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() */
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 (truncate_pagecache)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
67 * ->i_mmap_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)
108 * (code doesn't rely on that order, so you could switch it around)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
114 * Remove 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 __remove_from_page_cache(struct page *page)
120 struct address_space *mapping = page->mapping;
122 radix_tree_delete(&mapping->page_tree, page->index);
123 page->mapping = NULL;
125 __dec_zone_page_state(page, NR_FILE_PAGES);
126 if (PageSwapBacked(page))
127 __dec_zone_page_state(page, NR_SHMEM);
128 BUG_ON(page_mapped(page));
131 * Some filesystems seem to re-dirty the page even after
132 * the VM has canceled the dirty bit (eg ext3 journaling).
134 * Fix it up by doing a final dirty accounting check after
135 * having removed the page entirely.
137 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
138 dec_zone_page_state(page, NR_FILE_DIRTY);
139 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
143 void remove_from_page_cache(struct page *page)
145 struct address_space *mapping = page->mapping;
147 BUG_ON(!PageLocked(page));
149 spin_lock_irq(&mapping->tree_lock);
150 __remove_from_page_cache(page);
151 spin_unlock_irq(&mapping->tree_lock);
152 mem_cgroup_uncharge_cache_page(page);
154 EXPORT_SYMBOL(remove_from_page_cache);
156 static int sync_page(void *word)
158 struct address_space *mapping;
161 page = container_of((unsigned long *)word, struct page, flags);
164 * page_mapping() is being called without PG_locked held.
165 * Some knowledge of the state and use of the page is used to
166 * reduce the requirements down to a memory barrier.
167 * The danger here is of a stale page_mapping() return value
168 * indicating a struct address_space different from the one it's
169 * associated with when it is associated with one.
170 * After smp_mb(), it's either the correct page_mapping() for
171 * the page, or an old page_mapping() and the page's own
172 * page_mapping() has gone NULL.
173 * The ->sync_page() address_space operation must tolerate
174 * page_mapping() going NULL. By an amazing coincidence,
175 * this comes about because none of the users of the page
176 * in the ->sync_page() methods make essential use of the
177 * page_mapping(), merely passing the page down to the backing
178 * device's unplug functions when it's non-NULL, which in turn
179 * ignore it for all cases but swap, where only page_private(page) is
180 * of interest. When page_mapping() does go NULL, the entire
181 * call stack gracefully ignores the page and returns.
185 mapping = page_mapping(page);
186 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
187 mapping->a_ops->sync_page(page);
192 static int sync_page_killable(void *word)
195 return fatal_signal_pending(current) ? -EINTR : 0;
199 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
200 * @mapping: address space structure to write
201 * @start: offset in bytes where the range starts
202 * @end: offset in bytes where the range ends (inclusive)
203 * @sync_mode: enable synchronous operation
205 * Start writeback against all of a mapping's dirty pages that lie
206 * within the byte offsets <start, end> inclusive.
208 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
209 * opposed to a regular memory cleansing writeback. The difference between
210 * these two operations is that if a dirty page/buffer is encountered, it must
211 * be waited upon, and not just skipped over.
213 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
214 loff_t end, int sync_mode)
217 struct writeback_control wbc = {
218 .sync_mode = sync_mode,
219 .nr_to_write = LONG_MAX,
220 .range_start = start,
224 if (!mapping_cap_writeback_dirty(mapping))
227 ret = do_writepages(mapping, &wbc);
231 static inline int __filemap_fdatawrite(struct address_space *mapping,
234 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
237 int filemap_fdatawrite(struct address_space *mapping)
239 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
241 EXPORT_SYMBOL(filemap_fdatawrite);
243 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
246 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
248 EXPORT_SYMBOL(filemap_fdatawrite_range);
251 * filemap_flush - mostly a non-blocking flush
252 * @mapping: target address_space
254 * This is a mostly non-blocking flush. Not suitable for data-integrity
255 * purposes - I/O may not be started against all dirty pages.
257 int filemap_flush(struct address_space *mapping)
259 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
261 EXPORT_SYMBOL(filemap_flush);
264 * filemap_fdatawait_range - wait for writeback to complete
265 * @mapping: address space structure to wait for
266 * @start_byte: offset in bytes where the range starts
267 * @end_byte: offset in bytes where the range ends (inclusive)
269 * Walk the list of under-writeback pages of the given address space
270 * in the given range and wait for all of them.
272 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
275 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
276 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
281 if (end_byte < start_byte)
284 pagevec_init(&pvec, 0);
285 while ((index <= end) &&
286 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
287 PAGECACHE_TAG_WRITEBACK,
288 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
291 for (i = 0; i < nr_pages; i++) {
292 struct page *page = pvec.pages[i];
294 /* until radix tree lookup accepts end_index */
295 if (page->index > end)
298 wait_on_page_writeback(page);
302 pagevec_release(&pvec);
306 /* Check for outstanding write errors */
307 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
309 if (test_and_clear_bit(AS_EIO, &mapping->flags))
314 EXPORT_SYMBOL(filemap_fdatawait_range);
317 * filemap_fdatawait - wait for all under-writeback pages to complete
318 * @mapping: address space structure to wait for
320 * Walk the list of under-writeback pages of the given address space
321 * and wait for all of them.
323 int filemap_fdatawait(struct address_space *mapping)
325 loff_t i_size = i_size_read(mapping->host);
330 return filemap_fdatawait_range(mapping, 0, i_size - 1);
332 EXPORT_SYMBOL(filemap_fdatawait);
334 int filemap_write_and_wait(struct address_space *mapping)
338 if (mapping->nrpages) {
339 err = filemap_fdatawrite(mapping);
341 * Even if the above returned error, the pages may be
342 * written partially (e.g. -ENOSPC), so we wait for it.
343 * But the -EIO is special case, it may indicate the worst
344 * thing (e.g. bug) happened, so we avoid waiting for it.
347 int err2 = filemap_fdatawait(mapping);
354 EXPORT_SYMBOL(filemap_write_and_wait);
357 * filemap_write_and_wait_range - write out & wait on a file range
358 * @mapping: the address_space for the pages
359 * @lstart: offset in bytes where the range starts
360 * @lend: offset in bytes where the range ends (inclusive)
362 * Write out and wait upon file offsets lstart->lend, inclusive.
364 * Note that `lend' is inclusive (describes the last byte to be written) so
365 * that this function can be used to write to the very end-of-file (end = -1).
367 int filemap_write_and_wait_range(struct address_space *mapping,
368 loff_t lstart, loff_t lend)
372 if (mapping->nrpages) {
373 err = __filemap_fdatawrite_range(mapping, lstart, lend,
375 /* See comment of filemap_write_and_wait() */
377 int err2 = filemap_fdatawait_range(mapping,
385 EXPORT_SYMBOL(filemap_write_and_wait_range);
388 * add_to_page_cache_locked - add a locked page to the pagecache
390 * @mapping: the page's address_space
391 * @offset: page index
392 * @gfp_mask: page allocation mode
394 * This function is used to add a page to the pagecache. It must be locked.
