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 active_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_active_anon(page);
460 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
463 struct page *__page_cache_alloc(gfp_t gfp)
465 if (cpuset_do_page_mem_spread()) {
466 int n = cpuset_mem_spread_node();
467 return alloc_pages_exact_node(n, gfp, 0);
469 return alloc_pages(gfp, 0);
471 EXPORT_SYMBOL(__page_cache_alloc);
474 static int __sleep_on_page_lock(void *word)
481 * In order to wait for pages to become available there must be
482 * waitqueues associated with pages. By using a hash table of
483 * waitqueues where the bucket discipline is to maintain all
484 * waiters on the same queue and wake all when any of the pages
485 * become available, and for the woken contexts to check to be
486 * sure the appropriate page became available, this saves space
487 * at a cost of "thundering herd" phenomena during rare hash
490 static wait_queue_head_t *page_waitqueue(struct page *page)
492 const struct zone *zone = page_zone(page);
494 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
497 static inline void wake_up_page(struct page *page, int bit)
499 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
502 void wait_on_page_bit(struct page *page, int bit_nr)
504 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
506 if (test_bit(bit_nr, &page->flags))
507 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
508 TASK_UNINTERRUPTIBLE);
510 EXPORT_SYMBOL(wait_on_page_bit);
513 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
514 * @page: Page defining the wait queue of interest
515 * @waiter: Waiter to add to the queue
517 * Add an arbitrary @waiter to the wait queue for the nominated @page.
519 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
521 wait_queue_head_t *q = page_waitqueue(page);
524 spin_lock_irqsave(&q->lock, flags);
525 __add_wait_queue(q, waiter);
526 spin_unlock_irqrestore(&q->lock, flags);
528 EXPORT_SYMBOL_GPL(add_page_wait_queue);
531 * unlock_page - unlock a locked page
534 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
535 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
536 * mechananism between PageLocked pages and PageWriteback pages is shared.
537 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
539 * The mb is necessary to enforce ordering between the clear_bit and the read
540 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
542 void unlock_page(struct page *page)
544 VM_BUG_ON(!PageLocked(page));
545 clear_bit_unlock(PG_locked, &page->flags);
546 smp_mb__after_clear_bit();
547 wake_up_page(page, PG_locked);
549 EXPORT_SYMBOL(unlock_page);
552 * end_page_writeback - end writeback against a page
555 void end_page_writeback(struct page *page)
557 if (TestClearPageReclaim(page))
558 rotate_reclaimable_page(page);
560 if (!test_clear_page_writeback(page))
563 smp_mb__after_clear_bit();
564 wake_up_page(page, PG_writeback);
566 EXPORT_SYMBOL(end_page_writeback);
569 * __lock_page - get a lock on the page, assuming we need to sleep to get it
570 * @page: the page to lock
572 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
573 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
574 * chances are that on the second loop, the block layer's plug list is empty,
575 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
577 void __lock_page(struct page *page)
579 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
581 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
582 TASK_UNINTERRUPTIBLE);
584 EXPORT_SYMBOL(__lock_page);
586 int __lock_page_killable(struct page *page)
588 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
590 return __wait_on_bit_lock(page_waitqueue(page), &wait,
591 sync_page_killable, TASK_KILLABLE);
593 EXPORT_SYMBOL_GPL(__lock_page_killable);
596 * __lock_page_nosync - get a lock on the page, without calling sync_page()
597 * @page: the page to lock
599 * Variant of lock_page that does not require the caller to hold a reference
600 * on the page's mapping.
602 void __lock_page_nosync(struct page *page)
604 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
605 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
606 TASK_UNINTERRUPTIBLE);
610 * find_get_page - find and get a page reference
611 * @mapping: the address_space to search
612 * @offset: the page index
614 * Is there a pagecache struct page at the given (mapping, offset) tuple?
615 * If yes, increment its refcount and return it; if no, return NULL.
617 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
625 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
627 page = radix_tree_deref_slot(pagep);
628 if (unlikely(!page || page == RADIX_TREE_RETRY))
631 if (!page_cache_get_speculative(page))
635 * Has the page moved?
636 * This is part of the lockless pagecache protocol. See
637 * include/linux/pagemap.h for details.
639 if (unlikely(page != *pagep)) {
640 page_cache_release(page);
648 EXPORT_SYMBOL(find_get_page);
651 * find_lock_page - locate, pin and lock a pagecache page
652 * @mapping: the address_space to search
653 * @offset: the page index
655 * Locates the desired pagecache page, locks it, increments its reference
656 * count and returns its address.
658 * Returns zero if the page was not present. find_lock_page() may sleep.
660 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
665 page = find_get_page(mapping, offset);
668 /* Has the page been truncated? */
669 if (unlikely(page->mapping != mapping)) {
671 page_cache_release(page);
674 VM_BUG_ON(page->index != offset);
678 EXPORT_SYMBOL(find_lock_page);
681 * find_or_create_page - locate or add a pagecache page
682 * @mapping: the page's address_space
683 * @index: the page's index into the mapping
684 * @gfp_mask: page allocation mode
686 * Locates a page in the pagecache. If the page is not present, a new page
687 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
688 * LRU list. The returned page is locked and has its reference count
691 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
694 * find_or_create_page() returns the desired page's address, or zero on
697 struct page *find_or_create_page(struct address_space *mapping,
698 pgoff_t index, gfp_t gfp_mask)
703 page = find_lock_page(mapping, index);
705 page = __page_cache_alloc(gfp_mask);
709 * We want a regular kernel memory (not highmem or DMA etc)
710 * allocation for the radix tree nodes, but we need to honour
711 * the context-specific requirements the caller has asked for.
