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/config.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/compiler.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>
34 * FIXME: remove all knowledge of the buffer layer from the core VM
36 #include <linux/buffer_head.h> /* for generic_osync_inode */
38 #include <asm/uaccess.h>
42 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
43 loff_t offset, unsigned long nr_segs);
46 * Shared mappings implemented 30.11.1994. It's not fully working yet,
49 * Shared mappings now work. 15.8.1995 Bruno.
51 * finished 'unifying' the page and buffer cache and SMP-threaded the
52 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
54 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 * ->i_mmap_lock (vmtruncate)
61 * ->private_lock (__free_pte->__set_page_dirty_buffers)
62 * ->swap_lock (exclusive_swap_page, others)
63 * ->mapping->tree_lock
66 * ->i_mmap_lock (truncate->unmap_mapping_range)
70 * ->page_table_lock or pte_lock (various, mainly in memory.c)
71 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
74 * ->lock_page (access_process_vm)
80 * ->i_alloc_sem (various)
83 * ->sb_lock (fs/fs-writeback.c)
84 * ->mapping->tree_lock (__sync_single_inode)
87 * ->anon_vma.lock (vma_adjust)
90 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
92 * ->page_table_lock or pte_lock
93 * ->swap_lock (try_to_unmap_one)
94 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
97 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
98 * ->private_lock (page_remove_rmap->set_page_dirty)
99 * ->tree_lock (page_remove_rmap->set_page_dirty)
100 * ->inode_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (zap_pte_range->set_page_dirty)
102 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 * ->dcache_lock (proc_pid_lookup)
109 * Remove a page from the page cache and free it. Caller has to make
110 * sure the page is locked and that nobody else uses it - or that usage
111 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
113 void __remove_from_page_cache(struct page *page)
115 struct address_space *mapping = page->mapping;
117 radix_tree_delete(&mapping->page_tree, page->index);
118 page->mapping = NULL;
123 void remove_from_page_cache(struct page *page)
125 struct address_space *mapping = page->mapping;
127 BUG_ON(!PageLocked(page));
129 write_lock_irq(&mapping->tree_lock);
130 __remove_from_page_cache(page);
131 write_unlock_irq(&mapping->tree_lock);
134 static int sync_page(void *word)
136 struct address_space *mapping;
139 page = container_of((unsigned long *)word, struct page, flags);
142 * page_mapping() is being called without PG_locked held.
143 * Some knowledge of the state and use of the page is used to
144 * reduce the requirements down to a memory barrier.
145 * The danger here is of a stale page_mapping() return value
146 * indicating a struct address_space different from the one it's
147 * associated with when it is associated with one.
148 * After smp_mb(), it's either the correct page_mapping() for
149 * the page, or an old page_mapping() and the page's own
150 * page_mapping() has gone NULL.
151 * The ->sync_page() address_space operation must tolerate
152 * page_mapping() going NULL. By an amazing coincidence,
153 * this comes about because none of the users of the page
154 * in the ->sync_page() methods make essential use of the
155 * page_mapping(), merely passing the page down to the backing
156 * device's unplug functions when it's non-NULL, which in turn
157 * ignore it for all cases but swap, where only page_private(page) is
158 * of interest. When page_mapping() does go NULL, the entire
159 * call stack gracefully ignores the page and returns.
163 mapping = page_mapping(page);
164 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
165 mapping->a_ops->sync_page(page);
171 * filemap_fdatawrite_range - start writeback against all of a mapping's
172 * dirty pages that lie within the byte offsets <start, end>
173 * @mapping: address space structure to write
174 * @start: offset in bytes where the range starts
175 * @end: offset in bytes where the range ends
176 * @sync_mode: enable synchronous operation
178 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
179 * opposed to a regular memory * cleansing writeback. The difference between
180 * these two operations is that if a dirty page/buffer is encountered, it must
181 * be waited upon, and not just skipped over.
183 static int __filemap_fdatawrite_range(struct address_space *mapping,
184 loff_t start, loff_t end, int sync_mode)
187 struct writeback_control wbc = {
188 .sync_mode = sync_mode,
189 .nr_to_write = mapping->nrpages * 2,
194 if (!mapping_cap_writeback_dirty(mapping))
197 ret = do_writepages(mapping, &wbc);
201 static inline int __filemap_fdatawrite(struct address_space *mapping,
204 return __filemap_fdatawrite_range(mapping, 0, 0, sync_mode);
207 int filemap_fdatawrite(struct address_space *mapping)
209 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
211 EXPORT_SYMBOL(filemap_fdatawrite);
213 static int filemap_fdatawrite_range(struct address_space *mapping,
214 loff_t start, loff_t end)
216 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
220 * This is a mostly non-blocking flush. Not suitable for data-integrity
221 * purposes - I/O may not be started against all dirty pages.
223 int filemap_flush(struct address_space *mapping)
225 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
227 EXPORT_SYMBOL(filemap_flush);
230 * Wait for writeback to complete against pages indexed by start->end
233 static int wait_on_page_writeback_range(struct address_space *mapping,
234 pgoff_t start, pgoff_t end)
244 pagevec_init(&pvec, 0);
246 while ((index <= end) &&
247 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
248 PAGECACHE_TAG_WRITEBACK,
249 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
252 for (i = 0; i < nr_pages; i++) {
253 struct page *page = pvec.pages[i];
255 /* until radix tree lookup accepts end_index */
256 if (page->index > end)
259 wait_on_page_writeback(page);
263 pagevec_release(&pvec);
267 /* Check for outstanding write errors */
268 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
270 if (test_and_clear_bit(AS_EIO, &mapping->flags))
277 * Write and wait upon all the pages in the passed range. This is a "data
278 * integrity" operation. It waits upon in-flight writeout before starting and
279 * waiting upon new writeout. If there was an IO error, return it.
281 * We need to re-take i_mutex during the generic_osync_inode list walk because
282 * it is otherwise livelockable.
