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(zone) (follow_page->mark_page_accessed)
99 * ->zone_lru_lock(zone) (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_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
222 if (PageSwapBacked(page
)) {
223 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
224 if (PageTransHuge(page
))
225 __dec_node_page_state(page
, NR_SHMEM_THPS
);
227 VM_BUG_ON_PAGE(PageTransHuge(page
) && !PageHuge(page
), page
);
231 * At this point page must be either written or cleaned by truncate.
232 * Dirty page here signals a bug and loss of unwritten data.
234 * This fixes dirty accounting after removing the page entirely but
235 * leaves PageDirty set: it has no effect for truncated page and
236 * anyway will be cleared before returning page into buddy allocator.
238 if (WARN_ON_ONCE(PageDirty(page
)))
239 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
243 * delete_from_page_cache - delete page from page cache
244 * @page: the page which the kernel is trying to remove from page cache
246 * This must be called only on pages that have been verified to be in the page
247 * cache and locked. It will never put the page into the free list, the caller
248 * has a reference on the page.
250 void delete_from_page_cache(struct page
*page
)
252 struct address_space
*mapping
= page_mapping(page
);
254 void (*freepage
)(struct page
*);
256 BUG_ON(!PageLocked(page
));
258 freepage
= mapping
->a_ops
->freepage
;
260 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
261 __delete_from_page_cache(page
, NULL
);
262 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
267 if (PageTransHuge(page
) && !PageHuge(page
)) {
268 page_ref_sub(page
, HPAGE_PMD_NR
);
269 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
274 EXPORT_SYMBOL(delete_from_page_cache
);
276 int filemap_check_errors(struct address_space
*mapping
)
279 /* Check for outstanding write errors */
280 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
281 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
283 if (test_bit(AS_EIO
, &mapping
->flags
) &&
284 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
288 EXPORT_SYMBOL(filemap_check_errors
);
291 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
292 * @mapping: address space structure to write
293 * @start: offset in bytes where the range starts
294 * @end: offset in bytes where the range ends (inclusive)
295 * @sync_mode: enable synchronous operation
297 * Start writeback against all of a mapping's dirty pages that lie
298 * within the byte offsets <start, end> inclusive.
300 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
301 * opposed to a regular memory cleansing writeback. The difference between
302 * these two operations is that if a dirty page/buffer is encountered, it must
303 * be waited upon, and not just skipped over.
305 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
306 loff_t end
, int sync_mode
)
309 struct writeback_control wbc
= {
310 .sync_mode
= sync_mode
,
311 .nr_to_write
= LONG_MAX
,
312 .range_start
= start
,
316 if (!mapping_cap_writeback_dirty(mapping
))
319 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
320 ret
= do_writepages(mapping
, &wbc
);
321 wbc_detach_inode(&wbc
);
325 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
328 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
331 int filemap_fdatawrite(struct address_space
*mapping
)
333 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
335 EXPORT_SYMBOL(filemap_fdatawrite
);
337 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
340 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
342 EXPORT_SYMBOL(filemap_fdatawrite_range
);
345 * filemap_flush - mostly a non-blocking flush
346 * @mapping: target address_space
348 * This is a mostly non-blocking flush. Not suitable for data-integrity
349 * purposes - I/O may not be started against all dirty pages.
351 int filemap_flush(struct address_space
*mapping
)
353 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
355 EXPORT_SYMBOL(filemap_flush
);
357 static int __filemap_fdatawait_range(struct address_space
*mapping
,
358 loff_t start_byte
, loff_t end_byte
)
360 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
361 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
366 if (end_byte
< start_byte
)
369 pagevec_init(&pvec
, 0);
370 while ((index
<= end
) &&
371 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
372 PAGECACHE_TAG_WRITEBACK
,
373 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
376 for (i
= 0; i
< nr_pages
; i
++) {
377 struct page
*page
= pvec
.pages
[i
];
379 /* until radix tree lookup accepts end_index */
380 if (page
->index
> end
)
383 wait_on_page_writeback(page
);
384 if (TestClearPageError(page
))
387 pagevec_release(&pvec
);
395 * filemap_fdatawait_range - wait for writeback to complete
396 * @mapping: address space structure to wait for
397 * @start_byte: offset in bytes where the range starts
398 * @end_byte: offset in bytes where the range ends (inclusive)
400 * Walk the list of under-writeback pages of the given address space
401 * in the given range and wait for all of them. Check error status of
402 * the address space and return it.
404 * Since the error status of the address space is cleared by this function,
405 * callers are responsible for checking the return value and handling and/or
406 * reporting the error.
408 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
413 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
414 ret2
= filemap_check_errors(mapping
);
420 EXPORT_SYMBOL(filemap_fdatawait_range
);
423 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
424 * @mapping: address space structure to wait for
426 * Walk the list of under-writeback pages of the given address space
427 * and wait for all of them. Unlike filemap_fdatawait(), this function
428 * does not clear error status of the address space.
430 * Use this function if callers don't handle errors themselves. Expected
431 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
434 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
436 loff_t i_size
= i_size_read(mapping
->host
);
441 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
445 * filemap_fdatawait - wait for all under-writeback pages to complete
446 * @mapping: address space structure to wait for
448 * Walk the list of under-writeback pages of the given address space
449 * and wait for all of them. Check error status of the address space
452 * Since the error status of the address space is cleared by this function,
453 * callers are responsible for checking the return value and handling and/or
454 * reporting the error.
456 int filemap_fdatawait(struct address_space
*mapping
)
458 loff_t i_size
= i_size_read(mapping
->host
);
463 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
465 EXPORT_SYMBOL(filemap_fdatawait
);
467 int filemap_write_and_wait(struct address_space
*mapping
)
471 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
472 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
473 err
= filemap_fdatawrite(mapping
);
475 * Even if the above returned error, the pages may be
476 * written partially (e.g. -ENOSPC), so we wait for it.
477 * But the -EIO is special case, it may indicate the worst
478 * thing (e.g. bug) happened, so we avoid waiting for it.
