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/cpuset.h>
33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
34 #include <linux/memcontrol.h>
35 #include <linux/cleancache.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/filemap.h>
42 * FIXME: remove all knowledge of the buffer layer from the core VM
44 #include <linux/buffer_head.h> /* for try_to_free_buffers */
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * Shared mappings now work. 15.8.1995 Bruno.
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
63 * ->i_mmap_mutex (truncate_pagecache)
64 * ->private_lock (__free_pte->__set_page_dirty_buffers)
65 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
69 * ->i_mmap_mutex (truncate->unmap_mapping_range)
73 * ->page_table_lock or pte_lock (various, mainly in memory.c)
74 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 * ->lock_page (access_process_vm)
79 * ->i_mutex (generic_file_buffered_write)
80 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
83 * sb_lock (fs/fs-writeback.c)
84 * ->mapping->tree_lock (__sync_single_inode)
87 * ->anon_vma.lock (vma_adjust)
90 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
92 * ->page_table_lock or pte_lock
93 * ->swap_lock (try_to_unmap_one)
94 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
97 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
98 * ->private_lock (page_remove_rmap->set_page_dirty)
99 * ->tree_lock (page_remove_rmap->set_page_dirty)
100 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
101 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
103 * ->inode->i_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 * Delete 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 __delete_from_page_cache(struct page *page)
117 struct address_space *mapping = page->mapping;
119 trace_mm_filemap_delete_from_page_cache(page);
121 * if we're uptodate, flush out into the cleancache, otherwise
122 * invalidate any existing cleancache entries. We can't leave
123 * stale data around in the cleancache once our page is gone
125 if (PageUptodate(page) && PageMappedToDisk(page))
126 cleancache_put_page(page);
128 cleancache_invalidate_page(mapping, page);
130 radix_tree_delete(&mapping->page_tree, page->index);
131 page->mapping = NULL;
132 /* Leave page->index set: truncation lookup relies upon it */
134 __dec_zone_page_state(page, NR_FILE_PAGES);
135 if (PageSwapBacked(page))
136 __dec_zone_page_state(page, NR_SHMEM);
137 BUG_ON(page_mapped(page));
140 * Some filesystems seem to re-dirty the page even after
141 * the VM has canceled the dirty bit (eg ext3 journaling).
143 * Fix it up by doing a final dirty accounting check after
144 * having removed the page entirely.
146 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
147 dec_zone_page_state(page, NR_FILE_DIRTY);
148 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
153 * delete_from_page_cache - delete page from page cache
154 * @page: the page which the kernel is trying to remove from page cache
156 * This must be called only on pages that have been verified to be in the page
157 * cache and locked. It will never put the page into the free list, the caller
158 * has a reference on the page.
160 void delete_from_page_cache(struct page *page)
162 struct address_space *mapping = page->mapping;
163 void (*freepage)(struct page *);
165 BUG_ON(!PageLocked(page));
167 freepage = mapping->a_ops->freepage;
168 spin_lock_irq(&mapping->tree_lock);
169 __delete_from_page_cache(page);
170 spin_unlock_irq(&mapping->tree_lock);
171 mem_cgroup_uncharge_cache_page(page);
175 page_cache_release(page);
177 EXPORT_SYMBOL(delete_from_page_cache);
179 static int sleep_on_page(void *word)
185 static int sleep_on_page_killable(void *word)
188 return fatal_signal_pending(current) ? -EINTR : 0;
192 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
193 * @mapping: address space structure to write
194 * @start: offset in bytes where the range starts
195 * @end: offset in bytes where the range ends (inclusive)
196 * @sync_mode: enable synchronous operation
198 * Start writeback against all of a mapping's dirty pages that lie
199 * within the byte offsets <start, end> inclusive.
201 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
202 * opposed to a regular memory cleansing writeback. The difference between
203 * these two operations is that if a dirty page/buffer is encountered, it must
204 * be waited upon, and not just skipped over.
206 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
207 loff_t end, int sync_mode)
210 struct writeback_control wbc = {
211 .sync_mode = sync_mode,
212 .nr_to_write = LONG_MAX,
213 .range_start = start,
217 if (!mapping_cap_writeback_dirty(mapping))
220 ret = do_writepages(mapping, &wbc);
224 static inline int __filemap_fdatawrite(struct address_space *mapping,
227 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
230 int filemap_fdatawrite(struct address_space *mapping)
232 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
234 EXPORT_SYMBOL(filemap_fdatawrite);
236 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
239 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
241 EXPORT_SYMBOL(filemap_fdatawrite_range);
244 * filemap_flush - mostly a non-blocking flush
245 * @mapping: target address_space
247 * This is a mostly non-blocking flush. Not suitable for data-integrity
248 * purposes - I/O may not be started against all dirty pages.
250 int filemap_flush(struct address_space *mapping)
252 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
254 EXPORT_SYMBOL(filemap_flush);
257 * filemap_fdatawait_range - wait for writeback to complete
258 * @mapping: address space structure to wait for
259 * @start_byte: offset in bytes where the range starts
260 * @end_byte: offset in bytes where the range ends (inclusive)
262 * Walk the list of under-writeback pages of the given address space
263 * in the given range and wait for all of them.
265 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
268 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
269 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
274 if (end_byte < start_byte)
277 pagevec_init(&pvec, 0);
278 while ((index <= end) &&
279 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
280 PAGECACHE_TAG_WRITEBACK,
281 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
284 for (i = 0; i < nr_pages; i++) {
285 struct page *page = pvec.pages[i];
287 /* until radix tree lookup accepts end_index */
288 if (page->index > end)
291 wait_on_page_writeback(page);
292 if (TestClearPageError(page))
295 pagevec_release(&pvec);
299 /* Check for outstanding write errors */
300 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
302 if (test_and_clear_bit(AS_EIO, &mapping->flags))
307 EXPORT_SYMBOL(filemap_fdatawait_range);
310 * filemap_fdatawait - wait for all under-writeback pages to complete
311 * @mapping: address space structure to wait for
313 * Walk the list of under-writeback pages of the given address space
314 * and wait for all of them.
316 int filemap_fdatawait(struct address_space *mapping)
318 loff_t i_size = i_size_read(mapping->host);
323 return filemap_fdatawait_range(mapping, 0, i_size - 1);
325 EXPORT_SYMBOL(filemap_fdatawait);
327 int filemap_write_and_wait(struct address_space *mapping)
331 if (mapping->nrpages) {
332 err = filemap_fdatawrite(mapping);
334 * Even if the above returned error, the pages may be
335 * written partially (e.g. -ENOSPC), so we wait for it.
336 * But the -EIO is special case, it may indicate the worst
337 * thing (e.g. bug) happened, so we avoid waiting for it.
340 int err2 = filemap_fdatawait(mapping);
347 EXPORT_SYMBOL(filemap_write_and_wait);
350 * filemap_write_and_wait_range - write out & wait on a file range
351 * @mapping: the address_space for the pages
352 * @lstart: offset in bytes where the range starts
353 * @lend: offset in bytes where the range ends (inclusive)
355 * Write out and wait upon file offsets lstart->lend, inclusive.
