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/slab.h>
14 #include <linux/compiler.h>
16 #include <linux/uaccess.h>
17 #include <linux/aio.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/syscalls.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/memcontrol.h>
36 #include <linux/mm_inline.h> /* for page_is_file_cache() */
40 * FIXME: remove all knowledge of the buffer layer from the core VM
42 #include <linux/buffer_head.h> /* for try_to_free_buffers */
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
50 * Shared mappings now work. 15.8.1995 Bruno.
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
61 * ->i_mmap_lock (vmtruncate)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
67 * ->i_mmap_lock (truncate->unmap_mapping_range)
71 * ->page_table_lock or pte_lock (various, mainly in memory.c)
72 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
75 * ->lock_page (access_process_vm)
77 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
81 * ->i_alloc_sem (various)
84 * ->sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
106 * ->dcache_lock (proc_pid_lookup)
110 * Remove a page from the page cache and free it. Caller has to make
111 * sure the page is locked and that nobody else uses it - or that usage
112 * is safe. The caller must hold the mapping's tree_lock.
114 void __remove_from_page_cache(struct page *page)
116 struct address_space *mapping = page->mapping;
118 radix_tree_delete(&mapping->page_tree, page->index);
119 page->mapping = NULL;
121 __dec_zone_page_state(page, NR_FILE_PAGES);
122 BUG_ON(page_mapped(page));
125 * Some filesystems seem to re-dirty the page even after
126 * the VM has canceled the dirty bit (eg ext3 journaling).
128 * Fix it up by doing a final dirty accounting check after
129 * having removed the page entirely.
131 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
132 dec_zone_page_state(page, NR_FILE_DIRTY);
133 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
137 void remove_from_page_cache(struct page *page)
139 struct address_space *mapping = page->mapping;
141 BUG_ON(!PageLocked(page));
143 spin_lock_irq(&mapping->tree_lock);
144 __remove_from_page_cache(page);
145 spin_unlock_irq(&mapping->tree_lock);
146 mem_cgroup_uncharge_cache_page(page);
149 static int sync_page(void *word)
151 struct address_space *mapping;
154 page = container_of((unsigned long *)word, struct page, flags);
157 * page_mapping() is being called without PG_locked held.
158 * Some knowledge of the state and use of the page is used to
159 * reduce the requirements down to a memory barrier.
160 * The danger here is of a stale page_mapping() return value
161 * indicating a struct address_space different from the one it's
162 * associated with when it is associated with one.
163 * After smp_mb(), it's either the correct page_mapping() for
164 * the page, or an old page_mapping() and the page's own
165 * page_mapping() has gone NULL.
166 * The ->sync_page() address_space operation must tolerate
167 * page_mapping() going NULL. By an amazing coincidence,
168 * this comes about because none of the users of the page
169 * in the ->sync_page() methods make essential use of the
170 * page_mapping(), merely passing the page down to the backing
171 * device's unplug functions when it's non-NULL, which in turn
172 * ignore it for all cases but swap, where only page_private(page) is
173 * of interest. When page_mapping() does go NULL, the entire
174 * call stack gracefully ignores the page and returns.
178 mapping = page_mapping(page);
179 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
180 mapping->a_ops->sync_page(page);
185 static int sync_page_killable(void *word)
188 return fatal_signal_pending(current) ? -EINTR : 0;
192 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
193 * @mapping: address space structure to write
194 * @start: offset in bytes where the range starts
195 * @end: offset in bytes where the range ends (inclusive)
196 * @sync_mode: enable synchronous operation
198 * Start writeback against all of a mapping's dirty pages that lie
199 * within the byte offsets <start, end> inclusive.
201 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
202 * opposed to a regular memory cleansing writeback. The difference between
203 * these two operations is that if a dirty page/buffer is encountered, it must
204 * be waited upon, and not just skipped over.
206 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
207 loff_t end, int sync_mode)
210 struct writeback_control wbc = {
211 .sync_mode = sync_mode,
212 .nr_to_write = LONG_MAX,
213 .range_start = start,
217 if (!mapping_cap_writeback_dirty(mapping))
220 ret = do_writepages(mapping, &wbc);
224 static inline int __filemap_fdatawrite(struct address_space *mapping,
227 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
230 int filemap_fdatawrite(struct address_space *mapping)
232 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
234 EXPORT_SYMBOL(filemap_fdatawrite);
236 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
239 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
241 EXPORT_SYMBOL(filemap_fdatawrite_range);
244 * filemap_flush - mostly a non-blocking flush
245 * @mapping: target address_space
247 * This is a mostly non-blocking flush. Not suitable for data-integrity
248 * purposes - I/O may not be started against all dirty pages.
250 int filemap_flush(struct address_space *mapping)
252 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
254 EXPORT_SYMBOL(filemap_flush);
257 * wait_on_page_writeback_range - wait for writeback to complete
258 * @mapping: target address_space
259 * @start: beginning page index
260 * @end: ending page index
262 * Wait for writeback to complete against pages indexed by start->end
265 int wait_on_page_writeback_range(struct address_space *mapping,
266 pgoff_t start, pgoff_t end)
276 pagevec_init(&pvec, 0);
278 while ((index <= end) &&
279 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
280 PAGECACHE_TAG_WRITEBACK,
281 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
284 for (i = 0; i < nr_pages; i++) {
285 struct page *page = pvec.pages[i];
287 /* until radix tree lookup accepts end_index */
288 if (page->index > end)
291 wait_on_page_writeback(page);
295 pagevec_release(&pvec);
299 /* Check for outstanding write errors */
300 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
302 if (test_and_clear_bit(AS_EIO, &mapping->flags))
309 * filemap_fdatawait_range - wait for all under-writeback pages to complete in a given range
310 * @mapping: address space structure to wait for
311 * @start: offset in bytes where the range starts
312 * @end: offset in bytes where the range ends (inclusive)
314 * Walk the list of under-writeback pages of the given address space
315 * in the given range and wait for all of them.
317 * This is just a simple wrapper so that callers don't have to convert offsets
318 * to page indexes themselves
320 int filemap_fdatawait_range(struct address_space *mapping, loff_t start,
323 return wait_on_page_writeback_range(mapping, start >> PAGE_CACHE_SHIFT,
324 end >> PAGE_CACHE_SHIFT);
326 EXPORT_SYMBOL(filemap_fdatawait_range);
329 * sync_page_range - write and wait on all pages in the passed range
330 * @inode: target inode
331 * @mapping: target address_space
332 * @pos: beginning offset in pages to write
333 * @count: number of bytes to write
335 * Write and wait upon all the pages in the passed range. This is a "data
336 * integrity" operation. It waits upon in-flight writeout before starting and
337 * waiting upon new writeout. If there was an IO error, return it.
339 * We need to re-take i_mutex during the generic_osync_inode list walk because
340 * it is otherwise livelockable.
