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;
191 static int filemap_check_errors(struct address_space *mapping)
194 /* Check for outstanding write errors */
195 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
197 if (test_and_clear_bit(AS_EIO, &mapping->flags))
203 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
204 * @mapping: address space structure to write
205 * @start: offset in bytes where the range starts
206 * @end: offset in bytes where the range ends (inclusive)
207 * @sync_mode: enable synchronous operation
209 * Start writeback against all of a mapping's dirty pages that lie
210 * within the byte offsets <start, end> inclusive.
212 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
213 * opposed to a regular memory cleansing writeback. The difference between
214 * these two operations is that if a dirty page/buffer is encountered, it must
215 * be waited upon, and not just skipped over.
217 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
218 loff_t end, int sync_mode)
221 struct writeback_control wbc = {
222 .sync_mode = sync_mode,
223 .nr_to_write = LONG_MAX,
224 .range_start = start,
228 if (!mapping_cap_writeback_dirty(mapping))
231 ret = do_writepages(mapping, &wbc);
235 static inline int __filemap_fdatawrite(struct address_space *mapping,
238 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
241 int filemap_fdatawrite(struct address_space *mapping)
243 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
245 EXPORT_SYMBOL(filemap_fdatawrite);
247 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
250 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
252 EXPORT_SYMBOL(filemap_fdatawrite_range);
255 * filemap_flush - mostly a non-blocking flush
256 * @mapping: target address_space
258 * This is a mostly non-blocking flush. Not suitable for data-integrity
259 * purposes - I/O may not be started against all dirty pages.
261 int filemap_flush(struct address_space *mapping)
263 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
265 EXPORT_SYMBOL(filemap_flush);
268 * filemap_fdatawait_range - wait for writeback to complete
269 * @mapping: address space structure to wait for
270 * @start_byte: offset in bytes where the range starts
271 * @end_byte: offset in bytes where the range ends (inclusive)
273 * Walk the list of under-writeback pages of the given address space
274 * in the given range and wait for all of them.
276 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
279 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
280 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
285 if (end_byte < start_byte)
288 pagevec_init(&pvec, 0);
289 while ((index <= end) &&
290 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
291 PAGECACHE_TAG_WRITEBACK,
292 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
295 for (i = 0; i < nr_pages; i++) {
296 struct page *page = pvec.pages[i];
298 /* until radix tree lookup accepts end_index */
299 if (page->index > end)
302 wait_on_page_writeback(page);
303 if (TestClearPageError(page))
306 pagevec_release(&pvec);
310 ret2 = filemap_check_errors(mapping);
316 EXPORT_SYMBOL(filemap_fdatawait_range);
319 * filemap_fdatawait - wait for all under-writeback pages to complete
320 * @mapping: address space structure to wait for
322 * Walk the list of under-writeback pages of the given address space
323 * and wait for all of them.
325 int filemap_fdatawait(struct address_space *mapping)
327 loff_t i_size = i_size_read(mapping->host);
332 return filemap_fdatawait_range(mapping, 0, i_size - 1);
334 EXPORT_SYMBOL(filemap_fdatawait);
336 int filemap_write_and_wait(struct address_space *mapping)
340 if (mapping->nrpages) {
341 err = filemap_fdatawrite(mapping);
343 * Even if the above returned error, the pages may be
344 * written partially (e.g. -ENOSPC), so we wait for it.
345 * But the -EIO is special case, it may indicate the worst
346 * thing (e.g. bug) happened, so we avoid waiting for it.
349 int err2 = filemap_fdatawait(mapping);
354 err = filemap_check_errors(mapping);
358 EXPORT_SYMBOL(filemap_write_and_wait);
361 * filemap_write_and_wait_range - write out & wait on a file range
362 * @mapping: the address_space for the pages
363 * @lstart: offset in bytes where the range starts
364 * @lend: offset in bytes where the range ends (inclusive)
366 * Write out and wait upon file offsets lstart->lend, inclusive.
368 * Note that `lend' is inclusive (describes the last byte to be written) so
369 * that this function can be used to write to the very end-of-file (end = -1).
371 int filemap_write_and_wait_range(struct address_space *mapping,
372 loff_t lstart, loff_t lend)
376 if (mapping->nrpages) {
377 err = __filemap_fdatawrite_range(mapping, lstart, lend,
379 /* See comment of filemap_write_and_wait() */
381 int err2 = filemap_fdatawait_range(mapping,
387 err = filemap_check_errors(mapping);
391 EXPORT_SYMBOL(filemap_write_and_wait_range);
394 * replace_page_cache_page - replace a pagecache page with a new one
395 * @old: page to be replaced
396 * @new: page to replace with
397 * @gfp_mask: allocation mode
399 * This function replaces a page in the pagecache with a new one. On
400 * success it acquires the pagecache reference for the new page and
401 * drops it for the old page. Both the old and new pages must be
402 * locked. This function does not add the new page to the LRU, the
403 * caller must do that.
405 * The remove + add is atomic. The only way this function can fail is
406 * memory allocation failure.
408 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
412 VM_BUG_ON(!PageLocked(old));
413 VM_BUG_ON(!PageLocked(new));
414 VM_BUG_ON(new->mapping);
416 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
418 struct address_space *mapping = old->mapping;
419 void (*freepage)(struct page *);
421 pgoff_t offset = old->index;
422 freepage = mapping->a_ops->freepage;
425 new->mapping = mapping;
428 spin_lock_irq(&mapping->tree_lock);
429 __delete_from_page_cache(old);
430 error = radix_tree_insert(&mapping->page_tree, offset, new);
433 __inc_zone_page_state(new, NR_FILE_PAGES);
434 if (PageSwapBacked(new))
435 __inc_zone_page_state(new, NR_SHMEM);
436 spin_unlock_irq(&mapping->tree_lock);
437 /* mem_cgroup codes must not be called under tree_lock */
438 mem_cgroup_replace_page_cache(old, new);
439 radix_tree_preload_end();
442 page_cache_release(old);
447 EXPORT_SYMBOL_GPL(replace_page_cache_page);
450 * add_to_page_cache_locked - add a locked page to the pagecache
452 * @mapping: the page's address_space
453 * @offset: page index
454 * @gfp_mask: page allocation mode
456 * This function is used to add a page to the pagecache. It must be locked.
457 * This function does not add the page to the LRU. The caller must do that.