357 * Note that `lend' is inclusive (describes the last byte to be written) so
358 * that this function can be used to write to the very end-of-file (end = -1).
360 int filemap_write_and_wait_range(struct address_space *mapping,
361 loff_t lstart, loff_t lend)
365 if (mapping->nrpages) {
366 err = __filemap_fdatawrite_range(mapping, lstart, lend,
368 /* See comment of filemap_write_and_wait() */
370 int err2 = filemap_fdatawait_range(mapping,
378 EXPORT_SYMBOL(filemap_write_and_wait_range);
381 * replace_page_cache_page - replace a pagecache page with a new one
382 * @old: page to be replaced
383 * @new: page to replace with
384 * @gfp_mask: allocation mode
386 * This function replaces a page in the pagecache with a new one. On
387 * success it acquires the pagecache reference for the new page and
388 * drops it for the old page. Both the old and new pages must be
389 * locked. This function does not add the new page to the LRU, the
390 * caller must do that.
392 * The remove + add is atomic. The only way this function can fail is
393 * memory allocation failure.
395 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
399 VM_BUG_ON(!PageLocked(old));
400 VM_BUG_ON(!PageLocked(new));
401 VM_BUG_ON(new->mapping);
403 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
405 struct address_space *mapping = old->mapping;
406 void (*freepage)(struct page *);
408 pgoff_t offset = old->index;
409 freepage = mapping->a_ops->freepage;
412 new->mapping = mapping;
415 spin_lock_irq(&mapping->tree_lock);
416 __delete_from_page_cache(old);
417 error = radix_tree_insert(&mapping->page_tree, offset, new);
420 __inc_zone_page_state(new, NR_FILE_PAGES);
421 if (PageSwapBacked(new))
422 __inc_zone_page_state(new, NR_SHMEM);
423 spin_unlock_irq(&mapping->tree_lock);
424 /* mem_cgroup codes must not be called under tree_lock */
425 mem_cgroup_replace_page_cache(old, new);
426 radix_tree_preload_end();
429 page_cache_release(old);
434 EXPORT_SYMBOL_GPL(replace_page_cache_page);
437 * add_to_page_cache_locked - add a locked page to the pagecache
439 * @mapping: the page's address_space
440 * @offset: page index
441 * @gfp_mask: page allocation mode
443 * This function is used to add a page to the pagecache. It must be locked.
444 * This function does not add the page to the LRU. The caller must do that.
446 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
447 pgoff_t offset, gfp_t gfp_mask)
451 VM_BUG_ON(!PageLocked(page));
452 VM_BUG_ON(PageSwapBacked(page));
454 error = mem_cgroup_cache_charge(page, current->mm,
455 gfp_mask & GFP_RECLAIM_MASK);
459 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
461 page_cache_get(page);
462 page->mapping = mapping;
463 page->index = offset;
465 spin_lock_irq(&mapping->tree_lock);
466 error = radix_tree_insert(&mapping->page_tree, offset, page);
467 if (likely(!error)) {
469 __inc_zone_page_state(page, NR_FILE_PAGES);
470 trace_mm_filemap_add_to_page_cache(page);
471 spin_unlock_irq(&mapping->tree_lock);
473 page->mapping = NULL;
474 /* Leave page->index set: truncation relies upon it */
475 spin_unlock_irq(&mapping->tree_lock);
476 mem_cgroup_uncharge_cache_page(page);
477 page_cache_release(page);
479 radix_tree_preload_end();
481 mem_cgroup_uncharge_cache_page(page);
485 EXPORT_SYMBOL(add_to_page_cache_locked);
487 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
488 pgoff_t offset, gfp_t gfp_mask)
492 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
494 lru_cache_add_file(page);
497 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
500 struct page *__page_cache_alloc(gfp_t gfp)
505 if (cpuset_do_page_mem_spread()) {
506 unsigned int cpuset_mems_cookie;
508 cpuset_mems_cookie = get_mems_allowed();
509 n = cpuset_mem_spread_node();
510 page = alloc_pages_exact_node(n, gfp, 0);
511 } while (!put_mems_allowed(cpuset_mems_cookie) && !page);
515 return alloc_pages(gfp, 0);
517 EXPORT_SYMBOL(__page_cache_alloc);
521 * In order to wait for pages to become available there must be
522 * waitqueues associated with pages. By using a hash table of
523 * waitqueues where the bucket discipline is to maintain all
524 * waiters on the same queue and wake all when any of the pages
525 * become available, and for the woken contexts to check to be
526 * sure the appropriate page became available, this saves space
527 * at a cost of "thundering herd" phenomena during rare hash
530 static wait_queue_head_t *page_waitqueue(struct page *page)
532 const struct zone *zone = page_zone(page);
534 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
537 static inline void wake_up_page(struct page *page, int bit)
539 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
542 void wait_on_page_bit(struct page *page, int bit_nr)
544 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
546 if (test_bit(bit_nr, &page->flags))
547 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
548 TASK_UNINTERRUPTIBLE);
550 EXPORT_SYMBOL(wait_on_page_bit);
552 int wait_on_page_bit_killable(struct page *page, int bit_nr)
554 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
556 if (!test_bit(bit_nr, &page->flags))
559 return __wait_on_bit(page_waitqueue(page), &wait,
560 sleep_on_page_killable, TASK_KILLABLE);
564 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
565 * @page: Page defining the wait queue of interest
566 * @waiter: Waiter to add to the queue
568 * Add an arbitrary @waiter to the wait queue for the nominated @page.
570 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
572 wait_queue_head_t *q = page_waitqueue(page);
575 spin_lock_irqsave(&q->lock, flags);
576 __add_wait_queue(q, waiter);
577 spin_unlock_irqrestore(&q->lock, flags);
579 EXPORT_SYMBOL_GPL(add_page_wait_queue);
582 * unlock_page - unlock a locked page
585 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
586 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
587 * mechananism between PageLocked pages and PageWriteback pages is shared.
588 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
590 * The mb is necessary to enforce ordering between the clear_bit and the read
591 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
593 void unlock_page(struct page *page)
595 VM_BUG_ON(!PageLocked(page));
596 clear_bit_unlock(PG_locked, &page->flags);
597 smp_mb__after_clear_bit();
598 wake_up_page(page, PG_locked);
600 EXPORT_SYMBOL(unlock_page);
603 * end_page_writeback - end writeback against a page
606 void end_page_writeback(struct page *page)
608 if (TestClearPageReclaim(page))
609 rotate_reclaimable_page(page);
611 if (!test_clear_page_writeback(page))
614 smp_mb__after_clear_bit();
615 wake_up_page(page, PG_writeback);
617 EXPORT_SYMBOL(end_page_writeback);
620 * __lock_page - get a lock on the page, assuming we need to sleep to get it
621 * @page: the page to lock
623 void __lock_page(struct page *page)
625 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
627 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
628 TASK_UNINTERRUPTIBLE);
630 EXPORT_SYMBOL(__lock_page);
632 int __lock_page_killable(struct page *page)
634 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
636 return __wait_on_bit_lock(page_waitqueue(page), &wait,
637 sleep_on_page_killable, TASK_KILLABLE);
639 EXPORT_SYMBOL_GPL(__lock_page_killable);
641 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
644 if (flags & FAULT_FLAG_ALLOW_RETRY) {
646 * CAUTION! In this case, mmap_sem is not released
647 * even though return 0.