342 int sync_page_range(struct inode *inode, struct address_space *mapping,
343 loff_t pos, loff_t count)
345 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
346 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
349 if (!mapping_cap_writeback_dirty(mapping) || !count)
351 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
353 mutex_lock(&inode->i_mutex);
354 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
355 mutex_unlock(&inode->i_mutex);
358 ret = wait_on_page_writeback_range(mapping, start, end);
361 EXPORT_SYMBOL(sync_page_range);
364 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
365 * @inode: target inode
366 * @mapping: target address_space
367 * @pos: beginning offset in pages to write
368 * @count: number of bytes to write
370 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
371 * as it forces O_SYNC writers to different parts of the same file
372 * to be serialised right until io completion.
374 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
375 loff_t pos, loff_t count)
377 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
378 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
381 if (!mapping_cap_writeback_dirty(mapping) || !count)
383 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
385 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
387 ret = wait_on_page_writeback_range(mapping, start, end);
390 EXPORT_SYMBOL(sync_page_range_nolock);
393 * filemap_fdatawait - wait for all under-writeback pages to complete
394 * @mapping: address space structure to wait for
396 * Walk the list of under-writeback pages of the given address space
397 * and wait for all of them.
399 int filemap_fdatawait(struct address_space *mapping)
401 loff_t i_size = i_size_read(mapping->host);
406 return wait_on_page_writeback_range(mapping, 0,
407 (i_size - 1) >> PAGE_CACHE_SHIFT);
409 EXPORT_SYMBOL(filemap_fdatawait);
411 int filemap_write_and_wait(struct address_space *mapping)
415 if (mapping->nrpages) {
416 err = filemap_fdatawrite(mapping);
418 * Even if the above returned error, the pages may be
419 * written partially (e.g. -ENOSPC), so we wait for it.
420 * But the -EIO is special case, it may indicate the worst
421 * thing (e.g. bug) happened, so we avoid waiting for it.
424 int err2 = filemap_fdatawait(mapping);
431 EXPORT_SYMBOL(filemap_write_and_wait);
434 * filemap_write_and_wait_range - write out & wait on a file range
435 * @mapping: the address_space for the pages
436 * @lstart: offset in bytes where the range starts
437 * @lend: offset in bytes where the range ends (inclusive)
439 * Write out and wait upon file offsets lstart->lend, inclusive.
441 * Note that `lend' is inclusive (describes the last byte to be written) so
442 * that this function can be used to write to the very end-of-file (end = -1).
444 int filemap_write_and_wait_range(struct address_space *mapping,
445 loff_t lstart, loff_t lend)
449 if (mapping->nrpages) {
450 err = __filemap_fdatawrite_range(mapping, lstart, lend,
452 /* See comment of filemap_write_and_wait() */
454 int err2 = wait_on_page_writeback_range(mapping,
455 lstart >> PAGE_CACHE_SHIFT,
456 lend >> PAGE_CACHE_SHIFT);
463 EXPORT_SYMBOL(filemap_write_and_wait_range);
466 * add_to_page_cache_locked - add a locked page to the pagecache
468 * @mapping: the page's address_space
469 * @offset: page index
470 * @gfp_mask: page allocation mode
472 * This function is used to add a page to the pagecache. It must be locked.
473 * This function does not add the page to the LRU. The caller must do that.
475 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
476 pgoff_t offset, gfp_t gfp_mask)
480 VM_BUG_ON(!PageLocked(page));
482 error = mem_cgroup_cache_charge(page, current->mm,
483 gfp_mask & GFP_RECLAIM_MASK);
487 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
489 page_cache_get(page);
490 page->mapping = mapping;
491 page->index = offset;
493 spin_lock_irq(&mapping->tree_lock);
494 error = radix_tree_insert(&mapping->page_tree, offset, page);
495 if (likely(!error)) {
497 __inc_zone_page_state(page, NR_FILE_PAGES);
498 spin_unlock_irq(&mapping->tree_lock);
500 page->mapping = NULL;
501 spin_unlock_irq(&mapping->tree_lock);
502 mem_cgroup_uncharge_cache_page(page);
503 page_cache_release(page);
505 radix_tree_preload_end();
507 mem_cgroup_uncharge_cache_page(page);
511 EXPORT_SYMBOL(add_to_page_cache_locked);
513 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
514 pgoff_t offset, gfp_t gfp_mask)
519 * Splice_read and readahead add shmem/tmpfs pages into the page cache
520 * before shmem_readpage has a chance to mark them as SwapBacked: they
521 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
522 * (called in add_to_page_cache) needs to know where they're going too.
524 if (mapping_cap_swap_backed(mapping))
525 SetPageSwapBacked(page);
527 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
529 if (page_is_file_cache(page))
530 lru_cache_add_file(page);
532 lru_cache_add_active_anon(page);
536 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
539 struct page *__page_cache_alloc(gfp_t gfp)
541 if (cpuset_do_page_mem_spread()) {
542 int n = cpuset_mem_spread_node();
543 return alloc_pages_exact_node(n, gfp, 0);
545 return alloc_pages(gfp, 0);
547 EXPORT_SYMBOL(__page_cache_alloc);
550 static int __sleep_on_page_lock(void *word)
557 * In order to wait for pages to become available there must be
558 * waitqueues associated with pages. By using a hash table of
559 * waitqueues where the bucket discipline is to maintain all
560 * waiters on the same queue and wake all when any of the pages
561 * become available, and for the woken contexts to check to be
562 * sure the appropriate page became available, this saves space
563 * at a cost of "thundering herd" phenomena during rare hash
566 static wait_queue_head_t *page_waitqueue(struct page *page)
568 const struct zone *zone = page_zone(page);
570 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
573 static inline void wake_up_page(struct page *page, int bit)
575 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
578 void wait_on_page_bit(struct page *page, int bit_nr)
580 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
582 if (test_bit(bit_nr, &page->flags))
583 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
584 TASK_UNINTERRUPTIBLE);
586 EXPORT_SYMBOL(wait_on_page_bit);
589 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
590 * @page: Page defining the wait queue of interest
591 * @waiter: Waiter to add to the queue
593 * Add an arbitrary @waiter to the wait queue for the nominated @page.
595 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
597 wait_queue_head_t *q = page_waitqueue(page);
600 spin_lock_irqsave(&q->lock, flags);
601 __add_wait_queue(q, waiter);
602 spin_unlock_irqrestore(&q->lock, flags);
604 EXPORT_SYMBOL_GPL(add_page_wait_queue);
607 * unlock_page - unlock a locked page
610 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
611 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
612 * mechananism between PageLocked pages and PageWriteback pages is shared.
