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/kernel_stat.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
32 * FIXME: remove all knowledge of the buffer layer from the core VM
34 #include <linux/buffer_head.h> /* for generic_osync_inode */
36 #include <asm/uaccess.h>
40 * Shared mappings implemented 30.11.1994. It's not fully working yet,
43 * Shared mappings now work. 15.8.1995 Bruno.
45 * finished 'unifying' the page and buffer cache and SMP-threaded the
46 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
48 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
54 * ->i_mmap_lock (vmtruncate)
55 * ->private_lock (__free_pte->__set_page_dirty_buffers)
57 * ->swap_device_lock (exclusive_swap_page, others)
58 * ->mapping->tree_lock
61 * ->i_mmap_lock (truncate->unmap_mapping_range)
65 * ->page_table_lock (various places, mainly in mmap.c)
66 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
69 * ->lock_page (access_process_vm)
75 * ->i_alloc_sem (various)
78 * ->sb_lock (fs/fs-writeback.c)
79 * ->mapping->tree_lock (__sync_single_inode)
82 * ->anon_vma.lock (vma_adjust)
85 * ->page_table_lock (anon_vma_prepare and various)
88 * ->swap_device_lock (try_to_unmap_one)
89 * ->private_lock (try_to_unmap_one)
90 * ->tree_lock (try_to_unmap_one)
91 * ->zone.lru_lock (follow_page->mark_page_accessed)
92 * ->private_lock (page_remove_rmap->set_page_dirty)
93 * ->tree_lock (page_remove_rmap->set_page_dirty)
94 * ->inode_lock (page_remove_rmap->set_page_dirty)
95 * ->inode_lock (zap_pte_range->set_page_dirty)
96 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
99 * ->dcache_lock (proc_pid_lookup)
103 * Remove a page from the page cache and free it. Caller has to make
104 * sure the page is locked and that nobody else uses it - or that usage
105 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
107 void __remove_from_page_cache(struct page *page)
109 struct address_space *mapping = page->mapping;
111 radix_tree_delete(&mapping->page_tree, page->index);
112 page->mapping = NULL;
117 void remove_from_page_cache(struct page *page)
119 struct address_space *mapping = page->mapping;
121 BUG_ON(!PageLocked(page));
123 write_lock_irq(&mapping->tree_lock);
124 __remove_from_page_cache(page);
125 write_unlock_irq(&mapping->tree_lock);
128 static int sync_page(void *word)
130 struct address_space *mapping;
133 page = container_of((page_flags_t *)word, struct page, flags);
136 * page_mapping() is being called without PG_locked held.
137 * Some knowledge of the state and use of the page is used to
138 * reduce the requirements down to a memory barrier.
139 * The danger here is of a stale page_mapping() return value
140 * indicating a struct address_space different from the one it's
141 * associated with when it is associated with one.
142 * After smp_mb(), it's either the correct page_mapping() for
143 * the page, or an old page_mapping() and the page's own
144 * page_mapping() has gone NULL.
145 * The ->sync_page() address_space operation must tolerate
146 * page_mapping() going NULL. By an amazing coincidence,
147 * this comes about because none of the users of the page
148 * in the ->sync_page() methods make essential use of the
149 * page_mapping(), merely passing the page down to the backing
150 * device's unplug functions when it's non-NULL, which in turn
151 * ignore it for all cases but swap, where only page->private is
152 * of interest. When page_mapping() does go NULL, the entire
153 * call stack gracefully ignores the page and returns.
157 mapping = page_mapping(page);
158 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
159 mapping->a_ops->sync_page(page);
165 * filemap_fdatawrite_range - start writeback against all of a mapping's
166 * dirty pages that lie within the byte offsets <start, end>
167 * @mapping: address space structure to write
168 * @start: offset in bytes where the range starts
169 * @end: offset in bytes where the range ends
170 * @sync_mode: enable synchronous operation
172 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
173 * opposed to a regular memory * cleansing writeback. The difference between
174 * these two operations is that if a dirty page/buffer is encountered, it must
175 * be waited upon, and not just skipped over.
177 static int __filemap_fdatawrite_range(struct address_space *mapping,
178 loff_t start, loff_t end, int sync_mode)
181 struct writeback_control wbc = {
182 .sync_mode = sync_mode,
183 .nr_to_write = mapping->nrpages * 2,
188 if (!mapping_cap_writeback_dirty(mapping))
191 ret = do_writepages(mapping, &wbc);
195 static inline int __filemap_fdatawrite(struct address_space *mapping,
198 return __filemap_fdatawrite_range(mapping, 0, 0, sync_mode);
201 int filemap_fdatawrite(struct address_space *mapping)
203 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
205 EXPORT_SYMBOL(filemap_fdatawrite);
207 static int filemap_fdatawrite_range(struct address_space *mapping,
208 loff_t start, loff_t end)
210 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
214 * This is a mostly non-blocking flush. Not suitable for data-integrity
215 * purposes - I/O may not be started against all dirty pages.
217 int filemap_flush(struct address_space *mapping)
219 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
221 EXPORT_SYMBOL(filemap_flush);
224 * Wait for writeback to complete against pages indexed by start->end
227 static int wait_on_page_writeback_range(struct address_space *mapping,
228 pgoff_t start, pgoff_t end)
238 pagevec_init(&pvec, 0);
240 while ((index <= end) &&
241 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
242 PAGECACHE_TAG_WRITEBACK,
243 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
246 for (i = 0; i < nr_pages; i++) {
247 struct page *page = pvec.pages[i];
249 /* until radix tree lookup accepts end_index */
250 if (page->index > end)
253 wait_on_page_writeback(page);
257 pagevec_release(&pvec);
261 /* Check for outstanding write errors */
262 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
264 if (test_and_clear_bit(AS_EIO, &mapping->flags))
271 * Write and wait upon all the pages in the passed range. This is a "data
272 * integrity" operation. It waits upon in-flight writeout before starting and
273 * waiting upon new writeout. If there was an IO error, return it.