649 if (flags & FAULT_FLAG_RETRY_NOWAIT)
652 up_read(&mm->mmap_sem);
653 if (flags & FAULT_FLAG_KILLABLE)
654 wait_on_page_locked_killable(page);
656 wait_on_page_locked(page);
659 if (flags & FAULT_FLAG_KILLABLE) {
662 ret = __lock_page_killable(page);
664 up_read(&mm->mmap_sem);
674 * find_get_page - find and get a page reference
675 * @mapping: the address_space to search
676 * @offset: the page index
678 * Is there a pagecache struct page at the given (mapping, offset) tuple?
679 * If yes, increment its refcount and return it; if no, return NULL.
681 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
689 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
691 page = radix_tree_deref_slot(pagep);
694 if (radix_tree_exception(page)) {
695 if (radix_tree_deref_retry(page))
698 * Otherwise, shmem/tmpfs must be storing a swap entry
699 * here as an exceptional entry: so return it without
700 * attempting to raise page count.
704 if (!page_cache_get_speculative(page))
708 * Has the page moved?
709 * This is part of the lockless pagecache protocol. See
710 * include/linux/pagemap.h for details.
712 if (unlikely(page != *pagep)) {
713 page_cache_release(page);
722 EXPORT_SYMBOL(find_get_page);
725 * find_lock_page - locate, pin and lock a pagecache page
726 * @mapping: the address_space to search
727 * @offset: the page index
729 * Locates the desired pagecache page, locks it, increments its reference
730 * count and returns its address.
732 * Returns zero if the page was not present. find_lock_page() may sleep.
734 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
739 page = find_get_page(mapping, offset);
740 if (page && !radix_tree_exception(page)) {
742 /* Has the page been truncated? */
743 if (unlikely(page->mapping != mapping)) {
745 page_cache_release(page);
748 VM_BUG_ON(page->index != offset);
752 EXPORT_SYMBOL(find_lock_page);
755 * find_or_create_page - locate or add a pagecache page
756 * @mapping: the page's address_space
757 * @index: the page's index into the mapping
758 * @gfp_mask: page allocation mode
760 * Locates a page in the pagecache. If the page is not present, a new page
761 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
762 * LRU list. The returned page is locked and has its reference count
765 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
768 * find_or_create_page() returns the desired page's address, or zero on
771 struct page *find_or_create_page(struct address_space *mapping,
772 pgoff_t index, gfp_t gfp_mask)
777 page = find_lock_page(mapping, index);
779 page = __page_cache_alloc(gfp_mask);
783 * We want a regular kernel memory (not highmem or DMA etc)
784 * allocation for the radix tree nodes, but we need to honour
785 * the context-specific requirements the caller has asked for.
786 * GFP_RECLAIM_MASK collects those requirements.
788 err = add_to_page_cache_lru(page, mapping, index,
789 (gfp_mask & GFP_RECLAIM_MASK));
791 page_cache_release(page);
799 EXPORT_SYMBOL(find_or_create_page);
802 * find_get_pages - gang pagecache lookup
803 * @mapping: The address_space to search
804 * @start: The starting page index
805 * @nr_pages: The maximum number of pages
806 * @pages: Where the resulting pages are placed
808 * find_get_pages() will search for and return a group of up to
809 * @nr_pages pages in the mapping. The pages are placed at @pages.
810 * find_get_pages() takes a reference against the returned pages.
812 * The search returns a group of mapping-contiguous pages with ascending
813 * indexes. There may be holes in the indices due to not-present pages.
815 * find_get_pages() returns the number of pages which were found.
817 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
818 unsigned int nr_pages, struct page **pages)
820 struct radix_tree_iter iter;
824 if (unlikely(!nr_pages))
829 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
832 page = radix_tree_deref_slot(slot);
836 if (radix_tree_exception(page)) {
837 if (radix_tree_deref_retry(page)) {
839 * Transient condition which can only trigger
840 * when entry at index 0 moves out of or back
841 * to root: none yet gotten, safe to restart.
847 * Otherwise, shmem/tmpfs must be storing a swap entry
848 * here as an exceptional entry: so skip over it -
849 * we only reach this from invalidate_mapping_pages().
854 if (!page_cache_get_speculative(page))
857 /* Has the page moved? */
858 if (unlikely(page != *slot)) {
859 page_cache_release(page);
864 if (++ret == nr_pages)
873 * find_get_pages_contig - gang contiguous pagecache lookup
874 * @mapping: The address_space to search
875 * @index: The starting page index
876 * @nr_pages: The maximum number of pages
877 * @pages: Where the resulting pages are placed
879 * find_get_pages_contig() works exactly like find_get_pages(), except
880 * that the returned number of pages are guaranteed to be contiguous.
882 * find_get_pages_contig() returns the number of pages which were found.
884 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
885 unsigned int nr_pages, struct page **pages)
887 struct radix_tree_iter iter;
889 unsigned int ret = 0;
891 if (unlikely(!nr_pages))
896 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
899 page = radix_tree_deref_slot(slot);
900 /* The hole, there no reason to continue */
904 if (radix_tree_exception(page)) {
905 if (radix_tree_deref_retry(page)) {
907 * Transient condition which can only trigger
908 * when entry at index 0 moves out of or back
909 * to root: none yet gotten, safe to restart.
914 * Otherwise, shmem/tmpfs must be storing a swap entry
915 * here as an exceptional entry: so stop looking for
921 if (!page_cache_get_speculative(page))
924 /* Has the page moved? */
925 if (unlikely(page != *slot)) {
926 page_cache_release(page);
931 * must check mapping and index after taking the ref.
932 * otherwise we can get both false positives and false
933 * negatives, which is just confusing to the caller.
935 if (page->mapping == NULL || page->index != iter.index) {
936 page_cache_release(page);
941 if (++ret == nr_pages)
947 EXPORT_SYMBOL(find_get_pages_contig);
950 * find_get_pages_tag - find and return pages that match @tag
951 * @mapping: the address_space to search
952 * @index: the starting page index
953 * @tag: the tag index
954 * @nr_pages: the maximum number of pages
955 * @pages: where the resulting pages are placed
957 * Like find_get_pages, except we only return pages which are tagged with
958 * @tag. We update @index to index the next page for the traversal.
960 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
961 int tag, unsigned int nr_pages, struct page **pages)
963 struct radix_tree_iter iter;
967 if (unlikely(!nr_pages))
972 radix_tree_for_each_tagged(slot, &mapping->page_tree,
973 &iter, *index, tag) {
976 page = radix_tree_deref_slot(slot);
980 if (radix_tree_exception(page)) {
981 if (radix_tree_deref_retry(page)) {
983 * Transient condition which can only trigger
984 * when entry at index 0 moves out of or back
985 * to root: none yet gotten, safe to restart.