613 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
615 * The mb is necessary to enforce ordering between the clear_bit and the read
616 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
618 void unlock_page(struct page *page)
620 VM_BUG_ON(!PageLocked(page));
621 clear_bit_unlock(PG_locked, &page->flags);
622 smp_mb__after_clear_bit();
623 wake_up_page(page, PG_locked);
625 EXPORT_SYMBOL(unlock_page);
628 * end_page_writeback - end writeback against a page
631 void end_page_writeback(struct page *page)
633 if (TestClearPageReclaim(page))
634 rotate_reclaimable_page(page);
636 if (!test_clear_page_writeback(page))
639 smp_mb__after_clear_bit();
640 wake_up_page(page, PG_writeback);
642 EXPORT_SYMBOL(end_page_writeback);
645 * __lock_page - get a lock on the page, assuming we need to sleep to get it
646 * @page: the page to lock
648 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
649 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
650 * chances are that on the second loop, the block layer's plug list is empty,
651 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
653 void __lock_page(struct page *page)
655 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
657 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
658 TASK_UNINTERRUPTIBLE);
660 EXPORT_SYMBOL(__lock_page);
662 int __lock_page_killable(struct page *page)
664 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
666 return __wait_on_bit_lock(page_waitqueue(page), &wait,
667 sync_page_killable, TASK_KILLABLE);
669 EXPORT_SYMBOL_GPL(__lock_page_killable);
672 * __lock_page_nosync - get a lock on the page, without calling sync_page()
673 * @page: the page to lock
675 * Variant of lock_page that does not require the caller to hold a reference
676 * on the page's mapping.
678 void __lock_page_nosync(struct page *page)
680 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
681 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
682 TASK_UNINTERRUPTIBLE);
686 * find_get_page - find and get a page reference
687 * @mapping: the address_space to search
688 * @offset: the page index
690 * Is there a pagecache struct page at the given (mapping, offset) tuple?
691 * If yes, increment its refcount and return it; if no, return NULL.
693 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
701 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
703 page = radix_tree_deref_slot(pagep);
704 if (unlikely(!page || page == RADIX_TREE_RETRY))
707 if (!page_cache_get_speculative(page))
711 * Has the page moved?
712 * This is part of the lockless pagecache protocol. See
713 * include/linux/pagemap.h for details.
715 if (unlikely(page != *pagep)) {
716 page_cache_release(page);
724 EXPORT_SYMBOL(find_get_page);
727 * find_lock_page - locate, pin and lock a pagecache page
728 * @mapping: the address_space to search
729 * @offset: the page index
731 * Locates the desired pagecache page, locks it, increments its reference
732 * count and returns its address.
734 * Returns zero if the page was not present. find_lock_page() may sleep.
736 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
741 page = find_get_page(mapping, offset);
744 /* Has the page been truncated? */
745 if (unlikely(page->mapping != mapping)) {
747 page_cache_release(page);
750 VM_BUG_ON(page->index != offset);
754 EXPORT_SYMBOL(find_lock_page);
757 * find_or_create_page - locate or add a pagecache page
758 * @mapping: the page's address_space
759 * @index: the page's index into the mapping
760 * @gfp_mask: page allocation mode
762 * Locates a page in the pagecache. If the page is not present, a new page
763 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
764 * LRU list. The returned page is locked and has its reference count
767 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
770 * find_or_create_page() returns the desired page's address, or zero on
773 struct page *find_or_create_page(struct address_space *mapping,
774 pgoff_t index, gfp_t gfp_mask)
779 page = find_lock_page(mapping, index);
781 page = __page_cache_alloc(gfp_mask);
785 * We want a regular kernel memory (not highmem or DMA etc)
786 * allocation for the radix tree nodes, but we need to honour
787 * the context-specific requirements the caller has asked for.
788 * GFP_RECLAIM_MASK collects those requirements.
790 err = add_to_page_cache_lru(page, mapping, index,
791 (gfp_mask & GFP_RECLAIM_MASK));
793 page_cache_release(page);
801 EXPORT_SYMBOL(find_or_create_page);
804 * find_get_pages - gang pagecache lookup
805 * @mapping: The address_space to search
806 * @start: The starting page index
807 * @nr_pages: The maximum number of pages
808 * @pages: Where the resulting pages are placed
810 * find_get_pages() will search for and return a group of up to
811 * @nr_pages pages in the mapping. The pages are placed at @pages.
812 * find_get_pages() takes a reference against the returned pages.
814 * The search returns a group of mapping-contiguous pages with ascending
815 * indexes. There may be holes in the indices due to not-present pages.
817 * find_get_pages() returns the number of pages which were found.
819 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
820 unsigned int nr_pages, struct page **pages)
824 unsigned int nr_found;
828 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
829 (void ***)pages, start, nr_pages);
831 for (i = 0; i < nr_found; i++) {
834 page = radix_tree_deref_slot((void **)pages[i]);
838 * this can only trigger if nr_found == 1, making livelock
841 if (unlikely(page == RADIX_TREE_RETRY))
844 if (!page_cache_get_speculative(page))
847 /* Has the page moved? */
848 if (unlikely(page != *((void **)pages[i]))) {
849 page_cache_release(page);
861 * find_get_pages_contig - gang contiguous pagecache lookup
862 * @mapping: The address_space to search
863 * @index: The starting page index
864 * @nr_pages: The maximum number of pages
865 * @pages: Where the resulting pages are placed
867 * find_get_pages_contig() works exactly like find_get_pages(), except
868 * that the returned number of pages are guaranteed to be contiguous.
870 * find_get_pages_contig() returns the number of pages which were found.
872 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
873 unsigned int nr_pages, struct page **pages)
877 unsigned int nr_found;
881 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
882 (void ***)pages, index, nr_pages);
884 for (i = 0; i < nr_found; i++) {
887 page = radix_tree_deref_slot((void **)pages[i]);
891 * this can only trigger if nr_found == 1, making livelock
894 if (unlikely(page == RADIX_TREE_RETRY))
897 if (page->mapping == NULL || page->index != index)
900 if (!page_cache_get_speculative(page))
903 /* Has the page moved? */
904 if (unlikely(page != *((void **)pages[i]))) {
905 page_cache_release(page);
916 EXPORT_SYMBOL(find_get_pages_contig);
919 * find_get_pages_tag - find and return pages that match @tag
920 * @mapping: the address_space to search
921 * @index: the starting page index
922 * @tag: the tag index
923 * @nr_pages: the maximum number of pages
924 * @pages: where the resulting pages are placed
926 * Like find_get_pages, except we only return pages which are tagged with
927 * @tag. We update @index to index the next page for the traversal.
929 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
930 int tag, unsigned int nr_pages, struct page **pages)
934 unsigned int nr_found;
938 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
939 (void ***)pages, *index, nr_pages, tag);
941 for (i = 0; i < nr_found; i++) {
944 page = radix_tree_deref_slot((void **)pages[i]);
948 * this can only trigger if nr_found == 1, making livelock
951 if (unlikely(page == RADIX_TREE_RETRY))
954 if (!page_cache_get_speculative(page))
957 /* Has the page moved? */
958 if (unlikely(page != *((void **)pages[i]))) {
959 page_cache_release(page);
969 *index = pages[ret - 1]->index + 1;
973 EXPORT_SYMBOL(find_get_pages_tag);
976 * grab_cache_page_nowait - returns locked page at given index in given cache
977 * @mapping: target address_space
978 * @index: the page index
980 * Same as grab_cache_page(), but do not wait if the page is unavailable.