481 int err2
= filemap_fdatawait(mapping
);
486 err
= filemap_check_errors(mapping
);
490 EXPORT_SYMBOL(filemap_write_and_wait
);
493 * filemap_write_and_wait_range - write out & wait on a file range
494 * @mapping: the address_space for the pages
495 * @lstart: offset in bytes where the range starts
496 * @lend: offset in bytes where the range ends (inclusive)
498 * Write out and wait upon file offsets lstart->lend, inclusive.
500 * Note that `lend' is inclusive (describes the last byte to be written) so
501 * that this function can be used to write to the very end-of-file (end = -1).
503 int filemap_write_and_wait_range(struct address_space
*mapping
,
504 loff_t lstart
, loff_t lend
)
508 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
509 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
510 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
512 /* See comment of filemap_write_and_wait() */
514 int err2
= filemap_fdatawait_range(mapping
,
520 err
= filemap_check_errors(mapping
);
524 EXPORT_SYMBOL(filemap_write_and_wait_range
);
527 * replace_page_cache_page - replace a pagecache page with a new one
528 * @old: page to be replaced
529 * @new: page to replace with
530 * @gfp_mask: allocation mode
532 * This function replaces a page in the pagecache with a new one. On
533 * success it acquires the pagecache reference for the new page and
534 * drops it for the old page. Both the old and new pages must be
535 * locked. This function does not add the new page to the LRU, the
536 * caller must do that.
538 * The remove + add is atomic. The only way this function can fail is
539 * memory allocation failure.
541 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
545 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
546 VM_BUG_ON_PAGE(!PageLocked(new), new);
547 VM_BUG_ON_PAGE(new->mapping
, new);
549 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
551 struct address_space
*mapping
= old
->mapping
;
552 void (*freepage
)(struct page
*);
555 pgoff_t offset
= old
->index
;
556 freepage
= mapping
->a_ops
->freepage
;
559 new->mapping
= mapping
;
562 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
563 __delete_from_page_cache(old
, NULL
);
564 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
569 * hugetlb pages do not participate in page cache accounting.
572 __inc_node_page_state(new, NR_FILE_PAGES
);
573 if (PageSwapBacked(new))
574 __inc_node_page_state(new, NR_SHMEM
);
575 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
576 mem_cgroup_migrate(old
, new);
577 radix_tree_preload_end();
585 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
587 static int page_cache_tree_insert(struct address_space
*mapping
,
588 struct page
*page
, void **shadowp
)
590 struct radix_tree_node
*node
;
594 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
601 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
602 if (!radix_tree_exceptional_entry(p
))
605 mapping
->nrexceptional
--;
606 if (!dax_mapping(mapping
)) {
610 workingset_node_shadows_dec(node
);
612 /* DAX can replace empty locked entry with a hole */
614 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY
|
615 RADIX_DAX_ENTRY_LOCK
));
616 /* DAX accounts exceptional entries as normal pages */
618 workingset_node_pages_dec(node
);
619 /* Wakeup waiters for exceptional entry lock */
620 dax_wake_mapping_entry_waiter(mapping
, page
->index
,
624 radix_tree_replace_slot(slot
, page
);
627 workingset_node_pages_inc(node
);
629 * Don't track node that contains actual pages.
631 * Avoid acquiring the list_lru lock if already
632 * untracked. The list_empty() test is safe as
633 * node->private_list is protected by
634 * mapping->tree_lock.
636 if (!list_empty(&node
->private_list
))
637 list_lru_del(&workingset_shadow_nodes
,
638 &node
->private_list
);
643 static int __add_to_page_cache_locked(struct page
*page
,
644 struct address_space
*mapping
,
645 pgoff_t offset
, gfp_t gfp_mask
,
648 int huge
= PageHuge(page
);
649 struct mem_cgroup
*memcg
;
652 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
653 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
656 error
= mem_cgroup_try_charge(page
, current
->mm
,
657 gfp_mask
, &memcg
, false);
662 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
665 mem_cgroup_cancel_charge(page
, memcg
, false);
670 page
->mapping
= mapping
;
671 page
->index
= offset
;
673 spin_lock_irq(&mapping
->tree_lock
);
674 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
675 radix_tree_preload_end();
679 /* hugetlb pages do not participate in page cache accounting. */
681 __inc_node_page_state(page
, NR_FILE_PAGES
);
682 spin_unlock_irq(&mapping
->tree_lock
);
684 mem_cgroup_commit_charge(page
, memcg
, false, false);
685 trace_mm_filemap_add_to_page_cache(page
);
688 page
->mapping
= NULL
;
689 /* Leave page->index set: truncation relies upon it */
690 spin_unlock_irq(&mapping
->tree_lock
);
692 mem_cgroup_cancel_charge(page
, memcg
, false);
698 * add_to_page_cache_locked - add a locked page to the pagecache
700 * @mapping: the page's address_space
701 * @offset: page index
702 * @gfp_mask: page allocation mode
704 * This function is used to add a page to the pagecache. It must be locked.
705 * This function does not add the page to the LRU. The caller must do that.
707 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
708 pgoff_t offset
, gfp_t gfp_mask
)
710 return __add_to_page_cache_locked(page
, mapping
, offset
,
713 EXPORT_SYMBOL(add_to_page_cache_locked
);
715 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
716 pgoff_t offset
, gfp_t gfp_mask
)
721 __SetPageLocked(page
);
722 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
725 __ClearPageLocked(page
);
728 * The page might have been evicted from cache only
729 * recently, in which case it should be activated like
730 * any other repeatedly accessed page.
731 * The exception is pages getting rewritten; evicting other
732 * data from the working set, only to cache data that will
733 * get overwritten with something else, is a waste of memory.
