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/compiler.h>
15 #include <linux/uaccess.h>
16 #include <linux/aio.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.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/security.h>
32 #include <linux/syscalls.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/memcontrol.h>
36 #include <linux/mm_inline.h> /* for page_is_file_cache() */
40 * FIXME: remove all knowledge of the buffer layer from the core VM
42 #include <linux/buffer_head.h> /* for try_to_free_buffers */
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
50 * Shared mappings now work. 15.8.1995 Bruno.
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
61 * ->i_mmap_lock (truncate_pagecache)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
67 * ->i_mmap_lock (truncate->unmap_mapping_range)
71 * ->page_table_lock or pte_lock (various, mainly in memory.c)
72 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
75 * ->lock_page (access_process_vm)
77 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
81 * ->i_alloc_sem (various)
84 * ->sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
88 * ->anon_vma.lock (vma_adjust)
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 * (code doesn't rely on that order, so you could switch it around)
106 * ->tasklist_lock (memory_failure, collect_procs_ao)
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 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);
123 if (PageSwapBacked(page))
124 __dec_zone_page_state(page, NR_SHMEM);
125 BUG_ON(page_mapped(page));
128 * Some filesystems seem to re-dirty the page even after
129 * the VM has canceled the dirty bit (eg ext3 journaling).
131 * Fix it up by doing a final dirty accounting check after
132 * having removed the page entirely.
134 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
135 dec_zone_page_state(page, NR_FILE_DIRTY);
136 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
140 void remove_from_page_cache(struct page *page)
142 struct address_space *mapping = page->mapping;
143 void (*freepage)(struct page *);
145 BUG_ON(!PageLocked(page));
147 freepage = mapping->a_ops->freepage;
148 spin_lock_irq(&mapping->tree_lock);
149 __remove_from_page_cache(page);
150 spin_unlock_irq(&mapping->tree_lock);
151 mem_cgroup_uncharge_cache_page(page);
156 EXPORT_SYMBOL(remove_from_page_cache);
158 static int sleep_on_page(void *word)
164 static int sleep_on_page_killable(void *word)
167 return fatal_signal_pending(current) ? -EINTR : 0;
171 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
172 * @mapping: address space structure to write
173 * @start: offset in bytes where the range starts
174 * @end: offset in bytes where the range ends (inclusive)
175 * @sync_mode: enable synchronous operation
177 * Start writeback against all of a mapping's dirty pages that lie
178 * within the byte offsets <start, end> inclusive.
180 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
181 * opposed to a regular memory cleansing writeback. The difference between
182 * these two operations is that if a dirty page/buffer is encountered, it must
183 * be waited upon, and not just skipped over.
185 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
186 loff_t end, int sync_mode)
189 struct writeback_control wbc = {
190 .sync_mode = sync_mode,
191 .nr_to_write = LONG_MAX,
192 .range_start = start,
196 if (!mapping_cap_writeback_dirty(mapping))
199 ret = do_writepages(mapping, &wbc);
203 static inline int __filemap_fdatawrite(struct address_space *mapping,
206 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
209 int filemap_fdatawrite(struct address_space *mapping)
211 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
213 EXPORT_SYMBOL(filemap_fdatawrite);
215 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
218 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
220 EXPORT_SYMBOL(filemap_fdatawrite_range);
223 * filemap_flush - mostly a non-blocking flush
224 * @mapping: target address_space
226 * This is a mostly non-blocking flush. Not suitable for data-integrity
227 * purposes - I/O may not be started against all dirty pages.
229 int filemap_flush(struct address_space *mapping)
231 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
233 EXPORT_SYMBOL(filemap_flush);
236 * filemap_fdatawait_range - wait for writeback to complete
237 * @mapping: address space structure to wait for
238 * @start_byte: offset in bytes where the range starts
239 * @end_byte: offset in bytes where the range ends (inclusive)
241 * Walk the list of under-writeback pages of the given address space
242 * in the given range and wait for all of them.
244 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
247 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
248 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
253 if (end_byte < start_byte)
256 pagevec_init(&pvec, 0);
257 while ((index <= end) &&
258 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
259 PAGECACHE_TAG_WRITEBACK,
260 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
263 for (i = 0; i < nr_pages; i++) {
264 struct page *page = pvec.pages[i];
266 /* until radix tree lookup accepts end_index */
267 if (page->index > end)
270 wait_on_page_writeback(page);
271 if (TestClearPageError(page))
274 pagevec_release(&pvec);
278 /* Check for outstanding write errors */
279 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
281 if (test_and_clear_bit(AS_EIO, &mapping->flags))
286 EXPORT_SYMBOL(filemap_fdatawait_range);
289 * filemap_fdatawait - wait for all under-writeback pages to complete
290 * @mapping: address space structure to wait for
292 * Walk the list of under-writeback pages of the given address space
293 * and wait for all of them.
295 int filemap_fdatawait(struct address_space *mapping)
297 loff_t i_size = i_size_read(mapping->host);
302 return filemap_fdatawait_range(mapping, 0, i_size - 1);
304 EXPORT_SYMBOL(filemap_fdatawait);
306 int filemap_write_and_wait(struct address_space *mapping)
310 if (mapping->nrpages) {
311 err = filemap_fdatawrite(mapping);
313 * Even if the above returned error, the pages may be
314 * written partially (e.g. -ENOSPC), so we wait for it.
315 * But the -EIO is special case, it may indicate the worst
316 * thing (e.g. bug) happened, so we avoid waiting for it.
319 int err2 = filemap_fdatawait(mapping);
326 EXPORT_SYMBOL(filemap_write_and_wait);
329 * filemap_write_and_wait_range - write out & wait on a file range
330 * @mapping: the address_space for the pages
331 * @lstart: offset in bytes where the range starts
332 * @lend: offset in bytes where the range ends (inclusive)
334 * Write out and wait upon file offsets lstart->lend, inclusive.
336 * Note that `lend' is inclusive (describes the last byte to be written) so
337 * that this function can be used to write to the very end-of-file (end = -1).
339 int filemap_write_and_wait_range(struct address_space *mapping,
340 loff_t lstart, loff_t lend)
344 if (mapping->nrpages) {
345 err = __filemap_fdatawrite_range(mapping, lstart, lend,
347 /* See comment of filemap_write_and_wait() */
349 int err2 = filemap_fdatawait_range(mapping,
357 EXPORT_SYMBOL(filemap_write_and_wait_range);
360 * add_to_page_cache_locked - add a locked page to the pagecache
362 * @mapping: the page's address_space
363 * @offset: page index
364 * @gfp_mask: page allocation mode
366 * This function is used to add a page to the pagecache. It must be locked.