275 * We need to re-take i_sem during the generic_osync_inode list walk because
276 * it is otherwise livelockable.
278 int sync_page_range(struct inode *inode, struct address_space *mapping,
279 loff_t pos, size_t count)
281 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
282 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
285 if (!mapping_cap_writeback_dirty(mapping) || !count)
287 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
290 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
294 ret = wait_on_page_writeback_range(mapping, start, end);
297 EXPORT_SYMBOL(sync_page_range);
300 * Note: Holding i_sem across sync_page_range_nolock is not a good idea
301 * as it forces O_SYNC writers to different parts of the same file
302 * to be serialised right until io completion.
304 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
305 loff_t pos, size_t count)
307 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
308 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
311 if (!mapping_cap_writeback_dirty(mapping) || !count)
313 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
315 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
317 ret = wait_on_page_writeback_range(mapping, start, end);
320 EXPORT_SYMBOL(sync_page_range_nolock);
323 * filemap_fdatawait - walk the list of under-writeback pages of the given
324 * address space and wait for all of them.
326 * @mapping: address space structure to wait for
328 int filemap_fdatawait(struct address_space *mapping)
330 loff_t i_size = i_size_read(mapping->host);
335 return wait_on_page_writeback_range(mapping, 0,
336 (i_size - 1) >> PAGE_CACHE_SHIFT);
338 EXPORT_SYMBOL(filemap_fdatawait);
340 int filemap_write_and_wait(struct address_space *mapping)
344 if (mapping->nrpages) {
345 retval = filemap_fdatawrite(mapping);
347 retval = filemap_fdatawait(mapping);
352 int filemap_write_and_wait_range(struct address_space *mapping,
353 loff_t lstart, loff_t lend)
357 if (mapping->nrpages) {
358 retval = __filemap_fdatawrite_range(mapping, lstart, lend,
361 retval = wait_on_page_writeback_range(mapping,
362 lstart >> PAGE_CACHE_SHIFT,
363 lend >> PAGE_CACHE_SHIFT);
369 * This function is used to add newly allocated pagecache pages:
370 * the page is new, so we can just run SetPageLocked() against it.
371 * The other page state flags were set by rmqueue().
373 * This function does not add the page to the LRU. The caller must do that.
375 int add_to_page_cache(struct page *page, struct address_space *mapping,
376 pgoff_t offset, int gfp_mask)
378 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
381 write_lock_irq(&mapping->tree_lock);
382 error = radix_tree_insert(&mapping->page_tree, offset, page);
384 page_cache_get(page);
386 page->mapping = mapping;
387 page->index = offset;
391 write_unlock_irq(&mapping->tree_lock);
392 radix_tree_preload_end();
397 EXPORT_SYMBOL(add_to_page_cache);
399 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
400 pgoff_t offset, int gfp_mask)
402 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
409 * In order to wait for pages to become available there must be
410 * waitqueues associated with pages. By using a hash table of
411 * waitqueues where the bucket discipline is to maintain all
412 * waiters on the same queue and wake all when any of the pages
413 * become available, and for the woken contexts to check to be
414 * sure the appropriate page became available, this saves space
415 * at a cost of "thundering herd" phenomena during rare hash
418 static wait_queue_head_t *page_waitqueue(struct page *page)
420 const struct zone *zone = page_zone(page);
422 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
425 static inline void wake_up_page(struct page *page, int bit)
427 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
430 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
432 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
434 if (test_bit(bit_nr, &page->flags))
435 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
436 TASK_UNINTERRUPTIBLE);
438 EXPORT_SYMBOL(wait_on_page_bit);
441 * unlock_page() - unlock a locked page
445 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
446 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
447 * mechananism between PageLocked pages and PageWriteback pages is shared.
448 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
450 * The first mb is necessary to safely close the critical section opened by the
451 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
452 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
453 * parallel wait_on_page_locked()).
455 void fastcall unlock_page(struct page *page)
457 smp_mb__before_clear_bit();
458 if (!TestClearPageLocked(page))
460 smp_mb__after_clear_bit();
461 wake_up_page(page, PG_locked);
463 EXPORT_SYMBOL(unlock_page);
466 * End writeback against a page.
468 void end_page_writeback(struct page *page)
470 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
471 if (!test_clear_page_writeback(page))
474 smp_mb__after_clear_bit();
475 wake_up_page(page, PG_writeback);
477 EXPORT_SYMBOL(end_page_writeback);
480 * Get a lock on the page, assuming we need to sleep to get it.
482 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
483 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
484 * chances are that on the second loop, the block layer's plug list is empty,
485 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
487 void fastcall __lock_page(struct page *page)
489 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
491 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
492 TASK_UNINTERRUPTIBLE);
494 EXPORT_SYMBOL(__lock_page);
497 * a rather lightweight function, finding and getting a reference to a
498 * hashed page atomically.
500 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
504 read_lock_irq(&mapping->tree_lock);
505 page = radix_tree_lookup(&mapping->page_tree, offset);
507 page_cache_get(page);
508 read_unlock_irq(&mapping->tree_lock);
512 EXPORT_SYMBOL(find_get_page);
515 * Same as above, but trylock it instead of incrementing the count.
517 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
521 read_lock_irq(&mapping->tree_lock);
522 page = radix_tree_lookup(&mapping->page_tree, offset);
523 if (page && TestSetPageLocked(page))
525 read_unlock_irq(&mapping->tree_lock);
529 EXPORT_SYMBOL(find_trylock_page);
532 * find_lock_page - locate, pin and lock a pagecache page
534 * @mapping: the address_space to search
535 * @offset: the page index
537 * Locates the desired pagecache page, locks it, increments its reference
538 * count and returns its address.
540 * Returns zero if the page was not present. find_lock_page() may sleep.
