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/module.h>
13 #include <linux/slab.h>
14 #include <linux/compiler.h>
16 #include <linux/uaccess.h>
17 #include <linux/aio.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/cpuset.h>
37 * FIXME: remove all knowledge of the buffer layer from the core VM
39 #include <linux/buffer_head.h> /* for generic_osync_inode */
44 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
45 loff_t offset, unsigned long nr_segs);
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * Shared mappings now work. 15.8.1995 Bruno.
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
64 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
68 * ->i_mmap_lock (truncate->unmap_mapping_range)
72 * ->page_table_lock or pte_lock (various, mainly in memory.c)
73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 * ->lock_page (access_process_vm)
82 * ->i_alloc_sem (various)
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->dcache_lock (proc_pid_lookup)
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
113 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
115 void __remove_from_page_cache(struct page *page)
117 struct address_space *mapping = page->mapping;
119 radix_tree_delete(&mapping->page_tree, page->index);
120 page->mapping = NULL;
122 __dec_zone_page_state(page, NR_FILE_PAGES);
125 void remove_from_page_cache(struct page *page)
127 struct address_space *mapping = page->mapping;
129 BUG_ON(!PageLocked(page));
131 write_lock_irq(&mapping->tree_lock);
132 __remove_from_page_cache(page);
133 write_unlock_irq(&mapping->tree_lock);
136 static int sync_page(void *word)
138 struct address_space *mapping;
141 page = container_of((unsigned long *)word, struct page, flags);
144 * page_mapping() is being called without PG_locked held.
145 * Some knowledge of the state and use of the page is used to
146 * reduce the requirements down to a memory barrier.
147 * The danger here is of a stale page_mapping() return value
148 * indicating a struct address_space different from the one it's
149 * associated with when it is associated with one.
150 * After smp_mb(), it's either the correct page_mapping() for
151 * the page, or an old page_mapping() and the page's own
152 * page_mapping() has gone NULL.
153 * The ->sync_page() address_space operation must tolerate
154 * page_mapping() going NULL. By an amazing coincidence,
155 * this comes about because none of the users of the page
156 * in the ->sync_page() methods make essential use of the
157 * page_mapping(), merely passing the page down to the backing
158 * device's unplug functions when it's non-NULL, which in turn
159 * ignore it for all cases but swap, where only page_private(page) is
160 * of interest. When page_mapping() does go NULL, the entire
161 * call stack gracefully ignores the page and returns.
165 mapping = page_mapping(page);
166 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
167 mapping->a_ops->sync_page(page);
173 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
174 * @mapping: address space structure to write
175 * @start: offset in bytes where the range starts
176 * @end: offset in bytes where the range ends (inclusive)
177 * @sync_mode: enable synchronous operation
179 * Start writeback against all of a mapping's dirty pages that lie
180 * within the byte offsets <start, end> inclusive.
182 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
183 * opposed to a regular memory cleansing writeback. The difference between
184 * these two operations is that if a dirty page/buffer is encountered, it must
185 * be waited upon, and not just skipped over.
187 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
188 loff_t end, int sync_mode)
191 struct writeback_control wbc = {
192 .sync_mode = sync_mode,
193 .nr_to_write = mapping->nrpages * 2,
194 .range_start = start,
198 if (!mapping_cap_writeback_dirty(mapping))
201 ret = do_writepages(mapping, &wbc);
205 static inline int __filemap_fdatawrite(struct address_space *mapping,
208 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
211 int filemap_fdatawrite(struct address_space *mapping)
213 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
215 EXPORT_SYMBOL(filemap_fdatawrite);
217 static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
220 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
224 * filemap_flush - mostly a non-blocking flush
225 * @mapping: target address_space
227 * This is a mostly non-blocking flush. Not suitable for data-integrity
228 * purposes - I/O may not be started against all dirty pages.
230 int filemap_flush(struct address_space *mapping)
232 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
234 EXPORT_SYMBOL(filemap_flush);
237 * wait_on_page_writeback_range - wait for writeback to complete
238 * @mapping: target address_space
239 * @start: beginning page index
240 * @end: ending page index
242 * Wait for writeback to complete against pages indexed by start->end
245 int wait_on_page_writeback_range(struct address_space *mapping,
246 pgoff_t start, pgoff_t end)
256 pagevec_init(&pvec, 0);
258 while ((index <= end) &&
259 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
260 PAGECACHE_TAG_WRITEBACK,
261 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
264 for (i = 0; i < nr_pages; i++) {
265 struct page *page = pvec.pages[i];
267 /* until radix tree lookup accepts end_index */
268 if (page->index > end)
271 wait_on_page_writeback(page);
275 pagevec_release(&pvec);
279 /* Check for outstanding write errors */
280 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
282 if (test_and_clear_bit(AS_EIO, &mapping->flags))
289 * sync_page_range - write and wait on all pages in the passed range
290 * @inode: target inode
291 * @mapping: target address_space
292 * @pos: beginning offset in pages to write
293 * @count: number of bytes to write
295 * Write and wait upon all the pages in the passed range. This is a "data
296 * integrity" operation. It waits upon in-flight writeout before starting and
297 * waiting upon new writeout. If there was an IO error, return it.
299 * We need to re-take i_mutex during the generic_osync_inode list walk because
300 * it is otherwise livelockable.
302 int sync_page_range(struct inode *inode, struct address_space *mapping,
303 loff_t pos, loff_t count)
305 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
306 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
309 if (!mapping_cap_writeback_dirty(mapping) || !count)
311 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
313 mutex_lock(&inode->i_mutex);
314 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
315 mutex_unlock(&inode->i_mutex);
318 ret = wait_on_page_writeback_range(mapping, start, end);
321 EXPORT_SYMBOL(sync_page_range);
324 * sync_page_range_nolock
325 * @inode: target inode
326 * @mapping: target address_space
327 * @pos: beginning offset in pages to write
328 * @count: number of bytes to write
330 * Note: Holding i_mutex across sync_page_range_nolock is not a good idea
331 * as it forces O_SYNC writers to different parts of the same file
332 * to be serialised right until io completion.
334 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
335 loff_t pos, loff_t count)
337 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
338 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
341 if (!mapping_cap_writeback_dirty(mapping) || !count)
343 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
345 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
347 ret = wait_on_page_writeback_range(mapping, start, end);
350 EXPORT_SYMBOL(sync_page_range_nolock);
353 * filemap_fdatawait - wait for all under-writeback pages to complete
354 * @mapping: address space structure to wait for
356 * Walk the list of under-writeback pages of the given address space
357 * and wait for all of them.
