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/export.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/cpuset.h>
33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
34 #include <linux/memcontrol.h>
35 #include <linux/cleancache.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/filemap.h>
42 * FIXME: remove all knowledge of the buffer layer from the core VM
44 #include <linux/buffer_head.h> /* for try_to_free_buffers */
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * Shared mappings now work. 15.8.1995 Bruno.
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
63 * ->i_mmap_mutex (truncate_pagecache)
64 * ->private_lock (__free_pte->__set_page_dirty_buffers)
65 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
69 * ->i_mmap_mutex (truncate->unmap_mapping_range)
73 * ->page_table_lock or pte_lock (various, mainly in memory.c)
74 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 * ->lock_page (access_process_vm)
79 * ->i_mutex (generic_file_buffered_write)
80 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
83 * sb_lock (fs/fs-writeback.c)
84 * ->mapping->tree_lock (__sync_single_inode)
87 * ->anon_vma.lock (vma_adjust)
90 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
92 * ->page_table_lock or pte_lock
93 * ->swap_lock (try_to_unmap_one)
94 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
97 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
98 * ->private_lock (page_remove_rmap->set_page_dirty)
99 * ->tree_lock (page_remove_rmap->set_page_dirty)
100 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
101 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
103 * ->inode->i_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 * Delete 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 __delete_from_page_cache(struct page
*page
)
117 struct address_space
*mapping
= page
->mapping
;
119 trace_mm_filemap_delete_from_page_cache(page
);
121 * if we're uptodate, flush out into the cleancache, otherwise
122 * invalidate any existing cleancache entries. We can't leave
123 * stale data around in the cleancache once our page is gone
125 if (PageUptodate(page
) && PageMappedToDisk(page
))
126 cleancache_put_page(page
);
128 cleancache_invalidate_page(mapping
, page
);
130 radix_tree_delete(&mapping
->page_tree
, page
->index
);
131 page
->mapping
= NULL
;
132 /* Leave page->index set: truncation lookup relies upon it */
134 __dec_zone_page_state(page
, NR_FILE_PAGES
);
135 if (PageSwapBacked(page
))
136 __dec_zone_page_state(page
, NR_SHMEM
);
137 BUG_ON(page_mapped(page
));
140 * Some filesystems seem to re-dirty the page even after
141 * the VM has canceled the dirty bit (eg ext3 journaling).
143 * Fix it up by doing a final dirty accounting check after
144 * having removed the page entirely.
146 if (PageDirty(page
) && mapping_cap_account_dirty(mapping
)) {
147 dec_zone_page_state(page
, NR_FILE_DIRTY
);
148 dec_bdi_stat(mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
153 * delete_from_page_cache - delete page from page cache
154 * @page: the page which the kernel is trying to remove from page cache
156 * This must be called only on pages that have been verified to be in the page
157 * cache and locked. It will never put the page into the free list, the caller
158 * has a reference on the page.
160 void delete_from_page_cache(struct page
*page
)
162 struct address_space
*mapping
= page
->mapping
;
163 void (*freepage
)(struct page
*);
165 BUG_ON(!PageLocked(page
));
167 freepage
= mapping
->a_ops
->freepage
;
168 spin_lock_irq(&mapping
->tree_lock
);
169 __delete_from_page_cache(page
);
170 spin_unlock_irq(&mapping
->tree_lock
);
171 mem_cgroup_uncharge_cache_page(page
);
175 page_cache_release(page
);
177 EXPORT_SYMBOL(delete_from_page_cache
);
179 static int sleep_on_page(void *word
)
185 static int sleep_on_page_killable(void *word
)
188 return fatal_signal_pending(current
) ? -EINTR
: 0;
191 static int filemap_check_errors(struct address_space
*mapping
)
194 /* Check for outstanding write errors */
195 if (test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
197 if (test_and_clear_bit(AS_EIO
, &mapping
->flags
))
203 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
204 * @mapping: address space structure to write
205 * @start: offset in bytes where the range starts
206 * @end: offset in bytes where the range ends (inclusive)
207 * @sync_mode: enable synchronous operation
209 * Start writeback against all of a mapping's dirty pages that lie
210 * within the byte offsets <start, end> inclusive.
212 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
213 * opposed to a regular memory cleansing writeback. The difference between
214 * these two operations is that if a dirty page/buffer is encountered, it must
215 * be waited upon, and not just skipped over.
217 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
218 loff_t end
, int sync_mode
)
221 struct writeback_control wbc
= {
222 .sync_mode
= sync_mode
,
223 .nr_to_write
= LONG_MAX
,
224 .range_start
= start
,
228 if (!mapping_cap_writeback_dirty(mapping
))
231 ret
= do_writepages(mapping
, &wbc
);
235 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
238 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
241 int filemap_fdatawrite(struct address_space
*mapping
)
243 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
245 EXPORT_SYMBOL(filemap_fdatawrite
);
247 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
250 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
252 EXPORT_SYMBOL(filemap_fdatawrite_range
);
255 * filemap_flush - mostly a non-blocking flush
256 * @mapping: target address_space
258 * This is a mostly non-blocking flush. Not suitable for data-integrity
259 * purposes - I/O may not be started against all dirty pages.
261 int filemap_flush(struct address_space
*mapping
)
263 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
265 EXPORT_SYMBOL(filemap_flush
);
268 * filemap_fdatawait_range - wait for writeback to complete
269 * @mapping: address space structure to wait for
270 * @start_byte: offset in bytes where the range starts
271 * @end_byte: offset in bytes where the range ends (inclusive)
273 * Walk the list of under-writeback pages of the given address space
274 * in the given range and wait for all of them.
276 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
279 pgoff_t index
= start_byte
>> PAGE_CACHE_SHIFT
;
280 pgoff_t end
= end_byte
>> PAGE_CACHE_SHIFT
;
285 if (end_byte
< start_byte
)
288 pagevec_init(&pvec
, 0);
289 while ((index
<= end
) &&
290 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
291 PAGECACHE_TAG_WRITEBACK
,
292 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
295 for (i
= 0; i
< nr_pages
; i
++) {
296 struct page
*page
= pvec
.pages
[i
];
298 /* until radix tree lookup accepts end_index */
299 if (page
->index
> end
)
302 wait_on_page_writeback(page
);
303 if (TestClearPageError(page
))
306 pagevec_release(&pvec
);
310 ret2
= filemap_check_errors(mapping
);
316 EXPORT_SYMBOL(filemap_fdatawait_range
);
319 * filemap_fdatawait - wait for all under-writeback pages to complete
320 * @mapping: address space structure to wait for
322 * Walk the list of under-writeback pages of the given address space
323 * and wait for all of them.
325 int filemap_fdatawait(struct address_space
*mapping
)
327 loff_t i_size
= i_size_read(mapping
->host
);
332 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
334 EXPORT_SYMBOL(filemap_fdatawait
);
336 int filemap_write_and_wait(struct address_space
*mapping
)
340 if (mapping
->nrpages
) {
341 err
= filemap_fdatawrite(mapping
);
343 * Even if the above returned error, the pages may be
344 * written partially (e.g. -ENOSPC), so we wait for it.
345 * But the -EIO is special case, it may indicate the worst
346 * thing (e.g. bug) happened, so we avoid waiting for it.
349 int err2
= filemap_fdatawait(mapping
);
354 err
= filemap_check_errors(mapping
);
358 EXPORT_SYMBOL(filemap_write_and_wait
);
361 * filemap_write_and_wait_range - write out & wait on a file range
362 * @mapping: the address_space for the pages
363 * @lstart: offset in bytes where the range starts
364 * @lend: offset in bytes where the range ends (inclusive)
366 * Write out and wait upon file offsets lstart->lend, inclusive.
