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/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/cpuset.h>
37 * FIXME: remove all knowledge of the buffer layer from the core VM
39 #include <linux/buffer_head.h> /* for generic_osync_inode */
44 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
45 loff_t offset, unsigned long nr_segs);
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * Shared mappings now work. 15.8.1995 Bruno.
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
64 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
68 * ->i_mmap_lock (truncate->unmap_mapping_range)
72 * ->page_table_lock or pte_lock (various, mainly in memory.c)
73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 * ->lock_page (access_process_vm)
78 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
82 * ->i_alloc_sem (various)
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->dcache_lock (proc_pid_lookup)
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
115 void __remove_from_page_cache(struct page *page)
117 struct address_space *mapping = page->mapping;
119 radix_tree_delete(&mapping->page_tree, page->index);
120 page->mapping = NULL;
122 __dec_zone_page_state(page, NR_FILE_PAGES);
125 void remove_from_page_cache(struct page *page)
127 struct address_space *mapping = page->mapping;
129 BUG_ON(!PageLocked(page));
131 write_lock_irq(&mapping->tree_lock);
132 __remove_from_page_cache(page);
133 write_unlock_irq(&mapping->tree_lock);
136 static int sync_page(void *word)
138 struct address_space *mapping;
141 page = container_of((unsigned long *)word, struct page, flags);
144 * page_mapping() is being called without PG_locked held.
145 * Some knowledge of the state and use of the page is used to
146 * reduce the requirements down to a memory barrier.
147 * The danger here is of a stale page_mapping() return value
148 * indicating a struct address_space different from the one it's
149 * associated with when it is associated with one.
150 * After smp_mb(), it's either the correct page_mapping() for
151 * the page, or an old page_mapping() and the page's own
152 * page_mapping() has gone NULL.
153 * The ->sync_page() address_space operation must tolerate
154 * page_mapping() going NULL. By an amazing coincidence,
155 * this comes about because none of the users of the page
156 * in the ->sync_page() methods make essential use of the
157 * page_mapping(), merely passing the page down to the backing
158 * device's unplug functions when it's non-NULL, which in turn
159 * ignore it for all cases but swap, where only page_private(page) is
160 * of interest. When page_mapping() does go NULL, the entire
161 * call stack gracefully ignores the page and returns.
165 mapping = page_mapping(page);
166 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
167 mapping->a_ops->sync_page(page);
173 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
174 * @mapping: address space structure to write
175 * @start: offset in bytes where the range starts
176 * @end: offset in bytes where the range ends (inclusive)
177 * @sync_mode: enable synchronous operation
179 * Start writeback against all of a mapping's dirty pages that lie
180 * within the byte offsets <start, end> inclusive.
182 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
183 * opposed to a regular memory cleansing writeback. The difference between
184 * these two operations is that if a dirty page/buffer is encountered, it must
185 * be waited upon, and not just skipped over.
187 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
188 loff_t end, int sync_mode)
191 struct writeback_control wbc = {
192 .sync_mode = sync_mode,
193 .nr_to_write = mapping->nrpages * 2,
194 .range_start = start,
198 if (!mapping_cap_writeback_dirty(mapping))
201 ret = do_writepages(mapping, &wbc);
205 static inline int __filemap_fdatawrite(struct address_space *mapping,
208 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
211 int filemap_fdatawrite(struct address_space *mapping)
213 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
215 EXPORT_SYMBOL(filemap_fdatawrite);
217 static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
220 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
224 * filemap_flush - mostly a non-blocking flush
225 * @mapping: target address_space
227 * This is a mostly non-blocking flush. Not suitable for data-integrity
228 * purposes - I/O may not be started against all dirty pages.
230 int filemap_flush(struct address_space *mapping)
232 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
234 EXPORT_SYMBOL(filemap_flush);
237 * wait_on_page_writeback_range - wait for writeback to complete
238 * @mapping: target address_space
239 * @start: beginning page index
240 * @end: ending page index
242 * Wait for writeback to complete against pages indexed by start->end
245 int wait_on_page_writeback_range(struct address_space *mapping,
246 pgoff_t start, pgoff_t end)
256 pagevec_init(&pvec, 0);
258 while ((index <= end) &&
259 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
260 PAGECACHE_TAG_WRITEBACK,
261 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
264 for (i = 0; i < nr_pages; i++) {
265 struct page *page = pvec.pages[i];
267 /* until radix tree lookup accepts end_index */
268 if (page->index > end)
271 wait_on_page_writeback(page);
275 pagevec_release(&pvec);
279 /* Check for outstanding write errors */
280 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
282 if (test_and_clear_bit(AS_EIO, &mapping->flags))
289 * sync_page_range - write and wait on all pages in the passed range
290 * @inode: target inode
291 * @mapping: target address_space
292 * @pos: beginning offset in pages to write
293 * @count: number of bytes to write
295 * Write and wait upon all the pages in the passed range. This is a "data
296 * integrity" operation. It waits upon in-flight writeout before starting and
297 * waiting upon new writeout. If there was an IO error, return it.
299 * We need to re-take i_mutex during the generic_osync_inode list walk because
300 * it is otherwise livelockable.
302 int sync_page_range(struct inode *inode, struct address_space *mapping,
303 loff_t pos, loff_t count)
305 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
306 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
309 if (!mapping_cap_writeback_dirty(mapping) || !count)
311 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
313 mutex_lock(&inode->i_mutex);
314 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
315 mutex_unlock(&inode->i_mutex);
318 ret = wait_on_page_writeback_range(mapping, start, end);
321 EXPORT_SYMBOL(sync_page_range);
324 * sync_page_range_nolock
325 * @inode: target inode
326 * @mapping: target address_space
327 * @pos: beginning offset in pages to write
328 * @count: number of bytes to write
330 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
331 * as it forces O_SYNC writers to different parts of the same file
332 * to be serialised right until io completion.
334 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
335 loff_t pos, loff_t count)
337 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
338 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
341 if (!mapping_cap_writeback_dirty(mapping) || !count)
343 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
345 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
347 ret = wait_on_page_writeback_range(mapping, start, end);
350 EXPORT_SYMBOL(sync_page_range_nolock);
353 * filemap_fdatawait - wait for all under-writeback pages to complete
354 * @mapping: address space structure to wait for
356 * Walk the list of under-writeback pages of the given address space
357 * and wait for all of them.
359 int filemap_fdatawait(struct address_space *mapping)
361 loff_t i_size = i_size_read(mapping->host);
366 return wait_on_page_writeback_range(mapping, 0,
367 (i_size - 1) >> PAGE_CACHE_SHIFT);
369 EXPORT_SYMBOL(filemap_fdatawait);
371 int filemap_write_and_wait(struct address_space *mapping)
375 if (mapping->nrpages) {
376 err = filemap_fdatawrite(mapping);
378 * Even if the above returned error, the pages may be
379 * written partially (e.g. -ENOSPC), so we wait for it.
380 * But the -EIO is special case, it may indicate the worst
381 * thing (e.g. bug) happened, so we avoid waiting for it.
384 int err2 = filemap_fdatawait(mapping);
391 EXPORT_SYMBOL(filemap_write_and_wait);
394 * filemap_write_and_wait_range - write out & wait on a file range
395 * @mapping: the address_space for the pages
396 * @lstart: offset in bytes where the range starts
397 * @lend: offset in bytes where the range ends (inclusive)
399 * Write out and wait upon file offsets lstart->lend, inclusive.
