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_shared_sem (vmtruncate)
59 * ->private_lock (__free_pte->__set_page_dirty_buffers)
61 * ->swap_device_lock (exclusive_swap_page, others)
62 * ->mapping->tree_lock
65 * ->i_shared_sem (truncate->unmap_mapping_range)
68 * ->i_shared_sem (various places)
71 * ->lock_page (access_process_vm)
77 * ->i_alloc_sem (various)
80 * ->sb_lock (fs/fs-writeback.c)
81 * ->mapping->tree_lock (__sync_single_inode)
84 * ->swap_device_lock (try_to_unmap_one)
85 * ->private_lock (try_to_unmap_one)
86 * ->tree_lock (try_to_unmap_one)
87 * ->zone.lru_lock (follow_page->mark_page_accessed)
90 * ->dcache_lock (proc_pid_lookup)
94 * Remove a page from the page cache and free it. Caller has to make
95 * sure the page is locked and that nobody else uses it - or that usage
96 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
98 void __remove_from_page_cache(struct page *page)
100 struct address_space *mapping = page->mapping;
102 radix_tree_delete(&mapping->page_tree, page->index);
103 page->mapping = NULL;
108 void remove_from_page_cache(struct page *page)
110 struct address_space *mapping = page->mapping;
112 if (unlikely(!PageLocked(page)))
115 spin_lock_irq(&mapping->tree_lock);
116 __remove_from_page_cache(page);
117 spin_unlock_irq(&mapping->tree_lock);
120 static inline int sync_page(struct page *page)
122 struct address_space *mapping;
125 mapping = page_mapping(page);
127 if (mapping->a_ops && mapping->a_ops->sync_page)
128 return mapping->a_ops->sync_page(page);
129 } else if (PageSwapCache(page)) {
130 swap_unplug_io_fn(page);
136 * filemap_fdatawrite - start writeback against all of a mapping's dirty pages
137 * @mapping: address space structure to write
139 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
140 * opposed to a regular memory * cleansing writeback. The difference between
141 * these two operations is that if a dirty page/buffer is encountered, it must
142 * be waited upon, and not just skipped over.
144 static int __filemap_fdatawrite(struct address_space *mapping, int sync_mode)
147 struct writeback_control wbc = {
148 .sync_mode = sync_mode,
149 .nr_to_write = mapping->nrpages * 2,
152 if (mapping->backing_dev_info->memory_backed)
155 ret = do_writepages(mapping, &wbc);
159 int filemap_fdatawrite(struct address_space *mapping)
161 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
163 EXPORT_SYMBOL(filemap_fdatawrite);
166 * This is a mostly non-blocking flush. Not suitable for data-integrity
167 * purposes - I/O may not be started against all dirty pages.
169 int filemap_flush(struct address_space *mapping)
171 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
173 EXPORT_SYMBOL(filemap_flush);
176 * Wait for writeback to complete against pages indexed by start->end
179 static int wait_on_page_writeback_range(struct address_space *mapping,
180 pgoff_t start, pgoff_t end)
190 pagevec_init(&pvec, 0);
192 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
193 PAGECACHE_TAG_WRITEBACK,
194 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
197 for (i = 0; i < nr_pages; i++) {
198 struct page *page = pvec.pages[i];
200 wait_on_page_writeback(page);
204 pagevec_release(&pvec);
208 /* Check for outstanding write errors */
209 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
211 if (test_and_clear_bit(AS_EIO, &mapping->flags))
218 * filemap_fdatawait - walk the list of under-writeback pages of the given
219 * address space and wait for all of them.
221 * @mapping: address space structure to wait for
223 int filemap_fdatawait(struct address_space *mapping)
225 return wait_on_page_writeback_range(mapping, 0, -1);
228 EXPORT_SYMBOL(filemap_fdatawait);
230 int filemap_write_and_wait(struct address_space *mapping)
234 if (mapping->nrpages) {
235 retval = filemap_fdatawrite(mapping);
237 retval = filemap_fdatawait(mapping);
243 * This function is used to add newly allocated pagecache pages:
244 * the page is new, so we can just run SetPageLocked() against it.
245 * The other page state flags were set by rmqueue().
247 * This function does not add the page to the LRU. The caller must do that.
249 int add_to_page_cache(struct page *page, struct address_space *mapping,
250 pgoff_t offset, int gfp_mask)
252 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
255 page_cache_get(page);
256 spin_lock_irq(&mapping->tree_lock);
257 error = radix_tree_insert(&mapping->page_tree, offset, page);
260 page->mapping = mapping;
261 page->index = offset;
265 page_cache_release(page);
267 spin_unlock_irq(&mapping->tree_lock);
268 radix_tree_preload_end();
273 EXPORT_SYMBOL(add_to_page_cache);
275 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
276 pgoff_t offset, int gfp_mask)
278 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
285 * In order to wait for pages to become available there must be
286 * waitqueues associated with pages. By using a hash table of
287 * waitqueues where the bucket discipline is to maintain all
288 * waiters on the same queue and wake all when any of the pages
289 * become available, and for the woken contexts to check to be
290 * sure the appropriate page became available, this saves space
291 * at a cost of "thundering herd" phenomena during rare hash
294 static wait_queue_head_t *page_waitqueue(struct page *page)
296 const struct zone *zone = page_zone(page);
298 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
301 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
303 wait_queue_head_t *waitqueue = page_waitqueue(page);
307 prepare_to_wait(waitqueue, &wait, TASK_UNINTERRUPTIBLE);
308 if (test_bit(bit_nr, &page->flags)) {
312 } while (test_bit(bit_nr, &page->flags));
313 finish_wait(waitqueue, &wait);
316 EXPORT_SYMBOL(wait_on_page_bit);
319 * unlock_page() - unlock a locked page
323 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
324 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
325 * mechananism between PageLocked pages and PageWriteback pages is shared.