990 * This function is never used on a shmem/tmpfs
991 * mapping, so a swap entry won't be found here.
996 if (!page_cache_get_speculative(page))
999 /* Has the page moved? */
1000 if (unlikely(page != *slot)) {
1001 page_cache_release(page);
1006 if (++ret == nr_pages)
1013 *index = pages[ret - 1]->index + 1;
1017 EXPORT_SYMBOL(find_get_pages_tag);
1020 * grab_cache_page_nowait - returns locked page at given index in given cache
1021 * @mapping: target address_space
1022 * @index: the page index
1024 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1025 * This is intended for speculative data generators, where the data can
1026 * be regenerated if the page couldn't be grabbed. This routine should
1027 * be safe to call while holding the lock for another page.
1029 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1030 * and deadlock against the caller's locked page.
1033 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1035 struct page *page = find_get_page(mapping, index);
1038 if (trylock_page(page))
1040 page_cache_release(page);
1043 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1044 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1045 page_cache_release(page);
1050 EXPORT_SYMBOL(grab_cache_page_nowait);
1053 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1054 * a _large_ part of the i/o request. Imagine the worst scenario:
1056 * ---R__________________________________________B__________
1057 * ^ reading here ^ bad block(assume 4k)
1059 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1060 * => failing the whole request => read(R) => read(R+1) =>
1061 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1062 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1063 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1065 * It is going insane. Fix it by quickly scaling down the readahead size.
1067 static void shrink_readahead_size_eio(struct file *filp,
1068 struct file_ra_state *ra)
1074 * do_generic_file_read - generic file read routine
1075 * @filp: the file to read
1076 * @ppos: current file position
1077 * @desc: read_descriptor
1078 * @actor: read method
1080 * This is a generic file read routine, and uses the
1081 * mapping->a_ops->readpage() function for the actual low-level stuff.
1083 * This is really ugly. But the goto's actually try to clarify some
1084 * of the logic when it comes to error handling etc.
1086 static void do_generic_file_read(struct file *filp, loff_t *ppos,
1087 read_descriptor_t *desc, read_actor_t actor)
1089 struct address_space *mapping = filp->f_mapping;
1090 struct inode *inode = mapping->host;
1091 struct file_ra_state *ra = &filp->f_ra;
1095 unsigned long offset; /* offset into pagecache page */
1096 unsigned int prev_offset;
1099 index = *ppos >> PAGE_CACHE_SHIFT;
1100 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1101 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1102 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1103 offset = *ppos & ~PAGE_CACHE_MASK;
1109 unsigned long nr, ret;
1113 page = find_get_page(mapping, index);
1115 page_cache_sync_readahead(mapping,
1117 index, last_index - index);
1118 page = find_get_page(mapping, index);
1119 if (unlikely(page == NULL))
1120 goto no_cached_page;
1122 if (PageReadahead(page)) {
1123 page_cache_async_readahead(mapping,
1125 index, last_index - index);
1127 if (!PageUptodate(page)) {
1128 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1129 !mapping->a_ops->is_partially_uptodate)
1130 goto page_not_up_to_date;
1131 if (!trylock_page(page))
1132 goto page_not_up_to_date;
1133 /* Did it get truncated before we got the lock? */
1135 goto page_not_up_to_date_locked;
1136 if (!mapping->a_ops->is_partially_uptodate(page,
1138 goto page_not_up_to_date_locked;
1143 * i_size must be checked after we know the page is Uptodate.
1145 * Checking i_size after the check allows us to calculate
1146 * the correct value for "nr", which means the zero-filled
1147 * part of the page is not copied back to userspace (unless
1148 * another truncate extends the file - this is desired though).
1151 isize = i_size_read(inode);
1152 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1153 if (unlikely(!isize || index > end_index)) {
1154 page_cache_release(page);
1158 /* nr is the maximum number of bytes to copy from this page */
1159 nr = PAGE_CACHE_SIZE;
1160 if (index == end_index) {
1161 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1163 page_cache_release(page);
1169 /* If users can be writing to this page using arbitrary
1170 * virtual addresses, take care about potential aliasing
1171 * before reading the page on the kernel side.
1173 if (mapping_writably_mapped(mapping))
1174 flush_dcache_page(page);
1177 * When a sequential read accesses a page several times,
1178 * only mark it as accessed the first time.
1180 if (prev_index != index || offset != prev_offset)
1181 mark_page_accessed(page);
1185 * Ok, we have the page, and it's up-to-date, so
1186 * now we can copy it to user space...
1188 * The actor routine returns how many bytes were actually used..
1189 * NOTE! This may not be the same as how much of a user buffer
1190 * we filled up (we may be padding etc), so we can only update
1191 * "pos" here (the actor routine has to update the user buffer
1192 * pointers and the remaining count).
1194 ret = actor(desc, page, offset, nr);
1196 index += offset >> PAGE_CACHE_SHIFT;
1197 offset &= ~PAGE_CACHE_MASK;
1198 prev_offset = offset;
1200 page_cache_release(page);
1201 if (ret == nr && desc->count)
1205 page_not_up_to_date:
1206 /* Get exclusive access to the page ... */
1207 error = lock_page_killable(page);
1208 if (unlikely(error))
1209 goto readpage_error;
1211 page_not_up_to_date_locked:
1212 /* Did it get truncated before we got the lock? */
1213 if (!page->mapping) {
1215 page_cache_release(page);
1219 /* Did somebody else fill it already? */
1220 if (PageUptodate(page)) {
1227 * A previous I/O error may have been due to temporary
1228 * failures, eg. multipath errors.
1229 * PG_error will be set again if readpage fails.
1231 ClearPageError(page);
1232 /* Start the actual read. The read will unlock the page. */
1233 error = mapping->a_ops->readpage(filp, page);
1235 if (unlikely(error)) {
1236 if (error == AOP_TRUNCATED_PAGE) {
1237 page_cache_release(page);
1240 goto readpage_error;
1243 if (!PageUptodate(page)) {
1244 error = lock_page_killable(page);
1245 if (unlikely(error))
1246 goto readpage_error;
1247 if (!PageUptodate(page)) {
1248 if (page->mapping == NULL) {
1250 * invalidate_mapping_pages got it
1253 page_cache_release(page);
1257 shrink_readahead_size_eio(filp, ra);
1259 goto readpage_error;
1267 /* UHHUH! A synchronous read error occurred. Report it */
1268 desc->error = error;
1269 page_cache_release(page);
1274 * Ok, it wasn't cached, so we need to create a new
1277 page = page_cache_alloc_cold(mapping);
1279 desc->error = -ENOMEM;
1282 error = add_to_page_cache_lru(page, mapping,
1285 page_cache_release(page);
1286 if (error == -EEXIST)
1288 desc->error = error;
1295 ra->prev_pos = prev_index;
1296 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1297 ra->prev_pos |= prev_offset;
1299 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1300 file_accessed(filp);
1303 int file_read_actor(read_descriptor_t *desc, struct page *page,
1304 unsigned long offset, unsigned long size)
1307 unsigned long left, count = desc->count;
1313 * Faults on the destination of a read are common, so do it before
1316 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1317 kaddr = kmap_atomic(page);
1318 left = __copy_to_user_inatomic(desc->arg.buf,
1319 kaddr + offset, size);
1320 kunmap_atomic(kaddr);
1325 /* Do it the slow way */
1327 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1332 desc->error = -EFAULT;
1335 desc->count = count - size;
1336 desc->written += size;
1337 desc->arg.buf += size;
1342 * Performs necessary checks before doing a write
1343 * @iov: io vector request
1344 * @nr_segs: number of segments in the iovec
1345 * @count: number of bytes to write
1346 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1348 * Adjust number of segments and amount of bytes to write (nr_segs should be
1349 * properly initialized first). Returns appropriate error code that caller
1350 * should return or zero in case that write should be allowed.