401 * Note that `lend' is inclusive (describes the last byte to be written) so
402 * that this function can be used to write to the very end-of-file (end = -1).
404 int filemap_write_and_wait_range(struct address_space *mapping,
405 loff_t lstart, loff_t lend)
409 if (mapping->nrpages) {
410 err = __filemap_fdatawrite_range(mapping, lstart, lend,
412 /* See comment of filemap_write_and_wait() */
414 int err2 = wait_on_page_writeback_range(mapping,
415 lstart >> PAGE_CACHE_SHIFT,
416 lend >> PAGE_CACHE_SHIFT);
425 * add_to_page_cache - add newly allocated pagecache pages
427 * @mapping: the page's address_space
428 * @offset: page index
429 * @gfp_mask: page allocation mode
431 * This function is used to add newly allocated pagecache pages;
432 * the page is new, so we can just run SetPageLocked() against it.
433 * The other page state flags were set by rmqueue().
435 * This function does not add the page to the LRU. The caller must do that.
437 int add_to_page_cache(struct page *page, struct address_space *mapping,
438 pgoff_t offset, gfp_t gfp_mask)
440 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
443 write_lock_irq(&mapping->tree_lock);
444 error = radix_tree_insert(&mapping->page_tree, offset, page);
446 page_cache_get(page);
448 page->mapping = mapping;
449 page->index = offset;
451 __inc_zone_page_state(page, NR_FILE_PAGES);
453 write_unlock_irq(&mapping->tree_lock);
454 radix_tree_preload_end();
458 EXPORT_SYMBOL(add_to_page_cache);
460 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
461 pgoff_t offset, gfp_t gfp_mask)
463 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
470 struct page *__page_cache_alloc(gfp_t gfp)
472 if (cpuset_do_page_mem_spread()) {
473 int n = cpuset_mem_spread_node();
474 return alloc_pages_node(n, gfp, 0);
476 return alloc_pages(gfp, 0);
478 EXPORT_SYMBOL(__page_cache_alloc);
481 static int __sleep_on_page_lock(void *word)
488 * In order to wait for pages to become available there must be
489 * waitqueues associated with pages. By using a hash table of
490 * waitqueues where the bucket discipline is to maintain all
491 * waiters on the same queue and wake all when any of the pages
492 * become available, and for the woken contexts to check to be
493 * sure the appropriate page became available, this saves space
494 * at a cost of "thundering herd" phenomena during rare hash
497 static wait_queue_head_t *page_waitqueue(struct page *page)
499 const struct zone *zone = page_zone(page);
501 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
504 static inline void wake_up_page(struct page *page, int bit)
506 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
509 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
511 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
513 if (test_bit(bit_nr, &page->flags))
514 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
515 TASK_UNINTERRUPTIBLE);
517 EXPORT_SYMBOL(wait_on_page_bit);
519 void install_page_waitqueue_monitor(struct page *page, wait_queue_t *monitor)
521 wait_queue_head_t *q = page_waitqueue(page);
524 spin_lock_irqsave(&q->lock, flags);
525 __add_wait_queue(q, monitor);
526 spin_unlock_irqrestore(&q->lock, flags);
529 EXPORT_SYMBOL_GPL(install_page_waitqueue_monitor);
532 * unlock_page - unlock a locked page
535 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
536 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
537 * mechananism between PageLocked pages and PageWriteback pages is shared.
538 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
540 * The first mb is necessary to safely close the critical section opened by the
541 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
542 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
543 * parallel wait_on_page_locked()).
545 void fastcall unlock_page(struct page *page)
547 smp_mb__before_clear_bit();
548 if (!TestClearPageLocked(page))
550 smp_mb__after_clear_bit();
551 wake_up_page(page, PG_locked);
553 EXPORT_SYMBOL(unlock_page);
556 * end_page_writeback - end writeback against a page
559 void end_page_writeback(struct page *page)
561 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
562 if (!test_clear_page_writeback(page))
565 smp_mb__after_clear_bit();
566 wake_up_page(page, PG_writeback);
568 EXPORT_SYMBOL(end_page_writeback);
571 * __lock_page - get a lock on the page, assuming we need to sleep to get it
572 * @page: the page to lock
574 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
575 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
576 * chances are that on the second loop, the block layer's plug list is empty,
577 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
579 void fastcall __lock_page(struct page *page)
581 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
583 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
584 TASK_UNINTERRUPTIBLE);
586 EXPORT_SYMBOL(__lock_page);
589 * Variant of lock_page that does not require the caller to hold a reference
590 * on the page's mapping.
592 void fastcall __lock_page_nosync(struct page *page)
594 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
595 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
596 TASK_UNINTERRUPTIBLE);
600 * find_get_page - find and get a page reference
601 * @mapping: the address_space to search
602 * @offset: the page index
604 * Is there a pagecache struct page at the given (mapping, offset) tuple?
605 * If yes, increment its refcount and return it; if no, return NULL.
607 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
611 read_lock_irq(&mapping->tree_lock);
612 page = radix_tree_lookup(&mapping->page_tree, offset);
614 page_cache_get(page);
615 read_unlock_irq(&mapping->tree_lock);
618 EXPORT_SYMBOL(find_get_page);
621 * find_trylock_page - find and lock a page
622 * @mapping: the address_space to search
623 * @offset: the page index
625 * Same as find_get_page(), but trylock it instead of incrementing the count.
627 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
631 read_lock_irq(&mapping->tree_lock);
632 page = radix_tree_lookup(&mapping->page_tree, offset);
633 if (page && TestSetPageLocked(page))
635 read_unlock_irq(&mapping->tree_lock);
638 EXPORT_SYMBOL(find_trylock_page);
641 * find_lock_page - locate, pin and lock a pagecache page
642 * @mapping: the address_space to search
643 * @offset: the page index
645 * Locates the desired pagecache page, locks it, increments its reference
646 * count and returns its address.
648 * Returns zero if the page was not present. find_lock_page() may sleep.
650 struct page *find_lock_page(struct address_space *mapping,
651 unsigned long offset)
655 read_lock_irq(&mapping->tree_lock);
657 page = radix_tree_lookup(&mapping->page_tree, offset);
659 page_cache_get(page);
660 if (TestSetPageLocked(page)) {
661 read_unlock_irq(&mapping->tree_lock);
663 read_lock_irq(&mapping->tree_lock);
665 /* Has the page been truncated while we slept? */
666 if (unlikely(page->mapping != mapping ||
667 page->index != offset)) {
669 page_cache_release(page);
674 read_unlock_irq(&mapping->tree_lock);
677 EXPORT_SYMBOL(find_lock_page);
680 * find_or_create_page - locate or add a pagecache page
681 * @mapping: the page's address_space
682 * @index: the page's index into the mapping
683 * @gfp_mask: page allocation mode
685 * Locates a page in the pagecache. If the page is not present, a new page
686 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
687 * LRU list. The returned page is locked and has its reference count
690 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
693 * find_or_create_page() returns the desired page's address, or zero on
696 struct page *find_or_create_page(struct address_space *mapping,
697 unsigned long index, gfp_t gfp_mask)
699 struct page *page, *cached_page = NULL;
702 page = find_lock_page(mapping, index);
705 cached_page = alloc_page(gfp_mask);
709 err = add_to_page_cache_lru(cached_page, mapping,
714 } else if (err == -EEXIST)
718 page_cache_release(cached_page);
721 EXPORT_SYMBOL(find_or_create_page);
724 * find_get_pages - gang pagecache lookup
725 * @mapping: The address_space to search
726 * @start: The starting page index
727 * @nr_pages: The maximum number of pages
728 * @pages: Where the resulting pages are placed
730 * find_get_pages() will search for and return a group of up to
731 * @nr_pages pages in the mapping. The pages are placed at @pages.