326 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
328 * The first mb is necessary to safely close the critical section opened by the
329 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
330 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
331 * parallel wait_on_page_locked()).
333 void fastcall unlock_page(struct page *page)
335 wait_queue_head_t *waitqueue = page_waitqueue(page);
336 smp_mb__before_clear_bit();
337 if (!TestClearPageLocked(page))
339 smp_mb__after_clear_bit();
340 if (waitqueue_active(waitqueue))
341 wake_up_all(waitqueue);
344 EXPORT_SYMBOL(unlock_page);
345 EXPORT_SYMBOL(lock_page);
348 * End writeback against a page.
350 void end_page_writeback(struct page *page)
352 wait_queue_head_t *waitqueue = page_waitqueue(page);
354 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
355 if (!test_clear_page_writeback(page))
357 smp_mb__after_clear_bit();
359 if (waitqueue_active(waitqueue))
360 wake_up_all(waitqueue);
363 EXPORT_SYMBOL(end_page_writeback);
366 * Get a lock on the page, assuming we need to sleep to get it.
368 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
369 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
370 * chances are that on the second loop, the block layer's plug list is empty,
371 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
373 void fastcall __lock_page(struct page *page)
375 wait_queue_head_t *wqh = page_waitqueue(page);
378 while (TestSetPageLocked(page)) {
379 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
380 if (PageLocked(page)) {
385 finish_wait(wqh, &wait);
388 EXPORT_SYMBOL(__lock_page);
391 * a rather lightweight function, finding and getting a reference to a
392 * hashed page atomically.
394 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
399 * We scan the hash list read-only. Addition to and removal from
400 * the hash-list needs a held write-lock.
402 spin_lock_irq(&mapping->tree_lock);
403 page = radix_tree_lookup(&mapping->page_tree, offset);
405 page_cache_get(page);
406 spin_unlock_irq(&mapping->tree_lock);
410 EXPORT_SYMBOL(find_get_page);
413 * Same as above, but trylock it instead of incrementing the count.
415 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
419 spin_lock_irq(&mapping->tree_lock);
420 page = radix_tree_lookup(&mapping->page_tree, offset);
421 if (page && TestSetPageLocked(page))
423 spin_unlock_irq(&mapping->tree_lock);
427 EXPORT_SYMBOL(find_trylock_page);
430 * find_lock_page - locate, pin and lock a pagecache page
432 * @mapping - the address_space to search
433 * @offset - the page index
435 * Locates the desired pagecache page, locks it, increments its reference
436 * count and returns its address.
438 * Returns zero if the page was not present. find_lock_page() may sleep.
440 struct page *find_lock_page(struct address_space *mapping,
441 unsigned long offset)
445 spin_lock_irq(&mapping->tree_lock);
447 page = radix_tree_lookup(&mapping->page_tree, offset);
449 page_cache_get(page);
450 if (TestSetPageLocked(page)) {
451 spin_unlock_irq(&mapping->tree_lock);
453 spin_lock_irq(&mapping->tree_lock);
455 /* Has the page been truncated while we slept? */
456 if (page->mapping != mapping || page->index != offset) {
458 page_cache_release(page);
463 spin_unlock_irq(&mapping->tree_lock);
467 EXPORT_SYMBOL(find_lock_page);
470 * find_or_create_page - locate or add a pagecache page
472 * @mapping - the page's address_space
473 * @index - the page's index into the mapping
474 * @gfp_mask - page allocation mode
476 * Locates a page in the pagecache. If the page is not present, a new page
477 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
478 * LRU list. The returned page is locked and has its reference count
481 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
484 * find_or_create_page() returns the desired page's address, or zero on
487 struct page *find_or_create_page(struct address_space *mapping,
488 unsigned long index, unsigned int gfp_mask)
490 struct page *page, *cached_page = NULL;
493 page = find_lock_page(mapping, index);
496 cached_page = alloc_page(gfp_mask);
500 err = add_to_page_cache_lru(cached_page, mapping,
505 } else if (err == -EEXIST)
509 page_cache_release(cached_page);
513 EXPORT_SYMBOL(find_or_create_page);
516 * find_get_pages - gang pagecache lookup
517 * @mapping: The address_space to search
518 * @start: The starting page index
519 * @nr_pages: The maximum number of pages
520 * @pages: Where the resulting pages are placed
522 * find_get_pages() will search for and return a group of up to
523 * @nr_pages pages in the mapping. The pages are placed at @pages.
524 * find_get_pages() takes a reference against the returned pages.
