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>
31 * This is needed for the following functions:
32 * - try_to_release_page
33 * - block_invalidatepage
34 * - generic_osync_inode
36 * FIXME: remove all knowledge of the buffer layer from the core VM
38 #include <linux/buffer_head.h> /* for generic_osync_inode */
40 #include <asm/uaccess.h>
44 * Shared mappings implemented 30.11.1994. It's not fully working yet,
47 * Shared mappings now work. 15.8.1995 Bruno.
49 * finished 'unifying' the page and buffer cache and SMP-threaded the
50 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
52 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
58 * ->i_mmap_lock (vmtruncate)
59 * ->private_lock (__free_pte->__set_page_dirty_buffers)
61 * ->swap_device_lock (exclusive_swap_page, others)
62 * ->mapping->tree_lock
63 * ->page_map_lock() (try_to_unmap_file)
66 * ->i_mmap_lock (truncate->unmap_mapping_range)
70 * ->page_table_lock (various places, mainly in mmap.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 * ->swap_device_lock (try_to_unmap_one)
88 * ->private_lock (try_to_unmap_one)
89 * ->tree_lock (try_to_unmap_one)
90 * ->zone.lru_lock (follow_page->mark_page_accessed)
91 * ->page_map_lock() (page_add_anon_rmap)
92 * ->tree_lock (page_remove_rmap->set_page_dirty)
93 * ->private_lock (page_remove_rmap->set_page_dirty)
94 * ->inode_lock (page_remove_rmap->set_page_dirty)
95 * ->anon_vma.lock (anon_vma_prepare)
96 * ->inode_lock (zap_pte_range->set_page_dirty)
97 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
100 * ->dcache_lock (proc_pid_lookup)
104 * Remove a page from the page cache and free it. Caller has to make
105 * sure the page is locked and that nobody else uses it - or that usage
106 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
108 void __remove_from_page_cache(struct page *page)
110 struct address_space *mapping = page->mapping;
112 radix_tree_delete(&mapping->page_tree, page->index);
113 page->mapping = NULL;
118 void remove_from_page_cache(struct page *page)
120 struct address_space *mapping = page->mapping;
122 if (unlikely(!PageLocked(page)))
125 spin_lock_irq(&mapping->tree_lock);
126 __remove_from_page_cache(page);
127 spin_unlock_irq(&mapping->tree_lock);
130 static inline int sync_page(struct page *page)
132 struct address_space *mapping;
135 * FIXME, fercrissake. What is this barrier here for?
138 mapping = page_mapping(page);
139 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
140 return mapping->a_ops->sync_page(page);
145 * filemap_fdatawrite - start writeback against all of a mapping's dirty pages
146 * @mapping: address space structure to write
148 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
149 * opposed to a regular memory * cleansing writeback. The difference between
150 * these two operations is that if a dirty page/buffer is encountered, it must
151 * be waited upon, and not just skipped over.
153 static int __filemap_fdatawrite(struct address_space *mapping, int sync_mode)
156 struct writeback_control wbc = {
157 .sync_mode = sync_mode,
158 .nr_to_write = mapping->nrpages * 2,
161 if (mapping->backing_dev_info->memory_backed)
164 ret = do_writepages(mapping, &wbc);
168 int filemap_fdatawrite(struct address_space *mapping)
170 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
172 EXPORT_SYMBOL(filemap_fdatawrite);
175 * This is a mostly non-blocking flush. Not suitable for data-integrity
176 * purposes - I/O may not be started against all dirty pages.
178 int filemap_flush(struct address_space *mapping)
180 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
182 EXPORT_SYMBOL(filemap_flush);
185 * Wait for writeback to complete against pages indexed by start->end
188 static int wait_on_page_writeback_range(struct address_space *mapping,
189 pgoff_t start, pgoff_t end)
199 pagevec_init(&pvec, 0);
201 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
202 PAGECACHE_TAG_WRITEBACK,
203 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
206 for (i = 0; i < nr_pages; i++) {
207 struct page *page = pvec.pages[i];
209 wait_on_page_writeback(page);
213 pagevec_release(&pvec);
217 /* Check for outstanding write errors */
218 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
220 if (test_and_clear_bit(AS_EIO, &mapping->flags))
227 * filemap_fdatawait - walk the list of under-writeback pages of the given
228 * address space and wait for all of them.
230 * @mapping: address space structure to wait for
232 int filemap_fdatawait(struct address_space *mapping)
234 return wait_on_page_writeback_range(mapping, 0, -1);
237 EXPORT_SYMBOL(filemap_fdatawait);
239 int filemap_write_and_wait(struct address_space *mapping)
243 if (mapping->nrpages) {
244 retval = filemap_fdatawrite(mapping);
246 retval = filemap_fdatawait(mapping);
252 * This function is used to add newly allocated pagecache pages:
253 * the page is new, so we can just run SetPageLocked() against it.
254 * The other page state flags were set by rmqueue().
256 * This function does not add the page to the LRU. The caller must do that.