542 struct page *find_lock_page(struct address_space *mapping,
543 unsigned long offset)
547 read_lock_irq(&mapping->tree_lock);
549 page = radix_tree_lookup(&mapping->page_tree, offset);
551 page_cache_get(page);
552 if (TestSetPageLocked(page)) {
553 read_unlock_irq(&mapping->tree_lock);
555 read_lock_irq(&mapping->tree_lock);
557 /* Has the page been truncated while we slept? */
558 if (page->mapping != mapping || page->index != offset) {
560 page_cache_release(page);
565 read_unlock_irq(&mapping->tree_lock);
569 EXPORT_SYMBOL(find_lock_page);
572 * find_or_create_page - locate or add a pagecache page
574 * @mapping: the page's address_space
575 * @index: the page's index into the mapping
576 * @gfp_mask: page allocation mode
578 * Locates a page in the pagecache. If the page is not present, a new page
579 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
580 * LRU list. The returned page is locked and has its reference count
583 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
586 * find_or_create_page() returns the desired page's address, or zero on
589 struct page *find_or_create_page(struct address_space *mapping,
590 unsigned long index, unsigned int gfp_mask)
592 struct page *page, *cached_page = NULL;
595 page = find_lock_page(mapping, index);
598 cached_page = alloc_page(gfp_mask);
602 err = add_to_page_cache_lru(cached_page, mapping,
607 } else if (err == -EEXIST)
611 page_cache_release(cached_page);
615 EXPORT_SYMBOL(find_or_create_page);
618 * find_get_pages - gang pagecache lookup
619 * @mapping: The address_space to search
620 * @start: The starting page index
621 * @nr_pages: The maximum number of pages
622 * @pages: Where the resulting pages are placed
624 * find_get_pages() will search for and return a group of up to
625 * @nr_pages pages in the mapping. The pages are placed at @pages.
626 * find_get_pages() takes a reference against the returned pages.
628 * The search returns a group of mapping-contiguous pages with ascending
629 * indexes. There may be holes in the indices due to not-present pages.
631 * find_get_pages() returns the number of pages which were found.
633 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
634 unsigned int nr_pages, struct page **pages)
639 read_lock_irq(&mapping->tree_lock);
640 ret = radix_tree_gang_lookup(&mapping->page_tree,
641 (void **)pages, start, nr_pages);
642 for (i = 0; i < ret; i++)
643 page_cache_get(pages[i]);
644 read_unlock_irq(&mapping->tree_lock);
649 * Like find_get_pages, except we only return pages which are tagged with
650 * `tag'. We update *index to index the next page for the traversal.
652 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
653 int tag, unsigned int nr_pages, struct page **pages)
658 read_lock_irq(&mapping->tree_lock);
659 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
660 (void **)pages, *index, nr_pages, tag);
661 for (i = 0; i < ret; i++)
662 page_cache_get(pages[i]);
664 *index = pages[ret - 1]->index + 1;
665 read_unlock_irq(&mapping->tree_lock);
670 * Same as grab_cache_page, but do not wait if the page is unavailable.
671 * This is intended for speculative data generators, where the data can
672 * be regenerated if the page couldn't be grabbed. This routine should
673 * be safe to call while holding the lock for another page.
675 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
676 * and deadlock against the caller's locked page.
679 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
681 struct page *page = find_get_page(mapping, index);
682 unsigned int gfp_mask;
685 if (!TestSetPageLocked(page))
687 page_cache_release(page);
690 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
691 page = alloc_pages(gfp_mask, 0);
692 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
693 page_cache_release(page);
699 EXPORT_SYMBOL(grab_cache_page_nowait);
702 * This is a generic file read routine, and uses the
703 * mapping->a_ops->readpage() function for the actual low-level
706 * This is really ugly. But the goto's actually try to clarify some
707 * of the logic when it comes to error handling etc.
709 * Note the struct file* is only passed for the use of readpage. It may be
712 void do_generic_mapping_read(struct address_space *mapping,
713 struct file_ra_state *_ra,
716 read_descriptor_t *desc,
720 struct inode *inode = mapping->host;
722 unsigned long end_index;
723 unsigned long offset;
724 unsigned long last_index;
725 unsigned long next_index;
726 unsigned long prev_index;
728 struct page *cached_page;
730 struct file_ra_state ra = *_ra;
733 index = *ppos >> PAGE_CACHE_SHIFT;
735 prev_index = ra.prev_page;
736 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
737 offset = *ppos & ~PAGE_CACHE_MASK;
739 isize = i_size_read(inode);
743 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
746 unsigned long nr, ret;
748 /* nr is the maximum number of bytes to copy from this page */
749 nr = PAGE_CACHE_SIZE;
750 if (index >= end_index) {
751 if (index > end_index)
753 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
761 if (index == next_index)
762 next_index = page_cache_readahead(mapping, &ra, filp,
763 index, last_index - index);
766 page = find_get_page(mapping, index);
767 if (unlikely(page == NULL)) {
769 desc->error = -EWOULDBLOCKIO;
772 handle_ra_miss(mapping, &ra, index);
775 if (!PageUptodate(page)) {
777 page_cache_release(page);
778 desc->error = -EWOULDBLOCKIO;
781 goto page_not_up_to_date;
785 /* If users can be writing to this page using arbitrary
786 * virtual addresses, take care about potential aliasing
787 * before reading the page on the kernel side.
789 if (mapping_writably_mapped(mapping))
790 flush_dcache_page(page);
793 * When (part of) the same page is read multiple times
794 * in succession, only mark it as accessed the first time.
796 if (prev_index != index)
797 mark_page_accessed(page);
801 * Ok, we have the page, and it's up-to-date, so
802 * now we can copy it to user space...
804 * The actor routine returns how many bytes were actually used..
805 * NOTE! This may not be the same as how much of a user buffer
806 * we filled up (we may be padding etc), so we can only update
807 * "pos" here (the actor routine has to update the user buffer
808 * pointers and the remaining count).