368 * Note that `lend' is inclusive (describes the last byte to be written) so
369 * that this function can be used to write to the very end-of-file (end = -1).
371 int filemap_write_and_wait_range(struct address_space
*mapping
,
372 loff_t lstart
, loff_t lend
)
376 if (mapping
->nrpages
) {
377 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
379 /* See comment of filemap_write_and_wait() */
381 int err2
= filemap_fdatawait_range(mapping
,
387 err
= filemap_check_errors(mapping
);
391 EXPORT_SYMBOL(filemap_write_and_wait_range
);
394 * replace_page_cache_page - replace a pagecache page with a new one
395 * @old: page to be replaced
396 * @new: page to replace with
397 * @gfp_mask: allocation mode
399 * This function replaces a page in the pagecache with a new one. On
400 * success it acquires the pagecache reference for the new page and
401 * drops it for the old page. Both the old and new pages must be
402 * locked. This function does not add the new page to the LRU, the
403 * caller must do that.
405 * The remove + add is atomic. The only way this function can fail is
406 * memory allocation failure.
408 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
412 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
413 VM_BUG_ON_PAGE(!PageLocked(new), new);
414 VM_BUG_ON_PAGE(new->mapping
, new);
416 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
418 struct address_space
*mapping
= old
->mapping
;
419 void (*freepage
)(struct page
*);
421 pgoff_t offset
= old
->index
;
422 freepage
= mapping
->a_ops
->freepage
;
425 new->mapping
= mapping
;
428 spin_lock_irq(&mapping
->tree_lock
);
429 __delete_from_page_cache(old
);
430 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
433 __inc_zone_page_state(new, NR_FILE_PAGES
);
434 if (PageSwapBacked(new))
435 __inc_zone_page_state(new, NR_SHMEM
);
436 spin_unlock_irq(&mapping
->tree_lock
);
437 /* mem_cgroup codes must not be called under tree_lock */
438 mem_cgroup_replace_page_cache(old
, new);
439 radix_tree_preload_end();
442 page_cache_release(old
);
447 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
450 * add_to_page_cache_locked - add a locked page to the pagecache
452 * @mapping: the page's address_space
453 * @offset: page index
454 * @gfp_mask: page allocation mode
456 * This function is used to add a page to the pagecache. It must be locked.
457 * This function does not add the page to the LRU. The caller must do that.
459 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
460 pgoff_t offset
, gfp_t gfp_mask
)
464 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
465 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
467 error
= mem_cgroup_cache_charge(page
, current
->mm
,
468 gfp_mask
& GFP_RECLAIM_MASK
);
472 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
474 mem_cgroup_uncharge_cache_page(page
);
478 page_cache_get(page
);
479 page
->mapping
= mapping
;
480 page
->index
= offset
;
482 spin_lock_irq(&mapping
->tree_lock
);
483 error
= radix_tree_insert(&mapping
->page_tree
, offset
, page
);
484 radix_tree_preload_end();
488 __inc_zone_page_state(page
, NR_FILE_PAGES
);
489 spin_unlock_irq(&mapping
->tree_lock
);
490 trace_mm_filemap_add_to_page_cache(page
);
493 page
->mapping
= NULL
;
494 /* Leave page->index set: truncation relies upon it */
495 spin_unlock_irq(&mapping
->tree_lock
);
496 mem_cgroup_uncharge_cache_page(page
);
497 page_cache_release(page
);
500 EXPORT_SYMBOL(add_to_page_cache_locked
);
502 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
503 pgoff_t offset
, gfp_t gfp_mask
)
507 ret
= add_to_page_cache(page
, mapping
, offset
, gfp_mask
);
509 lru_cache_add_file(page
);
512 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
515 struct page
*__page_cache_alloc(gfp_t gfp
)
520 if (cpuset_do_page_mem_spread()) {
521 unsigned int cpuset_mems_cookie
;
523 cpuset_mems_cookie
= get_mems_allowed();
524 n
= cpuset_mem_spread_node();
525 page
= alloc_pages_exact_node(n
, gfp
, 0);
526 } while (!put_mems_allowed(cpuset_mems_cookie
) && !page
);
530 return alloc_pages(gfp
, 0);
532 EXPORT_SYMBOL(__page_cache_alloc
);
536 * In order to wait for pages to become available there must be
537 * waitqueues associated with pages. By using a hash table of
538 * waitqueues where the bucket discipline is to maintain all
539 * waiters on the same queue and wake all when any of the pages
540 * become available, and for the woken contexts to check to be
541 * sure the appropriate page became available, this saves space
542 * at a cost of "thundering herd" phenomena during rare hash
545 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
547 const struct zone
*zone
= page_zone(page
);
549 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
552 static inline void wake_up_page(struct page
*page
, int bit
)
554 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
557 void wait_on_page_bit(struct page
*page
, int bit_nr
)
559 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
561 if (test_bit(bit_nr
, &page
->flags
))
562 __wait_on_bit(page_waitqueue(page
), &wait
, sleep_on_page
,
563 TASK_UNINTERRUPTIBLE
);
565 EXPORT_SYMBOL(wait_on_page_bit
);
567 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
569 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
571 if (!test_bit(bit_nr
, &page
->flags
))
574 return __wait_on_bit(page_waitqueue(page
), &wait
,
575 sleep_on_page_killable
, TASK_KILLABLE
);
579 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
580 * @page: Page defining the wait queue of interest
581 * @waiter: Waiter to add to the queue
583 * Add an arbitrary @waiter to the wait queue for the nominated @page.
585 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
587 wait_queue_head_t
*q
= page_waitqueue(page
);
590 spin_lock_irqsave(&q
->lock
, flags
);
591 __add_wait_queue(q
, waiter
);
592 spin_unlock_irqrestore(&q
->lock
, flags
);
594 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
597 * unlock_page - unlock a locked page
600 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
601 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
602 * mechananism between PageLocked pages and PageWriteback pages is shared.
603 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
605 * The mb is necessary to enforce ordering between the clear_bit and the read
606 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
608 void unlock_page(struct page
*page
)
610 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
611 clear_bit_unlock(PG_locked
, &page
->flags
);
612 smp_mb__after_clear_bit();
613 wake_up_page(page
, PG_locked
);
615 EXPORT_SYMBOL(unlock_page
);
618 * end_page_writeback - end writeback against a page
621 void end_page_writeback(struct page
*page
)
623 if (TestClearPageReclaim(page
))
624 rotate_reclaimable_page(page
);
626 if (!test_clear_page_writeback(page
))
629 smp_mb__after_clear_bit();
630 wake_up_page(page
, PG_writeback
);
632 EXPORT_SYMBOL(end_page_writeback
);
635 * __lock_page - get a lock on the page, assuming we need to sleep to get it
636 * @page: the page to lock
638 void __lock_page(struct page
*page
)
640 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
642 __wait_on_bit_lock(page_waitqueue(page
), &wait
, sleep_on_page
,
643 TASK_UNINTERRUPTIBLE
);
645 EXPORT_SYMBOL(__lock_page
);
647 int __lock_page_killable(struct page
*page
)
649 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
651 return __wait_on_bit_lock(page_waitqueue(page
), &wait
,
652 sleep_on_page_killable
, TASK_KILLABLE
);
654 EXPORT_SYMBOL_GPL(__lock_page_killable
);
656 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
659 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
661 * CAUTION! In this case, mmap_sem is not released
662 * even though return 0.