359 int filemap_fdatawait(struct address_space *mapping)
361 loff_t i_size = i_size_read(mapping->host);
366 return wait_on_page_writeback_range(mapping, 0,
367 (i_size - 1) >> PAGE_CACHE_SHIFT);
369 EXPORT_SYMBOL(filemap_fdatawait);
371 int filemap_write_and_wait(struct address_space *mapping)
375 if (mapping->nrpages) {
376 err = filemap_fdatawrite(mapping);
378 * Even if the above returned error, the pages may be
379 * written partially (e.g. -ENOSPC), so we wait for it.
380 * But the -EIO is special case, it may indicate the worst
381 * thing (e.g. bug) happened, so we avoid waiting for it.
384 int err2 = filemap_fdatawait(mapping);
391 EXPORT_SYMBOL(filemap_write_and_wait);
394 * filemap_write_and_wait_range - write out & wait on a file range
395 * @mapping: the address_space for the pages
396 * @lstart: offset in bytes where the range starts
397 * @lend: offset in bytes where the range ends (inclusive)
399 * Write out and wait upon file offsets lstart->lend, inclusive.
401 * Note that `lend' is inclusive (describes the last byte to be written) so
402 * that this function can be used to write to the very end-of-file (end = -1).
404 int filemap_write_and_wait_range(struct address_space *mapping,
405 loff_t lstart, loff_t lend)
409 if (mapping->nrpages) {
410 err = __filemap_fdatawrite_range(mapping, lstart, lend,
412 /* See comment of filemap_write_and_wait() */
414 int err2 = wait_on_page_writeback_range(mapping,
415 lstart >> PAGE_CACHE_SHIFT,
416 lend >> PAGE_CACHE_SHIFT);
425 * add_to_page_cache - add newly allocated pagecache pages
427 * @mapping: the page's address_space
428 * @offset: page index
429 * @gfp_mask: page allocation mode
431 * This function is used to add newly allocated pagecache pages;
432 * the page is new, so we can just run SetPageLocked() against it.
433 * The other page state flags were set by rmqueue().
435 * This function does not add the page to the LRU. The caller must do that.
437 int add_to_page_cache(struct page *page, struct address_space *mapping,
438 pgoff_t offset, gfp_t gfp_mask)
440 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
443 write_lock_irq(&mapping->tree_lock);
444 error = radix_tree_insert(&mapping->page_tree, offset, page);
446 page_cache_get(page);
448 page->mapping = mapping;
449 page->index = offset;
451 __inc_zone_page_state(page, NR_FILE_PAGES);
453 write_unlock_irq(&mapping->tree_lock);
454 radix_tree_preload_end();
458 EXPORT_SYMBOL(add_to_page_cache);
460 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
461 pgoff_t offset, gfp_t gfp_mask)
463 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
470 struct page *page_cache_alloc(struct address_space *x)
472 if (cpuset_do_page_mem_spread()) {
473 int n = cpuset_mem_spread_node();
474 return alloc_pages_node(n, mapping_gfp_mask(x), 0);
476 return alloc_pages(mapping_gfp_mask(x), 0);
478 EXPORT_SYMBOL(page_cache_alloc);
480 struct page *page_cache_alloc_cold(struct address_space *x)
482 if (cpuset_do_page_mem_spread()) {
483 int n = cpuset_mem_spread_node();
484 return alloc_pages_node(n, mapping_gfp_mask(x)|__GFP_COLD, 0);
486 return alloc_pages(mapping_gfp_mask(x)|__GFP_COLD, 0);
488 EXPORT_SYMBOL(page_cache_alloc_cold);
491 static int __sleep_on_page_lock(void *word)
498 * In order to wait for pages to become available there must be
499 * waitqueues associated with pages. By using a hash table of
500 * waitqueues where the bucket discipline is to maintain all
501 * waiters on the same queue and wake all when any of the pages
502 * become available, and for the woken contexts to check to be
503 * sure the appropriate page became available, this saves space
504 * at a cost of "thundering herd" phenomena during rare hash
507 static wait_queue_head_t *page_waitqueue(struct page *page)
509 const struct zone *zone = page_zone(page);
511 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
514 static inline void wake_up_page(struct page *page, int bit)
516 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
519 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
521 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
523 if (test_bit(bit_nr, &page->flags))
524 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
525 TASK_UNINTERRUPTIBLE);
527 EXPORT_SYMBOL(wait_on_page_bit);
530 * unlock_page - unlock a locked page
533 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
534 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
535 * mechananism between PageLocked pages and PageWriteback pages is shared.
536 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
538 * The first mb is necessary to safely close the critical section opened by the
539 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
540 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
541 * parallel wait_on_page_locked()).
543 void fastcall unlock_page(struct page *page)
545 smp_mb__before_clear_bit();
546 if (!TestClearPageLocked(page))
548 smp_mb__after_clear_bit();
549 wake_up_page(page, PG_locked);
551 EXPORT_SYMBOL(unlock_page);
554 * end_page_writeback - end writeback against a page
557 void end_page_writeback(struct page *page)
559 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
560 if (!test_clear_page_writeback(page))
563 smp_mb__after_clear_bit();
564 wake_up_page(page, PG_writeback);
566 EXPORT_SYMBOL(end_page_writeback);
569 * __lock_page - get a lock on the page, assuming we need to sleep to get it
570 * @page: the page to lock
572 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
573 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
574 * chances are that on the second loop, the block layer's plug list is empty,
575 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
577 void fastcall __lock_page(struct page *page)
579 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
581 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
582 TASK_UNINTERRUPTIBLE);
584 EXPORT_SYMBOL(__lock_page);
587 * Variant of lock_page that does not require the caller to hold a reference
588 * on the page's mapping.
590 void fastcall __lock_page_nosync(struct page *page)
592 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
593 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
594 TASK_UNINTERRUPTIBLE);
598 * find_get_page - find and get a page reference
599 * @mapping: the address_space to search
600 * @offset: the page index
602 * Is there a pagecache struct page at the given (mapping, offset) tuple?
603 * If yes, increment its refcount and return it; if no, return NULL.
605 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
609 read_lock_irq(&mapping->tree_lock);
610 page = radix_tree_lookup(&mapping->page_tree, offset);
612 page_cache_get(page);
613 read_unlock_irq(&mapping->tree_lock);
616 EXPORT_SYMBOL(find_get_page);
619 * find_trylock_page - find and lock a page
620 * @mapping: the address_space to search
621 * @offset: the page index
623 * Same as find_get_page(), but trylock it instead of incrementing the count.