459 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
460 pgoff_t offset, gfp_t gfp_mask)
464 VM_BUG_ON(!PageLocked(page));
465 VM_BUG_ON(PageSwapBacked(page));
467 error = mem_cgroup_cache_charge(page, current->mm,
468 gfp_mask & GFP_RECLAIM_MASK);
472 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
474 page_cache_get(page);
475 page->mapping = mapping;
476 page->index = offset;
478 spin_lock_irq(&mapping->tree_lock);
479 error = radix_tree_insert(&mapping->page_tree, offset, page);
480 if (likely(!error)) {
482 __inc_zone_page_state(page, NR_FILE_PAGES);
483 spin_unlock_irq(&mapping->tree_lock);
484 trace_mm_filemap_add_to_page_cache(page);
486 page->mapping = NULL;
487 /* Leave page->index set: truncation relies upon it */
488 spin_unlock_irq(&mapping->tree_lock);
489 mem_cgroup_uncharge_cache_page(page);
490 page_cache_release(page);
492 radix_tree_preload_end();
494 mem_cgroup_uncharge_cache_page(page);
498 EXPORT_SYMBOL(add_to_page_cache_locked);
500 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
501 pgoff_t offset, gfp_t gfp_mask)
505 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
507 lru_cache_add_file(page);
510 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
513 struct page *__page_cache_alloc(gfp_t gfp)
518 if (cpuset_do_page_mem_spread()) {
519 unsigned int cpuset_mems_cookie;
521 cpuset_mems_cookie = get_mems_allowed();
522 n = cpuset_mem_spread_node();
523 page = alloc_pages_exact_node(n, gfp, 0);
524 } while (!put_mems_allowed(cpuset_mems_cookie) && !page);
528 return alloc_pages(gfp, 0);
530 EXPORT_SYMBOL(__page_cache_alloc);
534 * In order to wait for pages to become available there must be
535 * waitqueues associated with pages. By using a hash table of
536 * waitqueues where the bucket discipline is to maintain all
537 * waiters on the same queue and wake all when any of the pages
538 * become available, and for the woken contexts to check to be
539 * sure the appropriate page became available, this saves space
540 * at a cost of "thundering herd" phenomena during rare hash
543 static wait_queue_head_t *page_waitqueue(struct page *page)
545 const struct zone *zone = page_zone(page);
547 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
550 static inline void wake_up_page(struct page *page, int bit)
552 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
555 void wait_on_page_bit(struct page *page, int bit_nr)
557 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
559 if (test_bit(bit_nr, &page->flags))
560 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
561 TASK_UNINTERRUPTIBLE);
563 EXPORT_SYMBOL(wait_on_page_bit);
565 int wait_on_page_bit_killable(struct page *page, int bit_nr)
567 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
569 if (!test_bit(bit_nr, &page->flags))
572 return __wait_on_bit(page_waitqueue(page), &wait,
573 sleep_on_page_killable, TASK_KILLABLE);
577 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
578 * @page: Page defining the wait queue of interest
579 * @waiter: Waiter to add to the queue
581 * Add an arbitrary @waiter to the wait queue for the nominated @page.
583 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
585 wait_queue_head_t *q = page_waitqueue(page);
588 spin_lock_irqsave(&q->lock, flags);
589 __add_wait_queue(q, waiter);
590 spin_unlock_irqrestore(&q->lock, flags);
592 EXPORT_SYMBOL_GPL(add_page_wait_queue);
595 * unlock_page - unlock a locked page
598 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
599 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
600 * mechananism between PageLocked pages and PageWriteback pages is shared.
601 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
603 * The mb is necessary to enforce ordering between the clear_bit and the read
604 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
606 void unlock_page(struct page *page)
608 VM_BUG_ON(!PageLocked(page));
609 clear_bit_unlock(PG_locked, &page->flags);
610 smp_mb__after_clear_bit();
611 wake_up_page(page, PG_locked);
613 EXPORT_SYMBOL(unlock_page);
616 * end_page_writeback - end writeback against a page
619 void end_page_writeback(struct page *page)
621 if (TestClearPageReclaim(page))
622 rotate_reclaimable_page(page);
624 if (!test_clear_page_writeback(page))
627 smp_mb__after_clear_bit();
628 wake_up_page(page, PG_writeback);
630 EXPORT_SYMBOL(end_page_writeback);
633 * __lock_page - get a lock on the page, assuming we need to sleep to get it
634 * @page: the page to lock
636 void __lock_page(struct page *page)
638 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
640 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
641 TASK_UNINTERRUPTIBLE);
643 EXPORT_SYMBOL(__lock_page);
645 int __lock_page_killable(struct page *page)
647 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
649 return __wait_on_bit_lock(page_waitqueue(page), &wait,
650 sleep_on_page_killable, TASK_KILLABLE);
652 EXPORT_SYMBOL_GPL(__lock_page_killable);
654 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
657 if (flags & FAULT_FLAG_ALLOW_RETRY) {
659 * CAUTION! In this case, mmap_sem is not released
660 * even though return 0.
662 if (flags & FAULT_FLAG_RETRY_NOWAIT)
665 up_read(&mm->mmap_sem);
666 if (flags & FAULT_FLAG_KILLABLE)
667 wait_on_page_locked_killable(page);
669 wait_on_page_locked(page);
672 if (flags & FAULT_FLAG_KILLABLE) {
675 ret = __lock_page_killable(page);
677 up_read(&mm->mmap_sem);
687 * find_get_page - find and get a page reference
688 * @mapping: the address_space to search
689 * @offset: the page index
691 * Is there a pagecache struct page at the given (mapping, offset) tuple?
692 * If yes, increment its refcount and return it; if no, return NULL.
694 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
702 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
704 page = radix_tree_deref_slot(pagep);
707 if (radix_tree_exception(page)) {
708 if (radix_tree_deref_retry(page))
711 * Otherwise, shmem/tmpfs must be storing a swap entry
712 * here as an exceptional entry: so return it without
713 * attempting to raise page count.
717 if (!page_cache_get_speculative(page))
721 * Has the page moved?
722 * This is part of the lockless pagecache protocol. See
723 * include/linux/pagemap.h for details.
725 if (unlikely(page != *pagep)) {
726 page_cache_release(page);
735 EXPORT_SYMBOL(find_get_page);
738 * find_lock_page - locate, pin and lock a pagecache page
739 * @mapping: the address_space to search
740 * @offset: the page index
742 * Locates the desired pagecache page, locks it, increments its reference
743 * count and returns its address.
745 * Returns zero if the page was not present. find_lock_page() may sleep.
747 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
752 page = find_get_page(mapping, offset);
753 if (page && !radix_tree_exception(page)) {
755 /* Has the page been truncated? */
756 if (unlikely(page->mapping != mapping)) {
758 page_cache_release(page);
761 VM_BUG_ON(page->index != offset);
765 EXPORT_SYMBOL(find_lock_page);
768 * find_or_create_page - locate or add a pagecache page
769 * @mapping: the page's address_space
770 * @index: the page's index into the mapping
771 * @gfp_mask: page allocation mode
773 * Locates a page in the pagecache. If the page is not present, a new page
774 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
775 * LRU list. The returned page is locked and has its reference count
778 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
781 * find_or_create_page() returns the desired page's address, or zero on
784 struct page *find_or_create_page(struct address_space *mapping,
785 pgoff_t index, gfp_t gfp_mask)
790 page = find_lock_page(mapping, index);
792 page = __page_cache_alloc(gfp_mask);
796 * We want a regular kernel memory (not highmem or DMA etc)
797 * allocation for the radix tree nodes, but we need to honour
798 * the context-specific requirements the caller has asked for.
