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/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/backing-dev.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/cpuset.h>
35 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
39 * FIXME: remove all knowledge of the buffer layer from the core VM
41 #include <linux/buffer_head.h> /* for generic_osync_inode */
46 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
47 loff_t offset, unsigned long nr_segs);
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
53 * Shared mappings now work. 15.8.1995 Bruno.
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
64 * ->i_mmap_lock (vmtruncate)
65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
71 * ->i_mmap_lock (truncate->unmap_mapping_range)
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
79 * ->lock_page (access_process_vm)
81 * ->i_mutex (generic_file_buffered_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
85 * ->i_alloc_sem (various)
88 * ->sb_lock (fs/fs-writeback.c)
89 * ->mapping->tree_lock (__sync_single_inode)
92 * ->anon_vma.lock (vma_adjust)
95 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
97 * ->page_table_lock or pte_lock
98 * ->swap_lock (try_to_unmap_one)
99 * ->private_lock (try_to_unmap_one)
100 * ->tree_lock (try_to_unmap_one)
101 * ->zone.lru_lock (follow_page->mark_page_accessed)
102 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
103 * ->private_lock (page_remove_rmap->set_page_dirty)
104 * ->tree_lock (page_remove_rmap->set_page_dirty)
105 * ->inode_lock (page_remove_rmap->set_page_dirty)
106 * ->inode_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
110 * ->dcache_lock (proc_pid_lookup)
114 * Remove a page from the page cache and free it. Caller has to make
115 * sure the page is locked and that nobody else uses it - or that usage
116 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
118 void __remove_from_page_cache(struct page *page)
120 struct address_space *mapping = page->mapping;
122 radix_tree_delete(&mapping->page_tree, page->index);
123 page->mapping = NULL;
125 __dec_zone_page_state(page, NR_FILE_PAGES);
126 BUG_ON(page_mapped(page));
129 void remove_from_page_cache(struct page *page)
131 struct address_space *mapping = page->mapping;
133 BUG_ON(!PageLocked(page));
135 write_lock_irq(&mapping->tree_lock);
136 __remove_from_page_cache(page);
137 write_unlock_irq(&mapping->tree_lock);
140 static int sync_page(void *word)
142 struct address_space *mapping;
145 page = container_of((unsigned long *)word, struct page, flags);
148 * page_mapping() is being called without PG_locked held.
149 * Some knowledge of the state and use of the page is used to
150 * reduce the requirements down to a memory barrier.
151 * The danger here is of a stale page_mapping() return value
152 * indicating a struct address_space different from the one it's
153 * associated with when it is associated with one.
154 * After smp_mb(), it's either the correct page_mapping() for
155 * the page, or an old page_mapping() and the page's own
156 * page_mapping() has gone NULL.
157 * The ->sync_page() address_space operation must tolerate
158 * page_mapping() going NULL. By an amazing coincidence,
159 * this comes about because none of the users of the page
160 * in the ->sync_page() methods make essential use of the
161 * page_mapping(), merely passing the page down to the backing
162 * device's unplug functions when it's non-NULL, which in turn
163 * ignore it for all cases but swap, where only page_private(page) is
164 * of interest. When page_mapping() does go NULL, the entire
165 * call stack gracefully ignores the page and returns.
169 mapping = page_mapping(page);
170 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
171 mapping->a_ops->sync_page(page);
176 static int sync_page_killable(void *word)
179 return fatal_signal_pending(current) ? -EINTR : 0;
183 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
184 * @mapping: address space structure to write
185 * @start: offset in bytes where the range starts
186 * @end: offset in bytes where the range ends (inclusive)
187 * @sync_mode: enable synchronous operation
189 * Start writeback against all of a mapping's dirty pages that lie
190 * within the byte offsets <start, end> inclusive.
192 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
193 * opposed to a regular memory cleansing writeback. The difference between
194 * these two operations is that if a dirty page/buffer is encountered, it must
195 * be waited upon, and not just skipped over.
197 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
198 loff_t end, int sync_mode)
201 struct writeback_control wbc = {
202 .sync_mode = sync_mode,
203 .nr_to_write = mapping->nrpages * 2,
204 .range_start = start,
208 if (!mapping_cap_writeback_dirty(mapping))
211 ret = do_writepages(mapping, &wbc);
215 static inline int __filemap_fdatawrite(struct address_space *mapping,
218 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
221 int filemap_fdatawrite(struct address_space *mapping)
223 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
225 EXPORT_SYMBOL(filemap_fdatawrite);
227 static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
230 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
234 * filemap_flush - mostly a non-blocking flush
235 * @mapping: target address_space
237 * This is a mostly non-blocking flush. Not suitable for data-integrity
238 * purposes - I/O may not be started against all dirty pages.
240 int filemap_flush(struct address_space *mapping)
242 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
244 EXPORT_SYMBOL(filemap_flush);
247 * wait_on_page_writeback_range - wait for writeback to complete
248 * @mapping: target address_space
249 * @start: beginning page index
250 * @end: ending page index
252 * Wait for writeback to complete against pages indexed by start->end
255 int wait_on_page_writeback_range(struct address_space *mapping,
256 pgoff_t start, pgoff_t end)
266 pagevec_init(&pvec, 0);
268 while ((index <= end) &&
269 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
270 PAGECACHE_TAG_WRITEBACK,
271 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
274 for (i = 0; i < nr_pages; i++) {
275 struct page *page = pvec.pages[i];
277 /* until radix tree lookup accepts end_index */
278 if (page->index > end)
281 wait_on_page_writeback(page);
285 pagevec_release(&pvec);
289 /* Check for outstanding write errors */
290 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
292 if (test_and_clear_bit(AS_EIO, &mapping->flags))
299 * sync_page_range - write and wait on all pages in the passed range
300 * @inode: target inode
301 * @mapping: target address_space
302 * @pos: beginning offset in pages to write
303 * @count: number of bytes to write
305 * Write and wait upon all the pages in the passed range. This is a "data
306 * integrity" operation. It waits upon in-flight writeout before starting and
307 * waiting upon new writeout. If there was an IO error, return it.
309 * We need to re-take i_mutex during the generic_osync_inode list walk because
310 * it is otherwise livelockable.
312 int sync_page_range(struct inode *inode, struct address_space *mapping,
313 loff_t pos, loff_t count)
315 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
316 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
319 if (!mapping_cap_writeback_dirty(mapping) || !count)
321 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
323 mutex_lock(&inode->i_mutex);
324 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
325 mutex_unlock(&inode->i_mutex);
328 ret = wait_on_page_writeback_range(mapping, start, end);
331 EXPORT_SYMBOL(sync_page_range);
334 * sync_page_range_nolock
335 * @inode: target inode
336 * @mapping: target address_space
337 * @pos: beginning offset in pages to write
338 * @count: number of bytes to write
340 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
341 * as it forces O_SYNC writers to different parts of the same file
342 * to be serialised right until io completion.