732 * find_get_pages() takes a reference against the returned pages.
734 * The search returns a group of mapping-contiguous pages with ascending
735 * indexes. There may be holes in the indices due to not-present pages.
737 * find_get_pages() returns the number of pages which were found.
739 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
740 unsigned int nr_pages, struct page **pages)
745 read_lock_irq(&mapping->tree_lock);
746 ret = radix_tree_gang_lookup(&mapping->page_tree,
747 (void **)pages, start, nr_pages);
748 for (i = 0; i < ret; i++)
749 page_cache_get(pages[i]);
750 read_unlock_irq(&mapping->tree_lock);
755 * find_get_pages_contig - gang contiguous pagecache lookup
756 * @mapping: The address_space to search
757 * @index: The starting page index
758 * @nr_pages: The maximum number of pages
759 * @pages: Where the resulting pages are placed
761 * find_get_pages_contig() works exactly like find_get_pages(), except
762 * that the returned number of pages are guaranteed to be contiguous.
764 * find_get_pages_contig() returns the number of pages which were found.
766 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
767 unsigned int nr_pages, struct page **pages)
772 read_lock_irq(&mapping->tree_lock);
773 ret = radix_tree_gang_lookup(&mapping->page_tree,
774 (void **)pages, index, nr_pages);
775 for (i = 0; i < ret; i++) {
776 if (pages[i]->mapping == NULL || pages[i]->index != index)
779 page_cache_get(pages[i]);
782 read_unlock_irq(&mapping->tree_lock);
787 * find_get_pages_tag - find and return pages that match @tag
788 * @mapping: the address_space to search
789 * @index: the starting page index
790 * @tag: the tag index
791 * @nr_pages: the maximum number of pages
792 * @pages: where the resulting pages are placed
794 * Like find_get_pages, except we only return pages which are tagged with
795 * @tag. We update @index to index the next page for the traversal.
797 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
798 int tag, unsigned int nr_pages, struct page **pages)
803 read_lock_irq(&mapping->tree_lock);
804 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
805 (void **)pages, *index, nr_pages, tag);
806 for (i = 0; i < ret; i++)
807 page_cache_get(pages[i]);
809 *index = pages[ret - 1]->index + 1;
810 read_unlock_irq(&mapping->tree_lock);
815 * grab_cache_page_nowait - returns locked page at given index in given cache
816 * @mapping: target address_space
817 * @index: the page index
819 * Same as grab_cache_page, but do not wait if the page is unavailable.
820 * This is intended for speculative data generators, where the data can
821 * be regenerated if the page couldn't be grabbed. This routine should
822 * be safe to call while holding the lock for another page.
824 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
825 * and deadlock against the caller's locked page.
828 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
830 struct page *page = find_get_page(mapping, index);
833 if (!TestSetPageLocked(page))
835 page_cache_release(page);
838 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
839 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
840 page_cache_release(page);
845 EXPORT_SYMBOL(grab_cache_page_nowait);
848 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
849 * a _large_ part of the i/o request. Imagine the worst scenario:
851 * ---R__________________________________________B__________
852 * ^ reading here ^ bad block(assume 4k)
854 * read(R) => miss => readahead(R...B) => media error => frustrating retries
855 * => failing the whole request => read(R) => read(R+1) =>
856 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
857 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
858 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
860 * It is going insane. Fix it by quickly scaling down the readahead size.
862 static void shrink_readahead_size_eio(struct file *filp,
863 struct file_ra_state *ra)
872 * do_generic_mapping_read - generic file read routine
873 * @mapping: address_space to be read
874 * @_ra: file's readahead state
875 * @filp: the file to read
876 * @ppos: current file position
877 * @desc: read_descriptor
878 * @actor: read method
880 * This is a generic file read routine, and uses the
881 * mapping->a_ops->readpage() function for the actual low-level stuff.
883 * This is really ugly. But the goto's actually try to clarify some
884 * of the logic when it comes to error handling etc.
886 * Note the struct file* is only passed for the use of readpage.
889 void do_generic_mapping_read(struct address_space *mapping,
890 struct file_ra_state *_ra,
893 read_descriptor_t *desc,
896 struct inode *inode = mapping->host;
898 unsigned long end_index;
899 unsigned long offset;
900 unsigned long last_index;
901 unsigned long next_index;
902 unsigned long prev_index;
903 unsigned int prev_offset;
905 struct page *cached_page;
907 struct file_ra_state ra = *_ra;
910 index = *ppos >> PAGE_CACHE_SHIFT;
912 prev_index = ra.prev_page;
913 prev_offset = ra.offset;
914 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
915 offset = *ppos & ~PAGE_CACHE_MASK;
917 isize = i_size_read(inode);
921 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
924 unsigned long nr, ret;
926 /* nr is the maximum number of bytes to copy from this page */
927 nr = PAGE_CACHE_SIZE;
928 if (index >= end_index) {
929 if (index > end_index)
931 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
939 if (index == next_index)
940 next_index = page_cache_readahead(mapping, &ra, filp,
941 index, last_index - index);
944 page = find_get_page(mapping, index);
945 if (unlikely(page == NULL)) {
946 handle_ra_miss(mapping, &ra, index);
949 if (!PageUptodate(page))
950 goto page_not_up_to_date;
953 /* If users can be writing to this page using arbitrary
954 * virtual addresses, take care about potential aliasing
955 * before reading the page on the kernel side.
957 if (mapping_writably_mapped(mapping))
958 flush_dcache_page(page);
961 * When a sequential read accesses a page several times,
962 * only mark it as accessed the first time.
964 if (prev_index != index || offset != prev_offset)
965 mark_page_accessed(page);
969 * Ok, we have the page, and it's up-to-date, so
970 * now we can copy it to user space...
972 * The actor routine returns how many bytes were actually used..
973 * NOTE! This may not be the same as how much of a user buffer
974 * we filled up (we may be padding etc), so we can only update
975 * "pos" here (the actor routine has to update the user buffer
976 * pointers and the remaining count).
978 ret = actor(desc, page, offset, nr);
980 index += offset >> PAGE_CACHE_SHIFT;
981 offset &= ~PAGE_CACHE_MASK;
982 prev_offset = ra.offset = offset;
984 page_cache_release(page);
985 if (ret == nr && desc->count)
990 /* Get exclusive access to the page ... */
993 /* Did it get truncated before we got the lock? */
994 if (!page->mapping) {
996 page_cache_release(page);
1000 /* Did somebody else fill it already? */
1001 if (PageUptodate(page)) {
1007 /* Start the actual read. The read will unlock the page. */
1008 error = mapping->a_ops->readpage(filp, page);
1010 if (unlikely(error)) {
1011 if (error == AOP_TRUNCATED_PAGE) {
1012 page_cache_release(page);
1015 goto readpage_error;
1018 if (!PageUptodate(page)) {
1020 if (!PageUptodate(page)) {
1021 if (page->mapping == NULL) {
1023 * invalidate_inode_pages got it
1026 page_cache_release(page);
1031 shrink_readahead_size_eio(filp, &ra);
1032 goto readpage_error;
1038 * i_size must be checked after we have done ->readpage.
