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/uaccess.h>
18 #include <linux/aio.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
22 #include <linux/swap.h>
23 #include <linux/mman.h>
24 #include <linux/pagemap.h>
25 #include <linux/file.h>
26 #include <linux/uio.h>
27 #include <linux/hash.h>
28 #include <linux/writeback.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/syscalls.h>
33 #include <linux/cpuset.h>
38 * FIXME: remove all knowledge of the buffer layer from the core VM
40 #include <linux/buffer_head.h> /* for generic_osync_inode */
45 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
46 loff_t offset, unsigned long nr_segs);
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * Shared mappings now work. 15.8.1995 Bruno.
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
63 * ->i_mmap_lock (vmtruncate)
64 * ->private_lock (__free_pte->__set_page_dirty_buffers)
65 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
69 * ->i_mmap_lock (truncate->unmap_mapping_range)
73 * ->page_table_lock or pte_lock (various, mainly in memory.c)
74 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 * ->lock_page (access_process_vm)
83 * ->i_alloc_sem (various)
86 * ->sb_lock (fs/fs-writeback.c)
87 * ->mapping->tree_lock (__sync_single_inode)
90 * ->anon_vma.lock (vma_adjust)
93 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
95 * ->page_table_lock or pte_lock
96 * ->swap_lock (try_to_unmap_one)
97 * ->private_lock (try_to_unmap_one)
98 * ->tree_lock (try_to_unmap_one)
99 * ->zone.lru_lock (follow_page->mark_page_accessed)
100 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
101 * ->private_lock (page_remove_rmap->set_page_dirty)
102 * ->tree_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (page_remove_rmap->set_page_dirty)
104 * ->inode_lock (zap_pte_range->set_page_dirty)
105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 * ->dcache_lock (proc_pid_lookup)
112 * Remove a page from the page cache and free it. Caller has to make
113 * sure the page is locked and that nobody else uses it - or that usage
114 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
116 void __remove_from_page_cache(struct page *page)
118 struct address_space *mapping = page->mapping;
120 radix_tree_delete(&mapping->page_tree, page->index);
121 page->mapping = NULL;
126 void remove_from_page_cache(struct page *page)
128 struct address_space *mapping = page->mapping;
130 BUG_ON(!PageLocked(page));
132 write_lock_irq(&mapping->tree_lock);
133 __remove_from_page_cache(page);
134 write_unlock_irq(&mapping->tree_lock);
137 static int sync_page(void *word)
139 struct address_space *mapping;
142 page = container_of((unsigned long *)word, struct page, flags);
145 * page_mapping() is being called without PG_locked held.
146 * Some knowledge of the state and use of the page is used to
147 * reduce the requirements down to a memory barrier.
148 * The danger here is of a stale page_mapping() return value
149 * indicating a struct address_space different from the one it's
150 * associated with when it is associated with one.
151 * After smp_mb(), it's either the correct page_mapping() for
152 * the page, or an old page_mapping() and the page's own
153 * page_mapping() has gone NULL.
154 * The ->sync_page() address_space operation must tolerate
155 * page_mapping() going NULL. By an amazing coincidence,
156 * this comes about because none of the users of the page
157 * in the ->sync_page() methods make essential use of the
158 * page_mapping(), merely passing the page down to the backing
159 * device's unplug functions when it's non-NULL, which in turn
160 * ignore it for all cases but swap, where only page_private(page) is
161 * of interest. When page_mapping() does go NULL, the entire
162 * call stack gracefully ignores the page and returns.
166 mapping = page_mapping(page);
167 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
168 mapping->a_ops->sync_page(page);
174 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
175 * @mapping: address space structure to write
176 * @start: offset in bytes where the range starts
177 * @end: offset in bytes where the range ends (inclusive)
178 * @sync_mode: enable synchronous operation
180 * Start writeback against all of a mapping's dirty pages that lie
181 * within the byte offsets <start, end> inclusive.
183 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
184 * opposed to a regular memory cleansing writeback. The difference between
185 * these two operations is that if a dirty page/buffer is encountered, it must
186 * be waited upon, and not just skipped over.
188 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
189 loff_t end, int sync_mode)
192 struct writeback_control wbc = {
193 .sync_mode = sync_mode,
194 .nr_to_write = mapping->nrpages * 2,
195 .range_start = start,
199 if (!mapping_cap_writeback_dirty(mapping))
202 ret = do_writepages(mapping, &wbc);
206 static inline int __filemap_fdatawrite(struct address_space *mapping,
209 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
212 int filemap_fdatawrite(struct address_space *mapping)
214 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
216 EXPORT_SYMBOL(filemap_fdatawrite);
218 static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
221 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
225 * filemap_flush - mostly a non-blocking flush
226 * @mapping: target address_space
228 * This is a mostly non-blocking flush. Not suitable for data-integrity
229 * purposes - I/O may not be started against all dirty pages.
231 int filemap_flush(struct address_space *mapping)
233 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
235 EXPORT_SYMBOL(filemap_flush);
238 * wait_on_page_writeback_range - wait for writeback to complete
239 * @mapping: target address_space
240 * @start: beginning page index
241 * @end: ending page index
243 * Wait for writeback to complete against pages indexed by start->end
246 int wait_on_page_writeback_range(struct address_space *mapping,
247 pgoff_t start, pgoff_t end)
257 pagevec_init(&pvec, 0);
259 while ((index <= end) &&
260 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
261 PAGECACHE_TAG_WRITEBACK,
262 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
265 for (i = 0; i < nr_pages; i++) {
266 struct page *page = pvec.pages[i];
268 /* until radix tree lookup accepts end_index */
269 if (page->index > end)
272 wait_on_page_writeback(page);
276 pagevec_release(&pvec);
280 /* Check for outstanding write errors */
281 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
283 if (test_and_clear_bit(AS_EIO, &mapping->flags))
290 * sync_page_range - write and wait on all pages in the passed range
291 * @inode: target inode
292 * @mapping: target address_space
293 * @pos: beginning offset in pages to write
294 * @count: number of bytes to write
296 * Write and wait upon all the pages in the passed range. This is a "data
297 * integrity" operation. It waits upon in-flight writeout before starting and
298 * waiting upon new writeout. If there was an IO error, return it.
300 * We need to re-take i_mutex during the generic_osync_inode list walk because
301 * it is otherwise livelockable.