367 * This function does not add the page to the LRU. The caller must do that.
369 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
370 pgoff_t offset, gfp_t gfp_mask)
374 VM_BUG_ON(!PageLocked(page));
376 error = mem_cgroup_cache_charge(page, current->mm,
377 gfp_mask & GFP_RECLAIM_MASK);
381 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
383 page_cache_get(page);
384 page->mapping = mapping;
385 page->index = offset;
387 spin_lock_irq(&mapping->tree_lock);
388 error = radix_tree_insert(&mapping->page_tree, offset, page);
389 if (likely(!error)) {
391 __inc_zone_page_state(page, NR_FILE_PAGES);
392 if (PageSwapBacked(page))
393 __inc_zone_page_state(page, NR_SHMEM);
394 spin_unlock_irq(&mapping->tree_lock);
396 page->mapping = NULL;
397 spin_unlock_irq(&mapping->tree_lock);
398 mem_cgroup_uncharge_cache_page(page);
399 page_cache_release(page);
401 radix_tree_preload_end();
403 mem_cgroup_uncharge_cache_page(page);
407 EXPORT_SYMBOL(add_to_page_cache_locked);
409 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
410 pgoff_t offset, gfp_t gfp_mask)
415 * Splice_read and readahead add shmem/tmpfs pages into the page cache
416 * before shmem_readpage has a chance to mark them as SwapBacked: they
417 * need to go on the anon lru below, and mem_cgroup_cache_charge
418 * (called in add_to_page_cache) needs to know where they're going too.
420 if (mapping_cap_swap_backed(mapping))
421 SetPageSwapBacked(page);
423 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
425 if (page_is_file_cache(page))
426 lru_cache_add_file(page);
428 lru_cache_add_anon(page);
432 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
435 struct page *__page_cache_alloc(gfp_t gfp)
440 if (cpuset_do_page_mem_spread()) {
442 n = cpuset_mem_spread_node();
443 page = alloc_pages_exact_node(n, gfp, 0);
447 return alloc_pages(gfp, 0);
449 EXPORT_SYMBOL(__page_cache_alloc);
453 * In order to wait for pages to become available there must be
454 * waitqueues associated with pages. By using a hash table of
455 * waitqueues where the bucket discipline is to maintain all
456 * waiters on the same queue and wake all when any of the pages
457 * become available, and for the woken contexts to check to be
458 * sure the appropriate page became available, this saves space
459 * at a cost of "thundering herd" phenomena during rare hash
462 static wait_queue_head_t *page_waitqueue(struct page *page)
464 const struct zone *zone = page_zone(page);
466 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
469 static inline void wake_up_page(struct page *page, int bit)
471 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
474 void wait_on_page_bit(struct page *page, int bit_nr)
476 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
478 if (test_bit(bit_nr, &page->flags))
479 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
480 TASK_UNINTERRUPTIBLE);
482 EXPORT_SYMBOL(wait_on_page_bit);
485 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
486 * @page: Page defining the wait queue of interest
487 * @waiter: Waiter to add to the queue
489 * Add an arbitrary @waiter to the wait queue for the nominated @page.
491 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
493 wait_queue_head_t *q = page_waitqueue(page);
496 spin_lock_irqsave(&q->lock, flags);
497 __add_wait_queue(q, waiter);
498 spin_unlock_irqrestore(&q->lock, flags);
500 EXPORT_SYMBOL_GPL(add_page_wait_queue);
503 * unlock_page - unlock a locked page
506 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
507 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
508 * mechananism between PageLocked pages and PageWriteback pages is shared.
509 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
511 * The mb is necessary to enforce ordering between the clear_bit and the read
512 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
514 void unlock_page(struct page *page)
516 VM_BUG_ON(!PageLocked(page));
517 clear_bit_unlock(PG_locked, &page->flags);
518 smp_mb__after_clear_bit();
519 wake_up_page(page, PG_locked);
521 EXPORT_SYMBOL(unlock_page);
524 * end_page_writeback - end writeback against a page
527 void end_page_writeback(struct page *page)
529 if (TestClearPageReclaim(page))
530 rotate_reclaimable_page(page);
532 if (!test_clear_page_writeback(page))
535 smp_mb__after_clear_bit();
536 wake_up_page(page, PG_writeback);
538 EXPORT_SYMBOL(end_page_writeback);
541 * __lock_page - get a lock on the page, assuming we need to sleep to get it
542 * @page: the page to lock
544 void __lock_page(struct page *page)
546 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
548 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
549 TASK_UNINTERRUPTIBLE);
551 EXPORT_SYMBOL(__lock_page);
553 int __lock_page_killable(struct page *page)
555 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
557 return __wait_on_bit_lock(page_waitqueue(page), &wait,
558 sleep_on_page_killable, TASK_KILLABLE);
560 EXPORT_SYMBOL_GPL(__lock_page_killable);
562 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
565 if (!(flags & FAULT_FLAG_ALLOW_RETRY)) {
569 up_read(&mm->mmap_sem);
570 wait_on_page_locked(page);
576 * find_get_page - find and get a page reference
577 * @mapping: the address_space to search
578 * @offset: the page index
580 * Is there a pagecache struct page at the given (mapping, offset) tuple?
581 * If yes, increment its refcount and return it; if no, return NULL.
583 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
591 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
593 page = radix_tree_deref_slot(pagep);
596 if (radix_tree_deref_retry(page))
599 if (!page_cache_get_speculative(page))
603 * Has the page moved?
604 * This is part of the lockless pagecache protocol. See
605 * include/linux/pagemap.h for details.
607 if (unlikely(page != *pagep)) {
608 page_cache_release(page);
617 EXPORT_SYMBOL(find_get_page);
620 * find_lock_page - locate, pin and lock a pagecache page
621 * @mapping: the address_space to search
622 * @offset: the page index
624 * Locates the desired pagecache page, locks it, increments its reference
625 * count and returns its address.
627 * Returns zero if the page was not present. find_lock_page() may sleep.
629 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
634 page = find_get_page(mapping, offset);
637 /* Has the page been truncated? */
638 if (unlikely(page->mapping != mapping)) {
640 page_cache_release(page);
643 VM_BUG_ON(page->index != offset);
647 EXPORT_SYMBOL(find_lock_page);
650 * find_or_create_page - locate or add a pagecache page
651 * @mapping: the page's address_space
652 * @index: the page's index into the mapping
653 * @gfp_mask: page allocation mode
655 * Locates a page in the pagecache. If the page is not present, a new page
656 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
657 * LRU list. The returned page is locked and has its reference count
660 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
663 * find_or_create_page() returns the desired page's address, or zero on
666 struct page *find_or_create_page(struct address_space *mapping,
667 pgoff_t index, gfp_t gfp_mask)
672 page = find_lock_page(mapping, index);
674 page = __page_cache_alloc(gfp_mask);
678 * We want a regular kernel memory (not highmem or DMA etc)
679 * allocation for the radix tree nodes, but we need to honour
680 * the context-specific requirements the caller has asked for.