526 * The search returns a group of mapping-contiguous pages with ascending
527 * indexes. There may be holes in the indices due to not-present pages.
529 * find_get_pages() returns the number of pages which were found.
531 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
532 unsigned int nr_pages, struct page **pages)
537 spin_lock_irq(&mapping->tree_lock);
538 ret = radix_tree_gang_lookup(&mapping->page_tree,
539 (void **)pages, start, nr_pages);
540 for (i = 0; i < ret; i++)
541 page_cache_get(pages[i]);
542 spin_unlock_irq(&mapping->tree_lock);
547 * Like find_get_pages, except we only return pages which are tagged with
548 * `tag'. We update *index to index the next page for the traversal.
550 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
551 int tag, unsigned int nr_pages, struct page **pages)
556 spin_lock_irq(&mapping->tree_lock);
557 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
558 (void **)pages, *index, nr_pages, tag);
559 for (i = 0; i < ret; i++)
560 page_cache_get(pages[i]);
562 *index = pages[ret - 1]->index + 1;
563 spin_unlock_irq(&mapping->tree_lock);
568 * Same as grab_cache_page, but do not wait if the page is unavailable.
569 * This is intended for speculative data generators, where the data can
570 * be regenerated if the page couldn't be grabbed. This routine should
571 * be safe to call while holding the lock for another page.
573 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
574 * and deadlock against the caller's locked page.
577 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
579 struct page *page = find_get_page(mapping, index);
583 if (!TestSetPageLocked(page))
585 page_cache_release(page);
588 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
589 page = alloc_pages(gfp_mask, 0);
590 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
591 page_cache_release(page);
597 EXPORT_SYMBOL(grab_cache_page_nowait);
600 * This is a generic file read routine, and uses the
601 * mapping->a_ops->readpage() function for the actual low-level
604 * This is really ugly. But the goto's actually try to clarify some
605 * of the logic when it comes to error handling etc.
606 * - note the struct file * is only passed for the use of readpage
608 void do_generic_mapping_read(struct address_space *mapping,
609 struct file_ra_state *ra,
612 read_descriptor_t * desc,
615 struct inode *inode = mapping->host;
616 unsigned long index, offset;
617 struct page *cached_page;
621 index = *ppos >> PAGE_CACHE_SHIFT;
622 offset = *ppos & ~PAGE_CACHE_MASK;
626 unsigned long end_index, nr, ret;
627 loff_t isize = i_size_read(inode);
629 end_index = isize >> PAGE_CACHE_SHIFT;
631 if (index > end_index)
633 nr = PAGE_CACHE_SIZE;
634 if (index == end_index) {
635 nr = isize & ~PAGE_CACHE_MASK;
641 page_cache_readahead(mapping, ra, filp, index);
645 page = find_get_page(mapping, index);
646 if (unlikely(page == NULL)) {
647 handle_ra_miss(mapping, ra, index);
650 if (!PageUptodate(page))
651 goto page_not_up_to_date;
653 /* If users can be writing to this page using arbitrary
654 * virtual addresses, take care about potential aliasing
655 * before reading the page on the kernel side.
657 if (mapping_writably_mapped(mapping))
658 flush_dcache_page(page);
661 * Mark the page accessed if we read the beginning.
664 mark_page_accessed(page);
667 * Ok, we have the page, and it's up-to-date, so
668 * now we can copy it to user space...
670 * The actor routine returns how many bytes were actually used..
671 * NOTE! This may not be the same as how much of a user buffer
672 * we filled up (we may be padding etc), so we can only update
673 * "pos" here (the actor routine has to update the user buffer
674 * pointers and the remaining count).
676 ret = actor(desc, page, offset, nr);
678 index += offset >> PAGE_CACHE_SHIFT;
679 offset &= ~PAGE_CACHE_MASK;
681 page_cache_release(page);
682 if (ret == nr && desc->count)
687 if (PageUptodate(page))
690 /* Get exclusive access to the page ... */
693 /* Did it get unhashed before we got the lock? */
694 if (!page->mapping) {
696 page_cache_release(page);
700 /* Did somebody else fill it already? */
701 if (PageUptodate(page)) {
707 /* ... and start the actual read. The read will unlock the page. */
708 error = mapping->a_ops->readpage(filp, page);
711 if (PageUptodate(page))
713 wait_on_page_locked(page);
714 if (PageUptodate(page))
719 /* UHHUH! A synchronous read error occurred. Report it */
721 page_cache_release(page);
726 * Ok, it wasn't cached, so we need to create a new
730 cached_page = page_cache_alloc_cold(mapping);
732 desc->error = -ENOMEM;
736 error = add_to_page_cache_lru(cached_page, mapping,
739 if (error == -EEXIST)
749 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
751 page_cache_release(cached_page);
755 EXPORT_SYMBOL(do_generic_mapping_read);
757 int file_read_actor(read_descriptor_t *desc, struct page *page,
758 unsigned long offset, unsigned long size)
761 unsigned long left, count = desc->count;
767 * Faults on the destination of a read are common, so do it before
770 if (!fault_in_pages_writeable(desc->buf, size)) {
771 kaddr = kmap_atomic(page, KM_USER0);
772 left = __copy_to_user(desc->buf, kaddr + offset, size);
773 kunmap_atomic(kaddr, KM_USER0);
778 /* Do it the slow way */
780 left = __copy_to_user(desc->buf, kaddr + offset, size);
785 desc->error = -EFAULT;
788 desc->count = count - size;
789 desc->written += size;
795 * This is the "read()" routine for all filesystems
796 * that can use the page cache directly.