258 int add_to_page_cache(struct page *page, struct address_space *mapping,
259 pgoff_t offset, int gfp_mask)
261 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
264 spin_lock_irq(&mapping->tree_lock);
265 error = radix_tree_insert(&mapping->page_tree, offset, page);
267 page_cache_get(page);
269 page->mapping = mapping;
270 page->index = offset;
274 spin_unlock_irq(&mapping->tree_lock);
275 radix_tree_preload_end();
280 EXPORT_SYMBOL(add_to_page_cache);
282 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
283 pgoff_t offset, int gfp_mask)
285 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
292 * In order to wait for pages to become available there must be
293 * waitqueues associated with pages. By using a hash table of
294 * waitqueues where the bucket discipline is to maintain all
295 * waiters on the same queue and wake all when any of the pages
296 * become available, and for the woken contexts to check to be
297 * sure the appropriate page became available, this saves space
298 * at a cost of "thundering herd" phenomena during rare hash
301 struct page_wait_queue {
307 static int page_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
309 struct page *page = key;
310 struct page_wait_queue *wq;
312 wq = container_of(wait, struct page_wait_queue, wait);
313 if (wq->page != page || test_bit(wq->bit, &page->flags))
316 return autoremove_wake_function(wait, mode, sync, NULL);
319 #define __DEFINE_PAGE_WAIT(name, p, b, f) \
320 struct page_wait_queue name = { \
325 .func = page_wake_function, \
327 .task_list = LIST_HEAD_INIT(name.wait.task_list),\
331 #define DEFINE_PAGE_WAIT(name, p, b) __DEFINE_PAGE_WAIT(name, p, b, 0)
332 #define DEFINE_PAGE_WAIT_EXCLUSIVE(name, p, b) \
333 __DEFINE_PAGE_WAIT(name, p, b, WQ_FLAG_EXCLUSIVE)
335 static wait_queue_head_t *page_waitqueue(struct page *page)
337 const struct zone *zone = page_zone(page);
339 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
342 static void wake_up_page(struct page *page)
344 const unsigned int mode = TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE;
345 wait_queue_head_t *waitqueue = page_waitqueue(page);
347 if (waitqueue_active(waitqueue))
348 __wake_up(waitqueue, mode, 1, page);
351 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
353 wait_queue_head_t *waitqueue = page_waitqueue(page);
354 DEFINE_PAGE_WAIT(wait, page, bit_nr);
357 prepare_to_wait(waitqueue, &wait.wait, TASK_UNINTERRUPTIBLE);
358 if (test_bit(bit_nr, &page->flags)) {
362 } while (test_bit(bit_nr, &page->flags));
363 finish_wait(waitqueue, &wait.wait);
366 EXPORT_SYMBOL(wait_on_page_bit);
369 * unlock_page() - unlock a locked page
373 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
374 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
375 * mechananism between PageLocked pages and PageWriteback pages is shared.
376 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
378 * The first mb is necessary to safely close the critical section opened by the
379 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
380 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
381 * parallel wait_on_page_locked()).
383 void fastcall unlock_page(struct page *page)
385 smp_mb__before_clear_bit();
386 if (!TestClearPageLocked(page))
388 smp_mb__after_clear_bit();
392 EXPORT_SYMBOL(unlock_page);
393 EXPORT_SYMBOL(lock_page);
396 * End writeback against a page.
398 void end_page_writeback(struct page *page)
400 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
401 if (!test_clear_page_writeback(page))
403 smp_mb__after_clear_bit();
408 EXPORT_SYMBOL(end_page_writeback);
411 * Get a lock on the page, assuming we need to sleep to get it.
413 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
414 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
415 * chances are that on the second loop, the block layer's plug list is empty,
416 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
418 void fastcall __lock_page(struct page *page)
420 wait_queue_head_t *wqh = page_waitqueue(page);
421 DEFINE_PAGE_WAIT_EXCLUSIVE(wait, page, PG_locked);
423 while (TestSetPageLocked(page)) {
424 prepare_to_wait_exclusive(wqh, &wait.wait, TASK_UNINTERRUPTIBLE);
425 if (PageLocked(page)) {
430 finish_wait(wqh, &wait.wait);
433 EXPORT_SYMBOL(__lock_page);
436 * a rather lightweight function, finding and getting a reference to a
437 * hashed page atomically.
439 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
443 spin_lock_irq(&mapping->tree_lock);
444 page = radix_tree_lookup(&mapping->page_tree, offset);
446 page_cache_get(page);
447 spin_unlock_irq(&mapping->tree_lock);
451 EXPORT_SYMBOL(find_get_page);
454 * Same as above, but trylock it instead of incrementing the count.
456 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
460 spin_lock_irq(&mapping->tree_lock);
461 page = radix_tree_lookup(&mapping->page_tree, offset);
462 if (page && TestSetPageLocked(page))
464 spin_unlock_irq(&mapping->tree_lock);
468 EXPORT_SYMBOL(find_trylock_page);
471 * find_lock_page - locate, pin and lock a pagecache page
473 * @mapping - the address_space to search
474 * @offset - the page index
476 * Locates the desired pagecache page, locks it, increments its reference
477 * count and returns its address.
479 * Returns zero if the page was not present. find_lock_page() may sleep.
