4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/compiler.h>
16 #include <linux/uaccess.h>
17 #include <linux/aio.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/backing-dev.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/cpuset.h>
35 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
36 #include <linux/memcontrol.h>
40 * FIXME: remove all knowledge of the buffer layer from the core VM
42 #include <linux/buffer_head.h> /* for generic_osync_inode */
47 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
48 loff_t offset, unsigned long nr_segs);
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
54 * Shared mappings now work. 15.8.1995 Bruno.
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
65 * ->i_mmap_lock (vmtruncate)
66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
71 * ->i_mmap_lock (truncate->unmap_mapping_range)
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
79 * ->lock_page (access_process_vm)
81 * ->i_mutex (generic_file_buffered_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
85 * ->i_alloc_sem (various)
88 * ->sb_lock (fs/fs-writeback.c)
89 * ->mapping->tree_lock (__sync_single_inode)
92 * ->anon_vma.lock (vma_adjust)
95 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
97 * ->page_table_lock or pte_lock
98 * ->swap_lock (try_to_unmap_one)
99 * ->private_lock (try_to_unmap_one)
100 * ->tree_lock (try_to_unmap_one)
101 * ->zone.lru_lock (follow_page->mark_page_accessed)
102 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
103 * ->private_lock (page_remove_rmap->set_page_dirty)
104 * ->tree_lock (page_remove_rmap->set_page_dirty)
105 * ->inode_lock (page_remove_rmap->set_page_dirty)
106 * ->inode_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
110 * ->dcache_lock (proc_pid_lookup)
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 a write_lock on the mapping's tree_lock.
118 void __remove_from_page_cache(struct page *page)
120 struct address_space *mapping = page->mapping;
122 mem_cgroup_uncharge_page(page);
123 radix_tree_delete(&mapping->page_tree, page->index);
124 page->mapping = NULL;
126 __dec_zone_page_state(page, NR_FILE_PAGES);
127 BUG_ON(page_mapped(page));
130 * Some filesystems seem to re-dirty the page even after
131 * the VM has canceled the dirty bit (eg ext3 journaling).
133 * Fix it up by doing a final dirty accounting check after
134 * having removed the page entirely.
136 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
137 dec_zone_page_state(page, NR_FILE_DIRTY);
138 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
142 void remove_from_page_cache(struct page *page)
144 struct address_space *mapping = page->mapping;
146 BUG_ON(!PageLocked(page));
148 write_lock_irq(&mapping->tree_lock);
149 __remove_from_page_cache(page);
150 write_unlock_irq(&mapping->tree_lock);
153 static int sync_page(void *word)
155 struct address_space *mapping;
158 page = container_of((unsigned long *)word, struct page, flags);
161 * page_mapping() is being called without PG_locked held.
162 * Some knowledge of the state and use of the page is used to
163 * reduce the requirements down to a memory barrier.
164 * The danger here is of a stale page_mapping() return value
165 * indicating a struct address_space different from the one it's
166 * associated with when it is associated with one.
167 * After smp_mb(), it's either the correct page_mapping() for
168 * the page, or an old page_mapping() and the page's own
169 * page_mapping() has gone NULL.
170 * The ->sync_page() address_space operation must tolerate
171 * page_mapping() going NULL. By an amazing coincidence,
172 * this comes about because none of the users of the page
173 * in the ->sync_page() methods make essential use of the
174 * page_mapping(), merely passing the page down to the backing
175 * device's unplug functions when it's non-NULL, which in turn
176 * ignore it for all cases but swap, where only page_private(page) is
177 * of interest. When page_mapping() does go NULL, the entire
178 * call stack gracefully ignores the page and returns.
182 mapping = page_mapping(page);
183 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
184 mapping->a_ops->sync_page(page);
189 static int sync_page_killable(void *word)
192 return fatal_signal_pending(current) ? -EINTR : 0;
196 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
197 * @mapping: address space structure to write
198 * @start: offset in bytes where the range starts
199 * @end: offset in bytes where the range ends (inclusive)
200 * @sync_mode: enable synchronous operation
202 * Start writeback against all of a mapping's dirty pages that lie
203 * within the byte offsets <start, end> inclusive.
205 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
206 * opposed to a regular memory cleansing writeback. The difference between
207 * these two operations is that if a dirty page/buffer is encountered, it must
208 * be waited upon, and not just skipped over.
210 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
211 loff_t end, int sync_mode)
214 struct writeback_control wbc = {
215 .sync_mode = sync_mode,
216 .nr_to_write = mapping->nrpages * 2,
217 .range_start = start,
221 if (!mapping_cap_writeback_dirty(mapping))
224 ret = do_writepages(mapping, &wbc);
228 static inline int __filemap_fdatawrite(struct address_space *mapping,
231 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
234 int filemap_fdatawrite(struct address_space *mapping)
236 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
238 EXPORT_SYMBOL(filemap_fdatawrite);
240 static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
243 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
247 * filemap_flush - mostly a non-blocking flush
248 * @mapping: target address_space
250 * This is a mostly non-blocking flush. Not suitable for data-integrity
251 * purposes - I/O may not be started against all dirty pages.
253 int filemap_flush(struct address_space *mapping)
255 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
257 EXPORT_SYMBOL(filemap_flush);
260 * wait_on_page_writeback_range - wait for writeback to complete
261 * @mapping: target address_space
262 * @start: beginning page index
263 * @end: ending page index
265 * Wait for writeback to complete against pages indexed by start->end
268 int wait_on_page_writeback_range(struct address_space *mapping,
269 pgoff_t start, pgoff_t end)
279 pagevec_init(&pvec, 0);
281 while ((index <= end) &&
282 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
283 PAGECACHE_TAG_WRITEBACK,
284 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
287 for (i = 0; i < nr_pages; i++) {
288 struct page *page = pvec.pages[i];
290 /* until radix tree lookup accepts end_index */
291 if (page->index > end)
294 wait_on_page_writeback(page);
298 pagevec_release(&pvec);
302 /* Check for outstanding write errors */
303 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
305 if (test_and_clear_bit(AS_EIO, &mapping->flags))
312 * sync_page_range - write and wait on all pages in the passed range
313 * @inode: target inode
314 * @mapping: target address_space
315 * @pos: beginning offset in pages to write
316 * @count: number of bytes to write
318 * Write and wait upon all the pages in the passed range. This is a "data
319 * integrity" operation. It waits upon in-flight writeout before starting and
320 * waiting upon new writeout. If there was an IO error, return it.
322 * We need to re-take i_mutex during the generic_osync_inode list walk because
323 * it is otherwise livelockable.
325 int sync_page_range(struct inode *inode, struct address_space *mapping,
326 loff_t pos, loff_t count)
328 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
329 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
332 if (!mapping_cap_writeback_dirty(mapping) || !count)
334 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
336 mutex_lock(&inode->i_mutex);
337 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
338 mutex_unlock(&inode->i_mutex);
341 ret = wait_on_page_writeback_range(mapping, start, end);
344 EXPORT_SYMBOL(sync_page_range);
347 * sync_page_range_nolock
348 * @inode: target inode
349 * @mapping: target address_space
350 * @pos: beginning offset in pages to write
351 * @count: number of bytes to write
353 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
354 * as it forces O_SYNC writers to different parts of the same file
355 * to be serialised right until io completion.