664 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
667 up_read(&mm
->mmap_sem
);
668 if (flags
& FAULT_FLAG_KILLABLE
)
669 wait_on_page_locked_killable(page
);
671 wait_on_page_locked(page
);
674 if (flags
& FAULT_FLAG_KILLABLE
) {
677 ret
= __lock_page_killable(page
);
679 up_read(&mm
->mmap_sem
);
689 * find_get_page - find and get a page reference
690 * @mapping: the address_space to search
691 * @offset: the page index
693 * Is there a pagecache struct page at the given (mapping, offset) tuple?
694 * If yes, increment its refcount and return it; if no, return NULL.
696 struct page
*find_get_page(struct address_space
*mapping
, pgoff_t offset
)
704 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
706 page
= radix_tree_deref_slot(pagep
);
709 if (radix_tree_exception(page
)) {
710 if (radix_tree_deref_retry(page
))
713 * Otherwise, shmem/tmpfs must be storing a swap entry
714 * here as an exceptional entry: so return it without
715 * attempting to raise page count.
719 if (!page_cache_get_speculative(page
))
723 * Has the page moved?
724 * This is part of the lockless pagecache protocol. See
725 * include/linux/pagemap.h for details.
727 if (unlikely(page
!= *pagep
)) {
728 page_cache_release(page
);
737 EXPORT_SYMBOL(find_get_page
);
740 * find_lock_page - locate, pin and lock a pagecache page
741 * @mapping: the address_space to search
742 * @offset: the page index
744 * Locates the desired pagecache page, locks it, increments its reference
745 * count and returns its address.
747 * Returns zero if the page was not present. find_lock_page() may sleep.
749 struct page
*find_lock_page(struct address_space
*mapping
, pgoff_t offset
)
754 page
= find_get_page(mapping
, offset
);
755 if (page
&& !radix_tree_exception(page
)) {
757 /* Has the page been truncated? */
758 if (unlikely(page
->mapping
!= mapping
)) {
760 page_cache_release(page
);
763 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
767 EXPORT_SYMBOL(find_lock_page
);
770 * find_or_create_page - locate or add a pagecache page
771 * @mapping: the page's address_space
772 * @index: the page's index into the mapping
773 * @gfp_mask: page allocation mode
775 * Locates a page in the pagecache. If the page is not present, a new page
776 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
777 * LRU list. The returned page is locked and has its reference count
780 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
783 * find_or_create_page() returns the desired page's address, or zero on
786 struct page
*find_or_create_page(struct address_space
*mapping
,
787 pgoff_t index
, gfp_t gfp_mask
)
792 page
= find_lock_page(mapping
, index
);
794 page
= __page_cache_alloc(gfp_mask
);
798 * We want a regular kernel memory (not highmem or DMA etc)
799 * allocation for the radix tree nodes, but we need to honour
800 * the context-specific requirements the caller has asked for.
801 * GFP_RECLAIM_MASK collects those requirements.
803 err
= add_to_page_cache_lru(page
, mapping
, index
,
804 (gfp_mask
& GFP_RECLAIM_MASK
));
806 page_cache_release(page
);
814 EXPORT_SYMBOL(find_or_create_page
);
817 * find_get_pages - gang pagecache lookup
818 * @mapping: The address_space to search
819 * @start: The starting page index
820 * @nr_pages: The maximum number of pages
821 * @pages: Where the resulting pages are placed
823 * find_get_pages() will search for and return a group of up to
824 * @nr_pages pages in the mapping. The pages are placed at @pages.
825 * find_get_pages() takes a reference against the returned pages.
827 * The search returns a group of mapping-contiguous pages with ascending
828 * indexes. There may be holes in the indices due to not-present pages.
830 * find_get_pages() returns the number of pages which were found.
832 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
833 unsigned int nr_pages
, struct page
**pages
)
835 struct radix_tree_iter iter
;
839 if (unlikely(!nr_pages
))
844 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
847 page
= radix_tree_deref_slot(slot
);
851 if (radix_tree_exception(page
)) {
852 if (radix_tree_deref_retry(page
)) {
854 * Transient condition which can only trigger
855 * when entry at index 0 moves out of or back
856 * to root: none yet gotten, safe to restart.
862 * Otherwise, shmem/tmpfs must be storing a swap entry
863 * here as an exceptional entry: so skip over it -
864 * we only reach this from invalidate_mapping_pages().
869 if (!page_cache_get_speculative(page
))
872 /* Has the page moved? */
873 if (unlikely(page
!= *slot
)) {
874 page_cache_release(page
);
879 if (++ret
== nr_pages
)
888 * find_get_pages_contig - gang contiguous pagecache lookup
889 * @mapping: The address_space to search
890 * @index: The starting page index
891 * @nr_pages: The maximum number of pages
892 * @pages: Where the resulting pages are placed
894 * find_get_pages_contig() works exactly like find_get_pages(), except
895 * that the returned number of pages are guaranteed to be contiguous.
897 * find_get_pages_contig() returns the number of pages which were found.
899 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
900 unsigned int nr_pages
, struct page
**pages
)
902 struct radix_tree_iter iter
;
904 unsigned int ret
= 0;
906 if (unlikely(!nr_pages
))
911 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
914 page
= radix_tree_deref_slot(slot
);
915 /* The hole, there no reason to continue */
919 if (radix_tree_exception(page
)) {
920 if (radix_tree_deref_retry(page
)) {
922 * Transient condition which can only trigger
923 * when entry at index 0 moves out of or back
924 * to root: none yet gotten, safe to restart.
929 * Otherwise, shmem/tmpfs must be storing a swap entry
930 * here as an exceptional entry: so stop looking for
936 if (!page_cache_get_speculative(page
))
939 /* Has the page moved? */
940 if (unlikely(page
!= *slot
)) {
941 page_cache_release(page
);
946 * must check mapping and index after taking the ref.
947 * otherwise we can get both false positives and false
948 * negatives, which is just confusing to the caller.
950 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
951 page_cache_release(page
);
956 if (++ret
== nr_pages
)
962 EXPORT_SYMBOL(find_get_pages_contig
);
965 * find_get_pages_tag - find and return pages that match @tag
966 * @mapping: the address_space to search
967 * @index: the starting page index
968 * @tag: the tag index
969 * @nr_pages: the maximum number of pages
970 * @pages: where the resulting pages are placed
972 * Like find_get_pages, except we only return pages which are tagged with
973 * @tag. We update @index to index the next page for the traversal.
975 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
976 int tag
, unsigned int nr_pages
, struct page
**pages
)
978 struct radix_tree_iter iter
;
982 if (unlikely(!nr_pages
))
987 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
988 &iter
, *index
, tag
) {
991 page
= radix_tree_deref_slot(slot
);
995 if (radix_tree_exception(page
)) {
996 if (radix_tree_deref_retry(page
)) {
998 * Transient condition which can only trigger
999 * when entry at index 0 moves out of or back
1000 * to root: none yet gotten, safe to restart.
1005 * This function is never used on a shmem/tmpfs
1006 * mapping, so a swap entry won't be found here.
1011 if (!page_cache_get_speculative(page
))
1014 /* Has the page moved? */
1015 if (unlikely(page
!= *slot
)) {
1016 page_cache_release(page
);
1021 if (++ret
== nr_pages
)
1028 *index
= pages
[ret
- 1]->index
+ 1;
1032 EXPORT_SYMBOL(find_get_pages_tag
);
1035 * grab_cache_page_nowait - returns locked page at given index in given cache
1036 * @mapping: target address_space
1037 * @index: the page index
1039 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1040 * This is intended for speculative data generators, where the data can
1041 * be regenerated if the page couldn't be grabbed. This routine should
1042 * be safe to call while holding the lock for another page.