481 struct page *find_lock_page(struct address_space *mapping,
482 unsigned long offset)
486 spin_lock_irq(&mapping->tree_lock);
488 page = radix_tree_lookup(&mapping->page_tree, offset);
490 page_cache_get(page);
491 if (TestSetPageLocked(page)) {
492 spin_unlock_irq(&mapping->tree_lock);
494 spin_lock_irq(&mapping->tree_lock);
496 /* Has the page been truncated while we slept? */
497 if (page->mapping != mapping || page->index != offset) {
499 page_cache_release(page);
504 spin_unlock_irq(&mapping->tree_lock);
508 EXPORT_SYMBOL(find_lock_page);
511 * find_or_create_page - locate or add a pagecache page
513 * @mapping - the page's address_space
514 * @index - the page's index into the mapping
515 * @gfp_mask - page allocation mode
517 * Locates a page in the pagecache. If the page is not present, a new page
518 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
519 * LRU list. The returned page is locked and has its reference count
522 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
525 * find_or_create_page() returns the desired page's address, or zero on
528 struct page *find_or_create_page(struct address_space *mapping,
529 unsigned long index, unsigned int gfp_mask)
531 struct page *page, *cached_page = NULL;
534 page = find_lock_page(mapping, index);
537 cached_page = alloc_page(gfp_mask);
541 err = add_to_page_cache_lru(cached_page, mapping,
546 } else if (err == -EEXIST)
550 page_cache_release(cached_page);
554 EXPORT_SYMBOL(find_or_create_page);
557 * find_get_pages - gang pagecache lookup
558 * @mapping: The address_space to search
559 * @start: The starting page index
560 * @nr_pages: The maximum number of pages
561 * @pages: Where the resulting pages are placed
563 * find_get_pages() will search for and return a group of up to
564 * @nr_pages pages in the mapping. The pages are placed at @pages.
565 * find_get_pages() takes a reference against the returned pages.
567 * The search returns a group of mapping-contiguous pages with ascending
568 * indexes. There may be holes in the indices due to not-present pages.
570 * find_get_pages() returns the number of pages which were found.
572 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
573 unsigned int nr_pages, struct page **pages)
578 spin_lock_irq(&mapping->tree_lock);
579 ret = radix_tree_gang_lookup(&mapping->page_tree,
580 (void **)pages, start, nr_pages);
581 for (i = 0; i < ret; i++)
582 page_cache_get(pages[i]);
583 spin_unlock_irq(&mapping->tree_lock);
588 * Like find_get_pages, except we only return pages which are tagged with
589 * `tag'. We update *index to index the next page for the traversal.
591 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
592 int tag, unsigned int nr_pages, struct page **pages)
597 spin_lock_irq(&mapping->tree_lock);
598 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
599 (void **)pages, *index, nr_pages, tag);
600 for (i = 0; i < ret; i++)
601 page_cache_get(pages[i]);
603 *index = pages[ret - 1]->index + 1;
604 spin_unlock_irq(&mapping->tree_lock);
609 * Same as grab_cache_page, but do not wait if the page is unavailable.
610 * This is intended for speculative data generators, where the data can
611 * be regenerated if the page couldn't be grabbed. This routine should
612 * be safe to call while holding the lock for another page.
614 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
615 * and deadlock against the caller's locked page.
618 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
620 struct page *page = find_get_page(mapping, index);
624 if (!TestSetPageLocked(page))
626 page_cache_release(page);
629 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
630 page = alloc_pages(gfp_mask, 0);
631 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
632 page_cache_release(page);
638 EXPORT_SYMBOL(grab_cache_page_nowait);
641 * This is a generic file read routine, and uses the
642 * mapping->a_ops->readpage() function for the actual low-level
645 * This is really ugly. But the goto's actually try to clarify some
646 * of the logic when it comes to error handling etc.
647 * - note the struct file * is only passed for the use of readpage
649 void do_generic_mapping_read(struct address_space *mapping,
650 struct file_ra_state *_ra,
653 read_descriptor_t * desc,
657 struct inode *inode = mapping->host;
658 unsigned long index, end_index, offset;
660 struct page *cached_page;
662 struct file_ra_state ra = *_ra;
665 index = *ppos >> PAGE_CACHE_SHIFT;
666 offset = *ppos & ~PAGE_CACHE_MASK;
668 isize = i_size_read(inode);
669 end_index = isize >> PAGE_CACHE_SHIFT;
670 if (index > end_index)
675 unsigned long nr, ret;
678 page_cache_readahead(mapping, &ra, filp, index);
681 page = find_get_page(mapping, index);
682 if (unlikely(page == NULL)) {
684 desc->error = -EWOULDBLOCKIO;
687 handle_ra_miss(mapping, &ra, index);
690 if (!PageUptodate(page)) {
692 page_cache_release(page);
693 desc->error = -EWOULDBLOCKIO;
696 goto page_not_up_to_date;
699 /* nr is the maximum number of bytes to copy from this page */
700 nr = PAGE_CACHE_SIZE;
701 if (index == end_index) {
702 nr = isize & ~PAGE_CACHE_MASK;
704 page_cache_release(page);
710 /* If users can be writing to this page using arbitrary
711 * virtual addresses, take care about potential aliasing
712 * before reading the page on the kernel side.
714 if (mapping_writably_mapped(mapping))
715 flush_dcache_page(page);
718 * Mark the page accessed if we read the beginning.
721 mark_page_accessed(page);
724 * Ok, we have the page, and it's up-to-date, so
725 * now we can copy it to user space...
727 * The actor routine returns how many bytes were actually used..
728 * NOTE! This may not be the same as how much of a user buffer
729 * we filled up (we may be padding etc), so we can only update
730 * "pos" here (the actor routine has to update the user buffer
731 * pointers and the remaining count).