357 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
358 loff_t pos, loff_t count)
360 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
361 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
364 if (!mapping_cap_writeback_dirty(mapping) || !count)
366 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
368 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
370 ret = wait_on_page_writeback_range(mapping, start, end);
373 EXPORT_SYMBOL(sync_page_range_nolock);
376 * filemap_fdatawait - wait for all under-writeback pages to complete
377 * @mapping: address space structure to wait for
379 * Walk the list of under-writeback pages of the given address space
380 * and wait for all of them.
382 int filemap_fdatawait(struct address_space *mapping)
384 loff_t i_size = i_size_read(mapping->host);
389 return wait_on_page_writeback_range(mapping, 0,
390 (i_size - 1) >> PAGE_CACHE_SHIFT);
392 EXPORT_SYMBOL(filemap_fdatawait);
394 int filemap_write_and_wait(struct address_space *mapping)
398 if (mapping->nrpages) {
399 err = filemap_fdatawrite(mapping);
401 * Even if the above returned error, the pages may be
402 * written partially (e.g. -ENOSPC), so we wait for it.
403 * But the -EIO is special case, it may indicate the worst
404 * thing (e.g. bug) happened, so we avoid waiting for it.
407 int err2 = filemap_fdatawait(mapping);
414 EXPORT_SYMBOL(filemap_write_and_wait);
417 * filemap_write_and_wait_range - write out & wait on a file range
418 * @mapping: the address_space for the pages
419 * @lstart: offset in bytes where the range starts
420 * @lend: offset in bytes where the range ends (inclusive)
422 * Write out and wait upon file offsets lstart->lend, inclusive.
424 * Note that `lend' is inclusive (describes the last byte to be written) so
425 * that this function can be used to write to the very end-of-file (end = -1).
427 int filemap_write_and_wait_range(struct address_space *mapping,
428 loff_t lstart, loff_t lend)
432 if (mapping->nrpages) {
433 err = __filemap_fdatawrite_range(mapping, lstart, lend,
435 /* See comment of filemap_write_and_wait() */
437 int err2 = wait_on_page_writeback_range(mapping,
438 lstart >> PAGE_CACHE_SHIFT,
439 lend >> PAGE_CACHE_SHIFT);
448 * add_to_page_cache - add newly allocated pagecache pages
450 * @mapping: the page's address_space
451 * @offset: page index
452 * @gfp_mask: page allocation mode
454 * This function is used to add newly allocated pagecache pages;
455 * the page is new, so we can just run SetPageLocked() against it.
456 * The other page state flags were set by rmqueue().
458 * This function does not add the page to the LRU. The caller must do that.
460 int add_to_page_cache(struct page *page, struct address_space *mapping,
461 pgoff_t offset, gfp_t gfp_mask)
463 int error = mem_cgroup_cache_charge(page, current->mm, gfp_mask);
467 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
469 write_lock_irq(&mapping->tree_lock);
470 error = radix_tree_insert(&mapping->page_tree, offset, page);
472 page_cache_get(page);
474 page->mapping = mapping;
475 page->index = offset;
477 __inc_zone_page_state(page, NR_FILE_PAGES);
479 mem_cgroup_uncharge_page(page);
481 write_unlock_irq(&mapping->tree_lock);
482 radix_tree_preload_end();
484 mem_cgroup_uncharge_page(page);
488 EXPORT_SYMBOL(add_to_page_cache);
490 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
491 pgoff_t offset, gfp_t gfp_mask)
493 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
500 struct page *__page_cache_alloc(gfp_t gfp)
502 if (cpuset_do_page_mem_spread()) {
503 int n = cpuset_mem_spread_node();
504 return alloc_pages_node(n, gfp, 0);
506 return alloc_pages(gfp, 0);
508 EXPORT_SYMBOL(__page_cache_alloc);
511 static int __sleep_on_page_lock(void *word)
518 * In order to wait for pages to become available there must be
519 * waitqueues associated with pages. By using a hash table of
520 * waitqueues where the bucket discipline is to maintain all
521 * waiters on the same queue and wake all when any of the pages
522 * become available, and for the woken contexts to check to be
523 * sure the appropriate page became available, this saves space
524 * at a cost of "thundering herd" phenomena during rare hash
527 static wait_queue_head_t *page_waitqueue(struct page *page)
529 const struct zone *zone = page_zone(page);
531 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
534 static inline void wake_up_page(struct page *page, int bit)
536 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
539 void wait_on_page_bit(struct page *page, int bit_nr)
541 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
543 if (test_bit(bit_nr, &page->flags))
544 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
545 TASK_UNINTERRUPTIBLE);
547 EXPORT_SYMBOL(wait_on_page_bit);
550 * unlock_page - unlock a locked page
553 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
554 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
555 * mechananism between PageLocked pages and PageWriteback pages is shared.
556 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
558 * The first mb is necessary to safely close the critical section opened by the
559 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
560 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
561 * parallel wait_on_page_locked()).
563 void unlock_page(struct page *page)
565 smp_mb__before_clear_bit();
566 if (!TestClearPageLocked(page))
568 smp_mb__after_clear_bit();
569 wake_up_page(page, PG_locked);
571 EXPORT_SYMBOL(unlock_page);
574 * end_page_writeback - end writeback against a page
577 void end_page_writeback(struct page *page)
579 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
580 if (!test_clear_page_writeback(page))
583 smp_mb__after_clear_bit();
584 wake_up_page(page, PG_writeback);
586 EXPORT_SYMBOL(end_page_writeback);
589 * __lock_page - get a lock on the page, assuming we need to sleep to get it
590 * @page: the page to lock
592 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
593 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
594 * chances are that on the second loop, the block layer's plug list is empty,
595 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
597 void __lock_page(struct page *page)
599 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
601 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
602 TASK_UNINTERRUPTIBLE);
604 EXPORT_SYMBOL(__lock_page);
606 int fastcall __lock_page_killable(struct page *page)
608 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
610 return __wait_on_bit_lock(page_waitqueue(page), &wait,
611 sync_page_killable, TASK_KILLABLE);
615 * Variant of lock_page that does not require the caller to hold a reference
616 * on the page's mapping.
618 void __lock_page_nosync(struct page *page)
620 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
621 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
622 TASK_UNINTERRUPTIBLE);
626 * find_get_page - find and get a page reference
627 * @mapping: the address_space to search
628 * @offset: the page index
630 * Is there a pagecache struct page at the given (mapping, offset) tuple?
631 * If yes, increment its refcount and return it; if no, return NULL.
633 struct page * find_get_page(struct address_space *mapping, pgoff_t offset)
637 read_lock_irq(&mapping->tree_lock);
638 page = radix_tree_lookup(&mapping->page_tree, offset);
640 page_cache_get(page);
641 read_unlock_irq(&mapping->tree_lock);
644 EXPORT_SYMBOL(find_get_page);
647 * find_lock_page - locate, pin and lock a pagecache page
648 * @mapping: the address_space to search
649 * @offset: the page index
651 * Locates the desired pagecache page, locks it, increments its reference
652 * count and returns its address.