625 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
629 read_lock_irq(&mapping->tree_lock);
630 page = radix_tree_lookup(&mapping->page_tree, offset);
631 if (page && TestSetPageLocked(page))
633 read_unlock_irq(&mapping->tree_lock);
636 EXPORT_SYMBOL(find_trylock_page);
639 * find_lock_page - locate, pin and lock a pagecache page
640 * @mapping: the address_space to search
641 * @offset: the page index
643 * Locates the desired pagecache page, locks it, increments its reference
644 * count and returns its address.
646 * Returns zero if the page was not present. find_lock_page() may sleep.
648 struct page *find_lock_page(struct address_space *mapping,
649 unsigned long offset)
653 read_lock_irq(&mapping->tree_lock);
655 page = radix_tree_lookup(&mapping->page_tree, offset);
657 page_cache_get(page);
658 if (TestSetPageLocked(page)) {
659 read_unlock_irq(&mapping->tree_lock);
661 read_lock_irq(&mapping->tree_lock);
663 /* Has the page been truncated while we slept? */
664 if (unlikely(page->mapping != mapping ||
665 page->index != offset)) {
667 page_cache_release(page);
672 read_unlock_irq(&mapping->tree_lock);
675 EXPORT_SYMBOL(find_lock_page);
678 * find_or_create_page - locate or add a pagecache page
679 * @mapping: the page's address_space
680 * @index: the page's index into the mapping
681 * @gfp_mask: page allocation mode
683 * Locates a page in the pagecache. If the page is not present, a new page
684 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
685 * LRU list. The returned page is locked and has its reference count
688 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
691 * find_or_create_page() returns the desired page's address, or zero on
694 struct page *find_or_create_page(struct address_space *mapping,
695 unsigned long index, gfp_t gfp_mask)
697 struct page *page, *cached_page = NULL;
700 page = find_lock_page(mapping, index);
703 cached_page = alloc_page(gfp_mask);
707 err = add_to_page_cache_lru(cached_page, mapping,
712 } else if (err == -EEXIST)
716 page_cache_release(cached_page);
719 EXPORT_SYMBOL(find_or_create_page);
722 * find_get_pages - gang pagecache lookup
723 * @mapping: The address_space to search
724 * @start: The starting page index
725 * @nr_pages: The maximum number of pages
726 * @pages: Where the resulting pages are placed
728 * find_get_pages() will search for and return a group of up to
729 * @nr_pages pages in the mapping. The pages are placed at @pages.
730 * find_get_pages() takes a reference against the returned pages.
732 * The search returns a group of mapping-contiguous pages with ascending
733 * indexes. There may be holes in the indices due to not-present pages.
735 * find_get_pages() returns the number of pages which were found.
737 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
738 unsigned int nr_pages, struct page **pages)
743 read_lock_irq(&mapping->tree_lock);
744 ret = radix_tree_gang_lookup(&mapping->page_tree,
745 (void **)pages, start, nr_pages);
746 for (i = 0; i < ret; i++)
747 page_cache_get(pages[i]);
748 read_unlock_irq(&mapping->tree_lock);
753 * find_get_pages_contig - gang contiguous pagecache lookup
754 * @mapping: The address_space to search
755 * @index: The starting page index
756 * @nr_pages: The maximum number of pages
757 * @pages: Where the resulting pages are placed
759 * find_get_pages_contig() works exactly like find_get_pages(), except
760 * that the returned number of pages are guaranteed to be contiguous.
762 * find_get_pages_contig() returns the number of pages which were found.
764 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
765 unsigned int nr_pages, struct page **pages)
770 read_lock_irq(&mapping->tree_lock);
771 ret = radix_tree_gang_lookup(&mapping->page_tree,
772 (void **)pages, index, nr_pages);
773 for (i = 0; i < ret; i++) {
774 if (pages[i]->mapping == NULL || pages[i]->index != index)
777 page_cache_get(pages[i]);
780 read_unlock_irq(&mapping->tree_lock);
785 * find_get_pages_tag - find and return pages that match @tag
786 * @mapping: the address_space to search
787 * @index: the starting page index
788 * @tag: the tag index
789 * @nr_pages: the maximum number of pages
790 * @pages: where the resulting pages are placed
792 * Like find_get_pages, except we only return pages which are tagged with
793 * @tag. We update @index to index the next page for the traversal.
795 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
796 int tag, unsigned int nr_pages, struct page **pages)
801 read_lock_irq(&mapping->tree_lock);
802 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
803 (void **)pages, *index, nr_pages, tag);
804 for (i = 0; i < ret; i++)
805 page_cache_get(pages[i]);
807 *index = pages[ret - 1]->index + 1;
808 read_unlock_irq(&mapping->tree_lock);
813 * grab_cache_page_nowait - returns locked page at given index in given cache
814 * @mapping: target address_space
815 * @index: the page index
817 * Same as grab_cache_page, but do not wait if the page is unavailable.
818 * This is intended for speculative data generators, where the data can
819 * be regenerated if the page couldn't be grabbed. This routine should
820 * be safe to call while holding the lock for another page.
822 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
823 * and deadlock against the caller's locked page.
826 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
828 struct page *page = find_get_page(mapping, index);
832 if (!TestSetPageLocked(page))
834 page_cache_release(page);
837 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
838 page = alloc_pages(gfp_mask, 0);
839 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
840 page_cache_release(page);
845 EXPORT_SYMBOL(grab_cache_page_nowait);
848 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
849 * a _large_ part of the i/o request. Imagine the worst scenario:
851 * ---R__________________________________________B__________
852 * ^ reading here ^ bad block(assume 4k)
854 * read(R) => miss => readahead(R...B) => media error => frustrating retries
855 * => failing the whole request => read(R) => read(R+1) =>
856 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
857 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
858 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
860 * It is going insane. Fix it by quickly scaling down the readahead size.
862 static void shrink_readahead_size_eio(struct file *filp,
863 struct file_ra_state *ra)
872 * do_generic_mapping_read - generic file read routine
873 * @mapping: address_space to be read
874 * @_ra: file's readahead state
875 * @filp: the file to read
876 * @ppos: current file position
877 * @desc: read_descriptor
878 * @actor: read method
880 * This is a generic file read routine, and uses the
881 * mapping->a_ops->readpage() function for the actual low-level stuff.