681 * GFP_RECLAIM_MASK collects those requirements.
683 err = add_to_page_cache_lru(page, mapping, index,
684 (gfp_mask & GFP_RECLAIM_MASK));
686 page_cache_release(page);
694 EXPORT_SYMBOL(find_or_create_page);
697 * find_get_pages - gang pagecache lookup
698 * @mapping: The address_space to search
699 * @start: The starting page index
700 * @nr_pages: The maximum number of pages
701 * @pages: Where the resulting pages are placed
703 * find_get_pages() will search for and return a group of up to
704 * @nr_pages pages in the mapping. The pages are placed at @pages.
705 * find_get_pages() takes a reference against the returned pages.
707 * The search returns a group of mapping-contiguous pages with ascending
708 * indexes. There may be holes in the indices due to not-present pages.
710 * find_get_pages() returns the number of pages which were found.
712 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
713 unsigned int nr_pages, struct page **pages)
717 unsigned int nr_found;
721 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
722 (void ***)pages, start, nr_pages);
724 for (i = 0; i < nr_found; i++) {
727 page = radix_tree_deref_slot((void **)pages[i]);
730 if (radix_tree_deref_retry(page)) {
732 start = pages[ret-1]->index;
736 if (!page_cache_get_speculative(page))
739 /* Has the page moved? */
740 if (unlikely(page != *((void **)pages[i]))) {
741 page_cache_release(page);
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)
769 unsigned int nr_found;
773 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
774 (void ***)pages, index, nr_pages);
776 for (i = 0; i < nr_found; i++) {
779 page = radix_tree_deref_slot((void **)pages[i]);
782 if (radix_tree_deref_retry(page))
785 if (!page_cache_get_speculative(page))
788 /* Has the page moved? */
789 if (unlikely(page != *((void **)pages[i]))) {
790 page_cache_release(page);
795 * must check mapping and index after taking the ref.
796 * otherwise we can get both false positives and false
797 * negatives, which is just confusing to the caller.
799 if (page->mapping == NULL || page->index != index) {
800 page_cache_release(page);
811 EXPORT_SYMBOL(find_get_pages_contig);
814 * find_get_pages_tag - find and return pages that match @tag
815 * @mapping: the address_space to search
816 * @index: the starting page index
817 * @tag: the tag index
818 * @nr_pages: the maximum number of pages
819 * @pages: where the resulting pages are placed
821 * Like find_get_pages, except we only return pages which are tagged with
822 * @tag. We update @index to index the next page for the traversal.
824 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
825 int tag, unsigned int nr_pages, struct page **pages)
829 unsigned int nr_found;
833 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
834 (void ***)pages, *index, nr_pages, tag);
836 for (i = 0; i < nr_found; i++) {
839 page = radix_tree_deref_slot((void **)pages[i]);
842 if (radix_tree_deref_retry(page))
845 if (!page_cache_get_speculative(page))
848 /* Has the page moved? */
849 if (unlikely(page != *((void **)pages[i]))) {
850 page_cache_release(page);
860 *index = pages[ret - 1]->index + 1;
864 EXPORT_SYMBOL(find_get_pages_tag);
867 * grab_cache_page_nowait - returns locked page at given index in given cache
868 * @mapping: target address_space
869 * @index: the page index
871 * Same as grab_cache_page(), but do not wait if the page is unavailable.
872 * This is intended for speculative data generators, where the data can
873 * be regenerated if the page couldn't be grabbed. This routine should
874 * be safe to call while holding the lock for another page.
876 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
877 * and deadlock against the caller's locked page.
880 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
882 struct page *page = find_get_page(mapping, index);
885 if (trylock_page(page))
887 page_cache_release(page);
890 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
891 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
892 page_cache_release(page);
897 EXPORT_SYMBOL(grab_cache_page_nowait);
900 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
901 * a _large_ part of the i/o request. Imagine the worst scenario:
903 * ---R__________________________________________B__________
904 * ^ reading here ^ bad block(assume 4k)
906 * read(R) => miss => readahead(R...B) => media error => frustrating retries
907 * => failing the whole request => read(R) => read(R+1) =>
908 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
909 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
910 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
912 * It is going insane. Fix it by quickly scaling down the readahead size.
914 static void shrink_readahead_size_eio(struct file *filp,
915 struct file_ra_state *ra)
921 * do_generic_file_read - generic file read routine
922 * @filp: the file to read
923 * @ppos: current file position
924 * @desc: read_descriptor
925 * @actor: read method
927 * This is a generic file read routine, and uses the
928 * mapping->a_ops->readpage() function for the actual low-level stuff.
930 * This is really ugly. But the goto's actually try to clarify some
931 * of the logic when it comes to error handling etc.
933 static void do_generic_file_read(struct file *filp, loff_t *ppos,
934 read_descriptor_t *desc, read_actor_t actor)
936 struct address_space *mapping = filp->f_mapping;
937 struct inode *inode = mapping->host;
938 struct file_ra_state *ra = &filp->f_ra;
942 unsigned long offset; /* offset into pagecache page */
943 unsigned int prev_offset;
946 index = *ppos >> PAGE_CACHE_SHIFT;
947 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
948 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
949 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
950 offset = *ppos & ~PAGE_CACHE_MASK;
956 unsigned long nr, ret;
960 page = find_get_page(mapping, index);
962 page_cache_sync_readahead(mapping,
964 index, last_index - index);
965 page = find_get_page(mapping, index);
966 if (unlikely(page == NULL))
969 if (PageReadahead(page)) {
970 page_cache_async_readahead(mapping,
972 index, last_index - index);
974 if (!PageUptodate(page)) {
975 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
976 !mapping->a_ops->is_partially_uptodate)
977 goto page_not_up_to_date;
978 if (!trylock_page(page))
979 goto page_not_up_to_date;
980 /* Did it get truncated before we got the lock? */
982 goto page_not_up_to_date_locked;
983 if (!mapping->a_ops->is_partially_uptodate(page,
985 goto page_not_up_to_date_locked;
990 * i_size must be checked after we know the page is Uptodate.
992 * Checking i_size after the check allows us to calculate
993 * the correct value for "nr", which means the zero-filled
994 * part of the page is not copied back to userspace (unless
995 * another truncate extends the file - this is desired though).