799 * GFP_RECLAIM_MASK collects those requirements.
801 err = add_to_page_cache_lru(page, mapping, index,
802 (gfp_mask & GFP_RECLAIM_MASK));
804 page_cache_release(page);
812 EXPORT_SYMBOL(find_or_create_page);
815 * find_get_pages - gang pagecache lookup
816 * @mapping: The address_space to search
817 * @start: The starting page index
818 * @nr_pages: The maximum number of pages
819 * @pages: Where the resulting pages are placed
821 * find_get_pages() will search for and return a group of up to
822 * @nr_pages pages in the mapping. The pages are placed at @pages.
823 * find_get_pages() takes a reference against the returned pages.
825 * The search returns a group of mapping-contiguous pages with ascending
826 * indexes. There may be holes in the indices due to not-present pages.
828 * find_get_pages() returns the number of pages which were found.
830 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
831 unsigned int nr_pages, struct page **pages)
833 struct radix_tree_iter iter;
837 if (unlikely(!nr_pages))
842 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
845 page = radix_tree_deref_slot(slot);
849 if (radix_tree_exception(page)) {
850 if (radix_tree_deref_retry(page)) {
852 * Transient condition which can only trigger
853 * when entry at index 0 moves out of or back
854 * to root: none yet gotten, safe to restart.
860 * Otherwise, shmem/tmpfs must be storing a swap entry
861 * here as an exceptional entry: so skip over it -
862 * we only reach this from invalidate_mapping_pages().
867 if (!page_cache_get_speculative(page))
870 /* Has the page moved? */
871 if (unlikely(page != *slot)) {
872 page_cache_release(page);
877 if (++ret == nr_pages)
886 * find_get_pages_contig - gang contiguous pagecache lookup
887 * @mapping: The address_space to search
888 * @index: The starting page index
889 * @nr_pages: The maximum number of pages
890 * @pages: Where the resulting pages are placed
892 * find_get_pages_contig() works exactly like find_get_pages(), except
893 * that the returned number of pages are guaranteed to be contiguous.
895 * find_get_pages_contig() returns the number of pages which were found.
897 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
898 unsigned int nr_pages, struct page **pages)
900 struct radix_tree_iter iter;
902 unsigned int ret = 0;
904 if (unlikely(!nr_pages))
909 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
912 page = radix_tree_deref_slot(slot);
913 /* The hole, there no reason to continue */
917 if (radix_tree_exception(page)) {
918 if (radix_tree_deref_retry(page)) {
920 * Transient condition which can only trigger
921 * when entry at index 0 moves out of or back
922 * to root: none yet gotten, safe to restart.
927 * Otherwise, shmem/tmpfs must be storing a swap entry
928 * here as an exceptional entry: so stop looking for
934 if (!page_cache_get_speculative(page))
937 /* Has the page moved? */
938 if (unlikely(page != *slot)) {
939 page_cache_release(page);
944 * must check mapping and index after taking the ref.
945 * otherwise we can get both false positives and false
946 * negatives, which is just confusing to the caller.
948 if (page->mapping == NULL || page->index != iter.index) {
949 page_cache_release(page);
954 if (++ret == nr_pages)
960 EXPORT_SYMBOL(find_get_pages_contig);
963 * find_get_pages_tag - find and return pages that match @tag
964 * @mapping: the address_space to search
965 * @index: the starting page index
966 * @tag: the tag index
967 * @nr_pages: the maximum number of pages
968 * @pages: where the resulting pages are placed
970 * Like find_get_pages, except we only return pages which are tagged with
971 * @tag. We update @index to index the next page for the traversal.
973 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
974 int tag, unsigned int nr_pages, struct page **pages)
976 struct radix_tree_iter iter;
980 if (unlikely(!nr_pages))
985 radix_tree_for_each_tagged(slot, &mapping->page_tree,
986 &iter, *index, tag) {
989 page = radix_tree_deref_slot(slot);
993 if (radix_tree_exception(page)) {
994 if (radix_tree_deref_retry(page)) {
996 * Transient condition which can only trigger
997 * when entry at index 0 moves out of or back
998 * to root: none yet gotten, safe to restart.
1003 * This function is never used on a shmem/tmpfs
1004 * mapping, so a swap entry won't be found here.
1009 if (!page_cache_get_speculative(page))
1012 /* Has the page moved? */
1013 if (unlikely(page != *slot)) {
1014 page_cache_release(page);
1019 if (++ret == nr_pages)
1026 *index = pages[ret - 1]->index + 1;
1030 EXPORT_SYMBOL(find_get_pages_tag);
1033 * grab_cache_page_nowait - returns locked page at given index in given cache
1034 * @mapping: target address_space
1035 * @index: the page index
1037 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1038 * This is intended for speculative data generators, where the data can
1039 * be regenerated if the page couldn't be grabbed. This routine should
1040 * be safe to call while holding the lock for another page.
1042 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1043 * and deadlock against the caller's locked page.
1046 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1048 struct page *page = find_get_page(mapping, index);
1051 if (trylock_page(page))
1053 page_cache_release(page);
1056 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1057 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1058 page_cache_release(page);
1063 EXPORT_SYMBOL(grab_cache_page_nowait);
1066 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1067 * a _large_ part of the i/o request. Imagine the worst scenario:
1069 * ---R__________________________________________B__________
1070 * ^ reading here ^ bad block(assume 4k)
1072 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1073 * => failing the whole request => read(R) => read(R+1) =>
1074 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1075 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1076 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1078 * It is going insane. Fix it by quickly scaling down the readahead size.
1080 static void shrink_readahead_size_eio(struct file *filp,
1081 struct file_ra_state *ra)
1087 * do_generic_file_read - generic file read routine
1088 * @filp: the file to read
1089 * @ppos: current file position
1090 * @desc: read_descriptor
1091 * @actor: read method
1093 * This is a generic file read routine, and uses the
1094 * mapping->a_ops->readpage() function for the actual low-level stuff.
1096 * This is really ugly. But the goto's actually try to clarify some
1097 * of the logic when it comes to error handling etc.