712 * GFP_RECLAIM_MASK collects those requirements.
714 err = add_to_page_cache_lru(page, mapping, index,
715 (gfp_mask & GFP_RECLAIM_MASK));
717 page_cache_release(page);
725 EXPORT_SYMBOL(find_or_create_page);
728 * find_get_pages - gang pagecache lookup
729 * @mapping: The address_space to search
730 * @start: The starting page index
731 * @nr_pages: The maximum number of pages
732 * @pages: Where the resulting pages are placed
734 * find_get_pages() will search for and return a group of up to
735 * @nr_pages pages in the mapping. The pages are placed at @pages.
736 * find_get_pages() takes a reference against the returned pages.
738 * The search returns a group of mapping-contiguous pages with ascending
739 * indexes. There may be holes in the indices due to not-present pages.
741 * find_get_pages() returns the number of pages which were found.
743 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
744 unsigned int nr_pages, struct page **pages)
748 unsigned int nr_found;
752 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
753 (void ***)pages, start, nr_pages);
755 for (i = 0; i < nr_found; i++) {
758 page = radix_tree_deref_slot((void **)pages[i]);
762 * this can only trigger if nr_found == 1, making livelock
765 if (unlikely(page == RADIX_TREE_RETRY))
768 if (!page_cache_get_speculative(page))
771 /* Has the page moved? */
772 if (unlikely(page != *((void **)pages[i]))) {
773 page_cache_release(page);
785 * find_get_pages_contig - gang contiguous pagecache lookup
786 * @mapping: The address_space to search
787 * @index: The starting page index
788 * @nr_pages: The maximum number of pages
789 * @pages: Where the resulting pages are placed
791 * find_get_pages_contig() works exactly like find_get_pages(), except
792 * that the returned number of pages are guaranteed to be contiguous.
794 * find_get_pages_contig() returns the number of pages which were found.
796 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
797 unsigned int nr_pages, struct page **pages)
801 unsigned int nr_found;
805 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
806 (void ***)pages, index, nr_pages);
808 for (i = 0; i < nr_found; i++) {
811 page = radix_tree_deref_slot((void **)pages[i]);
815 * this can only trigger if nr_found == 1, making livelock
818 if (unlikely(page == RADIX_TREE_RETRY))
821 if (page->mapping == NULL || page->index != index)
824 if (!page_cache_get_speculative(page))
827 /* Has the page moved? */
828 if (unlikely(page != *((void **)pages[i]))) {
829 page_cache_release(page);
840 EXPORT_SYMBOL(find_get_pages_contig);
843 * find_get_pages_tag - find and return pages that match @tag
844 * @mapping: the address_space to search
845 * @index: the starting page index
846 * @tag: the tag index
847 * @nr_pages: the maximum number of pages
848 * @pages: where the resulting pages are placed
850 * Like find_get_pages, except we only return pages which are tagged with
851 * @tag. We update @index to index the next page for the traversal.
853 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
854 int tag, unsigned int nr_pages, struct page **pages)
858 unsigned int nr_found;
862 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
863 (void ***)pages, *index, nr_pages, tag);
865 for (i = 0; i < nr_found; i++) {
868 page = radix_tree_deref_slot((void **)pages[i]);
872 * this can only trigger if nr_found == 1, making livelock
875 if (unlikely(page == RADIX_TREE_RETRY))
878 if (!page_cache_get_speculative(page))
881 /* Has the page moved? */
882 if (unlikely(page != *((void **)pages[i]))) {
883 page_cache_release(page);
893 *index = pages[ret - 1]->index + 1;
897 EXPORT_SYMBOL(find_get_pages_tag);
900 * grab_cache_page_nowait - returns locked page at given index in given cache
901 * @mapping: target address_space
902 * @index: the page index
904 * Same as grab_cache_page(), but do not wait if the page is unavailable.
905 * This is intended for speculative data generators, where the data can
906 * be regenerated if the page couldn't be grabbed. This routine should
907 * be safe to call while holding the lock for another page.
909 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
910 * and deadlock against the caller's locked page.
913 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
915 struct page *page = find_get_page(mapping, index);
918 if (trylock_page(page))
920 page_cache_release(page);
923 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
924 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
925 page_cache_release(page);
930 EXPORT_SYMBOL(grab_cache_page_nowait);
933 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
934 * a _large_ part of the i/o request. Imagine the worst scenario:
936 * ---R__________________________________________B__________
937 * ^ reading here ^ bad block(assume 4k)
939 * read(R) => miss => readahead(R...B) => media error => frustrating retries
940 * => failing the whole request => read(R) => read(R+1) =>
941 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
942 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
943 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
945 * It is going insane. Fix it by quickly scaling down the readahead size.
947 static void shrink_readahead_size_eio(struct file *filp,
948 struct file_ra_state *ra)
954 * do_generic_file_read - generic file read routine
955 * @filp: the file to read
956 * @ppos: current file position
957 * @desc: read_descriptor
958 * @actor: read method
960 * This is a generic file read routine, and uses the
961 * mapping->a_ops->readpage() function for the actual low-level stuff.
963 * This is really ugly. But the goto's actually try to clarify some
964 * of the logic when it comes to error handling etc.