998 isize = i_size_read(inode);
999 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1000 if (unlikely(!isize || index > end_index)) {
1001 page_cache_release(page);
1005 /* nr is the maximum number of bytes to copy from this page */
1006 nr = PAGE_CACHE_SIZE;
1007 if (index == end_index) {
1008 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1010 page_cache_release(page);
1016 /* If users can be writing to this page using arbitrary
1017 * virtual addresses, take care about potential aliasing
1018 * before reading the page on the kernel side.
1020 if (mapping_writably_mapped(mapping))
1021 flush_dcache_page(page);
1024 * When a sequential read accesses a page several times,
1025 * only mark it as accessed the first time.
1027 if (prev_index != index || offset != prev_offset)
1028 mark_page_accessed(page);
1032 * Ok, we have the page, and it's up-to-date, so
1033 * now we can copy it to user space...
1035 * The actor routine returns how many bytes were actually used..
1036 * NOTE! This may not be the same as how much of a user buffer
1037 * we filled up (we may be padding etc), so we can only update
1038 * "pos" here (the actor routine has to update the user buffer
1039 * pointers and the remaining count).
1041 ret = actor(desc, page, offset, nr);
1043 index += offset >> PAGE_CACHE_SHIFT;
1044 offset &= ~PAGE_CACHE_MASK;
1045 prev_offset = offset;
1047 page_cache_release(page);
1048 if (ret == nr && desc->count)
1052 page_not_up_to_date:
1053 /* Get exclusive access to the page ... */
1054 error = lock_page_killable(page);
1055 if (unlikely(error))
1056 goto readpage_error;
1058 page_not_up_to_date_locked:
1059 /* Did it get truncated before we got the lock? */
1060 if (!page->mapping) {
1062 page_cache_release(page);
1066 /* Did somebody else fill it already? */
1067 if (PageUptodate(page)) {
1074 * A previous I/O error may have been due to temporary
1075 * failures, eg. multipath errors.
1076 * PG_error will be set again if readpage fails.
1078 ClearPageError(page);
1079 /* Start the actual read. The read will unlock the page. */
1080 error = mapping->a_ops->readpage(filp, page);
1082 if (unlikely(error)) {
1083 if (error == AOP_TRUNCATED_PAGE) {
1084 page_cache_release(page);
1087 goto readpage_error;
1090 if (!PageUptodate(page)) {
1091 error = lock_page_killable(page);
1092 if (unlikely(error))
1093 goto readpage_error;
1094 if (!PageUptodate(page)) {
1095 if (page->mapping == NULL) {
1097 * invalidate_mapping_pages got it
1100 page_cache_release(page);
1104 shrink_readahead_size_eio(filp, ra);
1106 goto readpage_error;
1114 /* UHHUH! A synchronous read error occurred. Report it */
1115 desc->error = error;
1116 page_cache_release(page);
1121 * Ok, it wasn't cached, so we need to create a new
1124 page = page_cache_alloc_cold(mapping);
1126 desc->error = -ENOMEM;
1129 error = add_to_page_cache_lru(page, mapping,
1132 page_cache_release(page);
1133 if (error == -EEXIST)
1135 desc->error = error;
1142 ra->prev_pos = prev_index;
1143 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1144 ra->prev_pos |= prev_offset;
1146 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1147 file_accessed(filp);
1150 int file_read_actor(read_descriptor_t *desc, struct page *page,
1151 unsigned long offset, unsigned long size)
1154 unsigned long left, count = desc->count;
1160 * Faults on the destination of a read are common, so do it before
1163 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1164 kaddr = kmap_atomic(page, KM_USER0);
1165 left = __copy_to_user_inatomic(desc->arg.buf,
1166 kaddr + offset, size);
1167 kunmap_atomic(kaddr, KM_USER0);
1172 /* Do it the slow way */
1174 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1179 desc->error = -EFAULT;
1182 desc->count = count - size;
1183 desc->written += size;
1184 desc->arg.buf += size;
1189 * Performs necessary checks before doing a write
1190 * @iov: io vector request
1191 * @nr_segs: number of segments in the iovec
1192 * @count: number of bytes to write
1193 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1195 * Adjust number of segments and amount of bytes to write (nr_segs should be
1196 * properly initialized first). Returns appropriate error code that caller
1197 * should return or zero in case that write should be allowed.
1199 int generic_segment_checks(const struct iovec *iov,
1200 unsigned long *nr_segs, size_t *count, int access_flags)
1204 for (seg = 0; seg < *nr_segs; seg++) {
1205 const struct iovec *iv = &iov[seg];
1208 * If any segment has a negative length, or the cumulative
1209 * length ever wraps negative then return -EINVAL.
1212 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1214 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1219 cnt -= iv->iov_len; /* This segment is no good */
1225 EXPORT_SYMBOL(generic_segment_checks);
1228 * generic_file_aio_read - generic filesystem read routine
1229 * @iocb: kernel I/O control block
1230 * @iov: io vector request
1231 * @nr_segs: number of segments in the iovec
1232 * @pos: current file position
1234 * This is the "read()" routine for all filesystems
1235 * that can use the page cache directly.
1238 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1239 unsigned long nr_segs, loff_t pos)
1241 struct file *filp = iocb->ki_filp;
1243 unsigned long seg = 0;
1245 loff_t *ppos = &iocb->ki_pos;
1246 struct blk_plug plug;
1249 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1253 blk_start_plug(&plug);
1255 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1256 if (filp->f_flags & O_DIRECT) {
1258 struct address_space *mapping;
1259 struct inode *inode;
1261 mapping = filp->f_mapping;
1262 inode = mapping->host;
1264 goto out; /* skip atime */
1265 size = i_size_read(inode);
1267 retval = filemap_write_and_wait_range(mapping, pos,
1268 pos + iov_length(iov, nr_segs) - 1);
1270 retval = mapping->a_ops->direct_IO(READ, iocb,
1274 *ppos = pos + retval;
1279 * Btrfs can have a short DIO read if we encounter
1280 * compressed extents, so if there was an error, or if
1281 * we've already read everything we wanted to, or if
1282 * there was a short read because we hit EOF, go ahead
1283 * and return. Otherwise fallthrough to buffered io for
1284 * the rest of the read.
1286 if (retval < 0 || !count || *ppos >= size) {
1287 file_accessed(filp);
1294 for (seg = 0; seg < nr_segs; seg++) {
1295 read_descriptor_t desc;
1299 * If we did a short DIO read we need to skip the section of the
1300 * iov that we've already read data into.