1099 static void do_generic_file_read(struct file *filp, loff_t *ppos,
1100 read_descriptor_t *desc, read_actor_t actor)
1102 struct address_space *mapping = filp->f_mapping;
1103 struct inode *inode = mapping->host;
1104 struct file_ra_state *ra = &filp->f_ra;
1108 unsigned long offset; /* offset into pagecache page */
1109 unsigned int prev_offset;
1112 index = *ppos >> PAGE_CACHE_SHIFT;
1113 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1114 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1115 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1116 offset = *ppos & ~PAGE_CACHE_MASK;
1122 unsigned long nr, ret;
1126 page = find_get_page(mapping, index);
1128 page_cache_sync_readahead(mapping,
1130 index, last_index - index);
1131 page = find_get_page(mapping, index);
1132 if (unlikely(page == NULL))
1133 goto no_cached_page;
1135 if (PageReadahead(page)) {
1136 page_cache_async_readahead(mapping,
1138 index, last_index - index);
1140 if (!PageUptodate(page)) {
1141 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1142 !mapping->a_ops->is_partially_uptodate)
1143 goto page_not_up_to_date;
1144 if (!trylock_page(page))
1145 goto page_not_up_to_date;
1146 /* Did it get truncated before we got the lock? */
1148 goto page_not_up_to_date_locked;
1149 if (!mapping->a_ops->is_partially_uptodate(page,
1151 goto page_not_up_to_date_locked;
1156 * i_size must be checked after we know the page is Uptodate.
1158 * Checking i_size after the check allows us to calculate
1159 * the correct value for "nr", which means the zero-filled
1160 * part of the page is not copied back to userspace (unless
1161 * another truncate extends the file - this is desired though).
1164 isize = i_size_read(inode);
1165 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1166 if (unlikely(!isize || index > end_index)) {
1167 page_cache_release(page);
1171 /* nr is the maximum number of bytes to copy from this page */
1172 nr = PAGE_CACHE_SIZE;
1173 if (index == end_index) {
1174 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1176 page_cache_release(page);
1182 /* If users can be writing to this page using arbitrary
1183 * virtual addresses, take care about potential aliasing
1184 * before reading the page on the kernel side.
1186 if (mapping_writably_mapped(mapping))
1187 flush_dcache_page(page);
1190 * When a sequential read accesses a page several times,
1191 * only mark it as accessed the first time.
1193 if (prev_index != index || offset != prev_offset)
1194 mark_page_accessed(page);
1198 * Ok, we have the page, and it's up-to-date, so
1199 * now we can copy it to user space...
1201 * The actor routine returns how many bytes were actually used..
1202 * NOTE! This may not be the same as how much of a user buffer
1203 * we filled up (we may be padding etc), so we can only update
1204 * "pos" here (the actor routine has to update the user buffer
1205 * pointers and the remaining count).
1207 ret = actor(desc, page, offset, nr);
1209 index += offset >> PAGE_CACHE_SHIFT;
1210 offset &= ~PAGE_CACHE_MASK;
1211 prev_offset = offset;
1213 page_cache_release(page);
1214 if (ret == nr && desc->count)
1218 page_not_up_to_date:
1219 /* Get exclusive access to the page ... */
1220 error = lock_page_killable(page);
1221 if (unlikely(error))
1222 goto readpage_error;
1224 page_not_up_to_date_locked:
1225 /* Did it get truncated before we got the lock? */
1226 if (!page->mapping) {
1228 page_cache_release(page);
1232 /* Did somebody else fill it already? */
1233 if (PageUptodate(page)) {
1240 * A previous I/O error may have been due to temporary
1241 * failures, eg. multipath errors.
1242 * PG_error will be set again if readpage fails.
1244 ClearPageError(page);
1245 /* Start the actual read. The read will unlock the page. */
1246 error = mapping->a_ops->readpage(filp, page);
1248 if (unlikely(error)) {
1249 if (error == AOP_TRUNCATED_PAGE) {
1250 page_cache_release(page);
1253 goto readpage_error;
1256 if (!PageUptodate(page)) {
1257 error = lock_page_killable(page);
1258 if (unlikely(error))
1259 goto readpage_error;
1260 if (!PageUptodate(page)) {
1261 if (page->mapping == NULL) {
1263 * invalidate_mapping_pages got it
1266 page_cache_release(page);
1270 shrink_readahead_size_eio(filp, ra);
1272 goto readpage_error;
1280 /* UHHUH! A synchronous read error occurred. Report it */
1281 desc->error = error;
1282 page_cache_release(page);
1287 * Ok, it wasn't cached, so we need to create a new
1290 page = page_cache_alloc_cold(mapping);
1292 desc->error = -ENOMEM;
1295 error = add_to_page_cache_lru(page, mapping,
1298 page_cache_release(page);
1299 if (error == -EEXIST)
1301 desc->error = error;
1308 ra->prev_pos = prev_index;
1309 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1310 ra->prev_pos |= prev_offset;
1312 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1313 file_accessed(filp);
1316 int file_read_actor(read_descriptor_t *desc, struct page *page,
1317 unsigned long offset, unsigned long size)
1320 unsigned long left, count = desc->count;
1326 * Faults on the destination of a read are common, so do it before
1329 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1330 kaddr = kmap_atomic(page);
1331 left = __copy_to_user_inatomic(desc->arg.buf,
1332 kaddr + offset, size);
1333 kunmap_atomic(kaddr);
1338 /* Do it the slow way */
1340 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1345 desc->error = -EFAULT;
1348 desc->count = count - size;
1349 desc->written += size;
1350 desc->arg.buf += size;
1355 * Performs necessary checks before doing a write
1356 * @iov: io vector request
1357 * @nr_segs: number of segments in the iovec
1358 * @count: number of bytes to write
1359 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1361 * Adjust number of segments and amount of bytes to write (nr_segs should be
1362 * properly initialized first). Returns appropriate error code that caller
1363 * should return or zero in case that write should be allowed.
1365 int generic_segment_checks(const struct iovec *iov,
1366 unsigned long *nr_segs, size_t *count, int access_flags)
1370 for (seg = 0; seg < *nr_segs; seg++) {
1371 const struct iovec *iv = &iov[seg];
1374 * If any segment has a negative length, or the cumulative
1375 * length ever wraps negative then return -EINVAL.
1378 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1380 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1385 cnt -= iv->iov_len; /* This segment is no good */
1391 EXPORT_SYMBOL(generic_segment_checks);
1394 * generic_file_aio_read - generic filesystem read routine
1395 * @iocb: kernel I/O control block
1396 * @iov: io vector request
1397 * @nr_segs: number of segments in the iovec
1398 * @pos: current file position
1400 * This is the "read()" routine for all filesystems
1401 * that can use the page cache directly.
