4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/module.h>
13 #include <linux/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/aio.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/syscalls.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/memcontrol.h>
36 #include <linux/mm_inline.h> /* for page_is_file_cache() */
37 #include <linux/cleancache.h>
41 * FIXME: remove all knowledge of the buffer layer from the core VM
43 #include <linux/buffer_head.h> /* for try_to_free_buffers */
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * Shared mappings now work. 15.8.1995 Bruno.
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
62 * ->i_mmap_mutex (truncate_pagecache)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
64 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
68 * ->i_mmap_mutex (truncate->unmap_mapping_range)
72 * ->page_table_lock or pte_lock (various, mainly in memory.c)
73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 * ->lock_page (access_process_vm)
78 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
82 * ->i_alloc_sem (various)
85 * sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * inode_wb_list_lock (page_remove_rmap->set_page_dirty)
103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
104 * inode_wb_list_lock (zap_pte_range->set_page_dirty)
105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 * (code doesn't rely on that order, so you could switch it around)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
114 * Delete a page from the page cache and free it. Caller has to make
115 * sure the page is locked and that nobody else uses it - or that usage
116 * is safe. The caller must hold the mapping's tree_lock.
118 void __delete_from_page_cache(struct page *page)
120 struct address_space *mapping = page->mapping;
123 * if we're uptodate, flush out into the cleancache, otherwise
124 * invalidate any existing cleancache entries. We can't leave
125 * stale data around in the cleancache once our page is gone
127 if (PageUptodate(page) && PageMappedToDisk(page))
128 cleancache_put_page(page);
130 cleancache_flush_page(mapping, page);
132 radix_tree_delete(&mapping->page_tree, page->index);
133 page->mapping = NULL;
135 __dec_zone_page_state(page, NR_FILE_PAGES);
136 if (PageSwapBacked(page))
137 __dec_zone_page_state(page, NR_SHMEM);
138 BUG_ON(page_mapped(page));
141 * Some filesystems seem to re-dirty the page even after
142 * the VM has canceled the dirty bit (eg ext3 journaling).
144 * Fix it up by doing a final dirty accounting check after
145 * having removed the page entirely.
147 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
148 dec_zone_page_state(page, NR_FILE_DIRTY);
149 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
154 * delete_from_page_cache - delete page from page cache
155 * @page: the page which the kernel is trying to remove from page cache
157 * This must be called only on pages that have been verified to be in the page
158 * cache and locked. It will never put the page into the free list, the caller
159 * has a reference on the page.
161 void delete_from_page_cache(struct page *page)
163 struct address_space *mapping = page->mapping;
164 void (*freepage)(struct page *);
166 BUG_ON(!PageLocked(page));
168 freepage = mapping->a_ops->freepage;
169 spin_lock_irq(&mapping->tree_lock);
170 __delete_from_page_cache(page);
171 spin_unlock_irq(&mapping->tree_lock);
172 mem_cgroup_uncharge_cache_page(page);
176 page_cache_release(page);
178 EXPORT_SYMBOL(delete_from_page_cache);
180 static int sleep_on_page(void *word)
186 static int sleep_on_page_killable(void *word)
189 return fatal_signal_pending(current) ? -EINTR : 0;
193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
194 * @mapping: address space structure to write
195 * @start: offset in bytes where the range starts
196 * @end: offset in bytes where the range ends (inclusive)
197 * @sync_mode: enable synchronous operation
199 * Start writeback against all of a mapping's dirty pages that lie
200 * within the byte offsets <start, end> inclusive.
202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
203 * opposed to a regular memory cleansing writeback. The difference between
204 * these two operations is that if a dirty page/buffer is encountered, it must
205 * be waited upon, and not just skipped over.
207 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
208 loff_t end, int sync_mode)
211 struct writeback_control wbc = {
212 .sync_mode = sync_mode,
213 .nr_to_write = LONG_MAX,
214 .range_start = start,
218 if (!mapping_cap_writeback_dirty(mapping))
221 ret = do_writepages(mapping, &wbc);
225 static inline int __filemap_fdatawrite(struct address_space *mapping,
228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
231 int filemap_fdatawrite(struct address_space *mapping)
233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
235 EXPORT_SYMBOL(filemap_fdatawrite);
237 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
242 EXPORT_SYMBOL(filemap_fdatawrite_range);
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
251 int filemap_flush(struct address_space *mapping)
253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
255 EXPORT_SYMBOL(filemap_flush);
258 * filemap_fdatawait_range - wait for writeback to complete
259 * @mapping: address space structure to wait for
260 * @start_byte: offset in bytes where the range starts
261 * @end_byte: offset in bytes where the range ends (inclusive)
263 * Walk the list of under-writeback pages of the given address space
264 * in the given range and wait for all of them.
266 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
269 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
270 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
275 if (end_byte < start_byte)
278 pagevec_init(&pvec, 0);
279 while ((index <= end) &&
280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
281 PAGECACHE_TAG_WRITEBACK,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
285 for (i = 0; i < nr_pages; i++) {
286 struct page *page = pvec.pages[i];
288 /* until radix tree lookup accepts end_index */
289 if (page->index > end)
292 wait_on_page_writeback(page);
293 if (TestClearPageError(page))
296 pagevec_release(&pvec);
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
303 if (test_and_clear_bit(AS_EIO, &mapping->flags))
308 EXPORT_SYMBOL(filemap_fdatawait_range);
311 * filemap_fdatawait - wait for all under-writeback pages to complete
312 * @mapping: address space structure to wait for
314 * Walk the list of under-writeback pages of the given address space
315 * and wait for all of them.
317 int filemap_fdatawait(struct address_space *mapping)
319 loff_t i_size = i_size_read(mapping->host);
324 return filemap_fdatawait_range(mapping, 0, i_size - 1);
326 EXPORT_SYMBOL(filemap_fdatawait);
328 int filemap_write_and_wait(struct address_space *mapping)
332 if (mapping->nrpages) {
333 err = filemap_fdatawrite(mapping);
335 * Even if the above returned error, the pages may be
336 * written partially (e.g. -ENOSPC), so we wait for it.