735 if (!(gfp_mask
& __GFP_WRITE
) &&
736 shadow
&& workingset_refault(shadow
)) {
738 workingset_activation(page
);
740 ClearPageActive(page
);
745 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
748 struct page
*__page_cache_alloc(gfp_t gfp
)
753 if (cpuset_do_page_mem_spread()) {
754 unsigned int cpuset_mems_cookie
;
756 cpuset_mems_cookie
= read_mems_allowed_begin();
757 n
= cpuset_mem_spread_node();
758 page
= __alloc_pages_node(n
, gfp
, 0);
759 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
763 return alloc_pages(gfp
, 0);
765 EXPORT_SYMBOL(__page_cache_alloc
);
769 * In order to wait for pages to become available there must be
770 * waitqueues associated with pages. By using a hash table of
771 * waitqueues where the bucket discipline is to maintain all
772 * waiters on the same queue and wake all when any of the pages
773 * become available, and for the woken contexts to check to be
774 * sure the appropriate page became available, this saves space
775 * at a cost of "thundering herd" phenomena during rare hash
778 wait_queue_head_t
*page_waitqueue(struct page
*page
)
780 const struct zone
*zone
= page_zone(page
);
782 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
784 EXPORT_SYMBOL(page_waitqueue
);
786 void wait_on_page_bit(struct page
*page
, int bit_nr
)
788 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
790 if (test_bit(bit_nr
, &page
->flags
))
791 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
792 TASK_UNINTERRUPTIBLE
);
794 EXPORT_SYMBOL(wait_on_page_bit
);
796 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
798 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
800 if (!test_bit(bit_nr
, &page
->flags
))
803 return __wait_on_bit(page_waitqueue(page
), &wait
,
804 bit_wait_io
, TASK_KILLABLE
);
807 int wait_on_page_bit_killable_timeout(struct page
*page
,
808 int bit_nr
, unsigned long timeout
)
810 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
812 wait
.key
.timeout
= jiffies
+ timeout
;
813 if (!test_bit(bit_nr
, &page
->flags
))
815 return __wait_on_bit(page_waitqueue(page
), &wait
,
816 bit_wait_io_timeout
, TASK_KILLABLE
);
818 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
821 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
822 * @page: Page defining the wait queue of interest
823 * @waiter: Waiter to add to the queue
825 * Add an arbitrary @waiter to the wait queue for the nominated @page.
827 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
829 wait_queue_head_t
*q
= page_waitqueue(page
);
832 spin_lock_irqsave(&q
->lock
, flags
);
833 __add_wait_queue(q
, waiter
);
834 spin_unlock_irqrestore(&q
->lock
, flags
);
836 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
839 * unlock_page - unlock a locked page
842 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
843 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
844 * mechanism between PageLocked pages and PageWriteback pages is shared.
845 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
847 * The mb is necessary to enforce ordering between the clear_bit and the read
848 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
850 void unlock_page(struct page
*page
)
852 page
= compound_head(page
);
853 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
854 clear_bit_unlock(PG_locked
, &page
->flags
);
855 smp_mb__after_atomic();
856 wake_up_page(page
, PG_locked
);
858 EXPORT_SYMBOL(unlock_page
);
861 * end_page_writeback - end writeback against a page
864 void end_page_writeback(struct page
*page
)
867 * TestClearPageReclaim could be used here but it is an atomic
868 * operation and overkill in this particular case. Failing to
869 * shuffle a page marked for immediate reclaim is too mild to
870 * justify taking an atomic operation penalty at the end of
871 * ever page writeback.
873 if (PageReclaim(page
)) {
874 ClearPageReclaim(page
);
875 rotate_reclaimable_page(page
);
878 if (!test_clear_page_writeback(page
))
881 smp_mb__after_atomic();
882 wake_up_page(page
, PG_writeback
);
884 EXPORT_SYMBOL(end_page_writeback
);
887 * After completing I/O on a page, call this routine to update the page
888 * flags appropriately
890 void page_endio(struct page
*page
, bool is_write
, int err
)
894 SetPageUptodate(page
);
896 ClearPageUptodate(page
);
904 mapping_set_error(page
->mapping
, err
);
906 end_page_writeback(page
);
909 EXPORT_SYMBOL_GPL(page_endio
);
912 * __lock_page - get a lock on the page, assuming we need to sleep to get it
913 * @page: the page to lock
915 void __lock_page(struct page
*page
)
917 struct page
*page_head
= compound_head(page
);
918 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
920 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
921 TASK_UNINTERRUPTIBLE
);
923 EXPORT_SYMBOL(__lock_page
);
925 int __lock_page_killable(struct page
*page
)
927 struct page
*page_head
= compound_head(page
);
928 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
930 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
931 bit_wait_io
, TASK_KILLABLE
);
933 EXPORT_SYMBOL_GPL(__lock_page_killable
);
937 * 1 - page is locked; mmap_sem is still held.
938 * 0 - page is not locked.
939 * mmap_sem has been released (up_read()), unless flags had both
940 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
941 * which case mmap_sem is still held.
943 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
944 * with the page locked and the mmap_sem unperturbed.
946 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
949 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
951 * CAUTION! In this case, mmap_sem is not released
952 * even though return 0.
954 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
957 up_read(&mm
->mmap_sem
);
958 if (flags
& FAULT_FLAG_KILLABLE
)
959 wait_on_page_locked_killable(page
);
961 wait_on_page_locked(page
);
964 if (flags
& FAULT_FLAG_KILLABLE
) {
967 ret
= __lock_page_killable(page
);
969 up_read(&mm
->mmap_sem
);
979 * page_cache_next_hole - find the next hole (not-present entry)
982 * @max_scan: maximum range to search
984 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
985 * lowest indexed hole.
987 * Returns: the index of the hole if found, otherwise returns an index
988 * outside of the set specified (in which case 'return - index >=
989 * max_scan' will be true). In rare cases of index wrap-around, 0 will
992 * page_cache_next_hole may be called under rcu_read_lock. However,
993 * like radix_tree_gang_lookup, this will not atomically search a
994 * snapshot of the tree at a single point in time. For example, if a
995 * hole is created at index 5, then subsequently a hole is created at
996 * index 10, page_cache_next_hole covering both indexes may return 10
997 * if called under rcu_read_lock.
