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>
14 #include <linux/dax.h>
16 #include <linux/uaccess.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/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/rmap.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/filemap.h>
44 * FIXME: remove all knowledge of the buffer layer from the core VM
46 #include <linux/buffer_head.h> /* for try_to_free_buffers */
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
54 * Shared mappings now work. 15.8.1995 Bruno.
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
65 * ->i_mmap_rwsem (truncate_pagecache)
66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
71 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
79 * ->lock_page (access_process_vm)
81 * ->i_mutex (generic_perform_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
85 * sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
104 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
110 * ->tasklist_lock (memory_failure, collect_procs_ao)
113 static void page_cache_tree_delete(struct address_space
*mapping
,
114 struct page
*page
, void *shadow
)
116 struct radix_tree_node
*node
;
117 int i
, nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
119 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
120 VM_BUG_ON_PAGE(PageTail(page
), page
);
121 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
124 mapping
->nrexceptional
+= nr
;
126 * Make sure the nrexceptional update is committed before
127 * the nrpages update so that final truncate racing
128 * with reclaim does not see both counters 0 at the
129 * same time and miss a shadow entry.
133 mapping
->nrpages
-= nr
;
135 for (i
= 0; i
< nr
; i
++) {
136 node
= radix_tree_replace_clear_tags(&mapping
->page_tree
,
137 page
->index
+ i
, shadow
);
139 VM_BUG_ON_PAGE(nr
!= 1, page
);
143 workingset_node_pages_dec(node
);
145 workingset_node_shadows_inc(node
);
147 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
151 * Track node that only contains shadow entries. DAX mappings
152 * contain no shadow entries and may contain other exceptional
153 * entries so skip those.
155 * Avoid acquiring the list_lru lock if already tracked.
156 * The list_empty() test is safe as node->private_list is
157 * protected by mapping->tree_lock.
159 if (!dax_mapping(mapping
) && !workingset_node_pages(node
) &&
160 list_empty(&node
->private_list
)) {
161 node
->private_data
= mapping
;
162 list_lru_add(&workingset_shadow_nodes
,
163 &node
->private_list
);
169 * Delete a page from the page cache and free it. Caller has to make
170 * sure the page is locked and that nobody else uses it - or that usage
171 * is safe. The caller must hold the mapping's tree_lock.
173 void __delete_from_page_cache(struct page
*page
, void *shadow
)
175 struct address_space
*mapping
= page
->mapping
;
176 int nr
= hpage_nr_pages(page
);
178 trace_mm_filemap_delete_from_page_cache(page
);
180 * if we're uptodate, flush out into the cleancache, otherwise
181 * invalidate any existing cleancache entries. We can't leave
182 * stale data around in the cleancache once our page is gone
184 if (PageUptodate(page
) && PageMappedToDisk(page
))
185 cleancache_put_page(page
);
187 cleancache_invalidate_page(mapping
, page
);
189 VM_BUG_ON_PAGE(PageTail(page
), page
);
190 VM_BUG_ON_PAGE(page_mapped(page
), page
);
191 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
194 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
195 current
->comm
, page_to_pfn(page
));
196 dump_page(page
, "still mapped when deleted");
198 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
200 mapcount
= page_mapcount(page
);
201 if (mapping_exiting(mapping
) &&
202 page_count(page
) >= mapcount
+ 2) {
204 * All vmas have already been torn down, so it's
205 * a good bet that actually the page is unmapped,
206 * and we'd prefer not to leak it: if we're wrong,
207 * some other bad page check should catch it later.
209 page_mapcount_reset(page
);
210 page_ref_sub(page
, mapcount
);
214 page_cache_tree_delete(mapping
, page
, shadow
);
216 page
->mapping
= NULL
;
217 /* Leave page->index set: truncation lookup relies upon it */
219 /* hugetlb pages do not participate in page cache accounting. */
221 __mod_zone_page_state(page_zone(page
), NR_FILE_PAGES
, -nr
);
222 if (PageSwapBacked(page
))
223 __mod_zone_page_state(page_zone(page
), NR_SHMEM
, -nr
);
226 * At this point page must be either written or cleaned by truncate.
227 * Dirty page here signals a bug and loss of unwritten data.
229 * This fixes dirty accounting after removing the page entirely but
230 * leaves PageDirty set: it has no effect for truncated page and
231 * anyway will be cleared before returning page into buddy allocator.
233 if (WARN_ON_ONCE(PageDirty(page
)))
234 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
238 * delete_from_page_cache - delete page from page cache
239 * @page: the page which the kernel is trying to remove from page cache
241 * This must be called only on pages that have been verified to be in the page
242 * cache and locked. It will never put the page into the free list, the caller
243 * has a reference on the page.
245 void delete_from_page_cache(struct page
*page
)
247 struct address_space
*mapping
= page_mapping(page
);
249 void (*freepage
)(struct page
*);
251 BUG_ON(!PageLocked(page
));
253 freepage
= mapping
->a_ops
->freepage
;
255 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
256 __delete_from_page_cache(page
, NULL
);
257 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
262 if (PageTransHuge(page
) && !PageHuge(page
)) {
263 page_ref_sub(page
, HPAGE_PMD_NR
);
264 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
269 EXPORT_SYMBOL(delete_from_page_cache
);
271 static int filemap_check_errors(struct address_space
*mapping
)
274 /* Check for outstanding write errors */
275 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
276 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
278 if (test_bit(AS_EIO
, &mapping
->flags
) &&
279 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
285 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
286 * @mapping: address space structure to write
287 * @start: offset in bytes where the range starts
288 * @end: offset in bytes where the range ends (inclusive)
289 * @sync_mode: enable synchronous operation
291 * Start writeback against all of a mapping's dirty pages that lie
292 * within the byte offsets <start, end> inclusive.
294 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
295 * opposed to a regular memory cleansing writeback. The difference between
296 * these two operations is that if a dirty page/buffer is encountered, it must
297 * be waited upon, and not just skipped over.
299 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
300 loff_t end
, int sync_mode
)
303 struct writeback_control wbc
= {
304 .sync_mode
= sync_mode
,
305 .nr_to_write
= LONG_MAX
,
306 .range_start
= start
,
310 if (!mapping_cap_writeback_dirty(mapping
))
313 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
314 ret
= do_writepages(mapping
, &wbc
);
315 wbc_detach_inode(&wbc
);
319 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
322 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
325 int filemap_fdatawrite(struct address_space
*mapping
)
327 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
329 EXPORT_SYMBOL(filemap_fdatawrite
);
331 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
334 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
336 EXPORT_SYMBOL(filemap_fdatawrite_range
);
339 * filemap_flush - mostly a non-blocking flush
340 * @mapping: target address_space
342 * This is a mostly non-blocking flush. Not suitable for data-integrity
343 * purposes - I/O may not be started against all dirty pages.
345 int filemap_flush(struct address_space
*mapping
)
347 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
349 EXPORT_SYMBOL(filemap_flush
);
351 static int __filemap_fdatawait_range(struct address_space
*mapping
,
352 loff_t start_byte
, loff_t end_byte
)
354 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
355 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
360 if (end_byte
< start_byte
)
363 pagevec_init(&pvec
, 0);
364 while ((index
<= end
) &&
365 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
366 PAGECACHE_TAG_WRITEBACK
,
367 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
370 for (i
= 0; i
< nr_pages
; i
++) {
371 struct page
*page
= pvec
.pages
[i
];
373 /* until radix tree lookup accepts end_index */
374 if (page
->index
> end
)
377 wait_on_page_writeback(page
);
378 if (TestClearPageError(page
))
381 pagevec_release(&pvec
);
389 * filemap_fdatawait_range - wait for writeback to complete
390 * @mapping: address space structure to wait for
391 * @start_byte: offset in bytes where the range starts
392 * @end_byte: offset in bytes where the range ends (inclusive)
394 * Walk the list of under-writeback pages of the given address space
395 * in the given range and wait for all of them. Check error status of
396 * the address space and return it.
398 * Since the error status of the address space is cleared by this function,
399 * callers are responsible for checking the return value and handling and/or
400 * reporting the error.