337 * But the -EIO is special case, it may indicate the worst
338 * thing (e.g. bug) happened, so we avoid waiting for it.
341 int err2 = filemap_fdatawait(mapping);
348 EXPORT_SYMBOL(filemap_write_and_wait);
351 * filemap_write_and_wait_range - write out & wait on a file range
352 * @mapping: the address_space for the pages
353 * @lstart: offset in bytes where the range starts
354 * @lend: offset in bytes where the range ends (inclusive)
356 * Write out and wait upon file offsets lstart->lend, inclusive.
358 * Note that `lend' is inclusive (describes the last byte to be written) so
359 * that this function can be used to write to the very end-of-file (end = -1).
361 int filemap_write_and_wait_range(struct address_space *mapping,
362 loff_t lstart, loff_t lend)
366 if (mapping->nrpages) {
367 err = __filemap_fdatawrite_range(mapping, lstart, lend,
369 /* See comment of filemap_write_and_wait() */
371 int err2 = filemap_fdatawait_range(mapping,
379 EXPORT_SYMBOL(filemap_write_and_wait_range);
382 * replace_page_cache_page - replace a pagecache page with a new one
383 * @old: page to be replaced
384 * @new: page to replace with
385 * @gfp_mask: allocation mode
387 * This function replaces a page in the pagecache with a new one. On
388 * success it acquires the pagecache reference for the new page and
389 * drops it for the old page. Both the old and new pages must be
390 * locked. This function does not add the new page to the LRU, the
391 * caller must do that.
393 * The remove + add is atomic. The only way this function can fail is
394 * memory allocation failure.
396 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
400 VM_BUG_ON(!PageLocked(old));
401 VM_BUG_ON(!PageLocked(new));
402 VM_BUG_ON(new->mapping);
404 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
406 struct address_space *mapping = old->mapping;
407 void (*freepage)(struct page *);
409 pgoff_t offset = old->index;
410 freepage = mapping->a_ops->freepage;
413 new->mapping = mapping;
416 spin_lock_irq(&mapping->tree_lock);
417 __delete_from_page_cache(old);
418 error = radix_tree_insert(&mapping->page_tree, offset, new);
421 __inc_zone_page_state(new, NR_FILE_PAGES);
422 if (PageSwapBacked(new))
423 __inc_zone_page_state(new, NR_SHMEM);
424 spin_unlock_irq(&mapping->tree_lock);
425 /* mem_cgroup codes must not be called under tree_lock */
426 mem_cgroup_replace_page_cache(old, new);
427 radix_tree_preload_end();
430 page_cache_release(old);
435 EXPORT_SYMBOL_GPL(replace_page_cache_page);
438 * add_to_page_cache_locked - add a locked page to the pagecache
440 * @mapping: the page's address_space
441 * @offset: page index
442 * @gfp_mask: page allocation mode
444 * This function is used to add a page to the pagecache. It must be locked.
445 * This function does not add the page to the LRU. The caller must do that.
447 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
448 pgoff_t offset, gfp_t gfp_mask)
452 VM_BUG_ON(!PageLocked(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 if (PageSwapBacked(page))
471 __inc_zone_page_state(page, NR_SHMEM);
472 spin_unlock_irq(&mapping->tree_lock);
474 page->mapping = NULL;
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)
493 * Splice_read and readahead add shmem/tmpfs pages into the page cache
494 * before shmem_readpage has a chance to mark them as SwapBacked: they
495 * need to go on the anon lru below, and mem_cgroup_cache_charge
496 * (called in add_to_page_cache) needs to know where they're going too.
498 if (mapping_cap_swap_backed(mapping))
499 SetPageSwapBacked(page);
501 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
503 if (page_is_file_cache(page))
504 lru_cache_add_file(page);
506 lru_cache_add_anon(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_deref_retry(page))
710 if (!page_cache_get_speculative(page))
714 * Has the page moved?
715 * This is part of the lockless pagecache protocol. See
716 * include/linux/pagemap.h for details.
718 if (unlikely(page != *pagep)) {
719 page_cache_release(page);
728 EXPORT_SYMBOL(find_get_page);
731 * find_lock_page - locate, pin and lock a pagecache page
732 * @mapping: the address_space to search
733 * @offset: the page index
735 * Locates the desired pagecache page, locks it, increments its reference
736 * count and returns its address.
738 * Returns zero if the page was not present. find_lock_page() may sleep.
740 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
745 page = find_get_page(mapping, offset);
748 /* Has the page been truncated? */
749 if (unlikely(page->mapping != mapping)) {
751 page_cache_release(page);
754 VM_BUG_ON(page->index != offset);
758 EXPORT_SYMBOL(find_lock_page);
761 * find_or_create_page - locate or add a pagecache page
762 * @mapping: the page's address_space
763 * @index: the page's index into the mapping
764 * @gfp_mask: page allocation mode
766 * Locates a page in the pagecache. If the page is not present, a new page
767 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
768 * LRU list. The returned page is locked and has its reference count
771 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
774 * find_or_create_page() returns the desired page's address, or zero on
777 struct page *find_or_create_page(struct address_space *mapping,
778 pgoff_t index, gfp_t gfp_mask)
783 page = find_lock_page(mapping, index);
785 page = __page_cache_alloc(gfp_mask);
789 * We want a regular kernel memory (not highmem or DMA etc)
790 * allocation for the radix tree nodes, but we need to honour
791 * the context-specific requirements the caller has asked for.
792 * GFP_RECLAIM_MASK collects those requirements.
794 err = add_to_page_cache_lru(page, mapping, index,
795 (gfp_mask & GFP_RECLAIM_MASK));
797 page_cache_release(page);
805 EXPORT_SYMBOL(find_or_create_page);
808 * find_get_pages - gang pagecache lookup
809 * @mapping: The address_space to search
810 * @start: The starting page index
811 * @nr_pages: The maximum number of pages
812 * @pages: Where the resulting pages are placed
814 * find_get_pages() will search for and return a group of up to
815 * @nr_pages pages in the mapping. The pages are placed at @pages.
816 * find_get_pages() takes a reference against the returned pages.
818 * The search returns a group of mapping-contiguous pages with ascending
819 * indexes. There may be holes in the indices due to not-present pages.