981 * This is intended for speculative data generators, where the data can
982 * be regenerated if the page couldn't be grabbed. This routine should
983 * be safe to call while holding the lock for another page.
985 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
986 * and deadlock against the caller's locked page.
989 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
991 struct page *page = find_get_page(mapping, index);
994 if (trylock_page(page))
996 page_cache_release(page);
999 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1000 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1001 page_cache_release(page);
1006 EXPORT_SYMBOL(grab_cache_page_nowait);
1009 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1010 * a _large_ part of the i/o request. Imagine the worst scenario:
1012 * ---R__________________________________________B__________
1013 * ^ reading here ^ bad block(assume 4k)
1015 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1016 * => failing the whole request => read(R) => read(R+1) =>
1017 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1018 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1019 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1021 * It is going insane. Fix it by quickly scaling down the readahead size.
1023 static void shrink_readahead_size_eio(struct file *filp,
1024 struct file_ra_state *ra)
1030 * do_generic_file_read - generic file read routine
1031 * @filp: the file to read
1032 * @ppos: current file position
1033 * @desc: read_descriptor
1034 * @actor: read method
1036 * This is a generic file read routine, and uses the
1037 * mapping->a_ops->readpage() function for the actual low-level stuff.
1039 * This is really ugly. But the goto's actually try to clarify some
1040 * of the logic when it comes to error handling etc.
1042 static void do_generic_file_read(struct file *filp, loff_t *ppos,
1043 read_descriptor_t *desc, read_actor_t actor)
1045 struct address_space *mapping = filp->f_mapping;
1046 struct inode *inode = mapping->host;
1047 struct file_ra_state *ra = &filp->f_ra;
1051 unsigned long offset; /* offset into pagecache page */
1052 unsigned int prev_offset;
1055 index = *ppos >> PAGE_CACHE_SHIFT;
1056 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1057 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1058 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1059 offset = *ppos & ~PAGE_CACHE_MASK;
1065 unsigned long nr, ret;
1069 page = find_get_page(mapping, index);
1071 page_cache_sync_readahead(mapping,
1073 index, last_index - index);
1074 page = find_get_page(mapping, index);
1075 if (unlikely(page == NULL))
1076 goto no_cached_page;
1078 if (PageReadahead(page)) {
1079 page_cache_async_readahead(mapping,
1081 index, last_index - index);
1083 if (!PageUptodate(page)) {
1084 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1085 !mapping->a_ops->is_partially_uptodate)
1086 goto page_not_up_to_date;
1087 if (!trylock_page(page))
1088 goto page_not_up_to_date;
1089 if (!mapping->a_ops->is_partially_uptodate(page,
1091 goto page_not_up_to_date_locked;
1096 * i_size must be checked after we know the page is Uptodate.
1098 * Checking i_size after the check allows us to calculate
1099 * the correct value for "nr", which means the zero-filled
1100 * part of the page is not copied back to userspace (unless
1101 * another truncate extends the file - this is desired though).
1104 isize = i_size_read(inode);
1105 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1106 if (unlikely(!isize || index > end_index)) {
1107 page_cache_release(page);
1111 /* nr is the maximum number of bytes to copy from this page */
1112 nr = PAGE_CACHE_SIZE;
1113 if (index == end_index) {
1114 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1116 page_cache_release(page);
1122 /* If users can be writing to this page using arbitrary
1123 * virtual addresses, take care about potential aliasing
1124 * before reading the page on the kernel side.
1126 if (mapping_writably_mapped(mapping))
1127 flush_dcache_page(page);
1130 * When a sequential read accesses a page several times,
1131 * only mark it as accessed the first time.
1133 if (prev_index != index || offset != prev_offset)
1134 mark_page_accessed(page);
1138 * Ok, we have the page, and it's up-to-date, so
1139 * now we can copy it to user space...
1141 * The actor routine returns how many bytes were actually used..
1142 * NOTE! This may not be the same as how much of a user buffer
1143 * we filled up (we may be padding etc), so we can only update
1144 * "pos" here (the actor routine has to update the user buffer
1145 * pointers and the remaining count).
1147 ret = actor(desc, page, offset, nr);
1149 index += offset >> PAGE_CACHE_SHIFT;
1150 offset &= ~PAGE_CACHE_MASK;
1151 prev_offset = offset;
1153 page_cache_release(page);
1154 if (ret == nr && desc->count)
1158 page_not_up_to_date:
1159 /* Get exclusive access to the page ... */
1160 error = lock_page_killable(page);
1161 if (unlikely(error))
1162 goto readpage_error;
1164 page_not_up_to_date_locked:
1165 /* Did it get truncated before we got the lock? */
1166 if (!page->mapping) {
1168 page_cache_release(page);
1172 /* Did somebody else fill it already? */
1173 if (PageUptodate(page)) {
1179 /* Start the actual read. The read will unlock the page. */
1180 error = mapping->a_ops->readpage(filp, page);
1182 if (unlikely(error)) {
1183 if (error == AOP_TRUNCATED_PAGE) {
1184 page_cache_release(page);
1187 goto readpage_error;
1190 if (!PageUptodate(page)) {
1191 error = lock_page_killable(page);
1192 if (unlikely(error))
1193 goto readpage_error;
1194 if (!PageUptodate(page)) {
1195 if (page->mapping == NULL) {
1197 * invalidate_inode_pages got it
1200 page_cache_release(page);
1204 shrink_readahead_size_eio(filp, ra);
1206 goto readpage_error;
1214 /* UHHUH! A synchronous read error occurred. Report it */
1215 desc->error = error;
1216 page_cache_release(page);
1221 * Ok, it wasn't cached, so we need to create a new
1224 page = page_cache_alloc_cold(mapping);
1226 desc->error = -ENOMEM;
1229 error = add_to_page_cache_lru(page, mapping,
1232 page_cache_release(page);
1233 if (error == -EEXIST)
1235 desc->error = error;
1242 ra->prev_pos = prev_index;
1243 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1244 ra->prev_pos |= prev_offset;
1246 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1247 file_accessed(filp);
1250 int file_read_actor(read_descriptor_t *desc, struct page *page,
1251 unsigned long offset, unsigned long size)
1254 unsigned long left, count = desc->count;
1260 * Faults on the destination of a read are common, so do it before
1263 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1264 kaddr = kmap_atomic(page, KM_USER0);
1265 left = __copy_to_user_inatomic(desc->arg.buf,
1266 kaddr + offset, size);
1267 kunmap_atomic(kaddr, KM_USER0);
1272 /* Do it the slow way */
1274 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1279 desc->error = -EFAULT;
1282 desc->count = count - size;
1283 desc->written += size;
1284 desc->arg.buf += size;
1289 * Performs necessary checks before doing a write
1290 * @iov: io vector request
1291 * @nr_segs: number of segments in the iovec
1292 * @count: number of bytes to write
1293 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1295 * Adjust number of segments and amount of bytes to write (nr_segs should be
1296 * properly initialized first). Returns appropriate error code that caller
1297 * should return or zero in case that write should be allowed.