1044 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1045 * and deadlock against the caller's locked page.
1048 grab_cache_page_nowait(struct address_space
*mapping
, pgoff_t index
)
1050 struct page
*page
= find_get_page(mapping
, index
);
1053 if (trylock_page(page
))
1055 page_cache_release(page
);
1058 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
1059 if (page
&& add_to_page_cache_lru(page
, mapping
, index
, GFP_NOFS
)) {
1060 page_cache_release(page
);
1065 EXPORT_SYMBOL(grab_cache_page_nowait
);
1068 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1069 * a _large_ part of the i/o request. Imagine the worst scenario:
1071 * ---R__________________________________________B__________
1072 * ^ reading here ^ bad block(assume 4k)
1074 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1075 * => failing the whole request => read(R) => read(R+1) =>
1076 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1077 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1078 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1080 * It is going insane. Fix it by quickly scaling down the readahead size.
1082 static void shrink_readahead_size_eio(struct file
*filp
,
1083 struct file_ra_state
*ra
)
1088 size_t copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
1091 size_t skip
, copy
, left
, wanted
;
1092 const struct iovec
*iov
;
1096 if (unlikely(bytes
> i
->count
))
1099 if (unlikely(!bytes
))
1104 skip
= i
->iov_offset
;
1105 buf
= iov
->iov_base
+ skip
;
1106 copy
= min(bytes
, iov
->iov_len
- skip
);
1108 if (!fault_in_pages_writeable(buf
, copy
)) {
1109 kaddr
= kmap_atomic(page
);
1110 from
= kaddr
+ offset
;
1112 /* first chunk, usually the only one */
1113 left
= __copy_to_user_inatomic(buf
, from
, copy
);
1119 while (unlikely(!left
&& bytes
)) {
1121 buf
= iov
->iov_base
;
1122 copy
= min(bytes
, iov
->iov_len
);
1123 left
= __copy_to_user_inatomic(buf
, from
, copy
);
1129 if (likely(!bytes
)) {
1130 kunmap_atomic(kaddr
);
1133 offset
= from
- kaddr
;
1135 kunmap_atomic(kaddr
);
1136 copy
= min(bytes
, iov
->iov_len
- skip
);
1138 /* Too bad - revert to non-atomic kmap */
1140 from
= kaddr
+ offset
;
1141 left
= __copy_to_user(buf
, from
, copy
);
1146 while (unlikely(!left
&& bytes
)) {
1148 buf
= iov
->iov_base
;
1149 copy
= min(bytes
, iov
->iov_len
);
1150 left
= __copy_to_user(buf
, from
, copy
);
1158 i
->count
-= wanted
- bytes
;
1159 i
->nr_segs
-= iov
- i
->iov
;
1161 i
->iov_offset
= skip
;
1162 return wanted
- bytes
;
1164 EXPORT_SYMBOL(copy_page_to_iter
);
1167 * do_generic_file_read - generic file read routine
1168 * @filp: the file to read
1169 * @ppos: current file position
1170 * @iter: data destination
1171 * @written: already copied
1173 * This is a generic file read routine, and uses the
1174 * mapping->a_ops->readpage() function for the actual low-level stuff.
1176 * This is really ugly. But the goto's actually try to clarify some
1177 * of the logic when it comes to error handling etc.
1179 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1180 struct iov_iter
*iter
, ssize_t written
)
1182 struct address_space
*mapping
= filp
->f_mapping
;
1183 struct inode
*inode
= mapping
->host
;
1184 struct file_ra_state
*ra
= &filp
->f_ra
;
1188 unsigned long offset
; /* offset into pagecache page */
1189 unsigned int prev_offset
;
1192 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1193 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1194 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1195 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1196 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1202 unsigned long nr
, ret
;
1206 page
= find_get_page(mapping
, index
);
1208 page_cache_sync_readahead(mapping
,
1210 index
, last_index
- index
);
1211 page
= find_get_page(mapping
, index
);
1212 if (unlikely(page
== NULL
))
1213 goto no_cached_page
;
1215 if (PageReadahead(page
)) {
1216 page_cache_async_readahead(mapping
,
1218 index
, last_index
- index
);
1220 if (!PageUptodate(page
)) {
1221 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1222 !mapping
->a_ops
->is_partially_uptodate
)
1223 goto page_not_up_to_date
;
1224 if (!trylock_page(page
))
1225 goto page_not_up_to_date
;
1226 /* Did it get truncated before we got the lock? */
1228 goto page_not_up_to_date_locked
;
1229 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1230 offset
, iter
->count
))
1231 goto page_not_up_to_date_locked
;
1236 * i_size must be checked after we know the page is Uptodate.
1238 * Checking i_size after the check allows us to calculate
1239 * the correct value for "nr", which means the zero-filled
1240 * part of the page is not copied back to userspace (unless
1241 * another truncate extends the file - this is desired though).
1244 isize
= i_size_read(inode
);
1245 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1246 if (unlikely(!isize
|| index
> end_index
)) {
1247 page_cache_release(page
);
1251 /* nr is the maximum number of bytes to copy from this page */
1252 nr
= PAGE_CACHE_SIZE
;
1253 if (index
== end_index
) {
1254 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1256 page_cache_release(page
);
1262 /* If users can be writing to this page using arbitrary
1263 * virtual addresses, take care about potential aliasing
1264 * before reading the page on the kernel side.
1266 if (mapping_writably_mapped(mapping
))
1267 flush_dcache_page(page
);
1270 * When a sequential read accesses a page several times,
1271 * only mark it as accessed the first time.
1273 if (prev_index
!= index
|| offset
!= prev_offset
)
1274 mark_page_accessed(page
);
1278 * Ok, we have the page, and it's up-to-date, so
1279 * now we can copy it to user space...
1282 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1284 index
+= offset
>> PAGE_CACHE_SHIFT
;
1285 offset
&= ~PAGE_CACHE_MASK
;
1286 prev_offset
= offset
;
1288 page_cache_release(page
);
1290 if (!iov_iter_count(iter
))
1298 page_not_up_to_date
:
1299 /* Get exclusive access to the page ... */
1300 error
= lock_page_killable(page
);
1301 if (unlikely(error
))
1302 goto readpage_error
;
1304 page_not_up_to_date_locked
:
1305 /* Did it get truncated before we got the lock? */
1306 if (!page
->mapping
) {
1308 page_cache_release(page
);
1312 /* Did somebody else fill it already? */
1313 if (PageUptodate(page
)) {
1320 * A previous I/O error may have been due to temporary
1321 * failures, eg. multipath errors.
1322 * PG_error will be set again if readpage fails.
1324 ClearPageError(page
);
1325 /* Start the actual read. The read will unlock the page. */
1326 error
= mapping
->a_ops
->readpage(filp
, page
);
1328 if (unlikely(error
)) {
1329 if (error
== AOP_TRUNCATED_PAGE
) {
1330 page_cache_release(page
);
1334 goto readpage_error
;
1337 if (!PageUptodate(page
)) {
1338 error
= lock_page_killable(page
);
1339 if (unlikely(error
))
1340 goto readpage_error
;
1341 if (!PageUptodate(page
)) {
1342 if (page
->mapping
== NULL
) {
1344 * invalidate_mapping_pages got it
1347 page_cache_release(page
);
1351 shrink_readahead_size_eio(filp
, ra
);
1353 goto readpage_error
;
1361 /* UHHUH! A synchronous read error occurred. Report it */
1362 page_cache_release(page
);
1367 * Ok, it wasn't cached, so we need to create a new
1370 page
= page_cache_alloc_cold(mapping
);
1375 error
= add_to_page_cache_lru(page
, mapping
,
1378 page_cache_release(page
);
1379 if (error
== -EEXIST
) {
1389 ra
->prev_pos
= prev_index
;
1390 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1391 ra
->prev_pos
|= prev_offset
;
1393 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1394 file_accessed(filp
);
1395 return written
? written
: error
;
1399 * Performs necessary checks before doing a write
1400 * @iov: io vector request
1401 * @nr_segs: number of segments in the iovec
1402 * @count: number of bytes to write
1403 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1405 * Adjust number of segments and amount of bytes to write (nr_segs should be
1406 * properly initialized first). Returns appropriate error code that caller
1407 * should return or zero in case that write should be allowed.