303 int sync_page_range(struct inode *inode, struct address_space *mapping,
304 loff_t pos, loff_t count)
306 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
307 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
310 if (!mapping_cap_writeback_dirty(mapping) || !count)
312 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
314 mutex_lock(&inode->i_mutex);
315 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
316 mutex_unlock(&inode->i_mutex);
319 ret = wait_on_page_writeback_range(mapping, start, end);
322 EXPORT_SYMBOL(sync_page_range);
325 * sync_page_range_nolock
326 * @inode: target inode
327 * @mapping: target address_space
328 * @pos: beginning offset in pages to write
329 * @count: number of bytes to write
331 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
332 * as it forces O_SYNC writers to different parts of the same file
333 * to be serialised right until io completion.
335 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
336 loff_t pos, loff_t count)
338 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
339 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
342 if (!mapping_cap_writeback_dirty(mapping) || !count)
344 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
346 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
348 ret = wait_on_page_writeback_range(mapping, start, end);
351 EXPORT_SYMBOL(sync_page_range_nolock);
354 * filemap_fdatawait - wait for all under-writeback pages to complete
355 * @mapping: address space structure to wait for
357 * Walk the list of under-writeback pages of the given address space
358 * and wait for all of them.
360 int filemap_fdatawait(struct address_space *mapping)
362 loff_t i_size = i_size_read(mapping->host);
367 return wait_on_page_writeback_range(mapping, 0,
368 (i_size - 1) >> PAGE_CACHE_SHIFT);
370 EXPORT_SYMBOL(filemap_fdatawait);
372 int filemap_write_and_wait(struct address_space *mapping)
376 if (mapping->nrpages) {
377 err = filemap_fdatawrite(mapping);
379 * Even if the above returned error, the pages may be
380 * written partially (e.g. -ENOSPC), so we wait for it.
381 * But the -EIO is special case, it may indicate the worst
382 * thing (e.g. bug) happened, so we avoid waiting for it.
385 int err2 = filemap_fdatawait(mapping);
392 EXPORT_SYMBOL(filemap_write_and_wait);
395 * filemap_write_and_wait_range - write out & wait on a file range
396 * @mapping: the address_space for the pages
397 * @lstart: offset in bytes where the range starts
398 * @lend: offset in bytes where the range ends (inclusive)
400 * Write out and wait upon file offsets lstart->lend, inclusive.
402 * Note that `lend' is inclusive (describes the last byte to be written) so
403 * that this function can be used to write to the very end-of-file (end = -1).
405 int filemap_write_and_wait_range(struct address_space *mapping,
406 loff_t lstart, loff_t lend)
410 if (mapping->nrpages) {
411 err = __filemap_fdatawrite_range(mapping, lstart, lend,
413 /* See comment of filemap_write_and_wait() */
415 int err2 = wait_on_page_writeback_range(mapping,
416 lstart >> PAGE_CACHE_SHIFT,
417 lend >> PAGE_CACHE_SHIFT);
426 * add_to_page_cache - add newly allocated pagecache pages
428 * @mapping: the page's address_space
429 * @offset: page index
430 * @gfp_mask: page allocation mode
432 * This function is used to add newly allocated pagecache pages;
433 * the page is new, so we can just run SetPageLocked() against it.
434 * The other page state flags were set by rmqueue().
436 * This function does not add the page to the LRU. The caller must do that.
438 int add_to_page_cache(struct page *page, struct address_space *mapping,
439 pgoff_t offset, gfp_t gfp_mask)
441 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
444 write_lock_irq(&mapping->tree_lock);
445 error = radix_tree_insert(&mapping->page_tree, offset, page);
447 page_cache_get(page);
449 page->mapping = mapping;
450 page->index = offset;
454 write_unlock_irq(&mapping->tree_lock);
455 radix_tree_preload_end();
459 EXPORT_SYMBOL(add_to_page_cache);
461 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
462 pgoff_t offset, gfp_t gfp_mask)
464 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
471 struct page *page_cache_alloc(struct address_space *x)
473 if (cpuset_do_page_mem_spread()) {
474 int n = cpuset_mem_spread_node();
475 return alloc_pages_node(n, mapping_gfp_mask(x), 0);
477 return alloc_pages(mapping_gfp_mask(x), 0);
479 EXPORT_SYMBOL(page_cache_alloc);
481 struct page *page_cache_alloc_cold(struct address_space *x)
483 if (cpuset_do_page_mem_spread()) {
484 int n = cpuset_mem_spread_node();
485 return alloc_pages_node(n, mapping_gfp_mask(x)|__GFP_COLD, 0);
487 return alloc_pages(mapping_gfp_mask(x)|__GFP_COLD, 0);
489 EXPORT_SYMBOL(page_cache_alloc_cold);
493 * In order to wait for pages to become available there must be
494 * waitqueues associated with pages. By using a hash table of
495 * waitqueues where the bucket discipline is to maintain all
496 * waiters on the same queue and wake all when any of the pages
497 * become available, and for the woken contexts to check to be
498 * sure the appropriate page became available, this saves space
499 * at a cost of "thundering herd" phenomena during rare hash
502 static wait_queue_head_t *page_waitqueue(struct page *page)
504 const struct zone *zone = page_zone(page);
506 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
509 static inline void wake_up_page(struct page *page, int bit)
511 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
514 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
516 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
518 if (test_bit(bit_nr, &page->flags))
519 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
520 TASK_UNINTERRUPTIBLE);
522 EXPORT_SYMBOL(wait_on_page_bit);
525 * unlock_page - unlock a locked page
528 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
529 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
530 * mechananism between PageLocked pages and PageWriteback pages is shared.
531 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
533 * The first mb is necessary to safely close the critical section opened by the
534 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
535 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
536 * parallel wait_on_page_locked()).
538 void fastcall unlock_page(struct page *page)
540 smp_mb__before_clear_bit();
541 if (!TestClearPageLocked(page))
543 smp_mb__after_clear_bit();
544 wake_up_page(page, PG_locked);
546 EXPORT_SYMBOL(unlock_page);
549 * end_page_writeback - end writeback against a page
552 void end_page_writeback(struct page *page)
554 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
555 if (!test_clear_page_writeback(page))
558 smp_mb__after_clear_bit();
559 wake_up_page(page, PG_writeback);
561 EXPORT_SYMBOL(end_page_writeback);
564 * __lock_page - get a lock on the page, assuming we need to sleep to get it
565 * @page: the page to lock
567 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
568 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
569 * chances are that on the second loop, the block layer's plug list is empty,
570 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
572 void fastcall __lock_page(struct page *page)
574 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
576 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
577 TASK_UNINTERRUPTIBLE);
579 EXPORT_SYMBOL(__lock_page);
582 * find_get_page - find and get a page reference
583 * @mapping: the address_space to search
584 * @offset: the page index
586 * A rather lightweight function, finding and getting a reference to a
587 * hashed page atomically.