654 * Returns zero if the page was not present. find_lock_page() may sleep.
656 struct page *find_lock_page(struct address_space *mapping,
662 read_lock_irq(&mapping->tree_lock);
663 page = radix_tree_lookup(&mapping->page_tree, offset);
665 page_cache_get(page);
666 if (TestSetPageLocked(page)) {
667 read_unlock_irq(&mapping->tree_lock);
670 /* Has the page been truncated while we slept? */
671 if (unlikely(page->mapping != mapping)) {
673 page_cache_release(page);
676 VM_BUG_ON(page->index != offset);
680 read_unlock_irq(&mapping->tree_lock);
684 EXPORT_SYMBOL(find_lock_page);
687 * find_or_create_page - locate or add a pagecache page
688 * @mapping: the page's address_space
689 * @index: the page's index into the mapping
690 * @gfp_mask: page allocation mode
692 * Locates a page in the pagecache. If the page is not present, a new page
693 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
694 * LRU list. The returned page is locked and has its reference count
697 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
700 * find_or_create_page() returns the desired page's address, or zero on
703 struct page *find_or_create_page(struct address_space *mapping,
704 pgoff_t index, gfp_t gfp_mask)
709 page = find_lock_page(mapping, index);
711 page = __page_cache_alloc(gfp_mask);
714 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
716 page_cache_release(page);
724 EXPORT_SYMBOL(find_or_create_page);
727 * find_get_pages - gang pagecache lookup
728 * @mapping: The address_space to search
729 * @start: The starting page index
730 * @nr_pages: The maximum number of pages
731 * @pages: Where the resulting pages are placed
733 * find_get_pages() will search for and return a group of up to
734 * @nr_pages pages in the mapping. The pages are placed at @pages.
735 * find_get_pages() takes a reference against the returned pages.
737 * The search returns a group of mapping-contiguous pages with ascending
738 * indexes. There may be holes in the indices due to not-present pages.
740 * find_get_pages() returns the number of pages which were found.
742 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
743 unsigned int nr_pages, struct page **pages)
748 read_lock_irq(&mapping->tree_lock);
749 ret = radix_tree_gang_lookup(&mapping->page_tree,
750 (void **)pages, start, nr_pages);
751 for (i = 0; i < ret; i++)
752 page_cache_get(pages[i]);
753 read_unlock_irq(&mapping->tree_lock);
758 * find_get_pages_contig - gang contiguous pagecache lookup
759 * @mapping: The address_space to search
760 * @index: The starting page index
761 * @nr_pages: The maximum number of pages
762 * @pages: Where the resulting pages are placed
764 * find_get_pages_contig() works exactly like find_get_pages(), except
765 * that the returned number of pages are guaranteed to be contiguous.
767 * find_get_pages_contig() returns the number of pages which were found.
769 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
770 unsigned int nr_pages, struct page **pages)
775 read_lock_irq(&mapping->tree_lock);
776 ret = radix_tree_gang_lookup(&mapping->page_tree,
777 (void **)pages, index, nr_pages);
778 for (i = 0; i < ret; i++) {
779 if (pages[i]->mapping == NULL || pages[i]->index != index)
782 page_cache_get(pages[i]);
785 read_unlock_irq(&mapping->tree_lock);
788 EXPORT_SYMBOL(find_get_pages_contig);
791 * find_get_pages_tag - find and return pages that match @tag
792 * @mapping: the address_space to search
793 * @index: the starting page index
794 * @tag: the tag index
795 * @nr_pages: the maximum number of pages
796 * @pages: where the resulting pages are placed
798 * Like find_get_pages, except we only return pages which are tagged with
799 * @tag. We update @index to index the next page for the traversal.
801 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
802 int tag, unsigned int nr_pages, struct page **pages)
807 read_lock_irq(&mapping->tree_lock);
808 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
809 (void **)pages, *index, nr_pages, tag);
810 for (i = 0; i < ret; i++)
811 page_cache_get(pages[i]);
813 *index = pages[ret - 1]->index + 1;
814 read_unlock_irq(&mapping->tree_lock);
817 EXPORT_SYMBOL(find_get_pages_tag);
820 * grab_cache_page_nowait - returns locked page at given index in given cache
821 * @mapping: target address_space
822 * @index: the page index
824 * Same as grab_cache_page(), but do not wait if the page is unavailable.
825 * This is intended for speculative data generators, where the data can
826 * be regenerated if the page couldn't be grabbed. This routine should
827 * be safe to call while holding the lock for another page.
829 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
830 * and deadlock against the caller's locked page.
833 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
835 struct page *page = find_get_page(mapping, index);
838 if (!TestSetPageLocked(page))
840 page_cache_release(page);
843 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
844 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
845 page_cache_release(page);
850 EXPORT_SYMBOL(grab_cache_page_nowait);
853 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
854 * a _large_ part of the i/o request. Imagine the worst scenario:
856 * ---R__________________________________________B__________
857 * ^ reading here ^ bad block(assume 4k)
859 * read(R) => miss => readahead(R...B) => media error => frustrating retries
860 * => failing the whole request => read(R) => read(R+1) =>
861 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
862 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
863 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
865 * It is going insane. Fix it by quickly scaling down the readahead size.
867 static void shrink_readahead_size_eio(struct file *filp,
868 struct file_ra_state *ra)
877 * do_generic_mapping_read - generic file read routine
878 * @mapping: address_space to be read
879 * @ra: file's readahead state
880 * @filp: the file to read
881 * @ppos: current file position
882 * @desc: read_descriptor
883 * @actor: read method
885 * This is a generic file read routine, and uses the
886 * mapping->a_ops->readpage() function for the actual low-level stuff.
888 * This is really ugly. But the goto's actually try to clarify some
889 * of the logic when it comes to error handling etc.
891 * Note the struct file* is only passed for the use of readpage.
894 void do_generic_mapping_read(struct address_space *mapping,
895 struct file_ra_state *ra,
898 read_descriptor_t *desc,
901 struct inode *inode = mapping->host;
905 unsigned long offset; /* offset into pagecache page */
906 unsigned int prev_offset;
909 index = *ppos >> PAGE_CACHE_SHIFT;
910 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
911 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
912 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
913 offset = *ppos & ~PAGE_CACHE_MASK;
919 unsigned long nr, ret;
923 page = find_get_page(mapping, index);
925 page_cache_sync_readahead(mapping,
927 index, last_index - index);
928 page = find_get_page(mapping, index);
929 if (unlikely(page == NULL))
932 if (PageReadahead(page)) {
933 page_cache_async_readahead(mapping,
935 index, last_index - index);
937 if (!PageUptodate(page))
938 goto page_not_up_to_date;
941 * i_size must be checked after we know the page is Uptodate.
943 * Checking i_size after the check allows us to calculate
944 * the correct value for "nr", which means the zero-filled
945 * part of the page is not copied back to userspace (unless
946 * another truncate extends the file - this is desired though).