999 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1000 pgoff_t index
, unsigned long max_scan
)
1004 for (i
= 0; i
< max_scan
; i
++) {
1007 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1008 if (!page
|| radix_tree_exceptional_entry(page
))
1017 EXPORT_SYMBOL(page_cache_next_hole
);
1020 * page_cache_prev_hole - find the prev hole (not-present entry)
1023 * @max_scan: maximum range to search
1025 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1028 * Returns: the index of the hole if found, otherwise returns an index
1029 * outside of the set specified (in which case 'index - return >=
1030 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1033 * page_cache_prev_hole may be called under rcu_read_lock. However,
1034 * like radix_tree_gang_lookup, this will not atomically search a
1035 * snapshot of the tree at a single point in time. For example, if a
1036 * hole is created at index 10, then subsequently a hole is created at
1037 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1038 * called under rcu_read_lock.
1040 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1041 pgoff_t index
, unsigned long max_scan
)
1045 for (i
= 0; i
< max_scan
; i
++) {
1048 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1049 if (!page
|| radix_tree_exceptional_entry(page
))
1052 if (index
== ULONG_MAX
)
1058 EXPORT_SYMBOL(page_cache_prev_hole
);
1061 * find_get_entry - find and get a page cache entry
1062 * @mapping: the address_space to search
1063 * @offset: the page cache index
1065 * Looks up the page cache slot at @mapping & @offset. If there is a
1066 * page cache page, it is returned with an increased refcount.
1068 * If the slot holds a shadow entry of a previously evicted page, or a
1069 * swap entry from shmem/tmpfs, it is returned.
1071 * Otherwise, %NULL is returned.
1073 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1076 struct page
*head
, *page
;
1081 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1083 page
= radix_tree_deref_slot(pagep
);
1084 if (unlikely(!page
))
1086 if (radix_tree_exception(page
)) {
1087 if (radix_tree_deref_retry(page
))
1090 * A shadow entry of a recently evicted page,
1091 * or a swap entry from shmem/tmpfs. Return
1092 * it without attempting to raise page count.
1097 head
= compound_head(page
);
1098 if (!page_cache_get_speculative(head
))
1101 /* The page was split under us? */
1102 if (compound_head(page
) != head
) {
1108 * Has the page moved?
1109 * This is part of the lockless pagecache protocol. See
1110 * include/linux/pagemap.h for details.
1112 if (unlikely(page
!= *pagep
)) {
1122 EXPORT_SYMBOL(find_get_entry
);
1125 * find_lock_entry - locate, pin and lock a page cache entry
1126 * @mapping: the address_space to search
1127 * @offset: the page cache index
1129 * Looks up the page cache slot at @mapping & @offset. If there is a
1130 * page cache page, it is returned locked and with an increased
1133 * If the slot holds a shadow entry of a previously evicted page, or a
1134 * swap entry from shmem/tmpfs, it is returned.
1136 * Otherwise, %NULL is returned.
1138 * find_lock_entry() may sleep.
1140 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1145 page
= find_get_entry(mapping
, offset
);
1146 if (page
&& !radix_tree_exception(page
)) {
1148 /* Has the page been truncated? */
1149 if (unlikely(page_mapping(page
) != mapping
)) {
1154 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1158 EXPORT_SYMBOL(find_lock_entry
);
1161 * pagecache_get_page - find and get a page reference
1162 * @mapping: the address_space to search
1163 * @offset: the page index
1164 * @fgp_flags: PCG flags
1165 * @gfp_mask: gfp mask to use for the page cache data page allocation
1167 * Looks up the page cache slot at @mapping & @offset.
1169 * PCG flags modify how the page is returned.
1171 * FGP_ACCESSED: the page will be marked accessed
1172 * FGP_LOCK: Page is return locked
1173 * FGP_CREAT: If page is not present then a new page is allocated using
1174 * @gfp_mask and added to the page cache and the VM's LRU
1175 * list. The page is returned locked and with an increased
1176 * refcount. Otherwise, %NULL is returned.
1178 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1179 * if the GFP flags specified for FGP_CREAT are atomic.
1181 * If there is a page cache page, it is returned with an increased refcount.
1183 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1184 int fgp_flags
, gfp_t gfp_mask
)
1189 page
= find_get_entry(mapping
, offset
);
1190 if (radix_tree_exceptional_entry(page
))
1195 if (fgp_flags
& FGP_LOCK
) {
1196 if (fgp_flags
& FGP_NOWAIT
) {
1197 if (!trylock_page(page
)) {
1205 /* Has the page been truncated? */
1206 if (unlikely(page
->mapping
!= mapping
)) {
1211 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1214 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1215 mark_page_accessed(page
);
1218 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1220 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1221 gfp_mask
|= __GFP_WRITE
;
1222 if (fgp_flags
& FGP_NOFS
)
1223 gfp_mask
&= ~__GFP_FS
;
1225 page
= __page_cache_alloc(gfp_mask
);
1229 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1230 fgp_flags
|= FGP_LOCK
;
1232 /* Init accessed so avoid atomic mark_page_accessed later */
1233 if (fgp_flags
& FGP_ACCESSED
)
1234 __SetPageReferenced(page
);
1236 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1237 gfp_mask
& GFP_RECLAIM_MASK
);
1238 if (unlikely(err
)) {
1248 EXPORT_SYMBOL(pagecache_get_page
);
1251 * find_get_entries - gang pagecache lookup
1252 * @mapping: The address_space to search
1253 * @start: The starting page cache index
1254 * @nr_entries: The maximum number of entries
1255 * @entries: Where the resulting entries are placed
1256 * @indices: The cache indices corresponding to the entries in @entries
1258 * find_get_entries() will search for and return a group of up to
1259 * @nr_entries entries in the mapping. The entries are placed at
1260 * @entries. find_get_entries() takes a reference against any actual
1263 * The search returns a group of mapping-contiguous page cache entries
1264 * with ascending indexes. There may be holes in the indices due to
1265 * not-present pages.
1267 * Any shadow entries of evicted pages, or swap entries from
1268 * shmem/tmpfs, are included in the returned array.