810 ret = actor(desc, page, offset, nr);
812 index += offset >> PAGE_CACHE_SHIFT;
813 offset &= ~PAGE_CACHE_MASK;
815 page_cache_release(page);
816 if (ret == nr && desc->count)
821 /* Get exclusive access to the page ... */
824 /* Did it get unhashed before we got the lock? */
825 if (!page->mapping) {
827 page_cache_release(page);
831 /* Did somebody else fill it already? */
832 if (PageUptodate(page)) {
838 /* Start the actual read. The read will unlock the page. */
839 error = mapping->a_ops->readpage(filp, page);
844 if (!PageUptodate(page)) {
846 if (!PageUptodate(page)) {
847 if (page->mapping == NULL) {
849 * invalidate_inode_pages got it
852 page_cache_release(page);
863 * i_size must be checked after we have done ->readpage.
865 * Checking i_size after the readpage allows us to calculate
866 * the correct value for "nr", which means the zero-filled
867 * part of the page is not copied back to userspace (unless
868 * another truncate extends the file - this is desired though).
870 isize = i_size_read(inode);
871 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
872 if (unlikely(!isize || index > end_index)) {
873 page_cache_release(page);
877 /* nr is the maximum number of bytes to copy from this page */
878 nr = PAGE_CACHE_SIZE;
879 if (index == end_index) {
880 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
882 page_cache_release(page);
890 /* UHHUH! A synchronous read error occurred. Report it */
892 page_cache_release(page);
897 * Ok, it wasn't cached, so we need to create a new
901 cached_page = page_cache_alloc_cold(mapping);
903 desc->error = -ENOMEM;
907 error = add_to_page_cache_lru(cached_page, mapping,
910 if (error == -EEXIST)
923 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
925 page_cache_release(cached_page);
930 EXPORT_SYMBOL(do_generic_mapping_read);
932 int file_read_actor(read_descriptor_t *desc, struct page *page,
933 unsigned long offset, unsigned long size)
936 unsigned long left, count = desc->count;
942 * Faults on the destination of a read are common, so do it before
945 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
946 kaddr = kmap_atomic(page, KM_USER0);
947 left = __copy_to_user_inatomic(desc->arg.buf,
948 kaddr + offset, size);
949 kunmap_atomic(kaddr, KM_USER0);
954 /* Do it the slow way */
956 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
961 desc->error = -EFAULT;
964 desc->count = count - size;
965 desc->written += size;
966 desc->arg.buf += size;
971 * This is the "read()" routine for all filesystems
972 * that can use the page cache directly.
975 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
976 unsigned long nr_segs, loff_t *ppos)
978 struct file *filp = iocb->ki_filp;
984 for (seg = 0; seg < nr_segs; seg++) {
985 const struct iovec *iv = &iov[seg];
988 * If any segment has a negative length, or the cumulative
989 * length ever wraps negative then return -EINVAL.
991 count += iv->iov_len;
992 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
994 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
999 count -= iv->iov_len; /* This segment is no good */
1003 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1004 if (filp->f_flags & O_DIRECT) {
1005 loff_t pos = *ppos, size;
1006 struct address_space *mapping;
1007 struct inode *inode;
1009 mapping = filp->f_mapping;
1010 inode = mapping->host;
1013 goto out; /* skip atime */
1014 size = i_size_read(inode);
1016 retval = generic_file_direct_IO(READ, iocb,
1018 if (retval > 0 && !is_sync_kiocb(iocb))
1019 retval = -EIOCBQUEUED;
1021 *ppos = pos + retval;
1023 file_accessed(filp);
1029 for (seg = 0; seg < nr_segs; seg++) {
1030 read_descriptor_t desc;
1033 desc.arg.buf = iov[seg].iov_base;
1034 desc.count = iov[seg].iov_len;
1035 if (desc.count == 0)
1038 do_generic_file_read(filp,ppos,&desc,file_read_actor,0);
1039 retval += desc.written;
1041 retval = desc.error;
1050 EXPORT_SYMBOL(__generic_file_aio_read);
1053 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1055 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1057 BUG_ON(iocb->ki_pos != pos);
1058 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1061 EXPORT_SYMBOL(generic_file_aio_read);
1064 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1066 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1070 init_sync_kiocb(&kiocb, filp);
1071 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1072 if (-EIOCBQUEUED == ret)
1073 ret = wait_on_sync_kiocb(&kiocb);
1077 EXPORT_SYMBOL(generic_file_read);
1079 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1082 unsigned long count = desc->count;
1083 struct file *file = desc->arg.data;
1088 written = file->f_op->sendpage(file, page, offset,
1089 size, &file->f_pos, size<count);
1091 desc->error = written;
1094 desc->count = count - written;
1095 desc->written += written;
1099 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1100 size_t count, read_actor_t actor, void *target)
1102 read_descriptor_t desc;
1109 desc.arg.data = target;
1112 do_generic_file_read(in_file, ppos, &desc, actor, 0);
1114 return desc.written;
1118 EXPORT_SYMBOL(generic_file_sendfile);
1121 do_readahead(struct address_space *mapping, struct file *filp,
1122 unsigned long index, unsigned long nr)
1124 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1127 force_page_cache_readahead(mapping, filp, index,
1128 max_sane_readahead(nr));
1132 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1140 if (file->f_mode & FMODE_READ) {
1141 struct address_space *mapping = file->f_mapping;
1142 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1143 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1144 unsigned long len = end - start + 1;
1145 ret = do_readahead(mapping, file, start, len);
1154 * This adds the requested page to the page cache if it isn't already there,
1155 * and schedules an I/O to read in its contents from disk.
1157 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1158 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1160 struct address_space *mapping = file->f_mapping;
1164 page = page_cache_alloc_cold(mapping);
1168 error = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1170 error = mapping->a_ops->readpage(file, page);
1171 page_cache_release(page);
1176 * We arrive here in the unlikely event that someone
1177 * raced with us and added our page to the cache first
1178 * or we are out of memory for radix-tree nodes.
