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)
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(struct address_space *x)
472 if (cpuset_do_page_mem_spread()) {
473 int n = cpuset_mem_spread_node();
474 return alloc_pages_node(n, mapping_gfp_mask(x), 0);
476 return alloc_pages(mapping_gfp_mask(x), 0);
478 EXPORT_SYMBOL(page_cache_alloc);
480 struct page *page_cache_alloc_cold(struct address_space *x)
482 if (cpuset_do_page_mem_spread()) {
483 int n = cpuset_mem_spread_node();
484 return alloc_pages_node(n, mapping_gfp_mask(x)|__GFP_COLD, 0);
486 return alloc_pages(mapping_gfp_mask(x)|__GFP_COLD, 0);
488 EXPORT_SYMBOL(page_cache_alloc_cold);
492 * In order to wait for pages to become available there must be
493 * waitqueues associated with pages. By using a hash table of
494 * waitqueues where the bucket discipline is to maintain all
495 * waiters on the same queue and wake all when any of the pages
496 * become available, and for the woken contexts to check to be
497 * sure the appropriate page became available, this saves space
498 * at a cost of "thundering herd" phenomena during rare hash
501 static wait_queue_head_t *page_waitqueue(struct page *page)
503 const struct zone *zone = page_zone(page);
505 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
508 static inline void wake_up_page(struct page *page, int bit)
510 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
513 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
515 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
517 if (test_bit(bit_nr, &page->flags))
518 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
519 TASK_UNINTERRUPTIBLE);
521 EXPORT_SYMBOL(wait_on_page_bit);
523 void install_page_waitqueue_monitor(struct page *page, wait_queue_t *monitor)
525 wait_queue_head_t *q = page_waitqueue(page);
528 spin_lock_irqsave(&q->lock, flags);
529 __add_wait_queue(q, monitor);
530 spin_unlock_irqrestore(&q->lock, flags);
533 EXPORT_SYMBOL_GPL(install_page_waitqueue_monitor);
536 * unlock_page - unlock a locked page
539 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
540 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
541 * mechananism between PageLocked pages and PageWriteback pages is shared.
542 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
544 * The first mb is necessary to safely close the critical section opened by the
545 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
546 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
547 * parallel wait_on_page_locked()).
549 void fastcall unlock_page(struct page *page)
551 smp_mb__before_clear_bit();
552 if (!TestClearPageLocked(page))
554 smp_mb__after_clear_bit();
555 wake_up_page(page, PG_locked);
557 EXPORT_SYMBOL(unlock_page);
560 * end_page_writeback - end writeback against a page
563 void end_page_writeback(struct page *page)
565 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
566 if (!test_clear_page_writeback(page))
569 smp_mb__after_clear_bit();
570 wake_up_page(page, PG_writeback);
572 EXPORT_SYMBOL(end_page_writeback);
575 * __lock_page - get a lock on the page, assuming we need to sleep to get it
576 * @page: the page to lock
578 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
579 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
580 * chances are that on the second loop, the block layer's plug list is empty,
581 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
583 void fastcall __lock_page(struct page *page)
585 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
587 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
588 TASK_UNINTERRUPTIBLE);
590 EXPORT_SYMBOL(__lock_page);
593 * Note completion of filesystem specific page synchronisation
595 * This is used to allow a page to be written to a filesystem cache in the
596 * background without holding up the completion of readpage
598 void fastcall end_page_fs_misc(struct page *page)
600 smp_mb__before_clear_bit();
601 if (!TestClearPageFsMisc(page))
603 smp_mb__after_clear_bit();
604 __wake_up_bit(page_waitqueue(page), &page->flags, PG_fs_misc);
607 EXPORT_SYMBOL(end_page_fs_misc);
610 * find_get_page - find and get a page reference
611 * @mapping: the address_space to search
612 * @offset: the page index
614 * A rather lightweight function, finding and getting a reference to a
615 * hashed page atomically.
617 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
621 read_lock_irq(&mapping->tree_lock);
622 page = radix_tree_lookup(&mapping->page_tree, offset);
624 page_cache_get(page);
625 read_unlock_irq(&mapping->tree_lock);
628 EXPORT_SYMBOL(find_get_page);
631 * find_trylock_page - find and lock a page
632 * @mapping: the address_space to search
633 * @offset: the page index
635 * Same as find_get_page(), but trylock it instead of incrementing the count.
637 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
641 read_lock_irq(&mapping->tree_lock);
642 page = radix_tree_lookup(&mapping->page_tree, offset);
643 if (page && TestSetPageLocked(page))
645 read_unlock_irq(&mapping->tree_lock);
648 EXPORT_SYMBOL(find_trylock_page);
651 * find_lock_page - locate, pin and lock a pagecache page
652 * @mapping: the address_space to search
653 * @offset: the page index
655 * Locates the desired pagecache page, locks it, increments its reference
656 * count and returns its address.
658 * Returns zero if the page was not present. find_lock_page() may sleep.