395 * This function does not add the page to the LRU. The caller must do that.
397 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
398 pgoff_t offset, gfp_t gfp_mask)
402 VM_BUG_ON(!PageLocked(page));
404 error = mem_cgroup_cache_charge(page, current->mm,
405 gfp_mask & GFP_RECLAIM_MASK);
409 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
411 page_cache_get(page);
412 page->mapping = mapping;
413 page->index = offset;
415 spin_lock_irq(&mapping->tree_lock);
416 error = radix_tree_insert(&mapping->page_tree, offset, page);
417 if (likely(!error)) {
419 __inc_zone_page_state(page, NR_FILE_PAGES);
420 if (PageSwapBacked(page))
421 __inc_zone_page_state(page, NR_SHMEM);
422 spin_unlock_irq(&mapping->tree_lock);
424 page->mapping = NULL;
425 spin_unlock_irq(&mapping->tree_lock);
426 mem_cgroup_uncharge_cache_page(page);
427 page_cache_release(page);
429 radix_tree_preload_end();
431 mem_cgroup_uncharge_cache_page(page);
435 EXPORT_SYMBOL(add_to_page_cache_locked);
437 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
438 pgoff_t offset, gfp_t gfp_mask)
443 * Splice_read and readahead add shmem/tmpfs pages into the page cache
444 * before shmem_readpage has a chance to mark them as SwapBacked: they
445 * need to go on the anon lru below, and mem_cgroup_cache_charge
446 * (called in add_to_page_cache) needs to know where they're going too.
448 if (mapping_cap_swap_backed(mapping))
449 SetPageSwapBacked(page);
451 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
453 if (page_is_file_cache(page))
454 lru_cache_add_file(page);
456 lru_cache_add_anon(page);
460 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
463 struct page *__page_cache_alloc(gfp_t gfp)
468 if (cpuset_do_page_mem_spread()) {
470 n = cpuset_mem_spread_node();
471 page = alloc_pages_exact_node(n, gfp, 0);
475 return alloc_pages(gfp, 0);
477 EXPORT_SYMBOL(__page_cache_alloc);
480 static int __sleep_on_page_lock(void *word)
487 * In order to wait for pages to become available there must be
488 * waitqueues associated with pages. By using a hash table of
489 * waitqueues where the bucket discipline is to maintain all
490 * waiters on the same queue and wake all when any of the pages
491 * become available, and for the woken contexts to check to be
492 * sure the appropriate page became available, this saves space
493 * at a cost of "thundering herd" phenomena during rare hash
496 static wait_queue_head_t *page_waitqueue(struct page *page)
498 const struct zone *zone = page_zone(page);
500 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
503 static inline void wake_up_page(struct page *page, int bit)
505 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
508 void wait_on_page_bit(struct page *page, int bit_nr)
510 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
512 if (test_bit(bit_nr, &page->flags))
513 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
514 TASK_UNINTERRUPTIBLE);
516 EXPORT_SYMBOL(wait_on_page_bit);
519 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
520 * @page: Page defining the wait queue of interest
521 * @waiter: Waiter to add to the queue
523 * Add an arbitrary @waiter to the wait queue for the nominated @page.
525 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
527 wait_queue_head_t *q = page_waitqueue(page);
530 spin_lock_irqsave(&q->lock, flags);
531 __add_wait_queue(q, waiter);
532 spin_unlock_irqrestore(&q->lock, flags);
534 EXPORT_SYMBOL_GPL(add_page_wait_queue);
537 * unlock_page - unlock a locked page
540 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
541 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
542 * mechananism between PageLocked pages and PageWriteback pages is shared.
543 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
545 * The mb is necessary to enforce ordering between the clear_bit and the read
546 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
548 void unlock_page(struct page *page)
550 VM_BUG_ON(!PageLocked(page));
551 clear_bit_unlock(PG_locked, &page->flags);
552 smp_mb__after_clear_bit();
553 wake_up_page(page, PG_locked);
555 EXPORT_SYMBOL(unlock_page);
558 * end_page_writeback - end writeback against a page
561 void end_page_writeback(struct page *page)
563 if (TestClearPageReclaim(page))
564 rotate_reclaimable_page(page);
566 if (!test_clear_page_writeback(page))
569 smp_mb__after_clear_bit();
570 wake_up_page(page, PG_writeback);
572 EXPORT_SYMBOL(end_page_writeback);
575 * __lock_page - get a lock on the page, assuming we need to sleep to get it
576 * @page: the page to lock
578 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
579 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
580 * chances are that on the second loop, the block layer's plug list is empty,
581 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
583 void __lock_page(struct page *page)
585 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
587 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
588 TASK_UNINTERRUPTIBLE);
590 EXPORT_SYMBOL(__lock_page);
592 int __lock_page_killable(struct page *page)
594 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
596 return __wait_on_bit_lock(page_waitqueue(page), &wait,
597 sync_page_killable, TASK_KILLABLE);
599 EXPORT_SYMBOL_GPL(__lock_page_killable);
602 * __lock_page_nosync - get a lock on the page, without calling sync_page()
603 * @page: the page to lock
605 * Variant of lock_page that does not require the caller to hold a reference
606 * on the page's mapping.
608 void __lock_page_nosync(struct page *page)
610 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
611 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
612 TASK_UNINTERRUPTIBLE);
615 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
618 if (!(flags & FAULT_FLAG_ALLOW_RETRY)) {
622 up_read(&mm->mmap_sem);
623 wait_on_page_locked(page);
629 * find_get_page - find and get a page reference
630 * @mapping: the address_space to search
631 * @offset: the page index
633 * Is there a pagecache struct page at the given (mapping, offset) tuple?
634 * If yes, increment its refcount and return it; if no, return NULL.
636 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
644 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
646 page = radix_tree_deref_slot(pagep);
647 if (unlikely(!page || page == RADIX_TREE_RETRY))
650 if (!page_cache_get_speculative(page))
654 * Has the page moved?
655 * This is part of the lockless pagecache protocol. See
656 * include/linux/pagemap.h for details.
658 if (unlikely(page != *pagep)) {
659 page_cache_release(page);
667 EXPORT_SYMBOL(find_get_page);
670 * find_lock_page - locate, pin and lock a pagecache page
671 * @mapping: the address_space to search
672 * @offset: the page index
674 * Locates the desired pagecache page, locks it, increments its reference
675 * count and returns its address.
677 * Returns zero if the page was not present. find_lock_page() may sleep.