589 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
593 read_lock_irq(&mapping->tree_lock);
594 page = radix_tree_lookup(&mapping->page_tree, offset);
596 page_cache_get(page);
597 read_unlock_irq(&mapping->tree_lock);
600 EXPORT_SYMBOL(find_get_page);
603 * find_trylock_page - find and lock a page
604 * @mapping: the address_space to search
605 * @offset: the page index
607 * Same as find_get_page(), but trylock it instead of incrementing the count.
609 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
613 read_lock_irq(&mapping->tree_lock);
614 page = radix_tree_lookup(&mapping->page_tree, offset);
615 if (page && TestSetPageLocked(page))
617 read_unlock_irq(&mapping->tree_lock);
620 EXPORT_SYMBOL(find_trylock_page);
623 * find_lock_page - locate, pin and lock a pagecache page
624 * @mapping: the address_space to search
625 * @offset: the page index
627 * Locates the desired pagecache page, locks it, increments its reference
628 * count and returns its address.
630 * Returns zero if the page was not present. find_lock_page() may sleep.
632 struct page *find_lock_page(struct address_space *mapping,
633 unsigned long offset)
637 read_lock_irq(&mapping->tree_lock);
639 page = radix_tree_lookup(&mapping->page_tree, offset);
641 page_cache_get(page);
642 if (TestSetPageLocked(page)) {
643 read_unlock_irq(&mapping->tree_lock);
645 read_lock_irq(&mapping->tree_lock);
647 /* Has the page been truncated while we slept? */
648 if (unlikely(page->mapping != mapping ||
649 page->index != offset)) {
651 page_cache_release(page);
656 read_unlock_irq(&mapping->tree_lock);
659 EXPORT_SYMBOL(find_lock_page);
662 * find_or_create_page - locate or add a pagecache page
663 * @mapping: the page's address_space
664 * @index: the page's index into the mapping
665 * @gfp_mask: page allocation mode
667 * Locates a page in the pagecache. If the page is not present, a new page
668 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
669 * LRU list. The returned page is locked and has its reference count
672 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
675 * find_or_create_page() returns the desired page's address, or zero on
678 struct page *find_or_create_page(struct address_space *mapping,
679 unsigned long index, gfp_t gfp_mask)
681 struct page *page, *cached_page = NULL;
684 page = find_lock_page(mapping, index);
687 cached_page = alloc_page(gfp_mask);
691 err = add_to_page_cache_lru(cached_page, mapping,
696 } else if (err == -EEXIST)
700 page_cache_release(cached_page);
703 EXPORT_SYMBOL(find_or_create_page);
706 * find_get_pages - gang pagecache lookup
707 * @mapping: The address_space to search
708 * @start: The starting page index
709 * @nr_pages: The maximum number of pages
710 * @pages: Where the resulting pages are placed
712 * find_get_pages() will search for and return a group of up to
713 * @nr_pages pages in the mapping. The pages are placed at @pages.
714 * find_get_pages() takes a reference against the returned pages.
716 * The search returns a group of mapping-contiguous pages with ascending
717 * indexes. There may be holes in the indices due to not-present pages.
719 * find_get_pages() returns the number of pages which were found.
721 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
722 unsigned int nr_pages, struct page **pages)
727 read_lock_irq(&mapping->tree_lock);
728 ret = radix_tree_gang_lookup(&mapping->page_tree,
729 (void **)pages, start, nr_pages);
730 for (i = 0; i < ret; i++)
731 page_cache_get(pages[i]);
732 read_unlock_irq(&mapping->tree_lock);
737 * find_get_pages_contig - gang contiguous pagecache lookup
738 * @mapping: The address_space to search
739 * @index: The starting page index
740 * @nr_pages: The maximum number of pages
741 * @pages: Where the resulting pages are placed
743 * find_get_pages_contig() works exactly like find_get_pages(), except
744 * that the returned number of pages are guaranteed to be contiguous.
746 * find_get_pages_contig() returns the number of pages which were found.
748 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
749 unsigned int nr_pages, struct page **pages)
754 read_lock_irq(&mapping->tree_lock);
755 ret = radix_tree_gang_lookup(&mapping->page_tree,
756 (void **)pages, index, nr_pages);
757 for (i = 0; i < ret; i++) {
758 if (pages[i]->mapping == NULL || pages[i]->index != index)
761 page_cache_get(pages[i]);
764 read_unlock_irq(&mapping->tree_lock);
769 * find_get_pages_tag - find and return pages that match @tag
770 * @mapping: the address_space to search
771 * @index: the starting page index
772 * @tag: the tag index
773 * @nr_pages: the maximum number of pages
774 * @pages: where the resulting pages are placed
776 * Like find_get_pages, except we only return pages which are tagged with
777 * @tag. We update @index to index the next page for the traversal.
779 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
780 int tag, unsigned int nr_pages, struct page **pages)
785 read_lock_irq(&mapping->tree_lock);
786 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
787 (void **)pages, *index, nr_pages, tag);
788 for (i = 0; i < ret; i++)
789 page_cache_get(pages[i]);
791 *index = pages[ret - 1]->index + 1;
792 read_unlock_irq(&mapping->tree_lock);
797 * grab_cache_page_nowait - returns locked page at given index in given cache
798 * @mapping: target address_space
799 * @index: the page index
801 * Same as grab_cache_page, but do not wait if the page is unavailable.
802 * This is intended for speculative data generators, where the data can
803 * be regenerated if the page couldn't be grabbed. This routine should
804 * be safe to call while holding the lock for another page.
806 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
807 * and deadlock against the caller's locked page.
810 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
812 struct page *page = find_get_page(mapping, index);
816 if (!TestSetPageLocked(page))
818 page_cache_release(page);
821 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
822 page = alloc_pages(gfp_mask, 0);
823 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
824 page_cache_release(page);
829 EXPORT_SYMBOL(grab_cache_page_nowait);
832 * do_generic_mapping_read - generic file read routine
833 * @mapping: address_space to be read
834 * @_ra: file's readahead state
835 * @filp: the file to read
836 * @ppos: current file position
837 * @desc: read_descriptor
838 * @actor: read method
840 * This is a generic file read routine, and uses the
841 * mapping->a_ops->readpage() function for the actual low-level stuff.