660 struct page *find_lock_page(struct address_space *mapping,
661 unsigned long offset)
665 read_lock_irq(&mapping->tree_lock);
667 page = radix_tree_lookup(&mapping->page_tree, offset);
669 page_cache_get(page);
670 if (TestSetPageLocked(page)) {
671 read_unlock_irq(&mapping->tree_lock);
673 read_lock_irq(&mapping->tree_lock);
675 /* Has the page been truncated while we slept? */
676 if (unlikely(page->mapping != mapping ||
677 page->index != offset)) {
679 page_cache_release(page);
684 read_unlock_irq(&mapping->tree_lock);
687 EXPORT_SYMBOL(find_lock_page);
690 * find_or_create_page - locate or add a pagecache page
691 * @mapping: the page's address_space
692 * @index: the page's index into the mapping
693 * @gfp_mask: page allocation mode
695 * Locates a page in the pagecache. If the page is not present, a new page
696 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
697 * LRU list. The returned page is locked and has its reference count
700 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
703 * find_or_create_page() returns the desired page's address, or zero on
706 struct page *find_or_create_page(struct address_space *mapping,
707 unsigned long index, gfp_t gfp_mask)
709 struct page *page, *cached_page = NULL;
712 page = find_lock_page(mapping, index);
715 cached_page = alloc_page(gfp_mask);
719 err = add_to_page_cache_lru(cached_page, mapping,
724 } else if (err == -EEXIST)
728 page_cache_release(cached_page);
731 EXPORT_SYMBOL(find_or_create_page);
734 * find_get_pages - gang pagecache lookup
735 * @mapping: The address_space to search
736 * @start: The starting page index
737 * @nr_pages: The maximum number of pages
738 * @pages: Where the resulting pages are placed
740 * find_get_pages() will search for and return a group of up to
741 * @nr_pages pages in the mapping. The pages are placed at @pages.
742 * find_get_pages() takes a reference against the returned pages.
744 * The search returns a group of mapping-contiguous pages with ascending
745 * indexes. There may be holes in the indices due to not-present pages.
747 * find_get_pages() returns the number of pages which were found.
749 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
750 unsigned int nr_pages, struct page **pages)
755 read_lock_irq(&mapping->tree_lock);
756 ret = radix_tree_gang_lookup(&mapping->page_tree,
757 (void **)pages, start, nr_pages);
758 for (i = 0; i < ret; i++)
759 page_cache_get(pages[i]);
760 read_unlock_irq(&mapping->tree_lock);
765 * find_get_pages_contig - gang contiguous pagecache lookup
766 * @mapping: The address_space to search
767 * @index: The starting page index
768 * @nr_pages: The maximum number of pages
769 * @pages: Where the resulting pages are placed
771 * find_get_pages_contig() works exactly like find_get_pages(), except
772 * that the returned number of pages are guaranteed to be contiguous.
774 * find_get_pages_contig() returns the number of pages which were found.
776 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
777 unsigned int nr_pages, struct page **pages)
782 read_lock_irq(&mapping->tree_lock);
783 ret = radix_tree_gang_lookup(&mapping->page_tree,
784 (void **)pages, index, nr_pages);
785 for (i = 0; i < ret; i++) {
786 if (pages[i]->mapping == NULL || pages[i]->index != index)
789 page_cache_get(pages[i]);
792 read_unlock_irq(&mapping->tree_lock);
797 * find_get_pages_tag - find and return pages that match @tag
798 * @mapping: the address_space to search
799 * @index: the starting page index
800 * @tag: the tag index
801 * @nr_pages: the maximum number of pages
802 * @pages: where the resulting pages are placed
804 * Like find_get_pages, except we only return pages which are tagged with
805 * @tag. We update @index to index the next page for the traversal.
807 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
808 int tag, unsigned int nr_pages, struct page **pages)
813 read_lock_irq(&mapping->tree_lock);
814 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
815 (void **)pages, *index, nr_pages, tag);
816 for (i = 0; i < ret; i++)
817 page_cache_get(pages[i]);
819 *index = pages[ret - 1]->index + 1;
820 read_unlock_irq(&mapping->tree_lock);
825 * grab_cache_page_nowait - returns locked page at given index in given cache
826 * @mapping: target address_space
827 * @index: the page index
829 * Same as grab_cache_page, but do not wait if the page is unavailable.
830 * This is intended for speculative data generators, where the data can
831 * be regenerated if the page couldn't be grabbed. This routine should
832 * be safe to call while holding the lock for another page.
834 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
835 * and deadlock against the caller's locked page.
838 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
840 struct page *page = find_get_page(mapping, index);
844 if (!TestSetPageLocked(page))
846 page_cache_release(page);
849 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
850 page = alloc_pages(gfp_mask, 0);
851 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
852 page_cache_release(page);
857 EXPORT_SYMBOL(grab_cache_page_nowait);
860 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
861 * a _large_ part of the i/o request. Imagine the worst scenario:
863 * ---R__________________________________________B__________
864 * ^ reading here ^ bad block(assume 4k)
866 * read(R) => miss => readahead(R...B) => media error => frustrating retries
867 * => failing the whole request => read(R) => read(R+1) =>
868 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
869 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
870 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
872 * It is going insane. Fix it by quickly scaling down the readahead size.
874 static void shrink_readahead_size_eio(struct file *filp,
875 struct file_ra_state *ra)
884 * do_generic_mapping_read - generic file read routine
885 * @mapping: address_space to be read
886 * @_ra: file's readahead state
887 * @filp: the file to read
888 * @ppos: current file position
889 * @desc: read_descriptor
890 * @actor: read method
892 * This is a generic file read routine, and uses the
893 * mapping->a_ops->readpage() function for the actual low-level stuff.
895 * This is really ugly. But the goto's actually try to clarify some
896 * of the logic when it comes to error handling etc.
898 * Note the struct file* is only passed for the use of readpage.