949 isize = i_size_read(inode);
950 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
951 if (unlikely(!isize || index > end_index)) {
952 page_cache_release(page);
956 /* nr is the maximum number of bytes to copy from this page */
957 nr = PAGE_CACHE_SIZE;
958 if (index == end_index) {
959 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
961 page_cache_release(page);
967 /* If users can be writing to this page using arbitrary
968 * virtual addresses, take care about potential aliasing
969 * before reading the page on the kernel side.
971 if (mapping_writably_mapped(mapping))
972 flush_dcache_page(page);
975 * When a sequential read accesses a page several times,
976 * only mark it as accessed the first time.
978 if (prev_index != index || offset != prev_offset)
979 mark_page_accessed(page);
983 * Ok, we have the page, and it's up-to-date, so
984 * now we can copy it to user space...
986 * The actor routine returns how many bytes were actually used..
987 * NOTE! This may not be the same as how much of a user buffer
988 * we filled up (we may be padding etc), so we can only update
989 * "pos" here (the actor routine has to update the user buffer
990 * pointers and the remaining count).
992 ret = actor(desc, page, offset, nr);
994 index += offset >> PAGE_CACHE_SHIFT;
995 offset &= ~PAGE_CACHE_MASK;
996 prev_offset = offset;
998 page_cache_release(page);
999 if (ret == nr && desc->count)
1003 page_not_up_to_date:
1004 /* Get exclusive access to the page ... */
1005 if (lock_page_killable(page))
1008 /* Did it get truncated before we got the lock? */
1009 if (!page->mapping) {
1011 page_cache_release(page);
1015 /* Did somebody else fill it already? */
1016 if (PageUptodate(page)) {
1022 /* Start the actual read. The read will unlock the page. */
1023 error = mapping->a_ops->readpage(filp, page);
1025 if (unlikely(error)) {
1026 if (error == AOP_TRUNCATED_PAGE) {
1027 page_cache_release(page);
1030 goto readpage_error;
1033 if (!PageUptodate(page)) {
1034 if (lock_page_killable(page))
1036 if (!PageUptodate(page)) {
1037 if (page->mapping == NULL) {
1039 * invalidate_inode_pages got it
1042 page_cache_release(page);
1046 shrink_readahead_size_eio(filp, ra);
1057 /* UHHUH! A synchronous read error occurred. Report it */
1058 desc->error = error;
1059 page_cache_release(page);
1064 * Ok, it wasn't cached, so we need to create a new
1067 page = page_cache_alloc_cold(mapping);
1069 desc->error = -ENOMEM;
1072 error = add_to_page_cache_lru(page, mapping,
1075 page_cache_release(page);
1076 if (error == -EEXIST)
1078 desc->error = error;
1085 ra->prev_pos = prev_index;
1086 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1087 ra->prev_pos |= prev_offset;
1089 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1091 file_accessed(filp);
1093 EXPORT_SYMBOL(do_generic_mapping_read);
1095 int file_read_actor(read_descriptor_t *desc, struct page *page,
1096 unsigned long offset, unsigned long size)
1099 unsigned long left, count = desc->count;
1105 * Faults on the destination of a read are common, so do it before
1108 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1109 kaddr = kmap_atomic(page, KM_USER0);
1110 left = __copy_to_user_inatomic(desc->arg.buf,
1111 kaddr + offset, size);
1112 kunmap_atomic(kaddr, KM_USER0);
1117 /* Do it the slow way */
1119 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1124 desc->error = -EFAULT;
1127 desc->count = count - size;
1128 desc->written += size;
1129 desc->arg.buf += size;
1134 * Performs necessary checks before doing a write
1135 * @iov: io vector request
1136 * @nr_segs: number of segments in the iovec
1137 * @count: number of bytes to write
1138 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1140 * Adjust number of segments and amount of bytes to write (nr_segs should be
1141 * properly initialized first). Returns appropriate error code that caller
1142 * should return or zero in case that write should be allowed.
1144 int generic_segment_checks(const struct iovec *iov,
1145 unsigned long *nr_segs, size_t *count, int access_flags)
1149 for (seg = 0; seg < *nr_segs; seg++) {
1150 const struct iovec *iv = &iov[seg];
1153 * If any segment has a negative length, or the cumulative
1154 * length ever wraps negative then return -EINVAL.
1157 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1159 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1164 cnt -= iv->iov_len; /* This segment is no good */
1170 EXPORT_SYMBOL(generic_segment_checks);
1173 * generic_file_aio_read - generic filesystem read routine
1174 * @iocb: kernel I/O control block
1175 * @iov: io vector request
1176 * @nr_segs: number of segments in the iovec
1177 * @pos: current file position
1179 * This is the "read()" routine for all filesystems
1180 * that can use the page cache directly.
1183 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1184 unsigned long nr_segs, loff_t pos)
1186 struct file *filp = iocb->ki_filp;
1190 loff_t *ppos = &iocb->ki_pos;
1193 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1197 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1198 if (filp->f_flags & O_DIRECT) {
1200 struct address_space *mapping;
1201 struct inode *inode;
1203 mapping = filp->f_mapping;
1204 inode = mapping->host;
1207 goto out; /* skip atime */
1208 size = i_size_read(inode);
1210 retval = generic_file_direct_IO(READ, iocb,
1213 *ppos = pos + retval;
1215 if (likely(retval != 0)) {
1216 file_accessed(filp);
1223 for (seg = 0; seg < nr_segs; seg++) {
1224 read_descriptor_t desc;
1227 desc.arg.buf = iov[seg].iov_base;
1228 desc.count = iov[seg].iov_len;
1229 if (desc.count == 0)
1232 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1233 retval += desc.written;
1235 retval = retval ?: desc.error;
1245 EXPORT_SYMBOL(generic_file_aio_read);
1248 do_readahead(struct address_space *mapping, struct file *filp,
1249 pgoff_t index, unsigned long nr)
1251 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1254 force_page_cache_readahead(mapping, filp, index,
1255 max_sane_readahead(nr));
1259 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1267 if (file->f_mode & FMODE_READ) {
1268 struct address_space *mapping = file->f_mapping;
1269 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1270 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1271 unsigned long len = end - start + 1;
1272 ret = do_readahead(mapping, file, start, len);
1281 * page_cache_read - adds requested page to the page cache if not already there
1282 * @file: file to read
1283 * @offset: page index
1285 * This adds the requested page to the page cache if it isn't already there,
1286 * and schedules an I/O to read in its contents from disk.
1288 static int page_cache_read(struct file *file, pgoff_t offset)
1290 struct address_space *mapping = file->f_mapping;
1295 page = page_cache_alloc_cold(mapping);
1299 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1301 ret = mapping->a_ops->readpage(file, page);
1302 else if (ret == -EEXIST)
1303 ret = 0; /* losing race to add is OK */
1305 page_cache_release(page);
1307 } while (ret == AOP_TRUNCATED_PAGE);
1312 #define MMAP_LOTSAMISS (100)
1315 * filemap_fault - read in file data for page fault handling
1316 * @vma: vma in which the fault was taken
1317 * @vmf: struct vm_fault containing details of the fault
1319 * filemap_fault() is invoked via the vma operations vector for a
1320 * mapped memory region to read in file data during a page fault.