733 ret = actor(desc, page, offset, nr);
735 index += offset >> PAGE_CACHE_SHIFT;
736 offset &= ~PAGE_CACHE_MASK;
738 page_cache_release(page);
739 if (ret == nr && desc->count)
744 /* Get exclusive access to the page ... */
747 /* Did it get unhashed before we got the lock? */
748 if (!page->mapping) {
750 page_cache_release(page);
754 /* Did somebody else fill it already? */
755 if (PageUptodate(page)) {
761 /* Start the actual read. The read will unlock the page. */
762 error = mapping->a_ops->readpage(filp, page);
767 if (!PageUptodate(page)) {
768 wait_on_page_locked(page);
769 if (!PageUptodate(page)) {
776 * i_size must be checked after we have done ->readpage.
778 * Checking i_size after the readpage allows us to calculate
779 * the correct value for "nr", which means the zero-filled
780 * part of the page is not copied back to userspace (unless
781 * another truncate extends the file - this is desired though).
783 isize = i_size_read(inode);
784 end_index = isize >> PAGE_CACHE_SHIFT;
785 if (index > end_index) {
786 page_cache_release(page);
792 /* UHHUH! A synchronous read error occurred. Report it */
794 page_cache_release(page);
799 * Ok, it wasn't cached, so we need to create a new
803 cached_page = page_cache_alloc_cold(mapping);
805 desc->error = -ENOMEM;
809 error = add_to_page_cache_lru(cached_page, mapping,
812 if (error == -EEXIST)
825 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
827 page_cache_release(cached_page);
831 EXPORT_SYMBOL(do_generic_mapping_read);
833 int file_read_actor(read_descriptor_t *desc, struct page *page,
834 unsigned long offset, unsigned long size)
837 unsigned long left, count = desc->count;
843 * Faults on the destination of a read are common, so do it before
846 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
847 kaddr = kmap_atomic(page, KM_USER0);
848 left = __copy_to_user_inatomic(desc->arg.buf, kaddr + offset, size);
849 kunmap_atomic(kaddr, KM_USER0);
854 /* Do it the slow way */
856 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
861 desc->error = -EFAULT;
864 desc->count = count - size;
865 desc->written += size;
866 desc->arg.buf += size;
871 * This is the "read()" routine for all filesystems
872 * that can use the page cache directly.
875 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
876 unsigned long nr_segs, loff_t *ppos)
878 struct file *filp = iocb->ki_filp;
884 for (seg = 0; seg < nr_segs; seg++) {
885 const struct iovec *iv = &iov[seg];
888 * If any segment has a negative length, or the cumulative
889 * length ever wraps negative then return -EINVAL.
891 count += iv->iov_len;
892 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
894 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
899 count -= iv->iov_len; /* This segment is no good */
903 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
904 if (filp->f_flags & O_DIRECT) {
905 loff_t pos = *ppos, size;
906 struct address_space *mapping;
909 mapping = filp->f_mapping;
910 inode = mapping->host;
913 goto out; /* skip atime */
914 size = i_size_read(inode);
916 retval = generic_file_direct_IO(READ, iocb,
918 if (retval >= 0 && !is_sync_kiocb(iocb))
919 retval = -EIOCBQUEUED;
921 *ppos = pos + retval;
929 for (seg = 0; seg < nr_segs; seg++) {
930 read_descriptor_t desc;
933 desc.arg.buf = iov[seg].iov_base;
934 desc.count = iov[seg].iov_len;
938 do_generic_file_read(filp,ppos,&desc,file_read_actor,0);
939 retval += desc.written;
950 EXPORT_SYMBOL(__generic_file_aio_read);
953 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
955 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
957 BUG_ON(iocb->ki_pos != pos);
958 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
961 EXPORT_SYMBOL(generic_file_aio_read);
964 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
966 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
970 init_sync_kiocb(&kiocb, filp);
971 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
972 if (-EIOCBQUEUED == ret)
973 ret = wait_on_sync_kiocb(&kiocb);
977 EXPORT_SYMBOL(generic_file_read);
979 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
982 unsigned long count = desc->count;
983 struct file *file = desc->arg.data;
988 written = file->f_op->sendpage(file, page, offset,
989 size, &file->f_pos, size<count);
991 desc->error = written;
994 desc->count = count - written;
995 desc->written += written;
999 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1000 size_t count, read_actor_t actor, void *target)
1002 read_descriptor_t desc;
1009 desc.arg.data = target;
1012 do_generic_file_read(in_file, ppos, &desc, actor, 0);
1014 return desc.written;
1018 EXPORT_SYMBOL(generic_file_sendfile);
1021 do_readahead(struct address_space *mapping, struct file *filp,
1022 unsigned long index, unsigned long nr)
1024 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1027 force_page_cache_readahead(mapping, filp, index,
1028 max_sane_readahead(nr));
1032 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1040 if (file->f_mode & FMODE_READ) {
1041 struct address_space *mapping = file->f_mapping;
1042 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1043 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1044 unsigned long len = end - start + 1;
1045 ret = do_readahead(mapping, file, start, len);
1054 * This adds the requested page to the page cache if it isn't already there,
1055 * and schedules an I/O to read in its contents from disk.
1057 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1058 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1060 struct address_space *mapping = file->f_mapping;
1064 page = page_cache_alloc_cold(mapping);
1068 error = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1070 error = mapping->a_ops->readpage(file, page);
1071 page_cache_release(page);
1076 * We arrive here in the unlikely event that someone
1077 * raced with us and added our page to the cache first
1078 * or we are out of memory for radix-tree nodes.
1080 page_cache_release(page);
1081 return error == -EEXIST ? 0 : error;
1084 #define MMAP_LOTSAMISS (100)
1087 * filemap_nopage() is invoked via the vma operations vector for a
1088 * mapped memory region to read in file data during a page fault.