1180 page_cache_release(page);
1181 return error == -EEXIST ? 0 : error;
1184 #define MMAP_LOTSAMISS (100)
1187 * filemap_nopage() is invoked via the vma operations vector for a
1188 * mapped memory region to read in file data during a page fault.
1190 * The goto's are kind of ugly, but this streamlines the normal case of having
1191 * it in the page cache, and handles the special cases reasonably without
1192 * having a lot of duplicated code.
1194 struct page *filemap_nopage(struct vm_area_struct *area,
1195 unsigned long address, int *type)
1198 struct file *file = area->vm_file;
1199 struct address_space *mapping = file->f_mapping;
1200 struct file_ra_state *ra = &file->f_ra;
1201 struct inode *inode = mapping->host;
1203 unsigned long size, pgoff;
1204 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1206 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1209 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1211 goto outside_data_content;
1213 /* If we don't want any read-ahead, don't bother */
1214 if (VM_RandomReadHint(area))
1215 goto no_cached_page;
1218 * The readahead code wants to be told about each and every page
1219 * so it can build and shrink its windows appropriately
1221 * For sequential accesses, we use the generic readahead logic.
1223 if (VM_SequentialReadHint(area))
1224 page_cache_readahead(mapping, ra, file, pgoff, 1);
1227 * Do we have something in the page cache already?
1230 page = find_get_page(mapping, pgoff);
1232 unsigned long ra_pages;
1234 if (VM_SequentialReadHint(area)) {
1235 handle_ra_miss(mapping, ra, pgoff);
1236 goto no_cached_page;
1241 * Do we miss much more than hit in this file? If so,
1242 * stop bothering with read-ahead. It will only hurt.
1244 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1245 goto no_cached_page;
1248 * To keep the pgmajfault counter straight, we need to
1249 * check did_readaround, as this is an inner loop.
1251 if (!did_readaround) {
1252 majmin = VM_FAULT_MAJOR;
1253 inc_page_state(pgmajfault);
1256 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1260 if (pgoff > ra_pages / 2)
1261 start = pgoff - ra_pages / 2;
1262 do_page_cache_readahead(mapping, file, start, ra_pages);
1264 page = find_get_page(mapping, pgoff);
1266 goto no_cached_page;
1269 if (!did_readaround)
1273 * Ok, found a page in the page cache, now we need to check
1274 * that it's up-to-date.
1276 if (!PageUptodate(page))
1277 goto page_not_uptodate;
1281 * Found the page and have a reference on it.
1283 mark_page_accessed(page);
1288 outside_data_content:
1290 * An external ptracer can access pages that normally aren't
1293 if (area->vm_mm == current->mm)
1295 /* Fall through to the non-read-ahead case */
1298 * We're only likely to ever get here if MADV_RANDOM is in
1301 error = page_cache_read(file, pgoff);
1305 * The page we want has now been added to the page cache.
1306 * In the unlikely event that someone removed it in the
1307 * meantime, we'll just come back here and read it again.
1313 * An error return from page_cache_read can result if the
1314 * system is low on memory, or a problem occurs while trying
1317 if (error == -ENOMEM)
1322 if (!did_readaround) {
1323 majmin = VM_FAULT_MAJOR;
1324 inc_page_state(pgmajfault);
1328 /* Did it get unhashed while we waited for it? */
1329 if (!page->mapping) {
1331 page_cache_release(page);
1335 /* Did somebody else get it up-to-date? */
1336 if (PageUptodate(page)) {
1341 if (!mapping->a_ops->readpage(file, page)) {
1342 wait_on_page_locked(page);
1343 if (PageUptodate(page))
1348 * Umm, take care of errors if the page isn't up-to-date.
1349 * Try to re-read it _once_. We do this synchronously,
1350 * because there really aren't any performance issues here
1351 * and we need to check for errors.
1355 /* Somebody truncated the page on us? */
1356 if (!page->mapping) {
1358 page_cache_release(page);
1362 /* Somebody else successfully read it in? */
1363 if (PageUptodate(page)) {
1367 ClearPageError(page);
1368 if (!mapping->a_ops->readpage(file, page)) {
1369 wait_on_page_locked(page);
1370 if (PageUptodate(page))
1375 * Things didn't work out. Return zero to tell the
1376 * mm layer so, possibly freeing the page cache page first.
1378 page_cache_release(page);
1382 EXPORT_SYMBOL(filemap_nopage);
1384 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1387 struct address_space *mapping = file->f_mapping;
1392 * Do we have something in the page cache already?
1395 page = find_get_page(mapping, pgoff);
1399 goto no_cached_page;
1403 * Ok, found a page in the page cache, now we need to check
1404 * that it's up-to-date.
1406 if (!PageUptodate(page)) {
1408 page_cache_release(page);
1411 goto page_not_uptodate;
1416 * Found the page and have a reference on it.
1418 mark_page_accessed(page);
1422 error = page_cache_read(file, pgoff);
1425 * The page we want has now been added to the page cache.
1426 * In the unlikely event that someone removed it in the
1427 * meantime, we'll just come back here and read it again.
1433 * An error return from page_cache_read can result if the
1434 * system is low on memory, or a problem occurs while trying
1442 /* Did it get unhashed while we waited for it? */
1443 if (!page->mapping) {
1448 /* Did somebody else get it up-to-date? */
1449 if (PageUptodate(page)) {
1454 if (!mapping->a_ops->readpage(file, page)) {
1455 wait_on_page_locked(page);
1456 if (PageUptodate(page))
1461 * Umm, take care of errors if the page isn't up-to-date.
1462 * Try to re-read it _once_. We do this synchronously,
1463 * because there really aren't any performance issues here
1464 * and we need to check for errors.
1468 /* Somebody truncated the page on us? */
1469 if (!page->mapping) {
1473 /* Somebody else successfully read it in? */
1474 if (PageUptodate(page)) {
1479 ClearPageError(page);
1480 if (!mapping->a_ops->readpage(file, page)) {
1481 wait_on_page_locked(page);
1482 if (PageUptodate(page))
1487 * Things didn't work out. Return zero to tell the
1488 * mm layer so, possibly freeing the page cache page first.