1270 * find_get_entries() returns the number of pages and shadow entries
1273 unsigned find_get_entries(struct address_space
*mapping
,
1274 pgoff_t start
, unsigned int nr_entries
,
1275 struct page
**entries
, pgoff_t
*indices
)
1278 unsigned int ret
= 0;
1279 struct radix_tree_iter iter
;
1285 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1286 struct page
*head
, *page
;
1288 page
= radix_tree_deref_slot(slot
);
1289 if (unlikely(!page
))
1291 if (radix_tree_exception(page
)) {
1292 if (radix_tree_deref_retry(page
)) {
1293 slot
= radix_tree_iter_retry(&iter
);
1297 * A shadow entry of a recently evicted page, a swap
1298 * entry from shmem/tmpfs or a DAX entry. Return it
1299 * without attempting to raise page count.
1304 head
= compound_head(page
);
1305 if (!page_cache_get_speculative(head
))
1308 /* The page was split under us? */
1309 if (compound_head(page
) != head
) {
1314 /* Has the page moved? */
1315 if (unlikely(page
!= *slot
)) {
1320 indices
[ret
] = iter
.index
;
1321 entries
[ret
] = page
;
1322 if (++ret
== nr_entries
)
1330 * find_get_pages - gang pagecache lookup
1331 * @mapping: The address_space to search
1332 * @start: The starting page index
1333 * @nr_pages: The maximum number of pages
1334 * @pages: Where the resulting pages are placed
1336 * find_get_pages() will search for and return a group of up to
1337 * @nr_pages pages in the mapping. The pages are placed at @pages.
1338 * find_get_pages() takes a reference against the returned pages.
1340 * The search returns a group of mapping-contiguous pages with ascending
1341 * indexes. There may be holes in the indices due to not-present pages.
1343 * find_get_pages() returns the number of pages which were found.
1345 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1346 unsigned int nr_pages
, struct page
**pages
)
1348 struct radix_tree_iter iter
;
1352 if (unlikely(!nr_pages
))
1356 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1357 struct page
*head
, *page
;
1359 page
= radix_tree_deref_slot(slot
);
1360 if (unlikely(!page
))
1363 if (radix_tree_exception(page
)) {
1364 if (radix_tree_deref_retry(page
)) {
1365 slot
= radix_tree_iter_retry(&iter
);
1369 * A shadow entry of a recently evicted page,
1370 * or a swap entry from shmem/tmpfs. Skip
1376 head
= compound_head(page
);
1377 if (!page_cache_get_speculative(head
))
1380 /* The page was split under us? */
1381 if (compound_head(page
) != head
) {
1386 /* Has the page moved? */
1387 if (unlikely(page
!= *slot
)) {
1393 if (++ret
== nr_pages
)
1402 * find_get_pages_contig - gang contiguous pagecache lookup
1403 * @mapping: The address_space to search
1404 * @index: The starting page index
1405 * @nr_pages: The maximum number of pages
1406 * @pages: Where the resulting pages are placed
1408 * find_get_pages_contig() works exactly like find_get_pages(), except
1409 * that the returned number of pages are guaranteed to be contiguous.
1411 * find_get_pages_contig() returns the number of pages which were found.
1413 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1414 unsigned int nr_pages
, struct page
**pages
)
1416 struct radix_tree_iter iter
;
1418 unsigned int ret
= 0;
1420 if (unlikely(!nr_pages
))
1424 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1425 struct page
*head
, *page
;
1427 page
= radix_tree_deref_slot(slot
);
1428 /* The hole, there no reason to continue */
1429 if (unlikely(!page
))
1432 if (radix_tree_exception(page
)) {
1433 if (radix_tree_deref_retry(page
)) {
1434 slot
= radix_tree_iter_retry(&iter
);
1438 * A shadow entry of a recently evicted page,
1439 * or a swap entry from shmem/tmpfs. Stop
1440 * looking for contiguous pages.
1445 head
= compound_head(page
);
1446 if (!page_cache_get_speculative(head
))
1449 /* The page was split under us? */
1450 if (compound_head(page
) != head
) {
1455 /* Has the page moved? */
1456 if (unlikely(page
!= *slot
)) {
1462 * must check mapping and index after taking the ref.
1463 * otherwise we can get both false positives and false
1464 * negatives, which is just confusing to the caller.
1466 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1472 if (++ret
== nr_pages
)
1478 EXPORT_SYMBOL(find_get_pages_contig
);
1481 * find_get_pages_tag - find and return pages that match @tag
1482 * @mapping: the address_space to search
1483 * @index: the starting page index
1484 * @tag: the tag index
1485 * @nr_pages: the maximum number of pages
1486 * @pages: where the resulting pages are placed
1488 * Like find_get_pages, except we only return pages which are tagged with
1489 * @tag. We update @index to index the next page for the traversal.
1491 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1492 int tag
, unsigned int nr_pages
, struct page
**pages
)
1494 struct radix_tree_iter iter
;
1498 if (unlikely(!nr_pages
))
1502 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1503 &iter
, *index
, tag
) {
1504 struct page
*head
, *page
;
1506 page
= radix_tree_deref_slot(slot
);
1507 if (unlikely(!page
))
1510 if (radix_tree_exception(page
)) {
1511 if (radix_tree_deref_retry(page
)) {
1512 slot
= radix_tree_iter_retry(&iter
);
1516 * A shadow entry of a recently evicted page.
1518 * Those entries should never be tagged, but
1519 * this tree walk is lockless and the tags are
1520 * looked up in bulk, one radix tree node at a
1521 * time, so there is a sizable window for page
1522 * reclaim to evict a page we saw tagged.