402 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
407 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
408 ret2
= filemap_check_errors(mapping
);
414 EXPORT_SYMBOL(filemap_fdatawait_range
);
417 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
418 * @mapping: address space structure to wait for
420 * Walk the list of under-writeback pages of the given address space
421 * and wait for all of them. Unlike filemap_fdatawait(), this function
422 * does not clear error status of the address space.
424 * Use this function if callers don't handle errors themselves. Expected
425 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
428 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
430 loff_t i_size
= i_size_read(mapping
->host
);
435 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
439 * filemap_fdatawait - wait for all under-writeback pages to complete
440 * @mapping: address space structure to wait for
442 * Walk the list of under-writeback pages of the given address space
443 * and wait for all of them. Check error status of the address space
446 * Since the error status of the address space is cleared by this function,
447 * callers are responsible for checking the return value and handling and/or
448 * reporting the error.
450 int filemap_fdatawait(struct address_space
*mapping
)
452 loff_t i_size
= i_size_read(mapping
->host
);
457 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
459 EXPORT_SYMBOL(filemap_fdatawait
);
461 int filemap_write_and_wait(struct address_space
*mapping
)
465 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
466 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
467 err
= filemap_fdatawrite(mapping
);
469 * Even if the above returned error, the pages may be
470 * written partially (e.g. -ENOSPC), so we wait for it.
471 * But the -EIO is special case, it may indicate the worst
472 * thing (e.g. bug) happened, so we avoid waiting for it.
475 int err2
= filemap_fdatawait(mapping
);
480 err
= filemap_check_errors(mapping
);
484 EXPORT_SYMBOL(filemap_write_and_wait
);
487 * filemap_write_and_wait_range - write out & wait on a file range
488 * @mapping: the address_space for the pages
489 * @lstart: offset in bytes where the range starts
490 * @lend: offset in bytes where the range ends (inclusive)
492 * Write out and wait upon file offsets lstart->lend, inclusive.
494 * Note that `lend' is inclusive (describes the last byte to be written) so
495 * that this function can be used to write to the very end-of-file (end = -1).
497 int filemap_write_and_wait_range(struct address_space
*mapping
,
498 loff_t lstart
, loff_t lend
)
502 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
503 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
504 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
506 /* See comment of filemap_write_and_wait() */
508 int err2
= filemap_fdatawait_range(mapping
,
514 err
= filemap_check_errors(mapping
);
518 EXPORT_SYMBOL(filemap_write_and_wait_range
);
521 * replace_page_cache_page - replace a pagecache page with a new one
522 * @old: page to be replaced
523 * @new: page to replace with
524 * @gfp_mask: allocation mode
526 * This function replaces a page in the pagecache with a new one. On
527 * success it acquires the pagecache reference for the new page and
528 * drops it for the old page. Both the old and new pages must be
529 * locked. This function does not add the new page to the LRU, the
530 * caller must do that.
532 * The remove + add is atomic. The only way this function can fail is
533 * memory allocation failure.
535 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
539 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
540 VM_BUG_ON_PAGE(!PageLocked(new), new);
541 VM_BUG_ON_PAGE(new->mapping
, new);
543 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
545 struct address_space
*mapping
= old
->mapping
;
546 void (*freepage
)(struct page
*);
549 pgoff_t offset
= old
->index
;
550 freepage
= mapping
->a_ops
->freepage
;
553 new->mapping
= mapping
;
556 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
557 __delete_from_page_cache(old
, NULL
);
558 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
563 * hugetlb pages do not participate in page cache accounting.
566 __inc_zone_page_state(new, NR_FILE_PAGES
);
567 if (PageSwapBacked(new))
568 __inc_zone_page_state(new, NR_SHMEM
);
569 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
570 mem_cgroup_migrate(old
, new);
571 radix_tree_preload_end();
579 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
581 static int page_cache_tree_insert(struct address_space
*mapping
,
582 struct page
*page
, void **shadowp
)
584 struct radix_tree_node
*node
;
588 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
595 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
596 if (!radix_tree_exceptional_entry(p
))
599 mapping
->nrexceptional
--;
600 if (!dax_mapping(mapping
)) {
604 workingset_node_shadows_dec(node
);
606 /* DAX can replace empty locked entry with a hole */
608 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY
|
609 RADIX_DAX_ENTRY_LOCK
));
610 /* DAX accounts exceptional entries as normal pages */
612 workingset_node_pages_dec(node
);
613 /* Wakeup waiters for exceptional entry lock */
614 dax_wake_mapping_entry_waiter(mapping
, page
->index
,
618 radix_tree_replace_slot(slot
, page
);
621 workingset_node_pages_inc(node
);
623 * Don't track node that contains actual pages.
625 * Avoid acquiring the list_lru lock if already
626 * untracked. The list_empty() test is safe as
627 * node->private_list is protected by
628 * mapping->tree_lock.
630 if (!list_empty(&node
->private_list
))
631 list_lru_del(&workingset_shadow_nodes
,
632 &node
->private_list
);
637 static int __add_to_page_cache_locked(struct page
*page
,
638 struct address_space
*mapping
,
639 pgoff_t offset
, gfp_t gfp_mask
,
642 int huge
= PageHuge(page
);
643 struct mem_cgroup
*memcg
;
646 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
647 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
650 error
= mem_cgroup_try_charge(page
, current
->mm
,
651 gfp_mask
, &memcg
, false);
656 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
659 mem_cgroup_cancel_charge(page
, memcg
, false);
664 page
->mapping
= mapping
;
665 page
->index
= offset
;
667 spin_lock_irq(&mapping
->tree_lock
);
668 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
669 radix_tree_preload_end();
673 /* hugetlb pages do not participate in page cache accounting. */
675 __inc_zone_page_state(page
, NR_FILE_PAGES
);
676 spin_unlock_irq(&mapping
->tree_lock
);
678 mem_cgroup_commit_charge(page
, memcg
, false, false);
679 trace_mm_filemap_add_to_page_cache(page
);
682 page
->mapping
= NULL
;
683 /* Leave page->index set: truncation relies upon it */
684 spin_unlock_irq(&mapping
->tree_lock
);
686 mem_cgroup_cancel_charge(page
, memcg
, false);
692 * add_to_page_cache_locked - add a locked page to the pagecache
694 * @mapping: the page's address_space
695 * @offset: page index
696 * @gfp_mask: page allocation mode
698 * This function is used to add a page to the pagecache. It must be locked.
699 * This function does not add the page to the LRU. The caller must do that.
701 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
702 pgoff_t offset
, gfp_t gfp_mask
)
704 return __add_to_page_cache_locked(page
, mapping
, offset
,
707 EXPORT_SYMBOL(add_to_page_cache_locked
);
709 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
710 pgoff_t offset
, gfp_t gfp_mask
)
715 __SetPageLocked(page
);
716 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
719 __ClearPageLocked(page
);
722 * The page might have been evicted from cache only
723 * recently, in which case it should be activated like
724 * any other repeatedly accessed page.
725 * The exception is pages getting rewritten; evicting other
726 * data from the working set, only to cache data that will
727 * get overwritten with something else, is a waste of memory.