1352 int generic_segment_checks(const struct iovec *iov,
1353 unsigned long *nr_segs, size_t *count, int access_flags)
1357 for (seg = 0; seg < *nr_segs; seg++) {
1358 const struct iovec *iv = &iov[seg];
1361 * If any segment has a negative length, or the cumulative
1362 * length ever wraps negative then return -EINVAL.
1365 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1367 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1372 cnt -= iv->iov_len; /* This segment is no good */
1378 EXPORT_SYMBOL(generic_segment_checks);
1381 * generic_file_aio_read - generic filesystem read routine
1382 * @iocb: kernel I/O control block
1383 * @iov: io vector request
1384 * @nr_segs: number of segments in the iovec
1385 * @pos: current file position
1387 * This is the "read()" routine for all filesystems
1388 * that can use the page cache directly.
1391 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1392 unsigned long nr_segs, loff_t pos)
1394 struct file *filp = iocb->ki_filp;
1396 unsigned long seg = 0;
1398 loff_t *ppos = &iocb->ki_pos;
1401 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1405 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1406 if (filp->f_flags & O_DIRECT) {
1408 struct address_space *mapping;
1409 struct inode *inode;
1411 mapping = filp->f_mapping;
1412 inode = mapping->host;
1414 goto out; /* skip atime */
1415 size = i_size_read(inode);
1417 retval = filemap_write_and_wait_range(mapping, pos,
1418 pos + iov_length(iov, nr_segs) - 1);
1420 retval = mapping->a_ops->direct_IO(READ, iocb,
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);
1483 * page_cache_read - adds requested page to the page cache if not already there
1484 * @file: file to read
1485 * @offset: page index
1487 * This adds the requested page to the page cache if it isn't already there,
1488 * and schedules an I/O to read in its contents from disk.
1490 static int page_cache_read(struct file *file, pgoff_t offset)
1492 struct address_space *mapping = file->f_mapping;
1497 page = page_cache_alloc_cold(mapping);
1501 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1503 ret = mapping->a_ops->readpage(file, page);
1504 else if (ret == -EEXIST)
1505 ret = 0; /* losing race to add is OK */
1507 page_cache_release(page);
1509 } while (ret == AOP_TRUNCATED_PAGE);
1514 #define MMAP_LOTSAMISS (100)
1517 * Synchronous readahead happens when we don't even find
1518 * a page in the page cache at all.
1520 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1521 struct file_ra_state *ra,
1525 unsigned long ra_pages;
1526 struct address_space *mapping = file->f_mapping;
1528 /* If we don't want any read-ahead, don't bother */
1529 if (VM_RandomReadHint(vma))
1534 if (VM_SequentialReadHint(vma)) {
1535 page_cache_sync_readahead(mapping, ra, file, offset,
1540 /* Avoid banging the cache line if not needed */
1541 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1545 * Do we miss much more than hit in this file? If so,
1546 * stop bothering with read-ahead. It will only hurt.
1548 if (ra->mmap_miss > MMAP_LOTSAMISS)
1554 ra_pages = max_sane_readahead(ra->ra_pages);
1555 ra->start = max_t(long, 0, offset - ra_pages / 2);
1556 ra->size = ra_pages;
1557 ra->async_size = ra_pages / 4;
1558 ra_submit(ra, mapping, file);
1562 * Asynchronous readahead happens when we find the page and PG_readahead,
1563 * so we want to possibly extend the readahead further..
1565 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1566 struct file_ra_state *ra,
1571 struct address_space *mapping = file->f_mapping;
1573 /* If we don't want any read-ahead, don't bother */
1574 if (VM_RandomReadHint(vma))
1576 if (ra->mmap_miss > 0)
1578 if (PageReadahead(page))
1579 page_cache_async_readahead(mapping, ra, file,
1580 page, offset, ra->ra_pages);
1584 * filemap_fault - read in file data for page fault handling
1585 * @vma: vma in which the fault was taken
1586 * @vmf: struct vm_fault containing details of the fault
1588 * filemap_fault() is invoked via the vma operations vector for a
1589 * mapped memory region to read in file data during a page fault.
1591 * The goto's are kind of ugly, but this streamlines the normal case of having
1592 * it in the page cache, and handles the special cases reasonably without
1593 * having a lot of duplicated code.
1595 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1598 struct file *file = vma->vm_file;
1599 struct address_space *mapping = file->f_mapping;
1600 struct file_ra_state *ra = &file->f_ra;
1601 struct inode *inode = mapping->host;
1602 pgoff_t offset = vmf->pgoff;
1607 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1609 return VM_FAULT_SIGBUS;
1612 * Do we have something in the page cache already?
1614 page = find_get_page(mapping, offset);
1615 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1617 * We found the page, so try async readahead before
1618 * waiting for the lock.
1620 do_async_mmap_readahead(vma, ra, file, page, offset);
1622 /* No page in the page cache at all */
1623 do_sync_mmap_readahead(vma, ra, file, offset);
1624 count_vm_event(PGMAJFAULT);
1625 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1626 ret = VM_FAULT_MAJOR;
1628 page = find_get_page(mapping, offset);
1630 goto no_cached_page;
1633 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1634 page_cache_release(page);
1635 return ret | VM_FAULT_RETRY;
1638 /* Did it get truncated? */
1639 if (unlikely(page->mapping != mapping)) {
1644 VM_BUG_ON(page->index != offset);
1647 * We have a locked page in the page cache, now we need to check
1648 * that it's up-to-date. If not, it is going to be due to an error.
1650 if (unlikely(!PageUptodate(page)))
1651 goto page_not_uptodate;
1654 * Found the page and have a reference on it.
1655 * We must recheck i_size under page lock.
1657 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1658 if (unlikely(offset >= size)) {
1660 page_cache_release(page);
1661 return VM_FAULT_SIGBUS;
1665 return ret | VM_FAULT_LOCKED;
1669 * We're only likely to ever get here if MADV_RANDOM is in
1672 error = page_cache_read(file, offset);
1675 * The page we want has now been added to the page cache.
1676 * In the unlikely event that someone removed it in the
1677 * meantime, we'll just come back here and read it again.