1529 head
= compound_head(page
);
1530 if (!page_cache_get_speculative(head
))
1533 /* The page was split under us? */
1534 if (compound_head(page
) != head
) {
1539 /* Has the page moved? */
1540 if (unlikely(page
!= *slot
)) {
1546 if (++ret
== nr_pages
)
1553 *index
= pages
[ret
- 1]->index
+ 1;
1557 EXPORT_SYMBOL(find_get_pages_tag
);
1560 * find_get_entries_tag - find and return entries that match @tag
1561 * @mapping: the address_space to search
1562 * @start: the starting page cache index
1563 * @tag: the tag index
1564 * @nr_entries: the maximum number of entries
1565 * @entries: where the resulting entries are placed
1566 * @indices: the cache indices corresponding to the entries in @entries
1568 * Like find_get_entries, except we only return entries which are tagged with
1571 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1572 int tag
, unsigned int nr_entries
,
1573 struct page
**entries
, pgoff_t
*indices
)
1576 unsigned int ret
= 0;
1577 struct radix_tree_iter iter
;
1583 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1584 &iter
, start
, tag
) {
1585 struct page
*head
, *page
;
1587 page
= radix_tree_deref_slot(slot
);
1588 if (unlikely(!page
))
1590 if (radix_tree_exception(page
)) {
1591 if (radix_tree_deref_retry(page
)) {
1592 slot
= radix_tree_iter_retry(&iter
);
1597 * A shadow entry of a recently evicted page, a swap
1598 * entry from shmem/tmpfs or a DAX entry. Return it
1599 * without attempting to raise page count.
1604 head
= compound_head(page
);
1605 if (!page_cache_get_speculative(head
))
1608 /* The page was split under us? */
1609 if (compound_head(page
) != head
) {
1614 /* Has the page moved? */
1615 if (unlikely(page
!= *slot
)) {
1620 indices
[ret
] = iter
.index
;
1621 entries
[ret
] = page
;
1622 if (++ret
== nr_entries
)
1628 EXPORT_SYMBOL(find_get_entries_tag
);
1631 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1632 * a _large_ part of the i/o request. Imagine the worst scenario:
1634 * ---R__________________________________________B__________
1635 * ^ reading here ^ bad block(assume 4k)
1637 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1638 * => failing the whole request => read(R) => read(R+1) =>
1639 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1640 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1641 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1643 * It is going insane. Fix it by quickly scaling down the readahead size.
1645 static void shrink_readahead_size_eio(struct file
*filp
,
1646 struct file_ra_state
*ra
)
1652 * do_generic_file_read - generic file read routine
1653 * @filp: the file to read
1654 * @ppos: current file position
1655 * @iter: data destination
1656 * @written: already copied
1658 * This is a generic file read routine, and uses the
1659 * mapping->a_ops->readpage() function for the actual low-level stuff.
1661 * This is really ugly. But the goto's actually try to clarify some
1662 * of the logic when it comes to error handling etc.
1664 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1665 struct iov_iter
*iter
, ssize_t written
)
1667 struct address_space
*mapping
= filp
->f_mapping
;
1668 struct inode
*inode
= mapping
->host
;
1669 struct file_ra_state
*ra
= &filp
->f_ra
;
1673 unsigned long offset
; /* offset into pagecache page */
1674 unsigned int prev_offset
;
1677 index
= *ppos
>> PAGE_SHIFT
;
1678 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1679 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1680 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1681 offset
= *ppos
& ~PAGE_MASK
;
1687 unsigned long nr
, ret
;
1691 page
= find_get_page(mapping
, index
);
1693 page_cache_sync_readahead(mapping
,
1695 index
, last_index
- index
);
1696 page
= find_get_page(mapping
, index
);
1697 if (unlikely(page
== NULL
))
1698 goto no_cached_page
;
1700 if (PageReadahead(page
)) {
1701 page_cache_async_readahead(mapping
,
1703 index
, last_index
- index
);
1705 if (!PageUptodate(page
)) {
1707 * See comment in do_read_cache_page on why
1708 * wait_on_page_locked is used to avoid unnecessarily
1709 * serialisations and why it's safe.
1711 error
= wait_on_page_locked_killable(page
);
1712 if (unlikely(error
))
1713 goto readpage_error
;
1714 if (PageUptodate(page
))
1717 if (inode
->i_blkbits
== PAGE_SHIFT
||
1718 !mapping
->a_ops
->is_partially_uptodate
)
1719 goto page_not_up_to_date
;
1720 if (!trylock_page(page
))
1721 goto page_not_up_to_date
;
1722 /* Did it get truncated before we got the lock? */
1724 goto page_not_up_to_date_locked
;
1725 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1726 offset
, iter
->count
))
1727 goto page_not_up_to_date_locked
;
1732 * i_size must be checked after we know the page is Uptodate.
1734 * Checking i_size after the check allows us to calculate
1735 * the correct value for "nr", which means the zero-filled
1736 * part of the page is not copied back to userspace (unless
1737 * another truncate extends the file - this is desired though).
1740 isize
= i_size_read(inode
);
1741 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1742 if (unlikely(!isize
|| index
> end_index
)) {
1747 /* nr is the maximum number of bytes to copy from this page */
1749 if (index
== end_index
) {
1750 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1758 /* If users can be writing to this page using arbitrary
1759 * virtual addresses, take care about potential aliasing
1760 * before reading the page on the kernel side.
1762 if (mapping_writably_mapped(mapping
))
1763 flush_dcache_page(page
);
1766 * When a sequential read accesses a page several times,
1767 * only mark it as accessed the first time.
1769 if (prev_index
!= index
|| offset
!= prev_offset
)
1770 mark_page_accessed(page
);
1774 * Ok, we have the page, and it's up-to-date, so
1775 * now we can copy it to user space...
1778 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1780 index
+= offset
>> PAGE_SHIFT
;
1781 offset
&= ~PAGE_MASK
;
1782 prev_offset
= offset
;
1786 if (!iov_iter_count(iter
))
1794 page_not_up_to_date
:
1795 /* Get exclusive access to the page ... */
1796 error
= lock_page_killable(page
);
1797 if (unlikely(error
))
1798 goto readpage_error
;
1800 page_not_up_to_date_locked
:
1801 /* Did it get truncated before we got the lock? */
1802 if (!page
->mapping
) {
1808 /* Did somebody else fill it already? */
1809 if (PageUptodate(page
)) {
1816 * A previous I/O error may have been due to temporary
1817 * failures, eg. multipath errors.
1818 * PG_error will be set again if readpage fails.