1409 int generic_segment_checks(const struct iovec
*iov
,
1410 unsigned long *nr_segs
, size_t *count
, int access_flags
)
1414 for (seg
= 0; seg
< *nr_segs
; seg
++) {
1415 const struct iovec
*iv
= &iov
[seg
];
1418 * If any segment has a negative length, or the cumulative
1419 * length ever wraps negative then return -EINVAL.
1422 if (unlikely((ssize_t
)(cnt
|iv
->iov_len
) < 0))
1424 if (access_ok(access_flags
, iv
->iov_base
, iv
->iov_len
))
1429 cnt
-= iv
->iov_len
; /* This segment is no good */
1435 EXPORT_SYMBOL(generic_segment_checks
);
1438 * generic_file_aio_read - generic filesystem read routine
1439 * @iocb: kernel I/O control block
1440 * @iov: io vector request
1441 * @nr_segs: number of segments in the iovec
1442 * @pos: current file position
1444 * This is the "read()" routine for all filesystems
1445 * that can use the page cache directly.
1448 generic_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1449 unsigned long nr_segs
, loff_t pos
)
1451 struct file
*filp
= iocb
->ki_filp
;
1454 loff_t
*ppos
= &iocb
->ki_pos
;
1458 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1461 iov_iter_init(&i
, iov
, nr_segs
, count
, 0);
1463 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1464 if (filp
->f_flags
& O_DIRECT
) {
1466 struct address_space
*mapping
;
1467 struct inode
*inode
;
1469 mapping
= filp
->f_mapping
;
1470 inode
= mapping
->host
;
1472 goto out
; /* skip atime */
1473 size
= i_size_read(inode
);
1474 retval
= filemap_write_and_wait_range(mapping
, pos
,
1475 pos
+ iov_length(iov
, nr_segs
) - 1);
1477 retval
= mapping
->a_ops
->direct_IO(READ
, iocb
,
1481 *ppos
= pos
+ retval
;
1484 * If we did a short DIO read we need to skip the
1485 * section of the iov that we've already read data into.
1487 iov_iter_advance(&i
, retval
);
1491 * Btrfs can have a short DIO read if we encounter
1492 * compressed extents, so if there was an error, or if
1493 * we've already read everything we wanted to, or if
1494 * there was a short read because we hit EOF, go ahead
1495 * and return. Otherwise fallthrough to buffered io for
1496 * the rest of the read.
1498 if (retval
< 0 || !count
|| *ppos
>= size
) {
1499 file_accessed(filp
);
1504 retval
= do_generic_file_read(filp
, ppos
, &i
, retval
);
1508 EXPORT_SYMBOL(generic_file_aio_read
);
1512 * page_cache_read - adds requested page to the page cache if not already there
1513 * @file: file to read
1514 * @offset: page index
1516 * This adds the requested page to the page cache if it isn't already there,
1517 * and schedules an I/O to read in its contents from disk.
1519 static int page_cache_read(struct file
*file
, pgoff_t offset
)
1521 struct address_space
*mapping
= file
->f_mapping
;
1526 page
= page_cache_alloc_cold(mapping
);
1530 ret
= add_to_page_cache_lru(page
, mapping
, offset
, GFP_KERNEL
);
1532 ret
= mapping
->a_ops
->readpage(file
, page
);
1533 else if (ret
== -EEXIST
)
1534 ret
= 0; /* losing race to add is OK */
1536 page_cache_release(page
);
1538 } while (ret
== AOP_TRUNCATED_PAGE
);
1543 #define MMAP_LOTSAMISS (100)
1546 * Synchronous readahead happens when we don't even find
1547 * a page in the page cache at all.
1549 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1550 struct file_ra_state
*ra
,
1554 unsigned long ra_pages
;
1555 struct address_space
*mapping
= file
->f_mapping
;
1557 /* If we don't want any read-ahead, don't bother */
1558 if (vma
->vm_flags
& VM_RAND_READ
)
1563 if (vma
->vm_flags
& VM_SEQ_READ
) {
1564 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1569 /* Avoid banging the cache line if not needed */
1570 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1574 * Do we miss much more than hit in this file? If so,
1575 * stop bothering with read-ahead. It will only hurt.
1577 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1583 ra_pages
= max_sane_readahead(ra
->ra_pages
);
1584 ra
->start
= max_t(long, 0, offset
- ra_pages
/ 2);
1585 ra
->size
= ra_pages
;
1586 ra
->async_size
= ra_pages
/ 4;
1587 ra_submit(ra
, mapping
, file
);
1591 * Asynchronous readahead happens when we find the page and PG_readahead,
1592 * so we want to possibly extend the readahead further..
1594 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1595 struct file_ra_state
*ra
,
1600 struct address_space
*mapping
= file
->f_mapping
;
1602 /* If we don't want any read-ahead, don't bother */
1603 if (vma
->vm_flags
& VM_RAND_READ
)
1605 if (ra
->mmap_miss
> 0)
1607 if (PageReadahead(page
))
1608 page_cache_async_readahead(mapping
, ra
, file
,
1609 page
, offset
, ra
->ra_pages
);
1613 * filemap_fault - read in file data for page fault handling
1614 * @vma: vma in which the fault was taken
1615 * @vmf: struct vm_fault containing details of the fault
1617 * filemap_fault() is invoked via the vma operations vector for a
1618 * mapped memory region to read in file data during a page fault.
1620 * The goto's are kind of ugly, but this streamlines the normal case of having
1621 * it in the page cache, and handles the special cases reasonably without
1622 * having a lot of duplicated code.
1624 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1627 struct file
*file
= vma
->vm_file
;
1628 struct address_space
*mapping
= file
->f_mapping
;
1629 struct file_ra_state
*ra
= &file
->f_ra
;
1630 struct inode
*inode
= mapping
->host
;
1631 pgoff_t offset
= vmf
->pgoff
;
1636 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1638 return VM_FAULT_SIGBUS
;
1641 * Do we have something in the page cache already?
1643 page
= find_get_page(mapping
, offset
);
1644 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
1646 * We found the page, so try async readahead before
1647 * waiting for the lock.
1649 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
1651 /* No page in the page cache at all */
1652 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
1653 count_vm_event(PGMAJFAULT
);
1654 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1655 ret
= VM_FAULT_MAJOR
;
1657 page
= find_get_page(mapping
, offset
);
1659 goto no_cached_page
;
1662 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
1663 page_cache_release(page
);
1664 return ret
| VM_FAULT_RETRY
;
1667 /* Did it get truncated? */
1668 if (unlikely(page
->mapping
!= mapping
)) {
1673 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1676 * We have a locked page in the page cache, now we need to check
1677 * that it's up-to-date. If not, it is going to be due to an error.
1679 if (unlikely(!PageUptodate(page
)))
1680 goto page_not_uptodate
;
1683 * Found the page and have a reference on it.
1684 * We must recheck i_size under page lock.