966 static void do_generic_file_read(struct file *filp, loff_t *ppos,
967 read_descriptor_t *desc, read_actor_t actor)
969 struct address_space *mapping = filp->f_mapping;
970 struct inode *inode = mapping->host;
971 struct file_ra_state *ra = &filp->f_ra;
975 unsigned long offset; /* offset into pagecache page */
976 unsigned int prev_offset;
979 index = *ppos >> PAGE_CACHE_SHIFT;
980 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
981 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
982 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
983 offset = *ppos & ~PAGE_CACHE_MASK;
989 unsigned long nr, ret;
993 page = find_get_page(mapping, index);
995 page_cache_sync_readahead(mapping,
997 index, last_index - index);
998 page = find_get_page(mapping, index);
999 if (unlikely(page == NULL))
1000 goto no_cached_page;
1002 if (PageReadahead(page)) {
1003 page_cache_async_readahead(mapping,
1005 index, last_index - index);
1007 if (!PageUptodate(page)) {
1008 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1009 !mapping->a_ops->is_partially_uptodate)
1010 goto page_not_up_to_date;
1011 if (!trylock_page(page))
1012 goto page_not_up_to_date;
1013 if (!mapping->a_ops->is_partially_uptodate(page,
1015 goto page_not_up_to_date_locked;
1020 * i_size must be checked after we know the page is Uptodate.
1022 * Checking i_size after the check allows us to calculate
1023 * the correct value for "nr", which means the zero-filled
1024 * part of the page is not copied back to userspace (unless
1025 * another truncate extends the file - this is desired though).
1028 isize = i_size_read(inode);
1029 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1030 if (unlikely(!isize || index > end_index)) {
1031 page_cache_release(page);
1035 /* nr is the maximum number of bytes to copy from this page */
1036 nr = PAGE_CACHE_SIZE;
1037 if (index == end_index) {
1038 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1040 page_cache_release(page);
1046 /* If users can be writing to this page using arbitrary
1047 * virtual addresses, take care about potential aliasing
1048 * before reading the page on the kernel side.
1050 if (mapping_writably_mapped(mapping))
1051 flush_dcache_page(page);
1054 * When a sequential read accesses a page several times,
1055 * only mark it as accessed the first time.
1057 if (prev_index != index || offset != prev_offset)
1058 mark_page_accessed(page);
1062 * Ok, we have the page, and it's up-to-date, so
1063 * now we can copy it to user space...
1065 * The actor routine returns how many bytes were actually used..
1066 * NOTE! This may not be the same as how much of a user buffer
1067 * we filled up (we may be padding etc), so we can only update
1068 * "pos" here (the actor routine has to update the user buffer
1069 * pointers and the remaining count).
1071 ret = actor(desc, page, offset, nr);
1073 index += offset >> PAGE_CACHE_SHIFT;
1074 offset &= ~PAGE_CACHE_MASK;
1075 prev_offset = offset;
1077 page_cache_release(page);
1078 if (ret == nr && desc->count)
1082 page_not_up_to_date:
1083 /* Get exclusive access to the page ... */
1084 error = lock_page_killable(page);
1085 if (unlikely(error))
1086 goto readpage_error;
1088 page_not_up_to_date_locked:
1089 /* Did it get truncated before we got the lock? */
1090 if (!page->mapping) {
1092 page_cache_release(page);
1096 /* Did somebody else fill it already? */
1097 if (PageUptodate(page)) {
1103 /* Start the actual read. The read will unlock the page. */
1104 error = mapping->a_ops->readpage(filp, page);
1106 if (unlikely(error)) {
1107 if (error == AOP_TRUNCATED_PAGE) {
1108 page_cache_release(page);
1111 goto readpage_error;
1114 if (!PageUptodate(page)) {
1115 error = lock_page_killable(page);
1116 if (unlikely(error))
1117 goto readpage_error;
1118 if (!PageUptodate(page)) {
1119 if (page->mapping == NULL) {
1121 * invalidate_mapping_pages got it
1124 page_cache_release(page);
1128 shrink_readahead_size_eio(filp, ra);
1130 goto readpage_error;
1138 /* UHHUH! A synchronous read error occurred. Report it */
1139 desc->error = error;
1140 page_cache_release(page);
1145 * Ok, it wasn't cached, so we need to create a new
1148 page = page_cache_alloc_cold(mapping);
1150 desc->error = -ENOMEM;
1153 error = add_to_page_cache_lru(page, mapping,
1156 page_cache_release(page);
1157 if (error == -EEXIST)
1159 desc->error = error;
1166 ra->prev_pos = prev_index;
1167 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1168 ra->prev_pos |= prev_offset;
1170 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1171 file_accessed(filp);
1174 int file_read_actor(read_descriptor_t *desc, struct page *page,
1175 unsigned long offset, unsigned long size)
1178 unsigned long left, count = desc->count;
1184 * Faults on the destination of a read are common, so do it before
1187 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1188 kaddr = kmap_atomic(page, KM_USER0);
1189 left = __copy_to_user_inatomic(desc->arg.buf,
1190 kaddr + offset, size);
1191 kunmap_atomic(kaddr, KM_USER0);
1196 /* Do it the slow way */
1198 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1203 desc->error = -EFAULT;
1206 desc->count = count - size;
1207 desc->written += size;
1208 desc->arg.buf += size;
1213 * Performs necessary checks before doing a write
1214 * @iov: io vector request
1215 * @nr_segs: number of segments in the iovec
1216 * @count: number of bytes to write
1217 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1219 * Adjust number of segments and amount of bytes to write (nr_segs should be
1220 * properly initialized first). Returns appropriate error code that caller
1221 * should return or zero in case that write should be allowed.
1223 int generic_segment_checks(const struct iovec *iov,
1224 unsigned long *nr_segs, size_t *count, int access_flags)
1228 for (seg = 0; seg < *nr_segs; seg++) {
1229 const struct iovec *iv = &iov[seg];
1232 * If any segment has a negative length, or the cumulative
1233 * length ever wraps negative then return -EINVAL.