1404 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1405 unsigned long nr_segs, loff_t pos)
1407 struct file *filp = iocb->ki_filp;
1409 unsigned long seg = 0;
1411 loff_t *ppos = &iocb->ki_pos;
1414 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1418 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1419 if (filp->f_flags & O_DIRECT) {
1421 struct address_space *mapping;
1422 struct inode *inode;
1424 mapping = filp->f_mapping;
1425 inode = mapping->host;
1427 goto out; /* skip atime */
1428 size = i_size_read(inode);
1430 retval = filemap_write_and_wait_range(mapping, pos,
1431 pos + iov_length(iov, nr_segs) - 1);
1433 retval = mapping->a_ops->direct_IO(READ, iocb,
1437 *ppos = pos + retval;
1442 * Btrfs can have a short DIO read if we encounter
1443 * compressed extents, so if there was an error, or if
1444 * we've already read everything we wanted to, or if
1445 * there was a short read because we hit EOF, go ahead
1446 * and return. Otherwise fallthrough to buffered io for
1447 * the rest of the read.
1449 if (retval < 0 || !count || *ppos >= size) {
1450 file_accessed(filp);
1457 for (seg = 0; seg < nr_segs; seg++) {
1458 read_descriptor_t desc;
1462 * If we did a short DIO read we need to skip the section of the
1463 * iov that we've already read data into.
1466 if (count > iov[seg].iov_len) {
1467 count -= iov[seg].iov_len;
1475 desc.arg.buf = iov[seg].iov_base + offset;
1476 desc.count = iov[seg].iov_len - offset;
1477 if (desc.count == 0)
1480 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1481 retval += desc.written;
1483 retval = retval ?: desc.error;
1492 EXPORT_SYMBOL(generic_file_aio_read);
1496 * page_cache_read - adds requested page to the page cache if not already there
1497 * @file: file to read
1498 * @offset: page index
1500 * This adds the requested page to the page cache if it isn't already there,
1501 * and schedules an I/O to read in its contents from disk.
1503 static int page_cache_read(struct file *file, pgoff_t offset)
1505 struct address_space *mapping = file->f_mapping;
1510 page = page_cache_alloc_cold(mapping);
1514 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1516 ret = mapping->a_ops->readpage(file, page);
1517 else if (ret == -EEXIST)
1518 ret = 0; /* losing race to add is OK */
1520 page_cache_release(page);
1522 } while (ret == AOP_TRUNCATED_PAGE);
1527 #define MMAP_LOTSAMISS (100)
1530 * Synchronous readahead happens when we don't even find
1531 * a page in the page cache at all.
1533 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1534 struct file_ra_state *ra,
1538 unsigned long ra_pages;
1539 struct address_space *mapping = file->f_mapping;
1541 /* If we don't want any read-ahead, don't bother */
1542 if (vma->vm_flags & VM_RAND_READ)
1547 if (vma->vm_flags & VM_SEQ_READ) {
1548 page_cache_sync_readahead(mapping, ra, file, offset,
1553 /* Avoid banging the cache line if not needed */
1554 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1558 * Do we miss much more than hit in this file? If so,
1559 * stop bothering with read-ahead. It will only hurt.
1561 if (ra->mmap_miss > MMAP_LOTSAMISS)
1567 ra_pages = max_sane_readahead(ra->ra_pages);
1568 ra->start = max_t(long, 0, offset - ra_pages / 2);
1569 ra->size = ra_pages;
1570 ra->async_size = ra_pages / 4;
1571 ra_submit(ra, mapping, file);
1575 * Asynchronous readahead happens when we find the page and PG_readahead,
1576 * so we want to possibly extend the readahead further..
1578 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1579 struct file_ra_state *ra,
1584 struct address_space *mapping = file->f_mapping;
1586 /* If we don't want any read-ahead, don't bother */
1587 if (vma->vm_flags & VM_RAND_READ)
1589 if (ra->mmap_miss > 0)
1591 if (PageReadahead(page))
1592 page_cache_async_readahead(mapping, ra, file,
1593 page, offset, ra->ra_pages);
1597 * filemap_fault - read in file data for page fault handling
1598 * @vma: vma in which the fault was taken
1599 * @vmf: struct vm_fault containing details of the fault
1601 * filemap_fault() is invoked via the vma operations vector for a
1602 * mapped memory region to read in file data during a page fault.
1604 * The goto's are kind of ugly, but this streamlines the normal case of having
1605 * it in the page cache, and handles the special cases reasonably without
1606 * having a lot of duplicated code.
1608 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1611 struct file *file = vma->vm_file;
1612 struct address_space *mapping = file->f_mapping;
1613 struct file_ra_state *ra = &file->f_ra;
1614 struct inode *inode = mapping->host;
1615 pgoff_t offset = vmf->pgoff;
1621 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1623 return VM_FAULT_SIGBUS;
1626 * Do we have something in the page cache already? Either
1627 * way, try readahead, but disable the memcg OOM killer for it
1628 * as readahead is optional and no errors are propagated up
1629 * the fault stack. The OOM killer is enabled while trying to
1630 * instantiate the faulting page individually below.
1632 page = find_get_page(mapping, offset);
1633 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1635 * We found the page, so try async readahead before
1636 * waiting for the lock.
1638 memcg_oom = mem_cgroup_toggle_oom(false);
1639 do_async_mmap_readahead(vma, ra, file, page, offset);
1640 mem_cgroup_toggle_oom(memcg_oom);
1642 /* No page in the page cache at all */
1643 memcg_oom = mem_cgroup_toggle_oom(false);
1644 do_sync_mmap_readahead(vma, ra, file, offset);
1645 mem_cgroup_toggle_oom(memcg_oom);
1646 count_vm_event(PGMAJFAULT);
1647 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1648 ret = VM_FAULT_MAJOR;
1650 page = find_get_page(mapping, offset);
1652 goto no_cached_page;
1655 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1656 page_cache_release(page);
1657 return ret | VM_FAULT_RETRY;
1660 /* Did it get truncated? */
1661 if (unlikely(page->mapping != mapping)) {
1666 VM_BUG_ON(page->index != offset);
1669 * We have a locked page in the page cache, now we need to check
1670 * that it's up-to-date. If not, it is going to be due to an error.
1672 if (unlikely(!PageUptodate(page)))
1673 goto page_not_uptodate;
1676 * Found the page and have a reference on it.
1677 * We must recheck i_size under page lock.
1679 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1680 if (unlikely(offset >= size)) {
1682 page_cache_release(page);
1683 return VM_FAULT_SIGBUS;
1687 return ret | VM_FAULT_LOCKED;
1691 * We're only likely to ever get here if MADV_RANDOM is in
1694 error = page_cache_read(file, offset);
1697 * The page we want has now been added to the page cache.
1698 * In the unlikely event that someone removed it in the
1699 * meantime, we'll just come back here and read it again.
1705 * An error return from page_cache_read can result if the
1706 * system is low on memory, or a problem occurs while trying
1709 if (error == -ENOMEM)
1710 return VM_FAULT_OOM;
1711 return VM_FAULT_SIGBUS;
1715 * Umm, take care of errors if the page isn't up-to-date.
1716 * Try to re-read it _once_. We do this synchronously,
1717 * because there really aren't any performance issues here
1718 * and we need to check for errors.