821 * find_get_pages() returns the number of pages which were found.
823 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
824 unsigned int nr_pages, struct page **pages)
828 unsigned int nr_found;
832 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
833 (void ***)pages, start, nr_pages);
835 for (i = 0; i < nr_found; i++) {
838 page = radix_tree_deref_slot((void **)pages[i]);
843 * This can only trigger when the entry at index 0 moves out
844 * of or back to the root: none yet gotten, safe to restart.
846 if (radix_tree_deref_retry(page)) {
851 if (!page_cache_get_speculative(page))
854 /* Has the page moved? */
855 if (unlikely(page != *((void **)pages[i]))) {
856 page_cache_release(page);
865 * If all entries were removed before we could secure them,
866 * try again, because callers stop trying once 0 is returned.
868 if (unlikely(!ret && nr_found))
875 * find_get_pages_contig - gang contiguous pagecache lookup
876 * @mapping: The address_space to search
877 * @index: The starting page index
878 * @nr_pages: The maximum number of pages
879 * @pages: Where the resulting pages are placed
881 * find_get_pages_contig() works exactly like find_get_pages(), except
882 * that the returned number of pages are guaranteed to be contiguous.
884 * find_get_pages_contig() returns the number of pages which were found.
886 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
887 unsigned int nr_pages, struct page **pages)
891 unsigned int nr_found;
895 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
896 (void ***)pages, index, nr_pages);
898 for (i = 0; i < nr_found; i++) {
901 page = radix_tree_deref_slot((void **)pages[i]);
906 * This can only trigger when the entry at index 0 moves out
907 * of or back to the root: none yet gotten, safe to restart.
909 if (radix_tree_deref_retry(page))
912 if (!page_cache_get_speculative(page))
915 /* Has the page moved? */
916 if (unlikely(page != *((void **)pages[i]))) {
917 page_cache_release(page);
922 * must check mapping and index after taking the ref.
923 * otherwise we can get both false positives and false
924 * negatives, which is just confusing to the caller.
926 if (page->mapping == NULL || page->index != index) {
927 page_cache_release(page);
938 EXPORT_SYMBOL(find_get_pages_contig);
941 * find_get_pages_tag - find and return pages that match @tag
942 * @mapping: the address_space to search
943 * @index: the starting page index
944 * @tag: the tag index
945 * @nr_pages: the maximum number of pages
946 * @pages: where the resulting pages are placed
948 * Like find_get_pages, except we only return pages which are tagged with
949 * @tag. We update @index to index the next page for the traversal.
951 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
952 int tag, unsigned int nr_pages, struct page **pages)
956 unsigned int nr_found;
960 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
961 (void ***)pages, *index, nr_pages, tag);
963 for (i = 0; i < nr_found; i++) {
966 page = radix_tree_deref_slot((void **)pages[i]);
971 * This can only trigger when the entry at index 0 moves out
972 * of or back to the root: none yet gotten, safe to restart.
974 if (radix_tree_deref_retry(page))
977 if (!page_cache_get_speculative(page))
980 /* Has the page moved? */
981 if (unlikely(page != *((void **)pages[i]))) {
982 page_cache_release(page);
991 * If all entries were removed before we could secure them,
992 * try again, because callers stop trying once 0 is returned.
994 if (unlikely(!ret && nr_found))
999 *index = pages[ret - 1]->index + 1;
1003 EXPORT_SYMBOL(find_get_pages_tag);
1006 * grab_cache_page_nowait - returns locked page at given index in given cache
1007 * @mapping: target address_space
1008 * @index: the page index
1010 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1011 * This is intended for speculative data generators, where the data can
1012 * be regenerated if the page couldn't be grabbed. This routine should
1013 * be safe to call while holding the lock for another page.
1015 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1016 * and deadlock against the caller's locked page.
1019 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1021 struct page *page = find_get_page(mapping, index);
1024 if (trylock_page(page))
1026 page_cache_release(page);
1029 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1030 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1031 page_cache_release(page);
1036 EXPORT_SYMBOL(grab_cache_page_nowait);
1039 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1040 * a _large_ part of the i/o request. Imagine the worst scenario:
1042 * ---R__________________________________________B__________
1043 * ^ reading here ^ bad block(assume 4k)
1045 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1046 * => failing the whole request => read(R) => read(R+1) =>
1047 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1048 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1049 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1051 * It is going insane. Fix it by quickly scaling down the readahead size.
1053 static void shrink_readahead_size_eio(struct file *filp,
1054 struct file_ra_state *ra)
1060 * do_generic_file_read - generic file read routine
1061 * @filp: the file to read
1062 * @ppos: current file position
1063 * @desc: read_descriptor
1064 * @actor: read method
1066 * This is a generic file read routine, and uses the
1067 * mapping->a_ops->readpage() function for the actual low-level stuff.
1069 * This is really ugly. But the goto's actually try to clarify some
1070 * of the logic when it comes to error handling etc.
1072 static void do_generic_file_read(struct file *filp, loff_t *ppos,
1073 read_descriptor_t *desc, read_actor_t actor)
1075 struct address_space *mapping = filp->f_mapping;
1076 struct inode *inode = mapping->host;
1077 struct file_ra_state *ra = &filp->f_ra;
1081 unsigned long offset; /* offset into pagecache page */
1082 unsigned int prev_offset;
1085 index = *ppos >> PAGE_CACHE_SHIFT;
1086 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1087 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1088 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1089 offset = *ppos & ~PAGE_CACHE_MASK;
1095 unsigned long nr, ret;
1099 page = find_get_page(mapping, index);
1101 page_cache_sync_readahead(mapping,
1103 index, last_index - index);
1104 page = find_get_page(mapping, index);
1105 if (unlikely(page == NULL))
1106 goto no_cached_page;
1108 if (PageReadahead(page)) {
1109 page_cache_async_readahead(mapping,
1111 index, last_index - index);
1113 if (!PageUptodate(page)) {
1114 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1115 !mapping->a_ops->is_partially_uptodate)
1116 goto page_not_up_to_date;
1117 if (!trylock_page(page))
1118 goto page_not_up_to_date;
1119 /* Did it get truncated before we got the lock? */
1121 goto page_not_up_to_date_locked;
1122 if (!mapping->a_ops->is_partially_uptodate(page,
1124 goto page_not_up_to_date_locked;
1129 * i_size must be checked after we know the page is Uptodate.