883 * This is really ugly. But the goto's actually try to clarify some
884 * of the logic when it comes to error handling etc.
886 * Note the struct file* is only passed for the use of readpage.
889 void do_generic_mapping_read(struct address_space *mapping,
890 struct file_ra_state *_ra,
893 read_descriptor_t *desc,
896 struct inode *inode = mapping->host;
898 unsigned long end_index;
899 unsigned long offset;
900 unsigned long last_index;
901 unsigned long next_index;
902 unsigned long prev_index;
904 struct page *cached_page;
906 struct file_ra_state ra = *_ra;
909 index = *ppos >> PAGE_CACHE_SHIFT;
911 prev_index = ra.prev_page;
912 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
913 offset = *ppos & ~PAGE_CACHE_MASK;
915 isize = i_size_read(inode);
919 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
922 unsigned long nr, ret;
924 /* nr is the maximum number of bytes to copy from this page */
925 nr = PAGE_CACHE_SIZE;
926 if (index >= end_index) {
927 if (index > end_index)
929 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
937 if (index == next_index)
938 next_index = page_cache_readahead(mapping, &ra, filp,
939 index, last_index - index);
942 page = find_get_page(mapping, index);
943 if (unlikely(page == NULL)) {
944 handle_ra_miss(mapping, &ra, index);
947 if (!PageUptodate(page))
948 goto page_not_up_to_date;
951 /* If users can be writing to this page using arbitrary
952 * virtual addresses, take care about potential aliasing
953 * before reading the page on the kernel side.
955 if (mapping_writably_mapped(mapping))
956 flush_dcache_page(page);
959 * When (part of) the same page is read multiple times
960 * in succession, only mark it as accessed the first time.
962 if (prev_index != index)
963 mark_page_accessed(page);
967 * Ok, we have the page, and it's up-to-date, so
968 * now we can copy it to user space...
970 * The actor routine returns how many bytes were actually used..
971 * NOTE! This may not be the same as how much of a user buffer
972 * we filled up (we may be padding etc), so we can only update
973 * "pos" here (the actor routine has to update the user buffer
974 * pointers and the remaining count).
976 ret = actor(desc, page, offset, nr);
978 index += offset >> PAGE_CACHE_SHIFT;
979 offset &= ~PAGE_CACHE_MASK;
981 page_cache_release(page);
982 if (ret == nr && desc->count)
987 /* Get exclusive access to the page ... */
990 /* Did it get truncated before we got the lock? */
991 if (!page->mapping) {
993 page_cache_release(page);
997 /* Did somebody else fill it already? */
998 if (PageUptodate(page)) {
1004 /* Start the actual read. The read will unlock the page. */
1005 error = mapping->a_ops->readpage(filp, page);
1007 if (unlikely(error)) {
1008 if (error == AOP_TRUNCATED_PAGE) {
1009 page_cache_release(page);
1012 goto readpage_error;
1015 if (!PageUptodate(page)) {
1017 if (!PageUptodate(page)) {
1018 if (page->mapping == NULL) {
1020 * invalidate_inode_pages got it
1023 page_cache_release(page);
1028 shrink_readahead_size_eio(filp, &ra);
1029 goto readpage_error;
1035 * i_size must be checked after we have done ->readpage.
1037 * Checking i_size after the readpage allows us to calculate
1038 * the correct value for "nr", which means the zero-filled
1039 * part of the page is not copied back to userspace (unless
1040 * another truncate extends the file - this is desired though).
1042 isize = i_size_read(inode);
1043 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1044 if (unlikely(!isize || index > end_index)) {
1045 page_cache_release(page);
1049 /* nr is the maximum number of bytes to copy from this page */
1050 nr = PAGE_CACHE_SIZE;
1051 if (index == end_index) {
1052 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1054 page_cache_release(page);
1062 /* UHHUH! A synchronous read error occurred. Report it */
1063 desc->error = error;
1064 page_cache_release(page);
1069 * Ok, it wasn't cached, so we need to create a new
1073 cached_page = page_cache_alloc_cold(mapping);
1075 desc->error = -ENOMEM;
1079 error = add_to_page_cache_lru(cached_page, mapping,
1082 if (error == -EEXIST)
1084 desc->error = error;
1095 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1097 page_cache_release(cached_page);
1099 file_accessed(filp);
1101 EXPORT_SYMBOL(do_generic_mapping_read);
1103 int file_read_actor(read_descriptor_t *desc, struct page *page,
1104 unsigned long offset, unsigned long size)
1107 unsigned long left, count = desc->count;
1113 * Faults on the destination of a read are common, so do it before
1116 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1117 kaddr = kmap_atomic(page, KM_USER0);
1118 left = __copy_to_user_inatomic(desc->arg.buf,
1119 kaddr + offset, size);
1120 kunmap_atomic(kaddr, KM_USER0);
1125 /* Do it the slow way */
1127 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1132 desc->error = -EFAULT;
1135 desc->count = count - size;
1136 desc->written += size;
1137 desc->arg.buf += size;
1142 * __generic_file_aio_read - generic filesystem read routine
1143 * @iocb: kernel I/O control block
1144 * @iov: io vector request
1145 * @nr_segs: number of segments in the iovec
1146 * @ppos: current file position
1148 * This is the "read()" routine for all filesystems
1149 * that can use the page cache directly.
1152 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1153 unsigned long nr_segs, loff_t *ppos)
1155 struct file *filp = iocb->ki_filp;
1161 for (seg = 0; seg < nr_segs; seg++) {
1162 const struct iovec *iv = &iov[seg];
1165 * If any segment has a negative length, or the cumulative
1166 * length ever wraps negative then return -EINVAL.