1236 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1238 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1243 cnt -= iv->iov_len; /* This segment is no good */
1249 EXPORT_SYMBOL(generic_segment_checks);
1252 * generic_file_aio_read - generic filesystem read routine
1253 * @iocb: kernel I/O control block
1254 * @iov: io vector request
1255 * @nr_segs: number of segments in the iovec
1256 * @pos: current file position
1258 * This is the "read()" routine for all filesystems
1259 * that can use the page cache directly.
1262 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1263 unsigned long nr_segs, loff_t pos)
1265 struct file *filp = iocb->ki_filp;
1269 loff_t *ppos = &iocb->ki_pos;
1272 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1276 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1277 if (filp->f_flags & O_DIRECT) {
1279 struct address_space *mapping;
1280 struct inode *inode;
1282 mapping = filp->f_mapping;
1283 inode = mapping->host;
1285 goto out; /* skip atime */
1286 size = i_size_read(inode);
1288 retval = filemap_write_and_wait_range(mapping, pos,
1289 pos + iov_length(iov, nr_segs) - 1);
1291 retval = mapping->a_ops->direct_IO(READ, iocb,
1295 *ppos = pos + retval;
1297 file_accessed(filp);
1303 for (seg = 0; seg < nr_segs; seg++) {
1304 read_descriptor_t desc;
1307 desc.arg.buf = iov[seg].iov_base;
1308 desc.count = iov[seg].iov_len;
1309 if (desc.count == 0)
1312 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1313 retval += desc.written;
1315 retval = retval ?: desc.error;
1324 EXPORT_SYMBOL(generic_file_aio_read);
1327 do_readahead(struct address_space *mapping, struct file *filp,
1328 pgoff_t index, unsigned long nr)
1330 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1333 force_page_cache_readahead(mapping, filp, index, nr);
1337 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1345 if (file->f_mode & FMODE_READ) {
1346 struct address_space *mapping = file->f_mapping;
1347 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1348 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1349 unsigned long len = end - start + 1;
1350 ret = do_readahead(mapping, file, start, len);
1356 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1357 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1359 return SYSC_readahead((int) fd, offset, (size_t) count);
1361 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1366 * page_cache_read - adds requested page to the page cache if not already there
1367 * @file: file to read
1368 * @offset: page index
1370 * This adds the requested page to the page cache if it isn't already there,
1371 * and schedules an I/O to read in its contents from disk.
1373 static int page_cache_read(struct file *file, pgoff_t offset)
1375 struct address_space *mapping = file->f_mapping;
1380 page = page_cache_alloc_cold(mapping);
1384 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1386 ret = mapping->a_ops->readpage(file, page);
1387 else if (ret == -EEXIST)
1388 ret = 0; /* losing race to add is OK */
1390 page_cache_release(page);
1392 } while (ret == AOP_TRUNCATED_PAGE);
1397 #define MMAP_LOTSAMISS (100)
1400 * Synchronous readahead happens when we don't even find
1401 * a page in the page cache at all.
1403 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1404 struct file_ra_state *ra,
1408 unsigned long ra_pages;
1409 struct address_space *mapping = file->f_mapping;
1411 /* If we don't want any read-ahead, don't bother */
1412 if (VM_RandomReadHint(vma))
1415 if (VM_SequentialReadHint(vma) ||
1416 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1417 page_cache_sync_readahead(mapping, ra, file, offset,
1422 if (ra->mmap_miss < INT_MAX)
1426 * Do we miss much more than hit in this file? If so,
1427 * stop bothering with read-ahead. It will only hurt.
1429 if (ra->mmap_miss > MMAP_LOTSAMISS)
1435 ra_pages = max_sane_readahead(ra->ra_pages);
1437 ra->start = max_t(long, 0, offset - ra_pages/2);
1438 ra->size = ra_pages;
1440 ra_submit(ra, mapping, file);
1445 * Asynchronous readahead happens when we find the page and PG_readahead,
1446 * so we want to possibly extend the readahead further..
1448 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1449 struct file_ra_state *ra,
1454 struct address_space *mapping = file->f_mapping;
1456 /* If we don't want any read-ahead, don't bother */
1457 if (VM_RandomReadHint(vma))
1459 if (ra->mmap_miss > 0)
1461 if (PageReadahead(page))
1462 page_cache_async_readahead(mapping, ra, file,
1463 page, offset, ra->ra_pages);
1467 * filemap_fault - read in file data for page fault handling
1468 * @vma: vma in which the fault was taken
1469 * @vmf: struct vm_fault containing details of the fault
1471 * filemap_fault() is invoked via the vma operations vector for a
1472 * mapped memory region to read in file data during a page fault.
1474 * The goto's are kind of ugly, but this streamlines the normal case of having
1475 * it in the page cache, and handles the special cases reasonably without
1476 * having a lot of duplicated code.
1478 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1481 struct file *file = vma->vm_file;
1482 struct address_space *mapping = file->f_mapping;
1483 struct file_ra_state *ra = &file->f_ra;
1484 struct inode *inode = mapping->host;
1485 pgoff_t offset = vmf->pgoff;
1490 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1492 return VM_FAULT_SIGBUS;
1495 * Do we have something in the page cache already?
1497 page = find_get_page(mapping, offset);
1500 * We found the page, so try async readahead before
1501 * waiting for the lock.