679 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
684 page = find_get_page(mapping, offset);
687 /* Has the page been truncated? */
688 if (unlikely(page->mapping != mapping)) {
690 page_cache_release(page);
693 VM_BUG_ON(page->index != offset);
697 EXPORT_SYMBOL(find_lock_page);
700 * find_or_create_page - locate or add a pagecache page
701 * @mapping: the page's address_space
702 * @index: the page's index into the mapping
703 * @gfp_mask: page allocation mode
705 * Locates a page in the pagecache. If the page is not present, a new page
706 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
707 * LRU list. The returned page is locked and has its reference count
710 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
713 * find_or_create_page() returns the desired page's address, or zero on
716 struct page *find_or_create_page(struct address_space *mapping,
717 pgoff_t index, gfp_t gfp_mask)
722 page = find_lock_page(mapping, index);
724 page = __page_cache_alloc(gfp_mask);
728 * We want a regular kernel memory (not highmem or DMA etc)
729 * allocation for the radix tree nodes, but we need to honour
730 * the context-specific requirements the caller has asked for.
731 * GFP_RECLAIM_MASK collects those requirements.
733 err = add_to_page_cache_lru(page, mapping, index,
734 (gfp_mask & GFP_RECLAIM_MASK));
736 page_cache_release(page);
744 EXPORT_SYMBOL(find_or_create_page);
747 * find_get_pages - gang pagecache lookup
748 * @mapping: The address_space to search
749 * @start: The starting page index
750 * @nr_pages: The maximum number of pages
751 * @pages: Where the resulting pages are placed
753 * find_get_pages() will search for and return a group of up to
754 * @nr_pages pages in the mapping. The pages are placed at @pages.
755 * find_get_pages() takes a reference against the returned pages.
757 * The search returns a group of mapping-contiguous pages with ascending
758 * indexes. There may be holes in the indices due to not-present pages.
760 * find_get_pages() returns the number of pages which were found.
762 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
763 unsigned int nr_pages, struct page **pages)
767 unsigned int nr_found;
771 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
772 (void ***)pages, start, nr_pages);
774 for (i = 0; i < nr_found; i++) {
777 page = radix_tree_deref_slot((void **)pages[i]);
781 * this can only trigger if nr_found == 1, making livelock
784 if (unlikely(page == RADIX_TREE_RETRY))
787 if (!page_cache_get_speculative(page))
790 /* Has the page moved? */
791 if (unlikely(page != *((void **)pages[i]))) {
792 page_cache_release(page);
804 * find_get_pages_contig - gang contiguous pagecache lookup
805 * @mapping: The address_space to search
806 * @index: The starting page index
807 * @nr_pages: The maximum number of pages
808 * @pages: Where the resulting pages are placed
810 * find_get_pages_contig() works exactly like find_get_pages(), except
811 * that the returned number of pages are guaranteed to be contiguous.
813 * find_get_pages_contig() returns the number of pages which were found.
815 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
816 unsigned int nr_pages, struct page **pages)
820 unsigned int nr_found;
824 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
825 (void ***)pages, index, nr_pages);
827 for (i = 0; i < nr_found; i++) {
830 page = radix_tree_deref_slot((void **)pages[i]);
834 * this can only trigger if nr_found == 1, making livelock
837 if (unlikely(page == RADIX_TREE_RETRY))
840 if (page->mapping == NULL || page->index != index)
843 if (!page_cache_get_speculative(page))
846 /* Has the page moved? */
847 if (unlikely(page != *((void **)pages[i]))) {
848 page_cache_release(page);
859 EXPORT_SYMBOL(find_get_pages_contig);
862 * find_get_pages_tag - find and return pages that match @tag
863 * @mapping: the address_space to search
864 * @index: the starting page index
865 * @tag: the tag index
866 * @nr_pages: the maximum number of pages
867 * @pages: where the resulting pages are placed
869 * Like find_get_pages, except we only return pages which are tagged with
870 * @tag. We update @index to index the next page for the traversal.
872 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
873 int tag, unsigned int nr_pages, struct page **pages)
877 unsigned int nr_found;
881 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
882 (void ***)pages, *index, nr_pages, tag);
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_cache_get_speculative(page))
900 /* Has the page moved? */
901 if (unlikely(page != *((void **)pages[i]))) {
902 page_cache_release(page);
912 *index = pages[ret - 1]->index + 1;
916 EXPORT_SYMBOL(find_get_pages_tag);
919 * grab_cache_page_nowait - returns locked page at given index in given cache
920 * @mapping: target address_space
921 * @index: the page index
923 * Same as grab_cache_page(), but do not wait if the page is unavailable.
924 * This is intended for speculative data generators, where the data can
925 * be regenerated if the page couldn't be grabbed. This routine should
926 * be safe to call while holding the lock for another page.
928 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
929 * and deadlock against the caller's locked page.
932 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
934 struct page *page = find_get_page(mapping, index);
937 if (trylock_page(page))
939 page_cache_release(page);
942 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
943 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
944 page_cache_release(page);
949 EXPORT_SYMBOL(grab_cache_page_nowait);
952 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
953 * a _large_ part of the i/o request. Imagine the worst scenario:
955 * ---R__________________________________________B__________
956 * ^ reading here ^ bad block(assume 4k)
958 * read(R) => miss => readahead(R...B) => media error => frustrating retries
959 * => failing the whole request => read(R) => read(R+1) =>
960 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
961 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
962 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
964 * It is going insane. Fix it by quickly scaling down the readahead size.
966 static void shrink_readahead_size_eio(struct file *filp,
967 struct file_ra_state *ra)
973 * do_generic_file_read - generic file read routine
974 * @filp: the file to read
975 * @ppos: current file position
976 * @desc: read_descriptor
977 * @actor: read method
979 * This is a generic file read routine, and uses the
980 * mapping->a_ops->readpage() function for the actual low-level stuff.
982 * This is really ugly. But the goto's actually try to clarify some
983 * of the logic when it comes to error handling etc.
985 static void do_generic_file_read(struct file *filp, loff_t *ppos,
986 read_descriptor_t *desc, read_actor_t actor)
988 struct address_space *mapping = filp->f_mapping;
989 struct inode *inode = mapping->host;
990 struct file_ra_state *ra = &filp->f_ra;
994 unsigned long offset; /* offset into pagecache page */
995 unsigned int prev_offset;
998 index = *ppos >> PAGE_CACHE_SHIFT;
999 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1000 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1001 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1002 offset = *ppos & ~PAGE_CACHE_MASK;
1008 unsigned long nr, ret;
1012 page = find_get_page(mapping, index);
1014 page_cache_sync_readahead(mapping,
1016 index, last_index - index);
1017 page = find_get_page(mapping, index);
1018 if (unlikely(page == NULL))
1019 goto no_cached_page;
1021 if (PageReadahead(page)) {
1022 page_cache_async_readahead(mapping,
1024 index, last_index - index);
1026 if (!PageUptodate(page)) {
1027 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1028 !mapping->a_ops->is_partially_uptodate)
1029 goto page_not_up_to_date;
1030 if (!trylock_page(page))
1031 goto page_not_up_to_date;
1032 /* Did it get truncated before we got the lock? */
1034 goto page_not_up_to_date_locked;
1035 if (!mapping->a_ops->is_partially_uptodate(page,
1037 goto page_not_up_to_date_locked;
1042 * i_size must be checked after we know the page is Uptodate.