1090 * The goto's are kind of ugly, but this streamlines the normal case of having
1091 * it in the page cache, and handles the special cases reasonably without
1092 * having a lot of duplicated code.
1094 struct page * filemap_nopage(struct vm_area_struct * area, unsigned long address, int *type)
1097 struct file *file = area->vm_file;
1098 struct address_space *mapping = file->f_mapping;
1099 struct file_ra_state *ra = &file->f_ra;
1100 struct inode *inode = mapping->host;
1102 unsigned long size, pgoff, endoff;
1103 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1105 pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1106 endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1109 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1111 goto outside_data_content;
1113 /* If we don't want any read-ahead, don't bother */
1114 if (VM_RandomReadHint(area))
1115 goto no_cached_page;
1118 * The "size" of the file, as far as mmap is concerned, isn't bigger
1125 * The readahead code wants to be told about each and every page
1126 * so it can build and shrink its windows appropriately
1128 * For sequential accesses, we use the generic readahead logic.
1130 if (VM_SequentialReadHint(area))
1131 page_cache_readahead(mapping, ra, file, pgoff);
1134 * Do we have something in the page cache already?
1137 page = find_get_page(mapping, pgoff);
1139 unsigned long ra_pages;
1141 if (VM_SequentialReadHint(area)) {
1142 handle_ra_miss(mapping, ra, pgoff);
1143 goto no_cached_page;
1148 * Do we miss much more than hit in this file? If so,
1149 * stop bothering with read-ahead. It will only hurt.
1151 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1152 goto no_cached_page;
1155 * To keep the pgmajfault counter straight, we need to
1156 * check did_readaround, as this is an inner loop.
1158 if (!did_readaround) {
1159 majmin = VM_FAULT_MAJOR;
1160 inc_page_state(pgmajfault);
1163 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1167 if (pgoff > ra_pages / 2)
1168 start = pgoff - ra_pages / 2;
1169 do_page_cache_readahead(mapping, file, start, ra_pages);
1171 page = find_get_page(mapping, pgoff);
1173 goto no_cached_page;
1176 if (!did_readaround)
1180 * Ok, found a page in the page cache, now we need to check
1181 * that it's up-to-date.
1183 if (!PageUptodate(page))
1184 goto page_not_uptodate;
1188 * Found the page and have a reference on it.
1190 mark_page_accessed(page);
1195 outside_data_content:
1197 * An external ptracer can access pages that normally aren't
1200 if (area->vm_mm == current->mm)
1202 /* Fall through to the non-read-ahead case */
1205 * We're only likely to ever get here if MADV_RANDOM is in
1208 error = page_cache_read(file, pgoff);
1212 * The page we want has now been added to the page cache.
1213 * In the unlikely event that someone removed it in the
1214 * meantime, we'll just come back here and read it again.
1220 * An error return from page_cache_read can result if the
1221 * system is low on memory, or a problem occurs while trying
1224 if (error == -ENOMEM)
1229 if (!did_readaround) {
1230 majmin = VM_FAULT_MAJOR;
1231 inc_page_state(pgmajfault);
1235 /* Did it get unhashed while we waited for it? */
1236 if (!page->mapping) {
1238 page_cache_release(page);
1242 /* Did somebody else get it up-to-date? */
1243 if (PageUptodate(page)) {
1248 if (!mapping->a_ops->readpage(file, page)) {
1249 wait_on_page_locked(page);
1250 if (PageUptodate(page))
1255 * Umm, take care of errors if the page isn't up-to-date.
1256 * Try to re-read it _once_. We do this synchronously,
1257 * because there really aren't any performance issues here
1258 * and we need to check for errors.
1262 /* Somebody truncated the page on us? */
1263 if (!page->mapping) {
1265 page_cache_release(page);
1269 /* Somebody else successfully read it in? */
1270 if (PageUptodate(page)) {
1274 ClearPageError(page);
1275 if (!mapping->a_ops->readpage(file, page)) {
1276 wait_on_page_locked(page);
1277 if (PageUptodate(page))
1282 * Things didn't work out. Return zero to tell the
1283 * mm layer so, possibly freeing the page cache page first.
1285 page_cache_release(page);
1289 EXPORT_SYMBOL(filemap_nopage);
1291 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1294 struct address_space *mapping = file->f_mapping;
1299 * Do we have something in the page cache already?
1302 page = find_get_page(mapping, pgoff);
1306 goto no_cached_page;
1310 * Ok, found a page in the page cache, now we need to check
1311 * that it's up-to-date.
1313 if (!PageUptodate(page))
1314 goto page_not_uptodate;
1318 * Found the page and have a reference on it.
1320 mark_page_accessed(page);
1324 error = page_cache_read(file, pgoff);
1327 * The page we want has now been added to the page cache.
1328 * In the unlikely event that someone removed it in the
1329 * meantime, we'll just come back here and read it again.
1335 * An error return from page_cache_read can result if the
1336 * system is low on memory, or a problem occurs while trying
1344 /* Did it get unhashed while we waited for it? */
1345 if (!page->mapping) {
1350 /* Did somebody else get it up-to-date? */
1351 if (PageUptodate(page)) {
1356 if (!mapping->a_ops->readpage(file, page)) {
1357 wait_on_page_locked(page);
1358 if (PageUptodate(page))
1363 * Umm, take care of errors if the page isn't up-to-date.
1364 * Try to re-read it _once_. We do this synchronously,
1365 * because there really aren't any performance issues here
1366 * and we need to check for errors.