1686 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1687 if (unlikely(offset
>= size
)) {
1689 page_cache_release(page
);
1690 return VM_FAULT_SIGBUS
;
1694 return ret
| VM_FAULT_LOCKED
;
1698 * We're only likely to ever get here if MADV_RANDOM is in
1701 error
= page_cache_read(file
, offset
);
1704 * The page we want has now been added to the page cache.
1705 * In the unlikely event that someone removed it in the
1706 * meantime, we'll just come back here and read it again.
1712 * An error return from page_cache_read can result if the
1713 * system is low on memory, or a problem occurs while trying
1716 if (error
== -ENOMEM
)
1717 return VM_FAULT_OOM
;
1718 return VM_FAULT_SIGBUS
;
1722 * Umm, take care of errors if the page isn't up-to-date.
1723 * Try to re-read it _once_. We do this synchronously,
1724 * because there really aren't any performance issues here
1725 * and we need to check for errors.
1727 ClearPageError(page
);
1728 error
= mapping
->a_ops
->readpage(file
, page
);
1730 wait_on_page_locked(page
);
1731 if (!PageUptodate(page
))
1734 page_cache_release(page
);
1736 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
1739 /* Things didn't work out. Return zero to tell the mm layer so. */
1740 shrink_readahead_size_eio(file
, ra
);
1741 return VM_FAULT_SIGBUS
;
1743 EXPORT_SYMBOL(filemap_fault
);
1745 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1747 struct page
*page
= vmf
->page
;
1748 struct inode
*inode
= file_inode(vma
->vm_file
);
1749 int ret
= VM_FAULT_LOCKED
;
1751 sb_start_pagefault(inode
->i_sb
);
1752 file_update_time(vma
->vm_file
);
1754 if (page
->mapping
!= inode
->i_mapping
) {
1756 ret
= VM_FAULT_NOPAGE
;
1760 * We mark the page dirty already here so that when freeze is in
1761 * progress, we are guaranteed that writeback during freezing will
1762 * see the dirty page and writeprotect it again.
1764 set_page_dirty(page
);
1765 wait_for_stable_page(page
);
1767 sb_end_pagefault(inode
->i_sb
);
1770 EXPORT_SYMBOL(filemap_page_mkwrite
);
1772 const struct vm_operations_struct generic_file_vm_ops
= {
1773 .fault
= filemap_fault
,
1774 .page_mkwrite
= filemap_page_mkwrite
,
1775 .remap_pages
= generic_file_remap_pages
,
1778 /* This is used for a general mmap of a disk file */
1780 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1782 struct address_space
*mapping
= file
->f_mapping
;
1784 if (!mapping
->a_ops
->readpage
)
1786 file_accessed(file
);
1787 vma
->vm_ops
= &generic_file_vm_ops
;
1792 * This is for filesystems which do not implement ->writepage.
1794 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1796 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
1798 return generic_file_mmap(file
, vma
);
1801 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1805 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
1809 #endif /* CONFIG_MMU */
1811 EXPORT_SYMBOL(generic_file_mmap
);
1812 EXPORT_SYMBOL(generic_file_readonly_mmap
);
1814 static struct page
*__read_cache_page(struct address_space
*mapping
,
1816 int (*filler
)(void *, struct page
*),
1823 page
= find_get_page(mapping
, index
);
1825 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
1827 return ERR_PTR(-ENOMEM
);
1828 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
1829 if (unlikely(err
)) {
1830 page_cache_release(page
);
1833 /* Presumably ENOMEM for radix tree node */
1834 return ERR_PTR(err
);
1836 err
= filler(data
, page
);
1838 page_cache_release(page
);
1839 page
= ERR_PTR(err
);
1845 static struct page
*do_read_cache_page(struct address_space
*mapping
,
1847 int (*filler
)(void *, struct page
*),
1856 page
= __read_cache_page(mapping
, index
, filler
, data
, gfp
);
1859 if (PageUptodate(page
))
1863 if (!page
->mapping
) {
1865 page_cache_release(page
);
1868 if (PageUptodate(page
)) {
1872 err
= filler(data
, page
);
1874 page_cache_release(page
);
1875 return ERR_PTR(err
);
1878 mark_page_accessed(page
);
1883 * read_cache_page_async - read into page cache, fill it if needed
1884 * @mapping: the page's address_space
1885 * @index: the page index
1886 * @filler: function to perform the read
1887 * @data: first arg to filler(data, page) function, often left as NULL
1889 * Same as read_cache_page, but don't wait for page to become unlocked
1890 * after submitting it to the filler.
1892 * Read into the page cache. If a page already exists, and PageUptodate() is
1893 * not set, try to fill the page but don't wait for it to become unlocked.
1895 * If the page does not get brought uptodate, return -EIO.
1897 struct page
*read_cache_page_async(struct address_space
*mapping
,
1899 int (*filler
)(void *, struct page
*),
1902 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
1904 EXPORT_SYMBOL(read_cache_page_async
);
1906 static struct page
*wait_on_page_read(struct page
*page
)
1908 if (!IS_ERR(page
)) {
1909 wait_on_page_locked(page
);
1910 if (!PageUptodate(page
)) {
1911 page_cache_release(page
);
1912 page
= ERR_PTR(-EIO
);
1919 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1920 * @mapping: the page's address_space
1921 * @index: the page index
1922 * @gfp: the page allocator flags to use if allocating
1924 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1925 * any new page allocations done using the specified allocation flags.
1927 * If the page does not get brought uptodate, return -EIO.
1929 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
1933 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
1935 return wait_on_page_read(do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
));
1937 EXPORT_SYMBOL(read_cache_page_gfp
);
1940 * read_cache_page - read into page cache, fill it if needed
1941 * @mapping: the page's address_space
1942 * @index: the page index
1943 * @filler: function to perform the read
1944 * @data: first arg to filler(data, page) function, often left as NULL
1946 * Read into the page cache. If a page already exists, and PageUptodate() is
1947 * not set, try to fill the page then wait for it to become unlocked.
1949 * If the page does not get brought uptodate, return -EIO.
1951 struct page
*read_cache_page(struct address_space
*mapping
,
1953 int (*filler
)(void *, struct page
*),
1956 return wait_on_page_read(read_cache_page_async(mapping
, index
, filler
, data
));
1958 EXPORT_SYMBOL(read_cache_page
);
1960 static size_t __iovec_copy_from_user_inatomic(char *vaddr
,
1961 const struct iovec
*iov
, size_t base
, size_t bytes
)
1963 size_t copied
= 0, left
= 0;
1966 char __user
*buf
= iov
->iov_base
+ base
;
1967 int copy
= min(bytes
, iov
->iov_len
- base
);
1970 left
= __copy_from_user_inatomic(vaddr
, buf
, copy
);
1979 return copied
- left
;
1983 * Copy as much as we can into the page and return the number of bytes which
1984 * were successfully copied. If a fault is encountered then return the number of
1985 * bytes which were copied.
1987 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
1988 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
1993 kaddr
= kmap_atomic(page
);
1994 if (likely(i
->nr_segs
== 1)) {
1996 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
1997 left
= __copy_from_user_inatomic(kaddr
+ offset
, buf
, bytes
);
1998 copied
= bytes
- left
;
2000 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2001 i
->iov
, i
->iov_offset
, bytes
);
2003 kunmap_atomic(kaddr
);
2007 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
2010 * This has the same sideeffects and return value as
2011 * iov_iter_copy_from_user_atomic().
2012 * The difference is that it attempts to resolve faults.
2013 * Page must not be locked.