1168 count += iv->iov_len;
1169 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1171 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1176 count -= iv->iov_len; /* This segment is no good */
1180 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1181 if (filp->f_flags & O_DIRECT) {
1182 loff_t pos = *ppos, size;
1183 struct address_space *mapping;
1184 struct inode *inode;
1186 mapping = filp->f_mapping;
1187 inode = mapping->host;
1190 goto out; /* skip atime */
1191 size = i_size_read(inode);
1193 retval = generic_file_direct_IO(READ, iocb,
1195 if (retval > 0 && !is_sync_kiocb(iocb))
1196 retval = -EIOCBQUEUED;
1198 *ppos = pos + retval;
1200 if (likely(retval != 0)) {
1201 file_accessed(filp);
1208 for (seg = 0; seg < nr_segs; seg++) {
1209 read_descriptor_t desc;
1212 desc.arg.buf = iov[seg].iov_base;
1213 desc.count = iov[seg].iov_len;
1214 if (desc.count == 0)
1217 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1218 retval += desc.written;
1220 retval = retval ?: desc.error;
1228 EXPORT_SYMBOL(__generic_file_aio_read);
1231 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1233 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1235 BUG_ON(iocb->ki_pos != pos);
1236 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1238 EXPORT_SYMBOL(generic_file_aio_read);
1241 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1243 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1247 init_sync_kiocb(&kiocb, filp);
1248 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1249 if (-EIOCBQUEUED == ret)
1250 ret = wait_on_sync_kiocb(&kiocb);
1253 EXPORT_SYMBOL(generic_file_read);
1255 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1258 unsigned long count = desc->count;
1259 struct file *file = desc->arg.data;
1264 written = file->f_op->sendpage(file, page, offset,
1265 size, &file->f_pos, size<count);
1267 desc->error = written;
1270 desc->count = count - written;
1271 desc->written += written;
1275 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1276 size_t count, read_actor_t actor, void *target)
1278 read_descriptor_t desc;
1285 desc.arg.data = target;
1288 do_generic_file_read(in_file, ppos, &desc, actor);
1290 return desc.written;
1293 EXPORT_SYMBOL(generic_file_sendfile);
1296 do_readahead(struct address_space *mapping, struct file *filp,
1297 unsigned long index, unsigned long nr)
1299 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1302 force_page_cache_readahead(mapping, filp, index,
1303 max_sane_readahead(nr));
1307 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1315 if (file->f_mode & FMODE_READ) {
1316 struct address_space *mapping = file->f_mapping;
1317 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1318 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1319 unsigned long len = end - start + 1;
1320 ret = do_readahead(mapping, file, start, len);
1328 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1330 * page_cache_read - adds requested page to the page cache if not already there
1331 * @file: file to read
1332 * @offset: page index
1334 * This adds the requested page to the page cache if it isn't already there,
1335 * and schedules an I/O to read in its contents from disk.
1337 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1339 struct address_space *mapping = file->f_mapping;
1344 page = page_cache_alloc_cold(mapping);
1348 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1350 ret = mapping->a_ops->readpage(file, page);
1351 else if (ret == -EEXIST)
1352 ret = 0; /* losing race to add is OK */
1354 page_cache_release(page);
1356 } while (ret == AOP_TRUNCATED_PAGE);
1361 #define MMAP_LOTSAMISS (100)
1364 * filemap_nopage - read in file data for page fault handling
1365 * @area: the applicable vm_area
1366 * @address: target address to read in
1367 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1369 * filemap_nopage() is invoked via the vma operations vector for a
1370 * mapped memory region to read in file data during a page fault.
1372 * The goto's are kind of ugly, but this streamlines the normal case of having
1373 * it in the page cache, and handles the special cases reasonably without
1374 * having a lot of duplicated code.
1376 struct page *filemap_nopage(struct vm_area_struct *area,
1377 unsigned long address, int *type)
1380 struct file *file = area->vm_file;
1381 struct address_space *mapping = file->f_mapping;
1382 struct file_ra_state *ra = &file->f_ra;
1383 struct inode *inode = mapping->host;
1385 unsigned long size, pgoff;
1386 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1388 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1391 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1393 goto outside_data_content;
1395 /* If we don't want any read-ahead, don't bother */
1396 if (VM_RandomReadHint(area))
1397 goto no_cached_page;
1400 * The readahead code wants to be told about each and every page
1401 * so it can build and shrink its windows appropriately
1403 * For sequential accesses, we use the generic readahead logic.
1405 if (VM_SequentialReadHint(area))
1406 page_cache_readahead(mapping, ra, file, pgoff, 1);
1409 * Do we have something in the page cache already?
1412 page = find_get_page(mapping, pgoff);
1414 unsigned long ra_pages;
1416 if (VM_SequentialReadHint(area)) {
1417 handle_ra_miss(mapping, ra, pgoff);
1418 goto no_cached_page;
1423 * Do we miss much more than hit in this file? If so,
1424 * stop bothering with read-ahead. It will only hurt.
1426 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1427 goto no_cached_page;
1430 * To keep the pgmajfault counter straight, we need to
1431 * check did_readaround, as this is an inner loop.
1433 if (!did_readaround) {
1434 majmin = VM_FAULT_MAJOR;
1435 count_vm_event(PGMAJFAULT);
1438 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1442 if (pgoff > ra_pages / 2)
1443 start = pgoff - ra_pages / 2;
1444 do_page_cache_readahead(mapping, file, start, ra_pages);
1446 page = find_get_page(mapping, pgoff);
1448 goto no_cached_page;
1451 if (!did_readaround)
1455 * Ok, found a page in the page cache, now we need to check
1456 * that it's up-to-date.
1458 if (!PageUptodate(page))
1459 goto page_not_uptodate;
1463 * Found the page and have a reference on it.
1465 mark_page_accessed(page);
1470 outside_data_content:
1472 * An external ptracer can access pages that normally aren't
1475 if (area->vm_mm == current->mm)
1477 /* Fall through to the non-read-ahead case */
1480 * We're only likely to ever get here if MADV_RANDOM is in
1483 error = page_cache_read(file, pgoff);
1487 * The page we want has now been added to the page cache.
1488 * In the unlikely event that someone removed it in the
1489 * meantime, we'll just come back here and read it again.
1495 * An error return from page_cache_read can result if the
1496 * system is low on memory, or a problem occurs while trying
1499 if (error == -ENOMEM)
1504 if (!did_readaround) {
1505 majmin = VM_FAULT_MAJOR;
1506 count_vm_event(PGMAJFAULT);
1510 /* Did it get unhashed while we waited for it? */
1511 if (!page->mapping) {
1513 page_cache_release(page);
1517 /* Did somebody else get it up-to-date? */
1518 if (PageUptodate(page)) {
1523 error = mapping->a_ops->readpage(file, page);
1525 wait_on_page_locked(page);
1526 if (PageUptodate(page))
1528 } else if (error == AOP_TRUNCATED_PAGE) {
1529 page_cache_release(page);
1534 * Umm, take care of errors if the page isn't up-to-date.
1535 * Try to re-read it _once_. We do this synchronously,
1536 * because there really aren't any performance issues here
1537 * and we need to check for errors.