1503 do_async_mmap_readahead(vma, ra, file, page, offset);
1506 /* Did it get truncated? */
1507 if (unlikely(page->mapping != mapping)) {
1510 goto no_cached_page;
1513 /* No page in the page cache at all */
1514 do_sync_mmap_readahead(vma, ra, file, offset);
1515 count_vm_event(PGMAJFAULT);
1516 ret = VM_FAULT_MAJOR;
1518 page = find_lock_page(mapping, offset);
1520 goto no_cached_page;
1524 * We have a locked page in the page cache, now we need to check
1525 * that it's up-to-date. If not, it is going to be due to an error.
1527 if (unlikely(!PageUptodate(page)))
1528 goto page_not_uptodate;
1531 * Found the page and have a reference on it.
1532 * We must recheck i_size under page lock.
1534 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1535 if (unlikely(offset >= size)) {
1537 page_cache_release(page);
1538 return VM_FAULT_SIGBUS;
1541 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1543 return ret | VM_FAULT_LOCKED;
1547 * We're only likely to ever get here if MADV_RANDOM is in
1550 error = page_cache_read(file, offset);
1553 * The page we want has now been added to the page cache.
1554 * In the unlikely event that someone removed it in the
1555 * meantime, we'll just come back here and read it again.
1561 * An error return from page_cache_read can result if the
1562 * system is low on memory, or a problem occurs while trying
1565 if (error == -ENOMEM)
1566 return VM_FAULT_OOM;
1567 return VM_FAULT_SIGBUS;
1571 * Umm, take care of errors if the page isn't up-to-date.
1572 * Try to re-read it _once_. We do this synchronously,
1573 * because there really aren't any performance issues here
1574 * and we need to check for errors.
1576 ClearPageError(page);
1577 error = mapping->a_ops->readpage(file, page);
1579 wait_on_page_locked(page);
1580 if (!PageUptodate(page))
1583 page_cache_release(page);
1585 if (!error || error == AOP_TRUNCATED_PAGE)
1588 /* Things didn't work out. Return zero to tell the mm layer so. */
1589 shrink_readahead_size_eio(file, ra);
1590 return VM_FAULT_SIGBUS;
1592 EXPORT_SYMBOL(filemap_fault);
1594 const struct vm_operations_struct generic_file_vm_ops = {
1595 .fault = filemap_fault,
1598 /* This is used for a general mmap of a disk file */
1600 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1602 struct address_space *mapping = file->f_mapping;
1604 if (!mapping->a_ops->readpage)
1606 file_accessed(file);
1607 vma->vm_ops = &generic_file_vm_ops;
1608 vma->vm_flags |= VM_CAN_NONLINEAR;
1613 * This is for filesystems which do not implement ->writepage.
1615 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1617 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1619 return generic_file_mmap(file, vma);
1622 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1626 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1630 #endif /* CONFIG_MMU */
1632 EXPORT_SYMBOL(generic_file_mmap);
1633 EXPORT_SYMBOL(generic_file_readonly_mmap);
1635 static struct page *__read_cache_page(struct address_space *mapping,
1637 int (*filler)(void *,struct page*),
1644 page = find_get_page(mapping, index);
1646 page = __page_cache_alloc(gfp | __GFP_COLD);
1648 return ERR_PTR(-ENOMEM);
1649 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1650 if (unlikely(err)) {
1651 page_cache_release(page);
1654 /* Presumably ENOMEM for radix tree node */
1655 return ERR_PTR(err);
1657 err = filler(data, page);
1659 page_cache_release(page);
1660 page = ERR_PTR(err);
1666 static struct page *do_read_cache_page(struct address_space *mapping,
1668 int (*filler)(void *,struct page*),
1677 page = __read_cache_page(mapping, index, filler, data, gfp);
1680 if (PageUptodate(page))
1684 if (!page->mapping) {
1686 page_cache_release(page);
1689 if (PageUptodate(page)) {
1693 err = filler(data, page);
1695 page_cache_release(page);
1696 return ERR_PTR(err);
1699 mark_page_accessed(page);
1704 * read_cache_page_async - read into page cache, fill it if needed
1705 * @mapping: the page's address_space
1706 * @index: the page index
1707 * @filler: function to perform the read
1708 * @data: destination for read data
1710 * Same as read_cache_page, but don't wait for page to become unlocked
1711 * after submitting it to the filler.
1713 * Read into the page cache. If a page already exists, and PageUptodate() is
1714 * not set, try to fill the page but don't wait for it to become unlocked.
1716 * If the page does not get brought uptodate, return -EIO.
1718 struct page *read_cache_page_async(struct address_space *mapping,
1720 int (*filler)(void *,struct page*),
1723 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1725 EXPORT_SYMBOL(read_cache_page_async);
1727 static struct page *wait_on_page_read(struct page *page)
1729 if (!IS_ERR(page)) {
1730 wait_on_page_locked(page);
1731 if (!PageUptodate(page)) {
1732 page_cache_release(page);
1733 page = ERR_PTR(-EIO);
1740 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1741 * @mapping: the page's address_space
1742 * @index: the page index
1743 * @gfp: the page allocator flags to use if allocating
1745 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1746 * any new page allocations done using the specified allocation flags. Note
1747 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1748 * expect to do this atomically or anything like that - but you can pass in
1749 * other page requirements.
1751 * If the page does not get brought uptodate, return -EIO.
1753 struct page *read_cache_page_gfp(struct address_space *mapping,
1757 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1759 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1761 EXPORT_SYMBOL(read_cache_page_gfp);
1764 * read_cache_page - read into page cache, fill it if needed
1765 * @mapping: the page's address_space
1766 * @index: the page index
1767 * @filler: function to perform the read
1768 * @data: destination for read data
1770 * Read into the page cache. If a page already exists, and PageUptodate() is
1771 * not set, try to fill the page then wait for it to become unlocked.