1040 * Checking i_size after the readpage allows us to calculate
1041 * the correct value for "nr", which means the zero-filled
1042 * part of the page is not copied back to userspace (unless
1043 * another truncate extends the file - this is desired though).
1045 isize = i_size_read(inode);
1046 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1047 if (unlikely(!isize || index > end_index)) {
1048 page_cache_release(page);
1052 /* nr is the maximum number of bytes to copy from this page */
1053 nr = PAGE_CACHE_SIZE;
1054 if (index == end_index) {
1055 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1057 page_cache_release(page);
1065 /* UHHUH! A synchronous read error occurred. Report it */
1066 desc->error = error;
1067 page_cache_release(page);
1072 * Ok, it wasn't cached, so we need to create a new
1076 cached_page = page_cache_alloc_cold(mapping);
1078 desc->error = -ENOMEM;
1082 error = add_to_page_cache_lru(cached_page, mapping,
1085 if (error == -EEXIST)
1087 desc->error = error;
1098 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1100 page_cache_release(cached_page);
1102 file_accessed(filp);
1104 EXPORT_SYMBOL(do_generic_mapping_read);
1106 int file_read_actor(read_descriptor_t *desc, struct page *page,
1107 unsigned long offset, unsigned long size)
1110 unsigned long left, count = desc->count;
1116 * Faults on the destination of a read are common, so do it before
1119 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1120 kaddr = kmap_atomic(page, KM_USER0);
1121 left = __copy_to_user_inatomic(desc->arg.buf,
1122 kaddr + offset, size);
1123 kunmap_atomic(kaddr, KM_USER0);
1128 /* Do it the slow way */
1130 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1135 desc->error = -EFAULT;
1138 desc->count = count - size;
1139 desc->written += size;
1140 desc->arg.buf += size;
1145 * generic_file_aio_read - generic filesystem read routine
1146 * @iocb: kernel I/O control block
1147 * @iov: io vector request
1148 * @nr_segs: number of segments in the iovec
1149 * @pos: current file position
1151 * This is the "read()" routine for all filesystems
1152 * that can use the page cache directly.
1155 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1156 unsigned long nr_segs, loff_t pos)
1158 struct file *filp = iocb->ki_filp;
1162 loff_t *ppos = &iocb->ki_pos;
1165 for (seg = 0; seg < nr_segs; seg++) {
1166 const struct iovec *iv = &iov[seg];
1169 * If any segment has a negative length, or the cumulative
1170 * length ever wraps negative then return -EINVAL.
1172 count += iv->iov_len;
1173 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1175 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1180 count -= iv->iov_len; /* This segment is no good */
1184 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1185 if (filp->f_flags & O_DIRECT) {
1187 struct address_space *mapping;
1188 struct inode *inode;
1190 mapping = filp->f_mapping;
1191 inode = mapping->host;
1194 goto out; /* skip atime */
1195 size = i_size_read(inode);
1197 retval = generic_file_direct_IO(READ, iocb,
1200 *ppos = pos + retval;
1202 if (likely(retval != 0)) {
1203 file_accessed(filp);
1210 for (seg = 0; seg < nr_segs; seg++) {
1211 read_descriptor_t desc;
1214 desc.arg.buf = iov[seg].iov_base;
1215 desc.count = iov[seg].iov_len;
1216 if (desc.count == 0)
1219 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1220 retval += desc.written;
1222 retval = retval ?: desc.error;
1230 EXPORT_SYMBOL(generic_file_aio_read);
1232 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1235 unsigned long count = desc->count;
1236 struct file *file = desc->arg.data;
1241 written = file->f_op->sendpage(file, page, offset,
1242 size, &file->f_pos, size<count);
1244 desc->error = written;
1247 desc->count = count - written;
1248 desc->written += written;
1252 /* FIXME: It would be as simple as this, if we had a (void __user*) to write.
1253 * We already have a kernel buffer, so it should be even simpler, right? ;)
1255 * Yes, sorta. After duplicating the complete path of generic_file_write(),
1256 * at least some special cases could be removed, so the copy is simpler than
1257 * the original. But it remains a copy, so overall complexity increases.
1260 generic_kernel_file_write(struct file *, const char *, size_t, loff_t *);
1262 ssize_t generic_file_sendpage(struct file *file, struct page *page,
1263 int offset, size_t size, loff_t *ppos, int more)
1269 ret = generic_kernel_file_write(file, kaddr + offset, size, ppos);
1275 EXPORT_SYMBOL(generic_file_sendpage);
1277 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1278 size_t count, read_actor_t actor, void *target)
1280 read_descriptor_t desc;
1287 desc.arg.data = target;
1290 do_generic_file_read(in_file, ppos, &desc, actor);
1292 return desc.written;
1295 EXPORT_SYMBOL(generic_file_sendfile);
1298 do_readahead(struct address_space *mapping, struct file *filp,
1299 unsigned long index, unsigned long nr)
1301 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1304 force_page_cache_readahead(mapping, filp, index,
1305 max_sane_readahead(nr));
1309 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1317 if (file->f_mode & FMODE_READ) {
1318 struct address_space *mapping = file->f_mapping;
1319 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1320 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1321 unsigned long len = end - start + 1;
1322 ret = do_readahead(mapping, file, start, len);
1330 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1332 * page_cache_read - adds requested page to the page cache if not already there
1333 * @file: file to read
1334 * @offset: page index
1336 * This adds the requested page to the page cache if it isn't already there,
1337 * and schedules an I/O to read in its contents from disk.
1339 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1341 struct address_space *mapping = file->f_mapping;
1346 page = page_cache_alloc_cold(mapping);
1350 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1352 ret = mapping->a_ops->readpage(file, page);
1353 else if (ret == -EEXIST)
1354 ret = 0; /* losing race to add is OK */
1356 page_cache_release(page);
1358 } while (ret == AOP_TRUNCATED_PAGE);
1363 #define MMAP_LOTSAMISS (100)
1366 * filemap_nopage - read in file data for page fault handling
1367 * @area: the applicable vm_area
1368 * @address: target address to read in
1369 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1371 * filemap_nopage() is invoked via the vma operations vector for a
1372 * mapped memory region to read in file data during a page fault.
1374 * The goto's are kind of ugly, but this streamlines the normal case of having
1375 * it in the page cache, and handles the special cases reasonably without
1376 * having a lot of duplicated code.
1378 struct page *filemap_nopage(struct vm_area_struct *area,
1379 unsigned long address, int *type)
1382 struct file *file = area->vm_file;
1383 struct address_space *mapping = file->f_mapping;
1384 struct file_ra_state *ra = &file->f_ra;
1385 struct inode *inode = mapping->host;
1387 unsigned long size, pgoff;
1388 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1390 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1393 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1395 goto outside_data_content;
1397 /* If we don't want any read-ahead, don't bother */
1398 if (VM_RandomReadHint(area))
1399 goto no_cached_page;
1402 * The readahead code wants to be told about each and every page
1403 * so it can build and shrink its windows appropriately
1405 * For sequential accesses, we use the generic readahead logic.
1407 if (VM_SequentialReadHint(area))
1408 page_cache_readahead(mapping, ra, file, pgoff, 1);
1411 * Do we have something in the page cache already?
1414 page = find_get_page(mapping, pgoff);
1416 unsigned long ra_pages;
1418 if (VM_SequentialReadHint(area)) {
1419 handle_ra_miss(mapping, ra, pgoff);
1420 goto no_cached_page;
1425 * Do we miss much more than hit in this file? If so,
1426 * stop bothering with read-ahead. It will only hurt.