1720 ClearPageError(page);
1721 error = mapping->a_ops->readpage(file, page);
1723 wait_on_page_locked(page);
1724 if (!PageUptodate(page))
1727 page_cache_release(page);
1729 if (!error || error == AOP_TRUNCATED_PAGE)
1732 /* Things didn't work out. Return zero to tell the mm layer so. */
1733 shrink_readahead_size_eio(file, ra);
1734 return VM_FAULT_SIGBUS;
1736 EXPORT_SYMBOL(filemap_fault);
1738 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1740 struct page *page = vmf->page;
1741 struct inode *inode = file_inode(vma->vm_file);
1742 int ret = VM_FAULT_LOCKED;
1744 sb_start_pagefault(inode->i_sb);
1745 file_update_time(vma->vm_file);
1747 if (page->mapping != inode->i_mapping) {
1749 ret = VM_FAULT_NOPAGE;
1753 * We mark the page dirty already here so that when freeze is in
1754 * progress, we are guaranteed that writeback during freezing will
1755 * see the dirty page and writeprotect it again.
1757 set_page_dirty(page);
1758 wait_for_stable_page(page);
1760 sb_end_pagefault(inode->i_sb);
1763 EXPORT_SYMBOL(filemap_page_mkwrite);
1765 const struct vm_operations_struct generic_file_vm_ops = {
1766 .fault = filemap_fault,
1767 .page_mkwrite = filemap_page_mkwrite,
1768 .remap_pages = generic_file_remap_pages,
1771 /* This is used for a general mmap of a disk file */
1773 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1775 struct address_space *mapping = file->f_mapping;
1777 if (!mapping->a_ops->readpage)
1779 file_accessed(file);
1780 vma->vm_ops = &generic_file_vm_ops;
1785 * This is for filesystems which do not implement ->writepage.
1787 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1789 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1791 return generic_file_mmap(file, vma);
1794 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1798 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1802 #endif /* CONFIG_MMU */
1804 EXPORT_SYMBOL(generic_file_mmap);
1805 EXPORT_SYMBOL(generic_file_readonly_mmap);
1807 static struct page *__read_cache_page(struct address_space *mapping,
1809 int (*filler)(void *, struct page *),
1816 page = find_get_page(mapping, index);
1818 page = __page_cache_alloc(gfp | __GFP_COLD);
1820 return ERR_PTR(-ENOMEM);
1821 err = add_to_page_cache_lru(page, mapping, index, gfp);
1822 if (unlikely(err)) {
1823 page_cache_release(page);
1826 /* Presumably ENOMEM for radix tree node */
1827 return ERR_PTR(err);
1829 err = filler(data, page);
1831 page_cache_release(page);
1832 page = ERR_PTR(err);
1838 static struct page *do_read_cache_page(struct address_space *mapping,
1840 int (*filler)(void *, struct page *),
1849 page = __read_cache_page(mapping, index, filler, data, gfp);
1852 if (PageUptodate(page))
1856 if (!page->mapping) {
1858 page_cache_release(page);
1861 if (PageUptodate(page)) {
1865 err = filler(data, page);
1867 page_cache_release(page);
1868 return ERR_PTR(err);
1871 mark_page_accessed(page);
1876 * read_cache_page_async - read into page cache, fill it if needed
1877 * @mapping: the page's address_space
1878 * @index: the page index
1879 * @filler: function to perform the read
1880 * @data: first arg to filler(data, page) function, often left as NULL
1882 * Same as read_cache_page, but don't wait for page to become unlocked
1883 * after submitting it to the filler.
1885 * Read into the page cache. If a page already exists, and PageUptodate() is
1886 * not set, try to fill the page but don't wait for it to become unlocked.
1888 * If the page does not get brought uptodate, return -EIO.
1890 struct page *read_cache_page_async(struct address_space *mapping,
1892 int (*filler)(void *, struct page *),
1895 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1897 EXPORT_SYMBOL(read_cache_page_async);
1899 static struct page *wait_on_page_read(struct page *page)
1901 if (!IS_ERR(page)) {
1902 wait_on_page_locked(page);
1903 if (!PageUptodate(page)) {
1904 page_cache_release(page);
1905 page = ERR_PTR(-EIO);
1912 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1913 * @mapping: the page's address_space
1914 * @index: the page index
1915 * @gfp: the page allocator flags to use if allocating
1917 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1918 * any new page allocations done using the specified allocation flags.
1920 * If the page does not get brought uptodate, return -EIO.
1922 struct page *read_cache_page_gfp(struct address_space *mapping,
1926 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1928 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1930 EXPORT_SYMBOL(read_cache_page_gfp);
1933 * read_cache_page - read into page cache, fill it if needed
1934 * @mapping: the page's address_space
1935 * @index: the page index
1936 * @filler: function to perform the read
1937 * @data: first arg to filler(data, page) function, often left as NULL
1939 * Read into the page cache. If a page already exists, and PageUptodate() is
1940 * not set, try to fill the page then wait for it to become unlocked.
1942 * If the page does not get brought uptodate, return -EIO.
1944 struct page *read_cache_page(struct address_space *mapping,
1946 int (*filler)(void *, struct page *),
1949 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1951 EXPORT_SYMBOL(read_cache_page);
1953 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1954 const struct iovec *iov, size_t base, size_t bytes)
1956 size_t copied = 0, left = 0;
1959 char __user *buf = iov->iov_base + base;
1960 int copy = min(bytes, iov->iov_len - base);
1963 left = __copy_from_user_inatomic(vaddr, buf, copy);
1972 return copied - left;
1976 * Copy as much as we can into the page and return the number of bytes which
1977 * were successfully copied. If a fault is encountered then return the number of
1978 * bytes which were copied.
1980 size_t iov_iter_copy_from_user_atomic(struct page *page,
1981 struct iov_iter *i, unsigned long offset, size_t bytes)
1986 BUG_ON(!in_atomic());
1987 kaddr = kmap_atomic(page);
1988 if (likely(i->nr_segs == 1)) {
1990 char __user *buf = i->iov->iov_base + i->iov_offset;
1991 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1992 copied = bytes - left;
1994 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1995 i->iov, i->iov_offset, bytes);
1997 kunmap_atomic(kaddr);
2001 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
2004 * This has the same sideeffects and return value as
2005 * iov_iter_copy_from_user_atomic().
2006 * The difference is that it attempts to resolve faults.
2007 * Page must not be locked.