901 void do_generic_mapping_read(struct address_space *mapping,
902 struct file_ra_state *_ra,
905 read_descriptor_t *desc,
909 struct inode *inode = mapping->host;
911 unsigned long end_index;
912 unsigned long offset;
913 unsigned long last_index;
914 unsigned long next_index;
915 unsigned long prev_index;
917 struct page *cached_page;
919 struct file_ra_state ra = *_ra;
922 index = *ppos >> PAGE_CACHE_SHIFT;
924 prev_index = ra.prev_page;
925 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
926 offset = *ppos & ~PAGE_CACHE_MASK;
928 isize = i_size_read(inode);
932 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
935 unsigned long nr, ret;
937 /* nr is the maximum number of bytes to copy from this page */
938 nr = PAGE_CACHE_SIZE;
939 if (index >= end_index) {
940 if (index > end_index)
942 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
950 if (index == next_index)
951 next_index = page_cache_readahead(mapping, &ra, filp,
952 index, last_index - index);
955 page = find_get_page(mapping, index);
956 if (unlikely(page == NULL)) {
958 desc->error = -EWOULDBLOCKIO;
961 handle_ra_miss(mapping, &ra, index);
964 if (!PageUptodate(page)) {
966 page_cache_release(page);
967 desc->error = -EWOULDBLOCKIO;
970 goto page_not_up_to_date;
974 /* If users can be writing to this page using arbitrary
975 * virtual addresses, take care about potential aliasing
976 * before reading the page on the kernel side.
978 if (mapping_writably_mapped(mapping))
979 flush_dcache_page(page);
982 * When (part of) the same page is read multiple times
983 * in succession, only mark it as accessed the first time.
985 if (prev_index != index)
986 mark_page_accessed(page);
990 * Ok, we have the page, and it's up-to-date, so
991 * now we can copy it to user space...
993 * The actor routine returns how many bytes were actually used..
994 * NOTE! This may not be the same as how much of a user buffer
995 * we filled up (we may be padding etc), so we can only update
996 * "pos" here (the actor routine has to update the user buffer
997 * pointers and the remaining count).
999 ret = actor(desc, page, offset, nr);
1001 index += offset >> PAGE_CACHE_SHIFT;
1002 offset &= ~PAGE_CACHE_MASK;
1004 page_cache_release(page);
1005 if (ret == nr && desc->count)
1009 page_not_up_to_date:
1010 /* Get exclusive access to the page ... */
1013 /* Did it get unhashed before we got the lock? */
1014 if (!page->mapping) {
1016 page_cache_release(page);
1020 /* Did somebody else fill it already? */
1021 if (PageUptodate(page)) {
1027 /* Start the actual read. The read will unlock the page. */
1028 error = mapping->a_ops->readpage(filp, page);
1030 if (unlikely(error)) {
1031 if (error == AOP_TRUNCATED_PAGE) {
1032 page_cache_release(page);
1035 goto readpage_error;
1038 if (!PageUptodate(page)) {
1040 if (!PageUptodate(page)) {
1041 if (page->mapping == NULL) {
1043 * invalidate_inode_pages got it
1046 page_cache_release(page);
1051 shrink_readahead_size_eio(filp, &ra);
1052 goto readpage_error;
1058 * i_size must be checked after we have done ->readpage.
1060 * Checking i_size after the readpage allows us to calculate
1061 * the correct value for "nr", which means the zero-filled
1062 * part of the page is not copied back to userspace (unless
1063 * another truncate extends the file - this is desired though).
1065 isize = i_size_read(inode);
1066 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1067 if (unlikely(!isize || index > end_index)) {
1068 page_cache_release(page);
1072 /* nr is the maximum number of bytes to copy from this page */
1073 nr = PAGE_CACHE_SIZE;
1074 if (index == end_index) {
1075 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1077 page_cache_release(page);
1085 /* UHHUH! A synchronous read error occurred. Report it */
1086 desc->error = error;
1087 page_cache_release(page);
1092 * Ok, it wasn't cached, so we need to create a new
1096 cached_page = page_cache_alloc_cold(mapping);
1098 desc->error = -ENOMEM;
1102 error = add_to_page_cache_lru(cached_page, mapping,
1105 if (error == -EEXIST)
1107 desc->error = error;
1118 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1120 page_cache_release(cached_page);
1122 file_accessed(filp);
1124 EXPORT_SYMBOL(do_generic_mapping_read);
1126 int file_read_actor(read_descriptor_t *desc, struct page *page,
1127 unsigned long offset, unsigned long size)
1130 unsigned long left, count = desc->count;
1136 * Faults on the destination of a read are common, so do it before
1139 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1140 kaddr = kmap_atomic(page, KM_USER0);
1141 left = __copy_to_user_inatomic(desc->arg.buf,
1142 kaddr + offset, size);
1143 kunmap_atomic(kaddr, KM_USER0);
1148 /* Do it the slow way */
1150 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1155 desc->error = -EFAULT;
1158 desc->count = count - size;
1159 desc->written += size;
1160 desc->arg.buf += size;
1165 * __generic_file_aio_read - generic filesystem read routine
1166 * @iocb: kernel I/O control block
1167 * @iov: io vector request
1168 * @nr_segs: number of segments in the iovec
1169 * @ppos: current file position
1171 * This is the "read()" routine for all filesystems
1172 * that can use the page cache directly.
1175 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1176 unsigned long nr_segs, loff_t *ppos)
1178 struct file *filp = iocb->ki_filp;
1184 for (seg = 0; seg < nr_segs; seg++) {
1185 const struct iovec *iv = &iov[seg];
1188 * If any segment has a negative length, or the cumulative
1189 * length ever wraps negative then return -EINVAL.