1044 * Checking i_size after the check allows us to calculate
1045 * the correct value for "nr", which means the zero-filled
1046 * part of the page is not copied back to userspace (unless
1047 * another truncate extends the file - this is desired though).
1050 isize = i_size_read(inode);
1051 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1052 if (unlikely(!isize || index > end_index)) {
1053 page_cache_release(page);
1057 /* nr is the maximum number of bytes to copy from this page */
1058 nr = PAGE_CACHE_SIZE;
1059 if (index == end_index) {
1060 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1062 page_cache_release(page);
1068 /* If users can be writing to this page using arbitrary
1069 * virtual addresses, take care about potential aliasing
1070 * before reading the page on the kernel side.
1072 if (mapping_writably_mapped(mapping))
1073 flush_dcache_page(page);
1076 * When a sequential read accesses a page several times,
1077 * only mark it as accessed the first time.
1079 if (prev_index != index || offset != prev_offset)
1080 mark_page_accessed(page);
1084 * Ok, we have the page, and it's up-to-date, so
1085 * now we can copy it to user space...
1087 * The actor routine returns how many bytes were actually used..
1088 * NOTE! This may not be the same as how much of a user buffer
1089 * we filled up (we may be padding etc), so we can only update
1090 * "pos" here (the actor routine has to update the user buffer
1091 * pointers and the remaining count).
1093 ret = actor(desc, page, offset, nr);
1095 index += offset >> PAGE_CACHE_SHIFT;
1096 offset &= ~PAGE_CACHE_MASK;
1097 prev_offset = offset;
1099 page_cache_release(page);
1100 if (ret == nr && desc->count)
1104 page_not_up_to_date:
1105 /* Get exclusive access to the page ... */
1106 error = lock_page_killable(page);
1107 if (unlikely(error))
1108 goto readpage_error;
1110 page_not_up_to_date_locked:
1111 /* Did it get truncated before we got the lock? */
1112 if (!page->mapping) {
1114 page_cache_release(page);
1118 /* Did somebody else fill it already? */
1119 if (PageUptodate(page)) {
1126 * A previous I/O error may have been due to temporary
1127 * failures, eg. multipath errors.
1128 * PG_error will be set again if readpage fails.
1130 ClearPageError(page);
1131 /* Start the actual read. The read will unlock the page. */
1132 error = mapping->a_ops->readpage(filp, page);
1134 if (unlikely(error)) {
1135 if (error == AOP_TRUNCATED_PAGE) {
1136 page_cache_release(page);
1139 goto readpage_error;
1142 if (!PageUptodate(page)) {
1143 error = lock_page_killable(page);
1144 if (unlikely(error))
1145 goto readpage_error;
1146 if (!PageUptodate(page)) {
1147 if (page->mapping == NULL) {
1149 * invalidate_mapping_pages got it
1152 page_cache_release(page);
1156 shrink_readahead_size_eio(filp, ra);
1158 goto readpage_error;
1166 /* UHHUH! A synchronous read error occurred. Report it */
1167 desc->error = error;
1168 page_cache_release(page);
1173 * Ok, it wasn't cached, so we need to create a new
1176 page = page_cache_alloc_cold(mapping);
1178 desc->error = -ENOMEM;
1181 error = add_to_page_cache_lru(page, mapping,
1184 page_cache_release(page);
1185 if (error == -EEXIST)
1187 desc->error = error;
1194 ra->prev_pos = prev_index;
1195 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1196 ra->prev_pos |= prev_offset;
1198 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1199 file_accessed(filp);
1202 int file_read_actor(read_descriptor_t *desc, struct page *page,
1203 unsigned long offset, unsigned long size)
1206 unsigned long left, count = desc->count;
1212 * Faults on the destination of a read are common, so do it before
1215 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1216 kaddr = kmap_atomic(page, KM_USER0);
1217 left = __copy_to_user_inatomic(desc->arg.buf,
1218 kaddr + offset, size);
1219 kunmap_atomic(kaddr, KM_USER0);
1224 /* Do it the slow way */
1226 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1231 desc->error = -EFAULT;
1234 desc->count = count - size;
1235 desc->written += size;
1236 desc->arg.buf += size;
1241 * Performs necessary checks before doing a write
1242 * @iov: io vector request
1243 * @nr_segs: number of segments in the iovec
1244 * @count: number of bytes to write
1245 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1247 * Adjust number of segments and amount of bytes to write (nr_segs should be
1248 * properly initialized first). Returns appropriate error code that caller
1249 * should return or zero in case that write should be allowed.
1251 int generic_segment_checks(const struct iovec *iov,
1252 unsigned long *nr_segs, size_t *count, int access_flags)
1256 for (seg = 0; seg < *nr_segs; seg++) {
1257 const struct iovec *iv = &iov[seg];
1260 * If any segment has a negative length, or the cumulative
1261 * length ever wraps negative then return -EINVAL.
1264 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1266 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1271 cnt -= iv->iov_len; /* This segment is no good */
1277 EXPORT_SYMBOL(generic_segment_checks);
1280 * generic_file_aio_read - generic filesystem read routine
1281 * @iocb: kernel I/O control block
1282 * @iov: io vector request
1283 * @nr_segs: number of segments in the iovec
1284 * @pos: current file position
1286 * This is the "read()" routine for all filesystems
1287 * that can use the page cache directly.
1290 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1291 unsigned long nr_segs, loff_t pos)
1293 struct file *filp = iocb->ki_filp;
1295 unsigned long seg = 0;
1297 loff_t *ppos = &iocb->ki_pos;
1300 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1304 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1305 if (filp->f_flags & O_DIRECT) {
1307 struct address_space *mapping;
1308 struct inode *inode;
1310 mapping = filp->f_mapping;
1311 inode = mapping->host;
1313 goto out; /* skip atime */
1314 size = i_size_read(inode);
1316 retval = filemap_write_and_wait_range(mapping, pos,
1317 pos + iov_length(iov, nr_segs) - 1);
1319 retval = mapping->a_ops->direct_IO(READ, iocb,
1323 *ppos = pos + retval;
1328 * Btrfs can have a short DIO read if we encounter
1329 * compressed extents, so if there was an error, or if
1330 * we've already read everything we wanted to, or if
1331 * there was a short read because we hit EOF, go ahead
1332 * and return. Otherwise fallthrough to buffered io for
1333 * the rest of the read.