2015 size_t iov_iter_copy_from_user(struct page
*page
,
2016 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
2022 if (likely(i
->nr_segs
== 1)) {
2024 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2025 left
= __copy_from_user(kaddr
+ offset
, buf
, bytes
);
2026 copied
= bytes
- left
;
2028 copied
= __iovec_copy_from_user_inatomic(kaddr
+ offset
,
2029 i
->iov
, i
->iov_offset
, bytes
);
2034 EXPORT_SYMBOL(iov_iter_copy_from_user
);
2036 void iov_iter_advance(struct iov_iter
*i
, size_t bytes
)
2038 BUG_ON(i
->count
< bytes
);
2040 if (likely(i
->nr_segs
== 1)) {
2041 i
->iov_offset
+= bytes
;
2044 const struct iovec
*iov
= i
->iov
;
2045 size_t base
= i
->iov_offset
;
2046 unsigned long nr_segs
= i
->nr_segs
;
2049 * The !iov->iov_len check ensures we skip over unlikely
2050 * zero-length segments (without overruning the iovec).
2052 while (bytes
|| unlikely(i
->count
&& !iov
->iov_len
)) {
2055 copy
= min(bytes
, iov
->iov_len
- base
);
2056 BUG_ON(!i
->count
|| i
->count
< copy
);
2060 if (iov
->iov_len
== base
) {
2067 i
->iov_offset
= base
;
2068 i
->nr_segs
= nr_segs
;
2071 EXPORT_SYMBOL(iov_iter_advance
);
2074 * Fault in the first iovec of the given iov_iter, to a maximum length
2075 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2076 * accessed (ie. because it is an invalid address).
2078 * writev-intensive code may want this to prefault several iovecs -- that
2079 * would be possible (callers must not rely on the fact that _only_ the
2080 * first iovec will be faulted with the current implementation).
2082 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
2084 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
2085 bytes
= min(bytes
, i
->iov
->iov_len
- i
->iov_offset
);
2086 return fault_in_pages_readable(buf
, bytes
);
2088 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
2091 * Return the count of just the current iov_iter segment.
2093 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
2095 const struct iovec
*iov
= i
->iov
;
2096 if (i
->nr_segs
== 1)
2099 return min(i
->count
, iov
->iov_len
- i
->iov_offset
);
2101 EXPORT_SYMBOL(iov_iter_single_seg_count
);
2104 * Performs necessary checks before doing a write
2106 * Can adjust writing position or amount of bytes to write.
2107 * Returns appropriate error code that caller should return or
2108 * zero in case that write should be allowed.
2110 inline int generic_write_checks(struct file
*file
, loff_t
*pos
, size_t *count
, int isblk
)
2112 struct inode
*inode
= file
->f_mapping
->host
;
2113 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2115 if (unlikely(*pos
< 0))
2119 /* FIXME: this is for backwards compatibility with 2.4 */
2120 if (file
->f_flags
& O_APPEND
)
2121 *pos
= i_size_read(inode
);
2123 if (limit
!= RLIM_INFINITY
) {
2124 if (*pos
>= limit
) {
2125 send_sig(SIGXFSZ
, current
, 0);
2128 if (*count
> limit
- (typeof(limit
))*pos
) {
2129 *count
= limit
- (typeof(limit
))*pos
;
2137 if (unlikely(*pos
+ *count
> MAX_NON_LFS
&&
2138 !(file
->f_flags
& O_LARGEFILE
))) {
2139 if (*pos
>= MAX_NON_LFS
) {
2142 if (*count
> MAX_NON_LFS
- (unsigned long)*pos
) {
2143 *count
= MAX_NON_LFS
- (unsigned long)*pos
;
2148 * Are we about to exceed the fs block limit ?
2150 * If we have written data it becomes a short write. If we have
2151 * exceeded without writing data we send a signal and return EFBIG.
2152 * Linus frestrict idea will clean these up nicely..
2154 if (likely(!isblk
)) {
2155 if (unlikely(*pos
>= inode
->i_sb
->s_maxbytes
)) {
2156 if (*count
|| *pos
> inode
->i_sb
->s_maxbytes
) {
2159 /* zero-length writes at ->s_maxbytes are OK */
2162 if (unlikely(*pos
+ *count
> inode
->i_sb
->s_maxbytes
))
2163 *count
= inode
->i_sb
->s_maxbytes
- *pos
;
2167 if (bdev_read_only(I_BDEV(inode
)))
2169 isize
= i_size_read(inode
);
2170 if (*pos
>= isize
) {
2171 if (*count
|| *pos
> isize
)
2175 if (*pos
+ *count
> isize
)
2176 *count
= isize
- *pos
;
2183 EXPORT_SYMBOL(generic_write_checks
);
2185 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2186 loff_t pos
, unsigned len
, unsigned flags
,
2187 struct page
**pagep
, void **fsdata
)
2189 const struct address_space_operations
*aops
= mapping
->a_ops
;
2191 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2194 EXPORT_SYMBOL(pagecache_write_begin
);
2196 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2197 loff_t pos
, unsigned len
, unsigned copied
,
2198 struct page
*page
, void *fsdata
)
2200 const struct address_space_operations
*aops
= mapping
->a_ops
;
2202 mark_page_accessed(page
);
2203 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2205 EXPORT_SYMBOL(pagecache_write_end
);
2208 generic_file_direct_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2209 unsigned long *nr_segs
, loff_t pos
, loff_t
*ppos
,
2210 size_t count
, size_t ocount
)
2212 struct file
*file
= iocb
->ki_filp
;
2213 struct address_space
*mapping
= file
->f_mapping
;
2214 struct inode
*inode
= mapping
->host
;
2219 if (count
!= ocount
)
2220 *nr_segs
= iov_shorten((struct iovec
*)iov
, *nr_segs
, count
);
2222 write_len
= iov_length(iov
, *nr_segs
);
2223 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2225 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2230 * After a write we want buffered reads to be sure to go to disk to get
2231 * the new data. We invalidate clean cached page from the region we're
2232 * about to write. We do this *before* the write so that we can return
2233 * without clobbering -EIOCBQUEUED from ->direct_IO().
2235 if (mapping
->nrpages
) {
2236 written
= invalidate_inode_pages2_range(mapping
,
2237 pos
>> PAGE_CACHE_SHIFT
, end
);
2239 * If a page can not be invalidated, return 0 to fall back
2240 * to buffered write.
2243 if (written
== -EBUSY
)
2249 written
= mapping
->a_ops
->direct_IO(WRITE
, iocb
, iov
, pos
, *nr_segs
);
2252 * Finally, try again to invalidate clean pages which might have been
2253 * cached by non-direct readahead, or faulted in by get_user_pages()
2254 * if the source of the write was an mmap'ed region of the file
2255 * we're writing. Either one is a pretty crazy thing to do,
2256 * so we don't support it 100%. If this invalidation
2257 * fails, tough, the write still worked...
2259 if (mapping
->nrpages
) {
2260 invalidate_inode_pages2_range(mapping
,
2261 pos
>> PAGE_CACHE_SHIFT
, end
);
2266 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2267 i_size_write(inode
, pos
);
2268 mark_inode_dirty(inode
);
2275 EXPORT_SYMBOL(generic_file_direct_write
);
2278 * Find or create a page at the given pagecache position. Return the locked
2279 * page. This function is specifically for buffered writes.