1191 count += iv->iov_len;
1192 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1194 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1199 count -= iv->iov_len; /* This segment is no good */
1203 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1204 if (filp->f_flags & O_DIRECT) {
1205 loff_t pos = *ppos, size;
1206 struct address_space *mapping;
1207 struct inode *inode;
1209 mapping = filp->f_mapping;
1210 inode = mapping->host;
1213 goto out; /* skip atime */
1214 size = i_size_read(inode);
1216 retval = generic_file_direct_IO(READ, iocb,
1218 if (retval > 0 && !is_sync_kiocb(iocb))
1219 retval = -EIOCBQUEUED;
1221 *ppos = pos + retval;
1223 file_accessed(filp);
1229 for (seg = 0; seg < nr_segs; seg++) {
1230 read_descriptor_t desc;
1233 desc.arg.buf = iov[seg].iov_base;
1234 desc.count = iov[seg].iov_len;
1235 if (desc.count == 0)
1238 do_generic_file_read(filp,ppos,&desc,file_read_actor,0);
1239 retval += desc.written;
1241 retval = retval ?: desc.error;
1249 EXPORT_SYMBOL(__generic_file_aio_read);
1252 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1254 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1256 BUG_ON(iocb->ki_pos != pos);
1257 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1259 EXPORT_SYMBOL(generic_file_aio_read);
1262 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1264 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1268 init_sync_kiocb(&kiocb, filp);
1269 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1270 if (-EIOCBQUEUED == ret)
1271 ret = wait_on_sync_kiocb(&kiocb);
1274 EXPORT_SYMBOL(generic_file_read);
1276 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1279 unsigned long count = desc->count;
1280 struct file *file = desc->arg.data;
1285 written = file->f_op->sendpage(file, page, offset,
1286 size, &file->f_pos, size<count);
1288 desc->error = written;
1291 desc->count = count - written;
1292 desc->written += written;
1296 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1297 size_t count, read_actor_t actor, void *target)
1299 read_descriptor_t desc;
1306 desc.arg.data = target;
1309 do_generic_file_read(in_file, ppos, &desc, actor, 0);
1311 return desc.written;
1314 EXPORT_SYMBOL(generic_file_sendfile);
1317 do_readahead(struct address_space *mapping, struct file *filp,
1318 unsigned long index, unsigned long nr)
1320 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1323 force_page_cache_readahead(mapping, filp, index,
1324 max_sane_readahead(nr));
1328 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1336 if (file->f_mode & FMODE_READ) {
1337 struct address_space *mapping = file->f_mapping;
1338 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1339 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1340 unsigned long len = end - start + 1;
1341 ret = do_readahead(mapping, file, start, len);
1349 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1351 * page_cache_read - adds requested page to the page cache if not already there
1352 * @file: file to read
1353 * @offset: page index
1355 * This adds the requested page to the page cache if it isn't already there,
1356 * and schedules an I/O to read in its contents from disk.
1358 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1360 struct address_space *mapping = file->f_mapping;
1365 page = page_cache_alloc_cold(mapping);
1369 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1371 ret = mapping->a_ops->readpage(file, page);
1372 else if (ret == -EEXIST)
1373 ret = 0; /* losing race to add is OK */
1375 page_cache_release(page);
1377 } while (ret == AOP_TRUNCATED_PAGE);
1382 #define MMAP_LOTSAMISS (100)
1385 * filemap_nopage - read in file data for page fault handling
1386 * @area: the applicable vm_area
1387 * @address: target address to read in
1388 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1390 * filemap_nopage() is invoked via the vma operations vector for a
1391 * mapped memory region to read in file data during a page fault.
1393 * The goto's are kind of ugly, but this streamlines the normal case of having
1394 * it in the page cache, and handles the special cases reasonably without
1395 * having a lot of duplicated code.
1397 struct page *filemap_nopage(struct vm_area_struct *area,
1398 unsigned long address, int *type)
1401 struct file *file = area->vm_file;
1402 struct address_space *mapping = file->f_mapping;
1403 struct file_ra_state *ra = &file->f_ra;
1404 struct inode *inode = mapping->host;
1406 unsigned long size, pgoff;
1407 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1409 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1412 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1414 goto outside_data_content;
1416 /* If we don't want any read-ahead, don't bother */
1417 if (VM_RandomReadHint(area))
1418 goto no_cached_page;
1421 * The readahead code wants to be told about each and every page
1422 * so it can build and shrink its windows appropriately
1424 * For sequential accesses, we use the generic readahead logic.
1426 if (VM_SequentialReadHint(area))
1427 page_cache_readahead(mapping, ra, file, pgoff, 1);
1430 * Do we have something in the page cache already?
1433 page = find_get_page(mapping, pgoff);
1435 unsigned long ra_pages;
1437 if (VM_SequentialReadHint(area)) {
1438 handle_ra_miss(mapping, ra, pgoff);
1439 goto no_cached_page;
1444 * Do we miss much more than hit in this file? If so,
1445 * stop bothering with read-ahead. It will only hurt.
1447 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1448 goto no_cached_page;
1451 * To keep the pgmajfault counter straight, we need to
1452 * check did_readaround, as this is an inner loop.
1454 if (!did_readaround) {
1455 majmin = VM_FAULT_MAJOR;
1456 count_vm_event(PGMAJFAULT);
1459 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1463 if (pgoff > ra_pages / 2)
1464 start = pgoff - ra_pages / 2;
1465 do_page_cache_readahead(mapping, file, start, ra_pages);
1467 page = find_get_page(mapping, pgoff);
1469 goto no_cached_page;
1472 if (!did_readaround)
1476 * Ok, found a page in the page cache, now we need to check
1477 * that it's up-to-date.
1479 if (!PageUptodate(page))
1480 goto page_not_uptodate;
1484 * Found the page and have a reference on it.
1486 mark_page_accessed(page);
1491 outside_data_content:
1493 * An external ptracer can access pages that normally aren't
1496 if (area->vm_mm == current->mm)
1498 /* Fall through to the non-read-ahead case */
1501 * We're only likely to ever get here if MADV_RANDOM is in
1504 error = page_cache_read(file, pgoff);
1508 * The page we want has now been added to the page cache.
1509 * In the unlikely event that someone removed it in the
1510 * meantime, we'll just come back here and read it again.