1303 if (count > iov[seg].iov_len) {
1304 count -= iov[seg].iov_len;
1312 desc.arg.buf = iov[seg].iov_base + offset;
1313 desc.count = iov[seg].iov_len - offset;
1314 if (desc.count == 0)
1317 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1318 retval += desc.written;
1320 retval = retval ?: desc.error;
1327 blk_finish_plug(&plug);
1330 EXPORT_SYMBOL(generic_file_aio_read);
1333 do_readahead(struct address_space *mapping, struct file *filp,
1334 pgoff_t index, unsigned long nr)
1336 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1339 force_page_cache_readahead(mapping, filp, index, nr);
1343 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1351 if (file->f_mode & FMODE_READ) {
1352 struct address_space *mapping = file->f_mapping;
1353 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1354 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1355 unsigned long len = end - start + 1;
1356 ret = do_readahead(mapping, file, start, len);
1362 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1363 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1365 return SYSC_readahead((int) fd, offset, (size_t) count);
1367 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1372 * page_cache_read - adds requested page to the page cache if not already there
1373 * @file: file to read
1374 * @offset: page index
1376 * This adds the requested page to the page cache if it isn't already there,
1377 * and schedules an I/O to read in its contents from disk.
1379 static int page_cache_read(struct file *file, pgoff_t offset)
1381 struct address_space *mapping = file->f_mapping;
1386 page = page_cache_alloc_cold(mapping);
1390 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1392 ret = mapping->a_ops->readpage(file, page);
1393 else if (ret == -EEXIST)
1394 ret = 0; /* losing race to add is OK */
1396 page_cache_release(page);
1398 } while (ret == AOP_TRUNCATED_PAGE);
1403 #define MMAP_LOTSAMISS (100)
1406 * Synchronous readahead happens when we don't even find
1407 * a page in the page cache at all.
1409 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1410 struct file_ra_state *ra,
1414 unsigned long ra_pages;
1415 struct address_space *mapping = file->f_mapping;
1417 /* If we don't want any read-ahead, don't bother */
1418 if (VM_RandomReadHint(vma))
1421 if (VM_SequentialReadHint(vma) ||
1422 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1423 page_cache_sync_readahead(mapping, ra, file, offset,
1428 if (ra->mmap_miss < INT_MAX)
1432 * Do we miss much more than hit in this file? If so,
1433 * stop bothering with read-ahead. It will only hurt.
1435 if (ra->mmap_miss > MMAP_LOTSAMISS)
1441 ra_pages = max_sane_readahead(ra->ra_pages);
1443 ra->start = max_t(long, 0, offset - ra_pages/2);
1444 ra->size = ra_pages;
1446 ra_submit(ra, mapping, file);
1451 * Asynchronous readahead happens when we find the page and PG_readahead,
1452 * so we want to possibly extend the readahead further..
1454 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1455 struct file_ra_state *ra,
1460 struct address_space *mapping = file->f_mapping;
1462 /* If we don't want any read-ahead, don't bother */
1463 if (VM_RandomReadHint(vma))
1465 if (ra->mmap_miss > 0)
1467 if (PageReadahead(page))
1468 page_cache_async_readahead(mapping, ra, file,
1469 page, offset, ra->ra_pages);
1473 * filemap_fault - read in file data for page fault handling
1474 * @vma: vma in which the fault was taken
1475 * @vmf: struct vm_fault containing details of the fault
1477 * filemap_fault() is invoked via the vma operations vector for a
1478 * mapped memory region to read in file data during a page fault.
1480 * The goto's are kind of ugly, but this streamlines the normal case of having
1481 * it in the page cache, and handles the special cases reasonably without
1482 * having a lot of duplicated code.
1484 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1487 struct file *file = vma->vm_file;
1488 struct address_space *mapping = file->f_mapping;
1489 struct file_ra_state *ra = &file->f_ra;
1490 struct inode *inode = mapping->host;
1491 pgoff_t offset = vmf->pgoff;
1496 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1498 return VM_FAULT_SIGBUS;
1501 * Do we have something in the page cache already?
1503 page = find_get_page(mapping, offset);
1506 * We found the page, so try async readahead before
1507 * waiting for the lock.
1509 do_async_mmap_readahead(vma, ra, file, page, offset);
1511 /* No page in the page cache at all */
1512 do_sync_mmap_readahead(vma, ra, file, offset);
1513 count_vm_event(PGMAJFAULT);
1514 ret = VM_FAULT_MAJOR;
1516 page = find_get_page(mapping, offset);
1518 goto no_cached_page;
1521 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1522 page_cache_release(page);
1523 return ret | VM_FAULT_RETRY;
1526 /* Did it get truncated? */
1527 if (unlikely(page->mapping != mapping)) {
1532 VM_BUG_ON(page->index != offset);
1535 * We have a locked page in the page cache, now we need to check
1536 * that it's up-to-date. If not, it is going to be due to an error.
1538 if (unlikely(!PageUptodate(page)))
1539 goto page_not_uptodate;
1542 * Found the page and have a reference on it.
1543 * We must recheck i_size under page lock.
1545 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1546 if (unlikely(offset >= size)) {
1548 page_cache_release(page);
1549 return VM_FAULT_SIGBUS;
1552 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1554 return ret | VM_FAULT_LOCKED;
1558 * We're only likely to ever get here if MADV_RANDOM is in
1561 error = page_cache_read(file, offset);
1564 * The page we want has now been added to the page cache.
1565 * In the unlikely event that someone removed it in the
1566 * meantime, we'll just come back here and read it again.
1572 * An error return from page_cache_read can result if the
1573 * system is low on memory, or a problem occurs while trying
1576 if (error == -ENOMEM)
1577 return VM_FAULT_OOM;
1578 return VM_FAULT_SIGBUS;
1582 * Umm, take care of errors if the page isn't up-to-date.
1583 * Try to re-read it _once_. We do this synchronously,
1584 * because there really aren't any performance issues here
1585 * and we need to check for errors.
1587 ClearPageError(page);
1588 error = mapping->a_ops->readpage(file, page);
1590 wait_on_page_locked(page);
1591 if (!PageUptodate(page))
1594 page_cache_release(page);
1596 if (!error || error == AOP_TRUNCATED_PAGE)
1599 /* Things didn't work out. Return zero to tell the mm layer so. */
1600 shrink_readahead_size_eio(file, ra);
1601 return VM_FAULT_SIGBUS;
1603 EXPORT_SYMBOL(filemap_fault);
1605 const struct vm_operations_struct generic_file_vm_ops = {
1606 .fault = filemap_fault,
1609 /* This is used for a general mmap of a disk file */
1611 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1613 struct address_space *mapping = file->f_mapping;
1615 if (!mapping->a_ops->readpage)
1617 file_accessed(file);
1618 vma->vm_ops = &generic_file_vm_ops;
1619 vma->vm_flags |= VM_CAN_NONLINEAR;
1624 * This is for filesystems which do not implement ->writepage.