1428 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1429 goto no_cached_page;
1432 * To keep the pgmajfault counter straight, we need to
1433 * check did_readaround, as this is an inner loop.
1435 if (!did_readaround) {
1436 majmin = VM_FAULT_MAJOR;
1437 count_vm_event(PGMAJFAULT);
1440 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1444 if (pgoff > ra_pages / 2)
1445 start = pgoff - ra_pages / 2;
1446 do_page_cache_readahead(mapping, file, start, ra_pages);
1448 page = find_get_page(mapping, pgoff);
1450 goto no_cached_page;
1453 if (!did_readaround)
1457 * Ok, found a page in the page cache, now we need to check
1458 * that it's up-to-date.
1460 if (!PageUptodate(page))
1461 goto page_not_uptodate;
1465 * Found the page and have a reference on it.
1467 mark_page_accessed(page);
1472 outside_data_content:
1474 * An external ptracer can access pages that normally aren't
1477 if (area->vm_mm == current->mm)
1478 return NOPAGE_SIGBUS;
1479 /* Fall through to the non-read-ahead case */
1482 * We're only likely to ever get here if MADV_RANDOM is in
1485 error = page_cache_read(file, pgoff);
1488 * The page we want has now been added to the page cache.
1489 * In the unlikely event that someone removed it in the
1490 * meantime, we'll just come back here and read it again.
1496 * An error return from page_cache_read can result if the
1497 * system is low on memory, or a problem occurs while trying
1500 if (error == -ENOMEM)
1502 return NOPAGE_SIGBUS;
1505 if (!did_readaround) {
1506 majmin = VM_FAULT_MAJOR;
1507 count_vm_event(PGMAJFAULT);
1511 /* Did it get unhashed while we waited for it? */
1512 if (!page->mapping) {
1514 page_cache_release(page);
1518 /* Did somebody else get it up-to-date? */
1519 if (PageUptodate(page)) {
1524 error = mapping->a_ops->readpage(file, page);
1526 wait_on_page_locked(page);
1527 if (PageUptodate(page))
1529 } else if (error == AOP_TRUNCATED_PAGE) {
1530 page_cache_release(page);
1535 * Umm, take care of errors if the page isn't up-to-date.
1536 * Try to re-read it _once_. We do this synchronously,
1537 * because there really aren't any performance issues here
1538 * and we need to check for errors.
1542 /* Somebody truncated the page on us? */
1543 if (!page->mapping) {
1545 page_cache_release(page);
1549 /* Somebody else successfully read it in? */
1550 if (PageUptodate(page)) {
1554 ClearPageError(page);
1555 error = mapping->a_ops->readpage(file, page);
1557 wait_on_page_locked(page);
1558 if (PageUptodate(page))
1560 } else if (error == AOP_TRUNCATED_PAGE) {
1561 page_cache_release(page);
1566 * Things didn't work out. Return zero to tell the
1567 * mm layer so, possibly freeing the page cache page first.
1569 shrink_readahead_size_eio(file, ra);
1570 page_cache_release(page);
1571 return NOPAGE_SIGBUS;
1573 EXPORT_SYMBOL(filemap_nopage);
1575 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1578 struct address_space *mapping = file->f_mapping;
1583 * Do we have something in the page cache already?
1586 page = find_get_page(mapping, pgoff);
1590 goto no_cached_page;
1594 * Ok, found a page in the page cache, now we need to check
1595 * that it's up-to-date.
1597 if (!PageUptodate(page)) {
1599 page_cache_release(page);
1602 goto page_not_uptodate;
1607 * Found the page and have a reference on it.
1609 mark_page_accessed(page);
1613 error = page_cache_read(file, pgoff);
1616 * The page we want has now been added to the page cache.
1617 * In the unlikely event that someone removed it in the
1618 * meantime, we'll just come back here and read it again.
1624 * An error return from page_cache_read can result if the
1625 * system is low on memory, or a problem occurs while trying
1633 /* Did it get truncated while we waited for it? */
1634 if (!page->mapping) {
1639 /* Did somebody else get it up-to-date? */
1640 if (PageUptodate(page)) {
1645 error = mapping->a_ops->readpage(file, page);
1647 wait_on_page_locked(page);
1648 if (PageUptodate(page))
1650 } else if (error == AOP_TRUNCATED_PAGE) {
1651 page_cache_release(page);
1656 * Umm, take care of errors if the page isn't up-to-date.
1657 * Try to re-read it _once_. We do this synchronously,
1658 * because there really aren't any performance issues here
1659 * and we need to check for errors.
1663 /* Somebody truncated the page on us? */
1664 if (!page->mapping) {
1668 /* Somebody else successfully read it in? */
1669 if (PageUptodate(page)) {
1674 ClearPageError(page);
1675 error = mapping->a_ops->readpage(file, page);
1677 wait_on_page_locked(page);
1678 if (PageUptodate(page))
1680 } else if (error == AOP_TRUNCATED_PAGE) {
1681 page_cache_release(page);
1686 * Things didn't work out. Return zero to tell the
1687 * mm layer so, possibly freeing the page cache page first.
1690 page_cache_release(page);
1695 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1696 unsigned long len, pgprot_t prot, unsigned long pgoff,
1699 struct file *file = vma->vm_file;
1700 struct address_space *mapping = file->f_mapping;
1701 struct inode *inode = mapping->host;
1703 struct mm_struct *mm = vma->vm_mm;
1708 force_page_cache_readahead(mapping, vma->vm_file,
1709 pgoff, len >> PAGE_CACHE_SHIFT);
1712 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1713 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1716 page = filemap_getpage(file, pgoff, nonblock);
1718 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1719 * done in shmem_populate calling shmem_getpage */
1720 if (!page && !nonblock)
1724 err = install_page(mm, vma, addr, page, prot);
1726 page_cache_release(page);
1729 } else if (vma->vm_flags & VM_NONLINEAR) {
1730 /* No page was found just because we can't read it in now (being
1731 * here implies nonblock != 0), but the page may exist, so set
1732 * the PTE to fault it in later. */
1733 err = install_file_pte(mm, vma, addr, pgoff, prot);
1746 EXPORT_SYMBOL(filemap_populate);
1748 struct vm_operations_struct generic_file_vm_ops = {
1749 .nopage = filemap_nopage,
1750 .populate = filemap_populate,
1753 /* This is used for a general mmap of a disk file */
1755 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1757 struct address_space *mapping = file->f_mapping;
1759 if (!mapping->a_ops->readpage)
1761 file_accessed(file);
1762 vma->vm_ops = &generic_file_vm_ops;
1767 * This is for filesystems which do not implement ->writepage.
1769 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1771 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1773 return generic_file_mmap(file, vma);
1776 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1780 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1784 #endif /* CONFIG_MMU */
1786 EXPORT_SYMBOL(generic_file_mmap);
1787 EXPORT_SYMBOL(generic_file_readonly_mmap);
1789 static inline struct page *__read_cache_page(struct address_space *mapping,
1790 unsigned long index,
1791 int (*filler)(void *,struct page*),
1794 struct page *page, *cached_page = NULL;
1797 page = find_get_page(mapping, index);
1800 cached_page = page_cache_alloc_cold(mapping);
1802 return ERR_PTR(-ENOMEM);
1804 err = add_to_page_cache_lru(cached_page, mapping,
1809 /* Presumably ENOMEM for radix tree node */
1810 page_cache_release(cached_page);
1811 return ERR_PTR(err);
1815 err = filler(data, page);
1817 page_cache_release(page);
1818 page = ERR_PTR(err);
1822 page_cache_release(cached_page);
1827 * read_cache_page - read into page cache, fill it if needed
1828 * @mapping: the page's address_space
1829 * @index: the page index
1830 * @filler: function to perform the read
1831 * @data: destination for read data
1833 * Read into the page cache. If a page already exists,
1834 * and PageUptodate() is not set, try to fill the page.