799 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
800 unsigned long nr_segs, loff_t *ppos)
802 struct file *filp = iocb->ki_filp;
808 for (seg = 0; seg < nr_segs; seg++) {
809 const struct iovec *iv = &iov[seg];
812 * If any segment has a negative length, or the cumulative
813 * length ever wraps negative then return -EINVAL.
815 count += iv->iov_len;
816 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
818 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
823 count -= iv->iov_len; /* This segment is no good */
827 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
828 if (filp->f_flags & O_DIRECT) {
829 loff_t pos = *ppos, size;
830 struct address_space *mapping;
833 mapping = filp->f_mapping;
834 inode = mapping->host;
837 goto out; /* skip atime */
838 size = i_size_read(inode);
840 retval = generic_file_direct_IO(READ, iocb,
842 if (retval >= 0 && !is_sync_kiocb(iocb))
843 retval = -EIOCBQUEUED;
845 *ppos = pos + retval;
853 for (seg = 0; seg < nr_segs; seg++) {
854 read_descriptor_t desc;
857 desc.buf = iov[seg].iov_base;
858 desc.count = iov[seg].iov_len;
862 do_generic_file_read(filp,ppos,&desc,file_read_actor);
863 retval += desc.written;
874 EXPORT_SYMBOL(__generic_file_aio_read);
877 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
879 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
881 BUG_ON(iocb->ki_pos != pos);
882 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
885 EXPORT_SYMBOL(generic_file_aio_read);
888 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
890 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
894 init_sync_kiocb(&kiocb, filp);
895 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
896 if (-EIOCBQUEUED == ret)
897 ret = wait_on_sync_kiocb(&kiocb);
901 EXPORT_SYMBOL(generic_file_read);
903 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
906 unsigned long count = desc->count;
907 struct file *file = (struct file *) desc->buf;
912 written = file->f_op->sendpage(file, page, offset,
913 size, &file->f_pos, size<count);
915 desc->error = written;
918 desc->count = count - written;
919 desc->written += written;
923 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
924 size_t count, read_actor_t actor, void __user *target)
926 read_descriptor_t desc;
936 do_generic_file_read(in_file, ppos, &desc, actor);
942 EXPORT_SYMBOL(generic_file_sendfile);
945 do_readahead(struct address_space *mapping, struct file *filp,
946 unsigned long index, unsigned long nr)
948 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
951 force_page_cache_readahead(mapping, filp, index,
952 max_sane_readahead(nr));
956 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
964 if (file->f_mode & FMODE_READ) {
965 struct address_space *mapping = file->f_mapping;
966 unsigned long start = offset >> PAGE_CACHE_SHIFT;
967 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
968 unsigned long len = end - start + 1;
969 ret = do_readahead(mapping, file, start, len);
978 * This adds the requested page to the page cache if it isn't already there,
979 * and schedules an I/O to read in its contents from disk.
981 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
982 static int fastcall page_cache_read(struct file * file, unsigned long offset)
984 struct address_space *mapping = file->f_mapping;
988 page = page_cache_alloc_cold(mapping);
992 error = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
994 error = mapping->a_ops->readpage(file, page);
995 page_cache_release(page);
1000 * We arrive here in the unlikely event that someone
1001 * raced with us and added our page to the cache first
1002 * or we are out of memory for radix-tree nodes.
1004 page_cache_release(page);
1005 return error == -EEXIST ? 0 : error;
1008 #define MMAP_LOTSAMISS (100)
1011 * filemap_nopage() is invoked via the vma operations vector for a
1012 * mapped memory region to read in file data during a page fault.
1014 * The goto's are kind of ugly, but this streamlines the normal case of having
1015 * it in the page cache, and handles the special cases reasonably without
1016 * having a lot of duplicated code.
1018 struct page * filemap_nopage(struct vm_area_struct * area, unsigned long address, int *type)
1021 struct file *file = area->vm_file;
1022 struct address_space *mapping = file->f_mapping;
1023 struct file_ra_state *ra = &file->f_ra;
1024 struct inode *inode = mapping->host;
1026 unsigned long size, pgoff, endoff;
1027 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1029 pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1030 endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1033 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1035 goto outside_data_content;
1037 /* If we don't want any read-ahead, don't bother */
1038 if (VM_RandomReadHint(area))
1039 goto no_cached_page;
1042 * The "size" of the file, as far as mmap is concerned, isn't bigger
1049 * The readahead code wants to be told about each and every page
1050 * so it can build and shrink its windows appropriately
1052 * For sequential accesses, we use the generic readahead logic.
1054 if (VM_SequentialReadHint(area))
1055 page_cache_readahead(mapping, ra, file, pgoff);
1058 * Do we have something in the page cache already?