1683 * An error return from page_cache_read can result if the
1684 * system is low on memory, or a problem occurs while trying
1687 if (error == -ENOMEM)
1688 return VM_FAULT_OOM;
1689 return VM_FAULT_SIGBUS;
1693 * Umm, take care of errors if the page isn't up-to-date.
1694 * Try to re-read it _once_. We do this synchronously,
1695 * because there really aren't any performance issues here
1696 * and we need to check for errors.
1698 ClearPageError(page);
1699 error = mapping->a_ops->readpage(file, page);
1701 wait_on_page_locked(page);
1702 if (!PageUptodate(page))
1705 page_cache_release(page);
1707 if (!error || error == AOP_TRUNCATED_PAGE)
1710 /* Things didn't work out. Return zero to tell the mm layer so. */
1711 shrink_readahead_size_eio(file, ra);
1712 return VM_FAULT_SIGBUS;
1714 EXPORT_SYMBOL(filemap_fault);
1716 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1718 struct page *page = vmf->page;
1719 struct inode *inode = file_inode(vma->vm_file);
1720 int ret = VM_FAULT_LOCKED;
1722 sb_start_pagefault(inode->i_sb);
1723 file_update_time(vma->vm_file);
1725 if (page->mapping != inode->i_mapping) {
1727 ret = VM_FAULT_NOPAGE;
1731 * We mark the page dirty already here so that when freeze is in
1732 * progress, we are guaranteed that writeback during freezing will
1733 * see the dirty page and writeprotect it again.
1735 set_page_dirty(page);
1736 wait_for_stable_page(page);
1738 sb_end_pagefault(inode->i_sb);
1741 EXPORT_SYMBOL(filemap_page_mkwrite);
1743 const struct vm_operations_struct generic_file_vm_ops = {
1744 .fault = filemap_fault,
1745 .page_mkwrite = filemap_page_mkwrite,
1746 .remap_pages = generic_file_remap_pages,
1749 /* This is used for a general mmap of a disk file */
1751 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1753 struct address_space *mapping = file->f_mapping;
1755 if (!mapping->a_ops->readpage)
1757 file_accessed(file);
1758 vma->vm_ops = &generic_file_vm_ops;
1763 * This is for filesystems which do not implement ->writepage.
1765 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1767 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1769 return generic_file_mmap(file, vma);
1772 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1776 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1780 #endif /* CONFIG_MMU */
1782 EXPORT_SYMBOL(generic_file_mmap);
1783 EXPORT_SYMBOL(generic_file_readonly_mmap);
1785 static struct page *__read_cache_page(struct address_space *mapping,
1787 int (*filler)(void *, struct page *),
1794 page = find_get_page(mapping, index);
1796 page = __page_cache_alloc(gfp | __GFP_COLD);
1798 return ERR_PTR(-ENOMEM);
1799 err = add_to_page_cache_lru(page, mapping, index, gfp);
1800 if (unlikely(err)) {
1801 page_cache_release(page);
1804 /* Presumably ENOMEM for radix tree node */
1805 return ERR_PTR(err);
1807 err = filler(data, page);
1809 page_cache_release(page);
1810 page = ERR_PTR(err);
1816 static struct page *do_read_cache_page(struct address_space *mapping,
1818 int (*filler)(void *, struct page *),
1827 page = __read_cache_page(mapping, index, filler, data, gfp);
1830 if (PageUptodate(page))
1834 if (!page->mapping) {
1836 page_cache_release(page);
1839 if (PageUptodate(page)) {
1843 err = filler(data, page);
1845 page_cache_release(page);
1846 return ERR_PTR(err);
1849 mark_page_accessed(page);
1854 * read_cache_page_async - read into page cache, fill it if needed
1855 * @mapping: the page's address_space
1856 * @index: the page index
1857 * @filler: function to perform the read
1858 * @data: first arg to filler(data, page) function, often left as NULL
1860 * Same as read_cache_page, but don't wait for page to become unlocked
1861 * after submitting it to the filler.
1863 * Read into the page cache. If a page already exists, and PageUptodate() is
1864 * not set, try to fill the page but don't wait for it to become unlocked.
1866 * If the page does not get brought uptodate, return -EIO.
1868 struct page *read_cache_page_async(struct address_space *mapping,
1870 int (*filler)(void *, struct page *),
1873 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1875 EXPORT_SYMBOL(read_cache_page_async);
1877 static struct page *wait_on_page_read(struct page *page)
1879 if (!IS_ERR(page)) {
1880 wait_on_page_locked(page);
1881 if (!PageUptodate(page)) {
1882 page_cache_release(page);
1883 page = ERR_PTR(-EIO);
1890 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1891 * @mapping: the page's address_space
1892 * @index: the page index
1893 * @gfp: the page allocator flags to use if allocating
1895 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1896 * any new page allocations done using the specified allocation flags.
1898 * If the page does not get brought uptodate, return -EIO.
1900 struct page *read_cache_page_gfp(struct address_space *mapping,
1904 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1906 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1908 EXPORT_SYMBOL(read_cache_page_gfp);
1911 * read_cache_page - read into page cache, fill it if needed
1912 * @mapping: the page's address_space
1913 * @index: the page index
1914 * @filler: function to perform the read
1915 * @data: first arg to filler(data, page) function, often left as NULL
1917 * Read into the page cache. If a page already exists, and PageUptodate() is
1918 * not set, try to fill the page then wait for it to become unlocked.
1920 * If the page does not get brought uptodate, return -EIO.
1922 struct page *read_cache_page(struct address_space *mapping,
1924 int (*filler)(void *, struct page *),
1927 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1929 EXPORT_SYMBOL(read_cache_page);
1931 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1932 const struct iovec *iov, size_t base, size_t bytes)
1934 size_t copied = 0, left = 0;
1937 char __user *buf = iov->iov_base + base;
1938 int copy = min(bytes, iov->iov_len - base);
1941 left = __copy_from_user_inatomic(vaddr, buf, copy);
1950 return copied - left;
1954 * Copy as much as we can into the page and return the number of bytes which
1955 * were successfully copied. If a fault is encountered then return the number of
1956 * bytes which were copied.
1958 size_t iov_iter_copy_from_user_atomic(struct page *page,
1959 struct iov_iter *i, unsigned long offset, size_t bytes)
1964 BUG_ON(!in_atomic());
1965 kaddr = kmap_atomic(page);
1966 if (likely(i->nr_segs == 1)) {
1968 char __user *buf = i->iov->iov_base + i->iov_offset;
1969 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1970 copied = bytes - left;
1972 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1973 i->iov, i->iov_offset, bytes);
1975 kunmap_atomic(kaddr);
1979 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1982 * This has the same sideeffects and return value as
1983 * iov_iter_copy_from_user_atomic().
1984 * The difference is that it attempts to resolve faults.
1985 * Page must not be locked.