344 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
345 loff_t pos, loff_t count)
347 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
348 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
351 if (!mapping_cap_writeback_dirty(mapping) || !count)
353 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
355 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
357 ret = wait_on_page_writeback_range(mapping, start, end);
360 EXPORT_SYMBOL(sync_page_range_nolock);
363 * filemap_fdatawait - wait for all under-writeback pages to complete
364 * @mapping: address space structure to wait for
366 * Walk the list of under-writeback pages of the given address space
367 * and wait for all of them.
369 int filemap_fdatawait(struct address_space *mapping)
371 loff_t i_size = i_size_read(mapping->host);
376 return wait_on_page_writeback_range(mapping, 0,
377 (i_size - 1) >> PAGE_CACHE_SHIFT);
379 EXPORT_SYMBOL(filemap_fdatawait);
381 int filemap_write_and_wait(struct address_space *mapping)
385 if (mapping->nrpages) {
386 err = filemap_fdatawrite(mapping);
388 * Even if the above returned error, the pages may be
389 * written partially (e.g. -ENOSPC), so we wait for it.
390 * But the -EIO is special case, it may indicate the worst
391 * thing (e.g. bug) happened, so we avoid waiting for it.
394 int err2 = filemap_fdatawait(mapping);
401 EXPORT_SYMBOL(filemap_write_and_wait);
404 * filemap_write_and_wait_range - write out & wait on a file range
405 * @mapping: the address_space for the pages
406 * @lstart: offset in bytes where the range starts
407 * @lend: offset in bytes where the range ends (inclusive)
409 * Write out and wait upon file offsets lstart->lend, inclusive.
411 * Note that `lend' is inclusive (describes the last byte to be written) so
412 * that this function can be used to write to the very end-of-file (end = -1).
414 int filemap_write_and_wait_range(struct address_space *mapping,
415 loff_t lstart, loff_t lend)
419 if (mapping->nrpages) {
420 err = __filemap_fdatawrite_range(mapping, lstart, lend,
422 /* See comment of filemap_write_and_wait() */
424 int err2 = wait_on_page_writeback_range(mapping,
425 lstart >> PAGE_CACHE_SHIFT,
426 lend >> PAGE_CACHE_SHIFT);
435 * add_to_page_cache - add newly allocated pagecache pages
437 * @mapping: the page's address_space
438 * @offset: page index
439 * @gfp_mask: page allocation mode
441 * This function is used to add newly allocated pagecache pages;
442 * the page is new, so we can just run SetPageLocked() against it.
443 * The other page state flags were set by rmqueue().
445 * This function does not add the page to the LRU. The caller must do that.
447 int add_to_page_cache(struct page *page, struct address_space *mapping,
448 pgoff_t offset, gfp_t gfp_mask)
450 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
453 write_lock_irq(&mapping->tree_lock);
454 error = radix_tree_insert(&mapping->page_tree, offset, page);
456 page_cache_get(page);
458 page->mapping = mapping;
459 page->index = offset;
461 __inc_zone_page_state(page, NR_FILE_PAGES);
463 write_unlock_irq(&mapping->tree_lock);
464 radix_tree_preload_end();
468 EXPORT_SYMBOL(add_to_page_cache);
470 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
471 pgoff_t offset, gfp_t gfp_mask)
473 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
480 struct page *__page_cache_alloc(gfp_t gfp)
482 if (cpuset_do_page_mem_spread()) {
483 int n = cpuset_mem_spread_node();
484 return alloc_pages_node(n, gfp, 0);
486 return alloc_pages(gfp, 0);
488 EXPORT_SYMBOL(__page_cache_alloc);
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);
586 int fastcall __lock_page_killable(struct page *page)
588 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
590 return __wait_on_bit_lock(page_waitqueue(page), &wait,
591 sync_page_killable, TASK_KILLABLE);
595 * Variant of lock_page that does not require the caller to hold a reference
596 * on the page's mapping.
598 void fastcall __lock_page_nosync(struct page *page)
600 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
601 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
602 TASK_UNINTERRUPTIBLE);
606 * find_get_page - find and get a page reference
607 * @mapping: the address_space to search
608 * @offset: the page index
610 * Is there a pagecache struct page at the given (mapping, offset) tuple?
611 * If yes, increment its refcount and return it; if no, return NULL.
613 struct page * find_get_page(struct address_space *mapping, pgoff_t offset)
617 read_lock_irq(&mapping->tree_lock);
618 page = radix_tree_lookup(&mapping->page_tree, offset);
620 page_cache_get(page);
621 read_unlock_irq(&mapping->tree_lock);
624 EXPORT_SYMBOL(find_get_page);
627 * find_lock_page - locate, pin and lock a pagecache page
628 * @mapping: the address_space to search
629 * @offset: the page index
631 * Locates the desired pagecache page, locks it, increments its reference
632 * count and returns its address.
634 * Returns zero if the page was not present. find_lock_page() may sleep.
636 struct page *find_lock_page(struct address_space *mapping,
642 read_lock_irq(&mapping->tree_lock);
643 page = radix_tree_lookup(&mapping->page_tree, offset);
645 page_cache_get(page);
646 if (TestSetPageLocked(page)) {
647 read_unlock_irq(&mapping->tree_lock);
650 /* Has the page been truncated while we slept? */
651 if (unlikely(page->mapping != mapping)) {
653 page_cache_release(page);
656 VM_BUG_ON(page->index != offset);
660 read_unlock_irq(&mapping->tree_lock);
664 EXPORT_SYMBOL(find_lock_page);
667 * find_or_create_page - locate or add a pagecache page
668 * @mapping: the page's address_space
669 * @index: the page's index into the mapping
670 * @gfp_mask: page allocation mode
672 * Locates a page in the pagecache. If the page is not present, a new page
673 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
674 * LRU list. The returned page is locked and has its reference count
677 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
680 * find_or_create_page() returns the desired page's address, or zero on
683 struct page *find_or_create_page(struct address_space *mapping,
684 pgoff_t index, gfp_t gfp_mask)
689 page = find_lock_page(mapping, index);
691 page = __page_cache_alloc(gfp_mask);
694 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
696 page_cache_release(page);
704 EXPORT_SYMBOL(find_or_create_page);
707 * find_get_pages - gang pagecache lookup
708 * @mapping: The address_space to search
709 * @start: The starting page index
710 * @nr_pages: The maximum number of pages
711 * @pages: Where the resulting pages are placed
713 * find_get_pages() will search for and return a group of up to
714 * @nr_pages pages in the mapping. The pages are placed at @pages.
715 * find_get_pages() takes a reference against the returned pages.
717 * The search returns a group of mapping-contiguous pages with ascending
718 * indexes. There may be holes in the indices due to not-present pages.