1299 int generic_segment_checks(const struct iovec *iov,
1300 unsigned long *nr_segs, size_t *count, int access_flags)
1304 for (seg = 0; seg < *nr_segs; seg++) {
1305 const struct iovec *iv = &iov[seg];
1308 * If any segment has a negative length, or the cumulative
1309 * length ever wraps negative then return -EINVAL.
1312 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1314 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1319 cnt -= iv->iov_len; /* This segment is no good */
1325 EXPORT_SYMBOL(generic_segment_checks);
1328 * generic_file_aio_read - generic filesystem read routine
1329 * @iocb: kernel I/O control block
1330 * @iov: io vector request
1331 * @nr_segs: number of segments in the iovec
1332 * @pos: current file position
1334 * This is the "read()" routine for all filesystems
1335 * that can use the page cache directly.
1338 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1339 unsigned long nr_segs, loff_t pos)
1341 struct file *filp = iocb->ki_filp;
1345 loff_t *ppos = &iocb->ki_pos;
1348 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1352 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1353 if (filp->f_flags & O_DIRECT) {
1355 struct address_space *mapping;
1356 struct inode *inode;
1358 mapping = filp->f_mapping;
1359 inode = mapping->host;
1361 goto out; /* skip atime */
1362 size = i_size_read(inode);
1364 retval = filemap_write_and_wait_range(mapping, pos,
1365 pos + iov_length(iov, nr_segs) - 1);
1367 retval = mapping->a_ops->direct_IO(READ, iocb,
1371 *ppos = pos + retval;
1373 file_accessed(filp);
1379 for (seg = 0; seg < nr_segs; seg++) {
1380 read_descriptor_t desc;
1383 desc.arg.buf = iov[seg].iov_base;
1384 desc.count = iov[seg].iov_len;
1385 if (desc.count == 0)
1388 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1389 retval += desc.written;
1391 retval = retval ?: desc.error;
1400 EXPORT_SYMBOL(generic_file_aio_read);
1403 do_readahead(struct address_space *mapping, struct file *filp,
1404 pgoff_t index, unsigned long nr)
1406 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1409 force_page_cache_readahead(mapping, filp, index, nr);
1413 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1421 if (file->f_mode & FMODE_READ) {
1422 struct address_space *mapping = file->f_mapping;
1423 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1424 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1425 unsigned long len = end - start + 1;
1426 ret = do_readahead(mapping, file, start, len);
1432 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1433 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1435 return SYSC_readahead((int) fd, offset, (size_t) count);
1437 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1442 * page_cache_read - adds requested page to the page cache if not already there
1443 * @file: file to read
1444 * @offset: page index
1446 * This adds the requested page to the page cache if it isn't already there,
1447 * and schedules an I/O to read in its contents from disk.
1449 static int page_cache_read(struct file *file, pgoff_t offset)
1451 struct address_space *mapping = file->f_mapping;
1456 page = page_cache_alloc_cold(mapping);
1460 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1462 ret = mapping->a_ops->readpage(file, page);
1463 else if (ret == -EEXIST)
1464 ret = 0; /* losing race to add is OK */
1466 page_cache_release(page);
1468 } while (ret == AOP_TRUNCATED_PAGE);
1473 #define MMAP_LOTSAMISS (100)
1476 * Synchronous readahead happens when we don't even find
1477 * a page in the page cache at all.
1479 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1480 struct file_ra_state *ra,
1484 unsigned long ra_pages;
1485 struct address_space *mapping = file->f_mapping;
1487 /* If we don't want any read-ahead, don't bother */
1488 if (VM_RandomReadHint(vma))
1491 if (VM_SequentialReadHint(vma) ||
1492 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1493 page_cache_sync_readahead(mapping, ra, file, offset,
1498 if (ra->mmap_miss < INT_MAX)
1502 * Do we miss much more than hit in this file? If so,
1503 * stop bothering with read-ahead. It will only hurt.
1505 if (ra->mmap_miss > MMAP_LOTSAMISS)
1511 ra_pages = max_sane_readahead(ra->ra_pages);
1513 ra->start = max_t(long, 0, offset - ra_pages/2);
1514 ra->size = ra_pages;
1516 ra_submit(ra, mapping, file);
1521 * Asynchronous readahead happens when we find the page and PG_readahead,
1522 * so we want to possibly extend the readahead further..
1524 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1525 struct file_ra_state *ra,
1530 struct address_space *mapping = file->f_mapping;
1532 /* If we don't want any read-ahead, don't bother */
1533 if (VM_RandomReadHint(vma))
1535 if (ra->mmap_miss > 0)
1537 if (PageReadahead(page))
1538 page_cache_async_readahead(mapping, ra, file,
1539 page, offset, ra->ra_pages);
1543 * filemap_fault - read in file data for page fault handling
1544 * @vma: vma in which the fault was taken
1545 * @vmf: struct vm_fault containing details of the fault
1547 * filemap_fault() is invoked via the vma operations vector for a
1548 * mapped memory region to read in file data during a page fault.
1550 * The goto's are kind of ugly, but this streamlines the normal case of having
1551 * it in the page cache, and handles the special cases reasonably without
1552 * having a lot of duplicated code.
1554 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1557 struct file *file = vma->vm_file;
1558 struct address_space *mapping = file->f_mapping;
1559 struct file_ra_state *ra = &file->f_ra;
1560 struct inode *inode = mapping->host;
1561 pgoff_t offset = vmf->pgoff;
1566 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1568 return VM_FAULT_SIGBUS;
1571 * Do we have something in the page cache already?
1573 page = find_get_page(mapping, offset);
1576 * We found the page, so try async readahead before
1577 * waiting for the lock.
1579 do_async_mmap_readahead(vma, ra, file, page, offset);
1582 /* Did it get truncated? */
1583 if (unlikely(page->mapping != mapping)) {
1586 goto no_cached_page;
1589 /* No page in the page cache at all */
1590 do_sync_mmap_readahead(vma, ra, file, offset);
1591 count_vm_event(PGMAJFAULT);
1592 ret = VM_FAULT_MAJOR;
1594 page = find_lock_page(mapping, offset);
1596 goto no_cached_page;
1600 * We have a locked page in the page cache, now we need to check
1601 * that it's up-to-date. If not, it is going to be due to an error.
1603 if (unlikely(!PageUptodate(page)))
1604 goto page_not_uptodate;
1607 * Found the page and have a reference on it.
1608 * We must recheck i_size under page lock.
1610 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1611 if (unlikely(offset >= size)) {
1613 page_cache_release(page);
1614 return VM_FAULT_SIGBUS;
1617 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1619 return ret | VM_FAULT_LOCKED;
1623 * We're only likely to ever get here if MADV_RANDOM is in
1626 error = page_cache_read(file, offset);
1629 * The page we want has now been added to the page cache.
1630 * In the unlikely event that someone removed it in the
1631 * meantime, we'll just come back here and read it again.