1987 size_t iov_iter_copy_from_user(struct page *page,
1988 struct iov_iter *i, unsigned long offset, size_t bytes)
1994 if (likely(i->nr_segs == 1)) {
1996 char __user *buf = i->iov->iov_base + i->iov_offset;
1997 left = __copy_from_user(kaddr + offset, buf, bytes);
1998 copied = bytes - left;
2000 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2001 i->iov, i->iov_offset, bytes);
2006 EXPORT_SYMBOL(iov_iter_copy_from_user);
2008 void iov_iter_advance(struct iov_iter *i, size_t bytes)
2010 BUG_ON(i->count < bytes);
2012 if (likely(i->nr_segs == 1)) {
2013 i->iov_offset += bytes;
2016 const struct iovec *iov = i->iov;
2017 size_t base = i->iov_offset;
2018 unsigned long nr_segs = i->nr_segs;
2021 * The !iov->iov_len check ensures we skip over unlikely
2022 * zero-length segments (without overruning the iovec).
2024 while (bytes || unlikely(i->count && !iov->iov_len)) {
2027 copy = min(bytes, iov->iov_len - base);
2028 BUG_ON(!i->count || i->count < copy);
2032 if (iov->iov_len == base) {
2039 i->iov_offset = base;
2040 i->nr_segs = nr_segs;
2043 EXPORT_SYMBOL(iov_iter_advance);
2046 * Fault in the first iovec of the given iov_iter, to a maximum length
2047 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2048 * accessed (ie. because it is an invalid address).
2050 * writev-intensive code may want this to prefault several iovecs -- that
2051 * would be possible (callers must not rely on the fact that _only_ the
2052 * first iovec will be faulted with the current implementation).
2054 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2056 char __user *buf = i->iov->iov_base + i->iov_offset;
2057 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2058 return fault_in_pages_readable(buf, bytes);
2060 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2063 * Return the count of just the current iov_iter segment.
2065 size_t iov_iter_single_seg_count(const struct iov_iter *i)
2067 const struct iovec *iov = i->iov;
2068 if (i->nr_segs == 1)
2071 return min(i->count, iov->iov_len - i->iov_offset);
2073 EXPORT_SYMBOL(iov_iter_single_seg_count);
2076 * Performs necessary checks before doing a write
2078 * Can adjust writing position or amount of bytes to write.
2079 * Returns appropriate error code that caller should return or
2080 * zero in case that write should be allowed.
2082 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2084 struct inode *inode = file->f_mapping->host;
2085 unsigned long limit = rlimit(RLIMIT_FSIZE);
2087 if (unlikely(*pos < 0))
2091 /* FIXME: this is for backwards compatibility with 2.4 */
2092 if (file->f_flags & O_APPEND)
2093 *pos = i_size_read(inode);
2095 if (limit != RLIM_INFINITY) {
2096 if (*pos >= limit) {
2097 send_sig(SIGXFSZ, current, 0);
2100 if (*count > limit - (typeof(limit))*pos) {
2101 *count = limit - (typeof(limit))*pos;
2109 if (unlikely(*pos + *count > MAX_NON_LFS &&
2110 !(file->f_flags & O_LARGEFILE))) {
2111 if (*pos >= MAX_NON_LFS) {
2114 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2115 *count = MAX_NON_LFS - (unsigned long)*pos;
2120 * Are we about to exceed the fs block limit ?
2122 * If we have written data it becomes a short write. If we have
2123 * exceeded without writing data we send a signal and return EFBIG.
2124 * Linus frestrict idea will clean these up nicely..
2126 if (likely(!isblk)) {
2127 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2128 if (*count || *pos > inode->i_sb->s_maxbytes) {
2131 /* zero-length writes at ->s_maxbytes are OK */
2134 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2135 *count = inode->i_sb->s_maxbytes - *pos;
2139 if (bdev_read_only(I_BDEV(inode)))
2141 isize = i_size_read(inode);
2142 if (*pos >= isize) {
2143 if (*count || *pos > isize)
2147 if (*pos + *count > isize)
2148 *count = isize - *pos;
2155 EXPORT_SYMBOL(generic_write_checks);
2157 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2158 loff_t pos, unsigned len, unsigned flags,
2159 struct page **pagep, void **fsdata)
2161 const struct address_space_operations *aops = mapping->a_ops;
2163 return aops->write_begin(file, mapping, pos, len, flags,
2166 EXPORT_SYMBOL(pagecache_write_begin);
2168 int pagecache_write_end(struct file *file, struct address_space *mapping,
2169 loff_t pos, unsigned len, unsigned copied,
2170 struct page *page, void *fsdata)
2172 const struct address_space_operations *aops = mapping->a_ops;
2174 mark_page_accessed(page);
2175 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2177 EXPORT_SYMBOL(pagecache_write_end);
2180 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2181 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2182 size_t count, size_t ocount)
2184 struct file *file = iocb->ki_filp;
2185 struct address_space *mapping = file->f_mapping;
2186 struct inode *inode = mapping->host;
2191 if (count != ocount)
2192 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2194 write_len = iov_length(iov, *nr_segs);
2195 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2197 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2202 * After a write we want buffered reads to be sure to go to disk to get
2203 * the new data. We invalidate clean cached page from the region we're
2204 * about to write. We do this *before* the write so that we can return
2205 * without clobbering -EIOCBQUEUED from ->direct_IO().
2207 if (mapping->nrpages) {
2208 written = invalidate_inode_pages2_range(mapping,
2209 pos >> PAGE_CACHE_SHIFT, end);
2211 * If a page can not be invalidated, return 0 to fall back
2212 * to buffered write.
2215 if (written == -EBUSY)
2221 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2224 * Finally, try again to invalidate clean pages which might have been
2225 * cached by non-direct readahead, or faulted in by get_user_pages()
2226 * if the source of the write was an mmap'ed region of the file
2227 * we're writing. Either one is a pretty crazy thing to do,
2228 * so we don't support it 100%. If this invalidation
2229 * fails, tough, the write still worked...
2231 if (mapping->nrpages) {
2232 invalidate_inode_pages2_range(mapping,
2233 pos >> PAGE_CACHE_SHIFT, end);
2238 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2239 i_size_write(inode, pos);
2240 mark_inode_dirty(inode);
2247 EXPORT_SYMBOL(generic_file_direct_write);
2250 * Find or create a page at the given pagecache position. Return the locked
2251 * page. This function is specifically for buffered writes.
2253 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2254 pgoff_t index, unsigned flags)
2259 gfp_t gfp_notmask = 0;
2261 gfp_mask = mapping_gfp_mask(mapping);
2262 if (mapping_cap_account_dirty(mapping))
2263 gfp_mask |= __GFP_WRITE;
2264 if (flags & AOP_FLAG_NOFS)
2265 gfp_notmask = __GFP_FS;
2267 page = find_lock_page(mapping, index);
2271 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2274 status = add_to_page_cache_lru(page, mapping, index,
2275 GFP_KERNEL & ~gfp_notmask);
2276 if (unlikely(status)) {
2277 page_cache_release(page);
2278 if (status == -EEXIST)
2283 wait_for_stable_page(page);
2286 EXPORT_SYMBOL(grab_cache_page_write_begin);
2288 static ssize_t generic_perform_write(struct file *file,
2289 struct iov_iter *i, loff_t pos)
2291 struct address_space *mapping = file->f_mapping;
2292 const struct address_space_operations *a_ops = mapping->a_ops;
2294 ssize_t written = 0;
2295 unsigned int flags = 0;
2298 * Copies from kernel address space cannot fail (NFSD is a big user).