1541 /* Somebody truncated the page on us? */
1542 if (!page->mapping) {
1544 page_cache_release(page);
1548 /* Somebody else successfully read it in? */
1549 if (PageUptodate(page)) {
1553 ClearPageError(page);
1554 error = mapping->a_ops->readpage(file, page);
1556 wait_on_page_locked(page);
1557 if (PageUptodate(page))
1559 } else if (error == AOP_TRUNCATED_PAGE) {
1560 page_cache_release(page);
1565 * Things didn't work out. Return zero to tell the
1566 * mm layer so, possibly freeing the page cache page first.
1568 shrink_readahead_size_eio(file, ra);
1569 page_cache_release(page);
1572 EXPORT_SYMBOL(filemap_nopage);
1574 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1577 struct address_space *mapping = file->f_mapping;
1582 * Do we have something in the page cache already?
1585 page = find_get_page(mapping, pgoff);
1589 goto no_cached_page;
1593 * Ok, found a page in the page cache, now we need to check
1594 * that it's up-to-date.
1596 if (!PageUptodate(page)) {
1598 page_cache_release(page);
1601 goto page_not_uptodate;
1606 * Found the page and have a reference on it.
1608 mark_page_accessed(page);
1612 error = page_cache_read(file, pgoff);
1615 * The page we want has now been added to the page cache.
1616 * In the unlikely event that someone removed it in the
1617 * meantime, we'll just come back here and read it again.
1623 * An error return from page_cache_read can result if the
1624 * system is low on memory, or a problem occurs while trying
1632 /* Did it get truncated while we waited for it? */
1633 if (!page->mapping) {
1638 /* Did somebody else get it up-to-date? */
1639 if (PageUptodate(page)) {
1644 error = mapping->a_ops->readpage(file, page);
1646 wait_on_page_locked(page);
1647 if (PageUptodate(page))
1649 } else if (error == AOP_TRUNCATED_PAGE) {
1650 page_cache_release(page);
1655 * Umm, take care of errors if the page isn't up-to-date.
1656 * Try to re-read it _once_. We do this synchronously,
1657 * because there really aren't any performance issues here
1658 * and we need to check for errors.
1662 /* Somebody truncated the page on us? */
1663 if (!page->mapping) {
1667 /* Somebody else successfully read it in? */
1668 if (PageUptodate(page)) {
1673 ClearPageError(page);
1674 error = mapping->a_ops->readpage(file, page);
1676 wait_on_page_locked(page);
1677 if (PageUptodate(page))
1679 } else if (error == AOP_TRUNCATED_PAGE) {
1680 page_cache_release(page);
1685 * Things didn't work out. Return zero to tell the
1686 * mm layer so, possibly freeing the page cache page first.
1689 page_cache_release(page);
1694 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1695 unsigned long len, pgprot_t prot, unsigned long pgoff,
1698 struct file *file = vma->vm_file;
1699 struct address_space *mapping = file->f_mapping;
1700 struct inode *inode = mapping->host;
1702 struct mm_struct *mm = vma->vm_mm;
1707 force_page_cache_readahead(mapping, vma->vm_file,
1708 pgoff, len >> PAGE_CACHE_SHIFT);
1711 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1712 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1715 page = filemap_getpage(file, pgoff, nonblock);
1717 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1718 * done in shmem_populate calling shmem_getpage */
1719 if (!page && !nonblock)
1723 err = install_page(mm, vma, addr, page, prot);
1725 page_cache_release(page);
1728 } else if (vma->vm_flags & VM_NONLINEAR) {
1729 /* No page was found just because we can't read it in now (being
1730 * here implies nonblock != 0), but the page may exist, so set
1731 * the PTE to fault it in later. */
1732 err = install_file_pte(mm, vma, addr, pgoff, prot);
1745 EXPORT_SYMBOL(filemap_populate);
1747 struct vm_operations_struct generic_file_vm_ops = {
1748 .nopage = filemap_nopage,
1749 .populate = filemap_populate,
1752 /* This is used for a general mmap of a disk file */
1754 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1756 struct address_space *mapping = file->f_mapping;
1758 if (!mapping->a_ops->readpage)
1760 file_accessed(file);
1761 vma->vm_ops = &generic_file_vm_ops;
1766 * This is for filesystems which do not implement ->writepage.
1768 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1770 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1772 return generic_file_mmap(file, vma);
1775 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1779 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1783 #endif /* CONFIG_MMU */
1785 EXPORT_SYMBOL(generic_file_mmap);
1786 EXPORT_SYMBOL(generic_file_readonly_mmap);
1788 static inline struct page *__read_cache_page(struct address_space *mapping,
1789 unsigned long index,
1790 int (*filler)(void *,struct page*),
1793 struct page *page, *cached_page = NULL;
1796 page = find_get_page(mapping, index);
1799 cached_page = page_cache_alloc_cold(mapping);
1801 return ERR_PTR(-ENOMEM);
1803 err = add_to_page_cache_lru(cached_page, mapping,
1808 /* Presumably ENOMEM for radix tree node */
1809 page_cache_release(cached_page);
1810 return ERR_PTR(err);
1814 err = filler(data, page);
1816 page_cache_release(page);
1817 page = ERR_PTR(err);
1821 page_cache_release(cached_page);
1826 * read_cache_page - read into page cache, fill it if needed
1827 * @mapping: the page's address_space
1828 * @index: the page index
1829 * @filler: function to perform the read
1830 * @data: destination for read data
1832 * Read into the page cache. If a page already exists,
1833 * and PageUptodate() is not set, try to fill the page.
1835 struct page *read_cache_page(struct address_space *mapping,
1836 unsigned long index,
1837 int (*filler)(void *,struct page*),
1844 page = __read_cache_page(mapping, index, filler, data);
1847 mark_page_accessed(page);
1848 if (PageUptodate(page))
1852 if (!page->mapping) {
1854 page_cache_release(page);
1857 if (PageUptodate(page)) {
1861 err = filler(data, page);
1863 page_cache_release(page);
1864 page = ERR_PTR(err);
1869 EXPORT_SYMBOL(read_cache_page);
1872 * If the page was newly created, increment its refcount and add it to the
1873 * caller's lru-buffering pagevec. This function is specifically for
1874 * generic_file_write().