1322 * The goto's are kind of ugly, but this streamlines the normal case of having
1323 * it in the page cache, and handles the special cases reasonably without
1324 * having a lot of duplicated code.
1326 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1329 struct file *file = vma->vm_file;
1330 struct address_space *mapping = file->f_mapping;
1331 struct file_ra_state *ra = &file->f_ra;
1332 struct inode *inode = mapping->host;
1335 int did_readaround = 0;
1338 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1339 if (vmf->pgoff >= size)
1340 return VM_FAULT_SIGBUS;
1342 /* If we don't want any read-ahead, don't bother */
1343 if (VM_RandomReadHint(vma))
1344 goto no_cached_page;
1347 * Do we have something in the page cache already?
1350 page = find_lock_page(mapping, vmf->pgoff);
1352 * For sequential accesses, we use the generic readahead logic.
1354 if (VM_SequentialReadHint(vma)) {
1356 page_cache_sync_readahead(mapping, ra, file,
1358 page = find_lock_page(mapping, vmf->pgoff);
1360 goto no_cached_page;
1362 if (PageReadahead(page)) {
1363 page_cache_async_readahead(mapping, ra, file, page,
1369 unsigned long ra_pages;
1374 * Do we miss much more than hit in this file? If so,
1375 * stop bothering with read-ahead. It will only hurt.
1377 if (ra->mmap_miss > MMAP_LOTSAMISS)
1378 goto no_cached_page;
1381 * To keep the pgmajfault counter straight, we need to
1382 * check did_readaround, as this is an inner loop.
1384 if (!did_readaround) {
1385 ret = VM_FAULT_MAJOR;
1386 count_vm_event(PGMAJFAULT);
1389 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1393 if (vmf->pgoff > ra_pages / 2)
1394 start = vmf->pgoff - ra_pages / 2;
1395 do_page_cache_readahead(mapping, file, start, ra_pages);
1397 page = find_lock_page(mapping, vmf->pgoff);
1399 goto no_cached_page;
1402 if (!did_readaround)
1406 * We have a locked page in the page cache, now we need to check
1407 * that it's up-to-date. If not, it is going to be due to an error.
1409 if (unlikely(!PageUptodate(page)))
1410 goto page_not_uptodate;
1412 /* Must recheck i_size under page lock */
1413 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1414 if (unlikely(vmf->pgoff >= size)) {
1416 page_cache_release(page);
1417 return VM_FAULT_SIGBUS;
1421 * Found the page and have a reference on it.
1423 mark_page_accessed(page);
1424 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
1426 return ret | VM_FAULT_LOCKED;
1430 * We're only likely to ever get here if MADV_RANDOM is in
1433 error = page_cache_read(file, vmf->pgoff);
1436 * The page we want has now been added to the page cache.
1437 * In the unlikely event that someone removed it in the
1438 * meantime, we'll just come back here and read it again.
1444 * An error return from page_cache_read can result if the
1445 * system is low on memory, or a problem occurs while trying
1448 if (error == -ENOMEM)
1449 return VM_FAULT_OOM;
1450 return VM_FAULT_SIGBUS;
1454 if (!did_readaround) {
1455 ret = VM_FAULT_MAJOR;
1456 count_vm_event(PGMAJFAULT);
1460 * Umm, take care of errors if the page isn't up-to-date.
1461 * Try to re-read it _once_. We do this synchronously,
1462 * because there really aren't any performance issues here
1463 * and we need to check for errors.
1465 ClearPageError(page);
1466 error = mapping->a_ops->readpage(file, page);
1467 page_cache_release(page);
1469 if (!error || error == AOP_TRUNCATED_PAGE)
1472 /* Things didn't work out. Return zero to tell the mm layer so. */
1473 shrink_readahead_size_eio(file, ra);
1474 return VM_FAULT_SIGBUS;
1476 EXPORT_SYMBOL(filemap_fault);
1478 struct vm_operations_struct generic_file_vm_ops = {
1479 .fault = filemap_fault,
1482 /* This is used for a general mmap of a disk file */
1484 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1486 struct address_space *mapping = file->f_mapping;
1488 if (!mapping->a_ops->readpage)
1490 file_accessed(file);
1491 vma->vm_ops = &generic_file_vm_ops;
1492 vma->vm_flags |= VM_CAN_NONLINEAR;
1497 * This is for filesystems which do not implement ->writepage.
1499 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1501 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1503 return generic_file_mmap(file, vma);
1506 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1510 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1514 #endif /* CONFIG_MMU */
1516 EXPORT_SYMBOL(generic_file_mmap);
1517 EXPORT_SYMBOL(generic_file_readonly_mmap);
1519 static struct page *__read_cache_page(struct address_space *mapping,
1521 int (*filler)(void *,struct page*),
1527 page = find_get_page(mapping, index);
1529 page = page_cache_alloc_cold(mapping);
1531 return ERR_PTR(-ENOMEM);
1532 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1533 if (unlikely(err)) {
1534 page_cache_release(page);
1537 /* Presumably ENOMEM for radix tree node */
1538 return ERR_PTR(err);
1540 err = filler(data, page);
1542 page_cache_release(page);
1543 page = ERR_PTR(err);
1550 * Same as read_cache_page, but don't wait for page to become unlocked
1551 * after submitting it to the filler.
1553 struct page *read_cache_page_async(struct address_space *mapping,
1555 int (*filler)(void *,struct page*),
1562 page = __read_cache_page(mapping, index, filler, data);
1565 if (PageUptodate(page))
1569 if (!page->mapping) {
1571 page_cache_release(page);
1574 if (PageUptodate(page)) {
1578 err = filler(data, page);
1580 page_cache_release(page);
1581 return ERR_PTR(err);
1584 mark_page_accessed(page);
1587 EXPORT_SYMBOL(read_cache_page_async);
1590 * read_cache_page - read into page cache, fill it if needed
1591 * @mapping: the page's address_space
1592 * @index: the page index
1593 * @filler: function to perform the read
1594 * @data: destination for read data
1596 * Read into the page cache. If a page already exists, and PageUptodate() is
1597 * not set, try to fill the page then wait for it to become unlocked.
1599 * If the page does not get brought uptodate, return -EIO.
1601 struct page *read_cache_page(struct address_space *mapping,
1603 int (*filler)(void *,struct page*),
1608 page = read_cache_page_async(mapping, index, filler, data);
1611 wait_on_page_locked(page);
1612 if (!PageUptodate(page)) {
1613 page_cache_release(page);
1614 page = ERR_PTR(-EIO);
1619 EXPORT_SYMBOL(read_cache_page);
1622 * The logic we want is
1624 * if suid or (sgid and xgrp)
1627 int should_remove_suid(struct dentry *dentry)
1629 mode_t mode = dentry->d_inode->i_mode;
1632 /* suid always must be killed */
1633 if (unlikely(mode & S_ISUID))
1634 kill = ATTR_KILL_SUID;
1637 * sgid without any exec bits is just a mandatory locking mark; leave
1638 * it alone. If some exec bits are set, it's a real sgid; kill it.