1773 * If the page does not get brought uptodate, return -EIO.
1775 struct page *read_cache_page(struct address_space *mapping,
1777 int (*filler)(void *,struct page*),
1780 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1782 EXPORT_SYMBOL(read_cache_page);
1785 * The logic we want is
1787 * if suid or (sgid and xgrp)
1790 int should_remove_suid(struct dentry *dentry)
1792 mode_t mode = dentry->d_inode->i_mode;
1795 /* suid always must be killed */
1796 if (unlikely(mode & S_ISUID))
1797 kill = ATTR_KILL_SUID;
1800 * sgid without any exec bits is just a mandatory locking mark; leave
1801 * it alone. If some exec bits are set, it's a real sgid; kill it.
1803 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1804 kill |= ATTR_KILL_SGID;
1806 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1811 EXPORT_SYMBOL(should_remove_suid);
1813 static int __remove_suid(struct dentry *dentry, int kill)
1815 struct iattr newattrs;
1817 newattrs.ia_valid = ATTR_FORCE | kill;
1818 return notify_change(dentry, &newattrs);
1821 int file_remove_suid(struct file *file)
1823 struct dentry *dentry = file->f_path.dentry;
1824 int killsuid = should_remove_suid(dentry);
1825 int killpriv = security_inode_need_killpriv(dentry);
1831 error = security_inode_killpriv(dentry);
1832 if (!error && killsuid)
1833 error = __remove_suid(dentry, killsuid);
1837 EXPORT_SYMBOL(file_remove_suid);
1839 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1840 const struct iovec *iov, size_t base, size_t bytes)
1842 size_t copied = 0, left = 0;
1845 char __user *buf = iov->iov_base + base;
1846 int copy = min(bytes, iov->iov_len - base);
1849 left = __copy_from_user_inatomic(vaddr, buf, copy);
1858 return copied - left;
1862 * Copy as much as we can into the page and return the number of bytes which
1863 * were successfully copied. If a fault is encountered then return the number of
1864 * bytes which were copied.
1866 size_t iov_iter_copy_from_user_atomic(struct page *page,
1867 struct iov_iter *i, unsigned long offset, size_t bytes)
1872 BUG_ON(!in_atomic());
1873 kaddr = kmap_atomic(page, KM_USER0);
1874 if (likely(i->nr_segs == 1)) {
1876 char __user *buf = i->iov->iov_base + i->iov_offset;
1877 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1878 copied = bytes - left;
1880 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1881 i->iov, i->iov_offset, bytes);
1883 kunmap_atomic(kaddr, KM_USER0);
1887 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1890 * This has the same sideeffects and return value as
1891 * iov_iter_copy_from_user_atomic().
1892 * The difference is that it attempts to resolve faults.
1893 * Page must not be locked.
1895 size_t iov_iter_copy_from_user(struct page *page,
1896 struct iov_iter *i, unsigned long offset, size_t bytes)
1902 if (likely(i->nr_segs == 1)) {
1904 char __user *buf = i->iov->iov_base + i->iov_offset;
1905 left = __copy_from_user(kaddr + offset, buf, bytes);
1906 copied = bytes - left;
1908 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1909 i->iov, i->iov_offset, bytes);
1914 EXPORT_SYMBOL(iov_iter_copy_from_user);
1916 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1918 BUG_ON(i->count < bytes);
1920 if (likely(i->nr_segs == 1)) {
1921 i->iov_offset += bytes;
1924 const struct iovec *iov = i->iov;
1925 size_t base = i->iov_offset;
1928 * The !iov->iov_len check ensures we skip over unlikely
1929 * zero-length segments (without overruning the iovec).
1931 while (bytes || unlikely(i->count && !iov->iov_len)) {
1934 copy = min(bytes, iov->iov_len - base);
1935 BUG_ON(!i->count || i->count < copy);
1939 if (iov->iov_len == base) {
1945 i->iov_offset = base;
1948 EXPORT_SYMBOL(iov_iter_advance);
1951 * Fault in the first iovec of the given iov_iter, to a maximum length
1952 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1953 * accessed (ie. because it is an invalid address).
1955 * writev-intensive code may want this to prefault several iovecs -- that
1956 * would be possible (callers must not rely on the fact that _only_ the
1957 * first iovec will be faulted with the current implementation).
1959 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
1961 char __user *buf = i->iov->iov_base + i->iov_offset;
1962 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1963 return fault_in_pages_readable(buf, bytes);
1965 EXPORT_SYMBOL(iov_iter_fault_in_readable);
1968 * Return the count of just the current iov_iter segment.
1970 size_t iov_iter_single_seg_count(struct iov_iter *i)
1972 const struct iovec *iov = i->iov;
1973 if (i->nr_segs == 1)
1976 return min(i->count, iov->iov_len - i->iov_offset);
1978 EXPORT_SYMBOL(iov_iter_single_seg_count);
1981 * Performs necessary checks before doing a write
1983 * Can adjust writing position or amount of bytes to write.
1984 * Returns appropriate error code that caller should return or
1985 * zero in case that write should be allowed.
1987 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1989 struct inode *inode = file->f_mapping->host;
1990 unsigned long limit = rlimit(RLIMIT_FSIZE);
1992 if (unlikely(*pos < 0))
1996 /* FIXME: this is for backwards compatibility with 2.4 */
1997 if (file->f_flags & O_APPEND)
1998 *pos = i_size_read(inode);
2000 if (limit != RLIM_INFINITY) {
2001 if (*pos >= limit) {
2002 send_sig(SIGXFSZ, current, 0);
2005 if (*count > limit - (typeof(limit))*pos) {
2006 *count = limit - (typeof(limit))*pos;
2014 if (unlikely(*pos + *count > MAX_NON_LFS &&
2015 !(file->f_flags & O_LARGEFILE))) {
2016 if (*pos >= MAX_NON_LFS) {
2019 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2020 *count = MAX_NON_LFS - (unsigned long)*pos;
2025 * Are we about to exceed the fs block limit ?