284 int sync_page_range(struct inode *inode, struct address_space *mapping,
285 loff_t pos, loff_t count)
287 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
288 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
291 if (!mapping_cap_writeback_dirty(mapping) || !count)
293 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
295 mutex_lock(&inode->i_mutex);
296 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
297 mutex_unlock(&inode->i_mutex);
300 ret = wait_on_page_writeback_range(mapping, start, end);
303 EXPORT_SYMBOL(sync_page_range);
306 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
307 * as it forces O_SYNC writers to different parts of the same file
308 * to be serialised right until io completion.
310 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
311 loff_t pos, loff_t count)
313 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
314 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
317 if (!mapping_cap_writeback_dirty(mapping) || !count)
319 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
321 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
323 ret = wait_on_page_writeback_range(mapping, start, end);
326 EXPORT_SYMBOL(sync_page_range_nolock);
329 * filemap_fdatawait - walk the list of under-writeback pages of the given
330 * address space and wait for all of them.
332 * @mapping: address space structure to wait for
334 int filemap_fdatawait(struct address_space *mapping)
336 loff_t i_size = i_size_read(mapping->host);
341 return wait_on_page_writeback_range(mapping, 0,
342 (i_size - 1) >> PAGE_CACHE_SHIFT);
344 EXPORT_SYMBOL(filemap_fdatawait);
346 int filemap_write_and_wait(struct address_space *mapping)
350 if (mapping->nrpages) {
351 err = filemap_fdatawrite(mapping);
353 * Even if the above returned error, the pages may be
354 * written partially (e.g. -ENOSPC), so we wait for it.
355 * But the -EIO is special case, it may indicate the worst
356 * thing (e.g. bug) happened, so we avoid waiting for it.
359 int err2 = filemap_fdatawait(mapping);
366 EXPORT_SYMBOL(filemap_write_and_wait);
368 int filemap_write_and_wait_range(struct address_space *mapping,
369 loff_t lstart, loff_t lend)
373 if (mapping->nrpages) {
374 err = __filemap_fdatawrite_range(mapping, lstart, lend,
376 /* See comment of filemap_write_and_wait() */
378 int err2 = wait_on_page_writeback_range(mapping,
379 lstart >> PAGE_CACHE_SHIFT,
380 lend >> PAGE_CACHE_SHIFT);
389 * This function is used to add newly allocated pagecache pages:
390 * the page is new, so we can just run SetPageLocked() against it.
391 * The other page state flags were set by rmqueue().
393 * This function does not add the page to the LRU. The caller must do that.
395 int add_to_page_cache(struct page *page, struct address_space *mapping,
396 pgoff_t offset, gfp_t gfp_mask)
398 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
401 write_lock_irq(&mapping->tree_lock);
402 error = radix_tree_insert(&mapping->page_tree, offset, page);
404 page_cache_get(page);
406 page->mapping = mapping;
407 page->index = offset;
411 write_unlock_irq(&mapping->tree_lock);
412 radix_tree_preload_end();
417 EXPORT_SYMBOL(add_to_page_cache);
419 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
420 pgoff_t offset, gfp_t gfp_mask)
422 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
429 * In order to wait for pages to become available there must be
430 * waitqueues associated with pages. By using a hash table of
431 * waitqueues where the bucket discipline is to maintain all
432 * waiters on the same queue and wake all when any of the pages
433 * become available, and for the woken contexts to check to be
434 * sure the appropriate page became available, this saves space
435 * at a cost of "thundering herd" phenomena during rare hash
438 static wait_queue_head_t *page_waitqueue(struct page *page)
440 const struct zone *zone = page_zone(page);
442 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
445 static inline void wake_up_page(struct page *page, int bit)
447 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
450 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
452 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
454 if (test_bit(bit_nr, &page->flags))
455 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
456 TASK_UNINTERRUPTIBLE);
458 EXPORT_SYMBOL(wait_on_page_bit);
461 * unlock_page() - unlock a locked page
465 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
466 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
467 * mechananism between PageLocked pages and PageWriteback pages is shared.
468 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
470 * The first mb is necessary to safely close the critical section opened by the
471 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
472 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
473 * parallel wait_on_page_locked()).
475 void fastcall unlock_page(struct page *page)
477 smp_mb__before_clear_bit();
478 if (!TestClearPageLocked(page))
480 smp_mb__after_clear_bit();
481 wake_up_page(page, PG_locked);
483 EXPORT_SYMBOL(unlock_page);
486 * End writeback against a page.
488 void end_page_writeback(struct page *page)
490 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
491 if (!test_clear_page_writeback(page))
494 smp_mb__after_clear_bit();
495 wake_up_page(page, PG_writeback);
497 EXPORT_SYMBOL(end_page_writeback);
500 * Get a lock on the page, assuming we need to sleep to get it.
502 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
503 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
504 * chances are that on the second loop, the block layer's plug list is empty,
505 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
507 void fastcall __lock_page(struct page *page)
509 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
511 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
512 TASK_UNINTERRUPTIBLE);
514 EXPORT_SYMBOL(__lock_page);
517 * a rather lightweight function, finding and getting a reference to a
518 * hashed page atomically.
520 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
524 read_lock_irq(&mapping->tree_lock);
525 page = radix_tree_lookup(&mapping->page_tree, offset);
527 page_cache_get(page);
528 read_unlock_irq(&mapping->tree_lock);
532 EXPORT_SYMBOL(find_get_page);
535 * Same as above, but trylock it instead of incrementing the count.
537 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
541 read_lock_irq(&mapping->tree_lock);
542 page = radix_tree_lookup(&mapping->page_tree, offset);
543 if (page && TestSetPageLocked(page))
545 read_unlock_irq(&mapping->tree_lock);
549 EXPORT_SYMBOL(find_trylock_page);
552 * find_lock_page - locate, pin and lock a pagecache page
554 * @mapping: the address_space to search
555 * @offset: the page index
557 * Locates the desired pagecache page, locks it, increments its reference
558 * count and returns its address.