1516 * An error return from page_cache_read can result if the
1517 * system is low on memory, or a problem occurs while trying
1520 if (error == -ENOMEM)
1525 if (!did_readaround) {
1526 majmin = VM_FAULT_MAJOR;
1527 count_vm_event(PGMAJFAULT);
1531 /* Did it get unhashed while we waited for it? */
1532 if (!page->mapping) {
1534 page_cache_release(page);
1538 /* Did somebody else get it up-to-date? */
1539 if (PageUptodate(page)) {
1544 error = mapping->a_ops->readpage(file, page);
1546 wait_on_page_locked(page);
1547 if (PageUptodate(page))
1549 } else if (error == AOP_TRUNCATED_PAGE) {
1550 page_cache_release(page);
1555 * Umm, take care of errors if the page isn't up-to-date.
1556 * Try to re-read it _once_. We do this synchronously,
1557 * because there really aren't any performance issues here
1558 * and we need to check for errors.
1562 /* Somebody truncated the page on us? */
1563 if (!page->mapping) {
1565 page_cache_release(page);
1569 /* Somebody else successfully read it in? */
1570 if (PageUptodate(page)) {
1574 ClearPageError(page);
1575 error = mapping->a_ops->readpage(file, page);
1577 wait_on_page_locked(page);
1578 if (PageUptodate(page))
1580 } else if (error == AOP_TRUNCATED_PAGE) {
1581 page_cache_release(page);
1586 * Things didn't work out. Return zero to tell the
1587 * mm layer so, possibly freeing the page cache page first.
1589 shrink_readahead_size_eio(file, ra);
1590 page_cache_release(page);
1593 EXPORT_SYMBOL(filemap_nopage);
1595 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1598 struct address_space *mapping = file->f_mapping;
1603 * Do we have something in the page cache already?
1606 page = find_get_page(mapping, pgoff);
1610 goto no_cached_page;
1614 * Ok, found a page in the page cache, now we need to check
1615 * that it's up-to-date.
1617 if (!PageUptodate(page)) {
1619 page_cache_release(page);
1622 goto page_not_uptodate;
1627 * Found the page and have a reference on it.
1629 mark_page_accessed(page);
1633 error = page_cache_read(file, pgoff);
1636 * The page we want has now been added to the page cache.
1637 * In the unlikely event that someone removed it in the
1638 * meantime, we'll just come back here and read it again.
1644 * An error return from page_cache_read can result if the
1645 * system is low on memory, or a problem occurs while trying
1653 /* Did it get unhashed while we waited for it? */
1654 if (!page->mapping) {
1659 /* Did somebody else get it up-to-date? */
1660 if (PageUptodate(page)) {
1665 error = mapping->a_ops->readpage(file, page);
1667 wait_on_page_locked(page);
1668 if (PageUptodate(page))
1670 } else if (error == AOP_TRUNCATED_PAGE) {
1671 page_cache_release(page);
1676 * Umm, take care of errors if the page isn't up-to-date.
1677 * Try to re-read it _once_. We do this synchronously,
1678 * because there really aren't any performance issues here
1679 * and we need to check for errors.
1683 /* Somebody truncated the page on us? */
1684 if (!page->mapping) {
1688 /* Somebody else successfully read it in? */
1689 if (PageUptodate(page)) {
1694 ClearPageError(page);
1695 error = mapping->a_ops->readpage(file, page);
1697 wait_on_page_locked(page);
1698 if (PageUptodate(page))
1700 } else if (error == AOP_TRUNCATED_PAGE) {
1701 page_cache_release(page);
1706 * Things didn't work out. Return zero to tell the
1707 * mm layer so, possibly freeing the page cache page first.
1710 page_cache_release(page);
1715 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1716 unsigned long len, pgprot_t prot, unsigned long pgoff,
1719 struct file *file = vma->vm_file;
1720 struct address_space *mapping = file->f_mapping;
1721 struct inode *inode = mapping->host;
1723 struct mm_struct *mm = vma->vm_mm;
1728 force_page_cache_readahead(mapping, vma->vm_file,
1729 pgoff, len >> PAGE_CACHE_SHIFT);
1732 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1733 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1736 page = filemap_getpage(file, pgoff, nonblock);
1738 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1739 * done in shmem_populate calling shmem_getpage */
1740 if (!page && !nonblock)
1744 err = install_page(mm, vma, addr, page, prot);
1746 page_cache_release(page);
1749 } else if (vma->vm_flags & VM_NONLINEAR) {
1750 /* No page was found just because we can't read it in now (being
1751 * here implies nonblock != 0), but the page may exist, so set
1752 * the PTE to fault it in later. */
1753 err = install_file_pte(mm, vma, addr, pgoff, prot);
1766 EXPORT_SYMBOL(filemap_populate);
1768 struct vm_operations_struct generic_file_vm_ops = {
1769 .nopage = filemap_nopage,
1770 .populate = filemap_populate,
1773 /* This is used for a general mmap of a disk file */
1775 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1777 struct address_space *mapping = file->f_mapping;
1779 if (!mapping->a_ops->readpage)
1781 file_accessed(file);
1782 vma->vm_ops = &generic_file_vm_ops;
1787 * This is for filesystems which do not implement ->writepage.