720 * find_get_pages() returns the number of pages which were found.
722 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
723 unsigned int nr_pages, struct page **pages)
728 read_lock_irq(&mapping->tree_lock);
729 ret = radix_tree_gang_lookup(&mapping->page_tree,
730 (void **)pages, start, nr_pages);
731 for (i = 0; i < ret; i++)
732 page_cache_get(pages[i]);
733 read_unlock_irq(&mapping->tree_lock);
738 * find_get_pages_contig - gang contiguous pagecache lookup
739 * @mapping: The address_space to search
740 * @index: The starting page index
741 * @nr_pages: The maximum number of pages
742 * @pages: Where the resulting pages are placed
744 * find_get_pages_contig() works exactly like find_get_pages(), except
745 * that the returned number of pages are guaranteed to be contiguous.
747 * find_get_pages_contig() returns the number of pages which were found.
749 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
750 unsigned int nr_pages, struct page **pages)
755 read_lock_irq(&mapping->tree_lock);
756 ret = radix_tree_gang_lookup(&mapping->page_tree,
757 (void **)pages, index, nr_pages);
758 for (i = 0; i < ret; i++) {
759 if (pages[i]->mapping == NULL || pages[i]->index != index)
762 page_cache_get(pages[i]);
765 read_unlock_irq(&mapping->tree_lock);
768 EXPORT_SYMBOL(find_get_pages_contig);
771 * find_get_pages_tag - find and return pages that match @tag
772 * @mapping: the address_space to search
773 * @index: the starting page index
774 * @tag: the tag index
775 * @nr_pages: the maximum number of pages
776 * @pages: where the resulting pages are placed
778 * Like find_get_pages, except we only return pages which are tagged with
779 * @tag. We update @index to index the next page for the traversal.
781 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
782 int tag, unsigned int nr_pages, struct page **pages)
787 read_lock_irq(&mapping->tree_lock);
788 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
789 (void **)pages, *index, nr_pages, tag);
790 for (i = 0; i < ret; i++)
791 page_cache_get(pages[i]);
793 *index = pages[ret - 1]->index + 1;
794 read_unlock_irq(&mapping->tree_lock);
797 EXPORT_SYMBOL(find_get_pages_tag);
800 * grab_cache_page_nowait - returns locked page at given index in given cache
801 * @mapping: target address_space
802 * @index: the page index
804 * Same as grab_cache_page(), but do not wait if the page is unavailable.
805 * This is intended for speculative data generators, where the data can
806 * be regenerated if the page couldn't be grabbed. This routine should
807 * be safe to call while holding the lock for another page.
809 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
810 * and deadlock against the caller's locked page.
813 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
815 struct page *page = find_get_page(mapping, index);
818 if (!TestSetPageLocked(page))
820 page_cache_release(page);
823 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
824 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
825 page_cache_release(page);
830 EXPORT_SYMBOL(grab_cache_page_nowait);
833 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
834 * a _large_ part of the i/o request. Imagine the worst scenario:
836 * ---R__________________________________________B__________
837 * ^ reading here ^ bad block(assume 4k)
839 * read(R) => miss => readahead(R...B) => media error => frustrating retries
840 * => failing the whole request => read(R) => read(R+1) =>
841 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
842 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
843 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
845 * It is going insane. Fix it by quickly scaling down the readahead size.
847 static void shrink_readahead_size_eio(struct file *filp,
848 struct file_ra_state *ra)
857 * do_generic_mapping_read - generic file read routine
858 * @mapping: address_space to be read
859 * @ra: file's readahead state
860 * @filp: the file to read
861 * @ppos: current file position
862 * @desc: read_descriptor
863 * @actor: read method
865 * This is a generic file read routine, and uses the
866 * mapping->a_ops->readpage() function for the actual low-level stuff.
868 * This is really ugly. But the goto's actually try to clarify some
869 * of the logic when it comes to error handling etc.
871 * Note the struct file* is only passed for the use of readpage.
874 void do_generic_mapping_read(struct address_space *mapping,
875 struct file_ra_state *ra,
878 read_descriptor_t *desc,
881 struct inode *inode = mapping->host;
885 unsigned long offset; /* offset into pagecache page */
886 unsigned int prev_offset;
889 index = *ppos >> PAGE_CACHE_SHIFT;
890 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
891 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
892 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
893 offset = *ppos & ~PAGE_CACHE_MASK;
899 unsigned long nr, ret;
903 page = find_get_page(mapping, index);
905 page_cache_sync_readahead(mapping,
907 index, last_index - index);
908 page = find_get_page(mapping, index);
909 if (unlikely(page == NULL))
912 if (PageReadahead(page)) {
913 page_cache_async_readahead(mapping,
915 index, last_index - index);
917 if (!PageUptodate(page))
918 goto page_not_up_to_date;
921 * i_size must be checked after we know the page is Uptodate.
923 * Checking i_size after the check allows us to calculate
924 * the correct value for "nr", which means the zero-filled
925 * part of the page is not copied back to userspace (unless
926 * another truncate extends the file - this is desired though).
929 isize = i_size_read(inode);
930 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
931 if (unlikely(!isize || index > end_index)) {
932 page_cache_release(page);
936 /* nr is the maximum number of bytes to copy from this page */
937 nr = PAGE_CACHE_SIZE;
938 if (index == end_index) {
939 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
941 page_cache_release(page);
947 /* If users can be writing to this page using arbitrary
948 * virtual addresses, take care about potential aliasing
949 * before reading the page on the kernel side.
951 if (mapping_writably_mapped(mapping))
952 flush_dcache_page(page);
955 * When a sequential read accesses a page several times,
956 * only mark it as accessed the first time.
958 if (prev_index != index || offset != prev_offset)
959 mark_page_accessed(page);
963 * Ok, we have the page, and it's up-to-date, so
964 * now we can copy it to user space...
966 * The actor routine returns how many bytes were actually used..
967 * NOTE! This may not be the same as how much of a user buffer
968 * we filled up (we may be padding etc), so we can only update
969 * "pos" here (the actor routine has to update the user buffer
970 * pointers and the remaining count).