1820 ClearPageError(page
);
1821 /* Start the actual read. The read will unlock the page. */
1822 error
= mapping
->a_ops
->readpage(filp
, page
);
1824 if (unlikely(error
)) {
1825 if (error
== AOP_TRUNCATED_PAGE
) {
1830 goto readpage_error
;
1833 if (!PageUptodate(page
)) {
1834 error
= lock_page_killable(page
);
1835 if (unlikely(error
))
1836 goto readpage_error
;
1837 if (!PageUptodate(page
)) {
1838 if (page
->mapping
== NULL
) {
1840 * invalidate_mapping_pages got it
1847 shrink_readahead_size_eio(filp
, ra
);
1849 goto readpage_error
;
1857 /* UHHUH! A synchronous read error occurred. Report it */
1863 * Ok, it wasn't cached, so we need to create a new
1866 page
= page_cache_alloc_cold(mapping
);
1871 error
= add_to_page_cache_lru(page
, mapping
, index
,
1872 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1875 if (error
== -EEXIST
) {
1885 ra
->prev_pos
= prev_index
;
1886 ra
->prev_pos
<<= PAGE_SHIFT
;
1887 ra
->prev_pos
|= prev_offset
;
1889 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
1890 file_accessed(filp
);
1891 return written
? written
: error
;
1895 * generic_file_read_iter - generic filesystem read routine
1896 * @iocb: kernel I/O control block
1897 * @iter: destination for the data read
1899 * This is the "read_iter()" routine for all filesystems
1900 * that can use the page cache directly.
1903 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1905 struct file
*file
= iocb
->ki_filp
;
1907 size_t count
= iov_iter_count(iter
);
1910 goto out
; /* skip atime */
1912 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1913 struct address_space
*mapping
= file
->f_mapping
;
1914 struct inode
*inode
= mapping
->host
;
1917 size
= i_size_read(inode
);
1918 retval
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
1919 iocb
->ki_pos
+ count
- 1);
1921 struct iov_iter data
= *iter
;
1922 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
);
1926 iocb
->ki_pos
+= retval
;
1927 iov_iter_advance(iter
, retval
);
1931 * Btrfs can have a short DIO read if we encounter
1932 * compressed extents, so if there was an error, or if
1933 * we've already read everything we wanted to, or if
1934 * there was a short read because we hit EOF, go ahead
1935 * and return. Otherwise fallthrough to buffered io for
1936 * the rest of the read. Buffered reads will not work for
1937 * DAX files, so don't bother trying.
1939 if (retval
< 0 || !iov_iter_count(iter
) || iocb
->ki_pos
>= size
||
1941 file_accessed(file
);
1946 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
1950 EXPORT_SYMBOL(generic_file_read_iter
);
1954 * page_cache_read - adds requested page to the page cache if not already there
1955 * @file: file to read
1956 * @offset: page index
1957 * @gfp_mask: memory allocation flags
1959 * This adds the requested page to the page cache if it isn't already there,
1960 * and schedules an I/O to read in its contents from disk.
1962 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1964 struct address_space
*mapping
= file
->f_mapping
;
1969 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1973 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1975 ret
= mapping
->a_ops
->readpage(file
, page
);
1976 else if (ret
== -EEXIST
)
1977 ret
= 0; /* losing race to add is OK */
1981 } while (ret
== AOP_TRUNCATED_PAGE
);
1986 #define MMAP_LOTSAMISS (100)
1989 * Synchronous readahead happens when we don't even find
1990 * a page in the page cache at all.
1992 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1993 struct file_ra_state
*ra
,
1997 struct address_space
*mapping
= file
->f_mapping
;
1999 /* If we don't want any read-ahead, don't bother */
2000 if (vma
->vm_flags
& VM_RAND_READ
)
2005 if (vma
->vm_flags
& VM_SEQ_READ
) {
2006 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2011 /* Avoid banging the cache line if not needed */
2012 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2016 * Do we miss much more than hit in this file? If so,
2017 * stop bothering with read-ahead. It will only hurt.
2019 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2025 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2026 ra
->size
= ra
->ra_pages
;
2027 ra
->async_size
= ra
->ra_pages
/ 4;
2028 ra_submit(ra
, mapping
, file
);
2032 * Asynchronous readahead happens when we find the page and PG_readahead,
2033 * so we want to possibly extend the readahead further..
2035 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2036 struct file_ra_state
*ra
,
2041 struct address_space
*mapping
= file
->f_mapping
;
2043 /* If we don't want any read-ahead, don't bother */
2044 if (vma
->vm_flags
& VM_RAND_READ
)
2046 if (ra
->mmap_miss
> 0)
2048 if (PageReadahead(page
))
2049 page_cache_async_readahead(mapping
, ra
, file
,
2050 page
, offset
, ra
->ra_pages
);
2054 * filemap_fault - read in file data for page fault handling
2055 * @vma: vma in which the fault was taken
2056 * @vmf: struct vm_fault containing details of the fault
2058 * filemap_fault() is invoked via the vma operations vector for a
2059 * mapped memory region to read in file data during a page fault.
2061 * The goto's are kind of ugly, but this streamlines the normal case of having
2062 * it in the page cache, and handles the special cases reasonably without
2063 * having a lot of duplicated code.
2065 * vma->vm_mm->mmap_sem must be held on entry.
2067 * If our return value has VM_FAULT_RETRY set, it's because
2068 * lock_page_or_retry() returned 0.
2069 * The mmap_sem has usually been released in this case.
2070 * See __lock_page_or_retry() for the exception.
2072 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2073 * has not been released.
2075 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2077 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2080 struct file
*file
= vma
->vm_file
;
2081 struct address_space
*mapping
= file
->f_mapping
;
2082 struct file_ra_state
*ra
= &file
->f_ra
;
2083 struct inode
*inode
= mapping
->host
;
2084 pgoff_t offset
= vmf
->pgoff
;
2089 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2090 if (offset
>= size
>> PAGE_SHIFT
)
2091 return VM_FAULT_SIGBUS
;
2094 * Do we have something in the page cache already?
2096 page
= find_get_page(mapping
, offset
);
2097 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2099 * We found the page, so try async readahead before
2100 * waiting for the lock.