843 * This is really ugly. But the goto's actually try to clarify some
844 * of the logic when it comes to error handling etc.
846 * Note the struct file* is only passed for the use of readpage.
849 void do_generic_mapping_read(struct address_space *mapping,
850 struct file_ra_state *_ra,
853 read_descriptor_t *desc,
856 struct inode *inode = mapping->host;
858 unsigned long end_index;
859 unsigned long offset;
860 unsigned long last_index;
861 unsigned long next_index;
862 unsigned long prev_index;
864 struct page *cached_page;
866 struct file_ra_state ra = *_ra;
869 index = *ppos >> PAGE_CACHE_SHIFT;
871 prev_index = ra.prev_page;
872 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
873 offset = *ppos & ~PAGE_CACHE_MASK;
875 isize = i_size_read(inode);
879 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
882 unsigned long nr, ret;
884 /* nr is the maximum number of bytes to copy from this page */
885 nr = PAGE_CACHE_SIZE;
886 if (index >= end_index) {
887 if (index > end_index)
889 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
897 if (index == next_index)
898 next_index = page_cache_readahead(mapping, &ra, filp,
899 index, last_index - index);
902 page = find_get_page(mapping, index);
903 if (unlikely(page == NULL)) {
904 handle_ra_miss(mapping, &ra, index);
907 if (!PageUptodate(page))
908 goto page_not_up_to_date;
911 /* If users can be writing to this page using arbitrary
912 * virtual addresses, take care about potential aliasing
913 * before reading the page on the kernel side.
915 if (mapping_writably_mapped(mapping))
916 flush_dcache_page(page);
919 * When (part of) the same page is read multiple times
920 * in succession, only mark it as accessed the first time.
922 if (prev_index != index)
923 mark_page_accessed(page);
927 * Ok, we have the page, and it's up-to-date, so
928 * now we can copy it to user space...
930 * The actor routine returns how many bytes were actually used..
931 * NOTE! This may not be the same as how much of a user buffer
932 * we filled up (we may be padding etc), so we can only update
933 * "pos" here (the actor routine has to update the user buffer
934 * pointers and the remaining count).
936 ret = actor(desc, page, offset, nr);
938 index += offset >> PAGE_CACHE_SHIFT;
939 offset &= ~PAGE_CACHE_MASK;
941 page_cache_release(page);
942 if (ret == nr && desc->count)
947 /* Get exclusive access to the page ... */
950 /* Did it get unhashed before we got the lock? */
951 if (!page->mapping) {
953 page_cache_release(page);
957 /* Did somebody else fill it already? */
958 if (PageUptodate(page)) {
964 /* Start the actual read. The read will unlock the page. */
965 error = mapping->a_ops->readpage(filp, page);
967 if (unlikely(error)) {
968 if (error == AOP_TRUNCATED_PAGE) {
969 page_cache_release(page);
975 if (!PageUptodate(page)) {
977 if (!PageUptodate(page)) {
978 if (page->mapping == NULL) {
980 * invalidate_inode_pages got it
983 page_cache_release(page);
994 * i_size must be checked after we have done ->readpage.
996 * Checking i_size after the readpage allows us to calculate
997 * the correct value for "nr", which means the zero-filled
998 * part of the page is not copied back to userspace (unless
999 * another truncate extends the file - this is desired though).
1001 isize = i_size_read(inode);
1002 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1003 if (unlikely(!isize || index > end_index)) {
1004 page_cache_release(page);
1008 /* nr is the maximum number of bytes to copy from this page */
1009 nr = PAGE_CACHE_SIZE;
1010 if (index == end_index) {
1011 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1013 page_cache_release(page);
1021 /* UHHUH! A synchronous read error occurred. Report it */
1022 desc->error = error;
1023 page_cache_release(page);
1028 * Ok, it wasn't cached, so we need to create a new
1032 cached_page = page_cache_alloc_cold(mapping);
1034 desc->error = -ENOMEM;
1038 error = add_to_page_cache_lru(cached_page, mapping,
1041 if (error == -EEXIST)
1043 desc->error = error;
1054 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1056 page_cache_release(cached_page);
1058 file_accessed(filp);
1060 EXPORT_SYMBOL(do_generic_mapping_read);
1062 int file_read_actor(read_descriptor_t *desc, struct page *page,
1063 unsigned long offset, unsigned long size)
1066 unsigned long left, count = desc->count;
1072 * Faults on the destination of a read are common, so do it before
1075 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1076 kaddr = kmap_atomic(page, KM_USER0);
1077 left = __copy_to_user_inatomic(desc->arg.buf,
1078 kaddr + offset, size);
1079 kunmap_atomic(kaddr, KM_USER0);
1084 /* Do it the slow way */
1086 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1091 desc->error = -EFAULT;
1094 desc->count = count - size;
1095 desc->written += size;
1096 desc->arg.buf += size;
1101 * __generic_file_aio_read - generic filesystem read routine
1102 * @iocb: kernel I/O control block
1103 * @iov: io vector request
1104 * @nr_segs: number of segments in the iovec
1105 * @ppos: current file position
1107 * This is the "read()" routine for all filesystems
1108 * that can use the page cache directly.
1111 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1112 unsigned long nr_segs, loff_t *ppos)
1114 struct file *filp = iocb->ki_filp;
1120 for (seg = 0; seg < nr_segs; seg++) {
1121 const struct iovec *iv = &iov[seg];
1124 * If any segment has a negative length, or the cumulative
1125 * length ever wraps negative then return -EINVAL.