1061 page = find_get_page(mapping, pgoff);
1063 unsigned long ra_pages;
1065 if (VM_SequentialReadHint(area)) {
1066 handle_ra_miss(mapping, ra, pgoff);
1067 goto no_cached_page;
1072 * Do we miss much more than hit in this file? If so,
1073 * stop bothering with read-ahead. It will only hurt.
1075 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1076 goto no_cached_page;
1079 * To keep the pgmajfault counter straight, we need to
1080 * check did_readaround, as this is an inner loop.
1082 if (!did_readaround) {
1083 majmin = VM_FAULT_MAJOR;
1084 inc_page_state(pgmajfault);
1087 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1091 start = pgoff - ra_pages / 2;
1094 do_page_cache_readahead(mapping, file, pgoff, ra_pages);
1096 page = find_get_page(mapping, pgoff);
1098 goto no_cached_page;
1101 if (!did_readaround)
1105 * Ok, found a page in the page cache, now we need to check
1106 * that it's up-to-date.
1108 if (!PageUptodate(page))
1109 goto page_not_uptodate;
1113 * Found the page and have a reference on it.
1115 mark_page_accessed(page);
1120 outside_data_content:
1122 * An external ptracer can access pages that normally aren't
1125 if (area->vm_mm == current->mm)
1127 /* Fall through to the non-read-ahead case */
1130 * We're only likely to ever get here if MADV_RANDOM is in
1133 error = page_cache_read(file, pgoff);
1136 * The page we want has now been added to the page cache.
1137 * In the unlikely event that someone removed it in the
1138 * meantime, we'll just come back here and read it again.
1144 * An error return from page_cache_read can result if the
1145 * system is low on memory, or a problem occurs while trying
1148 if (error == -ENOMEM)
1153 if (!did_readaround) {
1154 majmin = VM_FAULT_MAJOR;
1155 inc_page_state(pgmajfault);
1159 /* Did it get unhashed while we waited for it? */
1160 if (!page->mapping) {
1162 page_cache_release(page);
1166 /* Did somebody else get it up-to-date? */
1167 if (PageUptodate(page)) {
1172 if (!mapping->a_ops->readpage(file, page)) {
1173 wait_on_page_locked(page);
1174 if (PageUptodate(page))
1179 * Umm, take care of errors if the page isn't up-to-date.
1180 * Try to re-read it _once_. We do this synchronously,
1181 * because there really aren't any performance issues here
1182 * and we need to check for errors.
1186 /* Somebody truncated the page on us? */
1187 if (!page->mapping) {
1189 page_cache_release(page);
1193 /* Somebody else successfully read it in? */
1194 if (PageUptodate(page)) {
1198 ClearPageError(page);
1199 if (!mapping->a_ops->readpage(file, page)) {
1200 wait_on_page_locked(page);
1201 if (PageUptodate(page))
1206 * Things didn't work out. Return zero to tell the
1207 * mm layer so, possibly freeing the page cache page first.
1209 page_cache_release(page);
1213 EXPORT_SYMBOL(filemap_nopage);
1215 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1218 struct address_space *mapping = file->f_mapping;
1223 * Do we have something in the page cache already?
1226 page = find_get_page(mapping, pgoff);
1230 goto no_cached_page;
1234 * Ok, found a page in the page cache, now we need to check
1235 * that it's up-to-date.
1237 if (!PageUptodate(page))
1238 goto page_not_uptodate;
1242 * Found the page and have a reference on it.
1244 mark_page_accessed(page);
1248 error = page_cache_read(file, pgoff);
1251 * The page we want has now been added to the page cache.
1252 * In the unlikely event that someone removed it in the
1253 * meantime, we'll just come back here and read it again.
1259 * An error return from page_cache_read can result if the
1260 * system is low on memory, or a problem occurs while trying
1268 /* Did it get unhashed while we waited for it? */
1269 if (!page->mapping) {
1274 /* Did somebody else get it up-to-date? */
1275 if (PageUptodate(page)) {
1280 if (!mapping->a_ops->readpage(file, page)) {
1281 wait_on_page_locked(page);
1282 if (PageUptodate(page))
1287 * Umm, take care of errors if the page isn't up-to-date.
1288 * Try to re-read it _once_. We do this synchronously,
1289 * because there really aren't any performance issues here
1290 * and we need to check for errors.
1294 /* Somebody truncated the page on us? */
1295 if (!page->mapping) {
1299 /* Somebody else successfully read it in? */
1300 if (PageUptodate(page)) {
1305 ClearPageError(page);
1306 if (!mapping->a_ops->readpage(file, page)) {
1307 wait_on_page_locked(page);
1308 if (PageUptodate(page))
1313 * Things didn't work out. Return zero to tell the
1314 * mm layer so, possibly freeing the page cache page first.