729 if (!(gfp_mask
& __GFP_WRITE
) &&
730 shadow
&& workingset_refault(shadow
)) {
732 workingset_activation(page
);
734 ClearPageActive(page
);
739 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
742 struct page
*__page_cache_alloc(gfp_t gfp
)
747 if (cpuset_do_page_mem_spread()) {
748 unsigned int cpuset_mems_cookie
;
750 cpuset_mems_cookie
= read_mems_allowed_begin();
751 n
= cpuset_mem_spread_node();
752 page
= __alloc_pages_node(n
, gfp
, 0);
753 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
757 return alloc_pages(gfp
, 0);
759 EXPORT_SYMBOL(__page_cache_alloc
);
763 * In order to wait for pages to become available there must be
764 * waitqueues associated with pages. By using a hash table of
765 * waitqueues where the bucket discipline is to maintain all
766 * waiters on the same queue and wake all when any of the pages
767 * become available, and for the woken contexts to check to be
768 * sure the appropriate page became available, this saves space
769 * at a cost of "thundering herd" phenomena during rare hash
772 wait_queue_head_t
*page_waitqueue(struct page
*page
)
774 const struct zone
*zone
= page_zone(page
);
776 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
778 EXPORT_SYMBOL(page_waitqueue
);
780 void wait_on_page_bit(struct page
*page
, int bit_nr
)
782 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
784 if (test_bit(bit_nr
, &page
->flags
))
785 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
786 TASK_UNINTERRUPTIBLE
);
788 EXPORT_SYMBOL(wait_on_page_bit
);
790 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
792 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
794 if (!test_bit(bit_nr
, &page
->flags
))
797 return __wait_on_bit(page_waitqueue(page
), &wait
,
798 bit_wait_io
, TASK_KILLABLE
);
801 int wait_on_page_bit_killable_timeout(struct page
*page
,
802 int bit_nr
, unsigned long timeout
)
804 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
806 wait
.key
.timeout
= jiffies
+ timeout
;
807 if (!test_bit(bit_nr
, &page
->flags
))
809 return __wait_on_bit(page_waitqueue(page
), &wait
,
810 bit_wait_io_timeout
, TASK_KILLABLE
);
812 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
815 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
816 * @page: Page defining the wait queue of interest
817 * @waiter: Waiter to add to the queue
819 * Add an arbitrary @waiter to the wait queue for the nominated @page.
821 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
823 wait_queue_head_t
*q
= page_waitqueue(page
);
826 spin_lock_irqsave(&q
->lock
, flags
);
827 __add_wait_queue(q
, waiter
);
828 spin_unlock_irqrestore(&q
->lock
, flags
);
830 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
833 * unlock_page - unlock a locked page
836 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
837 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
838 * mechanism between PageLocked pages and PageWriteback pages is shared.
839 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
841 * The mb is necessary to enforce ordering between the clear_bit and the read
842 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
844 void unlock_page(struct page
*page
)
846 page
= compound_head(page
);
847 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
848 clear_bit_unlock(PG_locked
, &page
->flags
);
849 smp_mb__after_atomic();
850 wake_up_page(page
, PG_locked
);
852 EXPORT_SYMBOL(unlock_page
);
855 * end_page_writeback - end writeback against a page
858 void end_page_writeback(struct page
*page
)
861 * TestClearPageReclaim could be used here but it is an atomic
862 * operation and overkill in this particular case. Failing to
863 * shuffle a page marked for immediate reclaim is too mild to
864 * justify taking an atomic operation penalty at the end of
865 * ever page writeback.
867 if (PageReclaim(page
)) {
868 ClearPageReclaim(page
);
869 rotate_reclaimable_page(page
);
872 if (!test_clear_page_writeback(page
))
875 smp_mb__after_atomic();
876 wake_up_page(page
, PG_writeback
);
878 EXPORT_SYMBOL(end_page_writeback
);
881 * After completing I/O on a page, call this routine to update the page
882 * flags appropriately
884 void page_endio(struct page
*page
, int rw
, int err
)
888 SetPageUptodate(page
);
890 ClearPageUptodate(page
);
894 } else { /* rw == WRITE */
898 mapping_set_error(page
->mapping
, err
);
900 end_page_writeback(page
);
903 EXPORT_SYMBOL_GPL(page_endio
);
906 * __lock_page - get a lock on the page, assuming we need to sleep to get it
907 * @page: the page to lock
909 void __lock_page(struct page
*page
)
911 struct page
*page_head
= compound_head(page
);
912 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
914 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
915 TASK_UNINTERRUPTIBLE
);
917 EXPORT_SYMBOL(__lock_page
);
919 int __lock_page_killable(struct page
*page
)
921 struct page
*page_head
= compound_head(page
);
922 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
924 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
925 bit_wait_io
, TASK_KILLABLE
);
927 EXPORT_SYMBOL_GPL(__lock_page_killable
);
931 * 1 - page is locked; mmap_sem is still held.
932 * 0 - page is not locked.
933 * mmap_sem has been released (up_read()), unless flags had both
934 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
935 * which case mmap_sem is still held.
937 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
938 * with the page locked and the mmap_sem unperturbed.
940 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
943 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
945 * CAUTION! In this case, mmap_sem is not released
946 * even though return 0.
948 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
951 up_read(&mm
->mmap_sem
);
952 if (flags
& FAULT_FLAG_KILLABLE
)
953 wait_on_page_locked_killable(page
);
955 wait_on_page_locked(page
);
958 if (flags
& FAULT_FLAG_KILLABLE
) {
961 ret
= __lock_page_killable(page
);
963 up_read(&mm
->mmap_sem
);
973 * page_cache_next_hole - find the next hole (not-present entry)
976 * @max_scan: maximum range to search
978 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
979 * lowest indexed hole.
981 * Returns: the index of the hole if found, otherwise returns an index
982 * outside of the set specified (in which case 'return - index >=
983 * max_scan' will be true). In rare cases of index wrap-around, 0 will
986 * page_cache_next_hole may be called under rcu_read_lock. However,
987 * like radix_tree_gang_lookup, this will not atomically search a
988 * snapshot of the tree at a single point in time. For example, if a
989 * hole is created at index 5, then subsequently a hole is created at
990 * index 10, page_cache_next_hole covering both indexes may return 10
991 * if called under rcu_read_lock.
993 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
994 pgoff_t index
, unsigned long max_scan
)
998 for (i
= 0; i
< max_scan
; i
++) {
1001 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1002 if (!page
|| radix_tree_exceptional_entry(page
))
1011 EXPORT_SYMBOL(page_cache_next_hole
);
1014 * page_cache_prev_hole - find the prev hole (not-present entry)
1017 * @max_scan: maximum range to search
1019 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1022 * Returns: the index of the hole if found, otherwise returns an index
1023 * outside of the set specified (in which case 'index - return >=
1024 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1027 * page_cache_prev_hole may be called under rcu_read_lock. However,
1028 * like radix_tree_gang_lookup, this will not atomically search a
1029 * snapshot of the tree at a single point in time. For example, if a
1030 * hole is created at index 10, then subsequently a hole is created at
1031 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1032 * called under rcu_read_lock.
1034 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1035 pgoff_t index
, unsigned long max_scan
)
1039 for (i
= 0; i
< max_scan
; i
++) {
1042 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1043 if (!page
|| radix_tree_exceptional_entry(page
))
1046 if (index
== ULONG_MAX
)
1052 EXPORT_SYMBOL(page_cache_prev_hole
);
1055 * find_get_entry - find and get a page cache entry
1056 * @mapping: the address_space to search
1057 * @offset: the page cache index
1059 * Looks up the page cache slot at @mapping & @offset. If there is a
1060 * page cache page, it is returned with an increased refcount.
1062 * If the slot holds a shadow entry of a previously evicted page, or a
1063 * swap entry from shmem/tmpfs, it is returned.
1065 * Otherwise, %NULL is returned.
1067 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1070 struct page
*head
, *page
;
1075 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1077 page
= radix_tree_deref_slot(pagep
);
1078 if (unlikely(!page
))
1080 if (radix_tree_exception(page
)) {
1081 if (radix_tree_deref_retry(page
))
1084 * A shadow entry of a recently evicted page,
1085 * or a swap entry from shmem/tmpfs. Return
1086 * it without attempting to raise page count.
1091 head
= compound_head(page
);
1092 if (!page_cache_get_speculative(head
))
1095 /* The page was split under us? */
1096 if (compound_head(page
) != head
) {
1102 * Has the page moved?
1103 * This is part of the lockless pagecache protocol. See
1104 * include/linux/pagemap.h for details.
1106 if (unlikely(page
!= *pagep
)) {
1116 EXPORT_SYMBOL(find_get_entry
);
1119 * find_lock_entry - locate, pin and lock a page cache entry
1120 * @mapping: the address_space to search
1121 * @offset: the page cache index
1123 * Looks up the page cache slot at @mapping & @offset. If there is a
1124 * page cache page, it is returned locked and with an increased
1127 * If the slot holds a shadow entry of a previously evicted page, or a
1128 * swap entry from shmem/tmpfs, it is returned.