1131 * Checking i_size after the check allows us to calculate
1132 * the correct value for "nr", which means the zero-filled
1133 * part of the page is not copied back to userspace (unless
1134 * another truncate extends the file - this is desired though).
1137 isize = i_size_read(inode);
1138 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1139 if (unlikely(!isize || index > end_index)) {
1140 page_cache_release(page);
1144 /* nr is the maximum number of bytes to copy from this page */
1145 nr = PAGE_CACHE_SIZE;
1146 if (index == end_index) {
1147 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1149 page_cache_release(page);
1155 /* If users can be writing to this page using arbitrary
1156 * virtual addresses, take care about potential aliasing
1157 * before reading the page on the kernel side.
1159 if (mapping_writably_mapped(mapping))
1160 flush_dcache_page(page);
1163 * When a sequential read accesses a page several times,
1164 * only mark it as accessed the first time.
1166 if (prev_index != index || offset != prev_offset)
1167 mark_page_accessed(page);
1171 * Ok, we have the page, and it's up-to-date, so
1172 * now we can copy it to user space...
1174 * The actor routine returns how many bytes were actually used..
1175 * NOTE! This may not be the same as how much of a user buffer
1176 * we filled up (we may be padding etc), so we can only update
1177 * "pos" here (the actor routine has to update the user buffer
1178 * pointers and the remaining count).
1180 ret = actor(desc, page, offset, nr);
1182 index += offset >> PAGE_CACHE_SHIFT;
1183 offset &= ~PAGE_CACHE_MASK;
1184 prev_offset = offset;
1186 page_cache_release(page);
1187 if (ret == nr && desc->count)
1191 page_not_up_to_date:
1192 /* Get exclusive access to the page ... */
1193 error = lock_page_killable(page);
1194 if (unlikely(error))
1195 goto readpage_error;
1197 page_not_up_to_date_locked:
1198 /* Did it get truncated before we got the lock? */
1199 if (!page->mapping) {
1201 page_cache_release(page);
1205 /* Did somebody else fill it already? */
1206 if (PageUptodate(page)) {
1213 * A previous I/O error may have been due to temporary
1214 * failures, eg. multipath errors.
1215 * PG_error will be set again if readpage fails.
1217 ClearPageError(page);
1218 /* Start the actual read. The read will unlock the page. */
1219 error = mapping->a_ops->readpage(filp, page);
1221 if (unlikely(error)) {
1222 if (error == AOP_TRUNCATED_PAGE) {
1223 page_cache_release(page);
1226 goto readpage_error;
1229 if (!PageUptodate(page)) {
1230 error = lock_page_killable(page);
1231 if (unlikely(error))
1232 goto readpage_error;
1233 if (!PageUptodate(page)) {
1234 if (page->mapping == NULL) {
1236 * invalidate_mapping_pages got it
1239 page_cache_release(page);
1243 shrink_readahead_size_eio(filp, ra);
1245 goto readpage_error;
1253 /* UHHUH! A synchronous read error occurred. Report it */
1254 desc->error = error;
1255 page_cache_release(page);
1260 * Ok, it wasn't cached, so we need to create a new
1263 page = page_cache_alloc_cold(mapping);
1265 desc->error = -ENOMEM;
1268 error = add_to_page_cache_lru(page, mapping,
1271 page_cache_release(page);
1272 if (error == -EEXIST)
1274 desc->error = error;
1281 ra->prev_pos = prev_index;
1282 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1283 ra->prev_pos |= prev_offset;
1285 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1286 file_accessed(filp);
1289 int file_read_actor(read_descriptor_t *desc, struct page *page,
1290 unsigned long offset, unsigned long size)
1293 unsigned long left, count = desc->count;
1299 * Faults on the destination of a read are common, so do it before
1302 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1303 kaddr = kmap_atomic(page, KM_USER0);
1304 left = __copy_to_user_inatomic(desc->arg.buf,
1305 kaddr + offset, size);
1306 kunmap_atomic(kaddr, KM_USER0);
1311 /* Do it the slow way */
1313 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1318 desc->error = -EFAULT;
1321 desc->count = count - size;
1322 desc->written += size;
1323 desc->arg.buf += size;
1328 * Performs necessary checks before doing a write
1329 * @iov: io vector request
1330 * @nr_segs: number of segments in the iovec
1331 * @count: number of bytes to write
1332 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1334 * Adjust number of segments and amount of bytes to write (nr_segs should be
1335 * properly initialized first). Returns appropriate error code that caller
1336 * should return or zero in case that write should be allowed.
1338 int generic_segment_checks(const struct iovec *iov,
1339 unsigned long *nr_segs, size_t *count, int access_flags)
1343 for (seg = 0; seg < *nr_segs; seg++) {
1344 const struct iovec *iv = &iov[seg];
1347 * If any segment has a negative length, or the cumulative
1348 * length ever wraps negative then return -EINVAL.
1351 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1353 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1358 cnt -= iv->iov_len; /* This segment is no good */
1364 EXPORT_SYMBOL(generic_segment_checks);
1367 * generic_file_aio_read - generic filesystem read routine
1368 * @iocb: kernel I/O control block
1369 * @iov: io vector request
1370 * @nr_segs: number of segments in the iovec
1371 * @pos: current file position
1373 * This is the "read()" routine for all filesystems
1374 * that can use the page cache directly.