1789 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1791 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1793 return generic_file_mmap(file, vma);
1796 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1800 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1804 #endif /* CONFIG_MMU */
1806 EXPORT_SYMBOL(generic_file_mmap);
1807 EXPORT_SYMBOL(generic_file_readonly_mmap);
1809 static inline struct page *__read_cache_page(struct address_space *mapping,
1810 unsigned long index,
1811 int (*filler)(void *,struct page*),
1814 struct page *page, *cached_page = NULL;
1817 page = find_get_page(mapping, index);
1820 cached_page = page_cache_alloc_cold(mapping);
1822 return ERR_PTR(-ENOMEM);
1824 err = add_to_page_cache_lru(cached_page, mapping,
1829 /* Presumably ENOMEM for radix tree node */
1830 page_cache_release(cached_page);
1831 return ERR_PTR(err);
1835 err = filler(data, page);
1837 page_cache_release(page);
1838 page = ERR_PTR(err);
1842 page_cache_release(cached_page);
1847 * read_cache_page - read into page cache, fill it if needed
1848 * @mapping: the page's address_space
1849 * @index: the page index
1850 * @filler: function to perform the read
1851 * @data: destination for read data
1853 * Read into the page cache. If a page already exists,
1854 * and PageUptodate() is not set, try to fill the page.
1856 struct page *read_cache_page(struct address_space *mapping,
1857 unsigned long index,
1858 int (*filler)(void *,struct page*),
1865 page = __read_cache_page(mapping, index, filler, data);
1868 mark_page_accessed(page);
1869 if (PageUptodate(page))
1873 if (!page->mapping) {
1875 page_cache_release(page);
1878 if (PageUptodate(page)) {
1882 err = filler(data, page);
1884 page_cache_release(page);
1885 page = ERR_PTR(err);
1890 EXPORT_SYMBOL(read_cache_page);
1893 * If the page was newly created, increment its refcount and add it to the
1894 * caller's lru-buffering pagevec. This function is specifically for
1895 * generic_file_write().
1897 static inline struct page *
1898 __grab_cache_page(struct address_space *mapping, unsigned long index,
1899 struct page **cached_page, struct pagevec *lru_pvec)
1904 page = find_lock_page(mapping, index);
1906 if (!*cached_page) {
1907 *cached_page = page_cache_alloc(mapping);
1911 err = add_to_page_cache(*cached_page, mapping,
1916 page = *cached_page;
1917 page_cache_get(page);
1918 if (!pagevec_add(lru_pvec, page))
1919 __pagevec_lru_add(lru_pvec);
1920 *cached_page = NULL;
1927 * The logic we want is
1929 * if suid or (sgid and xgrp)
1932 int remove_suid(struct dentry *dentry)
1934 mode_t mode = dentry->d_inode->i_mode;
1938 /* suid always must be killed */
1939 if (unlikely(mode & S_ISUID))
1940 kill = ATTR_KILL_SUID;
1943 * sgid without any exec bits is just a mandatory locking mark; leave
1944 * it alone. If some exec bits are set, it's a real sgid; kill it.
1946 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1947 kill |= ATTR_KILL_SGID;
1949 if (unlikely(kill && !capable(CAP_FSETID))) {
1950 struct iattr newattrs;
1952 newattrs.ia_valid = ATTR_FORCE | kill;
1953 result = notify_change(dentry, &newattrs);
1957 EXPORT_SYMBOL(remove_suid);
1960 __filemap_copy_from_user_iovec_inatomic(char *vaddr,
1961 const struct iovec *iov, size_t base, size_t bytes)
1963 size_t copied = 0, left = 0;
1966 char __user *buf = iov->iov_base + base;
1967 int copy = min(bytes, iov->iov_len - base);
1970 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1979 return copied - left;
1983 * Performs necessary checks before doing a write
1985 * Can adjust writing position or amount of bytes to write.
1986 * Returns appropriate error code that caller should return or
1987 * zero in case that write should be allowed.
1989 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1991 struct inode *inode = file->f_mapping->host;
1992 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1994 if (unlikely(*pos < 0))
1998 /* FIXME: this is for backwards compatibility with 2.4 */
1999 if (file->f_flags & O_APPEND)
2000 *pos = i_size_read(inode);
2002 if (limit != RLIM_INFINITY) {
2003 if (*pos >= limit) {
2004 send_sig(SIGXFSZ, current, 0);
2007 if (*count > limit - (typeof(limit))*pos) {
2008 *count = limit - (typeof(limit))*pos;
2016 if (unlikely(*pos + *count > MAX_NON_LFS &&
2017 !(file->f_flags & O_LARGEFILE))) {
2018 if (*pos >= MAX_NON_LFS) {
2019 send_sig(SIGXFSZ, current, 0);
2022 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2023 *count = MAX_NON_LFS - (unsigned long)*pos;
2028 * Are we about to exceed the fs block limit ?
2030 * If we have written data it becomes a short write. If we have
2031 * exceeded without writing data we send a signal and return EFBIG.
2032 * Linus frestrict idea will clean these up nicely..
2034 if (likely(!isblk)) {
2035 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2036 if (*count || *pos > inode->i_sb->s_maxbytes) {
2037 send_sig(SIGXFSZ, current, 0);
2040 /* zero-length writes at ->s_maxbytes are OK */
2043 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2044 *count = inode->i_sb->s_maxbytes - *pos;
2047 if (bdev_read_only(I_BDEV(inode)))
2049 isize = i_size_read(inode);
2050 if (*pos >= isize) {
2051 if (*count || *pos > isize)
2055 if (*pos + *count > isize)
2056 *count = isize - *pos;
2060 EXPORT_SYMBOL(generic_write_checks);
2063 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2064 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2065 size_t count, size_t ocount)
2067 struct file *file = iocb->ki_filp;
2068 struct address_space *mapping = file->f_mapping;
2069 struct inode *inode = mapping->host;
2072 if (count != ocount)
2073 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2075 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2077 loff_t end = pos + written;
2078 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2079 i_size_write(inode, end);
2080 mark_inode_dirty(inode);
2086 * Sync the fs metadata but not the minor inode changes and
2087 * of course not the data as we did direct DMA for the IO.