1130 * Otherwise, %NULL is returned.
1132 * find_lock_entry() may sleep.
1134 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1139 page
= find_get_entry(mapping
, offset
);
1140 if (page
&& !radix_tree_exception(page
)) {
1142 /* Has the page been truncated? */
1143 if (unlikely(page_mapping(page
) != mapping
)) {
1148 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1152 EXPORT_SYMBOL(find_lock_entry
);
1155 * pagecache_get_page - find and get a page reference
1156 * @mapping: the address_space to search
1157 * @offset: the page index
1158 * @fgp_flags: PCG flags
1159 * @gfp_mask: gfp mask to use for the page cache data page allocation
1161 * Looks up the page cache slot at @mapping & @offset.
1163 * PCG flags modify how the page is returned.
1165 * FGP_ACCESSED: the page will be marked accessed
1166 * FGP_LOCK: Page is return locked
1167 * FGP_CREAT: If page is not present then a new page is allocated using
1168 * @gfp_mask and added to the page cache and the VM's LRU
1169 * list. The page is returned locked and with an increased
1170 * refcount. Otherwise, %NULL is returned.
1172 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1173 * if the GFP flags specified for FGP_CREAT are atomic.
1175 * If there is a page cache page, it is returned with an increased refcount.
1177 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1178 int fgp_flags
, gfp_t gfp_mask
)
1183 page
= find_get_entry(mapping
, offset
);
1184 if (radix_tree_exceptional_entry(page
))
1189 if (fgp_flags
& FGP_LOCK
) {
1190 if (fgp_flags
& FGP_NOWAIT
) {
1191 if (!trylock_page(page
)) {
1199 /* Has the page been truncated? */
1200 if (unlikely(page
->mapping
!= mapping
)) {
1205 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1208 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1209 mark_page_accessed(page
);
1212 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1214 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1215 gfp_mask
|= __GFP_WRITE
;
1216 if (fgp_flags
& FGP_NOFS
)
1217 gfp_mask
&= ~__GFP_FS
;
1219 page
= __page_cache_alloc(gfp_mask
);
1223 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1224 fgp_flags
|= FGP_LOCK
;
1226 /* Init accessed so avoid atomic mark_page_accessed later */
1227 if (fgp_flags
& FGP_ACCESSED
)
1228 __SetPageReferenced(page
);
1230 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1231 gfp_mask
& GFP_RECLAIM_MASK
);
1232 if (unlikely(err
)) {
1242 EXPORT_SYMBOL(pagecache_get_page
);
1245 * find_get_entries - gang pagecache lookup
1246 * @mapping: The address_space to search
1247 * @start: The starting page cache index
1248 * @nr_entries: The maximum number of entries
1249 * @entries: Where the resulting entries are placed
1250 * @indices: The cache indices corresponding to the entries in @entries
1252 * find_get_entries() will search for and return a group of up to
1253 * @nr_entries entries in the mapping. The entries are placed at
1254 * @entries. find_get_entries() takes a reference against any actual
1257 * The search returns a group of mapping-contiguous page cache entries
1258 * with ascending indexes. There may be holes in the indices due to
1259 * not-present pages.
1261 * Any shadow entries of evicted pages, or swap entries from
1262 * shmem/tmpfs, are included in the returned array.
1264 * find_get_entries() returns the number of pages and shadow entries
1267 unsigned find_get_entries(struct address_space
*mapping
,
1268 pgoff_t start
, unsigned int nr_entries
,
1269 struct page
**entries
, pgoff_t
*indices
)
1272 unsigned int ret
= 0;
1273 struct radix_tree_iter iter
;
1279 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1280 struct page
*head
, *page
;
1282 page
= radix_tree_deref_slot(slot
);
1283 if (unlikely(!page
))
1285 if (radix_tree_exception(page
)) {
1286 if (radix_tree_deref_retry(page
)) {
1287 slot
= radix_tree_iter_retry(&iter
);
1291 * A shadow entry of a recently evicted page, a swap
1292 * entry from shmem/tmpfs or a DAX entry. Return it
1293 * without attempting to raise page count.
1298 head
= compound_head(page
);
1299 if (!page_cache_get_speculative(head
))
1302 /* The page was split under us? */
1303 if (compound_head(page
) != head
) {
1308 /* Has the page moved? */
1309 if (unlikely(page
!= *slot
)) {
1314 indices
[ret
] = iter
.index
;
1315 entries
[ret
] = page
;
1316 if (++ret
== nr_entries
)
1324 * find_get_pages - gang pagecache lookup
1325 * @mapping: The address_space to search
1326 * @start: The starting page index
1327 * @nr_pages: The maximum number of pages
1328 * @pages: Where the resulting pages are placed
1330 * find_get_pages() will search for and return a group of up to
1331 * @nr_pages pages in the mapping. The pages are placed at @pages.
1332 * find_get_pages() takes a reference against the returned pages.
1334 * The search returns a group of mapping-contiguous pages with ascending
1335 * indexes. There may be holes in the indices due to not-present pages.
1337 * find_get_pages() returns the number of pages which were found.
1339 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1340 unsigned int nr_pages
, struct page
**pages
)
1342 struct radix_tree_iter iter
;
1346 if (unlikely(!nr_pages
))
1350 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1351 struct page
*head
, *page
;
1353 page
= radix_tree_deref_slot(slot
);
1354 if (unlikely(!page
))
1357 if (radix_tree_exception(page
)) {
1358 if (radix_tree_deref_retry(page
)) {
1359 slot
= radix_tree_iter_retry(&iter
);
1363 * A shadow entry of a recently evicted page,
1364 * or a swap entry from shmem/tmpfs. Skip
1370 head
= compound_head(page
);
1371 if (!page_cache_get_speculative(head
))
1374 /* The page was split under us? */
1375 if (compound_head(page
) != head
) {
1380 /* Has the page moved? */
1381 if (unlikely(page
!= *slot
)) {
1387 if (++ret
== nr_pages
)
1396 * find_get_pages_contig - gang contiguous pagecache lookup
1397 * @mapping: The address_space to search
1398 * @index: The starting page index
1399 * @nr_pages: The maximum number of pages
1400 * @pages: Where the resulting pages are placed
1402 * find_get_pages_contig() works exactly like find_get_pages(), except
1403 * that the returned number of pages are guaranteed to be contiguous.
1405 * find_get_pages_contig() returns the number of pages which were found.
1407 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1408 unsigned int nr_pages
, struct page
**pages
)
1410 struct radix_tree_iter iter
;
1412 unsigned int ret
= 0;
1414 if (unlikely(!nr_pages
))
1418 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1419 struct page
*head
, *page
;
1421 page
= radix_tree_deref_slot(slot
);
1422 /* The hole, there no reason to continue */
1423 if (unlikely(!page
))
1426 if (radix_tree_exception(page
)) {
1427 if (radix_tree_deref_retry(page
)) {
1428 slot
= radix_tree_iter_retry(&iter
);
1432 * A shadow entry of a recently evicted page,
1433 * or a swap entry from shmem/tmpfs. Stop
1434 * looking for contiguous pages.
1439 head
= compound_head(page
);
1440 if (!page_cache_get_speculative(head
))
1443 /* The page was split under us? */
1444 if (compound_head(page
) != head
) {
1449 /* Has the page moved? */
1450 if (unlikely(page
!= *slot
)) {
1456 * must check mapping and index after taking the ref.
1457 * otherwise we can get both false positives and false
1458 * negatives, which is just confusing to the caller.
1460 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1466 if (++ret
== nr_pages
)
1472 EXPORT_SYMBOL(find_get_pages_contig
);
1475 * find_get_pages_tag - find and return pages that match @tag
1476 * @mapping: the address_space to search
1477 * @index: the starting page index
1478 * @tag: the tag index
1479 * @nr_pages: the maximum number of pages
1480 * @pages: where the resulting pages are placed
1482 * Like find_get_pages, except we only return pages which are tagged with
1483 * @tag. We update @index to index the next page for the traversal.