1637 * An error return from page_cache_read can result if the
1638 * system is low on memory, or a problem occurs while trying
1641 if (error == -ENOMEM)
1642 return VM_FAULT_OOM;
1643 return VM_FAULT_SIGBUS;
1647 * Umm, take care of errors if the page isn't up-to-date.
1648 * Try to re-read it _once_. We do this synchronously,
1649 * because there really aren't any performance issues here
1650 * and we need to check for errors.
1652 ClearPageError(page);
1653 error = mapping->a_ops->readpage(file, page);
1655 wait_on_page_locked(page);
1656 if (!PageUptodate(page))
1659 page_cache_release(page);
1661 if (!error || error == AOP_TRUNCATED_PAGE)
1664 /* Things didn't work out. Return zero to tell the mm layer so. */
1665 shrink_readahead_size_eio(file, ra);
1666 return VM_FAULT_SIGBUS;
1668 EXPORT_SYMBOL(filemap_fault);
1670 struct vm_operations_struct generic_file_vm_ops = {
1671 .fault = filemap_fault,
1674 /* This is used for a general mmap of a disk file */
1676 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1678 struct address_space *mapping = file->f_mapping;
1680 if (!mapping->a_ops->readpage)
1682 file_accessed(file);
1683 vma->vm_ops = &generic_file_vm_ops;
1684 vma->vm_flags |= VM_CAN_NONLINEAR;
1689 * This is for filesystems which do not implement ->writepage.
1691 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1693 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1695 return generic_file_mmap(file, vma);
1698 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1702 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1706 #endif /* CONFIG_MMU */
1708 EXPORT_SYMBOL(generic_file_mmap);
1709 EXPORT_SYMBOL(generic_file_readonly_mmap);
1711 static struct page *__read_cache_page(struct address_space *mapping,
1713 int (*filler)(void *,struct page*),
1719 page = find_get_page(mapping, index);
1721 page = page_cache_alloc_cold(mapping);
1723 return ERR_PTR(-ENOMEM);
1724 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1725 if (unlikely(err)) {
1726 page_cache_release(page);
1729 /* Presumably ENOMEM for radix tree node */
1730 return ERR_PTR(err);
1732 err = filler(data, page);
1734 page_cache_release(page);
1735 page = ERR_PTR(err);
1742 * read_cache_page_async - read into page cache, fill it if needed
1743 * @mapping: the page's address_space
1744 * @index: the page index
1745 * @filler: function to perform the read
1746 * @data: destination for read data
1748 * Same as read_cache_page, but don't wait for page to become unlocked
1749 * after submitting it to the filler.
1751 * Read into the page cache. If a page already exists, and PageUptodate() is
1752 * not set, try to fill the page but don't wait for it to become unlocked.
1754 * If the page does not get brought uptodate, return -EIO.
1756 struct page *read_cache_page_async(struct address_space *mapping,
1758 int (*filler)(void *,struct page*),
1765 page = __read_cache_page(mapping, index, filler, data);
1768 if (PageUptodate(page))
1772 if (!page->mapping) {
1774 page_cache_release(page);
1777 if (PageUptodate(page)) {
1781 err = filler(data, page);
1783 page_cache_release(page);
1784 return ERR_PTR(err);
1787 mark_page_accessed(page);
1790 EXPORT_SYMBOL(read_cache_page_async);
1793 * read_cache_page - read into page cache, fill it if needed
1794 * @mapping: the page's address_space
1795 * @index: the page index
1796 * @filler: function to perform the read
1797 * @data: destination for read data
1799 * Read into the page cache. If a page already exists, and PageUptodate() is
1800 * not set, try to fill the page then wait for it to become unlocked.
1802 * If the page does not get brought uptodate, return -EIO.
1804 struct page *read_cache_page(struct address_space *mapping,
1806 int (*filler)(void *,struct page*),
1811 page = read_cache_page_async(mapping, index, filler, data);
1814 wait_on_page_locked(page);
1815 if (!PageUptodate(page)) {
1816 page_cache_release(page);
1817 page = ERR_PTR(-EIO);
1822 EXPORT_SYMBOL(read_cache_page);
1825 * The logic we want is
1827 * if suid or (sgid and xgrp)
1830 int should_remove_suid(struct dentry *dentry)
1832 mode_t mode = dentry->d_inode->i_mode;
1835 /* suid always must be killed */
1836 if (unlikely(mode & S_ISUID))
1837 kill = ATTR_KILL_SUID;
1840 * sgid without any exec bits is just a mandatory locking mark; leave
1841 * it alone. If some exec bits are set, it's a real sgid; kill it.
1843 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1844 kill |= ATTR_KILL_SGID;
1846 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1851 EXPORT_SYMBOL(should_remove_suid);
1853 static int __remove_suid(struct dentry *dentry, int kill)
1855 struct iattr newattrs;
1857 newattrs.ia_valid = ATTR_FORCE | kill;
1858 return notify_change(dentry, &newattrs);
1861 int file_remove_suid(struct file *file)
1863 struct dentry *dentry = file->f_path.dentry;
1864 int killsuid = should_remove_suid(dentry);
1865 int killpriv = security_inode_need_killpriv(dentry);
1871 error = security_inode_killpriv(dentry);
1872 if (!error && killsuid)
1873 error = __remove_suid(dentry, killsuid);
1877 EXPORT_SYMBOL(file_remove_suid);
1879 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1880 const struct iovec *iov, size_t base, size_t bytes)
1882 size_t copied = 0, left = 0;
1885 char __user *buf = iov->iov_base + base;
1886 int copy = min(bytes, iov->iov_len - base);
1889 left = __copy_from_user_inatomic(vaddr, buf, copy);
1898 return copied - left;
1902 * Copy as much as we can into the page and return the number of bytes which
1903 * were sucessfully copied. If a fault is encountered then return the number of
1904 * bytes which were copied.
1906 size_t iov_iter_copy_from_user_atomic(struct page *page,
1907 struct iov_iter *i, unsigned long offset, size_t bytes)
1912 BUG_ON(!in_atomic());
1913 kaddr = kmap_atomic(page, KM_USER0);
1914 if (likely(i->nr_segs == 1)) {
1916 char __user *buf = i->iov->iov_base + i->iov_offset;
1917 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1918 copied = bytes - left;
1920 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1921 i->iov, i->iov_offset, bytes);
1923 kunmap_atomic(kaddr, KM_USER0);
1927 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1930 * This has the same sideeffects and return value as
1931 * iov_iter_copy_from_user_atomic().
1932 * The difference is that it attempts to resolve faults.
1933 * Page must not be locked.