1626 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1628 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1630 return generic_file_mmap(file, vma);
1633 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1637 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1641 #endif /* CONFIG_MMU */
1643 EXPORT_SYMBOL(generic_file_mmap);
1644 EXPORT_SYMBOL(generic_file_readonly_mmap);
1646 static struct page *__read_cache_page(struct address_space *mapping,
1648 int (*filler)(void *,struct page*),
1655 page = find_get_page(mapping, index);
1657 page = __page_cache_alloc(gfp | __GFP_COLD);
1659 return ERR_PTR(-ENOMEM);
1660 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1661 if (unlikely(err)) {
1662 page_cache_release(page);
1665 /* Presumably ENOMEM for radix tree node */
1666 return ERR_PTR(err);
1668 err = filler(data, page);
1670 page_cache_release(page);
1671 page = ERR_PTR(err);
1677 static struct page *do_read_cache_page(struct address_space *mapping,
1679 int (*filler)(void *,struct page*),
1688 page = __read_cache_page(mapping, index, filler, data, gfp);
1691 if (PageUptodate(page))
1695 if (!page->mapping) {
1697 page_cache_release(page);
1700 if (PageUptodate(page)) {
1704 err = filler(data, page);
1706 page_cache_release(page);
1707 return ERR_PTR(err);
1710 mark_page_accessed(page);
1715 * read_cache_page_async - read into page cache, fill it if needed
1716 * @mapping: the page's address_space
1717 * @index: the page index
1718 * @filler: function to perform the read
1719 * @data: destination for read data
1721 * Same as read_cache_page, but don't wait for page to become unlocked
1722 * after submitting it to the filler.
1724 * Read into the page cache. If a page already exists, and PageUptodate() is
1725 * not set, try to fill the page but don't wait for it to become unlocked.
1727 * If the page does not get brought uptodate, return -EIO.
1729 struct page *read_cache_page_async(struct address_space *mapping,
1731 int (*filler)(void *,struct page*),
1734 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1736 EXPORT_SYMBOL(read_cache_page_async);
1738 static struct page *wait_on_page_read(struct page *page)
1740 if (!IS_ERR(page)) {
1741 wait_on_page_locked(page);
1742 if (!PageUptodate(page)) {
1743 page_cache_release(page);
1744 page = ERR_PTR(-EIO);
1751 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1752 * @mapping: the page's address_space
1753 * @index: the page index
1754 * @gfp: the page allocator flags to use if allocating
1756 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1757 * any new page allocations done using the specified allocation flags. Note
1758 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1759 * expect to do this atomically or anything like that - but you can pass in
1760 * other page requirements.
1762 * If the page does not get brought uptodate, return -EIO.
1764 struct page *read_cache_page_gfp(struct address_space *mapping,
1768 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1770 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1772 EXPORT_SYMBOL(read_cache_page_gfp);
1775 * read_cache_page - read into page cache, fill it if needed
1776 * @mapping: the page's address_space
1777 * @index: the page index
1778 * @filler: function to perform the read
1779 * @data: destination for read data
1781 * Read into the page cache. If a page already exists, and PageUptodate() is
1782 * not set, try to fill the page then wait for it to become unlocked.
1784 * If the page does not get brought uptodate, return -EIO.
1786 struct page *read_cache_page(struct address_space *mapping,
1788 int (*filler)(void *,struct page*),
1791 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1793 EXPORT_SYMBOL(read_cache_page);
1796 * The logic we want is
1798 * if suid or (sgid and xgrp)
1801 int should_remove_suid(struct dentry *dentry)
1803 mode_t mode = dentry->d_inode->i_mode;
1806 /* suid always must be killed */
1807 if (unlikely(mode & S_ISUID))
1808 kill = ATTR_KILL_SUID;
1811 * sgid without any exec bits is just a mandatory locking mark; leave
1812 * it alone. If some exec bits are set, it's a real sgid; kill it.
1814 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1815 kill |= ATTR_KILL_SGID;
1817 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1822 EXPORT_SYMBOL(should_remove_suid);
1824 static int __remove_suid(struct dentry *dentry, int kill)
1826 struct iattr newattrs;
1828 newattrs.ia_valid = ATTR_FORCE | kill;
1829 return notify_change(dentry, &newattrs);
1832 int file_remove_suid(struct file *file)
1834 struct dentry *dentry = file->f_path.dentry;
1835 int killsuid = should_remove_suid(dentry);
1836 int killpriv = security_inode_need_killpriv(dentry);
1842 error = security_inode_killpriv(dentry);
1843 if (!error && killsuid)
1844 error = __remove_suid(dentry, killsuid);
1848 EXPORT_SYMBOL(file_remove_suid);
1850 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1851 const struct iovec *iov, size_t base, size_t bytes)
1853 size_t copied = 0, left = 0;
1856 char __user *buf = iov->iov_base + base;
1857 int copy = min(bytes, iov->iov_len - base);
1860 left = __copy_from_user_inatomic(vaddr, buf, copy);
1869 return copied - left;
1873 * Copy as much as we can into the page and return the number of bytes which
1874 * were successfully copied. If a fault is encountered then return the number of
1875 * bytes which were copied.
1877 size_t iov_iter_copy_from_user_atomic(struct page *page,
1878 struct iov_iter *i, unsigned long offset, size_t bytes)
1883 BUG_ON(!in_atomic());
1884 kaddr = kmap_atomic(page, KM_USER0);
1885 if (likely(i->nr_segs == 1)) {
1887 char __user *buf = i->iov->iov_base + i->iov_offset;
1888 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1889 copied = bytes - left;
1891 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1892 i->iov, i->iov_offset, bytes);
1894 kunmap_atomic(kaddr, KM_USER0);
1898 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1901 * This has the same sideeffects and return value as
1902 * iov_iter_copy_from_user_atomic().
1903 * The difference is that it attempts to resolve faults.
1904 * Page must not be locked.
1906 size_t iov_iter_copy_from_user(struct page *page,
1907 struct iov_iter *i, unsigned long offset, size_t bytes)
1913 if (likely(i->nr_segs == 1)) {
1915 char __user *buf = i->iov->iov_base + i->iov_offset;
1916 left = __copy_from_user(kaddr + offset, buf, bytes);
1917 copied = bytes - left;
1919 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1920 i->iov, i->iov_offset, bytes);
1925 EXPORT_SYMBOL(iov_iter_copy_from_user);
1927 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1929 BUG_ON(i->count < bytes);
1931 if (likely(i->nr_segs == 1)) {
1932 i->iov_offset += bytes;
1935 const struct iovec *iov = i->iov;
1936 size_t base = i->iov_offset;
1939 * The !iov->iov_len check ensures we skip over unlikely
1940 * zero-length segments (without overruning the iovec).