1127 count += iv->iov_len;
1128 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1130 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1135 count -= iv->iov_len; /* This segment is no good */
1139 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1140 if (filp->f_flags & O_DIRECT) {
1141 loff_t pos = *ppos, size;
1142 struct address_space *mapping;
1143 struct inode *inode;
1145 mapping = filp->f_mapping;
1146 inode = mapping->host;
1149 goto out; /* skip atime */
1150 size = i_size_read(inode);
1152 retval = generic_file_direct_IO(READ, iocb,
1154 if (retval > 0 && !is_sync_kiocb(iocb))
1155 retval = -EIOCBQUEUED;
1157 *ppos = pos + retval;
1159 file_accessed(filp);
1165 for (seg = 0; seg < nr_segs; seg++) {
1166 read_descriptor_t desc;
1169 desc.arg.buf = iov[seg].iov_base;
1170 desc.count = iov[seg].iov_len;
1171 if (desc.count == 0)
1174 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1175 retval += desc.written;
1177 retval = retval ?: desc.error;
1185 EXPORT_SYMBOL(__generic_file_aio_read);
1188 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1190 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1192 BUG_ON(iocb->ki_pos != pos);
1193 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1195 EXPORT_SYMBOL(generic_file_aio_read);
1198 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1200 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1204 init_sync_kiocb(&kiocb, filp);
1205 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1206 if (-EIOCBQUEUED == ret)
1207 ret = wait_on_sync_kiocb(&kiocb);
1210 EXPORT_SYMBOL(generic_file_read);
1212 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1215 unsigned long count = desc->count;
1216 struct file *file = desc->arg.data;
1221 written = file->f_op->sendpage(file, page, offset,
1222 size, &file->f_pos, size<count);
1224 desc->error = written;
1227 desc->count = count - written;
1228 desc->written += written;
1232 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1233 size_t count, read_actor_t actor, void *target)
1235 read_descriptor_t desc;
1242 desc.arg.data = target;
1245 do_generic_file_read(in_file, ppos, &desc, actor);
1247 return desc.written;
1250 EXPORT_SYMBOL(generic_file_sendfile);
1253 do_readahead(struct address_space *mapping, struct file *filp,
1254 unsigned long index, unsigned long nr)
1256 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1259 force_page_cache_readahead(mapping, filp, index,
1260 max_sane_readahead(nr));
1264 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1272 if (file->f_mode & FMODE_READ) {
1273 struct address_space *mapping = file->f_mapping;
1274 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1275 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1276 unsigned long len = end - start + 1;
1277 ret = do_readahead(mapping, file, start, len);
1285 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1287 * page_cache_read - adds requested page to the page cache if not already there
1288 * @file: file to read
1289 * @offset: page index
1291 * This adds the requested page to the page cache if it isn't already there,
1292 * and schedules an I/O to read in its contents from disk.
1294 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1296 struct address_space *mapping = file->f_mapping;
1301 page = page_cache_alloc_cold(mapping);
1305 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1307 ret = mapping->a_ops->readpage(file, page);
1308 else if (ret == -EEXIST)
1309 ret = 0; /* losing race to add is OK */
1311 page_cache_release(page);
1313 } while (ret == AOP_TRUNCATED_PAGE);
1318 #define MMAP_LOTSAMISS (100)
1321 * filemap_nopage - read in file data for page fault handling
1322 * @area: the applicable vm_area
1323 * @address: target address to read in
1324 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1326 * filemap_nopage() is invoked via the vma operations vector for a
1327 * mapped memory region to read in file data during a page fault.
1329 * The goto's are kind of ugly, but this streamlines the normal case of having
1330 * it in the page cache, and handles the special cases reasonably without
1331 * having a lot of duplicated code.
1333 struct page *filemap_nopage(struct vm_area_struct *area,
1334 unsigned long address, int *type)
1337 struct file *file = area->vm_file;
1338 struct address_space *mapping = file->f_mapping;
1339 struct file_ra_state *ra = &file->f_ra;
1340 struct inode *inode = mapping->host;
1342 unsigned long size, pgoff;
1343 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1345 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1348 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1350 goto outside_data_content;
1352 /* If we don't want any read-ahead, don't bother */
1353 if (VM_RandomReadHint(area))
1354 goto no_cached_page;
1357 * The readahead code wants to be told about each and every page
1358 * so it can build and shrink its windows appropriately
1360 * For sequential accesses, we use the generic readahead logic.
1362 if (VM_SequentialReadHint(area))
1363 page_cache_readahead(mapping, ra, file, pgoff, 1);
1366 * Do we have something in the page cache already?
1369 page = find_get_page(mapping, pgoff);
1371 unsigned long ra_pages;
1373 if (VM_SequentialReadHint(area)) {
1374 handle_ra_miss(mapping, ra, pgoff);
1375 goto no_cached_page;
1380 * Do we miss much more than hit in this file? If so,
1381 * stop bothering with read-ahead. It will only hurt.
1383 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1384 goto no_cached_page;
1387 * To keep the pgmajfault counter straight, we need to
1388 * check did_readaround, as this is an inner loop.
1390 if (!did_readaround) {
1391 majmin = VM_FAULT_MAJOR;
1392 inc_page_state(pgmajfault);
1395 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1399 if (pgoff > ra_pages / 2)
1400 start = pgoff - ra_pages / 2;
1401 do_page_cache_readahead(mapping, file, start, ra_pages);
1403 page = find_get_page(mapping, pgoff);
1405 goto no_cached_page;
1408 if (!did_readaround)
1412 * Ok, found a page in the page cache, now we need to check
1413 * that it's up-to-date.
1415 if (!PageUptodate(page))
1416 goto page_not_uptodate;
1420 * Found the page and have a reference on it.
1422 mark_page_accessed(page);
1427 outside_data_content:
1429 * An external ptracer can access pages that normally aren't
1432 if (area->vm_mm == current->mm)
1434 /* Fall through to the non-read-ahead case */
1437 * We're only likely to ever get here if MADV_RANDOM is in
1440 error = page_cache_read(file, pgoff);
1444 * The page we want has now been added to the page cache.
1445 * In the unlikely event that someone removed it in the
1446 * meantime, we'll just come back here and read it again.
1452 * An error return from page_cache_read can result if the
1453 * system is low on memory, or a problem occurs while trying
1456 if (error == -ENOMEM)
1461 if (!did_readaround) {
1462 majmin = VM_FAULT_MAJOR;
1463 inc_page_state(pgmajfault);
1467 /* Did it get unhashed while we waited for it? */
1468 if (!page->mapping) {
1470 page_cache_release(page);
1474 /* Did somebody else get it up-to-date? */
1475 if (PageUptodate(page)) {
1480 error = mapping->a_ops->readpage(file, page);
1482 wait_on_page_locked(page);
1483 if (PageUptodate(page))
1485 } else if (error == AOP_TRUNCATED_PAGE) {
1486 page_cache_release(page);
1491 * Umm, take care of errors if the page isn't up-to-date.