972 ret = actor(desc, page, offset, nr);
974 index += offset >> PAGE_CACHE_SHIFT;
975 offset &= ~PAGE_CACHE_MASK;
976 prev_offset = offset;
978 page_cache_release(page);
979 if (ret == nr && desc->count)
984 /* Get exclusive access to the page ... */
985 if (lock_page_killable(page))
988 /* Did it get truncated before we got the lock? */
989 if (!page->mapping) {
991 page_cache_release(page);
995 /* Did somebody else fill it already? */
996 if (PageUptodate(page)) {
1002 /* Start the actual read. The read will unlock the page. */
1003 error = mapping->a_ops->readpage(filp, page);
1005 if (unlikely(error)) {
1006 if (error == AOP_TRUNCATED_PAGE) {
1007 page_cache_release(page);
1010 goto readpage_error;
1013 if (!PageUptodate(page)) {
1014 if (lock_page_killable(page))
1016 if (!PageUptodate(page)) {
1017 if (page->mapping == NULL) {
1019 * invalidate_inode_pages got it
1022 page_cache_release(page);
1026 shrink_readahead_size_eio(filp, ra);
1037 /* UHHUH! A synchronous read error occurred. Report it */
1038 desc->error = error;
1039 page_cache_release(page);
1044 * Ok, it wasn't cached, so we need to create a new
1047 page = page_cache_alloc_cold(mapping);
1049 desc->error = -ENOMEM;
1052 error = add_to_page_cache_lru(page, mapping,
1055 page_cache_release(page);
1056 if (error == -EEXIST)
1058 desc->error = error;
1065 ra->prev_pos = prev_index;
1066 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1067 ra->prev_pos |= prev_offset;
1069 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1071 file_accessed(filp);
1073 EXPORT_SYMBOL(do_generic_mapping_read);
1075 int file_read_actor(read_descriptor_t *desc, struct page *page,
1076 unsigned long offset, unsigned long size)
1079 unsigned long left, count = desc->count;
1085 * Faults on the destination of a read are common, so do it before
1088 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1089 kaddr = kmap_atomic(page, KM_USER0);
1090 left = __copy_to_user_inatomic(desc->arg.buf,
1091 kaddr + offset, size);
1092 kunmap_atomic(kaddr, KM_USER0);
1097 /* Do it the slow way */
1099 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1104 desc->error = -EFAULT;
1107 desc->count = count - size;
1108 desc->written += size;
1109 desc->arg.buf += size;
1114 * Performs necessary checks before doing a write
1115 * @iov: io vector request
1116 * @nr_segs: number of segments in the iovec
1117 * @count: number of bytes to write
1118 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1120 * Adjust number of segments and amount of bytes to write (nr_segs should be
1121 * properly initialized first). Returns appropriate error code that caller
1122 * should return or zero in case that write should be allowed.
1124 int generic_segment_checks(const struct iovec *iov,
1125 unsigned long *nr_segs, size_t *count, int access_flags)
1129 for (seg = 0; seg < *nr_segs; seg++) {
1130 const struct iovec *iv = &iov[seg];
1133 * If any segment has a negative length, or the cumulative
1134 * length ever wraps negative then return -EINVAL.
1137 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1139 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1144 cnt -= iv->iov_len; /* This segment is no good */
1150 EXPORT_SYMBOL(generic_segment_checks);
1153 * generic_file_aio_read - generic filesystem read routine
1154 * @iocb: kernel I/O control block
1155 * @iov: io vector request
1156 * @nr_segs: number of segments in the iovec
1157 * @pos: current file position
1159 * This is the "read()" routine for all filesystems
1160 * that can use the page cache directly.
1163 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1164 unsigned long nr_segs, loff_t pos)
1166 struct file *filp = iocb->ki_filp;
1170 loff_t *ppos = &iocb->ki_pos;
1173 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1177 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1178 if (filp->f_flags & O_DIRECT) {
1180 struct address_space *mapping;
1181 struct inode *inode;
1183 mapping = filp->f_mapping;
1184 inode = mapping->host;
1187 goto out; /* skip atime */
1188 size = i_size_read(inode);
1190 retval = generic_file_direct_IO(READ, iocb,
1193 *ppos = pos + retval;
1195 if (likely(retval != 0)) {
1196 file_accessed(filp);
1203 for (seg = 0; seg < nr_segs; seg++) {
1204 read_descriptor_t desc;
1207 desc.arg.buf = iov[seg].iov_base;
1208 desc.count = iov[seg].iov_len;
1209 if (desc.count == 0)
1212 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1213 retval += desc.written;
1215 retval = retval ?: desc.error;
1225 EXPORT_SYMBOL(generic_file_aio_read);
1228 do_readahead(struct address_space *mapping, struct file *filp,
1229 pgoff_t index, unsigned long nr)
1231 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1234 force_page_cache_readahead(mapping, filp, index,
1235 max_sane_readahead(nr));
1239 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1247 if (file->f_mode & FMODE_READ) {
1248 struct address_space *mapping = file->f_mapping;
1249 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1250 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1251 unsigned long len = end - start + 1;
1252 ret = do_readahead(mapping, file, start, len);
1261 * page_cache_read - adds requested page to the page cache if not already there
1262 * @file: file to read
1263 * @offset: page index
1265 * This adds the requested page to the page cache if it isn't already there,
1266 * and schedules an I/O to read in its contents from disk.
1268 static int fastcall page_cache_read(struct file * file, pgoff_t offset)
1270 struct address_space *mapping = file->f_mapping;
1275 page = page_cache_alloc_cold(mapping);
1279 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1281 ret = mapping->a_ops->readpage(file, page);
1282 else if (ret == -EEXIST)
1283 ret = 0; /* losing race to add is OK */
1285 page_cache_release(page);
1287 } while (ret == AOP_TRUNCATED_PAGE);
1292 #define MMAP_LOTSAMISS (100)
1295 * filemap_fault - read in file data for page fault handling
1296 * @vma: vma in which the fault was taken
1297 * @vmf: struct vm_fault containing details of the fault
1299 * filemap_fault() is invoked via the vma operations vector for a
1300 * mapped memory region to read in file data during a page fault.
1302 * The goto's are kind of ugly, but this streamlines the normal case of having
1303 * it in the page cache, and handles the special cases reasonably without
1304 * having a lot of duplicated code.
1306 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1309 struct file *file = vma->vm_file;
1310 struct address_space *mapping = file->f_mapping;
1311 struct file_ra_state *ra = &file->f_ra;
1312 struct inode *inode = mapping->host;
1315 int did_readaround = 0;
1318 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1319 if (vmf->pgoff >= size)
1320 return VM_FAULT_SIGBUS;
1322 /* If we don't want any read-ahead, don't bother */
1323 if (VM_RandomReadHint(vma))
1324 goto no_cached_page;
1327 * Do we have something in the page cache already?
1330 page = find_lock_page(mapping, vmf->pgoff);
1332 * For sequential accesses, we use the generic readahead logic.
1334 if (VM_SequentialReadHint(vma)) {
1336 page_cache_sync_readahead(mapping, ra, file,
1338 page = find_lock_page(mapping, vmf->pgoff);
1340 goto no_cached_page;
1342 if (PageReadahead(page)) {
1343 page_cache_async_readahead(mapping, ra, file, page,
1349 unsigned long ra_pages;
1354 * Do we miss much more than hit in this file? If so,
1355 * stop bothering with read-ahead. It will only hurt.