1942 while (bytes || unlikely(i->count && !iov->iov_len)) {
1945 copy = min(bytes, iov->iov_len - base);
1946 BUG_ON(!i->count || i->count < copy);
1950 if (iov->iov_len == base) {
1956 i->iov_offset = base;
1959 EXPORT_SYMBOL(iov_iter_advance);
1962 * Fault in the first iovec of the given iov_iter, to a maximum length
1963 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1964 * accessed (ie. because it is an invalid address).
1966 * writev-intensive code may want this to prefault several iovecs -- that
1967 * would be possible (callers must not rely on the fact that _only_ the
1968 * first iovec will be faulted with the current implementation).
1970 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
1972 char __user *buf = i->iov->iov_base + i->iov_offset;
1973 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1974 return fault_in_pages_readable(buf, bytes);
1976 EXPORT_SYMBOL(iov_iter_fault_in_readable);
1979 * Return the count of just the current iov_iter segment.
1981 size_t iov_iter_single_seg_count(struct iov_iter *i)
1983 const struct iovec *iov = i->iov;
1984 if (i->nr_segs == 1)
1987 return min(i->count, iov->iov_len - i->iov_offset);
1989 EXPORT_SYMBOL(iov_iter_single_seg_count);
1992 * Performs necessary checks before doing a write
1994 * Can adjust writing position or amount of bytes to write.
1995 * Returns appropriate error code that caller should return or
1996 * zero in case that write should be allowed.
1998 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2000 struct inode *inode = file->f_mapping->host;
2001 unsigned long limit = rlimit(RLIMIT_FSIZE);
2003 if (unlikely(*pos < 0))
2007 /* FIXME: this is for backwards compatibility with 2.4 */
2008 if (file->f_flags & O_APPEND)
2009 *pos = i_size_read(inode);
2011 if (limit != RLIM_INFINITY) {
2012 if (*pos >= limit) {
2013 send_sig(SIGXFSZ, current, 0);
2016 if (*count > limit - (typeof(limit))*pos) {
2017 *count = limit - (typeof(limit))*pos;
2025 if (unlikely(*pos + *count > MAX_NON_LFS &&
2026 !(file->f_flags & O_LARGEFILE))) {
2027 if (*pos >= MAX_NON_LFS) {
2030 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2031 *count = MAX_NON_LFS - (unsigned long)*pos;
2036 * Are we about to exceed the fs block limit ?
2038 * If we have written data it becomes a short write. If we have
2039 * exceeded without writing data we send a signal and return EFBIG.
2040 * Linus frestrict idea will clean these up nicely..
2042 if (likely(!isblk)) {
2043 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2044 if (*count || *pos > inode->i_sb->s_maxbytes) {
2047 /* zero-length writes at ->s_maxbytes are OK */
2050 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2051 *count = inode->i_sb->s_maxbytes - *pos;
2055 if (bdev_read_only(I_BDEV(inode)))
2057 isize = i_size_read(inode);
2058 if (*pos >= isize) {
2059 if (*count || *pos > isize)
2063 if (*pos + *count > isize)
2064 *count = isize - *pos;
2071 EXPORT_SYMBOL(generic_write_checks);
2073 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2074 loff_t pos, unsigned len, unsigned flags,
2075 struct page **pagep, void **fsdata)
2077 const struct address_space_operations *aops = mapping->a_ops;
2079 return aops->write_begin(file, mapping, pos, len, flags,
2082 EXPORT_SYMBOL(pagecache_write_begin);
2084 int pagecache_write_end(struct file *file, struct address_space *mapping,
2085 loff_t pos, unsigned len, unsigned copied,
2086 struct page *page, void *fsdata)
2088 const struct address_space_operations *aops = mapping->a_ops;
2090 mark_page_accessed(page);
2091 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2093 EXPORT_SYMBOL(pagecache_write_end);
2096 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2097 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2098 size_t count, size_t ocount)
2100 struct file *file = iocb->ki_filp;
2101 struct address_space *mapping = file->f_mapping;
2102 struct inode *inode = mapping->host;
2107 if (count != ocount)
2108 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2110 write_len = iov_length(iov, *nr_segs);
2111 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2113 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2118 * After a write we want buffered reads to be sure to go to disk to get
2119 * the new data. We invalidate clean cached page from the region we're
2120 * about to write. We do this *before* the write so that we can return
2121 * without clobbering -EIOCBQUEUED from ->direct_IO().
2123 if (mapping->nrpages) {
2124 written = invalidate_inode_pages2_range(mapping,
2125 pos >> PAGE_CACHE_SHIFT, end);
2127 * If a page can not be invalidated, return 0 to fall back
2128 * to buffered write.
2131 if (written == -EBUSY)
2137 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2140 * Finally, try again to invalidate clean pages which might have been
2141 * cached by non-direct readahead, or faulted in by get_user_pages()
2142 * if the source of the write was an mmap'ed region of the file
2143 * we're writing. Either one is a pretty crazy thing to do,
2144 * so we don't support it 100%. If this invalidation
2145 * fails, tough, the write still worked...
2147 if (mapping->nrpages) {
2148 invalidate_inode_pages2_range(mapping,
2149 pos >> PAGE_CACHE_SHIFT, end);
2154 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2155 i_size_write(inode, pos);
2156 mark_inode_dirty(inode);
2163 EXPORT_SYMBOL(generic_file_direct_write);
2166 * Find or create a page at the given pagecache position. Return the locked
2167 * page. This function is specifically for buffered writes.
2169 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2170 pgoff_t index, unsigned flags)
2174 gfp_t gfp_notmask = 0;
2175 if (flags & AOP_FLAG_NOFS)
2176 gfp_notmask = __GFP_FS;
2178 page = find_lock_page(mapping, index);
2182 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2185 status = add_to_page_cache_lru(page, mapping, index,
2186 GFP_KERNEL & ~gfp_notmask);
2187 if (unlikely(status)) {
2188 page_cache_release(page);
2189 if (status == -EEXIST)
2195 EXPORT_SYMBOL(grab_cache_page_write_begin);
2197 static ssize_t generic_perform_write(struct file *file,
2198 struct iov_iter *i, loff_t pos)
2200 struct address_space *mapping = file->f_mapping;
2201 const struct address_space_operations *a_ops = mapping->a_ops;
2203 ssize_t written = 0;
2204 unsigned int flags = 0;
2207 * Copies from kernel address space cannot fail (NFSD is a big user).