1876 static inline struct page *
1877 __grab_cache_page(struct address_space *mapping, unsigned long index,
1878 struct page **cached_page, struct pagevec *lru_pvec)
1883 page = find_lock_page(mapping, index);
1885 if (!*cached_page) {
1886 *cached_page = page_cache_alloc(mapping);
1890 err = add_to_page_cache(*cached_page, mapping,
1895 page = *cached_page;
1896 page_cache_get(page);
1897 if (!pagevec_add(lru_pvec, page))
1898 __pagevec_lru_add(lru_pvec);
1899 *cached_page = NULL;
1906 * The logic we want is
1908 * if suid or (sgid and xgrp)
1911 int remove_suid(struct dentry *dentry)
1913 mode_t mode = dentry->d_inode->i_mode;
1917 /* suid always must be killed */
1918 if (unlikely(mode & S_ISUID))
1919 kill = ATTR_KILL_SUID;
1922 * sgid without any exec bits is just a mandatory locking mark; leave
1923 * it alone. If some exec bits are set, it's a real sgid; kill it.
1925 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1926 kill |= ATTR_KILL_SGID;
1928 if (unlikely(kill && !capable(CAP_FSETID))) {
1929 struct iattr newattrs;
1931 newattrs.ia_valid = ATTR_FORCE | kill;
1932 result = notify_change(dentry, &newattrs);
1936 EXPORT_SYMBOL(remove_suid);
1939 __filemap_copy_from_user_iovec_inatomic(char *vaddr,
1940 const struct iovec *iov, size_t base, size_t bytes)
1942 size_t copied = 0, left = 0;
1945 char __user *buf = iov->iov_base + base;
1946 int copy = min(bytes, iov->iov_len - base);
1949 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1958 return copied - left;
1962 * Performs necessary checks before doing a write
1964 * Can adjust writing position or amount of bytes to write.
1965 * Returns appropriate error code that caller should return or
1966 * zero in case that write should be allowed.
1968 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1970 struct inode *inode = file->f_mapping->host;
1971 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1973 if (unlikely(*pos < 0))
1977 /* FIXME: this is for backwards compatibility with 2.4 */
1978 if (file->f_flags & O_APPEND)
1979 *pos = i_size_read(inode);
1981 if (limit != RLIM_INFINITY) {
1982 if (*pos >= limit) {
1983 send_sig(SIGXFSZ, current, 0);
1986 if (*count > limit - (typeof(limit))*pos) {
1987 *count = limit - (typeof(limit))*pos;
1995 if (unlikely(*pos + *count > MAX_NON_LFS &&
1996 !(file->f_flags & O_LARGEFILE))) {
1997 if (*pos >= MAX_NON_LFS) {
1998 send_sig(SIGXFSZ, current, 0);
2001 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2002 *count = MAX_NON_LFS - (unsigned long)*pos;
2007 * Are we about to exceed the fs block limit ?
2009 * If we have written data it becomes a short write. If we have
2010 * exceeded without writing data we send a signal and return EFBIG.
2011 * Linus frestrict idea will clean these up nicely..
2013 if (likely(!isblk)) {
2014 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2015 if (*count || *pos > inode->i_sb->s_maxbytes) {
2016 send_sig(SIGXFSZ, current, 0);
2019 /* zero-length writes at ->s_maxbytes are OK */
2022 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2023 *count = inode->i_sb->s_maxbytes - *pos;
2026 if (bdev_read_only(I_BDEV(inode)))
2028 isize = i_size_read(inode);
2029 if (*pos >= isize) {
2030 if (*count || *pos > isize)
2034 if (*pos + *count > isize)
2035 *count = isize - *pos;
2039 EXPORT_SYMBOL(generic_write_checks);
2042 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2043 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2044 size_t count, size_t ocount)
2046 struct file *file = iocb->ki_filp;
2047 struct address_space *mapping = file->f_mapping;
2048 struct inode *inode = mapping->host;
2051 if (count != ocount)
2052 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2054 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2056 loff_t end = pos + written;
2057 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2058 i_size_write(inode, end);
2059 mark_inode_dirty(inode);
2065 * Sync the fs metadata but not the minor inode changes and
2066 * of course not the data as we did direct DMA for the IO.
2067 * i_mutex is held, which protects generic_osync_inode() from
2070 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2071 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2075 if (written == count && !is_sync_kiocb(iocb))
2076 written = -EIOCBQUEUED;
2079 EXPORT_SYMBOL(generic_file_direct_write);
2082 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2083 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2084 size_t count, ssize_t written)
2086 struct file *file = iocb->ki_filp;
2087 struct address_space * mapping = file->f_mapping;
2088 const struct address_space_operations *a_ops = mapping->a_ops;
2089 struct inode *inode = mapping->host;
2092 struct page *cached_page = NULL;
2094 struct pagevec lru_pvec;
2095 const struct iovec *cur_iov = iov; /* current iovec */
2096 size_t iov_base = 0; /* offset in the current iovec */
2099 pagevec_init(&lru_pvec, 0);
2102 * handle partial DIO write. Adjust cur_iov if needed.
2104 if (likely(nr_segs == 1))
2105 buf = iov->iov_base + written;
2107 filemap_set_next_iovec(&cur_iov, &iov_base, written);
2108 buf = cur_iov->iov_base + iov_base;
2112 unsigned long index;
2113 unsigned long offset;
2116 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2117 index = pos >> PAGE_CACHE_SHIFT;
2118 bytes = PAGE_CACHE_SIZE - offset;
2120 /* Limit the size of the copy to the caller's write size */
2121 bytes = min(bytes, count);
2124 * Limit the size of the copy to that of the current segment,
2125 * because fault_in_pages_readable() doesn't know how to walk
2128 bytes = min(bytes, cur_iov->iov_len - iov_base);
2131 * Bring in the user page that we will copy from _first_.
2132 * Otherwise there's a nasty deadlock on copying from the
2133 * same page as we're writing to, without it being marked
2136 fault_in_pages_readable(buf, bytes);
2138 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2144 if (unlikely(bytes == 0)) {
2147 goto zero_length_segment;
2150 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2151 if (unlikely(status)) {
2152 loff_t isize = i_size_read(inode);
2154 if (status != AOP_TRUNCATED_PAGE)
2156 page_cache_release(page);
2157 if (status == AOP_TRUNCATED_PAGE)
2160 * prepare_write() may have instantiated a few blocks
2161 * outside i_size. Trim these off again.