2300 if (segment_eq(get_fs(), KERNEL_DS))
2301 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2305 unsigned long offset; /* Offset into pagecache page */
2306 unsigned long bytes; /* Bytes to write to page */
2307 size_t copied; /* Bytes copied from user */
2310 offset = (pos & (PAGE_CACHE_SIZE - 1));
2311 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2316 * Bring in the user page that we will copy from _first_.
2317 * Otherwise there's a nasty deadlock on copying from the
2318 * same page as we're writing to, without it being marked
2321 * Not only is this an optimisation, but it is also required
2322 * to check that the address is actually valid, when atomic
2323 * usercopies are used, below.
2325 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2330 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2332 if (unlikely(status))
2335 if (mapping_writably_mapped(mapping))
2336 flush_dcache_page(page);
2338 pagefault_disable();
2339 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2341 flush_dcache_page(page);
2343 mark_page_accessed(page);
2344 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2346 if (unlikely(status < 0))
2352 iov_iter_advance(i, copied);
2353 if (unlikely(copied == 0)) {
2355 * If we were unable to copy any data at all, we must
2356 * fall back to a single segment length write.
2358 * If we didn't fallback here, we could livelock
2359 * because not all segments in the iov can be copied at
2360 * once without a pagefault.
2362 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2363 iov_iter_single_seg_count(i));
2369 balance_dirty_pages_ratelimited(mapping);
2370 if (fatal_signal_pending(current)) {
2374 } while (iov_iter_count(i));
2376 return written ? written : status;
2380 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2381 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2382 size_t count, ssize_t written)
2384 struct file *file = iocb->ki_filp;
2388 iov_iter_init(&i, iov, nr_segs, count, written);
2389 status = generic_perform_write(file, &i, pos);
2391 if (likely(status >= 0)) {
2393 *ppos = pos + status;
2396 return written ? written : status;
2398 EXPORT_SYMBOL(generic_file_buffered_write);
2401 * __generic_file_aio_write - write data to a file
2402 * @iocb: IO state structure (file, offset, etc.)
2403 * @iov: vector with data to write
2404 * @nr_segs: number of segments in the vector
2405 * @ppos: position where to write
2407 * This function does all the work needed for actually writing data to a
2408 * file. It does all basic checks, removes SUID from the file, updates
2409 * modification times and calls proper subroutines depending on whether we
2410 * do direct IO or a standard buffered write.
2412 * It expects i_mutex to be grabbed unless we work on a block device or similar
2413 * object which does not need locking at all.
2415 * This function does *not* take care of syncing data in case of O_SYNC write.
2416 * A caller has to handle it. This is mainly due to the fact that we want to
2417 * avoid syncing under i_mutex.
2419 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2420 unsigned long nr_segs, loff_t *ppos)
2422 struct file *file = iocb->ki_filp;
2423 struct address_space * mapping = file->f_mapping;
2424 size_t ocount; /* original count */
2425 size_t count; /* after file limit checks */
2426 struct inode *inode = mapping->host;
2432 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2439 /* We can write back this queue in page reclaim */
2440 current->backing_dev_info = mapping->backing_dev_info;
2443 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2450 err = file_remove_suid(file);
2454 err = file_update_time(file);
2458 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2459 if (unlikely(file->f_flags & O_DIRECT)) {
2461 ssize_t written_buffered;
2463 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2464 ppos, count, ocount);
2465 if (written < 0 || written == count)
2468 * direct-io write to a hole: fall through to buffered I/O
2469 * for completing the rest of the request.
2473 written_buffered = generic_file_buffered_write(iocb, iov,
2474 nr_segs, pos, ppos, count,
2477 * If generic_file_buffered_write() retuned a synchronous error
2478 * then we want to return the number of bytes which were
2479 * direct-written, or the error code if that was zero. Note
2480 * that this differs from normal direct-io semantics, which
2481 * will return -EFOO even if some bytes were written.
2483 if (written_buffered < 0) {
2484 err = written_buffered;
2489 * We need to ensure that the page cache pages are written to
2490 * disk and invalidated to preserve the expected O_DIRECT
2493 endbyte = pos + written_buffered - written - 1;
2494 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2496 written = written_buffered;
2497 invalidate_mapping_pages(mapping,
2498 pos >> PAGE_CACHE_SHIFT,
2499 endbyte >> PAGE_CACHE_SHIFT);
2502 * We don't know how much we wrote, so just return
2503 * the number of bytes which were direct-written
2507 written = generic_file_buffered_write(iocb, iov, nr_segs,
2508 pos, ppos, count, written);
2511 current->backing_dev_info = NULL;
2512 return written ? written : err;
2514 EXPORT_SYMBOL(__generic_file_aio_write);
2517 * generic_file_aio_write - write data to a file
2518 * @iocb: IO state structure
2519 * @iov: vector with data to write
2520 * @nr_segs: number of segments in the vector
2521 * @pos: position in file where to write
2523 * This is a wrapper around __generic_file_aio_write() to be used by most
2524 * filesystems. It takes care of syncing the file in case of O_SYNC file
2525 * and acquires i_mutex as needed.
2527 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2528 unsigned long nr_segs, loff_t pos)
2530 struct file *file = iocb->ki_filp;
2531 struct inode *inode = file->f_mapping->host;
2534 BUG_ON(iocb->ki_pos != pos);
2536 if (!sb_start_file_write(file))
2538 mutex_lock(&inode->i_mutex);
2539 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2540 mutex_unlock(&inode->i_mutex);
2542 if (ret > 0 || ret == -EIOCBQUEUED) {
2545 err = generic_write_sync(file, pos, ret);
2546 if (err < 0 && ret > 0)
2549 sb_end_write(inode->i_sb);
2552 EXPORT_SYMBOL(generic_file_aio_write);
2555 * try_to_release_page() - release old fs-specific metadata on a page
2557 * @page: the page which the kernel is trying to free
2558 * @gfp_mask: memory allocation flags (and I/O mode)
2560 * The address_space is to try to release any data against the page
2561 * (presumably at page->private). If the release was successful, return `1'.
2562 * Otherwise return zero.
2564 * This may also be called if PG_fscache is set on a page, indicating that the
2565 * page is known to the local caching routines.
2567 * The @gfp_mask argument specifies whether I/O may be performed to release
2568 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2571 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2573 struct address_space * const mapping = page->mapping;
2575 BUG_ON(!PageLocked(page));
2576 if (PageWriteback(page))
2579 if (mapping && mapping->a_ops->releasepage)
2580 return mapping->a_ops->releasepage(page, gfp_mask);
2581 return try_to_free_buffers(page);
2584 EXPORT_SYMBOL(try_to_release_page);