1935 size_t iov_iter_copy_from_user(struct page *page,
1936 struct iov_iter *i, unsigned long offset, size_t bytes)
1942 if (likely(i->nr_segs == 1)) {
1944 char __user *buf = i->iov->iov_base + i->iov_offset;
1945 left = __copy_from_user(kaddr + offset, buf, bytes);
1946 copied = bytes - left;
1948 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1949 i->iov, i->iov_offset, bytes);
1954 EXPORT_SYMBOL(iov_iter_copy_from_user);
1956 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1958 BUG_ON(i->count < bytes);
1960 if (likely(i->nr_segs == 1)) {
1961 i->iov_offset += bytes;
1964 const struct iovec *iov = i->iov;
1965 size_t base = i->iov_offset;
1968 * The !iov->iov_len check ensures we skip over unlikely
1969 * zero-length segments (without overruning the iovec).
1971 while (bytes || unlikely(i->count && !iov->iov_len)) {
1974 copy = min(bytes, iov->iov_len - base);
1975 BUG_ON(!i->count || i->count < copy);
1979 if (iov->iov_len == base) {
1985 i->iov_offset = base;
1988 EXPORT_SYMBOL(iov_iter_advance);
1991 * Fault in the first iovec of the given iov_iter, to a maximum length
1992 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1993 * accessed (ie. because it is an invalid address).
1995 * writev-intensive code may want this to prefault several iovecs -- that
1996 * would be possible (callers must not rely on the fact that _only_ the
1997 * first iovec will be faulted with the current implementation).
1999 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2001 char __user *buf = i->iov->iov_base + i->iov_offset;
2002 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2003 return fault_in_pages_readable(buf, bytes);
2005 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2008 * Return the count of just the current iov_iter segment.
2010 size_t iov_iter_single_seg_count(struct iov_iter *i)
2012 const struct iovec *iov = i->iov;
2013 if (i->nr_segs == 1)
2016 return min(i->count, iov->iov_len - i->iov_offset);
2018 EXPORT_SYMBOL(iov_iter_single_seg_count);
2021 * Performs necessary checks before doing a write
2023 * Can adjust writing position or amount of bytes to write.
2024 * Returns appropriate error code that caller should return or
2025 * zero in case that write should be allowed.
2027 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2029 struct inode *inode = file->f_mapping->host;
2030 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2032 if (unlikely(*pos < 0))
2036 /* FIXME: this is for backwards compatibility with 2.4 */
2037 if (file->f_flags & O_APPEND)
2038 *pos = i_size_read(inode);
2040 if (limit != RLIM_INFINITY) {
2041 if (*pos >= limit) {
2042 send_sig(SIGXFSZ, current, 0);
2045 if (*count > limit - (typeof(limit))*pos) {
2046 *count = limit - (typeof(limit))*pos;
2054 if (unlikely(*pos + *count > MAX_NON_LFS &&
2055 !(file->f_flags & O_LARGEFILE))) {
2056 if (*pos >= MAX_NON_LFS) {
2059 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2060 *count = MAX_NON_LFS - (unsigned long)*pos;
2065 * Are we about to exceed the fs block limit ?
2067 * If we have written data it becomes a short write. If we have
2068 * exceeded without writing data we send a signal and return EFBIG.
2069 * Linus frestrict idea will clean these up nicely..
2071 if (likely(!isblk)) {
2072 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2073 if (*count || *pos > inode->i_sb->s_maxbytes) {
2076 /* zero-length writes at ->s_maxbytes are OK */
2079 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2080 *count = inode->i_sb->s_maxbytes - *pos;
2084 if (bdev_read_only(I_BDEV(inode)))
2086 isize = i_size_read(inode);
2087 if (*pos >= isize) {
2088 if (*count || *pos > isize)
2092 if (*pos + *count > isize)
2093 *count = isize - *pos;
2100 EXPORT_SYMBOL(generic_write_checks);
2102 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2103 loff_t pos, unsigned len, unsigned flags,
2104 struct page **pagep, void **fsdata)
2106 const struct address_space_operations *aops = mapping->a_ops;
2108 return aops->write_begin(file, mapping, pos, len, flags,
2111 EXPORT_SYMBOL(pagecache_write_begin);
2113 int pagecache_write_end(struct file *file, struct address_space *mapping,
2114 loff_t pos, unsigned len, unsigned copied,
2115 struct page *page, void *fsdata)
2117 const struct address_space_operations *aops = mapping->a_ops;
2119 mark_page_accessed(page);
2120 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2122 EXPORT_SYMBOL(pagecache_write_end);
2125 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2126 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2127 size_t count, size_t ocount)
2129 struct file *file = iocb->ki_filp;
2130 struct address_space *mapping = file->f_mapping;
2131 struct inode *inode = mapping->host;
2136 if (count != ocount)
2137 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2139 write_len = iov_length(iov, *nr_segs);
2140 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2142 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2147 * After a write we want buffered reads to be sure to go to disk to get
2148 * the new data. We invalidate clean cached page from the region we're
2149 * about to write. We do this *before* the write so that we can return
2150 * without clobbering -EIOCBQUEUED from ->direct_IO().
2152 if (mapping->nrpages) {
2153 written = invalidate_inode_pages2_range(mapping,
2154 pos >> PAGE_CACHE_SHIFT, end);
2156 * If a page can not be invalidated, return 0 to fall back
2157 * to buffered write.
2160 if (written == -EBUSY)
2166 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2169 * Finally, try again to invalidate clean pages which might have been
2170 * cached by non-direct readahead, or faulted in by get_user_pages()
2171 * if the source of the write was an mmap'ed region of the file
2172 * we're writing. Either one is a pretty crazy thing to do,
2173 * so we don't support it 100%. If this invalidation
2174 * fails, tough, the write still worked...
2176 if (mapping->nrpages) {
2177 invalidate_inode_pages2_range(mapping,
2178 pos >> PAGE_CACHE_SHIFT, end);
2182 loff_t end = pos + written;
2183 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2184 i_size_write(inode, end);
2185 mark_inode_dirty(inode);
2192 EXPORT_SYMBOL(generic_file_direct_write);
2195 * Find or create a page at the given pagecache position. Return the locked
2196 * page. This function is specifically for buffered writes.
2198 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2199 pgoff_t index, unsigned flags)
2203 gfp_t gfp_notmask = 0;
2204 if (flags & AOP_FLAG_NOFS)
2205 gfp_notmask = __GFP_FS;
2207 page = find_lock_page(mapping, index);
2211 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2214 status = add_to_page_cache_lru(page, mapping, index,
2215 GFP_KERNEL & ~gfp_notmask);
2216 if (unlikely(status)) {
2217 page_cache_release(page);
2218 if (status == -EEXIST)
2224 EXPORT_SYMBOL(grab_cache_page_write_begin);
2226 static ssize_t generic_perform_write(struct file *file,
2227 struct iov_iter *i, loff_t pos)
2229 struct address_space *mapping = file->f_mapping;
2230 const struct address_space_operations *a_ops = mapping->a_ops;
2232 ssize_t written = 0;
2233 unsigned int flags = 0;
2236 * Copies from kernel address space cannot fail (NFSD is a big user).