1836 struct page *read_cache_page(struct address_space *mapping,
1837 unsigned long index,
1838 int (*filler)(void *,struct page*),
1845 page = __read_cache_page(mapping, index, filler, data);
1848 mark_page_accessed(page);
1849 if (PageUptodate(page))
1853 if (!page->mapping) {
1855 page_cache_release(page);
1858 if (PageUptodate(page)) {
1862 err = filler(data, page);
1864 page_cache_release(page);
1865 page = ERR_PTR(err);
1870 EXPORT_SYMBOL(read_cache_page);
1873 * If the page was newly created, increment its refcount and add it to the
1874 * caller's lru-buffering pagevec. This function is specifically for
1875 * generic_file_write().
1877 static inline struct page *
1878 __grab_cache_page(struct address_space *mapping, unsigned long index,
1879 struct page **cached_page, struct pagevec *lru_pvec)
1884 page = find_lock_page(mapping, index);
1886 if (!*cached_page) {
1887 *cached_page = page_cache_alloc(mapping);
1891 err = add_to_page_cache(*cached_page, mapping,
1896 page = *cached_page;
1897 page_cache_get(page);
1898 if (!pagevec_add(lru_pvec, page))
1899 __pagevec_lru_add(lru_pvec);
1900 *cached_page = NULL;
1907 * The logic we want is
1909 * if suid or (sgid and xgrp)
1912 int should_remove_suid(struct dentry *dentry)
1914 mode_t mode = dentry->d_inode->i_mode;
1917 /* suid always must be killed */
1918 if (unlikely(mode & S_ISUID))
1919 kill = ATTR_KILL_SUID;
1922 * sgid without any exec bits is just a mandatory locking mark; leave
1923 * it alone. If some exec bits are set, it's a real sgid; kill it.
1925 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1926 kill |= ATTR_KILL_SGID;
1928 if (unlikely(kill && !capable(CAP_FSETID)))
1933 EXPORT_SYMBOL(should_remove_suid);
1935 int __remove_suid(struct dentry *dentry, int kill)
1937 struct iattr newattrs;
1939 newattrs.ia_valid = ATTR_FORCE | kill;
1940 return notify_change(dentry, &newattrs);
1943 int remove_suid(struct dentry *dentry)
1945 int kill = should_remove_suid(dentry);
1948 return __remove_suid(dentry, kill);
1952 EXPORT_SYMBOL(remove_suid);
1954 static inline size_t
1955 filemap_copy_from_kernel(struct page *page, unsigned long offset,
1956 const char *buf, unsigned bytes)
1961 memcpy(kaddr + offset, buf, bytes);
1968 __filemap_copy_from_user_iovec_inatomic(char *vaddr,
1969 const struct iovec *iov, size_t base, size_t bytes)
1971 size_t copied = 0, left = 0;
1974 char __user *buf = iov->iov_base + base;
1975 int copy = min(bytes, iov->iov_len - base);
1978 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1987 return copied - left;
1991 * Performs necessary checks before doing a write
1993 * Can adjust writing position or amount of bytes to write.
1994 * Returns appropriate error code that caller should return or
1995 * zero in case that write should be allowed.
1997 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1999 struct inode *inode = file->f_mapping->host;
2000 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2002 if (unlikely(*pos < 0))
2006 /* FIXME: this is for backwards compatibility with 2.4 */
2007 if (file->f_flags & O_APPEND)
2008 *pos = i_size_read(inode);
2010 if (limit != RLIM_INFINITY) {
2011 if (*pos >= limit) {
2012 send_sig(SIGXFSZ, current, 0);
2015 if (*count > limit - (typeof(limit))*pos) {
2016 *count = limit - (typeof(limit))*pos;
2024 if (unlikely(*pos + *count > MAX_NON_LFS &&
2025 !(file->f_flags & O_LARGEFILE))) {
2026 if (*pos >= MAX_NON_LFS) {
2027 send_sig(SIGXFSZ, current, 0);
2030 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2031 *count = MAX_NON_LFS - (unsigned long)*pos;
2036 * Are we about to exceed the fs block limit ?
2038 * If we have written data it becomes a short write. If we have
2039 * exceeded without writing data we send a signal and return EFBIG.
2040 * Linus frestrict idea will clean these up nicely..
2042 if (likely(!isblk)) {
2043 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2044 if (*count || *pos > inode->i_sb->s_maxbytes) {
2045 send_sig(SIGXFSZ, current, 0);
2048 /* zero-length writes at ->s_maxbytes are OK */
2051 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2052 *count = inode->i_sb->s_maxbytes - *pos;
2056 if (bdev_read_only(I_BDEV(inode)))
2058 isize = i_size_read(inode);
2059 if (*pos >= isize) {
2060 if (*count || *pos > isize)
2064 if (*pos + *count > isize)
2065 *count = isize - *pos;
2072 EXPORT_SYMBOL(generic_write_checks);
2075 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2076 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2077 size_t count, size_t ocount)
2079 struct file *file = iocb->ki_filp;
2080 struct address_space *mapping = file->f_mapping;
2081 struct inode *inode = mapping->host;
2084 if (count != ocount)
2085 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2087 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2089 loff_t end = pos + written;
2090 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2091 i_size_write(inode, end);
2092 mark_inode_dirty(inode);
2098 * Sync the fs metadata but not the minor inode changes and
2099 * of course not the data as we did direct DMA for the IO.
2100 * i_mutex is held, which protects generic_osync_inode() from
2101 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
2103 if ((written >= 0 || written == -EIOCBQUEUED) &&
2104 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2105 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2111 EXPORT_SYMBOL(generic_file_direct_write);
2114 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2115 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2116 size_t count, ssize_t written)
2118 struct file *file = iocb->ki_filp;
2119 struct address_space * mapping = file->f_mapping;
2120 const struct address_space_operations *a_ops = mapping->a_ops;
2121 struct inode *inode = mapping->host;
2124 struct page *cached_page = NULL;
2126 struct pagevec lru_pvec;
2127 const struct iovec *cur_iov = iov; /* current iovec */
2128 size_t iov_base = 0; /* offset in the current iovec */
2131 pagevec_init(&lru_pvec, 0);
2134 * handle partial DIO write. Adjust cur_iov if needed.
2136 if (likely(nr_segs == 1))
2137 buf = iov->iov_base + written;
2139 filemap_set_next_iovec(&cur_iov, &iov_base, written);
2140 buf = cur_iov->iov_base + iov_base;
2144 unsigned long index;
2145 unsigned long offset;
2148 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2149 index = pos >> PAGE_CACHE_SHIFT;
2150 bytes = PAGE_CACHE_SIZE - offset;
2152 /* Limit the size of the copy to the caller's write size */
2153 bytes = min(bytes, count);
2156 * Limit the size of the copy to that of the current segment,
2157 * because fault_in_pages_readable() doesn't know how to walk
2160 bytes = min(bytes, cur_iov->iov_len - iov_base);
2163 * Bring in the user page that we will copy from _first_.
