_mask);
2591 return try_to_free_buffers(page);
2594 EXPORT_SYMBOL(try_to_release_page);