1335 if (retval < 0 || !count || *ppos >= size) {
1336 file_accessed(filp);
1343 for (seg = 0; seg < nr_segs; seg++) {
1344 read_descriptor_t desc;
1348 * If we did a short DIO read we need to skip the section of the
1349 * iov that we've already read data into.
1352 if (count > iov[seg].iov_len) {
1353 count -= iov[seg].iov_len;
1361 desc.arg.buf = iov[seg].iov_base + offset;
1362 desc.count = iov[seg].iov_len - offset;
1363 if (desc.count == 0)
1366 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1367 retval += desc.written;
1369 retval = retval ?: desc.error;
1378 EXPORT_SYMBOL(generic_file_aio_read);
1381 do_readahead(struct address_space *mapping, struct file *filp,
1382 pgoff_t index, unsigned long nr)
1384 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1387 force_page_cache_readahead(mapping, filp, index, nr);
1391 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1399 if (file->f_mode & FMODE_READ) {
1400 struct address_space *mapping = file->f_mapping;
1401 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1402 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1403 unsigned long len = end - start + 1;
1404 ret = do_readahead(mapping, file, start, len);
1410 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1411 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1413 return SYSC_readahead((int) fd, offset, (size_t) count);
1415 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1420 * page_cache_read - adds requested page to the page cache if not already there
1421 * @file: file to read
1422 * @offset: page index
1424 * This adds the requested page to the page cache if it isn't already there,
1425 * and schedules an I/O to read in its contents from disk.
1427 static int page_cache_read(struct file *file, pgoff_t offset)
1429 struct address_space *mapping = file->f_mapping;
1434 page = page_cache_alloc_cold(mapping);
1438 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1440 ret = mapping->a_ops->readpage(file, page);
1441 else if (ret == -EEXIST)
1442 ret = 0; /* losing race to add is OK */
1444 page_cache_release(page);
1446 } while (ret == AOP_TRUNCATED_PAGE);
1451 #define MMAP_LOTSAMISS (100)
1454 * Synchronous readahead happens when we don't even find
1455 * a page in the page cache at all.
1457 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1458 struct file_ra_state *ra,
1462 unsigned long ra_pages;
1463 struct address_space *mapping = file->f_mapping;
1465 /* If we don't want any read-ahead, don't bother */
1466 if (VM_RandomReadHint(vma))
1469 if (VM_SequentialReadHint(vma) ||
1470 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1471 page_cache_sync_readahead(mapping, ra, file, offset,
1476 if (ra->mmap_miss < INT_MAX)
1480 * Do we miss much more than hit in this file? If so,
1481 * stop bothering with read-ahead. It will only hurt.
1483 if (ra->mmap_miss > MMAP_LOTSAMISS)
1489 ra_pages = max_sane_readahead(ra->ra_pages);
1491 ra->start = max_t(long, 0, offset - ra_pages/2);
1492 ra->size = ra_pages;
1494 ra_submit(ra, mapping, file);
1499 * Asynchronous readahead happens when we find the page and PG_readahead,
1500 * so we want to possibly extend the readahead further..
1502 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1503 struct file_ra_state *ra,
1508 struct address_space *mapping = file->f_mapping;
1510 /* If we don't want any read-ahead, don't bother */
1511 if (VM_RandomReadHint(vma))
1513 if (ra->mmap_miss > 0)
1515 if (PageReadahead(page))
1516 page_cache_async_readahead(mapping, ra, file,
1517 page, offset, ra->ra_pages);
1521 * filemap_fault - read in file data for page fault handling
1522 * @vma: vma in which the fault was taken
1523 * @vmf: struct vm_fault containing details of the fault
1525 * filemap_fault() is invoked via the vma operations vector for a
1526 * mapped memory region to read in file data during a page fault.
1528 * The goto's are kind of ugly, but this streamlines the normal case of having
1529 * it in the page cache, and handles the special cases reasonably without
1530 * having a lot of duplicated code.
1532 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1535 struct file *file = vma->vm_file;
1536 struct address_space *mapping = file->f_mapping;
1537 struct file_ra_state *ra = &file->f_ra;
1538 struct inode *inode = mapping->host;
1539 pgoff_t offset = vmf->pgoff;
1544 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1546 return VM_FAULT_SIGBUS;
1549 * Do we have something in the page cache already?
1551 page = find_get_page(mapping, offset);
1554 * We found the page, so try async readahead before
1555 * waiting for the lock.
1557 do_async_mmap_readahead(vma, ra, file, page, offset);
1559 /* No page in the page cache at all */
1560 do_sync_mmap_readahead(vma, ra, file, offset);
1561 count_vm_event(PGMAJFAULT);
1562 ret = VM_FAULT_MAJOR;
1564 page = find_get_page(mapping, offset);
1566 goto no_cached_page;
1569 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1570 page_cache_release(page);
1571 return ret | VM_FAULT_RETRY;
1574 /* Did it get truncated? */
1575 if (unlikely(page->mapping != mapping)) {
1580 VM_BUG_ON(page->index != offset);
1583 * We have a locked page in the page cache, now we need to check
1584 * that it's up-to-date. If not, it is going to be due to an error.
1586 if (unlikely(!PageUptodate(page)))
1587 goto page_not_uptodate;
1590 * Found the page and have a reference on it.
1591 * We must recheck i_size under page lock.
1593 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1594 if (unlikely(offset >= size)) {
1596 page_cache_release(page);
1597 return VM_FAULT_SIGBUS;
1600 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1602 return ret | VM_FAULT_LOCKED;
1606 * We're only likely to ever get here if MADV_RANDOM is in
1609 error = page_cache_read(file, offset);
1612 * The page we want has now been added to the page cache.
1613 * In the unlikely event that someone removed it in the
1614 * meantime, we'll just come back here and read it again.
1620 * An error return from page_cache_read can result if the
1621 * system is low on memory, or a problem occurs while trying
1624 if (error == -ENOMEM)
1625 return VM_FAULT_OOM;
1626 return VM_FAULT_SIGBUS;
1630 * Umm, take care of errors if the page isn't up-to-date.
1631 * Try to re-read it _once_. We do this synchronously,
1632 * because there really aren't any performance issues here
1633 * and we need to check for errors.