2238 if (segment_eq(get_fs(), KERNEL_DS))
2239 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2243 pgoff_t index; /* Pagecache index for current page */
2244 unsigned long offset; /* Offset into pagecache page */
2245 unsigned long bytes; /* Bytes to write to page */
2246 size_t copied; /* Bytes copied from user */
2249 offset = (pos & (PAGE_CACHE_SIZE - 1));
2250 index = pos >> PAGE_CACHE_SHIFT;
2251 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2257 * Bring in the user page that we will copy from _first_.
2258 * Otherwise there's a nasty deadlock on copying from the
2259 * same page as we're writing to, without it being marked
2262 * Not only is this an optimisation, but it is also required
2263 * to check that the address is actually valid, when atomic
2264 * usercopies are used, below.
2266 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2271 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2273 if (unlikely(status))
2276 pagefault_disable();
2277 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2279 flush_dcache_page(page);
2281 mark_page_accessed(page);
2282 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2284 if (unlikely(status < 0))
2290 iov_iter_advance(i, copied);
2291 if (unlikely(copied == 0)) {
2293 * If we were unable to copy any data at all, we must
2294 * fall back to a single segment length write.
2296 * If we didn't fallback here, we could livelock
2297 * because not all segments in the iov can be copied at
2298 * once without a pagefault.
2300 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2301 iov_iter_single_seg_count(i));
2307 balance_dirty_pages_ratelimited(mapping);
2309 } while (iov_iter_count(i));
2311 return written ? written : status;
2315 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2316 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2317 size_t count, ssize_t written)
2319 struct file *file = iocb->ki_filp;
2320 struct address_space *mapping = file->f_mapping;
2324 iov_iter_init(&i, iov, nr_segs, count, written);
2325 status = generic_perform_write(file, &i, pos);
2327 if (likely(status >= 0)) {
2329 *ppos = pos + status;
2333 * If we get here for O_DIRECT writes then we must have fallen through
2334 * to buffered writes (block instantiation inside i_size). So we sync
2335 * the file data here, to try to honour O_DIRECT expectations.
2337 if (unlikely(file->f_flags & O_DIRECT) && written)
2338 status = filemap_write_and_wait_range(mapping,
2339 pos, pos + written - 1);
2341 return written ? written : status;
2343 EXPORT_SYMBOL(generic_file_buffered_write);
2346 * __generic_file_aio_write - write data to a file
2347 * @iocb: IO state structure (file, offset, etc.)
2348 * @iov: vector with data to write
2349 * @nr_segs: number of segments in the vector
2350 * @ppos: position where to write
2352 * This function does all the work needed for actually writing data to a
2353 * file. It does all basic checks, removes SUID from the file, updates
2354 * modification times and calls proper subroutines depending on whether we
2355 * do direct IO or a standard buffered write.
2357 * It expects i_mutex to be grabbed unless we work on a block device or similar
2358 * object which does not need locking at all.
2360 * This function does *not* take care of syncing data in case of O_SYNC write.
2361 * A caller has to handle it. This is mainly due to the fact that we want to
2362 * avoid syncing under i_mutex.
2364 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2365 unsigned long nr_segs, loff_t *ppos)
2367 struct file *file = iocb->ki_filp;
2368 struct address_space * mapping = file->f_mapping;
2369 size_t ocount; /* original count */
2370 size_t count; /* after file limit checks */
2371 struct inode *inode = mapping->host;
2377 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2384 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2386 /* We can write back this queue in page reclaim */
2387 current->backing_dev_info = mapping->backing_dev_info;
2390 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2397 err = file_remove_suid(file);
2401 file_update_time(file);
2403 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2404 if (unlikely(file->f_flags & O_DIRECT)) {
2406 ssize_t written_buffered;
2408 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2409 ppos, count, ocount);
2410 if (written < 0 || written == count)
2413 * direct-io write to a hole: fall through to buffered I/O
2414 * for completing the rest of the request.
2418 written_buffered = generic_file_buffered_write(iocb, iov,
2419 nr_segs, pos, ppos, count,
2422 * If generic_file_buffered_write() retuned a synchronous error
2423 * then we want to return the number of bytes which were
2424 * direct-written, or the error code if that was zero. Note
2425 * that this differs from normal direct-io semantics, which
2426 * will return -EFOO even if some bytes were written.
2428 if (written_buffered < 0) {
2429 err = written_buffered;
2434 * We need to ensure that the page cache pages are written to
2435 * disk and invalidated to preserve the expected O_DIRECT
2438 endbyte = pos + written_buffered - written - 1;
2439 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2440 SYNC_FILE_RANGE_WAIT_BEFORE|
2441 SYNC_FILE_RANGE_WRITE|
2442 SYNC_FILE_RANGE_WAIT_AFTER);
2444 written = written_buffered;
2445 invalidate_mapping_pages(mapping,
2446 pos >> PAGE_CACHE_SHIFT,
2447 endbyte >> PAGE_CACHE_SHIFT);
2450 * We don't know how much we wrote, so just return
2451 * the number of bytes which were direct-written
2455 written = generic_file_buffered_write(iocb, iov, nr_segs,
2456 pos, ppos, count, written);
2459 current->backing_dev_info = NULL;
2460 return written ? written : err;
2462 EXPORT_SYMBOL(__generic_file_aio_write);
2465 * generic_file_aio_write - write data to a file
2466 * @iocb: IO state structure
2467 * @iov: vector with data to write
2468 * @nr_segs: number of segments in the vector
2469 * @pos: position in file where to write
2471 * This is a wrapper around __generic_file_aio_write() to be used by most
2472 * filesystems. It takes care of syncing the file in case of O_SYNC file
2473 * and acquires i_mutex as needed.
2475 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2476 unsigned long nr_segs, loff_t pos)
2478 struct file *file = iocb->ki_filp;
2479 struct inode *inode = file->f_mapping->host;
2482 BUG_ON(iocb->ki_pos != pos);
2484 mutex_lock(&inode->i_mutex);
2485 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2486 mutex_unlock(&inode->i_mutex);
2488 if (ret > 0 || ret == -EIOCBQUEUED) {
2491 err = generic_write_sync(file, pos, ret);
2492 if (err < 0 && ret > 0)
2497 EXPORT_SYMBOL(generic_file_aio_write);
2500 * try_to_release_page() - release old fs-specific metadata on a page
2502 * @page: the page which the kernel is trying to free
2503 * @gfp_mask: memory allocation flags (and I/O mode)
2505 * The address_space is to try to release any data against the page
2506 * (presumably at page->private). If the release was successful, return `1'.
2507 * Otherwise return zero.
2509 * This may also be called if PG_fscache is set on a page, indicating that the
2510 * page is known to the local caching routines.
2512 * The @gfp_mask argument specifies whether I/O may be performed to release
2513 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2516 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2518 struct address_space * const mapping = page->mapping;
2520 BUG_ON(!PageLocked(page));
2521 if (PageWriteback(page))
2524 if (mapping && mapping->a_ops->releasepage)
2525 return mapping->a_ops->releasepage(page, gfp_mask);
2526 return try_to_free_buffers(page);
2529 EXPORT_SYMBOL(try_to_release_page);