560 * Returns zero if the page was not present. find_lock_page() may sleep.
562 struct page *find_lock_page(struct address_space *mapping,
563 unsigned long offset)
567 read_lock_irq(&mapping->tree_lock);
569 page = radix_tree_lookup(&mapping->page_tree, offset);
571 page_cache_get(page);
572 if (TestSetPageLocked(page)) {
573 read_unlock_irq(&mapping->tree_lock);
575 read_lock_irq(&mapping->tree_lock);
577 /* Has the page been truncated while we slept? */
578 if (unlikely(page->mapping != mapping ||
579 page->index != offset)) {
581 page_cache_release(page);
586 read_unlock_irq(&mapping->tree_lock);
590 EXPORT_SYMBOL(find_lock_page);
593 * find_or_create_page - locate or add a pagecache page
595 * @mapping: the page's address_space
596 * @index: the page's index into the mapping
597 * @gfp_mask: page allocation mode
599 * Locates a page in the pagecache. If the page is not present, a new page
600 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
601 * LRU list. The returned page is locked and has its reference count
604 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
607 * find_or_create_page() returns the desired page's address, or zero on
610 struct page *find_or_create_page(struct address_space *mapping,
611 unsigned long index, gfp_t gfp_mask)
613 struct page *page, *cached_page = NULL;
616 page = find_lock_page(mapping, index);
619 cached_page = alloc_page(gfp_mask);
623 err = add_to_page_cache_lru(cached_page, mapping,
628 } else if (err == -EEXIST)
632 page_cache_release(cached_page);
636 EXPORT_SYMBOL(find_or_create_page);
639 * find_get_pages - gang pagecache lookup
640 * @mapping: The address_space to search
641 * @start: The starting page index
642 * @nr_pages: The maximum number of pages
643 * @pages: Where the resulting pages are placed
645 * find_get_pages() will search for and return a group of up to
646 * @nr_pages pages in the mapping. The pages are placed at @pages.
647 * find_get_pages() takes a reference against the returned pages.
649 * The search returns a group of mapping-contiguous pages with ascending
650 * indexes. There may be holes in the indices due to not-present pages.
652 * find_get_pages() returns the number of pages which were found.
654 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
655 unsigned int nr_pages, struct page **pages)
660 read_lock_irq(&mapping->tree_lock);
661 ret = radix_tree_gang_lookup(&mapping->page_tree,
662 (void **)pages, start, nr_pages);
663 for (i = 0; i < ret; i++)
664 page_cache_get(pages[i]);
665 read_unlock_irq(&mapping->tree_lock);
670 * Like find_get_pages, except we only return pages which are tagged with
671 * `tag'. We update *index to index the next page for the traversal.
673 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
674 int tag, unsigned int nr_pages, struct page **pages)
679 read_lock_irq(&mapping->tree_lock);
680 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
681 (void **)pages, *index, nr_pages, tag);
682 for (i = 0; i < ret; i++)
683 page_cache_get(pages[i]);
685 *index = pages[ret - 1]->index + 1;
686 read_unlock_irq(&mapping->tree_lock);
691 * Same as grab_cache_page, but do not wait if the page is unavailable.
692 * This is intended for speculative data generators, where the data can
693 * be regenerated if the page couldn't be grabbed. This routine should
694 * be safe to call while holding the lock for another page.
696 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
697 * and deadlock against the caller's locked page.
700 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
702 struct page *page = find_get_page(mapping, index);
706 if (!TestSetPageLocked(page))
708 page_cache_release(page);
711 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
712 page = alloc_pages(gfp_mask, 0);
713 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
714 page_cache_release(page);
720 EXPORT_SYMBOL(grab_cache_page_nowait);
723 * This is a generic file read routine, and uses the
724 * mapping->a_ops->readpage() function for the actual low-level
727 * This is really ugly. But the goto's actually try to clarify some
728 * of the logic when it comes to error handling etc.
730 * Note the struct file* is only passed for the use of readpage. It may be
733 void do_generic_mapping_read(struct address_space *mapping,
734 struct file_ra_state *_ra,
737 read_descriptor_t *desc,
740 struct inode *inode = mapping->host;
742 unsigned long end_index;
743 unsigned long offset;
744 unsigned long last_index;
745 unsigned long next_index;
746 unsigned long prev_index;
748 struct page *cached_page;
750 struct file_ra_state ra = *_ra;
753 index = *ppos >> PAGE_CACHE_SHIFT;
755 prev_index = ra.prev_page;
756 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
757 offset = *ppos & ~PAGE_CACHE_MASK;
759 isize = i_size_read(inode);
763 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
766 unsigned long nr, ret;
768 /* nr is the maximum number of bytes to copy from this page */
769 nr = PAGE_CACHE_SIZE;
770 if (index >= end_index) {
771 if (index > end_index)
773 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
781 if (index == next_index)
782 next_index = page_cache_readahead(mapping, &ra, filp,
783 index, last_index - index);
786 page = find_get_page(mapping, index);
787 if (unlikely(page == NULL)) {
788 handle_ra_miss(mapping, &ra, index);
791 if (!PageUptodate(page))
792 goto page_not_up_to_date;
795 /* If users can be writing to this page using arbitrary
796 * virtual addresses, take care about potential aliasing
797 * before reading the page on the kernel side.
799 if (mapping_writably_mapped(mapping))
800 flush_dcache_page(page);
803 * When (part of) the same page is read multiple times
804 * in succession, only mark it as accessed the first time.
806 if (prev_index != index)
807 mark_page_accessed(page);
811 * Ok, we have the page, and it's up-to-date, so
812 * now we can copy it to user space...
814 * The actor routine returns how many bytes were actually used..