1485 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1486 int tag
, unsigned int nr_pages
, struct page
**pages
)
1488 struct radix_tree_iter iter
;
1492 if (unlikely(!nr_pages
))
1496 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1497 &iter
, *index
, tag
) {
1498 struct page
*head
, *page
;
1500 page
= radix_tree_deref_slot(slot
);
1501 if (unlikely(!page
))
1504 if (radix_tree_exception(page
)) {
1505 if (radix_tree_deref_retry(page
)) {
1506 slot
= radix_tree_iter_retry(&iter
);
1510 * A shadow entry of a recently evicted page.
1512 * Those entries should never be tagged, but
1513 * this tree walk is lockless and the tags are
1514 * looked up in bulk, one radix tree node at a
1515 * time, so there is a sizable window for page
1516 * reclaim to evict a page we saw tagged.
1523 head
= compound_head(page
);
1524 if (!page_cache_get_speculative(head
))
1527 /* The page was split under us? */
1528 if (compound_head(page
) != head
) {
1533 /* Has the page moved? */
1534 if (unlikely(page
!= *slot
)) {
1540 if (++ret
== nr_pages
)
1547 *index
= pages
[ret
- 1]->index
+ 1;
1551 EXPORT_SYMBOL(find_get_pages_tag
);
1554 * find_get_entries_tag - find and return entries that match @tag
1555 * @mapping: the address_space to search
1556 * @start: the starting page cache index
1557 * @tag: the tag index
1558 * @nr_entries: the maximum number of entries
1559 * @entries: where the resulting entries are placed
1560 * @indices: the cache indices corresponding to the entries in @entries
1562 * Like find_get_entries, except we only return entries which are tagged with
1565 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1566 int tag
, unsigned int nr_entries
,
1567 struct page
**entries
, pgoff_t
*indices
)
1570 unsigned int ret
= 0;
1571 struct radix_tree_iter iter
;
1577 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1578 &iter
, start
, tag
) {
1579 struct page
*head
, *page
;
1581 page
= radix_tree_deref_slot(slot
);
1582 if (unlikely(!page
))
1584 if (radix_tree_exception(page
)) {
1585 if (radix_tree_deref_retry(page
)) {
1586 slot
= radix_tree_iter_retry(&iter
);
1591 * A shadow entry of a recently evicted page, a swap
1592 * entry from shmem/tmpfs or a DAX entry. Return it
1593 * without attempting to raise page count.
1598 head
= compound_head(page
);
1599 if (!page_cache_get_speculative(head
))
1602 /* The page was split under us? */
1603 if (compound_head(page
) != head
) {
1608 /* Has the page moved? */
1609 if (unlikely(page
!= *slot
)) {
1614 indices
[ret
] = iter
.index
;
1615 entries
[ret
] = page
;
1616 if (++ret
== nr_entries
)
1622 EXPORT_SYMBOL(find_get_entries_tag
);
1625 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1626 * a _large_ part of the i/o request. Imagine the worst scenario:
1628 * ---R__________________________________________B__________
1629 * ^ reading here ^ bad block(assume 4k)
1631 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1632 * => failing the whole request => read(R) => read(R+1) =>
1633 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1634 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1635 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1637 * It is going insane. Fix it by quickly scaling down the readahead size.
1639 static void shrink_readahead_size_eio(struct file
*filp
,
1640 struct file_ra_state
*ra
)
1646 * do_generic_file_read - generic file read routine
1647 * @filp: the file to read
1648 * @ppos: current file position
1649 * @iter: data destination
1650 * @written: already copied
1652 * This is a generic file read routine, and uses the
1653 * mapping->a_ops->readpage() function for the actual low-level stuff.
1655 * This is really ugly. But the goto's actually try to clarify some
1656 * of the logic when it comes to error handling etc.
1658 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1659 struct iov_iter
*iter
, ssize_t written
)
1661 struct address_space
*mapping
= filp
->f_mapping
;
1662 struct inode
*inode
= mapping
->host
;
1663 struct file_ra_state
*ra
= &filp
->f_ra
;
1667 unsigned long offset
; /* offset into pagecache page */
1668 unsigned int prev_offset
;
1671 index
= *ppos
>> PAGE_SHIFT
;
1672 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1673 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1674 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1675 offset
= *ppos
& ~PAGE_MASK
;
1681 unsigned long nr
, ret
;
1685 page
= find_get_page(mapping
, index
);
1687 page_cache_sync_readahead(mapping
,
1689 index
, last_index
- index
);
1690 page
= find_get_page(mapping
, index
);
1691 if (unlikely(page
== NULL
))
1692 goto no_cached_page
;
1694 if (PageReadahead(page
)) {
1695 page_cache_async_readahead(mapping
,
1697 index
, last_index
- index
);
1699 if (!PageUptodate(page
)) {
1701 * See comment in do_read_cache_page on why
1702 * wait_on_page_locked is used to avoid unnecessarily
1703 * serialisations and why it's safe.
1705 wait_on_page_locked_killable(page
);
1706 if (PageUptodate(page
))
1709 if (inode
->i_blkbits
== PAGE_SHIFT
||
1710 !mapping
->a_ops
->is_partially_uptodate
)
1711 goto page_not_up_to_date
;
1712 if (!trylock_page(page
))
1713 goto page_not_up_to_date
;
1714 /* Did it get truncated before we got the lock? */
1716 goto page_not_up_to_date_locked
;
1717 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1718 offset
, iter
->count
))
1719 goto page_not_up_to_date_locked
;
1724 * i_size must be checked after we know the page is Uptodate.
1726 * Checking i_size after the check allows us to calculate
1727 * the correct value for "nr", which means the zero-filled
1728 * part of the page is not copied back to userspace (unless
1729 * another truncate extends the file - this is desired though).
1732 isize
= i_size_read(inode
);
1733 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1734 if (unlikely(!isize
|| index
> end_index
)) {
1739 /* nr is the maximum number of bytes to copy from this page */
1741 if (index
== end_index
) {
1742 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1750 /* If users can be writing to this page using arbitrary
1751 * virtual addresses, take care about potential aliasing
1752 * before reading the page on the kernel side.
1754 if (mapping_writably_mapped(mapping
))
1755 flush_dcache_page(page
);
1758 * When a sequential read accesses a page several times,
1759 * only mark it as accessed the first time.
1761 if (prev_index
!= index
|| offset
!= prev_offset
)
1762 mark_page_accessed(page
);
1766 * Ok, we have the page, and it's up-to-date, so
1767 * now we can copy it to user space...
1770 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1772 index
+= offset
>> PAGE_SHIFT
;
1773 offset
&= ~PAGE_MASK
;
1774 prev_offset
= offset
;
1778 if (!iov_iter_count(iter
))
1786 page_not_up_to_date
:
1787 /* Get exclusive access to the page ... */
1788 error
= lock_page_killable(page
);
1789 if (unlikely(error
))
1790 goto readpage_error
;
1792 page_not_up_to_date_locked
:
1793 /* Did it get truncated before we got the lock? */
1794 if (!page
->mapping
) {
1800 /* Did somebody else fill it already? */
1801 if (PageUptodate(page
)) {
1808 * A previous I/O error may have been due to temporary
1809 * failures, eg. multipath errors.
1810 * PG_error will be set again if readpage fails.
1812 ClearPageError(page
);
1813 /* Start the actual read. The read will unlock the page. */
1814 error
= mapping
->a_ops
->readpage(filp
, page
);
1816 if (unlikely(error
)) {
1817 if (error
== AOP_TRUNCATED_PAGE
) {
1822 goto readpage_error
;
1825 if (!PageUptodate(page
)) {
1826 error
= lock_page_killable(page
);
1827 if (unlikely(error
))
1828 goto readpage_error
;
1829 if (!PageUptodate(page
)) {
1830 if (page
->mapping
== NULL
) {
1832 * invalidate_mapping_pages got it
1839 shrink_readahead_size_eio(filp
, ra
);
1841 goto readpage_error
;
1849 /* UHHUH! A synchronous read error occurred. Report it */
1855 * Ok, it wasn't cached, so we need to create a new
1858 page
= page_cache_alloc_cold(mapping
);
1863 error
= add_to_page_cache_lru(page
, mapping
, index
,
1864 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1867 if (error
== -EEXIST
) {
1877 ra
->prev_pos
= prev_index
;
1878 ra
->prev_pos
<<= PAGE_SHIFT
;
1879 ra
->prev_pos
|= prev_offset
;
1881 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
1882 file_accessed(filp
);
1883 return written
? written
: error
;
1887 * generic_file_read_iter - generic filesystem read routine
1888 * @iocb: kernel I/O control block
1889 * @iter: destination for the data read
1891 * This is the "read_iter()" routine for all filesystems
1892 * that can use the page cache directly.