2164 * Otherwise there's a nasty deadlock on copying from the
2165 * same page as we're writing to, without it being marked
2168 fault_in_pages_readable(buf, bytes);
2170 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2176 if (unlikely(bytes == 0)) {
2179 goto zero_length_segment;
2182 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2183 if (unlikely(status)) {
2184 loff_t isize = i_size_read(inode);
2186 if (status != AOP_TRUNCATED_PAGE)
2188 page_cache_release(page);
2189 if (status == AOP_TRUNCATED_PAGE)
2192 * prepare_write() may have instantiated a few blocks
2193 * outside i_size. Trim these off again.
2195 if (pos + bytes > isize)
2196 vmtruncate(inode, isize);
2199 if (likely(nr_segs == 1))
2200 copied = filemap_copy_from_user(page, offset,
2203 copied = filemap_copy_from_user_iovec(page, offset,
2204 cur_iov, iov_base, bytes);
2205 flush_dcache_page(page);
2206 status = a_ops->commit_write(file, page, offset, offset+bytes);
2207 if (status == AOP_TRUNCATED_PAGE) {
2208 page_cache_release(page);
2211 zero_length_segment:
2212 if (likely(copied >= 0)) {
2221 if (unlikely(nr_segs > 1)) {
2222 filemap_set_next_iovec(&cur_iov,
2225 buf = cur_iov->iov_base +
2232 if (unlikely(copied != bytes))
2236 mark_page_accessed(page);
2237 page_cache_release(page);
2240 balance_dirty_pages_ratelimited(mapping);
2246 page_cache_release(cached_page);
2249 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2251 if (likely(status >= 0)) {
2252 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2253 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2254 status = generic_osync_inode(inode, mapping,
2255 OSYNC_METADATA|OSYNC_DATA);
2260 * If we get here for O_DIRECT writes then we must have fallen through
2261 * to buffered writes (block instantiation inside i_size). So we sync
2262 * the file data here, to try to honour O_DIRECT expectations.
2264 if (unlikely(file->f_flags & O_DIRECT) && written)
2265 status = filemap_write_and_wait(mapping);
2267 pagevec_lru_add(&lru_pvec);
2268 return written ? written : status;
2270 EXPORT_SYMBOL(generic_file_buffered_write);
2273 * This writes the data from the source page to the specified page offset in
2274 * the nominated file
2275 * - the source page does not need to have any association with the file or the
2279 generic_file_buffered_write_one_kernel_page(struct address_space *mapping,
2283 const struct address_space_operations *a_ops = mapping->a_ops;
2284 struct pagevec lru_pvec;
2285 struct page *page, *cached_page = NULL;
2288 pagevec_init(&lru_pvec, 0);
2291 if (mapping->tree_lock.magic != RWLOCK_MAGIC)
2292 printk("RWLOCK magic incorrect: %x != %x\n",
2293 mapping->tree_lock.magic, RWLOCK_MAGIC);
2296 page = __grab_cache_page(mapping, index, &cached_page, &lru_pvec);
2298 BUG_ON(cached_page);
2302 status = a_ops->prepare_write(NULL, page, 0, PAGE_CACHE_SIZE);
2303 if (unlikely(status)) {
2304 loff_t isize = i_size_read(mapping->host);
2306 if (status != AOP_TRUNCATED_PAGE)
2308 page_cache_release(page);
2309 if (status == AOP_TRUNCATED_PAGE)
2312 /* prepare_write() may have instantiated a few blocks outside
2313 * i_size. Trim these off again.
2315 if ((1ULL << (index + 1)) > isize)
2316 vmtruncate(mapping->host, isize);
2320 copy_highpage(page, src);
2321 flush_dcache_page(page);
2323 status = a_ops->commit_write(NULL, page, 0, PAGE_CACHE_SIZE);
2324 if (status == AOP_TRUNCATED_PAGE) {
2325 page_cache_release(page);
2333 mark_page_accessed(page);
2334 page_cache_release(page);
2338 balance_dirty_pages_ratelimited(mapping);
2343 page_cache_release(cached_page);
2345 /* the caller must handle O_SYNC themselves, but we handle S_SYNC and
2346 * MS_SYNCHRONOUS here */
2347 if (unlikely(IS_SYNC(mapping->host)) && !a_ops->writepage)
2348 status = generic_osync_inode(mapping->host, mapping,
2349 OSYNC_METADATA | OSYNC_DATA);
2351 /* the caller must handle O_DIRECT for themselves */
2353 pagevec_lru_add(&lru_pvec);
2356 EXPORT_SYMBOL(generic_file_buffered_write_one_kernel_page);
2359 filemap_set_next_kvec(const struct kvec **iovp, size_t *basep, size_t bytes)
2361 const struct kvec *iov = *iovp;
2362 size_t base = *basep;
2365 int copy = min(bytes, iov->iov_len - base);
2369 if (iov->iov_len == base) {
2380 * This largely tries to copy generic_file_aio_write_nolock(), although it
2381 * doesn't have to be nearly as generic. A real cleanup should either
2382 * merge this into generic_file_aio_write_nolock() as well or keep it special
2383 * and remove as much code as possible.
2386 generic_kernel_file_aio_write_nolock(struct kiocb *iocb, const struct kvec*iov,
2387 unsigned long nr_segs, loff_t *ppos)
2389 struct file *file = iocb->ki_filp;
2390 struct address_space * mapping = file->f_mapping;
2391 const struct address_space_operations *a_ops = mapping->a_ops;
2392 size_t ocount; /* original count */
2393 size_t count; /* after file limit checks */
2394 struct inode *inode = mapping->host;
2398 struct page *cached_page = NULL;
2399 const int isblk = S_ISBLK(inode->i_mode);
2403 struct pagevec lru_pvec;
2404 const struct kvec *cur_iov = iov; /* current kvec */
2405 size_t iov_base = 0; /* offset in the current kvec */
2410 for (seg = 0; seg < nr_segs; seg++) {
2411 const struct kvec *iv = &iov[seg];
2414 * If any segment has a negative length, or the cumulative
2415 * length ever wraps negative then return -EINVAL.
2417 ocount += iv->iov_len;
2418 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2424 pagevec_init(&lru_pvec, 0);
2426 /* We can write back this queue in page reclaim */
2427 current->backing_dev_info = mapping->backing_dev_info;
2430 err = generic_write_checks(file, &pos, &count, isblk);
2438 remove_suid(file->f_dentry);
2439 file_update_time(file);
2441 /* There is no sane reason to use O_DIRECT */
2442 BUG_ON(file->f_flags & O_DIRECT);
2444 buf = iov->iov_base;
2446 unsigned long index;
2447 unsigned long offset;
2450 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2451 index = pos >> PAGE_CACHE_SHIFT;
2452 bytes = PAGE_CACHE_SIZE - offset;
2456 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2462 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2463 if (unlikely(status)) {
2464 loff_t isize = i_size_read(inode);
2466 * prepare_write() may have instantiated a few blocks
2467 * outside i_size. Trim these off again.