1640 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1641 kill |= ATTR_KILL_SGID;
1643 if (unlikely(kill && !capable(CAP_FSETID)))
1648 EXPORT_SYMBOL(should_remove_suid);
1650 int __remove_suid(struct dentry *dentry, int kill)
1652 struct iattr newattrs;
1654 newattrs.ia_valid = ATTR_FORCE | kill;
1655 return notify_change(dentry, &newattrs);
1658 int remove_suid(struct dentry *dentry)
1660 int killsuid = should_remove_suid(dentry);
1661 int killpriv = security_inode_need_killpriv(dentry);
1667 error = security_inode_killpriv(dentry);
1668 if (!error && killsuid)
1669 error = __remove_suid(dentry, killsuid);
1673 EXPORT_SYMBOL(remove_suid);
1675 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1676 const struct iovec *iov, size_t base, size_t bytes)
1678 size_t copied = 0, left = 0;
1681 char __user *buf = iov->iov_base + base;
1682 int copy = min(bytes, iov->iov_len - base);
1685 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1694 return copied - left;
1698 * Copy as much as we can into the page and return the number of bytes which
1699 * were sucessfully copied. If a fault is encountered then return the number of
1700 * bytes which were copied.
1702 size_t iov_iter_copy_from_user_atomic(struct page *page,
1703 struct iov_iter *i, unsigned long offset, size_t bytes)
1708 BUG_ON(!in_atomic());
1709 kaddr = kmap_atomic(page, KM_USER0);
1710 if (likely(i->nr_segs == 1)) {
1712 char __user *buf = i->iov->iov_base + i->iov_offset;
1713 left = __copy_from_user_inatomic_nocache(kaddr + offset,
1715 copied = bytes - left;
1717 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1718 i->iov, i->iov_offset, bytes);
1720 kunmap_atomic(kaddr, KM_USER0);
1724 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1727 * This has the same sideeffects and return value as
1728 * iov_iter_copy_from_user_atomic().
1729 * The difference is that it attempts to resolve faults.
1730 * Page must not be locked.
1732 size_t iov_iter_copy_from_user(struct page *page,
1733 struct iov_iter *i, unsigned long offset, size_t bytes)
1739 if (likely(i->nr_segs == 1)) {
1741 char __user *buf = i->iov->iov_base + i->iov_offset;
1742 left = __copy_from_user_nocache(kaddr + offset, buf, bytes);
1743 copied = bytes - left;
1745 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1746 i->iov, i->iov_offset, bytes);
1751 EXPORT_SYMBOL(iov_iter_copy_from_user);
1753 static void __iov_iter_advance_iov(struct iov_iter *i, size_t bytes)
1755 if (likely(i->nr_segs == 1)) {
1756 i->iov_offset += bytes;
1758 const struct iovec *iov = i->iov;
1759 size_t base = i->iov_offset;
1762 * The !iov->iov_len check ensures we skip over unlikely
1763 * zero-length segments.
1765 while (bytes || !iov->iov_len) {
1766 int copy = min(bytes, iov->iov_len - base);
1770 if (iov->iov_len == base) {
1776 i->iov_offset = base;
1780 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1782 BUG_ON(i->count < bytes);
1784 __iov_iter_advance_iov(i, bytes);
1787 EXPORT_SYMBOL(iov_iter_advance);
1790 * Fault in the first iovec of the given iov_iter, to a maximum length
1791 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1792 * accessed (ie. because it is an invalid address).
1794 * writev-intensive code may want this to prefault several iovecs -- that
1795 * would be possible (callers must not rely on the fact that _only_ the
1796 * first iovec will be faulted with the current implementation).
1798 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
1800 char __user *buf = i->iov->iov_base + i->iov_offset;
1801 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1802 return fault_in_pages_readable(buf, bytes);
1804 EXPORT_SYMBOL(iov_iter_fault_in_readable);
1807 * Return the count of just the current iov_iter segment.
1809 size_t iov_iter_single_seg_count(struct iov_iter *i)
1811 const struct iovec *iov = i->iov;
1812 if (i->nr_segs == 1)
1815 return min(i->count, iov->iov_len - i->iov_offset);
1817 EXPORT_SYMBOL(iov_iter_single_seg_count);
1820 * Performs necessary checks before doing a write
1822 * Can adjust writing position or amount of bytes to write.
1823 * Returns appropriate error code that caller should return or
1824 * zero in case that write should be allowed.
1826 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1828 struct inode *inode = file->f_mapping->host;
1829 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1831 if (unlikely(*pos < 0))
1835 /* FIXME: this is for backwards compatibility with 2.4 */
1836 if (file->f_flags & O_APPEND)
1837 *pos = i_size_read(inode);
1839 if (limit != RLIM_INFINITY) {
1840 if (*pos >= limit) {
1841 send_sig(SIGXFSZ, current, 0);
1844 if (*count > limit - (typeof(limit))*pos) {
1845 *count = limit - (typeof(limit))*pos;
1853 if (unlikely(*pos + *count > MAX_NON_LFS &&
1854 !(file->f_flags & O_LARGEFILE))) {
1855 if (*pos >= MAX_NON_LFS) {
1858 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1859 *count = MAX_NON_LFS - (unsigned long)*pos;
1864 * Are we about to exceed the fs block limit ?
1866 * If we have written data it becomes a short write. If we have
1867 * exceeded without writing data we send a signal and return EFBIG.
1868 * Linus frestrict idea will clean these up nicely..
1870 if (likely(!isblk)) {
1871 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1872 if (*count || *pos > inode->i_sb->s_maxbytes) {
1875 /* zero-length writes at ->s_maxbytes are OK */
1878 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1879 *count = inode->i_sb->s_maxbytes - *pos;
1883 if (bdev_read_only(I_BDEV(inode)))
1885 isize = i_size_read(inode);
1886 if (*pos >= isize) {
1887 if (*count || *pos > isize)
1891 if (*pos + *count > isize)
1892 *count = isize - *pos;
1899 EXPORT_SYMBOL(generic_write_checks);
1901 int pagecache_write_begin(struct file *file, struct address_space *mapping,
1902 loff_t pos, unsigned len, unsigned flags,
1903 struct page **pagep, void **fsdata)
1905 const struct address_space_operations *aops = mapping->a_ops;
1907 if (aops->write_begin) {
1908 return aops->write_begin(file, mapping, pos, len, flags,
1912 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1913 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1914 struct inode *inode = mapping->host;
1917 page = __grab_cache_page(mapping, index);
1922 if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) {
1924 * There is no way to resolve a short write situation
1925 * for a !Uptodate page (except by double copying in
1926 * the caller done by generic_perform_write_2copy).
1928 * Instead, we have to bring it uptodate here.