1895 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1897 struct file
*file
= iocb
->ki_filp
;
1899 size_t count
= iov_iter_count(iter
);
1902 goto out
; /* skip atime */
1904 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1905 struct address_space
*mapping
= file
->f_mapping
;
1906 struct inode
*inode
= mapping
->host
;
1909 size
= i_size_read(inode
);
1910 retval
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
1911 iocb
->ki_pos
+ count
- 1);
1913 struct iov_iter data
= *iter
;
1914 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
);
1918 iocb
->ki_pos
+= retval
;
1919 iov_iter_advance(iter
, retval
);
1923 * Btrfs can have a short DIO read if we encounter
1924 * compressed extents, so if there was an error, or if
1925 * we've already read everything we wanted to, or if
1926 * there was a short read because we hit EOF, go ahead
1927 * and return. Otherwise fallthrough to buffered io for
1928 * the rest of the read. Buffered reads will not work for
1929 * DAX files, so don't bother trying.
1931 if (retval
< 0 || !iov_iter_count(iter
) || iocb
->ki_pos
>= size
||
1933 file_accessed(file
);
1938 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
1942 EXPORT_SYMBOL(generic_file_read_iter
);
1946 * page_cache_read - adds requested page to the page cache if not already there
1947 * @file: file to read
1948 * @offset: page index
1949 * @gfp_mask: memory allocation flags
1951 * This adds the requested page to the page cache if it isn't already there,
1952 * and schedules an I/O to read in its contents from disk.
1954 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1956 struct address_space
*mapping
= file
->f_mapping
;
1961 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1965 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1967 ret
= mapping
->a_ops
->readpage(file
, page
);
1968 else if (ret
== -EEXIST
)
1969 ret
= 0; /* losing race to add is OK */
1973 } while (ret
== AOP_TRUNCATED_PAGE
);
1978 #define MMAP_LOTSAMISS (100)
1981 * Synchronous readahead happens when we don't even find
1982 * a page in the page cache at all.
1984 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1985 struct file_ra_state
*ra
,
1989 struct address_space
*mapping
= file
->f_mapping
;
1991 /* If we don't want any read-ahead, don't bother */
1992 if (vma
->vm_flags
& VM_RAND_READ
)
1997 if (vma
->vm_flags
& VM_SEQ_READ
) {
1998 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2003 /* Avoid banging the cache line if not needed */
2004 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2008 * Do we miss much more than hit in this file? If so,
2009 * stop bothering with read-ahead. It will only hurt.
2011 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2017 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2018 ra
->size
= ra
->ra_pages
;
2019 ra
->async_size
= ra
->ra_pages
/ 4;
2020 ra_submit(ra
, mapping
, file
);
2024 * Asynchronous readahead happens when we find the page and PG_readahead,
2025 * so we want to possibly extend the readahead further..
2027 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2028 struct file_ra_state
*ra
,
2033 struct address_space
*mapping
= file
->f_mapping
;
2035 /* If we don't want any read-ahead, don't bother */
2036 if (vma
->vm_flags
& VM_RAND_READ
)
2038 if (ra
->mmap_miss
> 0)
2040 if (PageReadahead(page
))
2041 page_cache_async_readahead(mapping
, ra
, file
,
2042 page
, offset
, ra
->ra_pages
);
2046 * filemap_fault - read in file data for page fault handling
2047 * @vma: vma in which the fault was taken
2048 * @vmf: struct vm_fault containing details of the fault
2050 * filemap_fault() is invoked via the vma operations vector for a
2051 * mapped memory region to read in file data during a page fault.
2053 * The goto's are kind of ugly, but this streamlines the normal case of having
2054 * it in the page cache, and handles the special cases reasonably without
2055 * having a lot of duplicated code.
2057 * vma->vm_mm->mmap_sem must be held on entry.
2059 * If our return value has VM_FAULT_RETRY set, it's because
2060 * lock_page_or_retry() returned 0.
2061 * The mmap_sem has usually been released in this case.
2062 * See __lock_page_or_retry() for the exception.
2064 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2065 * has not been released.
2067 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2069 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2072 struct file
*file
= vma
->vm_file
;
2073 struct address_space
*mapping
= file
->f_mapping
;
2074 struct file_ra_state
*ra
= &file
->f_ra
;
2075 struct inode
*inode
= mapping
->host
;
2076 pgoff_t offset
= vmf
->pgoff
;
2081 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2082 if (offset
>= size
>> PAGE_SHIFT
)
2083 return VM_FAULT_SIGBUS
;
2086 * Do we have something in the page cache already?
2088 page
= find_get_page(mapping
, offset
);
2089 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2091 * We found the page, so try async readahead before
2092 * waiting for the lock.
2094 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2096 /* No page in the page cache at all */
2097 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2098 count_vm_event(PGMAJFAULT
);
2099 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2100 ret
= VM_FAULT_MAJOR
;
2102 page
= find_get_page(mapping
, offset
);
2104 goto no_cached_page
;
2107 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2109 return ret
| VM_FAULT_RETRY
;
2112 /* Did it get truncated? */
2113 if (unlikely(page
->mapping
!= mapping
)) {
2118 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2121 * We have a locked page in the page cache, now we need to check
2122 * that it's up-to-date. If not, it is going to be due to an error.
2124 if (unlikely(!PageUptodate(page
)))
2125 goto page_not_uptodate
;
2128 * Found the page and have a reference on it.
2129 * We must recheck i_size under page lock.
2131 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2132 if (unlikely(offset
>= size
>> PAGE_SHIFT
)) {
2135 return VM_FAULT_SIGBUS
;
2139 return ret
| VM_FAULT_LOCKED
;
2143 * We're only likely to ever get here if MADV_RANDOM is in
2146 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2149 * The page we want has now been added to the page cache.
2150 * In the unlikely event that someone removed it in the
2151 * meantime, we'll just come back here and read it again.
2157 * An error return from page_cache_read can result if the
2158 * system is low on memory, or a problem occurs while trying
2161 if (error
== -ENOMEM
)
2162 return VM_FAULT_OOM
;
2163 return VM_FAULT_SIGBUS
;
2167 * Umm, take care of errors if the page isn't up-to-date.
2168 * Try to re-read it _once_. We do this synchronously,
2169 * because there really aren't any performance issues here
2170 * and we need to check for errors.