2470 page_cache_release(page);
2471 if (pos + bytes > isize)
2472 vmtruncate(inode, isize);
2476 BUG_ON(nr_segs != 1);
2477 copied = filemap_copy_from_kernel(page, offset, buf, bytes);
2479 flush_dcache_page(page);
2480 status = a_ops->commit_write(file, page, offset, offset+bytes);
2481 if (likely(copied > 0)) {
2490 if (unlikely(nr_segs > 1))
2491 filemap_set_next_kvec(&cur_iov,
2495 if (unlikely(copied != bytes))
2499 mark_page_accessed(page);
2500 page_cache_release(page);
2503 balance_dirty_pages_ratelimited(mapping);
2509 page_cache_release(cached_page);
2512 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2515 if ((file->f_flags & O_SYNC) || IS_SYNC(inode))
2516 status = generic_osync_inode(inode, mapping,
2517 OSYNC_METADATA|OSYNC_DATA);
2520 err = written ? written : status;
2522 pagevec_lru_add(&lru_pvec);
2523 current->backing_dev_info = 0;
2528 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2529 unsigned long nr_segs, loff_t *ppos)
2531 struct file *file = iocb->ki_filp;
2532 struct address_space * mapping = file->f_mapping;
2533 size_t ocount; /* original count */
2534 size_t count; /* after file limit checks */
2535 struct inode *inode = mapping->host;
2542 for (seg = 0; seg < nr_segs; seg++) {
2543 const struct iovec *iv = &iov[seg];
2546 * If any segment has a negative length, or the cumulative
2547 * length ever wraps negative then return -EINVAL.
2549 ocount += iv->iov_len;
2550 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2552 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2557 ocount -= iv->iov_len; /* This segment is no good */
2564 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2566 /* We can write back this queue in page reclaim */
2567 current->backing_dev_info = mapping->backing_dev_info;
2570 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2577 err = remove_suid(file->f_path.dentry);
2581 file_update_time(file);
2583 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2584 if (unlikely(file->f_flags & O_DIRECT)) {
2586 ssize_t written_buffered;
2588 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2589 ppos, count, ocount);
2590 if (written < 0 || written == count)
2593 * direct-io write to a hole: fall through to buffered I/O
2594 * for completing the rest of the request.
2598 written_buffered = generic_file_buffered_write(iocb, iov,
2599 nr_segs, pos, ppos, count,
2602 * If generic_file_buffered_write() retuned a synchronous error
2603 * then we want to return the number of bytes which were
2604 * direct-written, or the error code if that was zero. Note
2605 * that this differs from normal direct-io semantics, which
2606 * will return -EFOO even if some bytes were written.
2608 if (written_buffered < 0) {
2609 err = written_buffered;
2614 * We need to ensure that the page cache pages are written to
2615 * disk and invalidated to preserve the expected O_DIRECT
2618 endbyte = pos + written_buffered - written - 1;
2619 err = do_sync_file_range(file, pos, endbyte,
2620 SYNC_FILE_RANGE_WAIT_BEFORE|
2621 SYNC_FILE_RANGE_WRITE|
2622 SYNC_FILE_RANGE_WAIT_AFTER);
2624 written = written_buffered;
2625 invalidate_mapping_pages(mapping,
2626 pos >> PAGE_CACHE_SHIFT,
2627 endbyte >> PAGE_CACHE_SHIFT);
2630 * We don't know how much we wrote, so just return
2631 * the number of bytes which were direct-written
2635 written = generic_file_buffered_write(iocb, iov, nr_segs,
2636 pos, ppos, count, written);
2639 current->backing_dev_info = NULL;
2640 return written ? written : err;
2644 generic_kernel_file_write_nolock(struct file *file, const struct kvec *iov,
2645 unsigned long nr_segs, loff_t *ppos)
2650 init_sync_kiocb(&kiocb, file);
2651 ret = generic_kernel_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2652 if (ret == -EIOCBQUEUED)
2653 ret = wait_on_sync_kiocb(&kiocb);
2657 static ssize_t generic_kernel_file_write(struct file *file, const char *buf,
2658 size_t count, loff_t *ppos)
2660 struct inode *inode = file->f_mapping->host;
2662 struct kvec local_iov = { .iov_base = (char *) buf,
2665 mutex_lock(&inode->i_mutex);
2666 err = generic_kernel_file_write_nolock(file, &local_iov, 1, ppos);
2667 mutex_unlock(&inode->i_mutex);
2673 ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2674 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
2676 struct file *file = iocb->ki_filp;
2677 struct address_space *mapping = file->f_mapping;
2678 struct inode *inode = mapping->host;
2681 BUG_ON(iocb->ki_pos != pos);
2683 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2686 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2689 err = sync_page_range_nolock(inode, mapping, pos, ret);
2695 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2697 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2698 unsigned long nr_segs, loff_t pos)
2700 struct file *file = iocb->ki_filp;
2701 struct address_space *mapping = file->f_mapping;
2702 struct inode *inode = mapping->host;
2705 BUG_ON(iocb->ki_pos != pos);
2707 mutex_lock(&inode->i_mutex);
2708 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2710 mutex_unlock(&inode->i_mutex);
2712 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2715 err = sync_page_range(inode, mapping, pos, ret);
2721 EXPORT_SYMBOL(generic_file_aio_write);
2724 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2725 * went wrong during pagecache shootdown.
2728 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2729 loff_t offset, unsigned long nr_segs)
2731 struct file *file = iocb->ki_filp;
2732 struct address_space *mapping = file->f_mapping;
2735 pgoff_t end = 0; /* silence gcc */
2738 * If it's a write, unmap all mmappings of the file up-front. This
2739 * will cause any pte dirty bits to be propagated into the pageframes
2740 * for the subsequent filemap_write_and_wait().
2743 write_len = iov_length(iov, nr_segs);
2744 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
2745 if (mapping_mapped(mapping))
2746 unmap_mapping_range(mapping, offset, write_len, 0);
2749 retval = filemap_write_and_wait(mapping);
2754 * After a write we want buffered reads to be sure to go to disk to get
2755 * the new data. We invalidate clean cached page from the region we're
2756 * about to write. We do this *before* the write so that we can return
2757 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2759 if (rw == WRITE && mapping->nrpages) {
2760 retval = invalidate_inode_pages2_range(mapping,
2761 offset >> PAGE_CACHE_SHIFT, end);
2766 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2771 * Finally, try again to invalidate clean pages which might have been
2772 * faulted in by get_user_pages() if the source of the write was an
2773 * mmap()ed region of the file we're writing. That's a pretty crazy
2774 * thing to do, so we don't support it 100%. If this invalidation
2775 * fails and we have -EIOCBQUEUED we ignore the failure.
2777 if (rw == WRITE && mapping->nrpages) {
2778 int err = invalidate_inode_pages2_range(mapping,
2779 offset >> PAGE_CACHE_SHIFT, end);
2780 if (err && retval >= 0)
2788 * try_to_release_page() - release old fs-specific metadata on a page
2790 * @page: the page which the kernel is trying to free
2791 * @gfp_mask: memory allocation flags (and I/O mode)
2793 * The address_space is to try to release any data against the page
2794 * (presumably at page->private). If the release was successful, return `1'.
2795 * Otherwise return zero.
2797 * The @gfp_mask argument specifies whether I/O may be performed to release
2798 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2800 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2802 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2804 struct address_space * const mapping = page->mapping;
2806 BUG_ON(!PageLocked(page));
2807 if (PageWriteback(page))
2810 if (mapping && mapping->a_ops->releasepage)
2811 return mapping->a_ops->releasepage(page, gfp_mask);
2812 return try_to_free_buffers(page);
2815 EXPORT_SYMBOL(try_to_release_page);