1635 ClearPageError(page);
1636 error = mapping->a_ops->readpage(file, page);
1638 wait_on_page_locked(page);
1639 if (!PageUptodate(page))
1642 page_cache_release(page);
1644 if (!error || error == AOP_TRUNCATED_PAGE)
1647 /* Things didn't work out. Return zero to tell the mm layer so. */
1648 shrink_readahead_size_eio(file, ra);
1649 return VM_FAULT_SIGBUS;
1651 EXPORT_SYMBOL(filemap_fault);
1653 const struct vm_operations_struct generic_file_vm_ops = {
1654 .fault = filemap_fault,
1657 /* This is used for a general mmap of a disk file */
1659 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1661 struct address_space *mapping = file->f_mapping;
1663 if (!mapping->a_ops->readpage)
1665 file_accessed(file);
1666 vma->vm_ops = &generic_file_vm_ops;
1667 vma->vm_flags |= VM_CAN_NONLINEAR;
1672 * This is for filesystems which do not implement ->writepage.
1674 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1676 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1678 return generic_file_mmap(file, vma);
1681 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1685 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1689 #endif /* CONFIG_MMU */
1691 EXPORT_SYMBOL(generic_file_mmap);
1692 EXPORT_SYMBOL(generic_file_readonly_mmap);
1694 static struct page *__read_cache_page(struct address_space *mapping,
1696 int (*filler)(void *,struct page*),
1703 page = find_get_page(mapping, index);
1705 page = __page_cache_alloc(gfp | __GFP_COLD);
1707 return ERR_PTR(-ENOMEM);
1708 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1709 if (unlikely(err)) {
1710 page_cache_release(page);
1713 /* Presumably ENOMEM for radix tree node */
1714 return ERR_PTR(err);
1716 err = filler(data, page);
1718 page_cache_release(page);
1719 page = ERR_PTR(err);
1725 static struct page *do_read_cache_page(struct address_space *mapping,
1727 int (*filler)(void *,struct page*),
1736 page = __read_cache_page(mapping, index, filler, data, gfp);
1739 if (PageUptodate(page))
1743 if (!page->mapping) {
1745 page_cache_release(page);
1748 if (PageUptodate(page)) {
1752 err = filler(data, page);
1754 page_cache_release(page);
1755 return ERR_PTR(err);
1758 mark_page_accessed(page);
1763 * read_cache_page_async - read into page cache, fill it if needed
1764 * @mapping: the page's address_space
1765 * @index: the page index
1766 * @filler: function to perform the read
1767 * @data: destination for read data
1769 * Same as read_cache_page, but don't wait for page to become unlocked
1770 * after submitting it to the filler.
1772 * Read into the page cache. If a page already exists, and PageUptodate() is
1773 * not set, try to fill the page but don't wait for it to become unlocked.
1775 * If the page does not get brought uptodate, return -EIO.
1777 struct page *read_cache_page_async(struct address_space *mapping,
1779 int (*filler)(void *,struct page*),
1782 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1784 EXPORT_SYMBOL(read_cache_page_async);
1786 static struct page *wait_on_page_read(struct page *page)
1788 if (!IS_ERR(page)) {
1789 wait_on_page_locked(page);
1790 if (!PageUptodate(page)) {
1791 page_cache_release(page);
1792 page = ERR_PTR(-EIO);
1799 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1800 * @mapping: the page's address_space
1801 * @index: the page index
1802 * @gfp: the page allocator flags to use if allocating
1804 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1805 * any new page allocations done using the specified allocation flags. Note
1806 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1807 * expect to do this atomically or anything like that - but you can pass in
1808 * other page requirements.
1810 * If the page does not get brought uptodate, return -EIO.
1812 struct page *read_cache_page_gfp(struct address_space *mapping,
1816 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1818 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1820 EXPORT_SYMBOL(read_cache_page_gfp);
1823 * read_cache_page - read into page cache, fill it if needed
1824 * @mapping: the page's address_space
1825 * @index: the page index
1826 * @filler: function to perform the read
1827 * @data: destination for read data
1829 * Read into the page cache. If a page already exists, and PageUptodate() is
1830 * not set, try to fill the page then wait for it to become unlocked.
1832 * If the page does not get brought uptodate, return -EIO.
1834 struct page *read_cache_page(struct address_space *mapping,
1836 int (*filler)(void *,struct page*),
1839 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1841 EXPORT_SYMBOL(read_cache_page);
1844 * The logic we want is
1846 * if suid or (sgid and xgrp)
1849 int should_remove_suid(struct dentry *dentry)
1851 mode_t mode = dentry->d_inode->i_mode;
1854 /* suid always must be killed */
1855 if (unlikely(mode & S_ISUID))
1856 kill = ATTR_KILL_SUID;
1859 * sgid without any exec bits is just a mandatory locking mark; leave
1860 * it alone. If some exec bits are set, it's a real sgid; kill it.
1862 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1863 kill |= ATTR_KILL_SGID;
1865 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1870 EXPORT_SYMBOL(should_remove_suid);
1872 static int __remove_suid(struct dentry *dentry, int kill)
1874 struct iattr newattrs;
1876 newattrs.ia_valid = ATTR_FORCE | kill;
1877 return notify_change(dentry, &newattrs);
1880 int file_remove_suid(struct file *file)
1882 struct dentry *dentry = file->f_path.dentry;
1883 int killsuid = should_remove_suid(dentry);
1884 int killpriv = security_inode_need_killpriv(dentry);
1890 error = security_inode_killpriv(dentry);
1891 if (!error && killsuid)
1892 error = __remove_suid(dentry, killsuid);
1896 EXPORT_SYMBOL(file_remove_suid);
1898 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1899 const struct iovec *iov, size_t base, size_t bytes)
1901 size_t copied = 0, left = 0;
1904 char __user *buf = iov->iov_base + base;
1905 int copy = min(bytes, iov->iov_len - base);
1908 left = __copy_from_user_inatomic(vaddr, buf, copy);
1917 return copied - left;
1921 * Copy as much as we can into the page and return the number of bytes which
1922 * were successfully copied. If a fault is encountered then return the number of
1923 * bytes which were copied.
1925 size_t iov_iter_copy_from_user_atomic(struct page *page,
1926 struct iov_iter *i, unsigned long offset, size_t bytes)
1931 BUG_ON(!in_atomic());
1932 kaddr = kmap_atomic(page, KM_USER0);
1933 if (likely(i->nr_segs == 1)) {
1935 char __user *buf = i->iov->iov_base + i->iov_offset;
1936 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1937 copied = bytes - left;
1939 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1940 i->iov, i->iov_offset, bytes);
1942 kunmap_atomic(kaddr, KM_USER0);
1946 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1949 * This has the same sideeffects and return value as
1950 * iov_iter_copy_from_user_atomic().
1951 * The difference is that it attempts to resolve faults.
1952 * Page must not be locked.