2172 ClearPageError(page
);
2173 error
= mapping
->a_ops
->readpage(file
, page
);
2175 wait_on_page_locked(page
);
2176 if (!PageUptodate(page
))
2181 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2184 /* Things didn't work out. Return zero to tell the mm layer so. */
2185 shrink_readahead_size_eio(file
, ra
);
2186 return VM_FAULT_SIGBUS
;
2188 EXPORT_SYMBOL(filemap_fault
);
2190 void filemap_map_pages(struct fault_env
*fe
,
2191 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2193 struct radix_tree_iter iter
;
2195 struct file
*file
= fe
->vma
->vm_file
;
2196 struct address_space
*mapping
= file
->f_mapping
;
2197 pgoff_t last_pgoff
= start_pgoff
;
2199 struct page
*head
, *page
;
2202 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2204 if (iter
.index
> end_pgoff
)
2207 page
= radix_tree_deref_slot(slot
);
2208 if (unlikely(!page
))
2210 if (radix_tree_exception(page
)) {
2211 if (radix_tree_deref_retry(page
)) {
2212 slot
= radix_tree_iter_retry(&iter
);
2218 head
= compound_head(page
);
2219 if (!page_cache_get_speculative(head
))
2222 /* The page was split under us? */
2223 if (compound_head(page
) != head
) {
2228 /* Has the page moved? */
2229 if (unlikely(page
!= *slot
)) {
2234 if (!PageUptodate(page
) ||
2235 PageReadahead(page
) ||
2238 if (!trylock_page(page
))
2241 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2244 size
= round_up(i_size_read(mapping
->host
), PAGE_SIZE
);
2245 if (page
->index
>= size
>> PAGE_SHIFT
)
2248 if (file
->f_ra
.mmap_miss
> 0)
2249 file
->f_ra
.mmap_miss
--;
2251 fe
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2253 fe
->pte
+= iter
.index
- last_pgoff
;
2254 last_pgoff
= iter
.index
;
2255 if (alloc_set_pte(fe
, NULL
, page
))
2264 /* Huge page is mapped? No need to proceed. */
2265 if (pmd_trans_huge(*fe
->pmd
))
2267 if (iter
.index
== end_pgoff
)
2272 EXPORT_SYMBOL(filemap_map_pages
);
2274 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2276 struct page
*page
= vmf
->page
;
2277 struct inode
*inode
= file_inode(vma
->vm_file
);
2278 int ret
= VM_FAULT_LOCKED
;
2280 sb_start_pagefault(inode
->i_sb
);
2281 file_update_time(vma
->vm_file
);
2283 if (page
->mapping
!= inode
->i_mapping
) {
2285 ret
= VM_FAULT_NOPAGE
;
2289 * We mark the page dirty already here so that when freeze is in
2290 * progress, we are guaranteed that writeback during freezing will
2291 * see the dirty page and writeprotect it again.
2293 set_page_dirty(page
);
2294 wait_for_stable_page(page
);
2296 sb_end_pagefault(inode
->i_sb
);
2299 EXPORT_SYMBOL(filemap_page_mkwrite
);
2301 const struct vm_operations_struct generic_file_vm_ops
= {
2302 .fault
= filemap_fault
,
2303 .map_pages
= filemap_map_pages
,
2304 .page_mkwrite
= filemap_page_mkwrite
,
2307 /* This is used for a general mmap of a disk file */
2309 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2311 struct address_space
*mapping
= file
->f_mapping
;
2313 if (!mapping
->a_ops
->readpage
)
2315 file_accessed(file
);
2316 vma
->vm_ops
= &generic_file_vm_ops
;
2321 * This is for filesystems which do not implement ->writepage.
2323 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2325 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2327 return generic_file_mmap(file
, vma
);
2330 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2334 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2338 #endif /* CONFIG_MMU */
2340 EXPORT_SYMBOL(generic_file_mmap
);
2341 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2343 static struct page
*wait_on_page_read(struct page
*page
)
2345 if (!IS_ERR(page
)) {
2346 wait_on_page_locked(page
);
2347 if (!PageUptodate(page
)) {
2349 page
= ERR_PTR(-EIO
);
2355 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2357 int (*filler
)(void *, struct page
*),
2364 page
= find_get_page(mapping
, index
);
2366 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2368 return ERR_PTR(-ENOMEM
);
2369 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2370 if (unlikely(err
)) {
2374 /* Presumably ENOMEM for radix tree node */
2375 return ERR_PTR(err
);
2379 err
= filler(data
, page
);
2382 return ERR_PTR(err
);
2385 page
= wait_on_page_read(page
);
2390 if (PageUptodate(page
))
2394 * Page is not up to date and may be locked due one of the following
2395 * case a: Page is being filled and the page lock is held
2396 * case b: Read/write error clearing the page uptodate status
2397 * case c: Truncation in progress (page locked)
2398 * case d: Reclaim in progress
2400 * Case a, the page will be up to date when the page is unlocked.
2401 * There is no need to serialise on the page lock here as the page
2402 * is pinned so the lock gives no additional protection. Even if the
2403 * the page is truncated, the data is still valid if PageUptodate as
2404 * it's a race vs truncate race.
2405 * Case b, the page will not be up to date
2406 * Case c, the page may be truncated but in itself, the data may still
2407 * be valid after IO completes as it's a read vs truncate race. The
2408 * operation must restart if the page is not uptodate on unlock but
2409 * otherwise serialising on page lock to stabilise the mapping gives
2410 * no additional guarantees to the caller as the page lock is
2411 * released before return.
2412 * Case d, similar to truncation. If reclaim holds the page lock, it
2413 * will be a race with remove_mapping that determines if the mapping
2414 * is valid on unlock but otherwise the data is valid and there is
2415 * no need to serialise with page lock.
2417 * As the page lock gives no additional guarantee, we optimistically
2418 * wait on the page to be unlocked and check if it's up to date and
2419 * use the page if it is. Otherwise, the page lock is required to
2420 * distinguish between the different cases. The motivation is that we
2421 * avoid spurious serialisations and wakeups when multiple processes
2422 * wait on the same page for IO to complete.
2424 wait_on_page_locked(page
);
2425 if (PageUptodate(page
))
2428 /* Distinguish between all the cases under the safety of the lock */
2431 /* Case c or d, restart the operation */
2432 if (!page
->mapping
) {
2438 /* Someone else locked and filled the page in a very small window */
2439 if (PageUptodate(page
)) {
2446 mark_page_accessed(page
);
2451 * read_cache_page - read into page cache, fill it if needed
2452 * @mapping: the page's address_space
2453 * @index: the page index
2454 * @filler: function to perform the read
2455 * @data: first arg to filler(data, page) function, often left as NULL
2457 * Read into the page cache. If a page already exists, and PageUptodate() is
2458 * not set, try to fill the page and wait for it to become unlocked.
2460 * If the page does not get brought uptodate, return -EIO.
2462 struct page
*read_cache_page(struct address_space
*mapping
,
2464 int (*filler
)(void *, struct page
*),
2467 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2469 EXPORT_SYMBOL(read_cache_page
);
2472 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2473 * @mapping: the page's address_space
2474 * @index: the page index
2475 * @gfp: the page allocator flags to use if allocating
2477 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2478 * any new page allocations done using the specified allocation flags.
2480 * If the page does not get brought uptodate, return -EIO.
2482 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2486 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2488 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2490 EXPORT_SYMBOL(read_cache_page_gfp
);
2493 * Performs necessary checks before doing a write
2495 * Can adjust writing position or amount of bytes to write.
2496 * Returns appropriate error code that caller should return or
2497 * zero in case that write should be allowed.
2499 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2501 struct file
*file
= iocb
->ki_filp
;
2502 struct inode
*inode
= file
->f_mapping
->host
;
2503 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2506 if (!iov_iter_count(from
))
2509 /* FIXME: this is for backwards compatibility with 2.4 */
2510 if (iocb
->ki_flags
& IOCB_APPEND
)
2511 iocb
->ki_pos
= i_size_read(inode
);
2515 if (limit
!= RLIM_INFINITY
) {
2516 if (iocb
->ki_pos
>= limit
) {
2517 send_sig(SIGXFSZ
, current
, 0);
2520 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2526 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2527 !(file
->f_flags
& O_LARGEFILE
))) {
2528 if (pos
>= MAX_NON_LFS
)
2530 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2534 * Are we about to exceed the fs block limit ?
2536 * If we have written data it becomes a short write. If we have
2537 * exceeded without writing data we send a signal and return EFBIG.
2538 * Linus frestrict idea will clean these up nicely..
2540 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2543 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2544 return iov_iter_count(from
);
2546 EXPORT_SYMBOL(generic_write_checks
);
2548 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2549 loff_t pos
, unsigned len
, unsigned flags
,
2550 struct page
**pagep
, void **fsdata
)
2552 const struct address_space_operations
*aops
= mapping
->a_ops
;
2554 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2557 EXPORT_SYMBOL(pagecache_write_begin
);
2559 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2560 loff_t pos
, unsigned len
, unsigned copied
,
2561 struct page
*page
, void *fsdata
)
2563 const struct address_space_operations
*aops
= mapping
->a_ops
;
2565 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2567 EXPORT_SYMBOL(pagecache_write_end
);
2570 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2572 struct file
*file
= iocb
->ki_filp
;
2573 struct address_space
*mapping
= file
->f_mapping
;
2574 struct inode
*inode
= mapping
->host
;
2575 loff_t pos
= iocb
->ki_pos
;
2579 struct iov_iter data
;
2581 write_len
= iov_iter_count(from
);
2582 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2584 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2589 * After a write we want buffered reads to be sure to go to disk to get
2590 * the new data. We invalidate clean cached page from the region we're
2591 * about to write. We do this *before* the write so that we can return
2592 * without clobbering -EIOCBQUEUED from ->direct_IO().