2088 * i_mutex is held, which protects generic_osync_inode() from
2091 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2092 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2096 if (written == count && !is_sync_kiocb(iocb))
2097 written = -EIOCBQUEUED;
2100 EXPORT_SYMBOL(generic_file_direct_write);
2103 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2104 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2105 size_t count, ssize_t written)
2107 struct file *file = iocb->ki_filp;
2108 struct address_space * mapping = file->f_mapping;
2109 const struct address_space_operations *a_ops = mapping->a_ops;
2110 struct inode *inode = mapping->host;
2113 struct page *cached_page = NULL;
2115 struct pagevec lru_pvec;
2116 const struct iovec *cur_iov = iov; /* current iovec */
2117 size_t iov_base = 0; /* offset in the current iovec */
2120 pagevec_init(&lru_pvec, 0);
2123 * handle partial DIO write. Adjust cur_iov if needed.
2125 if (likely(nr_segs == 1))
2126 buf = iov->iov_base + written;
2128 filemap_set_next_iovec(&cur_iov, &iov_base, written);
2129 buf = cur_iov->iov_base + iov_base;
2133 unsigned long index;
2134 unsigned long offset;
2137 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2138 index = pos >> PAGE_CACHE_SHIFT;
2139 bytes = PAGE_CACHE_SIZE - offset;
2141 /* Limit the size of the copy to the caller's write size */
2142 bytes = min(bytes, count);
2145 * Limit the size of the copy to that of the current segment,
2146 * because fault_in_pages_readable() doesn't know how to walk
2149 bytes = min(bytes, cur_iov->iov_len - iov_base);
2152 * Bring in the user page that we will copy from _first_.
2153 * Otherwise there's a nasty deadlock on copying from the
2154 * same page as we're writing to, without it being marked
2157 fault_in_pages_readable(buf, bytes);
2159 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2165 if (unlikely(bytes == 0)) {
2168 goto zero_length_segment;
2171 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2172 if (unlikely(status)) {
2173 loff_t isize = i_size_read(inode);
2175 if (status != AOP_TRUNCATED_PAGE)
2177 page_cache_release(page);
2178 if (status == AOP_TRUNCATED_PAGE)
2181 * prepare_write() may have instantiated a few blocks
2182 * outside i_size. Trim these off again.
2184 if (pos + bytes > isize)
2185 vmtruncate(inode, isize);
2188 if (likely(nr_segs == 1))
2189 copied = filemap_copy_from_user(page, offset,
2192 copied = filemap_copy_from_user_iovec(page, offset,
2193 cur_iov, iov_base, bytes);
2194 flush_dcache_page(page);
2195 status = a_ops->commit_write(file, page, offset, offset+bytes);
2196 if (status == AOP_TRUNCATED_PAGE) {
2197 page_cache_release(page);
2200 zero_length_segment:
2201 if (likely(copied >= 0)) {
2210 if (unlikely(nr_segs > 1)) {
2211 filemap_set_next_iovec(&cur_iov,
2214 buf = cur_iov->iov_base +
2221 if (unlikely(copied != bytes))
2225 mark_page_accessed(page);
2226 page_cache_release(page);
2229 balance_dirty_pages_ratelimited(mapping);
2235 page_cache_release(cached_page);
2238 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2240 if (likely(status >= 0)) {
2241 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2242 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2243 status = generic_osync_inode(inode, mapping,
2244 OSYNC_METADATA|OSYNC_DATA);
2249 * If we get here for O_DIRECT writes then we must have fallen through
2250 * to buffered writes (block instantiation inside i_size). So we sync
2251 * the file data here, to try to honour O_DIRECT expectations.
2253 if (unlikely(file->f_flags & O_DIRECT) && written)
2254 status = filemap_write_and_wait(mapping);
2256 pagevec_lru_add(&lru_pvec);
2257 return written ? written : status;
2259 EXPORT_SYMBOL(generic_file_buffered_write);
2262 * This writes the data from the source page to the specified page offset in
2263 * the nominated file
2264 * - the source page does not need to have any association with the file or the
2268 generic_file_buffered_write_one_kernel_page(struct address_space *mapping,
2272 const struct address_space_operations *a_ops = mapping->a_ops;
2273 struct pagevec lru_pvec;
2274 struct page *page, *cached_page = NULL;
2277 pagevec_init(&lru_pvec, 0);
2280 if (mapping->tree_lock.magic != RWLOCK_MAGIC)
2281 printk("RWLOCK magic incorrect: %x != %x\n",
2282 mapping->tree_lock.magic, RWLOCK_MAGIC);
2285 page = __grab_cache_page(mapping, index, &cached_page, &lru_pvec);
2287 BUG_ON(cached_page);
2291 status = a_ops->prepare_write(NULL, page, 0, PAGE_CACHE_SIZE);
2292 if (unlikely(status)) {
2293 loff_t isize = i_size_read(mapping->host);
2295 if (status != AOP_TRUNCATED_PAGE)
2297 page_cache_release(page);
2298 if (status == AOP_TRUNCATED_PAGE)
2301 /* prepare_write() may have instantiated a few blocks outside
2302 * i_size. Trim these off again.