1357 if (ra->mmap_miss > MMAP_LOTSAMISS)
1358 goto no_cached_page;
1361 * To keep the pgmajfault counter straight, we need to
1362 * check did_readaround, as this is an inner loop.
1364 if (!did_readaround) {
1365 ret = VM_FAULT_MAJOR;
1366 count_vm_event(PGMAJFAULT);
1369 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1373 if (vmf->pgoff > ra_pages / 2)
1374 start = vmf->pgoff - ra_pages / 2;
1375 do_page_cache_readahead(mapping, file, start, ra_pages);
1377 page = find_lock_page(mapping, vmf->pgoff);
1379 goto no_cached_page;
1382 if (!did_readaround)
1386 * We have a locked page in the page cache, now we need to check
1387 * that it's up-to-date. If not, it is going to be due to an error.
1389 if (unlikely(!PageUptodate(page)))
1390 goto page_not_uptodate;
1392 /* Must recheck i_size under page lock */
1393 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1394 if (unlikely(vmf->pgoff >= size)) {
1396 page_cache_release(page);
1397 return VM_FAULT_SIGBUS;
1401 * Found the page and have a reference on it.
1403 mark_page_accessed(page);
1404 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
1406 return ret | VM_FAULT_LOCKED;
1410 * We're only likely to ever get here if MADV_RANDOM is in
1413 error = page_cache_read(file, vmf->pgoff);
1416 * The page we want has now been added to the page cache.
1417 * In the unlikely event that someone removed it in the
1418 * meantime, we'll just come back here and read it again.
1424 * An error return from page_cache_read can result if the
1425 * system is low on memory, or a problem occurs while trying
1428 if (error == -ENOMEM)
1429 return VM_FAULT_OOM;
1430 return VM_FAULT_SIGBUS;
1434 if (!did_readaround) {
1435 ret = VM_FAULT_MAJOR;
1436 count_vm_event(PGMAJFAULT);
1440 * Umm, take care of errors if the page isn't up-to-date.
1441 * Try to re-read it _once_. We do this synchronously,
1442 * because there really aren't any performance issues here
1443 * and we need to check for errors.
1445 ClearPageError(page);
1446 error = mapping->a_ops->readpage(file, page);
1447 page_cache_release(page);
1449 if (!error || error == AOP_TRUNCATED_PAGE)
1452 /* Things didn't work out. Return zero to tell the mm layer so. */
1453 shrink_readahead_size_eio(file, ra);
1454 return VM_FAULT_SIGBUS;
1456 EXPORT_SYMBOL(filemap_fault);
1458 struct vm_operations_struct generic_file_vm_ops = {
1459 .fault = filemap_fault,
1462 /* This is used for a general mmap of a disk file */
1464 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1466 struct address_space *mapping = file->f_mapping;
1468 if (!mapping->a_ops->readpage)
1470 file_accessed(file);
1471 vma->vm_ops = &generic_file_vm_ops;
1472 vma->vm_flags |= VM_CAN_NONLINEAR;
1477 * This is for filesystems which do not implement ->writepage.
1479 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1481 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1483 return generic_file_mmap(file, vma);
1486 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1490 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1494 #endif /* CONFIG_MMU */
1496 EXPORT_SYMBOL(generic_file_mmap);
1497 EXPORT_SYMBOL(generic_file_readonly_mmap);
1499 static struct page *__read_cache_page(struct address_space *mapping,
1501 int (*filler)(void *,struct page*),
1507 page = find_get_page(mapping, index);
1509 page = page_cache_alloc_cold(mapping);
1511 return ERR_PTR(-ENOMEM);
1512 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1513 if (unlikely(err)) {
1514 page_cache_release(page);
1517 /* Presumably ENOMEM for radix tree node */
1518 return ERR_PTR(err);
1520 err = filler(data, page);
1522 page_cache_release(page);
1523 page = ERR_PTR(err);
1530 * Same as read_cache_page, but don't wait for page to become unlocked
1531 * after submitting it to the filler.
1533 struct page *read_cache_page_async(struct address_space *mapping,
1535 int (*filler)(void *,struct page*),
1542 page = __read_cache_page(mapping, index, filler, data);
1545 if (PageUptodate(page))
1549 if (!page->mapping) {
1551 page_cache_release(page);
1554 if (PageUptodate(page)) {
1558 err = filler(data, page);
1560 page_cache_release(page);
1561 return ERR_PTR(err);
1564 mark_page_accessed(page);
1567 EXPORT_SYMBOL(read_cache_page_async);
1570 * read_cache_page - read into page cache, fill it if needed
1571 * @mapping: the page's address_space
1572 * @index: the page index
1573 * @filler: function to perform the read
1574 * @data: destination for read data
1576 * Read into the page cache. If a page already exists, and PageUptodate() is
1577 * not set, try to fill the page then wait for it to become unlocked.
1579 * If the page does not get brought uptodate, return -EIO.
1581 struct page *read_cache_page(struct address_space *mapping,
1583 int (*filler)(void *,struct page*),
1588 page = read_cache_page_async(mapping, index, filler, data);
1591 wait_on_page_locked(page);
1592 if (!PageUptodate(page)) {
1593 page_cache_release(page);
1594 page = ERR_PTR(-EIO);
1599 EXPORT_SYMBOL(read_cache_page);
1602 * The logic we want is
1604 * if suid or (sgid and xgrp)
1607 int should_remove_suid(struct dentry *dentry)
1609 mode_t mode = dentry->d_inode->i_mode;
1612 /* suid always must be killed */
1613 if (unlikely(mode & S_ISUID))
1614 kill = ATTR_KILL_SUID;
1617 * sgid without any exec bits is just a mandatory locking mark; leave
1618 * it alone. If some exec bits are set, it's a real sgid; kill it.
1620 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1621 kill |= ATTR_KILL_SGID;
1623 if (unlikely(kill && !capable(CAP_FSETID)))
1628 EXPORT_SYMBOL(should_remove_suid);
1630 int __remove_suid(struct dentry *dentry, int kill)
1632 struct iattr newattrs;
1634 newattrs.ia_valid = ATTR_FORCE | kill;
1635 return notify_change(dentry, &newattrs);
1638 int remove_suid(struct dentry *dentry)
1640 int killsuid = should_remove_suid(dentry);
1641 int killpriv = security_inode_need_killpriv(dentry);
1647 error = security_inode_killpriv(dentry);
1648 if (!error && killsuid)
1649 error = __remove_suid(dentry, killsuid);
1653 EXPORT_SYMBOL(remove_suid);
1655 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1656 const struct iovec *iov, size_t base, size_t bytes)
1658 size_t copied = 0, left = 0;
1661 char __user *buf = iov->iov_base + base;
1662 int copy = min(bytes, iov->iov_len - base);
1665 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1674 return copied - left;
1678 * Copy as much as we can into the page and return the number of bytes which
1679 * were sucessfully copied. If a fault is encountered then return the number of
1680 * bytes which were copied.