1492 * Try to re-read it _once_. We do this synchronously,
1493 * because there really aren't any performance issues here
1494 * and we need to check for errors.
1498 /* Somebody truncated the page on us? */
1499 if (!page->mapping) {
1501 page_cache_release(page);
1505 /* Somebody else successfully read it in? */
1506 if (PageUptodate(page)) {
1510 ClearPageError(page);
1511 error = mapping->a_ops->readpage(file, page);
1513 wait_on_page_locked(page);
1514 if (PageUptodate(page))
1516 } else if (error == AOP_TRUNCATED_PAGE) {
1517 page_cache_release(page);
1522 * Things didn't work out. Return zero to tell the
1523 * mm layer so, possibly freeing the page cache page first.
1525 page_cache_release(page);
1528 EXPORT_SYMBOL(filemap_nopage);
1530 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1533 struct address_space *mapping = file->f_mapping;
1538 * Do we have something in the page cache already?
1541 page = find_get_page(mapping, pgoff);
1545 goto no_cached_page;
1549 * Ok, found a page in the page cache, now we need to check
1550 * that it's up-to-date.
1552 if (!PageUptodate(page)) {
1554 page_cache_release(page);
1557 goto page_not_uptodate;
1562 * Found the page and have a reference on it.
1564 mark_page_accessed(page);
1568 error = page_cache_read(file, pgoff);
1571 * The page we want has now been added to the page cache.
1572 * In the unlikely event that someone removed it in the
1573 * meantime, we'll just come back here and read it again.
1579 * An error return from page_cache_read can result if the
1580 * system is low on memory, or a problem occurs while trying
1588 /* Did it get unhashed while we waited for it? */
1589 if (!page->mapping) {
1594 /* Did somebody else get it up-to-date? */
1595 if (PageUptodate(page)) {
1600 error = mapping->a_ops->readpage(file, page);
1602 wait_on_page_locked(page);
1603 if (PageUptodate(page))
1605 } else if (error == AOP_TRUNCATED_PAGE) {
1606 page_cache_release(page);
1611 * Umm, take care of errors if the page isn't up-to-date.
1612 * Try to re-read it _once_. We do this synchronously,
1613 * because there really aren't any performance issues here
1614 * and we need to check for errors.
1618 /* Somebody truncated the page on us? */
1619 if (!page->mapping) {
1623 /* Somebody else successfully read it in? */
1624 if (PageUptodate(page)) {
1629 ClearPageError(page);
1630 error = mapping->a_ops->readpage(file, page);
1632 wait_on_page_locked(page);
1633 if (PageUptodate(page))
1635 } else if (error == AOP_TRUNCATED_PAGE) {
1636 page_cache_release(page);
1641 * Things didn't work out. Return zero to tell the
1642 * mm layer so, possibly freeing the page cache page first.
1645 page_cache_release(page);
1650 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1651 unsigned long len, pgprot_t prot, unsigned long pgoff,
1654 struct file *file = vma->vm_file;
1655 struct address_space *mapping = file->f_mapping;
1656 struct inode *inode = mapping->host;
1658 struct mm_struct *mm = vma->vm_mm;
1663 force_page_cache_readahead(mapping, vma->vm_file,
1664 pgoff, len >> PAGE_CACHE_SHIFT);
1667 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1668 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1671 page = filemap_getpage(file, pgoff, nonblock);
1673 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1674 * done in shmem_populate calling shmem_getpage */
1675 if (!page && !nonblock)
1679 err = install_page(mm, vma, addr, page, prot);
1681 page_cache_release(page);
1684 } else if (vma->vm_flags & VM_NONLINEAR) {
1685 /* No page was found just because we can't read it in now (being
1686 * here implies nonblock != 0), but the page may exist, so set
1687 * the PTE to fault it in later. */
1688 err = install_file_pte(mm, vma, addr, pgoff, prot);
1701 EXPORT_SYMBOL(filemap_populate);
1703 struct vm_operations_struct generic_file_vm_ops = {
1704 .nopage = filemap_nopage,
1705 .populate = filemap_populate,
1708 /* This is used for a general mmap of a disk file */
1710 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1712 struct address_space *mapping = file->f_mapping;
1714 if (!mapping->a_ops->readpage)
1716 file_accessed(file);
1717 vma->vm_ops = &generic_file_vm_ops;
1722 * This is for filesystems which do not implement ->writepage.
1724 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1726 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1728 return generic_file_mmap(file, vma);
1731 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1735 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1739 #endif /* CONFIG_MMU */
1741 EXPORT_SYMBOL(generic_file_mmap);
1742 EXPORT_SYMBOL(generic_file_readonly_mmap);
1744 static inline struct page *__read_cache_page(struct address_space *mapping,
1745 unsigned long index,
1746 int (*filler)(void *,struct page*),
1749 struct page *page, *cached_page = NULL;
1752 page = find_get_page(mapping, index);
1755 cached_page = page_cache_alloc_cold(mapping);
1757 return ERR_PTR(-ENOMEM);
1759 err = add_to_page_cache_lru(cached_page, mapping,
1764 /* Presumably ENOMEM for radix tree node */
1765 page_cache_release(cached_page);
1766 return ERR_PTR(err);
1770 err = filler(data, page);
1772 page_cache_release(page);
1773 page = ERR_PTR(err);
1777 page_cache_release(cached_page);
1782 * read_cache_page - read into page cache, fill it if needed
1783 * @mapping: the page's address_space
1784 * @index: the page index
1785 * @filler: function to perform the read
1786 * @data: destination for read data
1788 * Read into the page cache. If a page already exists,
1789 * and PageUptodate() is not set, try to fill the page.
1791 struct page *read_cache_page(struct address_space *mapping,
1792 unsigned long index,
1793 int (*filler)(void *,struct page*),
1800 page = __read_cache_page(mapping, index, filler, data);
1803 mark_page_accessed(page);
1804 if (PageUptodate(page))
1808 if (!page->mapping) {
1810 page_cache_release(page);
1813 if (PageUptodate(page)) {
1817 err = filler(data, page);
1819 page_cache_release(page);
1820 page = ERR_PTR(err);
1825 EXPORT_SYMBOL(read_cache_page);
1828 * If the page was newly created, increment its refcount and add it to the
1829 * caller's lru-buffering pagevec. This function is specifically for
1830 * generic_file_write().