2594 if (mapping
->nrpages
) {
2595 written
= invalidate_inode_pages2_range(mapping
,
2596 pos
>> PAGE_SHIFT
, end
);
2598 * If a page can not be invalidated, return 0 to fall back
2599 * to buffered write.
2602 if (written
== -EBUSY
)
2609 written
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2612 * Finally, try again to invalidate clean pages which might have been
2613 * cached by non-direct readahead, or faulted in by get_user_pages()
2614 * if the source of the write was an mmap'ed region of the file
2615 * we're writing. Either one is a pretty crazy thing to do,
2616 * so we don't support it 100%. If this invalidation
2617 * fails, tough, the write still worked...
2619 if (mapping
->nrpages
) {
2620 invalidate_inode_pages2_range(mapping
,
2621 pos
>> PAGE_SHIFT
, end
);
2626 iov_iter_advance(from
, written
);
2627 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2628 i_size_write(inode
, pos
);
2629 mark_inode_dirty(inode
);
2636 EXPORT_SYMBOL(generic_file_direct_write
);
2639 * Find or create a page at the given pagecache position. Return the locked
2640 * page. This function is specifically for buffered writes.
2642 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2643 pgoff_t index
, unsigned flags
)
2646 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2648 if (flags
& AOP_FLAG_NOFS
)
2649 fgp_flags
|= FGP_NOFS
;
2651 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2652 mapping_gfp_mask(mapping
));
2654 wait_for_stable_page(page
);
2658 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2660 ssize_t
generic_perform_write(struct file
*file
,
2661 struct iov_iter
*i
, loff_t pos
)
2663 struct address_space
*mapping
= file
->f_mapping
;
2664 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2666 ssize_t written
= 0;
2667 unsigned int flags
= 0;
2670 * Copies from kernel address space cannot fail (NFSD is a big user).
2672 if (!iter_is_iovec(i
))
2673 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2677 unsigned long offset
; /* Offset into pagecache page */
2678 unsigned long bytes
; /* Bytes to write to page */
2679 size_t copied
; /* Bytes copied from user */
2682 offset
= (pos
& (PAGE_SIZE
- 1));
2683 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2688 * Bring in the user page that we will copy from _first_.
2689 * Otherwise there's a nasty deadlock on copying from the
2690 * same page as we're writing to, without it being marked
2693 * Not only is this an optimisation, but it is also required
2694 * to check that the address is actually valid, when atomic
2695 * usercopies are used, below.
2697 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2702 if (fatal_signal_pending(current
)) {
2707 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2709 if (unlikely(status
< 0))
2712 if (mapping_writably_mapped(mapping
))
2713 flush_dcache_page(page
);
2715 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2716 flush_dcache_page(page
);
2718 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2720 if (unlikely(status
< 0))
2726 iov_iter_advance(i
, copied
);
2727 if (unlikely(copied
== 0)) {
2729 * If we were unable to copy any data at all, we must
2730 * fall back to a single segment length write.
2732 * If we didn't fallback here, we could livelock
2733 * because not all segments in the iov can be copied at
2734 * once without a pagefault.
2736 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2737 iov_iter_single_seg_count(i
));
2743 balance_dirty_pages_ratelimited(mapping
);
2744 } while (iov_iter_count(i
));
2746 return written
? written
: status
;
2748 EXPORT_SYMBOL(generic_perform_write
);
2751 * __generic_file_write_iter - write data to a file
2752 * @iocb: IO state structure (file, offset, etc.)
2753 * @from: iov_iter with data to write
2755 * This function does all the work needed for actually writing data to a
2756 * file. It does all basic checks, removes SUID from the file, updates
2757 * modification times and calls proper subroutines depending on whether we
2758 * do direct IO or a standard buffered write.
2760 * It expects i_mutex to be grabbed unless we work on a block device or similar
2761 * object which does not need locking at all.
2763 * This function does *not* take care of syncing data in case of O_SYNC write.
2764 * A caller has to handle it. This is mainly due to the fact that we want to
2765 * avoid syncing under i_mutex.
2767 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2769 struct file
*file
= iocb
->ki_filp
;
2770 struct address_space
* mapping
= file
->f_mapping
;
2771 struct inode
*inode
= mapping
->host
;
2772 ssize_t written
= 0;
2776 /* We can write back this queue in page reclaim */
2777 current
->backing_dev_info
= inode_to_bdi(inode
);
2778 err
= file_remove_privs(file
);
2782 err
= file_update_time(file
);
2786 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2787 loff_t pos
, endbyte
;
2789 written
= generic_file_direct_write(iocb
, from
);
2791 * If the write stopped short of completing, fall back to
2792 * buffered writes. Some filesystems do this for writes to
2793 * holes, for example. For DAX files, a buffered write will
2794 * not succeed (even if it did, DAX does not handle dirty
2795 * page-cache pages correctly).
2797 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2800 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2802 * If generic_perform_write() returned a synchronous error
2803 * then we want to return the number of bytes which were
2804 * direct-written, or the error code if that was zero. Note
2805 * that this differs from normal direct-io semantics, which
2806 * will return -EFOO even if some bytes were written.
2808 if (unlikely(status
< 0)) {
2813 * We need to ensure that the page cache pages are written to
2814 * disk and invalidated to preserve the expected O_DIRECT
2817 endbyte
= pos
+ status
- 1;
2818 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2820 iocb
->ki_pos
= endbyte
+ 1;
2822 invalidate_mapping_pages(mapping
,
2824 endbyte
>> PAGE_SHIFT
);
2827 * We don't know how much we wrote, so just return
2828 * the number of bytes which were direct-written
2832 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2833 if (likely(written
> 0))
2834 iocb
->ki_pos
+= written
;
2837 current
->backing_dev_info
= NULL
;
2838 return written
? written
: err
;
2840 EXPORT_SYMBOL(__generic_file_write_iter
);
2843 * generic_file_write_iter - write data to a file
2844 * @iocb: IO state structure
2845 * @from: iov_iter with data to write
2847 * This is a wrapper around __generic_file_write_iter() to be used by most
2848 * filesystems. It takes care of syncing the file in case of O_SYNC file
2849 * and acquires i_mutex as needed.
2851 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2853 struct file
*file
= iocb
->ki_filp
;
2854 struct inode
*inode
= file
->f_mapping
->host
;
2858 ret
= generic_write_checks(iocb
, from
);
2860 ret
= __generic_file_write_iter(iocb
, from
);
2861 inode_unlock(inode
);
2864 ret
= generic_write_sync(iocb
, ret
);
2867 EXPORT_SYMBOL(generic_file_write_iter
);
2870 * try_to_release_page() - release old fs-specific metadata on a page
2872 * @page: the page which the kernel is trying to free
2873 * @gfp_mask: memory allocation flags (and I/O mode)
2875 * The address_space is to try to release any data against the page
2876 * (presumably at page->private). If the release was successful, return `1'.
2877 * Otherwise return zero.
2879 * This may also be called if PG_fscache is set on a page, indicating that the
2880 * page is known to the local caching routines.
2882 * The @gfp_mask argument specifies whether I/O may be performed to release
2883 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2886 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2888 struct address_space
* const mapping
= page
->mapping
;
2890 BUG_ON(!PageLocked(page
));
2891 if (PageWriteback(page
))
2894 if (mapping
&& mapping
->a_ops
->releasepage
)
2895 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2896 return try_to_free_buffers(page
);
2899 EXPORT_SYMBOL(try_to_release_page
);