2027 * If we have written data it becomes a short write. If we have
2028 * exceeded without writing data we send a signal and return EFBIG.
2029 * Linus frestrict idea will clean these up nicely..
2031 if (likely(!isblk)) {
2032 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2033 if (*count || *pos > inode->i_sb->s_maxbytes) {
2036 /* zero-length writes at ->s_maxbytes are OK */
2039 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2040 *count = inode->i_sb->s_maxbytes - *pos;
2044 if (bdev_read_only(I_BDEV(inode)))
2046 isize = i_size_read(inode);
2047 if (*pos >= isize) {
2048 if (*count || *pos > isize)
2052 if (*pos + *count > isize)
2053 *count = isize - *pos;
2060 EXPORT_SYMBOL(generic_write_checks);
2062 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2063 loff_t pos, unsigned len, unsigned flags,
2064 struct page **pagep, void **fsdata)
2066 const struct address_space_operations *aops = mapping->a_ops;
2068 return aops->write_begin(file, mapping, pos, len, flags,
2071 EXPORT_SYMBOL(pagecache_write_begin);
2073 int pagecache_write_end(struct file *file, struct address_space *mapping,
2074 loff_t pos, unsigned len, unsigned copied,
2075 struct page *page, void *fsdata)
2077 const struct address_space_operations *aops = mapping->a_ops;
2079 mark_page_accessed(page);
2080 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2082 EXPORT_SYMBOL(pagecache_write_end);
2085 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2086 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2087 size_t count, size_t ocount)
2089 struct file *file = iocb->ki_filp;
2090 struct address_space *mapping = file->f_mapping;
2091 struct inode *inode = mapping->host;
2096 if (count != ocount)
2097 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2099 write_len = iov_length(iov, *nr_segs);
2100 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2102 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2107 * After a write we want buffered reads to be sure to go to disk to get
2108 * the new data. We invalidate clean cached page from the region we're
2109 * about to write. We do this *before* the write so that we can return
2110 * without clobbering -EIOCBQUEUED from ->direct_IO().
2112 if (mapping->nrpages) {
2113 written = invalidate_inode_pages2_range(mapping,
2114 pos >> PAGE_CACHE_SHIFT, end);
2116 * If a page can not be invalidated, return 0 to fall back
2117 * to buffered write.
2120 if (written == -EBUSY)
2126 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2129 * Finally, try again to invalidate clean pages which might have been
2130 * cached by non-direct readahead, or faulted in by get_user_pages()
2131 * if the source of the write was an mmap'ed region of the file
2132 * we're writing. Either one is a pretty crazy thing to do,
2133 * so we don't support it 100%. If this invalidation
2134 * fails, tough, the write still worked...
2136 if (mapping->nrpages) {
2137 invalidate_inode_pages2_range(mapping,
2138 pos >> PAGE_CACHE_SHIFT, end);
2142 loff_t end = pos + written;
2143 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2144 i_size_write(inode, end);
2145 mark_inode_dirty(inode);
2152 EXPORT_SYMBOL(generic_file_direct_write);
2155 * Find or create a page at the given pagecache position. Return the locked
2156 * page. This function is specifically for buffered writes.
2158 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2159 pgoff_t index, unsigned flags)
2163 gfp_t gfp_notmask = 0;
2164 if (flags & AOP_FLAG_NOFS)
2165 gfp_notmask = __GFP_FS;
2167 page = find_lock_page(mapping, index);
2171 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2174 status = add_to_page_cache_lru(page, mapping, index,
2175 GFP_KERNEL & ~gfp_notmask);
2176 if (unlikely(status)) {
2177 page_cache_release(page);
2178 if (status == -EEXIST)
2184 EXPORT_SYMBOL(grab_cache_page_write_begin);
2186 static ssize_t generic_perform_write(struct file *file,
2187 struct iov_iter *i, loff_t pos)
2189 struct address_space *mapping = file->f_mapping;
2190 const struct address_space_operations *a_ops = mapping->a_ops;
2192 ssize_t written = 0;
2193 unsigned int flags = 0;
2196 * Copies from kernel address space cannot fail (NFSD is a big user).
2198 if (segment_eq(get_fs(), KERNEL_DS))
2199 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2203 pgoff_t index; /* Pagecache index for current page */
2204 unsigned long offset; /* Offset into pagecache page */
2205 unsigned long bytes; /* Bytes to write to page */
2206 size_t copied; /* Bytes copied from user */
2209 offset = (pos & (PAGE_CACHE_SIZE - 1));
2210 index = pos >> PAGE_CACHE_SHIFT;
2211 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2217 * Bring in the user page that we will copy from _first_.
2218 * Otherwise there's a nasty deadlock on copying from the
2219 * same page as we're writing to, without it being marked
2222 * Not only is this an optimisation, but it is also required
2223 * to check that the address is actually valid, when atomic
2224 * usercopies are used, below.
2226 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2231 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2233 if (unlikely(status))
2236 if (mapping_writably_mapped(mapping))
2237 flush_dcache_page(page);
2239 pagefault_disable();
2240 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2242 flush_dcache_page(page);
2244 mark_page_accessed(page);
2245 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2247 if (unlikely(status < 0))
2253 iov_iter_advance(i, copied);
2254 if (unlikely(copied == 0)) {
2256 * If we were unable to copy any data at all, we must
2257 * fall back to a single segment length write.