1377 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1378 unsigned long nr_segs, loff_t pos)
1380 struct file *filp = iocb->ki_filp;
1382 unsigned long seg = 0;
1384 loff_t *ppos = &iocb->ki_pos;
1387 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1391 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1392 if (filp->f_flags & O_DIRECT) {
1394 struct address_space *mapping;
1395 struct inode *inode;
1397 mapping = filp->f_mapping;
1398 inode = mapping->host;
1400 goto out; /* skip atime */
1401 size = i_size_read(inode);
1403 retval = filemap_write_and_wait_range(mapping, pos,
1404 pos + iov_length(iov, nr_segs) - 1);
1406 struct blk_plug plug;
1408 blk_start_plug(&plug);
1409 retval = mapping->a_ops->direct_IO(READ, iocb,
1411 blk_finish_plug(&plug);
1414 *ppos = pos + retval;
1419 * Btrfs can have a short DIO read if we encounter
1420 * compressed extents, so if there was an error, or if
1421 * we've already read everything we wanted to, or if
1422 * there was a short read because we hit EOF, go ahead
1423 * and return. Otherwise fallthrough to buffered io for
1424 * the rest of the read.
1426 if (retval < 0 || !count || *ppos >= size) {
1427 file_accessed(filp);
1434 for (seg = 0; seg < nr_segs; seg++) {
1435 read_descriptor_t desc;
1439 * If we did a short DIO read we need to skip the section of the
1440 * iov that we've already read data into.
1443 if (count > iov[seg].iov_len) {
1444 count -= iov[seg].iov_len;
1452 desc.arg.buf = iov[seg].iov_base + offset;
1453 desc.count = iov[seg].iov_len - offset;
1454 if (desc.count == 0)
1457 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1458 retval += desc.written;
1460 retval = retval ?: desc.error;
1469 EXPORT_SYMBOL(generic_file_aio_read);
1472 do_readahead(struct address_space *mapping, struct file *filp,
1473 pgoff_t index, unsigned long nr)
1475 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1478 force_page_cache_readahead(mapping, filp, index, nr);
1482 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1490 if (file->f_mode & FMODE_READ) {
1491 struct address_space *mapping = file->f_mapping;
1492 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1493 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1494 unsigned long len = end - start + 1;
1495 ret = do_readahead(mapping, file, start, len);
1501 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1502 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1504 return SYSC_readahead((int) fd, offset, (size_t) count);
1506 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1511 * page_cache_read - adds requested page to the page cache if not already there
1512 * @file: file to read
1513 * @offset: page index
1515 * This adds the requested page to the page cache if it isn't already there,
1516 * and schedules an I/O to read in its contents from disk.
1518 static int page_cache_read(struct file *file, pgoff_t offset)
1520 struct address_space *mapping = file->f_mapping;
1525 page = page_cache_alloc_cold(mapping);
1529 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1531 ret = mapping->a_ops->readpage(file, page);
1532 else if (ret == -EEXIST)
1533 ret = 0; /* losing race to add is OK */
1535 page_cache_release(page);
1537 } while (ret == AOP_TRUNCATED_PAGE);
1542 #define MMAP_LOTSAMISS (100)
1545 * Synchronous readahead happens when we don't even find
1546 * a page in the page cache at all.
1548 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1549 struct file_ra_state *ra,
1553 unsigned long ra_pages;
1554 struct address_space *mapping = file->f_mapping;
1556 /* If we don't want any read-ahead, don't bother */
1557 if (VM_RandomReadHint(vma))
1562 if (VM_SequentialReadHint(vma)) {
1563 page_cache_sync_readahead(mapping, ra, file, offset,
1568 /* Avoid banging the cache line if not needed */
1569 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1573 * Do we miss much more than hit in this file? If so,
1574 * stop bothering with read-ahead. It will only hurt.
1576 if (ra->mmap_miss > MMAP_LOTSAMISS)
1582 ra_pages = max_sane_readahead(ra->ra_pages);
1583 ra->start = max_t(long, 0, offset - ra_pages / 2);
1584 ra->size = ra_pages;
1585 ra->async_size = ra_pages / 4;
1586 ra_submit(ra, mapping, file);
1590 * Asynchronous readahead happens when we find the page and PG_readahead,
1591 * so we want to possibly extend the readahead further..
1593 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1594 struct file_ra_state *ra,
1599 struct address_space *mapping = file->f_mapping;
1601 /* If we don't want any read-ahead, don't bother */
1602 if (VM_RandomReadHint(vma))
1604 if (ra->mmap_miss > 0)
1606 if (PageReadahead(page))
1607 page_cache_async_readahead(mapping, ra, file,
1608 page, offset, ra->ra_pages);
1612 * filemap_fault - read in file data for page fault handling
1613 * @vma: vma in which the fault was taken
1614 * @vmf: struct vm_fault containing details of the fault
1616 * filemap_fault() is invoked via the vma operations vector for a
1617 * mapped memory region to read in file data during a page fault.
1619 * The goto's are kind of ugly, but this streamlines the normal case of having
1620 * it in the page cache, and handles the special cases reasonably without
1621 * having a lot of duplicated code.
1623 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1626 struct file *file = vma->vm_file;
1627 struct address_space *mapping = file->f_mapping;
1628 struct file_ra_state *ra = &file->f_ra;
1629 struct inode *inode = mapping->host;
1630 pgoff_t offset = vmf->pgoff;
1635 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1637 return VM_FAULT_SIGBUS;
1640 * Do we have something in the page cache already?
1642 page = find_get_page(mapping, offset);
1645 * We found the page, so try async readahead before
1646 * waiting for the lock.
1648 do_async_mmap_readahead(vma, ra, file, page, offset);
1650 /* No page in the page cache at all */
1651 do_sync_mmap_readahead(vma, ra, file, offset);
1652 count_vm_event(PGMAJFAULT);
1653 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1654 ret = VM_FAULT_MAJOR;
1656 page = find_get_page(mapping, offset);
1658 goto no_cached_page;
1661 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1662 page_cache_release(page);
1663 return ret | VM_FAULT_RETRY;
1666 /* Did it get truncated? */
1667 if (unlikely(page->mapping != mapping)) {
1672 VM_BUG_ON(page->index != offset);
1675 * We have a locked page in the page cache, now we need to check
1676 * that it's up-to-date. If not, it is going to be due to an error.
1678 if (unlikely(!PageUptodate(page)))
1679 goto page_not_uptodate;
1682 * Found the page and have a reference on it.
1683 * We must recheck i_size under page lock.