1954 size_t iov_iter_copy_from_user(struct page *page,
1955 struct iov_iter *i, unsigned long offset, size_t bytes)
1961 if (likely(i->nr_segs == 1)) {
1963 char __user *buf = i->iov->iov_base + i->iov_offset;
1964 left = __copy_from_user(kaddr + offset, buf, bytes);
1965 copied = bytes - left;
1967 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1968 i->iov, i->iov_offset, bytes);
1973 EXPORT_SYMBOL(iov_iter_copy_from_user);
1975 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1977 BUG_ON(i->count < bytes);
1979 if (likely(i->nr_segs == 1)) {
1980 i->iov_offset += bytes;
1983 const struct iovec *iov = i->iov;
1984 size_t base = i->iov_offset;
1987 * The !iov->iov_len check ensures we skip over unlikely
1988 * zero-length segments (without overruning the iovec).
1990 while (bytes || unlikely(i->count && !iov->iov_len)) {
1993 copy = min(bytes, iov->iov_len - base);
1994 BUG_ON(!i->count || i->count < copy);
1998 if (iov->iov_len == base) {
2004 i->iov_offset = base;
2007 EXPORT_SYMBOL(iov_iter_advance);
2010 * Fault in the first iovec of the given iov_iter, to a maximum length
2011 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2012 * accessed (ie. because it is an invalid address).
2014 * writev-intensive code may want this to prefault several iovecs -- that
2015 * would be possible (callers must not rely on the fact that _only_ the
2016 * first iovec will be faulted with the current implementation).
2018 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2020 char __user *buf = i->iov->iov_base + i->iov_offset;
2021 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2022 return fault_in_pages_readable(buf, bytes);
2024 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2027 * Return the count of just the current iov_iter segment.
2029 size_t iov_iter_single_seg_count(struct iov_iter *i)
2031 const struct iovec *iov = i->iov;
2032 if (i->nr_segs == 1)
2035 return min(i->count, iov->iov_len - i->iov_offset);
2037 EXPORT_SYMBOL(iov_iter_single_seg_count);
2040 * Performs necessary checks before doing a write
2042 * Can adjust writing position or amount of bytes to write.
2043 * Returns appropriate error code that caller should return or
2044 * zero in case that write should be allowed.
2046 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2048 struct inode *inode = file->f_mapping->host;
2049 unsigned long limit = rlimit(RLIMIT_FSIZE);
2051 if (unlikely(*pos < 0))
2055 /* FIXME: this is for backwards compatibility with 2.4 */
2056 if (file->f_flags & O_APPEND)
2057 *pos = i_size_read(inode);
2059 if (limit != RLIM_INFINITY) {
2060 if (*pos >= limit) {
2061 send_sig(SIGXFSZ, current, 0);
2064 if (*count > limit - (typeof(limit))*pos) {
2065 *count = limit - (typeof(limit))*pos;
2073 if (unlikely(*pos + *count > MAX_NON_LFS &&
2074 !(file->f_flags & O_LARGEFILE))) {
2075 if (*pos >= MAX_NON_LFS) {
2078 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2079 *count = MAX_NON_LFS - (unsigned long)*pos;
2084 * Are we about to exceed the fs block limit ?
2086 * If we have written data it becomes a short write. If we have
2087 * exceeded without writing data we send a signal and return EFBIG.
2088 * Linus frestrict idea will clean these up nicely..
2090 if (likely(!isblk)) {
2091 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2092 if (*count || *pos > inode->i_sb->s_maxbytes) {
2095 /* zero-length writes at ->s_maxbytes are OK */
2098 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2099 *count = inode->i_sb->s_maxbytes - *pos;
2103 if (bdev_read_only(I_BDEV(inode)))
2105 isize = i_size_read(inode);
2106 if (*pos >= isize) {
2107 if (*count || *pos > isize)
2111 if (*pos + *count > isize)
2112 *count = isize - *pos;
2119 EXPORT_SYMBOL(generic_write_checks);
2121 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2122 loff_t pos, unsigned len, unsigned flags,
2123 struct page **pagep, void **fsdata)
2125 const struct address_space_operations *aops = mapping->a_ops;
2127 return aops->write_begin(file, mapping, pos, len, flags,
2130 EXPORT_SYMBOL(pagecache_write_begin);
2132 int pagecache_write_end(struct file *file, struct address_space *mapping,
2133 loff_t pos, unsigned len, unsigned copied,
2134 struct page *page, void *fsdata)
2136 const struct address_space_operations *aops = mapping->a_ops;
2138 mark_page_accessed(page);
2139 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2141 EXPORT_SYMBOL(pagecache_write_end);
2144 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2145 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2146 size_t count, size_t ocount)
2148 struct file *file = iocb->ki_filp;
2149 struct address_space *mapping = file->f_mapping;
2150 struct inode *inode = mapping->host;
2155 if (count != ocount)
2156 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2158 write_len = iov_length(iov, *nr_segs);
2159 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2161 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2166 * After a write we want buffered reads to be sure to go to disk to get
2167 * the new data. We invalidate clean cached page from the region we're
2168 * about to write. We do this *before* the write so that we can return
2169 * without clobbering -EIOCBQUEUED from ->direct_IO().
2171 if (mapping->nrpages) {
2172 written = invalidate_inode_pages2_range(mapping,
2173 pos >> PAGE_CACHE_SHIFT, end);
2175 * If a page can not be invalidated, return 0 to fall back
2176 * to buffered write.
2179 if (written == -EBUSY)
2185 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2188 * Finally, try again to invalidate clean pages which might have been
2189 * cached by non-direct readahead, or faulted in by get_user_pages()
2190 * if the source of the write was an mmap'ed region of the file
2191 * we're writing. Either one is a pretty crazy thing to do,
2192 * so we don't support it 100%. If this invalidation
2193 * fails, tough, the write still worked...
2195 if (mapping->nrpages) {
2196 invalidate_inode_pages2_range(mapping,
2197 pos >> PAGE_CACHE_SHIFT, end);
2202 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2203 i_size_write(inode, pos);
2204 mark_inode_dirty(inode);
2211 EXPORT_SYMBOL(generic_file_direct_write);
2214 * Find or create a page at the given pagecache position. Return the locked
2215 * page. This function is specifically for buffered writes.
2217 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2218 pgoff_t index, unsigned flags)
2222 gfp_t gfp_notmask = 0;
2223 if (flags & AOP_FLAG_NOFS)
2224 gfp_notmask = __GFP_FS;
2226 page = find_lock_page(mapping, index);
2230 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2233 status = add_to_page_cache_lru(page, mapping, index,
2234 GFP_KERNEL & ~gfp_notmask);
2235 if (unlikely(status)) {
2236 page_cache_release(page);
2237 if (status == -EEXIST)
2243 EXPORT_SYMBOL(grab_cache_page_write_begin);
2245 static ssize_t generic_perform_write(struct file *file,
2246 struct iov_iter *i, loff_t pos)
2248 struct address_space *mapping = file->f_mapping;
2249 const struct address_space_operations *a_ops = mapping->a_ops;
2251 ssize_t written = 0;
2252 unsigned int flags = 0;
2255 * Copies from kernel address space cannot fail (NFSD is a big user).