1491 page_cache_release(page);
1496 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1497 unsigned long len, pgprot_t prot, unsigned long pgoff,
1500 struct file *file = vma->vm_file;
1501 struct address_space *mapping = file->f_mapping;
1502 struct inode *inode = mapping->host;
1504 struct mm_struct *mm = vma->vm_mm;
1509 force_page_cache_readahead(mapping, vma->vm_file,
1510 pgoff, len >> PAGE_CACHE_SHIFT);
1513 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1514 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1517 page = filemap_getpage(file, pgoff, nonblock);
1518 if (!page && !nonblock)
1521 err = install_page(mm, vma, addr, page, prot);
1523 page_cache_release(page);
1527 err = install_file_pte(mm, vma, addr, pgoff, prot);
1541 struct vm_operations_struct generic_file_vm_ops = {
1542 .nopage = filemap_nopage,
1543 .populate = filemap_populate,
1546 /* This is used for a general mmap of a disk file */
1548 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1550 struct address_space *mapping = file->f_mapping;
1552 if (!mapping->a_ops->readpage)
1554 file_accessed(file);
1555 vma->vm_ops = &generic_file_vm_ops;
1558 EXPORT_SYMBOL(filemap_populate);
1561 * This is for filesystems which do not implement ->writepage.
1563 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1565 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1567 return generic_file_mmap(file, vma);
1570 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1574 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1578 #endif /* CONFIG_MMU */
1580 EXPORT_SYMBOL(generic_file_mmap);
1581 EXPORT_SYMBOL(generic_file_readonly_mmap);
1583 static inline struct page *__read_cache_page(struct address_space *mapping,
1584 unsigned long index,
1585 int (*filler)(void *,struct page*),
1588 struct page *page, *cached_page = NULL;
1591 page = find_get_page(mapping, index);
1594 cached_page = page_cache_alloc_cold(mapping);
1596 return ERR_PTR(-ENOMEM);
1598 err = add_to_page_cache_lru(cached_page, mapping,
1603 /* Presumably ENOMEM for radix tree node */
1604 page_cache_release(cached_page);
1605 return ERR_PTR(err);
1609 err = filler(data, page);
1611 page_cache_release(page);
1612 page = ERR_PTR(err);
1616 page_cache_release(cached_page);
1621 * Read into the page cache. If a page already exists,
1622 * and PageUptodate() is not set, try to fill the page.
1624 struct page *read_cache_page(struct address_space *mapping,
1625 unsigned long index,
1626 int (*filler)(void *,struct page*),
1633 page = __read_cache_page(mapping, index, filler, data);
1636 mark_page_accessed(page);
1637 if (PageUptodate(page))
1641 if (!page->mapping) {
1643 page_cache_release(page);
1646 if (PageUptodate(page)) {
1650 err = filler(data, page);
1652 page_cache_release(page);
1653 page = ERR_PTR(err);
1659 EXPORT_SYMBOL(read_cache_page);
1662 * If the page was newly created, increment its refcount and add it to the
1663 * caller's lru-buffering pagevec. This function is specifically for
1664 * generic_file_write().
1666 static inline struct page *
1667 __grab_cache_page(struct address_space *mapping, unsigned long index,
1668 struct page **cached_page, struct pagevec *lru_pvec)
1673 page = find_lock_page(mapping, index);
1675 if (!*cached_page) {
1676 *cached_page = page_cache_alloc(mapping);
1680 err = add_to_page_cache(*cached_page, mapping,
1685 page = *cached_page;
1686 page_cache_get(page);
1687 if (!pagevec_add(lru_pvec, page))
1688 __pagevec_lru_add(lru_pvec);
1689 *cached_page = NULL;
1696 * The logic we want is
1698 * if suid or (sgid and xgrp)
1701 int remove_suid(struct dentry *dentry)
1703 mode_t mode = dentry->d_inode->i_mode;
1707 /* suid always must be killed */
1708 if (unlikely(mode & S_ISUID))
1709 kill = ATTR_KILL_SUID;
1712 * sgid without any exec bits is just a mandatory locking mark; leave
1713 * it alone. If some exec bits are set, it's a real sgid; kill it.
1715 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1716 kill |= ATTR_KILL_SGID;
1718 if (unlikely(kill && !capable(CAP_FSETID))) {
1719 struct iattr newattrs;
1721 newattrs.ia_valid = ATTR_FORCE | kill;
1722 result = notify_change(dentry, &newattrs);
1726 EXPORT_SYMBOL(remove_suid);
1729 * Copy as much as we can into the page and return the number of bytes which
1730 * were sucessfully copied. If a fault is encountered then clear the page
1731 * out to (offset+bytes) and return the number of bytes which were copied.
1733 static inline size_t
1734 filemap_copy_from_user(struct page *page, unsigned long offset,
1735 const char __user *buf, unsigned bytes)
1740 kaddr = kmap_atomic(page, KM_USER0);
1741 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1742 kunmap_atomic(kaddr, KM_USER0);
1745 /* Do it the slow way */
1747 left = __copy_from_user(kaddr + offset, buf, bytes);
1750 return bytes - left;
1754 __filemap_copy_from_user_iovec(char *vaddr,
1755 const struct iovec *iov, size_t base, size_t bytes)
1757 size_t copied = 0, left = 0;
1760 char __user *buf = iov->iov_base + base;
1761 int copy = min(bytes, iov->iov_len - base);
1764 left = __copy_from_user_inatomic(vaddr, buf, copy);
1770 if (unlikely(left)) {
1771 /* zero the rest of the target like __copy_from_user */
1773 memset(vaddr, 0, bytes);
1777 return copied - left;
1781 * This has the same sideeffects and return value as filemap_copy_from_user().
1782 * The difference is that on a fault we need to memset the remainder of the
1783 * page (out to offset+bytes), to emulate filemap_copy_from_user()'s
1784 * single-segment behaviour.