815 * NOTE! This may not be the same as how much of a user buffer
816 * we filled up (we may be padding etc), so we can only update
817 * "pos" here (the actor routine has to update the user buffer
818 * pointers and the remaining count).
820 ret = actor(desc, page, offset, nr);
822 index += offset >> PAGE_CACHE_SHIFT;
823 offset &= ~PAGE_CACHE_MASK;
825 page_cache_release(page);
826 if (ret == nr && desc->count)
831 /* Get exclusive access to the page ... */
834 /* Did it get unhashed before we got the lock? */
835 if (!page->mapping) {
837 page_cache_release(page);
841 /* Did somebody else fill it already? */
842 if (PageUptodate(page)) {
848 /* Start the actual read. The read will unlock the page. */
849 error = mapping->a_ops->readpage(filp, page);
851 if (unlikely(error)) {
852 if (error == AOP_TRUNCATED_PAGE) {
853 page_cache_release(page);
859 if (!PageUptodate(page)) {
861 if (!PageUptodate(page)) {
862 if (page->mapping == NULL) {
864 * invalidate_inode_pages got it
867 page_cache_release(page);
878 * i_size must be checked after we have done ->readpage.
880 * Checking i_size after the readpage allows us to calculate
881 * the correct value for "nr", which means the zero-filled
882 * part of the page is not copied back to userspace (unless
883 * another truncate extends the file - this is desired though).
885 isize = i_size_read(inode);
886 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
887 if (unlikely(!isize || index > end_index)) {
888 page_cache_release(page);
892 /* nr is the maximum number of bytes to copy from this page */
893 nr = PAGE_CACHE_SIZE;
894 if (index == end_index) {
895 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
897 page_cache_release(page);
905 /* UHHUH! A synchronous read error occurred. Report it */
907 page_cache_release(page);
912 * Ok, it wasn't cached, so we need to create a new
916 cached_page = page_cache_alloc_cold(mapping);
918 desc->error = -ENOMEM;
922 error = add_to_page_cache_lru(cached_page, mapping,
925 if (error == -EEXIST)
938 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
940 page_cache_release(cached_page);
945 EXPORT_SYMBOL(do_generic_mapping_read);
947 int file_read_actor(read_descriptor_t *desc, struct page *page,
948 unsigned long offset, unsigned long size)
951 unsigned long left, count = desc->count;
957 * Faults on the destination of a read are common, so do it before
960 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
961 kaddr = kmap_atomic(page, KM_USER0);
962 left = __copy_to_user_inatomic(desc->arg.buf,
963 kaddr + offset, size);
964 kunmap_atomic(kaddr, KM_USER0);
969 /* Do it the slow way */
971 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
976 desc->error = -EFAULT;
979 desc->count = count - size;
980 desc->written += size;
981 desc->arg.buf += size;
984 EXPORT_SYMBOL(file_read_actor);
987 * This is the "read()" routine for all filesystems
988 * that can use the page cache directly.
991 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
992 unsigned long nr_segs, loff_t *ppos)
994 struct file *filp = iocb->ki_filp;
1000 for (seg = 0; seg < nr_segs; seg++) {
1001 const struct iovec *iv = &iov[seg];
1004 * If any segment has a negative length, or the cumulative
1005 * length ever wraps negative then return -EINVAL.
1007 count += iv->iov_len;
1008 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1010 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1015 count -= iv->iov_len; /* This segment is no good */
1019 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1020 if (filp->f_flags & O_DIRECT) {
1021 loff_t pos = *ppos, size;
1022 struct address_space *mapping;
1023 struct inode *inode;
1025 mapping = filp->f_mapping;
1026 inode = mapping->host;
1029 goto out; /* skip atime */
1030 size = i_size_read(inode);
1032 retval = generic_file_direct_IO(READ, iocb,
1034 if (retval > 0 && !is_sync_kiocb(iocb))
1035 retval = -EIOCBQUEUED;
1037 *ppos = pos + retval;
1039 file_accessed(filp);
1045 for (seg = 0; seg < nr_segs; seg++) {
1046 read_descriptor_t desc;
1049 desc.arg.buf = iov[seg].iov_base;
1050 desc.count = iov[seg].iov_len;
1051 if (desc.count == 0)
1054 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1055 retval += desc.written;
1057 retval = retval ?: desc.error;
1066 EXPORT_SYMBOL(__generic_file_aio_read);
1069 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1071 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1073 BUG_ON(iocb->ki_pos != pos);
1074 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1077 EXPORT_SYMBOL(generic_file_aio_read);
1080 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1082 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1086 init_sync_kiocb(&kiocb, filp);
1087 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1088 if (-EIOCBQUEUED == ret)
1089 ret = wait_on_sync_kiocb(&kiocb);
1093 EXPORT_SYMBOL(generic_file_read);
1095 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1098 unsigned long count = desc->count;
1099 struct file *file = desc->arg.data;
1104 written = file->f_op->sendpage(file, page, offset,
1105 size, &file->f_pos, size<count);
1107 desc->error = written;
1110 desc->count = count - written;
1111 desc->written += written;
1115 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1116 size_t count, read_actor_t actor, void *target)
1118 read_descriptor_t desc;
1125 desc.arg.data = target;
1128 do_generic_file_read(in_file, ppos, &desc, actor);
1130 return desc.written;
1134 EXPORT_SYMBOL(generic_file_sendfile);
1137 do_readahead(struct address_space *mapping, struct file *filp,
1138 unsigned long index, unsigned long nr)
1140 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1143 force_page_cache_readahead(mapping, filp, index,
1144 max_sane_readahead(nr));
1148 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1156 if (file->f_mode & FMODE_READ) {
1157 struct address_space *mapping = file->f_mapping;
1158 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1159 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1160 unsigned long len = end - start + 1;
1161 ret = do_readahead(mapping, file, start, len);
1170 * This adds the requested page to the page cache if it isn't already there,
1171 * and schedules an I/O to read in its contents from disk.