1930 ret = aops->readpage(file, page);
1931 page_cache_release(page);
1933 if (ret == AOP_TRUNCATED_PAGE)
1940 ret = aops->prepare_write(file, page, offset, offset+len);
1943 page_cache_release(page);
1944 if (pos + len > inode->i_size)
1945 vmtruncate(inode, inode->i_size);
1950 EXPORT_SYMBOL(pagecache_write_begin);
1952 int pagecache_write_end(struct file *file, struct address_space *mapping,
1953 loff_t pos, unsigned len, unsigned copied,
1954 struct page *page, void *fsdata)
1956 const struct address_space_operations *aops = mapping->a_ops;
1959 if (aops->write_end) {
1960 mark_page_accessed(page);
1961 ret = aops->write_end(file, mapping, pos, len, copied,
1964 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1965 struct inode *inode = mapping->host;
1967 flush_dcache_page(page);
1968 ret = aops->commit_write(file, page, offset, offset+len);
1970 mark_page_accessed(page);
1971 page_cache_release(page);
1974 if (pos + len > inode->i_size)
1975 vmtruncate(inode, inode->i_size);
1977 ret = min_t(size_t, copied, ret);
1984 EXPORT_SYMBOL(pagecache_write_end);
1987 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1988 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1989 size_t count, size_t ocount)
1991 struct file *file = iocb->ki_filp;
1992 struct address_space *mapping = file->f_mapping;
1993 struct inode *inode = mapping->host;
1996 if (count != ocount)
1997 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1999 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2001 loff_t end = pos + written;
2002 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2003 i_size_write(inode, end);
2004 mark_inode_dirty(inode);
2010 * Sync the fs metadata but not the minor inode changes and
2011 * of course not the data as we did direct DMA for the IO.
2012 * i_mutex is held, which protects generic_osync_inode() from
2013 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
2015 if ((written >= 0 || written == -EIOCBQUEUED) &&
2016 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2017 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2023 EXPORT_SYMBOL(generic_file_direct_write);
2026 * Find or create a page at the given pagecache position. Return the locked
2027 * page. This function is specifically for buffered writes.
2029 struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index)
2034 page = find_lock_page(mapping, index);
2038 page = page_cache_alloc(mapping);
2041 status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
2042 if (unlikely(status)) {
2043 page_cache_release(page);
2044 if (status == -EEXIST)
2050 EXPORT_SYMBOL(__grab_cache_page);
2052 static ssize_t generic_perform_write_2copy(struct file *file,
2053 struct iov_iter *i, loff_t pos)
2055 struct address_space *mapping = file->f_mapping;
2056 const struct address_space_operations *a_ops = mapping->a_ops;
2057 struct inode *inode = mapping->host;
2059 ssize_t written = 0;
2062 struct page *src_page;
2064 pgoff_t index; /* Pagecache index for current page */
2065 unsigned long offset; /* Offset into pagecache page */
2066 unsigned long bytes; /* Bytes to write to page */
2067 size_t copied; /* Bytes copied from user */
2069 offset = (pos & (PAGE_CACHE_SIZE - 1));
2070 index = pos >> PAGE_CACHE_SHIFT;
2071 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2075 * a non-NULL src_page indicates that we're doing the
2076 * copy via get_user_pages and kmap.
2081 * Bring in the user page that we will copy from _first_.
2082 * Otherwise there's a nasty deadlock on copying from the
2083 * same page as we're writing to, without it being marked
2086 * Not only is this an optimisation, but it is also required
2087 * to check that the address is actually valid, when atomic
2088 * usercopies are used, below.
2090 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2095 page = __grab_cache_page(mapping, index);
2102 * non-uptodate pages cannot cope with short copies, and we
2103 * cannot take a pagefault with the destination page locked.
2104 * So pin the source page to copy it.
2106 if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) {
2109 src_page = alloc_page(GFP_KERNEL);
2111 page_cache_release(page);
2117 * Cannot get_user_pages with a page locked for the
2118 * same reason as we can't take a page fault with a
2119 * page locked (as explained below).
2121 copied = iov_iter_copy_from_user(src_page, i,
2123 if (unlikely(copied == 0)) {
2125 page_cache_release(page);
2126 page_cache_release(src_page);
2133 * Can't handle the page going uptodate here, because
2134 * that means we would use non-atomic usercopies, which
2135 * zero out the tail of the page, which can cause
2136 * zeroes to become transiently visible. We could just
2137 * use a non-zeroing copy, but the APIs aren't too
2140 if (unlikely(!page->mapping || PageUptodate(page))) {
2142 page_cache_release(page);
2143 page_cache_release(src_page);
2148 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2149 if (unlikely(status))
2150 goto fs_write_aop_error;
2154 * Must not enter the pagefault handler here, because
2155 * we hold the page lock, so we might recursively
2156 * deadlock on the same lock, or get an ABBA deadlock
2157 * against a different lock, or against the mmap_sem
2158 * (which nests outside the page lock). So increment
2159 * preempt count, and use _atomic usercopies.
2161 * The page is uptodate so we are OK to encounter a
2162 * short copy: if unmodified parts of the page are
2163 * marked dirty and written out to disk, it doesn't
2166 pagefault_disable();
2167 copied = iov_iter_copy_from_user_atomic(page, i,
2172 src = kmap_atomic(src_page, KM_USER0);
2173 dst = kmap_atomic(page, KM_USER1);
2174 memcpy(dst + offset, src + offset, bytes);
2175 kunmap_atomic(dst, KM_USER1);
2176 kunmap_atomic(src, KM_USER0);
2179 flush_dcache_page(page);
2181 status = a_ops->commit_write(file, page, offset, offset+bytes);
2182 if (unlikely(status < 0))
2183 goto fs_write_aop_error;
2184 if (unlikely(status > 0)) /* filesystem did partial write */
2185 copied = min_t(size_t, copied, status);
2188 mark_page_accessed(page);
2189 page_cache_release(page);
2191 page_cache_release(src_page);
2193 iov_iter_advance(i, copied);
2197 balance_dirty_pages_ratelimited(mapping);
2203 page_cache_release(page);
2205 page_cache_release(src_page);
2208 * prepare_write() may have instantiated a few blocks
2209 * outside i_size. Trim these off again. Don't need
2210 * i_size_read because we hold i_mutex.
2212 if (pos + bytes > inode->i_size)
2213 vmtruncate(inode, inode->i_size);
2215 } while (iov_iter_count(i));
2217 return written ? written : status;
2220 static ssize_t generic_perform_write(struct file *file,
2221 struct iov_iter *i, loff_t pos)
2223 struct address_space *mapping = file->f_mapping;
2224 const struct address_space_operations *a_ops = mapping->a_ops;
2226 ssize_t written = 0;
2227 unsigned int flags = 0;
2230 * Copies from kernel address space cannot fail (NFSD is a big user).
2232 if (segment_eq(get_fs(), KERNEL_DS))
2233 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2237 pgoff_t index; /* Pagecache index for current page */
2238 unsigned long offset; /* Offset into pagecache page */
2239 unsigned long bytes; /* Bytes to write to page */
2240 size_t copied; /* Bytes copied from user */
2243 offset = (pos & (PAGE_CACHE_SIZE - 1));
2244 index = pos >> PAGE_CACHE_SHIFT;
2245 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2251 * Bring in the user page that we will copy from _first_.
2252 * Otherwise there's a nasty deadlock on copying from the
2253 * same page as we're writing to, without it being marked
2256 * Not only is this an optimisation, but it is also required
2257 * to check that the address is actually valid, when atomic
2258 * usercopies are used, below.
2260 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2265 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2267 if (unlikely(status))
2270 pagefault_disable();
2271 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2273 flush_dcache_page(page);
2275 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2277 if (unlikely(status < 0))
2283 iov_iter_advance(i, copied);
2284 if (unlikely(copied == 0)) {
2286 * If we were unable to copy any data at all, we must
2287 * fall back to a single segment length write.