1685 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1686 if (unlikely(offset >= size)) {
1688 page_cache_release(page);
1689 return VM_FAULT_SIGBUS;
1693 return ret | VM_FAULT_LOCKED;
1697 * We're only likely to ever get here if MADV_RANDOM is in
1700 error = page_cache_read(file, offset);
1703 * The page we want has now been added to the page cache.
1704 * In the unlikely event that someone removed it in the
1705 * meantime, we'll just come back here and read it again.
1711 * An error return from page_cache_read can result if the
1712 * system is low on memory, or a problem occurs while trying
1715 if (error == -ENOMEM)
1716 return VM_FAULT_OOM;
1717 return VM_FAULT_SIGBUS;
1721 * Umm, take care of errors if the page isn't up-to-date.
1722 * Try to re-read it _once_. We do this synchronously,
1723 * because there really aren't any performance issues here
1724 * and we need to check for errors.
1726 ClearPageError(page);
1727 error = mapping->a_ops->readpage(file, page);
1729 wait_on_page_locked(page);
1730 if (!PageUptodate(page))
1733 page_cache_release(page);
1735 if (!error || error == AOP_TRUNCATED_PAGE)
1738 /* Things didn't work out. Return zero to tell the mm layer so. */
1739 shrink_readahead_size_eio(file, ra);
1740 return VM_FAULT_SIGBUS;
1742 EXPORT_SYMBOL(filemap_fault);
1744 const struct vm_operations_struct generic_file_vm_ops = {
1745 .fault = filemap_fault,
1748 /* This is used for a general mmap of a disk file */
1750 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1752 struct address_space *mapping = file->f_mapping;
1754 if (!mapping->a_ops->readpage)
1756 file_accessed(file);
1757 vma->vm_ops = &generic_file_vm_ops;
1758 vma->vm_flags |= VM_CAN_NONLINEAR;
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: destination for read data
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: destination for read data
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);
1932 * The logic we want is
1934 * if suid or (sgid and xgrp)
1937 int should_remove_suid(struct dentry *dentry)
1939 mode_t mode = dentry->d_inode->i_mode;
1942 /* suid always must be killed */
1943 if (unlikely(mode & S_ISUID))
1944 kill = ATTR_KILL_SUID;
1947 * sgid without any exec bits is just a mandatory locking mark; leave
1948 * it alone. If some exec bits are set, it's a real sgid; kill it.
1950 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1951 kill |= ATTR_KILL_SGID;
1953 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1958 EXPORT_SYMBOL(should_remove_suid);
1960 static int __remove_suid(struct dentry *dentry, int kill)
1962 struct iattr newattrs;
1964 newattrs.ia_valid = ATTR_FORCE | kill;
1965 return notify_change(dentry, &newattrs);
1968 int file_remove_suid(struct file *file)
1970 struct dentry *dentry = file->f_path.dentry;
1971 struct inode *inode = dentry->d_inode;
1976 /* Fast path for nothing security related */
1977 if (IS_NOSEC(inode))
1980 killsuid = should_remove_suid(dentry);
1981 killpriv = security_inode_need_killpriv(dentry);
1986 error = security_inode_killpriv(dentry);
1987 if (!error && killsuid)
1988 error = __remove_suid(dentry, killsuid);
1989 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1990 inode->i_flags |= S_NOSEC;
1994 EXPORT_SYMBOL(file_remove_suid);
1996 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1997 const struct iovec *iov, size_t base, size_t bytes)
1999 size_t copied = 0, left = 0;
2002 char __user *buf = iov->iov_base + base;
2003 int copy = min(bytes, iov->iov_len - base);
2006 left = __copy_from_user_inatomic(vaddr, buf, copy);
2015 return copied - left;
2019 * Copy as much as we can into the page and return the number of bytes which
2020 * were successfully copied. If a fault is encountered then return the number of
2021 * bytes which were copied.
2023 size_t iov_iter_copy_from_user_atomic(struct page *page,
2024 struct iov_iter *i, unsigned long offset, size_t bytes)
2029 BUG_ON(!in_atomic());
2030 kaddr = kmap_atomic(page, KM_USER0);
2031 if (likely(i->nr_segs == 1)) {
2033 char __user *buf = i->iov->iov_base + i->iov_offset;
2034 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
2035 copied = bytes - left;
2037 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2038 i->iov, i->iov_offset, bytes);
2040 kunmap_atomic(kaddr, KM_USER0);
2044 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
2047 * This has the same sideeffects and return value as
2048 * iov_iter_copy_from_user_atomic().
2049 * The difference is that it attempts to resolve faults.
2050 * Page must not be locked.
2052 size_t iov_iter_copy_from_user(struct page *page,
2053 struct iov_iter *i, unsigned long offset, size_t bytes)
2059 if (likely(i->nr_segs == 1)) {
2061 char __user *buf = i->iov->iov_base + i->iov_offset;
2062 left = __copy_from_user(kaddr + offset, buf, bytes);
2063 copied = bytes - left;
2065 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2066 i->iov, i->iov_offset, bytes);
2071 EXPORT_SYMBOL(iov_iter_copy_from_user);
2073 void iov_iter_advance(struct iov_iter *i, size_t bytes)
2075 BUG_ON(i->count < bytes);
2077 if (likely(i->nr_segs == 1)) {
2078 i->iov_offset += bytes;
2081 const struct iovec *iov = i->iov;
2082 size_t base = i->iov_offset;
2085 * The !iov->iov_len check ensures we skip over unlikely
2086 * zero-length segments (without overruning the iovec).
2088 while (bytes || unlikely(i->count && !iov->iov_len)) {
2091 copy = min(bytes, iov->iov_len - base);
2092 BUG_ON(!i->count || i->count < copy);
2096 if (iov->iov_len == base) {
2102 i->iov_offset = base;
2105 EXPORT_SYMBOL(iov_iter_advance);
2108 * Fault in the first iovec of the given iov_iter, to a maximum length
2109 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2110 * accessed (ie. because it is an invalid address).
2112 * writev-intensive code may want this to prefault several iovecs -- that
2113 * would be possible (callers must not rely on the fact that _only_ the
2114 * first iovec will be faulted with the current implementation).
2116 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2118 char __user *buf = i->iov->iov_base + i->iov_offset;
2119 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2120 return fault_in_pages_readable(buf, bytes);
2122 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2125 * Return the count of just the current iov_iter segment.