1173 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1174 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1176 struct address_space *mapping = file->f_mapping;
1181 page = page_cache_alloc_cold(mapping);
1185 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1187 ret = mapping->a_ops->readpage(file, page);
1188 else if (ret == -EEXIST)
1189 ret = 0; /* losing race to add is OK */
1191 page_cache_release(page);
1193 } while (ret == AOP_TRUNCATED_PAGE);
1198 #define MMAP_LOTSAMISS (100)
1201 * filemap_nopage() is invoked via the vma operations vector for a
1202 * mapped memory region to read in file data during a page fault.
1204 * The goto's are kind of ugly, but this streamlines the normal case of having
1205 * it in the page cache, and handles the special cases reasonably without
1206 * having a lot of duplicated code.
1208 struct page *filemap_nopage(struct vm_area_struct *area,
1209 unsigned long address, int *type)
1212 struct file *file = area->vm_file;
1213 struct address_space *mapping = file->f_mapping;
1214 struct file_ra_state *ra = &file->f_ra;
1215 struct inode *inode = mapping->host;
1217 unsigned long size, pgoff;
1218 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1220 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1223 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1225 goto outside_data_content;
1227 /* If we don't want any read-ahead, don't bother */
1228 if (VM_RandomReadHint(area))
1229 goto no_cached_page;
1232 * The readahead code wants to be told about each and every page
1233 * so it can build and shrink its windows appropriately
1235 * For sequential accesses, we use the generic readahead logic.
1237 if (VM_SequentialReadHint(area))
1238 page_cache_readahead(mapping, ra, file, pgoff, 1);
1241 * Do we have something in the page cache already?
1244 page = find_get_page(mapping, pgoff);
1246 unsigned long ra_pages;
1248 if (VM_SequentialReadHint(area)) {
1249 handle_ra_miss(mapping, ra, pgoff);
1250 goto no_cached_page;
1255 * Do we miss much more than hit in this file? If so,
1256 * stop bothering with read-ahead. It will only hurt.
1258 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1259 goto no_cached_page;
1262 * To keep the pgmajfault counter straight, we need to
1263 * check did_readaround, as this is an inner loop.
1265 if (!did_readaround) {
1266 majmin = VM_FAULT_MAJOR;
1267 inc_page_state(pgmajfault);
1270 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1274 if (pgoff > ra_pages / 2)
1275 start = pgoff - ra_pages / 2;
1276 do_page_cache_readahead(mapping, file, start, ra_pages);
1278 page = find_get_page(mapping, pgoff);
1280 goto no_cached_page;
1283 if (!did_readaround)
1287 * Ok, found a page in the page cache, now we need to check
1288 * that it's up-to-date.
1290 if (!PageUptodate(page))
1291 goto page_not_uptodate;
1295 * Found the page and have a reference on it.
1297 mark_page_accessed(page);
1302 outside_data_content:
1304 * An external ptracer can access pages that normally aren't
1307 if (area->vm_mm == current->mm)
1309 /* Fall through to the non-read-ahead case */
1312 * We're only likely to ever get here if MADV_RANDOM is in
1315 error = page_cache_read(file, pgoff);
1319 * The page we want has now been added to the page cache.
1320 * In the unlikely event that someone removed it in the
1321 * meantime, we'll just come back here and read it again.
1327 * An error return from page_cache_read can result if the
1328 * system is low on memory, or a problem occurs while trying
1331 if (error == -ENOMEM)
1336 if (!did_readaround) {
1337 majmin = VM_FAULT_MAJOR;
1338 inc_page_state(pgmajfault);
1342 /* Did it get unhashed while we waited for it? */
1343 if (!page->mapping) {
1345 page_cache_release(page);
1349 /* Did somebody else get it up-to-date? */
1350 if (PageUptodate(page)) {
1355 error = mapping->a_ops->readpage(file, page);
1357 wait_on_page_locked(page);
1358 if (PageUptodate(page))
1360 } else if (error == AOP_TRUNCATED_PAGE) {
1361 page_cache_release(page);
1366 * Umm, take care of errors if the page isn't up-to-date.
1367 * Try to re-read it _once_. We do this synchronously,
1368 * because there really aren't any performance issues here
1369 * and we need to check for errors.
1373 /* Somebody truncated the page on us? */
1374 if (!page->mapping) {
1376 page_cache_release(page);
1380 /* Somebody else successfully read it in? */
1381 if (PageUptodate(page)) {
1385 ClearPageError(page);
1386 error = mapping->a_ops->readpage(file, page);
1388 wait_on_page_locked(page);
1389 if (PageUptodate(page))
1391 } else if (error == AOP_TRUNCATED_PAGE) {
1392 page_cache_release(page);
1397 * Things didn't work out. Return zero to tell the
1398 * mm layer so, possibly freeing the page cache page first.
1400 page_cache_release(page);
1404 EXPORT_SYMBOL(filemap_nopage);
1406 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1409 struct address_space *mapping = file->f_mapping;
1414 * Do we have something in the page cache already?
1417 page = find_get_page(mapping, pgoff);
1421 goto no_cached_page;
1425 * Ok, found a page in the page cache, now we need to check
1426 * that it's up-to-date.
1428 if (!PageUptodate(page)) {
1430 page_cache_release(page);
1433 goto page_not_uptodate;
1438 * Found the page and have a reference on it.
1440 mark_page_accessed(page);
1444 error = page_cache_read(file, pgoff);
1447 * The page we want has now been added to the page cache.
1448 * In the unlikely event that someone removed it in the
1449 * meantime, we'll just come back here and read it again.
1455 * An error return from page_cache_read can result if the
1456 * system is low on memory, or a problem occurs while trying
1464 /* Did it get unhashed while we waited for it? */
1465 if (!page->mapping) {
1470 /* Did somebody else get it up-to-date? */
1471 if (PageUptodate(page)) {
1476 error = mapping->a_ops->readpage(file, page);
1478 wait_on_page_locked(page);
1479 if (PageUptodate(page))
1481 } else if (error == AOP_TRUNCATED_PAGE) {
1482 page_cache_release(page);
1487 * Umm, take care of errors if the page isn't up-to-date.