2009 size_t iov_iter_copy_from_user(struct page *page,
2010 struct iov_iter *i, unsigned long offset, size_t bytes)
2016 if (likely(i->nr_segs == 1)) {
2018 char __user *buf = i->iov->iov_base + i->iov_offset;
2019 left = __copy_from_user(kaddr + offset, buf, bytes);
2020 copied = bytes - left;
2022 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2023 i->iov, i->iov_offset, bytes);
2028 EXPORT_SYMBOL(iov_iter_copy_from_user);
2030 void iov_iter_advance(struct iov_iter *i, size_t bytes)
2032 BUG_ON(i->count < bytes);
2034 if (likely(i->nr_segs == 1)) {
2035 i->iov_offset += bytes;
2038 const struct iovec *iov = i->iov;
2039 size_t base = i->iov_offset;
2040 unsigned long nr_segs = i->nr_segs;
2043 * The !iov->iov_len check ensures we skip over unlikely
2044 * zero-length segments (without overruning the iovec).
2046 while (bytes || unlikely(i->count && !iov->iov_len)) {
2049 copy = min(bytes, iov->iov_len - base);
2050 BUG_ON(!i->count || i->count < copy);
2054 if (iov->iov_len == base) {
2061 i->iov_offset = base;
2062 i->nr_segs = nr_segs;
2065 EXPORT_SYMBOL(iov_iter_advance);
2068 * Fault in the first iovec of the given iov_iter, to a maximum length
2069 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2070 * accessed (ie. because it is an invalid address).
2072 * writev-intensive code may want this to prefault several iovecs -- that
2073 * would be possible (callers must not rely on the fact that _only_ the
2074 * first iovec will be faulted with the current implementation).
2076 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2078 char __user *buf = i->iov->iov_base + i->iov_offset;
2079 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2080 return fault_in_pages_readable(buf, bytes);
2082 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2085 * Return the count of just the current iov_iter segment.
2087 size_t iov_iter_single_seg_count(const struct iov_iter *i)
2089 const struct iovec *iov = i->iov;
2090 if (i->nr_segs == 1)
2093 return min(i->count, iov->iov_len - i->iov_offset);
2095 EXPORT_SYMBOL(iov_iter_single_seg_count);
2098 * Performs necessary checks before doing a write
2100 * Can adjust writing position or amount of bytes to write.
2101 * Returns appropriate error code that caller should return or
2102 * zero in case that write should be allowed.
2104 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2106 struct inode *inode = file->f_mapping->host;
2107 unsigned long limit = rlimit(RLIMIT_FSIZE);
2109 if (unlikely(*pos < 0))
2113 /* FIXME: this is for backwards compatibility with 2.4 */
2114 if (file->f_flags & O_APPEND)
2115 *pos = i_size_read(inode);
2117 if (limit != RLIM_INFINITY) {
2118 if (*pos >= limit) {
2119 send_sig(SIGXFSZ, current, 0);
2122 if (*count > limit - (typeof(limit))*pos) {
2123 *count = limit - (typeof(limit))*pos;
2131 if (unlikely(*pos + *count > MAX_NON_LFS &&
2132 !(file->f_flags & O_LARGEFILE))) {
2133 if (*pos >= MAX_NON_LFS) {
2136 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2137 *count = MAX_NON_LFS - (unsigned long)*pos;
2142 * Are we about to exceed the fs block limit ?
2144 * If we have written data it becomes a short write. If we have
2145 * exceeded without writing data we send a signal and return EFBIG.
2146 * Linus frestrict idea will clean these up nicely..
2148 if (likely(!isblk)) {
2149 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2150 if (*count || *pos > inode->i_sb->s_maxbytes) {
2153 /* zero-length writes at ->s_maxbytes are OK */
2156 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2157 *count = inode->i_sb->s_maxbytes - *pos;
2161 if (bdev_read_only(I_BDEV(inode)))
2163 isize = i_size_read(inode);
2164 if (*pos >= isize) {
2165 if (*count || *pos > isize)
2169 if (*pos + *count > isize)
2170 *count = isize - *pos;
2177 EXPORT_SYMBOL(generic_write_checks);
2179 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2180 loff_t pos, unsigned len, unsigned flags,
2181 struct page **pagep, void **fsdata)
2183 const struct address_space_operations *aops = mapping->a_ops;
2185 return aops->write_begin(file, mapping, pos, len, flags,
2188 EXPORT_SYMBOL(pagecache_write_begin);
2190 int pagecache_write_end(struct file *file, struct address_space *mapping,
2191 loff_t pos, unsigned len, unsigned copied,
2192 struct page *page, void *fsdata)
2194 const struct address_space_operations *aops = mapping->a_ops;
2196 mark_page_accessed(page);
2197 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2199 EXPORT_SYMBOL(pagecache_write_end);
2202 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2203 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2204 size_t count, size_t ocount)
2206 struct file *file = iocb->ki_filp;
2207 struct address_space *mapping = file->f_mapping;
2208 struct inode *inode = mapping->host;
2213 if (count != ocount)
2214 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2216 write_len = iov_length(iov, *nr_segs);
2217 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2219 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2224 * After a write we want buffered reads to be sure to go to disk to get
2225 * the new data. We invalidate clean cached page from the region we're
2226 * about to write. We do this *before* the write so that we can return
2227 * without clobbering -EIOCBQUEUED from ->direct_IO().
2229 if (mapping->nrpages) {
2230 written = invalidate_inode_pages2_range(mapping,
2231 pos >> PAGE_CACHE_SHIFT, end);
2233 * If a page can not be invalidated, return 0 to fall back
2234 * to buffered write.
2237 if (written == -EBUSY)
2243 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2246 * Finally, try again to invalidate clean pages which might have been
2247 * cached by non-direct readahead, or faulted in by get_user_pages()
2248 * if the source of the write was an mmap'ed region of the file
2249 * we're writing. Either one is a pretty crazy thing to do,
2250 * so we don't support it 100%. If this invalidation
2251 * fails, tough, the write still worked...
2253 if (mapping->nrpages) {
2254 invalidate_inode_pages2_range(mapping,
2255 pos >> PAGE_CACHE_SHIFT, end);
2260 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2261 i_size_write(inode, pos);
2262 mark_inode_dirty(inode);
2269 EXPORT_SYMBOL(generic_file_direct_write);
2272 * Find or create a page at the given pagecache position. Return the locked
2273 * page. This function is specifically for buffered writes.
2275 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2276 pgoff_t index, unsigned flags)
2281 gfp_t gfp_notmask = 0;
2283 gfp_mask = mapping_gfp_mask(mapping);
2284 if (mapping_cap_account_dirty(mapping))
2285 gfp_mask |= __GFP_WRITE;
2286 if (flags & AOP_FLAG_NOFS)
2287 gfp_notmask = __GFP_FS;
2289 page = find_lock_page(mapping, index);
2293 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2296 status = add_to_page_cache_lru(page, mapping, index,
2297 GFP_KERNEL & ~gfp_notmask);
2298 if (unlikely(status)) {
2299 page_cache_release(page);
2300 if (status == -EEXIST)
2305 wait_for_stable_page(page);
2308 EXPORT_SYMBOL(grab_cache_page_write_begin);
2310 static ssize_t generic_perform_write(struct file *file,
2311 struct iov_iter *i, loff_t pos)
2313 struct address_space *mapping = file->f_mapping;
2314 const struct address_space_operations *a_ops = mapping->a_ops;
2316 ssize_t written = 0;
2317 unsigned int flags = 0;
2320 * Copies from kernel address space cannot fail (NFSD is a big user).