2304 if ((1ULL << (index + 1)) > isize)
2305 vmtruncate(mapping->host, isize);
2309 copy_highpage(page, src);
2310 flush_dcache_page(page);
2312 status = a_ops->commit_write(NULL, page, 0, PAGE_CACHE_SIZE);
2313 if (status == AOP_TRUNCATED_PAGE) {
2314 page_cache_release(page);
2322 mark_page_accessed(page);
2323 page_cache_release(page);
2327 balance_dirty_pages_ratelimited(mapping);
2332 page_cache_release(cached_page);
2334 /* the caller must handle O_SYNC themselves, but we handle S_SYNC and
2335 * MS_SYNCHRONOUS here */
2336 if (unlikely(IS_SYNC(mapping->host)) && !a_ops->writepage)
2337 status = generic_osync_inode(mapping->host, mapping,
2338 OSYNC_METADATA | OSYNC_DATA);
2340 /* the caller must handle O_DIRECT for themselves */
2342 pagevec_lru_add(&lru_pvec);
2345 EXPORT_SYMBOL(generic_file_buffered_write_one_kernel_page);
2348 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2349 unsigned long nr_segs, loff_t *ppos)
2351 struct file *file = iocb->ki_filp;
2352 const struct address_space * mapping = file->f_mapping;
2353 size_t ocount; /* original count */
2354 size_t count; /* after file limit checks */
2355 struct inode *inode = mapping->host;
2362 for (seg = 0; seg < nr_segs; seg++) {
2363 const struct iovec *iv = &iov[seg];
2366 * If any segment has a negative length, or the cumulative
2367 * length ever wraps negative then return -EINVAL.
2369 ocount += iv->iov_len;
2370 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2372 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2377 ocount -= iv->iov_len; /* This segment is no good */
2384 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2386 /* We can write back this queue in page reclaim */
2387 current->backing_dev_info = mapping->backing_dev_info;
2390 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2397 err = remove_suid(file->f_dentry);
2401 file_update_time(file);
2403 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2404 if (unlikely(file->f_flags & O_DIRECT)) {
2405 written = generic_file_direct_write(iocb, iov,
2406 &nr_segs, pos, ppos, count, ocount);
2407 if (written < 0 || written == count)
2410 * direct-io write to a hole: fall through to buffered I/O
2411 * for completing the rest of the request.
2417 written = generic_file_buffered_write(iocb, iov, nr_segs,
2418 pos, ppos, count, written);
2420 current->backing_dev_info = NULL;
2421 return written ? written : err;
2423 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2426 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2427 unsigned long nr_segs, loff_t *ppos)
2429 struct file *file = iocb->ki_filp;
2430 struct address_space *mapping = file->f_mapping;
2431 struct inode *inode = mapping->host;
2435 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2437 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2440 err = sync_page_range_nolock(inode, mapping, pos, ret);
2448 __generic_file_write_nolock(struct file *file, const struct iovec *iov,
2449 unsigned long nr_segs, loff_t *ppos)
2454 init_sync_kiocb(&kiocb, file);
2455 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2456 if (ret == -EIOCBQUEUED)
2457 ret = wait_on_sync_kiocb(&kiocb);
2462 generic_file_write_nolock(struct file *file, const struct iovec *iov,
2463 unsigned long nr_segs, loff_t *ppos)
2468 init_sync_kiocb(&kiocb, file);
2469 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2470 if (-EIOCBQUEUED == ret)
2471 ret = wait_on_sync_kiocb(&kiocb);
2474 EXPORT_SYMBOL(generic_file_write_nolock);
2476 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2477 size_t count, loff_t pos)
2479 struct file *file = iocb->ki_filp;
2480 struct address_space *mapping = file->f_mapping;
2481 struct inode *inode = mapping->host;
2483 struct iovec local_iov = { .iov_base = (void __user *)buf,
2486 BUG_ON(iocb->ki_pos != pos);
2488 mutex_lock(&inode->i_mutex);
2489 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2491 mutex_unlock(&inode->i_mutex);
2493 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2496 err = sync_page_range(inode, mapping, pos, ret);
2502 EXPORT_SYMBOL(generic_file_aio_write);
2504 ssize_t generic_file_write(struct file *file, const char __user *buf,
2505 size_t count, loff_t *ppos)
2507 struct address_space *mapping = file->f_mapping;
2508 struct inode *inode = mapping->host;
2510 struct iovec local_iov = { .iov_base = (void __user *)buf,
2513 mutex_lock(&inode->i_mutex);
2514 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
2515 mutex_unlock(&inode->i_mutex);
2517 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2520 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2526 EXPORT_SYMBOL(generic_file_write);
2528 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2529 unsigned long nr_segs, loff_t *ppos)
2534 init_sync_kiocb(&kiocb, filp);
2535 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2536 if (-EIOCBQUEUED == ret)
2537 ret = wait_on_sync_kiocb(&kiocb);
2540 EXPORT_SYMBOL(generic_file_readv);
2542 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2543 unsigned long nr_segs, loff_t *ppos)
2545 struct address_space *mapping = file->f_mapping;
2546 struct inode *inode = mapping->host;
2549 mutex_lock(&inode->i_mutex);
2550 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
2551 mutex_unlock(&inode->i_mutex);
2553 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2556 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2562 EXPORT_SYMBOL(generic_file_writev);
2565 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2566 * went wrong during pagecache shootdown.
2569 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2570 loff_t offset, unsigned long nr_segs)
2572 struct file *file = iocb->ki_filp;
2573 struct address_space *mapping = file->f_mapping;
2575 size_t write_len = 0;
2578 * If it's a write, unmap all mmappings of the file up-front. This
2579 * will cause any pte dirty bits to be propagated into the pageframes
2580 * for the subsequent filemap_write_and_wait().
2583 write_len = iov_length(iov, nr_segs);
2584 if (mapping_mapped(mapping))
2585 unmap_mapping_range(mapping, offset, write_len, 0);
2588 retval = filemap_write_and_wait(mapping);
2590 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2592 if (rw == WRITE && mapping->nrpages) {
2593 pgoff_t end = (offset + write_len - 1)
2594 >> PAGE_CACHE_SHIFT;
2595 int err = invalidate_inode_pages2_range(mapping,
2596 offset >> PAGE_CACHE_SHIFT, end);