1488 * Try to re-read it _once_. We do this synchronously,
1489 * because there really aren't any performance issues here
1490 * and we need to check for errors.
1494 /* Somebody truncated the page on us? */
1495 if (!page->mapping) {
1499 /* Somebody else successfully read it in? */
1500 if (PageUptodate(page)) {
1505 ClearPageError(page);
1506 error = mapping->a_ops->readpage(file, page);
1508 wait_on_page_locked(page);
1509 if (PageUptodate(page))
1511 } else if (error == AOP_TRUNCATED_PAGE) {
1512 page_cache_release(page);
1517 * Things didn't work out. Return zero to tell the
1518 * mm layer so, possibly freeing the page cache page first.
1521 page_cache_release(page);
1526 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1527 unsigned long len, pgprot_t prot, unsigned long pgoff,
1530 struct file *file = vma->vm_file;
1531 struct address_space *mapping = file->f_mapping;
1532 struct inode *inode = mapping->host;
1534 struct mm_struct *mm = vma->vm_mm;
1539 force_page_cache_readahead(mapping, vma->vm_file,
1540 pgoff, len >> PAGE_CACHE_SHIFT);
1543 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1544 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1547 page = filemap_getpage(file, pgoff, nonblock);
1549 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1550 * done in shmem_populate calling shmem_getpage */
1551 if (!page && !nonblock)
1555 err = install_page(mm, vma, addr, page, prot);
1557 page_cache_release(page);
1560 } else if (vma->vm_flags & VM_NONLINEAR) {
1561 /* No page was found just because we can't read it in now (being
1562 * here implies nonblock != 0), but the page may exist, so set
1563 * the PTE to fault it in later. */
1564 err = install_file_pte(mm, vma, addr, pgoff, prot);
1577 EXPORT_SYMBOL(filemap_populate);
1579 struct vm_operations_struct generic_file_vm_ops = {
1580 .nopage = filemap_nopage,
1581 .populate = filemap_populate,
1584 /* This is used for a general mmap of a disk file */
1586 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1588 struct address_space *mapping = file->f_mapping;
1590 if (!mapping->a_ops->readpage)
1592 file_accessed(file);
1593 vma->vm_ops = &generic_file_vm_ops;
1598 * This is for filesystems which do not implement ->writepage.
1600 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1602 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1604 return generic_file_mmap(file, vma);
1607 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1611 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1615 #endif /* CONFIG_MMU */
1617 EXPORT_SYMBOL(generic_file_mmap);
1618 EXPORT_SYMBOL(generic_file_readonly_mmap);
1620 static inline struct page *__read_cache_page(struct address_space *mapping,
1621 unsigned long index,
1622 int (*filler)(void *,struct page*),
1625 struct page *page, *cached_page = NULL;
1628 page = find_get_page(mapping, index);
1631 cached_page = page_cache_alloc_cold(mapping);
1633 return ERR_PTR(-ENOMEM);
1635 err = add_to_page_cache_lru(cached_page, mapping,
1640 /* Presumably ENOMEM for radix tree node */
1641 page_cache_release(cached_page);
1642 return ERR_PTR(err);
1646 err = filler(data, page);
1648 page_cache_release(page);
1649 page = ERR_PTR(err);
1653 page_cache_release(cached_page);
1658 * Read into the page cache. If a page already exists,
1659 * and PageUptodate() is not set, try to fill the page.
1661 struct page *read_cache_page(struct address_space *mapping,
1662 unsigned long index,
1663 int (*filler)(void *,struct page*),
1670 page = __read_cache_page(mapping, index, filler, data);
1673 mark_page_accessed(page);
1674 if (PageUptodate(page))
1678 if (!page->mapping) {
1680 page_cache_release(page);
1683 if (PageUptodate(page)) {
1687 err = filler(data, page);
1689 page_cache_release(page);
1690 page = ERR_PTR(err);
1696 EXPORT_SYMBOL(read_cache_page);
1699 * If the page was newly created, increment its refcount and add it to the
1700 * caller's lru-buffering pagevec. This function is specifically for
1701 * generic_file_write().
1703 static inline struct page *
1704 __grab_cache_page(struct address_space *mapping, unsigned long index,
1705 struct page **cached_page, struct pagevec *lru_pvec)
1710 page = find_lock_page(mapping, index);
1712 if (!*cached_page) {
1713 *cached_page = page_cache_alloc(mapping);
1717 err = add_to_page_cache(*cached_page, mapping,
1722 page = *cached_page;
1723 page_cache_get(page);
1724 if (!pagevec_add(lru_pvec, page))
1725 __pagevec_lru_add(lru_pvec);
1726 *cached_page = NULL;
1733 * The logic we want is
1735 * if suid or (sgid and xgrp)
1738 int remove_suid(struct dentry *dentry)
1740 mode_t mode = dentry->d_inode->i_mode;
1744 /* suid always must be killed */
1745 if (unlikely(mode & S_ISUID))
1746 kill = ATTR_KILL_SUID;
1749 * sgid without any exec bits is just a mandatory locking mark; leave
1750 * it alone. If some exec bits are set, it's a real sgid; kill it.
1752 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1753 kill |= ATTR_KILL_SGID;
1755 if (unlikely(kill && !capable(CAP_FSETID))) {
1756 struct iattr newattrs;
1758 newattrs.ia_valid = ATTR_FORCE | kill;
1759 result = notify_change(dentry, &newattrs);
1763 EXPORT_SYMBOL(remove_suid);
1766 __filemap_copy_from_user_iovec(char *vaddr,
1767 const struct iovec *iov, size_t base, size_t bytes)
1769 size_t copied = 0, left = 0;
1772 char __user *buf = iov->iov_base + base;
1773 int copy = min(bytes, iov->iov_len - base);
1776 left = __copy_from_user_inatomic(vaddr, buf, copy);
1782 if (unlikely(left)) {
1783 /* zero the rest of the target like __copy_from_user */
1785 memset(vaddr, 0, bytes);
1789 return copied - left;
1793 * Performs necessary checks before doing a write
1795 * Can adjust writing position aor amount of bytes to write.