2257 if (segment_eq(get_fs(), KERNEL_DS))
2258 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2262 unsigned long offset; /* Offset into pagecache page */
2263 unsigned long bytes; /* Bytes to write to page */
2264 size_t copied; /* Bytes copied from user */
2267 offset = (pos & (PAGE_CACHE_SIZE - 1));
2268 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2274 * Bring in the user page that we will copy from _first_.
2275 * Otherwise there's a nasty deadlock on copying from the
2276 * same page as we're writing to, without it being marked
2279 * Not only is this an optimisation, but it is also required
2280 * to check that the address is actually valid, when atomic
2281 * usercopies are used, below.
2283 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2288 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2290 if (unlikely(status))
2293 if (mapping_writably_mapped(mapping))
2294 flush_dcache_page(page);
2296 pagefault_disable();
2297 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2299 flush_dcache_page(page);
2301 mark_page_accessed(page);
2302 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2304 if (unlikely(status < 0))
2310 iov_iter_advance(i, copied);
2311 if (unlikely(copied == 0)) {
2313 * If we were unable to copy any data at all, we must
2314 * fall back to a single segment length write.
2316 * If we didn't fallback here, we could livelock
2317 * because not all segments in the iov can be copied at
2318 * once without a pagefault.
2320 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2321 iov_iter_single_seg_count(i));
2327 balance_dirty_pages_ratelimited(mapping);
2329 } while (iov_iter_count(i));
2331 return written ? written : status;
2335 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2336 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2337 size_t count, ssize_t written)
2339 struct file *file = iocb->ki_filp;
2343 iov_iter_init(&i, iov, nr_segs, count, written);
2344 status = generic_perform_write(file, &i, pos);
2346 if (likely(status >= 0)) {
2348 *ppos = pos + status;
2351 return written ? written : status;
2353 EXPORT_SYMBOL(generic_file_buffered_write);
2356 * __generic_file_aio_write - write data to a file
2357 * @iocb: IO state structure (file, offset, etc.)
2358 * @iov: vector with data to write
2359 * @nr_segs: number of segments in the vector
2360 * @ppos: position where to write
2362 * This function does all the work needed for actually writing data to a
2363 * file. It does all basic checks, removes SUID from the file, updates
2364 * modification times and calls proper subroutines depending on whether we
2365 * do direct IO or a standard buffered write.
2367 * It expects i_mutex to be grabbed unless we work on a block device or similar
2368 * object which does not need locking at all.
2370 * This function does *not* take care of syncing data in case of O_SYNC write.
2371 * A caller has to handle it. This is mainly due to the fact that we want to
2372 * avoid syncing under i_mutex.
2374 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2375 unsigned long nr_segs, loff_t *ppos)
2377 struct file *file = iocb->ki_filp;
2378 struct address_space * mapping = file->f_mapping;
2379 size_t ocount; /* original count */
2380 size_t count; /* after file limit checks */
2381 struct inode *inode = mapping->host;
2387 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2394 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2396 /* We can write back this queue in page reclaim */
2397 current->backing_dev_info = mapping->backing_dev_info;
2400 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2407 err = file_remove_suid(file);
2411 file_update_time(file);
2413 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2414 if (unlikely(file->f_flags & O_DIRECT)) {
2416 ssize_t written_buffered;
2418 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2419 ppos, count, ocount);
2420 if (written < 0 || written == count)
2423 * direct-io write to a hole: fall through to buffered I/O
2424 * for completing the rest of the request.
2428 written_buffered = generic_file_buffered_write(iocb, iov,
2429 nr_segs, pos, ppos, count,
2432 * If generic_file_buffered_write() retuned a synchronous error
2433 * then we want to return the number of bytes which were
2434 * direct-written, or the error code if that was zero. Note
2435 * that this differs from normal direct-io semantics, which
2436 * will return -EFOO even if some bytes were written.
2438 if (written_buffered < 0) {
2439 err = written_buffered;
2444 * We need to ensure that the page cache pages are written to
2445 * disk and invalidated to preserve the expected O_DIRECT
2448 endbyte = pos + written_buffered - written - 1;
2449 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2451 written = written_buffered;
2452 invalidate_mapping_pages(mapping,
2453 pos >> PAGE_CACHE_SHIFT,
2454 endbyte >> PAGE_CACHE_SHIFT);
2457 * We don't know how much we wrote, so just return
2458 * the number of bytes which were direct-written
2462 written = generic_file_buffered_write(iocb, iov, nr_segs,
2463 pos, ppos, count, written);
2466 current->backing_dev_info = NULL;
2467 return written ? written : err;
2469 EXPORT_SYMBOL(__generic_file_aio_write);
2472 * generic_file_aio_write - write data to a file
2473 * @iocb: IO state structure
2474 * @iov: vector with data to write
2475 * @nr_segs: number of segments in the vector
2476 * @pos: position in file where to write
2478 * This is a wrapper around __generic_file_aio_write() to be used by most
2479 * filesystems. It takes care of syncing the file in case of O_SYNC file
2480 * and acquires i_mutex as needed.
2482 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2483 unsigned long nr_segs, loff_t pos)
2485 struct file *file = iocb->ki_filp;
2486 struct inode *inode = file->f_mapping->host;
2489 BUG_ON(iocb->ki_pos != pos);
2491 mutex_lock(&inode->i_mutex);
2492 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2493 mutex_unlock(&inode->i_mutex);
2495 if (ret > 0 || ret == -EIOCBQUEUED) {
2498 err = generic_write_sync(file, pos, ret);
2499 if (err < 0 && ret > 0)
2504 EXPORT_SYMBOL(generic_file_aio_write);
2507 * try_to_release_page() - release old fs-specific metadata on a page
2509 * @page: the page which the kernel is trying to free
2510 * @gfp_mask: memory allocation flags (and I/O mode)
2512 * The address_space is to try to release any data against the page
2513 * (presumably at page->private). If the release was successful, return `1'.
2514 * Otherwise return zero.
2516 * This may also be called if PG_fscache is set on a page, indicating that the
2517 * page is known to the local caching routines.
2519 * The @gfp_mask argument specifies whether I/O may be performed to release
2520 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2523 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2525 struct address_space * const mapping = page->mapping;
2527 BUG_ON(!PageLocked(page));
2528 if (PageWriteback(page))
2531 if (mapping && mapping->a_ops->releasepage)
2532 return mapping->a_ops->releasepage(page, gfp_mask);
2533 return try_to_free_buffers(page);
2536 EXPORT_SYMBOL(try_to_release_page);