1370 /* Somebody truncated the page on us? */
1371 if (!page->mapping) {
1375 /* Somebody else successfully read it in? */
1376 if (PageUptodate(page)) {
1381 ClearPageError(page);
1382 if (!mapping->a_ops->readpage(file, page)) {
1383 wait_on_page_locked(page);
1384 if (PageUptodate(page))
1389 * Things didn't work out. Return zero to tell the
1390 * mm layer so, possibly freeing the page cache page first.
1393 page_cache_release(page);
1398 static int filemap_populate(struct vm_area_struct *vma,
1402 unsigned long pgoff,
1405 struct file *file = vma->vm_file;
1406 struct address_space *mapping = file->f_mapping;
1407 struct inode *inode = mapping->host;
1409 struct mm_struct *mm = vma->vm_mm;
1414 force_page_cache_readahead(mapping, vma->vm_file,
1415 pgoff, len >> PAGE_CACHE_SHIFT);
1418 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1419 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1422 page = filemap_getpage(file, pgoff, nonblock);
1423 if (!page && !nonblock)
1426 err = install_page(mm, vma, addr, page, prot);
1428 page_cache_release(page);
1432 err = install_file_pte(mm, vma, addr, pgoff, prot);
1446 static struct vm_operations_struct generic_file_vm_ops = {
1447 .nopage = filemap_nopage,
1448 .populate = filemap_populate,
1451 /* This is used for a general mmap of a disk file */
1453 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1455 struct address_space *mapping = file->f_mapping;
1457 if (!mapping->a_ops->readpage)
1459 file_accessed(file);
1460 vma->vm_ops = &generic_file_vm_ops;
1465 * This is for filesystems which do not implement ->writepage.
1467 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1469 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1471 return generic_file_mmap(file, vma);
1474 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1478 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1482 #endif /* CONFIG_MMU */
1484 EXPORT_SYMBOL(generic_file_mmap);
1485 EXPORT_SYMBOL(generic_file_readonly_mmap);
1487 static inline struct page *__read_cache_page(struct address_space *mapping,
1488 unsigned long index,
1489 int (*filler)(void *,struct page*),
1492 struct page *page, *cached_page = NULL;
1495 page = find_get_page(mapping, index);
1498 cached_page = page_cache_alloc_cold(mapping);
1500 return ERR_PTR(-ENOMEM);
1502 err = add_to_page_cache_lru(cached_page, mapping,
1507 /* Presumably ENOMEM for radix tree node */
1508 page_cache_release(cached_page);
1509 return ERR_PTR(err);
1513 err = filler(data, page);
1515 page_cache_release(page);
1516 page = ERR_PTR(err);
1520 page_cache_release(cached_page);
1525 * Read into the page cache. If a page already exists,
1526 * and PageUptodate() is not set, try to fill the page.
1528 struct page *read_cache_page(struct address_space *mapping,
1529 unsigned long index,
1530 int (*filler)(void *,struct page*),
1537 page = __read_cache_page(mapping, index, filler, data);
1540 mark_page_accessed(page);
1541 if (PageUptodate(page))
1545 if (!page->mapping) {
1547 page_cache_release(page);
1550 if (PageUptodate(page)) {
1554 err = filler(data, page);
1556 page_cache_release(page);
1557 page = ERR_PTR(err);
1563 EXPORT_SYMBOL(read_cache_page);
1566 * If the page was newly created, increment its refcount and add it to the
1567 * caller's lru-buffering pagevec. This function is specifically for
1568 * generic_file_write().
1570 static inline struct page *
1571 __grab_cache_page(struct address_space *mapping, unsigned long index,
1572 struct page **cached_page, struct pagevec *lru_pvec)
1577 page = find_lock_page(mapping, index);
1579 if (!*cached_page) {
1580 *cached_page = page_cache_alloc(mapping);
1584 err = add_to_page_cache(*cached_page, mapping,
1589 page = *cached_page;
1590 page_cache_get(page);
1591 if (!pagevec_add(lru_pvec, page))
1592 __pagevec_lru_add(lru_pvec);
1593 *cached_page = NULL;
1600 * The logic we want is
1602 * if suid or (sgid and xgrp)
1605 int remove_suid(struct dentry *dentry)
1607 mode_t mode = dentry->d_inode->i_mode;
1611 /* suid always must be killed */
1612 if (unlikely(mode & S_ISUID))
1613 kill = ATTR_KILL_SUID;
1616 * sgid without any exec bits is just a mandatory locking mark; leave
1617 * it alone. If some exec bits are set, it's a real sgid; kill it.
1619 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1620 kill |= ATTR_KILL_SGID;
1622 if (unlikely(kill && !capable(CAP_FSETID))) {
1623 struct iattr newattrs;
1625 newattrs.ia_valid = ATTR_FORCE | kill;
1626 result = notify_change(dentry, &newattrs);
1630 EXPORT_SYMBOL(remove_suid);
1633 * Copy as much as we can into the page and return the number of bytes which
1634 * were sucessfully copied. If a fault is encountered then clear the page
1635 * out to (offset+bytes) and return the number of bytes which were copied.