1832 static inline struct page *
1833 __grab_cache_page(struct address_space *mapping, unsigned long index,
1834 struct page **cached_page, struct pagevec *lru_pvec)
1839 page = find_lock_page(mapping, index);
1841 if (!*cached_page) {
1842 *cached_page = page_cache_alloc(mapping);
1846 err = add_to_page_cache(*cached_page, mapping,
1851 page = *cached_page;
1852 page_cache_get(page);
1853 if (!pagevec_add(lru_pvec, page))
1854 __pagevec_lru_add(lru_pvec);
1855 *cached_page = NULL;
1862 * The logic we want is
1864 * if suid or (sgid and xgrp)
1867 int remove_suid(struct dentry *dentry)
1869 mode_t mode = dentry->d_inode->i_mode;
1873 /* suid always must be killed */
1874 if (unlikely(mode & S_ISUID))
1875 kill = ATTR_KILL_SUID;
1878 * sgid without any exec bits is just a mandatory locking mark; leave
1879 * it alone. If some exec bits are set, it's a real sgid; kill it.
1881 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1882 kill |= ATTR_KILL_SGID;
1884 if (unlikely(kill && !capable(CAP_FSETID))) {
1885 struct iattr newattrs;
1887 newattrs.ia_valid = ATTR_FORCE | kill;
1888 result = notify_change(dentry, &newattrs);
1892 EXPORT_SYMBOL(remove_suid);
1895 __filemap_copy_from_user_iovec_inatomic(char *vaddr,
1896 const struct iovec *iov, size_t base, size_t bytes)
1898 size_t copied = 0, left = 0;
1901 char __user *buf = iov->iov_base + base;
1902 int copy = min(bytes, iov->iov_len - base);
1905 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1914 return copied - left;
1918 * Performs necessary checks before doing a write
1920 * Can adjust writing position or amount of bytes to write.
1921 * Returns appropriate error code that caller should return or
1922 * zero in case that write should be allowed.
1924 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1926 struct inode *inode = file->f_mapping->host;
1927 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1929 if (unlikely(*pos < 0))
1933 /* FIXME: this is for backwards compatibility with 2.4 */
1934 if (file->f_flags & O_APPEND)
1935 *pos = i_size_read(inode);
1937 if (limit != RLIM_INFINITY) {
1938 if (*pos >= limit) {
1939 send_sig(SIGXFSZ, current, 0);
1942 if (*count > limit - (typeof(limit))*pos) {
1943 *count = limit - (typeof(limit))*pos;
1951 if (unlikely(*pos + *count > MAX_NON_LFS &&
1952 !(file->f_flags & O_LARGEFILE))) {
1953 if (*pos >= MAX_NON_LFS) {
1954 send_sig(SIGXFSZ, current, 0);
1957 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1958 *count = MAX_NON_LFS - (unsigned long)*pos;
1963 * Are we about to exceed the fs block limit ?
1965 * If we have written data it becomes a short write. If we have
1966 * exceeded without writing data we send a signal and return EFBIG.
1967 * Linus frestrict idea will clean these up nicely..
1969 if (likely(!isblk)) {
1970 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1971 if (*count || *pos > inode->i_sb->s_maxbytes) {
1972 send_sig(SIGXFSZ, current, 0);
1975 /* zero-length writes at ->s_maxbytes are OK */
1978 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1979 *count = inode->i_sb->s_maxbytes - *pos;
1982 if (bdev_read_only(I_BDEV(inode)))
1984 isize = i_size_read(inode);
1985 if (*pos >= isize) {
1986 if (*count || *pos > isize)
1990 if (*pos + *count > isize)
1991 *count = isize - *pos;
1995 EXPORT_SYMBOL(generic_write_checks);
1998 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1999 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2000 size_t count, size_t ocount)
2002 struct file *file = iocb->ki_filp;
2003 struct address_space *mapping = file->f_mapping;
2004 struct inode *inode = mapping->host;
2007 if (count != ocount)
2008 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2010 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2012 loff_t end = pos + written;
2013 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2014 i_size_write(inode, end);
2015 mark_inode_dirty(inode);
2021 * Sync the fs metadata but not the minor inode changes and
2022 * of course not the data as we did direct DMA for the IO.
2023 * i_mutex is held, which protects generic_osync_inode() from
2026 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2027 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2031 if (written == count && !is_sync_kiocb(iocb))
2032 written = -EIOCBQUEUED;
2035 EXPORT_SYMBOL(generic_file_direct_write);
2038 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2039 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2040 size_t count, ssize_t written)
2042 struct file *file = iocb->ki_filp;
2043 struct address_space * mapping = file->f_mapping;
2044 struct address_space_operations *a_ops = mapping->a_ops;
2045 struct inode *inode = mapping->host;
2048 struct page *cached_page = NULL;
2050 struct pagevec lru_pvec;
2051 const struct iovec *cur_iov = iov; /* current iovec */
2052 size_t iov_base = 0; /* offset in the current iovec */
2055 pagevec_init(&lru_pvec, 0);
2058 * handle partial DIO write. Adjust cur_iov if needed.
2060 if (likely(nr_segs == 1))
2061 buf = iov->iov_base + written;
2063 filemap_set_next_iovec(&cur_iov, &iov_base, written);
2064 buf = cur_iov->iov_base + iov_base;
2068 unsigned long index;
2069 unsigned long offset;
2070 unsigned long maxlen;
2073 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2074 index = pos >> PAGE_CACHE_SHIFT;
2075 bytes = PAGE_CACHE_SIZE - offset;
2080 * Bring in the user page that we will copy from _first_.
2081 * Otherwise there's a nasty deadlock on copying from the
2082 * same page as we're writing to, without it being marked
2085 maxlen = cur_iov->iov_len - iov_base;
2088 fault_in_pages_readable(buf, maxlen);
2090 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2096 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2097 if (unlikely(status)) {
2098 loff_t isize = i_size_read(inode);
2100 if (status != AOP_TRUNCATED_PAGE)
2102 page_cache_release(page);
2103 if (status == AOP_TRUNCATED_PAGE)
2106 * prepare_write() may have instantiated a few blocks
2107 * outside i_size. Trim these off again.