2322 if (segment_eq(get_fs(), KERNEL_DS))
2323 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2327 unsigned long offset; /* Offset into pagecache page */
2328 unsigned long bytes; /* Bytes to write to page */
2329 size_t copied; /* Bytes copied from user */
2332 offset = (pos & (PAGE_CACHE_SIZE - 1));
2333 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2338 * Bring in the user page that we will copy from _first_.
2339 * Otherwise there's a nasty deadlock on copying from the
2340 * same page as we're writing to, without it being marked
2343 * Not only is this an optimisation, but it is also required
2344 * to check that the address is actually valid, when atomic
2345 * usercopies are used, below.
2347 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2352 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2354 if (unlikely(status))
2357 if (mapping_writably_mapped(mapping))
2358 flush_dcache_page(page);
2360 pagefault_disable();
2361 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2363 flush_dcache_page(page);
2365 mark_page_accessed(page);
2366 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2368 if (unlikely(status < 0))
2374 iov_iter_advance(i, copied);
2375 if (unlikely(copied == 0)) {
2377 * If we were unable to copy any data at all, we must
2378 * fall back to a single segment length write.
2380 * If we didn't fallback here, we could livelock
2381 * because not all segments in the iov can be copied at
2382 * once without a pagefault.
2384 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2385 iov_iter_single_seg_count(i));
2391 balance_dirty_pages_ratelimited(mapping);
2392 if (fatal_signal_pending(current)) {
2396 } while (iov_iter_count(i));
2398 return written ? written : status;
2402 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2403 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2404 size_t count, ssize_t written)
2406 struct file *file = iocb->ki_filp;
2410 iov_iter_init(&i, iov, nr_segs, count, written);
2411 status = generic_perform_write(file, &i, pos);
2413 if (likely(status >= 0)) {
2415 *ppos = pos + status;
2418 return written ? written : status;
2420 EXPORT_SYMBOL(generic_file_buffered_write);
2423 * __generic_file_aio_write - write data to a file
2424 * @iocb: IO state structure (file, offset, etc.)
2425 * @iov: vector with data to write
2426 * @nr_segs: number of segments in the vector
2427 * @ppos: position where to write
2429 * This function does all the work needed for actually writing data to a
2430 * file. It does all basic checks, removes SUID from the file, updates
2431 * modification times and calls proper subroutines depending on whether we
2432 * do direct IO or a standard buffered write.
2434 * It expects i_mutex to be grabbed unless we work on a block device or similar
2435 * object which does not need locking at all.
2437 * This function does *not* take care of syncing data in case of O_SYNC write.
2438 * A caller has to handle it. This is mainly due to the fact that we want to
2439 * avoid syncing under i_mutex.
2441 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2442 unsigned long nr_segs, loff_t *ppos)
2444 struct file *file = iocb->ki_filp;
2445 struct address_space * mapping = file->f_mapping;
2446 size_t ocount; /* original count */
2447 size_t count; /* after file limit checks */
2448 struct inode *inode = mapping->host;
2454 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2461 /* We can write back this queue in page reclaim */
2462 current->backing_dev_info = mapping->backing_dev_info;
2465 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2472 err = file_remove_suid(file);
2476 err = file_update_time(file);
2480 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2481 if (unlikely(file->f_flags & O_DIRECT)) {
2483 ssize_t written_buffered;
2485 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2486 ppos, count, ocount);
2487 if (written < 0 || written == count)
2490 * direct-io write to a hole: fall through to buffered I/O
2491 * for completing the rest of the request.
2495 written_buffered = generic_file_buffered_write(iocb, iov,
2496 nr_segs, pos, ppos, count,
2499 * If generic_file_buffered_write() retuned a synchronous error
2500 * then we want to return the number of bytes which were
2501 * direct-written, or the error code if that was zero. Note
2502 * that this differs from normal direct-io semantics, which
2503 * will return -EFOO even if some bytes were written.
2505 if (written_buffered < 0) {
2506 err = written_buffered;
2511 * We need to ensure that the page cache pages are written to
2512 * disk and invalidated to preserve the expected O_DIRECT
2515 endbyte = pos + written_buffered - written - 1;
2516 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2518 written = written_buffered;
2519 invalidate_mapping_pages(mapping,
2520 pos >> PAGE_CACHE_SHIFT,
2521 endbyte >> PAGE_CACHE_SHIFT);
2524 * We don't know how much we wrote, so just return
2525 * the number of bytes which were direct-written
2529 written = generic_file_buffered_write(iocb, iov, nr_segs,
2530 pos, ppos, count, written);
2533 current->backing_dev_info = NULL;
2534 return written ? written : err;
2536 EXPORT_SYMBOL(__generic_file_aio_write);
2539 * generic_file_aio_write - write data to a file
2540 * @iocb: IO state structure
2541 * @iov: vector with data to write
2542 * @nr_segs: number of segments in the vector
2543 * @pos: position in file where to write
2545 * This is a wrapper around __generic_file_aio_write() to be used by most
2546 * filesystems. It takes care of syncing the file in case of O_SYNC file
2547 * and acquires i_mutex as needed.
2549 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2550 unsigned long nr_segs, loff_t pos)
2552 struct file *file = iocb->ki_filp;
2553 struct inode *inode = file->f_mapping->host;
2556 BUG_ON(iocb->ki_pos != pos);
2558 mutex_lock(&inode->i_mutex);
2559 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2560 mutex_unlock(&inode->i_mutex);
2565 err = generic_write_sync(file, pos, ret);
2566 if (err < 0 && ret > 0)
2571 EXPORT_SYMBOL(generic_file_aio_write);
2574 * try_to_release_page() - release old fs-specific metadata on a page
2576 * @page: the page which the kernel is trying to free
2577 * @gfp_mask: memory allocation flags (and I/O mode)
2579 * The address_space is to try to release any data against the page
2580 * (presumably at page->private). If the release was successful, return `1'.
2581 * Otherwise return zero.
2583 * This may also be called if PG_fscache is set on a page, indicating that the
2584 * page is known to the local caching routines.
2586 * The @gfp_mask argument specifies whether I/O may be performed to release
2587 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2590 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2592 struct address_space * const mapping = page->mapping;
2594 BUG_ON(!PageLocked(page));
2595 if (PageWriteback(page))
2598 if (mapping && mapping->a_ops->releasepage)
2599 return mapping->a_ops->releasepage(page, gfp_mask);
2600 return try_to_free_buffers(page);
2603 EXPORT_SYMBOL(try_to_release_page);