1637 static inline size_t
1638 filemap_copy_from_user(struct page *page, unsigned long offset,
1639 const char __user *buf, unsigned bytes)
1644 kaddr = kmap_atomic(page, KM_USER0);
1645 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1646 kunmap_atomic(kaddr, KM_USER0);
1649 /* Do it the slow way */
1651 left = __copy_from_user(kaddr + offset, buf, bytes);
1654 return bytes - left;
1658 __filemap_copy_from_user_iovec(char *vaddr,
1659 const struct iovec *iov, size_t base, size_t bytes)
1661 size_t copied = 0, left = 0;
1664 char __user *buf = iov->iov_base + base;
1665 int copy = min(bytes, iov->iov_len - base);
1668 left = __copy_from_user_inatomic(vaddr, buf, copy);
1674 if (unlikely(left)) {
1675 /* zero the rest of the target like __copy_from_user */
1677 memset(vaddr, 0, bytes);
1681 return copied - left;
1685 * This has the same sideeffects and return value as filemap_copy_from_user().
1686 * The difference is that on a fault we need to memset the remainder of the
1687 * page (out to offset+bytes), to emulate filemap_copy_from_user()'s
1688 * single-segment behaviour.
1690 static inline size_t
1691 filemap_copy_from_user_iovec(struct page *page, unsigned long offset,
1692 const struct iovec *iov, size_t base, size_t bytes)
1697 kaddr = kmap_atomic(page, KM_USER0);
1698 copied = __filemap_copy_from_user_iovec(kaddr + offset, iov,
1700 kunmap_atomic(kaddr, KM_USER0);
1701 if (copied != bytes) {
1703 copied = __filemap_copy_from_user_iovec(kaddr + offset, iov,
1711 filemap_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1713 const struct iovec *iov = *iovp;
1714 size_t base = *basep;
1717 int copy = min(bytes, iov->iov_len - base);
1721 if (iov->iov_len == base) {
1731 * Performs necessary checks before doing a write
1733 * Can adjust writing position aor amount of bytes to write.
1734 * Returns appropriate error code that caller should return or
1735 * zero in case that write should be allowed.
1737 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1739 struct inode *inode = file->f_mapping->host;
1740 unsigned long limit = current->rlim[RLIMIT_FSIZE].rlim_cur;
1742 if (unlikely(*pos < 0))
1745 if (unlikely(file->f_error)) {
1746 int err = file->f_error;
1752 /* FIXME: this is for backwards compatibility with 2.4 */
1753 if (file->f_flags & O_APPEND)
1754 *pos = i_size_read(inode);
1756 if (limit != RLIM_INFINITY) {
1757 if (*pos >= limit) {
1758 send_sig(SIGXFSZ, current, 0);
1761 if (*count > limit - (typeof(limit))*pos) {
1762 *count = limit - (typeof(limit))*pos;
1770 if (unlikely(*pos + *count > MAX_NON_LFS &&
1771 !(file->f_flags & O_LARGEFILE))) {
1772 if (*pos >= MAX_NON_LFS) {
1773 send_sig(SIGXFSZ, current, 0);
1776 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1777 *count = MAX_NON_LFS - (unsigned long)*pos;
1782 * Are we about to exceed the fs block limit ?
1784 * If we have written data it becomes a short write. If we have
1785 * exceeded without writing data we send a signal and return EFBIG.
1786 * Linus frestrict idea will clean these up nicely..
1788 if (likely(!isblk)) {
1789 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1790 if (*count || *pos > inode->i_sb->s_maxbytes) {
1791 send_sig(SIGXFSZ, current, 0);
1794 /* zero-length writes at ->s_maxbytes are OK */
1797 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1798 *count = inode->i_sb->s_maxbytes - *pos;
1801 if (bdev_read_only(I_BDEV(inode)))
1803 isize = i_size_read(inode);
1804 if (*pos >= isize) {
1805 if (*count || *pos > isize)
1809 if (*pos + *count > isize)
1810 *count = isize - *pos;
1815 EXPORT_SYMBOL(generic_write_checks);
1818 * Write to a file through the page cache.
1819 * Called under i_sem for S_ISREG files.
1821 * We put everything into the page cache prior to writing it. This is not a
1822 * problem when writing full pages. With partial pages, however, we first have
1823 * to read the data into the cache, then dirty the page, and finally schedule
1824 * it for writing by marking it dirty.
1828 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
1829 unsigned long nr_segs, loff_t *ppos)
1831 struct file *file = iocb->ki_filp;
1832 struct address_space * mapping = file->f_mapping;
1833 struct address_space_operations *a_ops = mapping->a_ops;
1834 size_t ocount; /* original count */
1835 size_t count; /* after file limit checks */
1836 struct inode *inode = mapping->host;
1840 struct page *cached_page = NULL;
1841 const int isblk = S_ISBLK(inode->i_mode);
1845 struct pagevec lru_pvec;
1846 const struct iovec *cur_iov = iov; /* current iovec */
1847 size_t iov_base = 0; /* offset in the current iovec */
1852 for (seg = 0; seg < nr_segs; seg++) {
1853 const struct iovec *iv = &iov[seg];
1856 * If any segment has a negative length, or the cumulative
1857 * length ever wraps negative then return -EINVAL.