1317 page_cache_release(page);
1322 static int filemap_populate(struct vm_area_struct *vma,
1326 unsigned long pgoff,
1329 struct file *file = vma->vm_file;
1330 struct address_space *mapping = file->f_mapping;
1331 struct inode *inode = mapping->host;
1333 struct mm_struct *mm = vma->vm_mm;
1338 force_page_cache_readahead(mapping, vma->vm_file,
1339 pgoff, len >> PAGE_CACHE_SHIFT);
1342 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1343 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1346 page = filemap_getpage(file, pgoff, nonblock);
1347 if (!page && !nonblock)
1350 err = install_page(mm, vma, addr, page, prot);
1352 page_cache_release(page);
1357 * If a nonlinear mapping then store the file page offset
1360 if (pgoff != linear_page_index(vma, addr)) {
1361 err = install_file_pte(mm, vma, addr, pgoff, prot);
1376 static struct vm_operations_struct generic_file_vm_ops = {
1377 .nopage = filemap_nopage,
1378 .populate = filemap_populate,
1381 /* This is used for a general mmap of a disk file */
1383 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1385 struct address_space *mapping = file->f_mapping;
1387 if (!mapping->a_ops->readpage)
1389 file_accessed(file);
1390 vma->vm_ops = &generic_file_vm_ops;
1395 * This is for filesystems which do not implement ->writepage.
1397 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1399 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1401 return generic_file_mmap(file, vma);
1404 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1408 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1412 #endif /* CONFIG_MMU */
1414 EXPORT_SYMBOL(generic_file_mmap);
1415 EXPORT_SYMBOL(generic_file_readonly_mmap);
1417 static inline struct page *__read_cache_page(struct address_space *mapping,
1418 unsigned long index,
1419 int (*filler)(void *,struct page*),
1422 struct page *page, *cached_page = NULL;
1425 page = find_get_page(mapping, index);
1428 cached_page = page_cache_alloc_cold(mapping);
1430 return ERR_PTR(-ENOMEM);
1432 err = add_to_page_cache_lru(cached_page, mapping,
1437 /* Presumably ENOMEM for radix tree node */
1438 page_cache_release(cached_page);
1439 return ERR_PTR(err);
1443 err = filler(data, page);
1445 page_cache_release(page);
1446 page = ERR_PTR(err);
1450 page_cache_release(cached_page);
1455 * Read into the page cache. If a page already exists,
1456 * and PageUptodate() is not set, try to fill the page.
1458 struct page *read_cache_page(struct address_space *mapping,
1459 unsigned long index,
1460 int (*filler)(void *,struct page*),
1467 page = __read_cache_page(mapping, index, filler, data);
1470 mark_page_accessed(page);
1471 if (PageUptodate(page))
1475 if (!page->mapping) {
1477 page_cache_release(page);
1480 if (PageUptodate(page)) {
1484 err = filler(data, page);
1486 page_cache_release(page);
1487 page = ERR_PTR(err);
1493 EXPORT_SYMBOL(read_cache_page);
1496 * If the page was newly created, increment its refcount and add it to the
1497 * caller's lru-buffering pagevec. This function is specifically for
1498 * generic_file_write().
1500 static inline struct page *
1501 __grab_cache_page(struct address_space *mapping, unsigned long index,
1502 struct page **cached_page, struct pagevec *lru_pvec)
1507 page = find_lock_page(mapping, index);
1509 if (!*cached_page) {
1510 *cached_page = page_cache_alloc(mapping);
1514 err = add_to_page_cache(*cached_page, mapping,
1519 page = *cached_page;
1520 page_cache_get(page);
1521 if (!pagevec_add(lru_pvec, page))
1522 __pagevec_lru_add(lru_pvec);
1523 *cached_page = NULL;
1530 * The logic we want is
1532 * if suid or (sgid and xgrp)
1535 int remove_suid(struct dentry *dentry)
1537 mode_t mode = dentry->d_inode->i_mode;
1541 /* suid always must be killed */
1542 if (unlikely(mode & S_ISUID))
1543 kill = ATTR_KILL_SUID;
1546 * sgid without any exec bits is just a mandatory locking mark; leave
1547 * it alone. If some exec bits are set, it's a real sgid; kill it.
1549 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1550 kill |= ATTR_KILL_SGID;
1552 if (unlikely(kill && !capable(CAP_FSETID))) {
1553 struct iattr newattrs;
1555 newattrs.ia_valid = ATTR_FORCE | kill;
1556 result = notify_change(dentry, &newattrs);
1560 EXPORT_SYMBOL(remove_suid);
1563 * Copy as much as we can into the page and return the number of bytes which
1564 * were sucessfully copied. If a fault is encountered then clear the page
1565 * out to (offset+bytes) and return the number of bytes which were copied.
1567 static inline size_t
1568 filemap_copy_from_user(struct page *page, unsigned long offset,
1569 const char __user *buf, unsigned bytes)
1574 kaddr = kmap_atomic(page, KM_USER0);
1575 left = __copy_from_user(kaddr + offset, buf, bytes);
1576 kunmap_atomic(kaddr, KM_USER0);
1579 /* Do it the slow way */
1581 left = __copy_from_user(kaddr + offset, buf, bytes);
1584 return bytes - left;
1588 __filemap_copy_from_user_iovec(char *vaddr,
1589 const struct iovec *iov, size_t base, size_t bytes)
1591 size_t copied = 0, left = 0;
1594 char __user *buf = iov->iov_base + base;
1595 int copy = min(bytes, iov->iov_len - base);
1598 left = __copy_from_user(vaddr, buf, copy);
1604 if (unlikely(left)) {
1605 /* zero the rest of the target like __copy_from_user */
1607 memset(vaddr, 0, bytes);
1611 return copied - left;
1615 * This has the same sideeffects and return value as filemap_copy_from_user().