1786 static inline size_t
1787 filemap_copy_from_user_iovec(struct page *page, unsigned long offset,
1788 const struct iovec *iov, size_t base, size_t bytes)
1793 kaddr = kmap_atomic(page, KM_USER0);
1794 copied = __filemap_copy_from_user_iovec(kaddr + offset, iov,
1796 kunmap_atomic(kaddr, KM_USER0);
1797 if (copied != bytes) {
1799 copied = __filemap_copy_from_user_iovec(kaddr + offset, iov,
1807 filemap_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1809 const struct iovec *iov = *iovp;
1810 size_t base = *basep;
1813 int copy = min(bytes, iov->iov_len - base);
1817 if (iov->iov_len == base) {
1827 * Performs necessary checks before doing a write
1829 * Can adjust writing position aor amount of bytes to write.
1830 * Returns appropriate error code that caller should return or
1831 * zero in case that write should be allowed.
1833 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1835 struct inode *inode = file->f_mapping->host;
1836 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1838 if (unlikely(*pos < 0))
1841 if (unlikely(file->f_error)) {
1842 int err = file->f_error;
1848 /* FIXME: this is for backwards compatibility with 2.4 */
1849 if (file->f_flags & O_APPEND)
1850 *pos = i_size_read(inode);
1852 if (limit != RLIM_INFINITY) {
1853 if (*pos >= limit) {
1854 send_sig(SIGXFSZ, current, 0);
1857 if (*count > limit - (typeof(limit))*pos) {
1858 *count = limit - (typeof(limit))*pos;
1866 if (unlikely(*pos + *count > MAX_NON_LFS &&
1867 !(file->f_flags & O_LARGEFILE))) {
1868 if (*pos >= MAX_NON_LFS) {
1869 send_sig(SIGXFSZ, current, 0);
1872 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1873 *count = MAX_NON_LFS - (unsigned long)*pos;
1878 * Are we about to exceed the fs block limit ?
1880 * If we have written data it becomes a short write. If we have
1881 * exceeded without writing data we send a signal and return EFBIG.
1882 * Linus frestrict idea will clean these up nicely..
1884 if (likely(!isblk)) {
1885 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1886 if (*count || *pos > inode->i_sb->s_maxbytes) {
1887 send_sig(SIGXFSZ, current, 0);
1890 /* zero-length writes at ->s_maxbytes are OK */
1893 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1894 *count = inode->i_sb->s_maxbytes - *pos;
1897 if (bdev_read_only(I_BDEV(inode)))
1899 isize = i_size_read(inode);
1900 if (*pos >= isize) {
1901 if (*count || *pos > isize)
1905 if (*pos + *count > isize)
1906 *count = isize - *pos;
1910 EXPORT_SYMBOL(generic_write_checks);
1913 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1914 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1915 size_t count, size_t ocount)
1917 struct file *file = iocb->ki_filp;
1918 struct address_space *mapping = file->f_mapping;
1919 struct inode *inode = mapping->host;
1922 if (count != ocount)
1923 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1925 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1927 loff_t end = pos + written;
1928 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1929 i_size_write(inode, end);
1930 mark_inode_dirty(inode);
1936 * Sync the fs metadata but not the minor inode changes and
1937 * of course not the data as we did direct DMA for the IO.
1938 * i_sem is held, which protects generic_osync_inode() from
1941 if (written >= 0 && file->f_flags & O_SYNC)
1942 generic_osync_inode(inode, mapping, OSYNC_METADATA);
1943 if (written == count && !is_sync_kiocb(iocb))
1944 written = -EIOCBQUEUED;
1947 EXPORT_SYMBOL(generic_file_direct_write);
1950 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
1951 unsigned long nr_segs, loff_t pos, loff_t *ppos,
1952 size_t count, ssize_t written)
1954 struct file *file = iocb->ki_filp;
1955 struct address_space * mapping = file->f_mapping;
1956 struct address_space_operations *a_ops = mapping->a_ops;
1957 struct inode *inode = mapping->host;
1960 struct page *cached_page = NULL;
1962 struct pagevec lru_pvec;
1963 const struct iovec *cur_iov = iov; /* current iovec */
1964 size_t iov_base = 0; /* offset in the current iovec */
1967 pagevec_init(&lru_pvec, 0);
1970 * handle partial DIO write. Adjust cur_iov if needed.
1972 if (likely(nr_segs == 1))
1973 buf = iov->iov_base + written;
1975 filemap_set_next_iovec(&cur_iov, &iov_base, written);
1976 buf = cur_iov->iov_base + iov_base;
1980 unsigned long index;
1981 unsigned long offset;
1982 unsigned long maxlen;
1985 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1986 index = pos >> PAGE_CACHE_SHIFT;
1987 bytes = PAGE_CACHE_SIZE - offset;
1992 * Bring in the user page that we will copy from _first_.
1993 * Otherwise there's a nasty deadlock on copying from the
1994 * same page as we're writing to, without it being marked
1997 maxlen = cur_iov->iov_len - iov_base;
2000 fault_in_pages_readable(buf, maxlen);
2002 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2008 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2009 if (unlikely(status)) {
2010 loff_t isize = i_size_read(inode);
2012 * prepare_write() may have instantiated a few blocks
2013 * outside i_size. Trim these off again.