1796 * Returns appropriate error code that caller should return or
1797 * zero in case that write should be allowed.
1799 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1801 struct inode *inode = file->f_mapping->host;
1802 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1804 if (unlikely(*pos < 0))
1808 /* FIXME: this is for backwards compatibility with 2.4 */
1809 if (file->f_flags & O_APPEND)
1810 *pos = i_size_read(inode);
1812 if (limit != RLIM_INFINITY) {
1813 if (*pos >= limit) {
1814 send_sig(SIGXFSZ, current, 0);
1817 if (*count > limit - (typeof(limit))*pos) {
1818 *count = limit - (typeof(limit))*pos;
1826 if (unlikely(*pos + *count > MAX_NON_LFS &&
1827 !(file->f_flags & O_LARGEFILE))) {
1828 if (*pos >= MAX_NON_LFS) {
1829 send_sig(SIGXFSZ, current, 0);
1832 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1833 *count = MAX_NON_LFS - (unsigned long)*pos;
1838 * Are we about to exceed the fs block limit ?
1840 * If we have written data it becomes a short write. If we have
1841 * exceeded without writing data we send a signal and return EFBIG.
1842 * Linus frestrict idea will clean these up nicely..
1844 if (likely(!isblk)) {
1845 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1846 if (*count || *pos > inode->i_sb->s_maxbytes) {
1847 send_sig(SIGXFSZ, current, 0);
1850 /* zero-length writes at ->s_maxbytes are OK */
1853 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1854 *count = inode->i_sb->s_maxbytes - *pos;
1857 if (bdev_read_only(I_BDEV(inode)))
1859 isize = i_size_read(inode);
1860 if (*pos >= isize) {
1861 if (*count || *pos > isize)
1865 if (*pos + *count > isize)
1866 *count = isize - *pos;
1870 EXPORT_SYMBOL(generic_write_checks);
1873 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1874 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1875 size_t count, size_t ocount)
1877 struct file *file = iocb->ki_filp;
1878 struct address_space *mapping = file->f_mapping;
1879 struct inode *inode = mapping->host;
1882 if (count != ocount)
1883 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1885 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1887 loff_t end = pos + written;
1888 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1889 i_size_write(inode, end);
1890 mark_inode_dirty(inode);
1896 * Sync the fs metadata but not the minor inode changes and
1897 * of course not the data as we did direct DMA for the IO.
1898 * i_mutex is held, which protects generic_osync_inode() from
1901 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
1902 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1906 if (written == count && !is_sync_kiocb(iocb))
1907 written = -EIOCBQUEUED;
1910 EXPORT_SYMBOL(generic_file_direct_write);
1913 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
1914 unsigned long nr_segs, loff_t pos, loff_t *ppos,
1915 size_t count, ssize_t written)
1917 struct file *file = iocb->ki_filp;
1918 struct address_space * mapping = file->f_mapping;
1919 struct address_space_operations *a_ops = mapping->a_ops;
1920 struct inode *inode = mapping->host;
1923 struct page *cached_page = NULL;
1925 struct pagevec lru_pvec;
1926 const struct iovec *cur_iov = iov; /* current iovec */
1927 size_t iov_base = 0; /* offset in the current iovec */
1930 pagevec_init(&lru_pvec, 0);
1933 * handle partial DIO write. Adjust cur_iov if needed.
1935 if (likely(nr_segs == 1))
1936 buf = iov->iov_base + written;
1938 filemap_set_next_iovec(&cur_iov, &iov_base, written);
1939 buf = cur_iov->iov_base + iov_base;
1943 unsigned long index;
1944 unsigned long offset;
1945 unsigned long maxlen;
1948 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1949 index = pos >> PAGE_CACHE_SHIFT;
1950 bytes = PAGE_CACHE_SIZE - offset;
1955 * Bring in the user page that we will copy from _first_.
1956 * Otherwise there's a nasty deadlock on copying from the
1957 * same page as we're writing to, without it being marked
1960 maxlen = cur_iov->iov_len - iov_base;
1963 fault_in_pages_readable(buf, maxlen);
1965 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
1971 status = a_ops->prepare_write(file, page, offset, offset+bytes);
1972 if (unlikely(status)) {
1973 loff_t isize = i_size_read(inode);
1975 if (status != AOP_TRUNCATED_PAGE)
1977 page_cache_release(page);
1978 if (status == AOP_TRUNCATED_PAGE)
1981 * prepare_write() may have instantiated a few blocks
1982 * outside i_size. Trim these off again.
1984 if (pos + bytes > isize)
1985 vmtruncate(inode, isize);
1988 if (likely(nr_segs == 1))
1989 copied = filemap_copy_from_user(page, offset,
1992 copied = filemap_copy_from_user_iovec(page, offset,
1993 cur_iov, iov_base, bytes);
1994 flush_dcache_page(page);
1995 status = a_ops->commit_write(file, page, offset, offset+bytes);
1996 if (status == AOP_TRUNCATED_PAGE) {
1997 page_cache_release(page);
2000 if (likely(copied > 0)) {
2009 if (unlikely(nr_segs > 1)) {
2010 filemap_set_next_iovec(&cur_iov,
2013 buf = cur_iov->iov_base +
2020 if (unlikely(copied != bytes))
2024 mark_page_accessed(page);
2025 page_cache_release(page);
2028 balance_dirty_pages_ratelimited(mapping);
2034 page_cache_release(cached_page);
2037 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2039 if (likely(status >= 0)) {
2040 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2041 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2042 status = generic_osync_inode(inode, mapping,
2043 OSYNC_METADATA|OSYNC_DATA);
2048 * If we get here for O_DIRECT writes then we must have fallen through
2049 * to buffered writes (block instantiation inside i_size). So we sync
2050 * the file data here, to try to honour O_DIRECT expectations.