2289 * If we didn't fallback here, we could livelock
2290 * because not all segments in the iov can be copied at
2291 * once without a pagefault.
2293 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2294 iov_iter_single_seg_count(i));
2300 balance_dirty_pages_ratelimited(mapping);
2302 } while (iov_iter_count(i));
2304 return written ? written : status;
2308 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2309 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2310 size_t count, ssize_t written)
2312 struct file *file = iocb->ki_filp;
2313 struct address_space *mapping = file->f_mapping;
2314 const struct address_space_operations *a_ops = mapping->a_ops;
2315 struct inode *inode = mapping->host;
2319 iov_iter_init(&i, iov, nr_segs, count, written);
2320 if (a_ops->write_begin)
2321 status = generic_perform_write(file, &i, pos);
2323 status = generic_perform_write_2copy(file, &i, pos);
2325 if (likely(status >= 0)) {
2327 *ppos = pos + status;
2330 * For now, when the user asks for O_SYNC, we'll actually give
2333 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2334 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2335 status = generic_osync_inode(inode, mapping,
2336 OSYNC_METADATA|OSYNC_DATA);
2341 * If we get here for O_DIRECT writes then we must have fallen through
2342 * to buffered writes (block instantiation inside i_size). So we sync
2343 * the file data here, to try to honour O_DIRECT expectations.
2345 if (unlikely(file->f_flags & O_DIRECT) && written)
2346 status = filemap_write_and_wait(mapping);
2348 return written ? written : status;
2350 EXPORT_SYMBOL(generic_file_buffered_write);
2353 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2354 unsigned long nr_segs, loff_t *ppos)
2356 struct file *file = iocb->ki_filp;
2357 struct address_space * mapping = file->f_mapping;
2358 size_t ocount; /* original count */
2359 size_t count; /* after file limit checks */
2360 struct inode *inode = mapping->host;
2366 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2373 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2375 /* We can write back this queue in page reclaim */
2376 current->backing_dev_info = mapping->backing_dev_info;
2379 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2386 err = remove_suid(file->f_path.dentry);
2390 file_update_time(file);
2392 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2393 if (unlikely(file->f_flags & O_DIRECT)) {
2395 ssize_t written_buffered;
2397 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2398 ppos, count, ocount);
2399 if (written < 0 || written == count)
2402 * direct-io write to a hole: fall through to buffered I/O
2403 * for completing the rest of the request.
2407 written_buffered = generic_file_buffered_write(iocb, iov,
2408 nr_segs, pos, ppos, count,
2411 * If generic_file_buffered_write() retuned a synchronous error
2412 * then we want to return the number of bytes which were
2413 * direct-written, or the error code if that was zero. Note
2414 * that this differs from normal direct-io semantics, which
2415 * will return -EFOO even if some bytes were written.
2417 if (written_buffered < 0) {
2418 err = written_buffered;
2423 * We need to ensure that the page cache pages are written to
2424 * disk and invalidated to preserve the expected O_DIRECT
2427 endbyte = pos + written_buffered - written - 1;
2428 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2429 SYNC_FILE_RANGE_WAIT_BEFORE|
2430 SYNC_FILE_RANGE_WRITE|
2431 SYNC_FILE_RANGE_WAIT_AFTER);
2433 written = written_buffered;
2434 invalidate_mapping_pages(mapping,
2435 pos >> PAGE_CACHE_SHIFT,
2436 endbyte >> PAGE_CACHE_SHIFT);
2439 * We don't know how much we wrote, so just return
2440 * the number of bytes which were direct-written
2444 written = generic_file_buffered_write(iocb, iov, nr_segs,
2445 pos, ppos, count, written);
2448 current->backing_dev_info = NULL;
2449 return written ? written : err;
2452 ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2453 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
2455 struct file *file = iocb->ki_filp;
2456 struct address_space *mapping = file->f_mapping;
2457 struct inode *inode = mapping->host;
2460 BUG_ON(iocb->ki_pos != pos);
2462 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2465 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2468 err = sync_page_range_nolock(inode, mapping, pos, ret);
2474 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2476 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2477 unsigned long nr_segs, loff_t pos)
2479 struct file *file = iocb->ki_filp;
2480 struct address_space *mapping = file->f_mapping;
2481 struct inode *inode = mapping->host;
2484 BUG_ON(iocb->ki_pos != pos);
2486 mutex_lock(&inode->i_mutex);
2487 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2489 mutex_unlock(&inode->i_mutex);
2491 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2494 err = sync_page_range(inode, mapping, pos, ret);
2500 EXPORT_SYMBOL(generic_file_aio_write);
2503 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2504 * went wrong during pagecache shootdown.
2507 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2508 loff_t offset, unsigned long nr_segs)
2510 struct file *file = iocb->ki_filp;
2511 struct address_space *mapping = file->f_mapping;
2514 pgoff_t end = 0; /* silence gcc */
2517 * If it's a write, unmap all mmappings of the file up-front. This
2518 * will cause any pte dirty bits to be propagated into the pageframes
2519 * for the subsequent filemap_write_and_wait().
2522 write_len = iov_length(iov, nr_segs);
2523 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
2524 if (mapping_mapped(mapping))
2525 unmap_mapping_range(mapping, offset, write_len, 0);
2528 retval = filemap_write_and_wait(mapping);
2533 * After a write we want buffered reads to be sure to go to disk to get
2534 * the new data. We invalidate clean cached page from the region we're
2535 * about to write. We do this *before* the write so that we can return
2536 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2538 if (rw == WRITE && mapping->nrpages) {
2539 retval = invalidate_inode_pages2_range(mapping,
2540 offset >> PAGE_CACHE_SHIFT, end);
2545 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2548 * Finally, try again to invalidate clean pages which might have been
2549 * cached by non-direct readahead, or faulted in by get_user_pages()
2550 * if the source of the write was an mmap'ed region of the file
2551 * we're writing. Either one is a pretty crazy thing to do,
2552 * so we don't support it 100%. If this invalidation
2553 * fails, tough, the write still worked...
2555 if (rw == WRITE && mapping->nrpages) {
2556 invalidate_inode_pages2_range(mapping, offset >> PAGE_CACHE_SHIFT, end);
2563 * try_to_release_page() - release old fs-specific metadata on a page
2565 * @page: the page which the kernel is trying to free
2566 * @gfp_mask: memory allocation flags (and I/O mode)
2568 * The address_space is to try to release any data against the page
2569 * (presumably at page->private). If the release was successful, return `1'.
2570 * Otherwise return zero.
2572 * The @gfp_mask argument specifies whether I/O may be performed to release
2573 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2575 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2577 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2579 struct address_space * const mapping = page->mapping;
2581 BUG_ON(!PageLocked(page));
2582 if (PageWriteback(page))
2585 if (mapping && mapping->a_ops->releasepage)
2586 return mapping->a_ops->releasepage(page, gfp_mask);
2587 return try_to_free_buffers(page);
2590 EXPORT_SYMBOL(try_to_release_page);