2259 * If we didn't fallback here, we could livelock
2260 * because not all segments in the iov can be copied at
2261 * once without a pagefault.
2263 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2264 iov_iter_single_seg_count(i));
2270 balance_dirty_pages_ratelimited(mapping);
2272 } while (iov_iter_count(i));
2274 return written ? written : status;
2278 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2279 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2280 size_t count, ssize_t written)
2282 struct file *file = iocb->ki_filp;
2286 iov_iter_init(&i, iov, nr_segs, count, written);
2287 status = generic_perform_write(file, &i, pos);
2289 if (likely(status >= 0)) {
2291 *ppos = pos + status;
2294 return written ? written : status;
2296 EXPORT_SYMBOL(generic_file_buffered_write);
2299 * __generic_file_aio_write - write data to a file
2300 * @iocb: IO state structure (file, offset, etc.)
2301 * @iov: vector with data to write
2302 * @nr_segs: number of segments in the vector
2303 * @ppos: position where to write
2305 * This function does all the work needed for actually writing data to a
2306 * file. It does all basic checks, removes SUID from the file, updates
2307 * modification times and calls proper subroutines depending on whether we
2308 * do direct IO or a standard buffered write.
2310 * It expects i_mutex to be grabbed unless we work on a block device or similar
2311 * object which does not need locking at all.
2313 * This function does *not* take care of syncing data in case of O_SYNC write.
2314 * A caller has to handle it. This is mainly due to the fact that we want to
2315 * avoid syncing under i_mutex.
2317 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2318 unsigned long nr_segs, loff_t *ppos)
2320 struct file *file = iocb->ki_filp;
2321 struct address_space * mapping = file->f_mapping;
2322 size_t ocount; /* original count */
2323 size_t count; /* after file limit checks */
2324 struct inode *inode = mapping->host;
2330 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2337 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2339 /* We can write back this queue in page reclaim */
2340 current->backing_dev_info = mapping->backing_dev_info;
2343 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2350 err = file_remove_suid(file);
2354 file_update_time(file);
2356 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2357 if (unlikely(file->f_flags & O_DIRECT)) {
2359 ssize_t written_buffered;
2361 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2362 ppos, count, ocount);
2363 if (written < 0 || written == count)
2366 * direct-io write to a hole: fall through to buffered I/O
2367 * for completing the rest of the request.
2371 written_buffered = generic_file_buffered_write(iocb, iov,
2372 nr_segs, pos, ppos, count,
2375 * If generic_file_buffered_write() retuned a synchronous error
2376 * then we want to return the number of bytes which were
2377 * direct-written, or the error code if that was zero. Note
2378 * that this differs from normal direct-io semantics, which
2379 * will return -EFOO even if some bytes were written.
2381 if (written_buffered < 0) {
2382 err = written_buffered;
2387 * We need to ensure that the page cache pages are written to
2388 * disk and invalidated to preserve the expected O_DIRECT
2391 endbyte = pos + written_buffered - written - 1;
2392 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2394 written = written_buffered;
2395 invalidate_mapping_pages(mapping,
2396 pos >> PAGE_CACHE_SHIFT,
2397 endbyte >> PAGE_CACHE_SHIFT);
2400 * We don't know how much we wrote, so just return
2401 * the number of bytes which were direct-written
2405 written = generic_file_buffered_write(iocb, iov, nr_segs,
2406 pos, ppos, count, written);
2409 current->backing_dev_info = NULL;
2410 return written ? written : err;
2412 EXPORT_SYMBOL(__generic_file_aio_write);
2415 * generic_file_aio_write - write data to a file
2416 * @iocb: IO state structure
2417 * @iov: vector with data to write
2418 * @nr_segs: number of segments in the vector
2419 * @pos: position in file where to write
2421 * This is a wrapper around __generic_file_aio_write() to be used by most
2422 * filesystems. It takes care of syncing the file in case of O_SYNC file
2423 * and acquires i_mutex as needed.
2425 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2426 unsigned long nr_segs, loff_t pos)
2428 struct file *file = iocb->ki_filp;
2429 struct inode *inode = file->f_mapping->host;
2432 BUG_ON(iocb->ki_pos != pos);
2434 mutex_lock(&inode->i_mutex);
2435 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2436 mutex_unlock(&inode->i_mutex);
2438 if (ret > 0 || ret == -EIOCBQUEUED) {
2441 err = generic_write_sync(file, pos, ret);
2442 if (err < 0 && ret > 0)
2447 EXPORT_SYMBOL(generic_file_aio_write);
2450 * try_to_release_page() - release old fs-specific metadata on a page
2452 * @page: the page which the kernel is trying to free
2453 * @gfp_mask: memory allocation flags (and I/O mode)
2455 * The address_space is to try to release any data against the page
2456 * (presumably at page->private). If the release was successful, return `1'.
2457 * Otherwise return zero.
2459 * This may also be called if PG_fscache is set on a page, indicating that the
2460 * page is known to the local caching routines.
2462 * The @gfp_mask argument specifies whether I/O may be performed to release
2463 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2466 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2468 struct address_space * const mapping = page->mapping;
2470 BUG_ON(!PageLocked(page));
2471 if (PageWriteback(page))
2474 if (mapping && mapping->a_ops->releasepage)
2475 return mapping->a_ops->releasepage(page, gfp_mask);
2476 return try_to_free_buffers(page);
2479 EXPORT_SYMBOL(try_to_release_page);