2209 if (segment_eq(get_fs(), KERNEL_DS))
2210 flags |= AOP_FLAG_UNINTERRUPTIBLE;
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 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2226 * Bring in the user page that we will copy from _first_.
2227 * Otherwise there's a nasty deadlock on copying from the
2228 * same page as we're writing to, without it being marked
2231 * Not only is this an optimisation, but it is also required
2232 * to check that the address is actually valid, when atomic
2233 * usercopies are used, below.
2235 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2240 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2242 if (unlikely(status))
2245 if (mapping_writably_mapped(mapping))
2246 flush_dcache_page(page);
2248 pagefault_disable();
2249 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2251 flush_dcache_page(page);
2253 mark_page_accessed(page);
2254 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2256 if (unlikely(status < 0))
2262 iov_iter_advance(i, copied);
2263 if (unlikely(copied == 0)) {
2265 * If we were unable to copy any data at all, we must
2266 * fall back to a single segment length write.
2268 * If we didn't fallback here, we could livelock
2269 * because not all segments in the iov can be copied at
2270 * once without a pagefault.
2272 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2273 iov_iter_single_seg_count(i));
2279 balance_dirty_pages_ratelimited(mapping);
2281 } while (iov_iter_count(i));
2283 return written ? written : status;
2287 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2288 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2289 size_t count, ssize_t written)
2291 struct file *file = iocb->ki_filp;
2295 iov_iter_init(&i, iov, nr_segs, count, written);
2296 status = generic_perform_write(file, &i, pos);
2298 if (likely(status >= 0)) {
2300 *ppos = pos + status;
2303 return written ? written : status;
2305 EXPORT_SYMBOL(generic_file_buffered_write);
2308 * __generic_file_aio_write - write data to a file
2309 * @iocb: IO state structure (file, offset, etc.)
2310 * @iov: vector with data to write
2311 * @nr_segs: number of segments in the vector
2312 * @ppos: position where to write
2314 * This function does all the work needed for actually writing data to a
2315 * file. It does all basic checks, removes SUID from the file, updates
2316 * modification times and calls proper subroutines depending on whether we
2317 * do direct IO or a standard buffered write.
2319 * It expects i_mutex to be grabbed unless we work on a block device or similar
2320 * object which does not need locking at all.
2322 * This function does *not* take care of syncing data in case of O_SYNC write.
2323 * A caller has to handle it. This is mainly due to the fact that we want to
2324 * avoid syncing under i_mutex.
2326 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2327 unsigned long nr_segs, loff_t *ppos)
2329 struct file *file = iocb->ki_filp;
2330 struct address_space * mapping = file->f_mapping;
2331 size_t ocount; /* original count */
2332 size_t count; /* after file limit checks */
2333 struct inode *inode = mapping->host;
2339 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2346 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2348 /* We can write back this queue in page reclaim */
2349 current->backing_dev_info = mapping->backing_dev_info;
2352 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2359 err = file_remove_suid(file);
2363 file_update_time(file);
2365 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2366 if (unlikely(file->f_flags & O_DIRECT)) {
2368 ssize_t written_buffered;
2370 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2371 ppos, count, ocount);
2372 if (written < 0 || written == count)
2375 * direct-io write to a hole: fall through to buffered I/O
2376 * for completing the rest of the request.
2380 written_buffered = generic_file_buffered_write(iocb, iov,
2381 nr_segs, pos, ppos, count,
2384 * If generic_file_buffered_write() retuned a synchronous error
2385 * then we want to return the number of bytes which were
2386 * direct-written, or the error code if that was zero. Note
2387 * that this differs from normal direct-io semantics, which
2388 * will return -EFOO even if some bytes were written.
2390 if (written_buffered < 0) {
2391 err = written_buffered;
2396 * We need to ensure that the page cache pages are written to
2397 * disk and invalidated to preserve the expected O_DIRECT
2400 endbyte = pos + written_buffered - written - 1;
2401 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2403 written = written_buffered;
2404 invalidate_mapping_pages(mapping,
2405 pos >> PAGE_CACHE_SHIFT,
2406 endbyte >> PAGE_CACHE_SHIFT);
2409 * We don't know how much we wrote, so just return
2410 * the number of bytes which were direct-written
2414 written = generic_file_buffered_write(iocb, iov, nr_segs,
2415 pos, ppos, count, written);
2418 current->backing_dev_info = NULL;
2419 return written ? written : err;
2421 EXPORT_SYMBOL(__generic_file_aio_write);
2424 * generic_file_aio_write - write data to a file
2425 * @iocb: IO state structure
2426 * @iov: vector with data to write
2427 * @nr_segs: number of segments in the vector
2428 * @pos: position in file where to write
2430 * This is a wrapper around __generic_file_aio_write() to be used by most
2431 * filesystems. It takes care of syncing the file in case of O_SYNC file
2432 * and acquires i_mutex as needed.
2434 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2435 unsigned long nr_segs, loff_t pos)
2437 struct file *file = iocb->ki_filp;
2438 struct inode *inode = file->f_mapping->host;
2439 struct blk_plug plug;
2442 BUG_ON(iocb->ki_pos != pos);
2444 mutex_lock(&inode->i_mutex);
2445 blk_start_plug(&plug);
2446 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2447 mutex_unlock(&inode->i_mutex);
2449 if (ret > 0 || ret == -EIOCBQUEUED) {
2452 err = generic_write_sync(file, pos, ret);
2453 if (err < 0 && ret > 0)
2456 blk_finish_plug(&plug);
2459 EXPORT_SYMBOL(generic_file_aio_write);
2462 * try_to_release_page() - release old fs-specific metadata on a page
2464 * @page: the page which the kernel is trying to free
2465 * @gfp_mask: memory allocation flags (and I/O mode)
2467 * The address_space is to try to release any data against the page
2468 * (presumably at page->private). If the release was successful, return `1'.
2469 * Otherwise return zero.
2471 * This may also be called if PG_fscache is set on a page, indicating that the
2472 * page is known to the local caching routines.
2474 * The @gfp_mask argument specifies whether I/O may be performed to release
2475 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2478 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2480 struct address_space * const mapping = page->mapping;
2482 BUG_ON(!PageLocked(page));
2483 if (PageWriteback(page))
2486 if (mapping && mapping->a_ops->releasepage)
2487 return mapping->a_ops->releasepage(page, gfp_mask);
2488 return try_to_free_buffers(page);
2491 EXPORT_SYMBOL(try_to_release_page);