1682 size_t iov_iter_copy_from_user_atomic(struct page *page,
1683 struct iov_iter *i, unsigned long offset, size_t bytes)
1688 BUG_ON(!in_atomic());
1689 kaddr = kmap_atomic(page, KM_USER0);
1690 if (likely(i->nr_segs == 1)) {
1692 char __user *buf = i->iov->iov_base + i->iov_offset;
1693 left = __copy_from_user_inatomic_nocache(kaddr + offset,
1695 copied = bytes - left;
1697 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1698 i->iov, i->iov_offset, bytes);
1700 kunmap_atomic(kaddr, KM_USER0);
1704 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1707 * This has the same sideeffects and return value as
1708 * iov_iter_copy_from_user_atomic().
1709 * The difference is that it attempts to resolve faults.
1710 * Page must not be locked.
1712 size_t iov_iter_copy_from_user(struct page *page,
1713 struct iov_iter *i, unsigned long offset, size_t bytes)
1719 if (likely(i->nr_segs == 1)) {
1721 char __user *buf = i->iov->iov_base + i->iov_offset;
1722 left = __copy_from_user_nocache(kaddr + offset, buf, bytes);
1723 copied = bytes - left;
1725 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1726 i->iov, i->iov_offset, bytes);
1731 EXPORT_SYMBOL(iov_iter_copy_from_user);
1733 static void __iov_iter_advance_iov(struct iov_iter *i, size_t bytes)
1735 if (likely(i->nr_segs == 1)) {
1736 i->iov_offset += bytes;
1738 const struct iovec *iov = i->iov;
1739 size_t base = i->iov_offset;
1742 int copy = min(bytes, iov->iov_len - base);
1746 if (iov->iov_len == base) {
1752 i->iov_offset = base;
1756 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1758 BUG_ON(i->count < bytes);
1760 __iov_iter_advance_iov(i, bytes);
1763 EXPORT_SYMBOL(iov_iter_advance);
1766 * Fault in the first iovec of the given iov_iter, to a maximum length
1767 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1768 * accessed (ie. because it is an invalid address).
1770 * writev-intensive code may want this to prefault several iovecs -- that
1771 * would be possible (callers must not rely on the fact that _only_ the
1772 * first iovec will be faulted with the current implementation).
1774 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
1776 char __user *buf = i->iov->iov_base + i->iov_offset;
1777 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1778 return fault_in_pages_readable(buf, bytes);
1780 EXPORT_SYMBOL(iov_iter_fault_in_readable);
1783 * Return the count of just the current iov_iter segment.
1785 size_t iov_iter_single_seg_count(struct iov_iter *i)
1787 const struct iovec *iov = i->iov;
1788 if (i->nr_segs == 1)
1791 return min(i->count, iov->iov_len - i->iov_offset);
1793 EXPORT_SYMBOL(iov_iter_single_seg_count);
1796 * Performs necessary checks before doing a write
1798 * Can adjust writing position or amount of bytes to write.
1799 * Returns appropriate error code that caller should return or
1800 * zero in case that write should be allowed.
1802 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1804 struct inode *inode = file->f_mapping->host;
1805 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1807 if (unlikely(*pos < 0))
1811 /* FIXME: this is for backwards compatibility with 2.4 */
1812 if (file->f_flags & O_APPEND)
1813 *pos = i_size_read(inode);
1815 if (limit != RLIM_INFINITY) {
1816 if (*pos >= limit) {
1817 send_sig(SIGXFSZ, current, 0);
1820 if (*count > limit - (typeof(limit))*pos) {
1821 *count = limit - (typeof(limit))*pos;
1829 if (unlikely(*pos + *count > MAX_NON_LFS &&
1830 !(file->f_flags & O_LARGEFILE))) {
1831 if (*pos >= MAX_NON_LFS) {
1834 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1835 *count = MAX_NON_LFS - (unsigned long)*pos;
1840 * Are we about to exceed the fs block limit ?
1842 * If we have written data it becomes a short write. If we have
1843 * exceeded without writing data we send a signal and return EFBIG.
1844 * Linus frestrict idea will clean these up nicely..
1846 if (likely(!isblk)) {
1847 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1848 if (*count || *pos > inode->i_sb->s_maxbytes) {
1851 /* zero-length writes at ->s_maxbytes are OK */
1854 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1855 *count = inode->i_sb->s_maxbytes - *pos;
1859 if (bdev_read_only(I_BDEV(inode)))
1861 isize = i_size_read(inode);
1862 if (*pos >= isize) {
1863 if (*count || *pos > isize)
1867 if (*pos + *count > isize)
1868 *count = isize - *pos;
1875 EXPORT_SYMBOL(generic_write_checks);
1877 int pagecache_write_begin(struct file *file, struct address_space *mapping,
1878 loff_t pos, unsigned len, unsigned flags,
1879 struct page **pagep, void **fsdata)
1881 const struct address_space_operations *aops = mapping->a_ops;
1883 if (aops->write_begin) {
1884 return aops->write_begin(file, mapping, pos, len, flags,
1888 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1889 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1890 struct inode *inode = mapping->host;
1893 page = __grab_cache_page(mapping, index);
1898 if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) {
1900 * There is no way to resolve a short write situation
1901 * for a !Uptodate page (except by double copying in
1902 * the caller done by generic_perform_write_2copy).
1904 * Instead, we have to bring it uptodate here.
1906 ret = aops->readpage(file, page);
1907 page_cache_release(page);
1909 if (ret == AOP_TRUNCATED_PAGE)
1916 ret = aops->prepare_write(file, page, offset, offset+len);
1919 page_cache_release(page);
1920 if (pos + len > inode->i_size)
1921 vmtruncate(inode, inode->i_size);
1926 EXPORT_SYMBOL(pagecache_write_begin);
1928 int pagecache_write_end(struct file *file, struct address_space *mapping,
1929 loff_t pos, unsigned len, unsigned copied,
1930 struct page *page, void *fsdata)
1932 const struct address_space_operations *aops = mapping->a_ops;
1935 if (aops->write_end) {
1936 mark_page_accessed(page);
1937 ret = aops->write_end(file, mapping, pos, len, copied,
1940 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1941 struct inode *inode = mapping->host;
1943 flush_dcache_page(page);
1944 ret = aops->commit_write(file, page, offset, offset+len);
1946 mark_page_accessed(page);
1947 page_cache_release(page);
1950 if (pos + len > inode->i_size)
1951 vmtruncate(inode, inode->i_size);
1953 ret = min_t(size_t, copied, ret);
1960 EXPORT_SYMBOL(pagecache_write_end);
1963 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1964 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1965 size_t count, size_t ocount)
1967 struct file *file = iocb->ki_filp;
1968 struct address_space *mapping = file->f_mapping;
1969 struct inode *inode = mapping->host;
1972 if (count != ocount)
1973 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1975 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1977 loff_t end = pos + written;
1978 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1979 i_size_write(inode, end);
1980 mark_inode_dirty(inode);
1986 * Sync the fs metadata but not the minor inode changes and
1987 * of course not the data as we did direct DMA for the IO.