1859 ocount += iv->iov_len;
1860 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
1862 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1867 ocount -= iv->iov_len; /* This segment is no good */
1873 pagevec_init(&lru_pvec, 0);
1875 /* We can write back this queue in page reclaim */
1876 current->backing_dev_info = mapping->backing_dev_info;
1879 err = generic_write_checks(file, &pos, &count, isblk);
1886 err = remove_suid(file->f_dentry);
1890 inode_update_time(inode, file->f_vfsmnt, 1);
1892 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1893 if (unlikely(file->f_flags & O_DIRECT)) {
1894 if (count != ocount)
1895 nr_segs = iov_shorten((struct iovec *)iov,
1897 written = generic_file_direct_IO(WRITE, iocb,
1900 loff_t end = pos + written;
1901 if (end > i_size_read(inode) && !isblk) {
1902 i_size_write(inode, end);
1903 mark_inode_dirty(inode);
1908 * Sync the fs metadata but not the minor inode changes and
1909 * of course not the data as we did direct DMA for the IO.
1910 * i_sem is held, which protects generic_osync_inode() from
1913 if (written >= 0 && file->f_flags & O_SYNC)
1914 status = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1915 if (written == count && !is_sync_kiocb(iocb))
1916 written = -EIOCBQUEUED;
1917 if (written < 0 || written == count)
1920 * direct-io write to a hole: fall through to buffered I/O
1921 * for completing the rest of the request.
1927 buf = iov->iov_base + written; /* handle partial DIO write */
1929 unsigned long index;
1930 unsigned long offset;
1933 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1934 index = pos >> PAGE_CACHE_SHIFT;
1935 bytes = PAGE_CACHE_SIZE - offset;
1940 * Bring in the user page that we will copy from _first_.
1941 * Otherwise there's a nasty deadlock on copying from the
1942 * same page as we're writing to, without it being marked
1945 fault_in_pages_readable(buf, bytes);
1947 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
1953 status = a_ops->prepare_write(file, page, offset, offset+bytes);
1954 if (unlikely(status)) {
1955 loff_t isize = i_size_read(inode);
1957 * prepare_write() may have instantiated a few blocks
1958 * outside i_size. Trim these off again.
1961 page_cache_release(page);
1962 if (pos + bytes > isize)
1963 vmtruncate(inode, isize);
1966 if (likely(nr_segs == 1))
1967 copied = filemap_copy_from_user(page, offset,
1970 copied = filemap_copy_from_user_iovec(page, offset,
1971 cur_iov, iov_base, bytes);
1972 flush_dcache_page(page);
1973 status = a_ops->commit_write(file, page, offset, offset+bytes);
1974 if (likely(copied > 0)) {
1983 if (unlikely(nr_segs > 1))
1984 filemap_set_next_iovec(&cur_iov,
1988 if (unlikely(copied != bytes))
1992 mark_page_accessed(page);
1993 page_cache_release(page);
1996 balance_dirty_pages_ratelimited(mapping);
2002 page_cache_release(cached_page);
2005 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2008 if ((file->f_flags & O_SYNC) || IS_SYNC(inode))
2009 status = generic_osync_inode(inode, mapping,
2010 OSYNC_METADATA|OSYNC_DATA);
2014 * If we get here for O_DIRECT writes then we must have fallen through
2015 * to buffered writes (block instantiation inside i_size). So we sync
2016 * the file data here, to try to honour O_DIRECT expectations.
2018 if (unlikely(file->f_flags & O_DIRECT) && written)
2019 status = filemap_write_and_wait(mapping);
2022 err = written ? written : status;
2024 pagevec_lru_add(&lru_pvec);
2025 current->backing_dev_info = NULL;
2029 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2032 generic_file_write_nolock(struct file *file, const struct iovec *iov,
2033 unsigned long nr_segs, loff_t *ppos)
2038 init_sync_kiocb(&kiocb, file);
2039 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2040 if (-EIOCBQUEUED == ret)
2041 ret = wait_on_sync_kiocb(&kiocb);
2045 EXPORT_SYMBOL(generic_file_write_nolock);
2047 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2048 size_t count, loff_t pos)
2050 struct file *file = iocb->ki_filp;
2051 struct inode *inode = file->f_mapping->host;
2053 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
2055 BUG_ON(iocb->ki_pos != pos);
2057 down(&inode->i_sem);
2058 err = generic_file_aio_write_nolock(iocb, &local_iov, 1,
2065 EXPORT_SYMBOL(generic_file_aio_write);
2067 ssize_t generic_file_write(struct file *file, const char __user *buf,
2068 size_t count, loff_t *ppos)
2070 struct inode *inode = file->f_mapping->host;
2072 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
2074 down(&inode->i_sem);
2075 err = generic_file_write_nolock(file, &local_iov, 1, ppos);
2081 EXPORT_SYMBOL(generic_file_write);
2083 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2084 unsigned long nr_segs, loff_t *ppos)
2089 init_sync_kiocb(&kiocb, filp);
2090 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2091 if (-EIOCBQUEUED == ret)
2092 ret = wait_on_sync_kiocb(&kiocb);
2096 EXPORT_SYMBOL(generic_file_readv);
2098 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2099 unsigned long nr_segs, loff_t * ppos)
2101 struct inode *inode = file->f_mapping->host;
2104 down(&inode->i_sem);
2105 ret = generic_file_write_nolock(file, iov, nr_segs, ppos);
2110 EXPORT_SYMBOL(generic_file_writev);
2113 * Called under i_sem for writes to S_ISREG files
2116 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2117 loff_t offset, unsigned long nr_segs)
2119 struct file *file = iocb->ki_filp;
2120 struct address_space *mapping = file->f_mapping;
2123 retval = filemap_write_and_wait(mapping);
2125 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2127 if (rw == WRITE && mapping->nrpages)
2128 invalidate_inode_pages2(mapping);
2133 EXPORT_SYMBOL_GPL(generic_file_direct_IO);