1616 * The difference is that on a fault we need to memset the remainder of the
1617 * page (out to offset+bytes), to emulate filemap_copy_from_user()'s
1618 * single-segment behaviour.
1620 static inline size_t
1621 filemap_copy_from_user_iovec(struct page *page, unsigned long offset,
1622 const struct iovec *iov, size_t base, size_t bytes)
1627 kaddr = kmap_atomic(page, KM_USER0);
1628 copied = __filemap_copy_from_user_iovec(kaddr + offset, iov,
1630 kunmap_atomic(kaddr, KM_USER0);
1631 if (copied != bytes) {
1633 copied = __filemap_copy_from_user_iovec(kaddr + offset, iov,
1641 filemap_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1643 const struct iovec *iov = *iovp;
1644 size_t base = *basep;
1647 int copy = min(bytes, iov->iov_len - base);
1651 if (iov->iov_len == base) {
1661 * Performs necessary checks before doing a write
1663 * Can adjust writing position aor amount of bytes to write.
1664 * Returns appropriate error code that caller should return or
1665 * zero in case that write should be allowed.
1667 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1669 struct inode *inode = file->f_mapping->host;
1670 unsigned long limit = current->rlim[RLIMIT_FSIZE].rlim_cur;
1672 if (unlikely(*pos < 0))
1675 if (unlikely(file->f_error)) {
1676 int err = file->f_error;
1682 /* FIXME: this is for backwards compatibility with 2.4 */
1683 if (file->f_flags & O_APPEND)
1684 *pos = i_size_read(inode);
1686 if (limit != RLIM_INFINITY) {
1687 if (*pos >= limit) {
1688 send_sig(SIGXFSZ, current, 0);
1691 if (*count > limit - (typeof(limit))*pos) {
1692 *count = limit - (typeof(limit))*pos;
1700 if (unlikely(*pos + *count > MAX_NON_LFS &&
1701 !(file->f_flags & O_LARGEFILE))) {
1702 if (*pos >= MAX_NON_LFS) {
1703 send_sig(SIGXFSZ, current, 0);
1706 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1707 *count = MAX_NON_LFS - (unsigned long)*pos;
1712 * Are we about to exceed the fs block limit ?
1714 * If we have written data it becomes a short write. If we have
1715 * exceeded without writing data we send a signal and return EFBIG.
1716 * Linus frestrict idea will clean these up nicely..
1718 if (likely(!isblk)) {
1719 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1720 if (*count || *pos > inode->i_sb->s_maxbytes) {
1721 send_sig(SIGXFSZ, current, 0);
1724 /* zero-length writes at ->s_maxbytes are OK */
1727 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1728 *count = inode->i_sb->s_maxbytes - *pos;
1731 if (bdev_read_only(I_BDEV(inode)))
1733 isize = i_size_read(inode);
1734 if (*pos >= isize) {
1735 if (*count || *pos > isize)
1739 if (*pos + *count > isize)
1740 *count = isize - *pos;
1745 EXPORT_SYMBOL(generic_write_checks);
1748 * Write to a file through the page cache.
1749 * Called under i_sem for S_ISREG files.
1751 * We put everything into the page cache prior to writing it. This is not a
1752 * problem when writing full pages. With partial pages, however, we first have
1753 * to read the data into the cache, then dirty the page, and finally schedule
1754 * it for writing by marking it dirty.
1758 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
1759 unsigned long nr_segs, loff_t *ppos)
1761 struct file *file = iocb->ki_filp;
1762 struct address_space * mapping = file->f_mapping;
1763 struct address_space_operations *a_ops = mapping->a_ops;
1764 size_t ocount; /* original count */
1765 size_t count; /* after file limit checks */
1766 struct inode *inode = mapping->host;
1770 struct page *cached_page = NULL;
1771 const int isblk = S_ISBLK(inode->i_mode);
1775 struct pagevec lru_pvec;
1776 const struct iovec *cur_iov = iov; /* current iovec */
1777 size_t iov_base = 0; /* offset in the current iovec */
1782 for (seg = 0; seg < nr_segs; seg++) {
1783 const struct iovec *iv = &iov[seg];
1786 * If any segment has a negative length, or the cumulative
1787 * length ever wraps negative then return -EINVAL.
1789 ocount += iv->iov_len;
1790 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
1792 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1797 ocount -= iv->iov_len; /* This segment is no good */
1803 pagevec_init(&lru_pvec, 0);
1805 /* We can write back this queue in page reclaim */
1806 current->backing_dev_info = mapping->backing_dev_info;
1809 err = generic_write_checks(file, &pos, &count, isblk);
1816 err = remove_suid(file->f_dentry);
1820 inode_update_time(inode, 1);
1822 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1823 if (unlikely(file->f_flags & O_DIRECT)) {
1824 if (count != ocount)
1825 nr_segs = iov_shorten((struct iovec *)iov,
1827 written = generic_file_direct_IO(WRITE, iocb,
1830 loff_t end = pos + written;
1831 if (end > i_size_read(inode) && !isblk) {
1832 i_size_write(inode, end);
1833 mark_inode_dirty(inode);
1838 * Sync the fs metadata but not the minor inode changes and
1839 * of course not the data as we did direct DMA for the IO.