1988 * i_mutex is held, which protects generic_osync_inode() from
1989 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
1991 if ((written >= 0 || written == -EIOCBQUEUED) &&
1992 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
1993 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1999 EXPORT_SYMBOL(generic_file_direct_write);
2002 * Find or create a page at the given pagecache position. Return the locked
2003 * page. This function is specifically for buffered writes.
2005 struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index)
2010 page = find_lock_page(mapping, index);
2014 page = page_cache_alloc(mapping);
2017 status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
2018 if (unlikely(status)) {
2019 page_cache_release(page);
2020 if (status == -EEXIST)
2026 EXPORT_SYMBOL(__grab_cache_page);
2028 static ssize_t generic_perform_write_2copy(struct file *file,
2029 struct iov_iter *i, loff_t pos)
2031 struct address_space *mapping = file->f_mapping;
2032 const struct address_space_operations *a_ops = mapping->a_ops;
2033 struct inode *inode = mapping->host;
2035 ssize_t written = 0;
2038 struct page *src_page;
2040 pgoff_t index; /* Pagecache index for current page */
2041 unsigned long offset; /* Offset into pagecache page */
2042 unsigned long bytes; /* Bytes to write to page */
2043 size_t copied; /* Bytes copied from user */
2045 offset = (pos & (PAGE_CACHE_SIZE - 1));
2046 index = pos >> PAGE_CACHE_SHIFT;
2047 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2051 * a non-NULL src_page indicates that we're doing the
2052 * copy via get_user_pages and kmap.
2057 * Bring in the user page that we will copy from _first_.
2058 * Otherwise there's a nasty deadlock on copying from the
2059 * same page as we're writing to, without it being marked
2062 * Not only is this an optimisation, but it is also required
2063 * to check that the address is actually valid, when atomic
2064 * usercopies are used, below.
2066 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2071 page = __grab_cache_page(mapping, index);
2078 * non-uptodate pages cannot cope with short copies, and we
2079 * cannot take a pagefault with the destination page locked.
2080 * So pin the source page to copy it.
2082 if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) {
2085 src_page = alloc_page(GFP_KERNEL);
2087 page_cache_release(page);
2093 * Cannot get_user_pages with a page locked for the
2094 * same reason as we can't take a page fault with a
2095 * page locked (as explained below).
2097 copied = iov_iter_copy_from_user(src_page, i,
2099 if (unlikely(copied == 0)) {
2101 page_cache_release(page);
2102 page_cache_release(src_page);
2109 * Can't handle the page going uptodate here, because
2110 * that means we would use non-atomic usercopies, which
2111 * zero out the tail of the page, which can cause
2112 * zeroes to become transiently visible. We could just
2113 * use a non-zeroing copy, but the APIs aren't too
2116 if (unlikely(!page->mapping || PageUptodate(page))) {
2118 page_cache_release(page);
2119 page_cache_release(src_page);
2124 status = a_ops->prepare_write(file, page, offset, offset+bytes);
2125 if (unlikely(status))
2126 goto fs_write_aop_error;
2130 * Must not enter the pagefault handler here, because
2131 * we hold the page lock, so we might recursively
2132 * deadlock on the same lock, or get an ABBA deadlock
2133 * against a different lock, or against the mmap_sem
2134 * (which nests outside the page lock). So increment
2135 * preempt count, and use _atomic usercopies.
2137 * The page is uptodate so we are OK to encounter a
2138 * short copy: if unmodified parts of the page are
2139 * marked dirty and written out to disk, it doesn't
2142 pagefault_disable();
2143 copied = iov_iter_copy_from_user_atomic(page, i,
2148 src = kmap_atomic(src_page, KM_USER0);
2149 dst = kmap_atomic(page, KM_USER1);
2150 memcpy(dst + offset, src + offset, bytes);
2151 kunmap_atomic(dst, KM_USER1);
2152 kunmap_atomic(src, KM_USER0);
2155 flush_dcache_page(page);
2157 status = a_ops->commit_write(file, page, offset, offset+bytes);
2158 if (unlikely(status < 0))
2159 goto fs_write_aop_error;
2160 if (unlikely(status > 0)) /* filesystem did partial write */
2161 copied = min_t(size_t, copied, status);
2164 mark_page_accessed(page);
2165 page_cache_release(page);
2167 page_cache_release(src_page);
2169 iov_iter_advance(i, copied);
2173 balance_dirty_pages_ratelimited(mapping);
2179 page_cache_release(page);
2181 page_cache_release(src_page);
2184 * prepare_write() may have instantiated a few blocks
2185 * outside i_size. Trim these off again. Don't need
2186 * i_size_read because we hold i_mutex.
2188 if (pos + bytes > inode->i_size)
2189 vmtruncate(inode, inode->i_size);
2191 } while (iov_iter_count(i));
2193 return written ? written : status;
2196 static ssize_t generic_perform_write(struct file *file,
2197 struct iov_iter *i, loff_t pos)
2199 struct address_space *mapping = file->f_mapping;
2200 const struct address_space_operations *a_ops = mapping->a_ops;
2202 ssize_t written = 0;
2203 unsigned int flags = 0;
2206 * Copies from kernel address space cannot fail (NFSD is a big user).
2208 if (segment_eq(get_fs(), KERNEL_DS))
2209 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2213 pgoff_t index; /* Pagecache index for current page */
2214 unsigned long offset; /* Offset into pagecache page */
2215 unsigned long bytes; /* Bytes to write to page */
2216 size_t copied; /* Bytes copied from user */
2219 offset = (pos & (PAGE_CACHE_SIZE - 1));
2220 index = pos >> PAGE_CACHE_SHIFT;
2221 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2227 * Bring in the user page that we will copy from _first_.
2228 * Otherwise there's a nasty deadlock on copying from the
2229 * same page as we're writing to, without it being marked
2232 * Not only is this an optimisation, but it is also required
2233 * to check that the address is actually valid, when atomic
2234 * usercopies are used, below.
2236 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2241 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2243 if (unlikely(status))
2246 pagefault_disable();
2247 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2249 flush_dcache_page(page);
2251 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2253 if (unlikely(status < 0))
2259 if (unlikely(copied == 0)) {
2261 * If we were unable to copy any data at all, we must
2262 * fall back to a single segment length write.
2264 * If we didn't fallback here, we could livelock
2265 * because not all segments in the iov can be copied at
2266 * once without a pagefault.