2109 if (pos + bytes > isize)
2110 vmtruncate(inode, isize);
2113 if (likely(nr_segs == 1))
2114 copied = filemap_copy_from_user(page, offset,
2117 copied = filemap_copy_from_user_iovec(page, offset,
2118 cur_iov, iov_base, bytes);
2119 flush_dcache_page(page);
2120 status = a_ops->commit_write(file, page, offset, offset+bytes);
2121 if (status == AOP_TRUNCATED_PAGE) {
2122 page_cache_release(page);
2125 if (likely(copied > 0)) {
2134 if (unlikely(nr_segs > 1)) {
2135 filemap_set_next_iovec(&cur_iov,
2138 buf = cur_iov->iov_base +
2145 if (unlikely(copied != bytes))
2149 mark_page_accessed(page);
2150 page_cache_release(page);
2153 balance_dirty_pages_ratelimited(mapping);
2159 page_cache_release(cached_page);
2162 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2164 if (likely(status >= 0)) {
2165 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2166 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2167 status = generic_osync_inode(inode, mapping,
2168 OSYNC_METADATA|OSYNC_DATA);
2173 * If we get here for O_DIRECT writes then we must have fallen through
2174 * to buffered writes (block instantiation inside i_size). So we sync
2175 * the file data here, to try to honour O_DIRECT expectations.
2177 if (unlikely(file->f_flags & O_DIRECT) && written)
2178 status = filemap_write_and_wait(mapping);
2180 pagevec_lru_add(&lru_pvec);
2181 return written ? written : status;
2183 EXPORT_SYMBOL(generic_file_buffered_write);
2186 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2187 unsigned long nr_segs, loff_t *ppos)
2189 struct file *file = iocb->ki_filp;
2190 struct address_space * mapping = file->f_mapping;
2191 size_t ocount; /* original count */
2192 size_t count; /* after file limit checks */
2193 struct inode *inode = mapping->host;
2200 for (seg = 0; seg < nr_segs; seg++) {
2201 const struct iovec *iv = &iov[seg];
2204 * If any segment has a negative length, or the cumulative
2205 * length ever wraps negative then return -EINVAL.
2207 ocount += iv->iov_len;
2208 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2210 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2215 ocount -= iv->iov_len; /* This segment is no good */
2222 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2224 /* We can write back this queue in page reclaim */
2225 current->backing_dev_info = mapping->backing_dev_info;
2228 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2235 err = remove_suid(file->f_dentry);
2239 file_update_time(file);
2241 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2242 if (unlikely(file->f_flags & O_DIRECT)) {
2243 written = generic_file_direct_write(iocb, iov,
2244 &nr_segs, pos, ppos, count, ocount);
2245 if (written < 0 || written == count)
2248 * direct-io write to a hole: fall through to buffered I/O
2249 * for completing the rest of the request.
2255 written = generic_file_buffered_write(iocb, iov, nr_segs,
2256 pos, ppos, count, written);
2258 current->backing_dev_info = NULL;
2259 return written ? written : err;
2261 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2264 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2265 unsigned long nr_segs, loff_t *ppos)
2267 struct file *file = iocb->ki_filp;
2268 struct address_space *mapping = file->f_mapping;
2269 struct inode *inode = mapping->host;
2273 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2275 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2278 err = sync_page_range_nolock(inode, mapping, pos, ret);
2286 __generic_file_write_nolock(struct file *file, const struct iovec *iov,
2287 unsigned long nr_segs, loff_t *ppos)
2292 init_sync_kiocb(&kiocb, file);
2293 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2294 if (ret == -EIOCBQUEUED)
2295 ret = wait_on_sync_kiocb(&kiocb);
2300 generic_file_write_nolock(struct file *file, const struct iovec *iov,
2301 unsigned long nr_segs, loff_t *ppos)
2306 init_sync_kiocb(&kiocb, file);
2307 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2308 if (-EIOCBQUEUED == ret)
2309 ret = wait_on_sync_kiocb(&kiocb);
2312 EXPORT_SYMBOL(generic_file_write_nolock);
2314 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2315 size_t count, loff_t pos)
2317 struct file *file = iocb->ki_filp;
2318 struct address_space *mapping = file->f_mapping;
2319 struct inode *inode = mapping->host;
2321 struct iovec local_iov = { .iov_base = (void __user *)buf,
2324 BUG_ON(iocb->ki_pos != pos);
2326 mutex_lock(&inode->i_mutex);
2327 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2329 mutex_unlock(&inode->i_mutex);
2331 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2334 err = sync_page_range(inode, mapping, pos, ret);
2340 EXPORT_SYMBOL(generic_file_aio_write);
2342 ssize_t generic_file_write(struct file *file, const char __user *buf,
2343 size_t count, loff_t *ppos)
2345 struct address_space *mapping = file->f_mapping;
2346 struct inode *inode = mapping->host;
2348 struct iovec local_iov = { .iov_base = (void __user *)buf,
2351 mutex_lock(&inode->i_mutex);
2352 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
2353 mutex_unlock(&inode->i_mutex);
2355 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2358 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2364 EXPORT_SYMBOL(generic_file_write);
2366 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2367 unsigned long nr_segs, loff_t *ppos)
2372 init_sync_kiocb(&kiocb, filp);
2373 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2374 if (-EIOCBQUEUED == ret)
2375 ret = wait_on_sync_kiocb(&kiocb);
2378 EXPORT_SYMBOL(generic_file_readv);
2380 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2381 unsigned long nr_segs, loff_t *ppos)
2383 struct address_space *mapping = file->f_mapping;
2384 struct inode *inode = mapping->host;
2387 mutex_lock(&inode->i_mutex);
2388 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
2389 mutex_unlock(&inode->i_mutex);
2391 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2394 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2400 EXPORT_SYMBOL(generic_file_writev);
2403 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2404 * went wrong during pagecache shootdown.
2407 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2408 loff_t offset, unsigned long nr_segs)
2410 struct file *file = iocb->ki_filp;
2411 struct address_space *mapping = file->f_mapping;
2413 size_t write_len = 0;
2416 * If it's a write, unmap all mmappings of the file up-front. This
2417 * will cause any pte dirty bits to be propagated into the pageframes
2418 * for the subsequent filemap_write_and_wait().
2421 write_len = iov_length(iov, nr_segs);
2422 if (mapping_mapped(mapping))
2423 unmap_mapping_range(mapping, offset, write_len, 0);
2426 retval = filemap_write_and_wait(mapping);
2428 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2430 if (rw == WRITE && mapping->nrpages) {
2431 pgoff_t end = (offset + write_len - 1)
2432 >> PAGE_CACHE_SHIFT;
2433 int err = invalidate_inode_pages2_range(mapping,
2434 offset >> PAGE_CACHE_SHIFT, end);