2052 if (unlikely(file->f_flags & O_DIRECT) && written)
2053 status = filemap_write_and_wait(mapping);
2055 pagevec_lru_add(&lru_pvec);
2056 return written ? written : status;
2058 EXPORT_SYMBOL(generic_file_buffered_write);
2061 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2062 unsigned long nr_segs, loff_t *ppos)
2064 struct file *file = iocb->ki_filp;
2065 struct address_space * mapping = file->f_mapping;
2066 size_t ocount; /* original count */
2067 size_t count; /* after file limit checks */
2068 struct inode *inode = mapping->host;
2075 for (seg = 0; seg < nr_segs; seg++) {
2076 const struct iovec *iv = &iov[seg];
2079 * If any segment has a negative length, or the cumulative
2080 * length ever wraps negative then return -EINVAL.
2082 ocount += iv->iov_len;
2083 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2085 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2090 ocount -= iv->iov_len; /* This segment is no good */
2097 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2099 /* We can write back this queue in page reclaim */
2100 current->backing_dev_info = mapping->backing_dev_info;
2103 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2110 err = remove_suid(file->f_dentry);
2114 file_update_time(file);
2116 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2117 if (unlikely(file->f_flags & O_DIRECT)) {
2118 written = generic_file_direct_write(iocb, iov,
2119 &nr_segs, pos, ppos, count, ocount);
2120 if (written < 0 || written == count)
2123 * direct-io write to a hole: fall through to buffered I/O
2124 * for completing the rest of the request.
2130 written = generic_file_buffered_write(iocb, iov, nr_segs,
2131 pos, ppos, count, written);
2133 current->backing_dev_info = NULL;
2134 return written ? written : err;
2136 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2139 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2140 unsigned long nr_segs, loff_t *ppos)
2142 struct file *file = iocb->ki_filp;
2143 struct address_space *mapping = file->f_mapping;
2144 struct inode *inode = mapping->host;
2148 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2150 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2153 err = sync_page_range_nolock(inode, mapping, pos, ret);
2161 __generic_file_write_nolock(struct file *file, const struct iovec *iov,
2162 unsigned long nr_segs, loff_t *ppos)
2167 init_sync_kiocb(&kiocb, file);
2168 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2169 if (ret == -EIOCBQUEUED)
2170 ret = wait_on_sync_kiocb(&kiocb);
2175 generic_file_write_nolock(struct file *file, const struct iovec *iov,
2176 unsigned long nr_segs, loff_t *ppos)
2181 init_sync_kiocb(&kiocb, file);
2182 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2183 if (-EIOCBQUEUED == ret)
2184 ret = wait_on_sync_kiocb(&kiocb);
2187 EXPORT_SYMBOL(generic_file_write_nolock);
2189 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2190 size_t count, loff_t pos)
2192 struct file *file = iocb->ki_filp;
2193 struct address_space *mapping = file->f_mapping;
2194 struct inode *inode = mapping->host;
2196 struct iovec local_iov = { .iov_base = (void __user *)buf,
2199 BUG_ON(iocb->ki_pos != pos);
2201 mutex_lock(&inode->i_mutex);
2202 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2204 mutex_unlock(&inode->i_mutex);
2206 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2209 err = sync_page_range(inode, mapping, pos, ret);
2215 EXPORT_SYMBOL(generic_file_aio_write);
2217 ssize_t generic_file_write(struct file *file, const char __user *buf,
2218 size_t count, loff_t *ppos)
2220 struct address_space *mapping = file->f_mapping;
2221 struct inode *inode = mapping->host;
2223 struct iovec local_iov = { .iov_base = (void __user *)buf,
2226 mutex_lock(&inode->i_mutex);
2227 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
2228 mutex_unlock(&inode->i_mutex);
2230 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2233 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2239 EXPORT_SYMBOL(generic_file_write);
2241 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2242 unsigned long nr_segs, loff_t *ppos)
2247 init_sync_kiocb(&kiocb, filp);
2248 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2249 if (-EIOCBQUEUED == ret)
2250 ret = wait_on_sync_kiocb(&kiocb);
2253 EXPORT_SYMBOL(generic_file_readv);
2255 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2256 unsigned long nr_segs, loff_t *ppos)
2258 struct address_space *mapping = file->f_mapping;
2259 struct inode *inode = mapping->host;
2262 mutex_lock(&inode->i_mutex);
2263 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
2264 mutex_unlock(&inode->i_mutex);
2266 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2269 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2275 EXPORT_SYMBOL(generic_file_writev);
2278 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2279 * went wrong during pagecache shootdown.
2282 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2283 loff_t offset, unsigned long nr_segs)
2285 struct file *file = iocb->ki_filp;
2286 struct address_space *mapping = file->f_mapping;
2288 size_t write_len = 0;
2291 * If it's a write, unmap all mmappings of the file up-front. This
2292 * will cause any pte dirty bits to be propagated into the pageframes
2293 * for the subsequent filemap_write_and_wait().
2296 write_len = iov_length(iov, nr_segs);
2297 if (mapping_mapped(mapping))
2298 unmap_mapping_range(mapping, offset, write_len, 0);
2301 retval = filemap_write_and_wait(mapping);
2303 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2305 if (rw == WRITE && mapping->nrpages) {
2306 pgoff_t end = (offset + write_len - 1)
2307 >> PAGE_CACHE_SHIFT;
2308 int err = invalidate_inode_pages2_range(mapping,
2309 offset >> PAGE_CACHE_SHIFT, end);