2127 size_t iov_iter_single_seg_count(struct iov_iter *i)
2129 const struct iovec *iov = i->iov;
2130 if (i->nr_segs == 1)
2133 return min(i->count, iov->iov_len - i->iov_offset);
2135 EXPORT_SYMBOL(iov_iter_single_seg_count);
2138 * Performs necessary checks before doing a write
2140 * Can adjust writing position or amount of bytes to write.
2141 * Returns appropriate error code that caller should return or
2142 * zero in case that write should be allowed.
2144 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2146 struct inode *inode = file->f_mapping->host;
2147 unsigned long limit = rlimit(RLIMIT_FSIZE);
2149 if (unlikely(*pos < 0))
2153 /* FIXME: this is for backwards compatibility with 2.4 */
2154 if (file->f_flags & O_APPEND)
2155 *pos = i_size_read(inode);
2157 if (limit != RLIM_INFINITY) {
2158 if (*pos >= limit) {
2159 send_sig(SIGXFSZ, current, 0);
2162 if (*count > limit - (typeof(limit))*pos) {
2163 *count = limit - (typeof(limit))*pos;
2171 if (unlikely(*pos + *count > MAX_NON_LFS &&
2172 !(file->f_flags & O_LARGEFILE))) {
2173 if (*pos >= MAX_NON_LFS) {
2176 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2177 *count = MAX_NON_LFS - (unsigned long)*pos;
2182 * Are we about to exceed the fs block limit ?
2184 * If we have written data it becomes a short write. If we have
2185 * exceeded without writing data we send a signal and return EFBIG.
2186 * Linus frestrict idea will clean these up nicely..
2188 if (likely(!isblk)) {
2189 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2190 if (*count || *pos > inode->i_sb->s_maxbytes) {
2193 /* zero-length writes at ->s_maxbytes are OK */
2196 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2197 *count = inode->i_sb->s_maxbytes - *pos;
2201 if (bdev_read_only(I_BDEV(inode)))
2203 isize = i_size_read(inode);
2204 if (*pos >= isize) {
2205 if (*count || *pos > isize)
2209 if (*pos + *count > isize)
2210 *count = isize - *pos;
2217 EXPORT_SYMBOL(generic_write_checks);
2219 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2220 loff_t pos, unsigned len, unsigned flags,
2221 struct page **pagep, void **fsdata)
2223 const struct address_space_operations *aops = mapping->a_ops;
2225 return aops->write_begin(file, mapping, pos, len, flags,
2228 EXPORT_SYMBOL(pagecache_write_begin);
2230 int pagecache_write_end(struct file *file, struct address_space *mapping,
2231 loff_t pos, unsigned len, unsigned copied,
2232 struct page *page, void *fsdata)
2234 const struct address_space_operations *aops = mapping->a_ops;
2236 mark_page_accessed(page);
2237 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2239 EXPORT_SYMBOL(pagecache_write_end);
2242 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2243 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2244 size_t count, size_t ocount)
2246 struct file *file = iocb->ki_filp;
2247 struct address_space *mapping = file->f_mapping;
2248 struct inode *inode = mapping->host;
2253 if (count != ocount)
2254 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2256 write_len = iov_length(iov, *nr_segs);
2257 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2259 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2264 * After a write we want buffered reads to be sure to go to disk to get
2265 * the new data. We invalidate clean cached page from the region we're
2266 * about to write. We do this *before* the write so that we can return
2267 * without clobbering -EIOCBQUEUED from ->direct_IO().
2269 if (mapping->nrpages) {
2270 written = invalidate_inode_pages2_range(mapping,
2271 pos >> PAGE_CACHE_SHIFT, end);
2273 * If a page can not be invalidated, return 0 to fall back
2274 * to buffered write.
2277 if (written == -EBUSY)
2283 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2286 * Finally, try again to invalidate clean pages which might have been
2287 * cached by non-direct readahead, or faulted in by get_user_pages()
2288 * if the source of the write was an mmap'ed region of the file
2289 * we're writing. Either one is a pretty crazy thing to do,
2290 * so we don't support it 100%. If this invalidation
2291 * fails, tough, the write still worked...
2293 if (mapping->nrpages) {
2294 invalidate_inode_pages2_range(mapping,
2295 pos >> PAGE_CACHE_SHIFT, end);
2300 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2301 i_size_write(inode, pos);
2302 mark_inode_dirty(inode);
2309 EXPORT_SYMBOL(generic_file_direct_write);
2312 * Find or create a page at the given pagecache position. Return the locked
2313 * page. This function is specifically for buffered writes.
2315 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2316 pgoff_t index, unsigned flags)
2320 gfp_t gfp_notmask = 0;
2321 if (flags & AOP_FLAG_NOFS)
2322 gfp_notmask = __GFP_FS;
2324 page = find_lock_page(mapping, index);
2328 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2331 status = add_to_page_cache_lru(page, mapping, index,
2332 GFP_KERNEL & ~gfp_notmask);
2333 if (unlikely(status)) {
2334 page_cache_release(page);
2335 if (status == -EEXIST)
2340 wait_on_page_writeback(page);
2343 EXPORT_SYMBOL(grab_cache_page_write_begin);
2345 static ssize_t generic_perform_write(struct file *file,
2346 struct iov_iter *i, loff_t pos)
2348 struct address_space *mapping = file->f_mapping;
2349 const struct address_space_operations *a_ops = mapping->a_ops;
2351 ssize_t written = 0;
2352 unsigned int flags = 0;
2355 * Copies from kernel address space cannot fail (NFSD is a big user).
2357 if (segment_eq(get_fs(), KERNEL_DS))
2358 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2362 unsigned long offset; /* Offset into pagecache page */
2363 unsigned long bytes; /* Bytes to write to page */
2364 size_t copied; /* Bytes copied from user */
2367 offset = (pos & (PAGE_CACHE_SIZE - 1));
2368 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2374 * Bring in the user page that we will copy from _first_.