2281 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2282 pgoff_t index
, unsigned flags
)
2287 gfp_t gfp_notmask
= 0;
2289 gfp_mask
= mapping_gfp_mask(mapping
);
2290 if (mapping_cap_account_dirty(mapping
))
2291 gfp_mask
|= __GFP_WRITE
;
2292 if (flags
& AOP_FLAG_NOFS
)
2293 gfp_notmask
= __GFP_FS
;
2295 page
= find_lock_page(mapping
, index
);
2299 page
= __page_cache_alloc(gfp_mask
& ~gfp_notmask
);
2302 status
= add_to_page_cache_lru(page
, mapping
, index
,
2303 GFP_KERNEL
& ~gfp_notmask
);
2304 if (unlikely(status
)) {
2305 page_cache_release(page
);
2306 if (status
== -EEXIST
)
2311 wait_for_stable_page(page
);
2314 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2316 static ssize_t
generic_perform_write(struct file
*file
,
2317 struct iov_iter
*i
, loff_t pos
)
2319 struct address_space
*mapping
= file
->f_mapping
;
2320 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2322 ssize_t written
= 0;
2323 unsigned int flags
= 0;
2326 * Copies from kernel address space cannot fail (NFSD is a big user).
2328 if (segment_eq(get_fs(), KERNEL_DS
))
2329 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2333 unsigned long offset
; /* Offset into pagecache page */
2334 unsigned long bytes
; /* Bytes to write to page */
2335 size_t copied
; /* Bytes copied from user */
2338 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2339 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2344 * Bring in the user page that we will copy from _first_.
2345 * Otherwise there's a nasty deadlock on copying from the
2346 * same page as we're writing to, without it being marked
2349 * Not only is this an optimisation, but it is also required
2350 * to check that the address is actually valid, when atomic
2351 * usercopies are used, below.
2353 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2358 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2360 if (unlikely(status
))
2363 if (mapping_writably_mapped(mapping
))
2364 flush_dcache_page(page
);
2366 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2367 flush_dcache_page(page
);
2369 mark_page_accessed(page
);
2370 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2372 if (unlikely(status
< 0))
2378 iov_iter_advance(i
, copied
);
2379 if (unlikely(copied
== 0)) {
2381 * If we were unable to copy any data at all, we must
2382 * fall back to a single segment length write.
2384 * If we didn't fallback here, we could livelock
2385 * because not all segments in the iov can be copied at
2386 * once without a pagefault.
2388 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2389 iov_iter_single_seg_count(i
));
2395 balance_dirty_pages_ratelimited(mapping
);
2396 if (fatal_signal_pending(current
)) {
2400 } while (iov_iter_count(i
));
2402 return written
? written
: status
;
2406 generic_file_buffered_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2407 unsigned long nr_segs
, loff_t pos
, loff_t
*ppos
,
2408 size_t count
, ssize_t written
)
2410 struct file
*file
= iocb
->ki_filp
;
2414 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
2415 status
= generic_perform_write(file
, &i
, pos
);
2417 if (likely(status
>= 0)) {
2419 *ppos
= pos
+ status
;
2422 return written
? written
: status
;
2424 EXPORT_SYMBOL(generic_file_buffered_write
);
2427 * __generic_file_aio_write - write data to a file
2428 * @iocb: IO state structure (file, offset, etc.)
2429 * @iov: vector with data to write
2430 * @nr_segs: number of segments in the vector
2431 * @ppos: position where to write
2433 * This function does all the work needed for actually writing data to a
2434 * file. It does all basic checks, removes SUID from the file, updates
2435 * modification times and calls proper subroutines depending on whether we
2436 * do direct IO or a standard buffered write.
2438 * It expects i_mutex to be grabbed unless we work on a block device or similar
2439 * object which does not need locking at all.
2441 * This function does *not* take care of syncing data in case of O_SYNC write.
2442 * A caller has to handle it. This is mainly due to the fact that we want to
2443 * avoid syncing under i_mutex.
2445 ssize_t
__generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2446 unsigned long nr_segs
, loff_t
*ppos
)
2448 struct file
*file
= iocb
->ki_filp
;
2449 struct address_space
* mapping
= file
->f_mapping
;
2450 size_t ocount
; /* original count */
2451 size_t count
; /* after file limit checks */
2452 struct inode
*inode
= mapping
->host
;
2458 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
2465 /* We can write back this queue in page reclaim */
2466 current
->backing_dev_info
= mapping
->backing_dev_info
;
2469 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
2476 err
= file_remove_suid(file
);
2480 err
= file_update_time(file
);
2484 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2485 if (unlikely(file
->f_flags
& O_DIRECT
)) {
2487 ssize_t written_buffered
;
2489 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
,
2490 ppos
, count
, ocount
);
2491 if (written
< 0 || written
== count
)
2494 * direct-io write to a hole: fall through to buffered I/O
2495 * for completing the rest of the request.
2499 written_buffered
= generic_file_buffered_write(iocb
, iov
,
2500 nr_segs
, pos
, ppos
, count
,
2503 * If generic_file_buffered_write() retuned a synchronous error
2504 * then we want to return the number of bytes which were
2505 * direct-written, or the error code if that was zero. Note
2506 * that this differs from normal direct-io semantics, which
2507 * will return -EFOO even if some bytes were written.
2509 if (written_buffered
< 0) {
2510 err
= written_buffered
;
2515 * We need to ensure that the page cache pages are written to
2516 * disk and invalidated to preserve the expected O_DIRECT
2519 endbyte
= pos
+ written_buffered
- written
- 1;
2520 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
2522 written
= written_buffered
;
2523 invalidate_mapping_pages(mapping
,
2524 pos
>> PAGE_CACHE_SHIFT
,
2525 endbyte
>> PAGE_CACHE_SHIFT
);
2528 * We don't know how much we wrote, so just return
2529 * the number of bytes which were direct-written
2533 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
,
2534 pos
, ppos
, count
, written
);
2537 current
->backing_dev_info
= NULL
;
2538 return written
? written
: err
;
2540 EXPORT_SYMBOL(__generic_file_aio_write
);
2543 * generic_file_aio_write - write data to a file
2544 * @iocb: IO state structure
2545 * @iov: vector with data to write
2546 * @nr_segs: number of segments in the vector
2547 * @pos: position in file where to write
2549 * This is a wrapper around __generic_file_aio_write() to be used by most
2550 * filesystems. It takes care of syncing the file in case of O_SYNC file
2551 * and acquires i_mutex as needed.
2553 ssize_t
generic_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
2554 unsigned long nr_segs
, loff_t pos
)
2556 struct file
*file
= iocb
->ki_filp
;
2557 struct inode
*inode
= file
->f_mapping
->host
;
2560 BUG_ON(iocb
->ki_pos
!= pos
);
2562 mutex_lock(&inode
->i_mutex
);
2563 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
, &iocb
->ki_pos
);
2564 mutex_unlock(&inode
->i_mutex
);
2569 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2575 EXPORT_SYMBOL(generic_file_aio_write
);
2578 * try_to_release_page() - release old fs-specific metadata on a page
2580 * @page: the page which the kernel is trying to free
2581 * @gfp_mask: memory allocation flags (and I/O mode)
2583 * The address_space is to try to release any data against the page
2584 * (presumably at page->private). If the release was successful, return `1'.
2585 * Otherwise return zero.
2587 * This may also be called if PG_fscache is set on a page, indicating that the
2588 * page is known to the local caching routines.
2590 * The @gfp_mask argument specifies whether I/O may be performed to release
2591 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2594 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2596 struct address_space
* const mapping
= page
->mapping
;
2598 BUG_ON(!PageLocked(page
));
2599 if (PageWriteback(page
))
2602 if (mapping
&& mapping
->a_ops
->releasepage
)
2603 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2604 return try_to_free_buffers(page
);
2607 EXPORT_SYMBOL(try_to_release_page
);