2268 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2269 iov_iter_single_seg_count(i));
2272 iov_iter_advance(i, copied);
2276 balance_dirty_pages_ratelimited(mapping);
2278 } while (iov_iter_count(i));
2280 return written ? written : status;
2284 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2285 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2286 size_t count, ssize_t written)
2288 struct file *file = iocb->ki_filp;
2289 struct address_space *mapping = file->f_mapping;
2290 const struct address_space_operations *a_ops = mapping->a_ops;
2291 struct inode *inode = mapping->host;
2295 iov_iter_init(&i, iov, nr_segs, count, written);
2296 if (a_ops->write_begin)
2297 status = generic_perform_write(file, &i, pos);
2299 status = generic_perform_write_2copy(file, &i, pos);
2301 if (likely(status >= 0)) {
2303 *ppos = pos + status;
2306 * For now, when the user asks for O_SYNC, we'll actually give
2309 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2310 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2311 status = generic_osync_inode(inode, mapping,
2312 OSYNC_METADATA|OSYNC_DATA);
2317 * If we get here for O_DIRECT writes then we must have fallen through
2318 * to buffered writes (block instantiation inside i_size). So we sync
2319 * the file data here, to try to honour O_DIRECT expectations.
2321 if (unlikely(file->f_flags & O_DIRECT) && written)
2322 status = filemap_write_and_wait(mapping);
2324 return written ? written : status;
2326 EXPORT_SYMBOL(generic_file_buffered_write);
2329 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2330 unsigned long nr_segs, loff_t *ppos)
2332 struct file *file = iocb->ki_filp;
2333 struct address_space * mapping = file->f_mapping;
2334 size_t ocount; /* original count */
2335 size_t count; /* after file limit checks */
2336 struct inode *inode = mapping->host;
2342 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2349 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2351 /* We can write back this queue in page reclaim */
2352 current->backing_dev_info = mapping->backing_dev_info;
2355 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2362 err = remove_suid(file->f_path.dentry);
2366 file_update_time(file);
2368 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2369 if (unlikely(file->f_flags & O_DIRECT)) {
2371 ssize_t written_buffered;
2373 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2374 ppos, count, ocount);
2375 if (written < 0 || written == count)
2378 * direct-io write to a hole: fall through to buffered I/O
2379 * for completing the rest of the request.
2383 written_buffered = generic_file_buffered_write(iocb, iov,
2384 nr_segs, pos, ppos, count,
2387 * If generic_file_buffered_write() retuned a synchronous error
2388 * then we want to return the number of bytes which were
2389 * direct-written, or the error code if that was zero. Note
2390 * that this differs from normal direct-io semantics, which
2391 * will return -EFOO even if some bytes were written.
2393 if (written_buffered < 0) {
2394 err = written_buffered;
2399 * We need to ensure that the page cache pages are written to
2400 * disk and invalidated to preserve the expected O_DIRECT
2403 endbyte = pos + written_buffered - written - 1;
2404 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2405 SYNC_FILE_RANGE_WAIT_BEFORE|
2406 SYNC_FILE_RANGE_WRITE|
2407 SYNC_FILE_RANGE_WAIT_AFTER);
2409 written = written_buffered;
2410 invalidate_mapping_pages(mapping,
2411 pos >> PAGE_CACHE_SHIFT,
2412 endbyte >> PAGE_CACHE_SHIFT);
2415 * We don't know how much we wrote, so just return
2416 * the number of bytes which were direct-written
2420 written = generic_file_buffered_write(iocb, iov, nr_segs,
2421 pos, ppos, count, written);
2424 current->backing_dev_info = NULL;
2425 return written ? written : err;
2428 ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2429 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
2431 struct file *file = iocb->ki_filp;
2432 struct address_space *mapping = file->f_mapping;
2433 struct inode *inode = mapping->host;
2436 BUG_ON(iocb->ki_pos != pos);
2438 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2441 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2444 err = sync_page_range_nolock(inode, mapping, pos, ret);
2450 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2452 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2453 unsigned long nr_segs, loff_t pos)
2455 struct file *file = iocb->ki_filp;
2456 struct address_space *mapping = file->f_mapping;
2457 struct inode *inode = mapping->host;
2460 BUG_ON(iocb->ki_pos != pos);
2462 mutex_lock(&inode->i_mutex);
2463 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2465 mutex_unlock(&inode->i_mutex);
2467 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2470 err = sync_page_range(inode, mapping, pos, ret);
2476 EXPORT_SYMBOL(generic_file_aio_write);
2479 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something
2480 * went wrong during pagecache shootdown.
2483 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2484 loff_t offset, unsigned long nr_segs)
2486 struct file *file = iocb->ki_filp;
2487 struct address_space *mapping = file->f_mapping;
2490 pgoff_t end = 0; /* silence gcc */
2493 * If it's a write, unmap all mmappings of the file up-front. This
2494 * will cause any pte dirty bits to be propagated into the pageframes
2495 * for the subsequent filemap_write_and_wait().
2498 write_len = iov_length(iov, nr_segs);
2499 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
2500 if (mapping_mapped(mapping))
2501 unmap_mapping_range(mapping, offset, write_len, 0);
2504 retval = filemap_write_and_wait(mapping);
2509 * After a write we want buffered reads to be sure to go to disk to get
2510 * the new data. We invalidate clean cached page from the region we're
2511 * about to write. We do this *before* the write so that we can return
2512 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2514 if (rw == WRITE && mapping->nrpages) {
2515 retval = invalidate_inode_pages2_range(mapping,
2516 offset >> PAGE_CACHE_SHIFT, end);
2521 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2524 * Finally, try again to invalidate clean pages which might have been
2525 * cached by non-direct readahead, or faulted in by get_user_pages()
2526 * if the source of the write was an mmap'ed region of the file
2527 * we're writing. Either one is a pretty crazy thing to do,
2528 * so we don't support it 100%. If this invalidation
2529 * fails, tough, the write still worked...
2531 if (rw == WRITE && mapping->nrpages) {
2532 invalidate_inode_pages2_range(mapping, offset >> PAGE_CACHE_SHIFT, end);
2539 * try_to_release_page() - release old fs-specific metadata on a page
2541 * @page: the page which the kernel is trying to free
2542 * @gfp_mask: memory allocation flags (and I/O mode)
2544 * The address_space is to try to release any data against the page
2545 * (presumably at page->private). If the release was successful, return `1'.
2546 * Otherwise return zero.
2548 * The @gfp_mask argument specifies whether I/O may be performed to release
2549 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2551 * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2553 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2555 struct address_space * const mapping = page->mapping;
2557 BUG_ON(!PageLocked(page));
2558 if (PageWriteback(page))
2561 if (mapping && mapping->a_ops->releasepage)
2562 return mapping->a_ops->releasepage(page, gfp_mask);
2563 return try_to_free_buffers(page);
2566 EXPORT_SYMBOL(try_to_release_page);