2375 * Otherwise there's a nasty deadlock on copying from the
2376 * same page as we're writing to, without it being marked
2379 * Not only is this an optimisation, but it is also required
2380 * to check that the address is actually valid, when atomic
2381 * usercopies are used, below.
2383 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2388 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2390 if (unlikely(status))
2393 if (mapping_writably_mapped(mapping))
2394 flush_dcache_page(page);
2396 pagefault_disable();
2397 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2399 flush_dcache_page(page);
2401 mark_page_accessed(page);
2402 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2404 if (unlikely(status < 0))
2410 iov_iter_advance(i, copied);
2411 if (unlikely(copied == 0)) {
2413 * If we were unable to copy any data at all, we must
2414 * fall back to a single segment length write.
2416 * If we didn't fallback here, we could livelock
2417 * because not all segments in the iov can be copied at
2418 * once without a pagefault.
2420 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2421 iov_iter_single_seg_count(i));
2427 balance_dirty_pages_ratelimited(mapping);
2429 } while (iov_iter_count(i));
2431 return written ? written : status;
2435 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2436 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2437 size_t count, ssize_t written)
2439 struct file *file = iocb->ki_filp;
2443 iov_iter_init(&i, iov, nr_segs, count, written);
2444 status = generic_perform_write(file, &i, pos);
2446 if (likely(status >= 0)) {
2448 *ppos = pos + status;
2451 return written ? written : status;
2453 EXPORT_SYMBOL(generic_file_buffered_write);
2456 * __generic_file_aio_write - write data to a file
2457 * @iocb: IO state structure (file, offset, etc.)
2458 * @iov: vector with data to write
2459 * @nr_segs: number of segments in the vector
2460 * @ppos: position where to write
2462 * This function does all the work needed for actually writing data to a
2463 * file. It does all basic checks, removes SUID from the file, updates
2464 * modification times and calls proper subroutines depending on whether we
2465 * do direct IO or a standard buffered write.
2467 * It expects i_mutex to be grabbed unless we work on a block device or similar
2468 * object which does not need locking at all.
2470 * This function does *not* take care of syncing data in case of O_SYNC write.
2471 * A caller has to handle it. This is mainly due to the fact that we want to
2472 * avoid syncing under i_mutex.
2474 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2475 unsigned long nr_segs, loff_t *ppos)
2477 struct file *file = iocb->ki_filp;
2478 struct address_space * mapping = file->f_mapping;
2479 size_t ocount; /* original count */
2480 size_t count; /* after file limit checks */
2481 struct inode *inode = mapping->host;
2487 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2494 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2496 /* We can write back this queue in page reclaim */
2497 current->backing_dev_info = mapping->backing_dev_info;
2500 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2507 err = file_remove_suid(file);
2511 file_update_time(file);
2513 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2514 if (unlikely(file->f_flags & O_DIRECT)) {
2516 ssize_t written_buffered;
2518 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2519 ppos, count, ocount);
2520 if (written < 0 || written == count)
2523 * direct-io write to a hole: fall through to buffered I/O
2524 * for completing the rest of the request.
2528 written_buffered = generic_file_buffered_write(iocb, iov,
2529 nr_segs, pos, ppos, count,
2532 * If generic_file_buffered_write() retuned a synchronous error
2533 * then we want to return the number of bytes which were
2534 * direct-written, or the error code if that was zero. Note
2535 * that this differs from normal direct-io semantics, which
2536 * will return -EFOO even if some bytes were written.
2538 if (written_buffered < 0) {
2539 err = written_buffered;
2544 * We need to ensure that the page cache pages are written to
2545 * disk and invalidated to preserve the expected O_DIRECT
2548 endbyte = pos + written_buffered - written - 1;
2549 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2551 written = written_buffered;
2552 invalidate_mapping_pages(mapping,
2553 pos >> PAGE_CACHE_SHIFT,
2554 endbyte >> PAGE_CACHE_SHIFT);
2557 * We don't know how much we wrote, so just return
2558 * the number of bytes which were direct-written
2562 written = generic_file_buffered_write(iocb, iov, nr_segs,
2563 pos, ppos, count, written);
2566 current->backing_dev_info = NULL;
2567 return written ? written : err;
2569 EXPORT_SYMBOL(__generic_file_aio_write);
2572 * generic_file_aio_write - write data to a file
2573 * @iocb: IO state structure
2574 * @iov: vector with data to write
2575 * @nr_segs: number of segments in the vector
2576 * @pos: position in file where to write
2578 * This is a wrapper around __generic_file_aio_write() to be used by most
2579 * filesystems. It takes care of syncing the file in case of O_SYNC file
2580 * and acquires i_mutex as needed.
2582 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2583 unsigned long nr_segs, loff_t pos)
2585 struct file *file = iocb->ki_filp;
2586 struct inode *inode = file->f_mapping->host;
2587 struct blk_plug plug;
2590 BUG_ON(iocb->ki_pos != pos);
2592 mutex_lock(&inode->i_mutex);
2593 blk_start_plug(&plug);
2594 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2595 mutex_unlock(&inode->i_mutex);
2597 if (ret > 0 || ret == -EIOCBQUEUED) {
2600 err = generic_write_sync(file, pos, ret);
2601 if (err < 0 && ret > 0)
2604 blk_finish_plug(&plug);
2607 EXPORT_SYMBOL(generic_file_aio_write);
2610 * try_to_release_page() - release old fs-specific metadata on a page
2612 * @page: the page which the kernel is trying to free
2613 * @gfp_mask: memory allocation flags (and I/O mode)
2615 * The address_space is to try to release any data against the page
2616 * (presumably at page->private). If the release was successful, return `1'.
2617 * Otherwise return zero.
2619 * This may also be called if PG_fscache is set on a page, indicating that the
2620 * page is known to the local caching routines.
2622 * The @gfp_mask argument specifies whether I/O may be performed to release
2623 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2626 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2628 struct address_space * const mapping = page->mapping;
2630 BUG_ON(!PageLocked(page));
2631 if (PageWriteback(page))
2634 if (mapping && mapping->a_ops->releasepage)
2635 return mapping->a_ops->releasepage(page, gfp_mask);
2636 return try_to_free_buffers(page);
2639 EXPORT_SYMBOL(try_to_release_page);