2016 page_cache_release(page);
2017 if (pos + bytes > isize)
2018 vmtruncate(inode, isize);
2021 if (likely(nr_segs == 1))
2022 copied = filemap_copy_from_user(page, offset,
2025 copied = filemap_copy_from_user_iovec(page, offset,
2026 cur_iov, iov_base, bytes);
2027 flush_dcache_page(page);
2028 status = a_ops->commit_write(file, page, offset, offset+bytes);
2029 if (likely(copied > 0)) {
2038 if (unlikely(nr_segs > 1)) {
2039 filemap_set_next_iovec(&cur_iov,
2041 buf = cur_iov->iov_base + iov_base;
2047 if (unlikely(copied != bytes))
2051 mark_page_accessed(page);
2052 page_cache_release(page);
2055 balance_dirty_pages_ratelimited(mapping);
2061 page_cache_release(cached_page);
2064 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2066 if (likely(status >= 0)) {
2067 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2068 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2069 status = generic_osync_inode(inode, mapping,
2070 OSYNC_METADATA|OSYNC_DATA);
2075 * If we get here for O_DIRECT writes then we must have fallen through
2076 * to buffered writes (block instantiation inside i_size). So we sync
2077 * the file data here, to try to honour O_DIRECT expectations.
2079 if (unlikely(file->f_flags & O_DIRECT) && written)
2080 status = filemap_write_and_wait(mapping);
2082 pagevec_lru_add(&lru_pvec);
2083 return written ? written : status;
2085 EXPORT_SYMBOL(generic_file_buffered_write);
2088 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2089 unsigned long nr_segs, loff_t *ppos)
2091 struct file *file = iocb->ki_filp;
2092 struct address_space * mapping = file->f_mapping;
2093 size_t ocount; /* original count */
2094 size_t count; /* after file limit checks */
2095 struct inode *inode = mapping->host;
2102 for (seg = 0; seg < nr_segs; seg++) {
2103 const struct iovec *iv = &iov[seg];
2106 * If any segment has a negative length, or the cumulative
2107 * length ever wraps negative then return -EINVAL.
2109 ocount += iv->iov_len;
2110 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2112 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2117 ocount -= iv->iov_len; /* This segment is no good */
2124 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2126 /* We can write back this queue in page reclaim */
2127 current->backing_dev_info = mapping->backing_dev_info;
2130 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2137 err = remove_suid(file->f_dentry);
2141 inode_update_time(inode, file->f_vfsmnt, 1);
2143 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2144 if (unlikely(file->f_flags & O_DIRECT)) {
2145 written = generic_file_direct_write(iocb, iov,
2146 &nr_segs, pos, ppos, count, ocount);
2147 if (written < 0 || written == count)
2150 * direct-io write to a hole: fall through to buffered I/O
2151 * for completing the rest of the request.
2157 written = generic_file_buffered_write(iocb, iov, nr_segs,
2158 pos, ppos, count, written);
2160 current->backing_dev_info = NULL;
2161 return written ? written : err;
2163 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2166 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2167 unsigned long nr_segs, loff_t *ppos)
2169 struct file *file = iocb->ki_filp;
2170 struct address_space *mapping = file->f_mapping;
2171 struct inode *inode = mapping->host;
2175 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2177 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2180 err = sync_page_range_nolock(inode, mapping, pos, ret);
2188 __generic_file_write_nolock(struct file *file, const struct iovec *iov,
2189 unsigned long nr_segs, loff_t *ppos)
2194 init_sync_kiocb(&kiocb, file);
2195 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2196 if (ret == -EIOCBQUEUED)
2197 ret = wait_on_sync_kiocb(&kiocb);
2202 generic_file_write_nolock(struct file *file, const struct iovec *iov,
2203 unsigned long nr_segs, loff_t *ppos)
2208 init_sync_kiocb(&kiocb, file);
2209 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2210 if (-EIOCBQUEUED == ret)
2211 ret = wait_on_sync_kiocb(&kiocb);
2214 EXPORT_SYMBOL(generic_file_write_nolock);
2216 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2217 size_t count, loff_t pos)
2219 struct file *file = iocb->ki_filp;
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 BUG_ON(iocb->ki_pos != pos);
2228 down(&inode->i_sem);
2229 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2233 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2236 err = sync_page_range(inode, mapping, pos, ret);
2242 EXPORT_SYMBOL(generic_file_aio_write);
2244 ssize_t generic_file_write(struct file *file, const char __user *buf,
2245 size_t count, loff_t *ppos)
2247 struct address_space *mapping = file->f_mapping;
2248 struct inode *inode = mapping->host;
2250 struct iovec local_iov = { .iov_base = (void __user *)buf,
2253 down(&inode->i_sem);
2254 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
2257 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2260 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2266 EXPORT_SYMBOL(generic_file_write);
2268 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2269 unsigned long nr_segs, loff_t *ppos)
2274 init_sync_kiocb(&kiocb, filp);
2275 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2276 if (-EIOCBQUEUED == ret)
2277 ret = wait_on_sync_kiocb(&kiocb);
2280 EXPORT_SYMBOL(generic_file_readv);
2282 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2283 unsigned long nr_segs, loff_t *ppos)
2285 struct address_space *mapping = file->f_mapping;
2286 struct inode *inode = mapping->host;
2289 down(&inode->i_sem);
2290 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
2293 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2296 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2302 EXPORT_SYMBOL(generic_file_writev);
2305 * Called under i_sem for writes to S_ISREG files. Returns -EIO if something
2306 * went wrong during pagecache shootdown.
2309 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2310 loff_t offset, unsigned long nr_segs)
2312 struct file *file = iocb->ki_filp;
2313 struct address_space *mapping = file->f_mapping;
2315 size_t write_len = 0;
2318 * If it's a write, unmap all mmappings of the file up-front. This
2319 * will cause any pte dirty bits to be propagated into the pageframes
2320 * for the subsequent filemap_write_and_wait().
2323 write_len = iov_length(iov, nr_segs);
2324 if (mapping_mapped(mapping))
2325 unmap_mapping_range(mapping, offset, write_len, 0);
2328 retval = filemap_write_and_wait(mapping);
2330 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2332 if (rw == WRITE && mapping->nrpages) {
2333 pgoff_t end = (offset + write_len - 1)
2334 >> PAGE_CACHE_SHIFT;
2335 int err = invalidate_inode_pages2_range(mapping,
2336 offset >> PAGE_CACHE_SHIFT, end);
2343 EXPORT_SYMBOL_GPL(generic_file_direct_IO);