2163 if (pos + bytes > isize)
2164 vmtruncate(inode, isize);
2167 if (likely(nr_segs == 1))
2168 copied = filemap_copy_from_user(page, offset,
2171 copied = filemap_copy_from_user_iovec(page, offset,
2172 cur_iov, iov_base, bytes);
2173 flush_dcache_page(page);
2174 status = a_ops->commit_write(file, page, offset, offset+bytes);
2175 if (status == AOP_TRUNCATED_PAGE) {
2176 page_cache_release(page);
2179 zero_length_segment:
2180 if (likely(copied >= 0)) {
2189 if (unlikely(nr_segs > 1)) {
2190 filemap_set_next_iovec(&cur_iov,
2193 buf = cur_iov->iov_base +
2200 if (unlikely(copied != bytes))
2204 mark_page_accessed(page);
2205 page_cache_release(page);
2208 balance_dirty_pages_ratelimited(mapping);
2214 page_cache_release(cached_page);
2217 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2219 if (likely(status >= 0)) {
2220 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2221 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2222 status = generic_osync_inode(inode, mapping,
2223 OSYNC_METADATA|OSYNC_DATA);
2228 * If we get here for O_DIRECT writes then we must have fallen through
2229 * to buffered writes (block instantiation inside i_size). So we sync
2230 * the file data here, to try to honour O_DIRECT expectations.
2232 if (unlikely(file->f_flags & O_DIRECT) && written)
2233 status = filemap_write_and_wait(mapping);
2235 pagevec_lru_add(&lru_pvec);
2236 return written ? written : status;
2238 EXPORT_SYMBOL(generic_file_buffered_write);
2241 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2242 unsigned long nr_segs, loff_t *ppos)
2244 struct file *file = iocb->ki_filp;
2245 const struct address_space * mapping = file->f_mapping;
2246 size_t ocount; /* original count */
2247 size_t count; /* after file limit checks */
2248 struct inode *inode = mapping->host;
2255 for (seg = 0; seg < nr_segs; seg++) {
2256 const struct iovec *iv = &iov[seg];
2259 * If any segment has a negative length, or the cumulative
2260 * length ever wraps negative then return -EINVAL.
2262 ocount += iv->iov_len;
2263 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2265 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2270 ocount -= iv->iov_len; /* This segment is no good */
2277 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2279 /* We can write back this queue in page reclaim */
2280 current->backing_dev_info = mapping->backing_dev_info;
2283 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2290 err = remove_suid(file->f_dentry);
2294 file_update_time(file);
2296 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2297 if (unlikely(file->f_flags & O_DIRECT)) {
2298 written = generic_file_direct_write(iocb, iov,
2299 &nr_segs, pos, ppos, count, ocount);
2300 if (written < 0 || written == count)
2303 * direct-io write to a hole: fall through to buffered I/O
2304 * for completing the rest of the request.
2310 written = generic_file_buffered_write(iocb, iov, nr_segs,
2311 pos, ppos, count, written);
2313 current->backing_dev_info = NULL;
2314 return written ? written : err;
2316 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2319 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2320 unsigned long nr_segs, loff_t *ppos)
2322 struct file *file = iocb->ki_filp;
2323 struct address_space *mapping = file->f_mapping;
2324 struct inode *inode = mapping->host;
2328 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2330 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2333 err = sync_page_range_nolock(inode, mapping, pos, ret);
2341 __generic_file_write_nolock(struct file *file, const struct iovec *iov,
2342 unsigned long nr_segs, loff_t *ppos)
2347 init_sync_kiocb(&kiocb, file);
2348 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2349 if (ret == -EIOCBQUEUED)
2350 ret = wait_on_sync_kiocb(&kiocb);
2355 generic_file_write_nolock(struct file *file, const struct iovec *iov,
2356 unsigned long nr_segs, loff_t *ppos)
2361 init_sync_kiocb(&kiocb, file);
2362 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2363 if (-EIOCBQUEUED == ret)
2364 ret = wait_on_sync_kiocb(&kiocb);
2367 EXPORT_SYMBOL(generic_file_write_nolock);
2369 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2370 size_t count, loff_t pos)
2372 struct file *file = iocb->ki_filp;
2373 struct address_space *mapping = file->f_mapping;
2374 struct inode *inode = mapping->host;
2376 struct iovec local_iov = { .iov_base = (void __user *)buf,
2379 BUG_ON(iocb->ki_pos != pos);
2381 mutex_lock(&inode->i_mutex);
2382 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2384 mutex_unlock(&inode->i_mutex);
2386 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2389 err = sync_page_range(inode, mapping, pos, ret);
2395 EXPORT_SYMBOL(generic_file_aio_write);
2397 ssize_t generic_file_write(struct file *file, const char __user *buf,
2398 size_t count, loff_t *ppos)
2400 struct address_space *mapping = file->f_mapping;
2401 struct inode *inode = mapping->host;
2403 struct iovec local_iov = { .iov_base = (void __user *)buf,
2406 mutex_lock(&inode->i_mutex);
2407 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
2408 mutex_unlock(&inode->i_mutex);
2410 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2413 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2419 EXPORT_SYMBOL(generic_file_write);
2421 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2422 unsigned long nr_segs, loff_t *ppos)
2427 init_sync_kiocb(&kiocb, filp);
2428 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2429 if (-EIOCBQUEUED == ret)
2430 ret = wait_on_sync_kiocb(&kiocb);
2433 EXPORT_SYMBOL(generic_file_readv);
2435 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2436 unsigned long nr_segs, loff_t *ppos)
2438 struct address_space *mapping = file->f_mapping;
2439 struct inode *inode = mapping->host;
2442 mutex_lock(&inode->i_mutex);
2443 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
2444 mutex_unlock(&inode->i_mutex);
2446 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2449 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2455 EXPORT_SYMBOL(generic_file_writev);
2458 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2459 * went wrong during pagecache shootdown.
2462 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2463 loff_t offset, unsigned long nr_segs)
2465 struct file *file = iocb->ki_filp;
2466 struct address_space *mapping = file->f_mapping;
2468 size_t write_len = 0;
2471 * If it's a write, unmap all mmappings of the file up-front. This
2472 * will cause any pte dirty bits to be propagated into the pageframes
2473 * for the subsequent filemap_write_and_wait().
2476 write_len = iov_length(iov, nr_segs);
2477 if (mapping_mapped(mapping))
2478 unmap_mapping_range(mapping, offset, write_len, 0);
2481 retval = filemap_write_and_wait(mapping);
2483 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2485 if (rw == WRITE && mapping->nrpages) {
2486 pgoff_t end = (offset + write_len - 1)
2487 >> PAGE_CACHE_SHIFT;
2488 int err = invalidate_inode_pages2_range(mapping,
2489 offset >> PAGE_CACHE_SHIFT, end);