1840 * i_sem is held, which protects generic_osync_inode() from
1843 if (written >= 0 && file->f_flags & O_SYNC)
1844 status = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1845 if (written == count && !is_sync_kiocb(iocb))
1846 written = -EIOCBQUEUED;
1847 if (written < 0 || written == count)
1850 * direct-io write to a hole: fall through to buffered I/O
1851 * for completing the rest of the request.
1857 buf = iov->iov_base;
1859 unsigned long index;
1860 unsigned long offset;
1863 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1864 index = pos >> PAGE_CACHE_SHIFT;
1865 bytes = PAGE_CACHE_SIZE - offset;
1870 * Bring in the user page that we will copy from _first_.
1871 * Otherwise there's a nasty deadlock on copying from the
1872 * same page as we're writing to, without it being marked
1875 fault_in_pages_readable(buf, bytes);
1877 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
1883 status = a_ops->prepare_write(file, page, offset, offset+bytes);
1884 if (unlikely(status)) {
1885 loff_t isize = i_size_read(inode);
1887 * prepare_write() may have instantiated a few blocks
1888 * outside i_size. Trim these off again.
1891 page_cache_release(page);
1892 if (pos + bytes > isize)
1893 vmtruncate(inode, isize);
1896 if (likely(nr_segs == 1))
1897 copied = filemap_copy_from_user(page, offset,
1900 copied = filemap_copy_from_user_iovec(page, offset,
1901 cur_iov, iov_base, bytes);
1902 flush_dcache_page(page);
1903 status = a_ops->commit_write(file, page, offset, offset+bytes);
1904 if (likely(copied > 0)) {
1913 if (unlikely(nr_segs > 1))
1914 filemap_set_next_iovec(&cur_iov,
1918 if (unlikely(copied != bytes))
1922 mark_page_accessed(page);
1923 page_cache_release(page);
1926 balance_dirty_pages_ratelimited(mapping);
1932 page_cache_release(cached_page);
1935 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
1938 if ((file->f_flags & O_SYNC) || IS_SYNC(inode))
1939 status = generic_osync_inode(inode, mapping,
1940 OSYNC_METADATA|OSYNC_DATA);
1944 * If we get here for O_DIRECT writes then we must have fallen through
1945 * to buffered writes (block instantiation inside i_size). So we sync
1946 * the file data here, to try to honour O_DIRECT expectations.
1948 if (unlikely(file->f_flags & O_DIRECT) && written)
1949 status = filemap_write_and_wait(mapping);
1952 err = written ? written : status;
1954 pagevec_lru_add(&lru_pvec);
1955 current->backing_dev_info = 0;
1959 EXPORT_SYMBOL(generic_file_aio_write_nolock);
1962 generic_file_write_nolock(struct file *file, const struct iovec *iov,
1963 unsigned long nr_segs, loff_t *ppos)
1968 init_sync_kiocb(&kiocb, file);
1969 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
1970 if (-EIOCBQUEUED == ret)
1971 ret = wait_on_sync_kiocb(&kiocb);
1975 EXPORT_SYMBOL(generic_file_write_nolock);
1977 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
1978 size_t count, loff_t pos)
1980 struct file *file = iocb->ki_filp;
1981 struct inode *inode = file->f_mapping->host;
1983 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1985 BUG_ON(iocb->ki_pos != pos);
1987 down(&inode->i_sem);
1988 err = generic_file_aio_write_nolock(iocb, &local_iov, 1,
1995 EXPORT_SYMBOL(generic_file_aio_write);
1997 ssize_t generic_file_write(struct file *file, const char __user *buf,
1998 size_t count, loff_t *ppos)
2000 struct inode *inode = file->f_mapping->host;
2002 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
2004 down(&inode->i_sem);
2005 err = generic_file_write_nolock(file, &local_iov, 1, ppos);
2011 EXPORT_SYMBOL(generic_file_write);
2013 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2014 unsigned long nr_segs, loff_t *ppos)
2019 init_sync_kiocb(&kiocb, filp);
2020 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2021 if (-EIOCBQUEUED == ret)
2022 ret = wait_on_sync_kiocb(&kiocb);
2026 EXPORT_SYMBOL(generic_file_readv);
2028 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2029 unsigned long nr_segs, loff_t * ppos)
2031 struct inode *inode = file->f_mapping->host;
2034 down(&inode->i_sem);
2035 ret = generic_file_write_nolock(file, iov, nr_segs, ppos);
2040 EXPORT_SYMBOL(generic_file_writev);
2043 * Called under i_sem for writes to S_ISREG files
2046 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2047 loff_t offset, unsigned long nr_segs)
2049 struct file *file = iocb->ki_filp;
2050 struct address_space *mapping = file->f_mapping;
2053 retval = filemap_write_and_wait(mapping);
2055 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2057 if (rw == WRITE && mapping->nrpages)
2058 invalidate_inode_pages2(mapping);
2063 EXPORT_SYMBOL_GPL(generic_file_direct_IO);