2102 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2104 /* No page in the page cache at all */
2105 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2106 count_vm_event(PGMAJFAULT
);
2107 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2108 ret
= VM_FAULT_MAJOR
;
2110 page
= find_get_page(mapping
, offset
);
2112 goto no_cached_page
;
2115 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2117 return ret
| VM_FAULT_RETRY
;
2120 /* Did it get truncated? */
2121 if (unlikely(page
->mapping
!= mapping
)) {
2126 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2129 * We have a locked page in the page cache, now we need to check
2130 * that it's up-to-date. If not, it is going to be due to an error.
2132 if (unlikely(!PageUptodate(page
)))
2133 goto page_not_uptodate
;
2136 * Found the page and have a reference on it.
2137 * We must recheck i_size under page lock.
2139 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2140 if (unlikely(offset
>= size
>> PAGE_SHIFT
)) {
2143 return VM_FAULT_SIGBUS
;
2147 return ret
| VM_FAULT_LOCKED
;
2151 * We're only likely to ever get here if MADV_RANDOM is in
2154 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2157 * The page we want has now been added to the page cache.
2158 * In the unlikely event that someone removed it in the
2159 * meantime, we'll just come back here and read it again.
2165 * An error return from page_cache_read can result if the
2166 * system is low on memory, or a problem occurs while trying
2169 if (error
== -ENOMEM
)
2170 return VM_FAULT_OOM
;
2171 return VM_FAULT_SIGBUS
;
2175 * Umm, take care of errors if the page isn't up-to-date.
2176 * Try to re-read it _once_. We do this synchronously,
2177 * because there really aren't any performance issues here
2178 * and we need to check for errors.
2180 ClearPageError(page
);
2181 error
= mapping
->a_ops
->readpage(file
, page
);
2183 wait_on_page_locked(page
);
2184 if (!PageUptodate(page
))
2189 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2192 /* Things didn't work out. Return zero to tell the mm layer so. */
2193 shrink_readahead_size_eio(file
, ra
);
2194 return VM_FAULT_SIGBUS
;
2196 EXPORT_SYMBOL(filemap_fault
);
2198 void filemap_map_pages(struct fault_env
*fe
,
2199 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2201 struct radix_tree_iter iter
;
2203 struct file
*file
= fe
->vma
->vm_file
;
2204 struct address_space
*mapping
= file
->f_mapping
;
2205 pgoff_t last_pgoff
= start_pgoff
;
2207 struct page
*head
, *page
;
2210 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2212 if (iter
.index
> end_pgoff
)
2215 page
= radix_tree_deref_slot(slot
);
2216 if (unlikely(!page
))
2218 if (radix_tree_exception(page
)) {
2219 if (radix_tree_deref_retry(page
)) {
2220 slot
= radix_tree_iter_retry(&iter
);
2226 head
= compound_head(page
);
2227 if (!page_cache_get_speculative(head
))
2230 /* The page was split under us? */
2231 if (compound_head(page
) != head
) {
2236 /* Has the page moved? */
2237 if (unlikely(page
!= *slot
)) {
2242 if (!PageUptodate(page
) ||
2243 PageReadahead(page
) ||
2246 if (!trylock_page(page
))
2249 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2252 size
= round_up(i_size_read(mapping
->host
), PAGE_SIZE
);
2253 if (page
->index
>= size
>> PAGE_SHIFT
)
2256 if (file
->f_ra
.mmap_miss
> 0)
2257 file
->f_ra
.mmap_miss
--;
2259 fe
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2261 fe
->pte
+= iter
.index
- last_pgoff
;
2262 last_pgoff
= iter
.index
;
2263 if (alloc_set_pte(fe
, NULL
, page
))
2272 /* Huge page is mapped? No need to proceed. */
2273 if (pmd_trans_huge(*fe
->pmd
))
2275 if (iter
.index
== end_pgoff
)
2280 EXPORT_SYMBOL(filemap_map_pages
);
2282 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2284 struct page
*page
= vmf
->page
;
2285 struct inode
*inode
= file_inode(vma
->vm_file
);
2286 int ret
= VM_FAULT_LOCKED
;
2288 sb_start_pagefault(inode
->i_sb
);
2289 file_update_time(vma
->vm_file
);
2291 if (page
->mapping
!= inode
->i_mapping
) {
2293 ret
= VM_FAULT_NOPAGE
;
2297 * We mark the page dirty already here so that when freeze is in
2298 * progress, we are guaranteed that writeback during freezing will
2299 * see the dirty page and writeprotect it again.
2301 set_page_dirty(page
);
2302 wait_for_stable_page(page
);
2304 sb_end_pagefault(inode
->i_sb
);
2307 EXPORT_SYMBOL(filemap_page_mkwrite
);
2309 const struct vm_operations_struct generic_file_vm_ops
= {
2310 .fault
= filemap_fault
,
2311 .map_pages
= filemap_map_pages
,
2312 .page_mkwrite
= filemap_page_mkwrite
,
2315 /* This is used for a general mmap of a disk file */
2317 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2319 struct address_space
*mapping
= file
->f_mapping
;
2321 if (!mapping
->a_ops
->readpage
)
2323 file_accessed(file
);
2324 vma
->vm_ops
= &generic_file_vm_ops
;
2329 * This is for filesystems which do not implement ->writepage.
2331 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2333 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2335 return generic_file_mmap(file
, vma
);
2338 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2342 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2346 #endif /* CONFIG_MMU */
2348 EXPORT_SYMBOL(generic_file_mmap
);
2349 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2351 static struct page
*wait_on_page_read(struct page
*page
)
2353 if (!IS_ERR(page
)) {
2354 wait_on_page_locked(page
);
2355 if (!PageUptodate(page
)) {
2357 page
= ERR_PTR(-EIO
);
2363 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2365 int (*filler
)(void *, struct page
*),
2372 page
= find_get_page(mapping
, index
);
2374 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2376 return ERR_PTR(-ENOMEM
);
2377 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2378 if (unlikely(err
)) {
2382 /* Presumably ENOMEM for radix tree node */
2383 return ERR_PTR(err
);
2387 err
= filler(data
, page
);
2390 return ERR_PTR(err
);
2393 page
= wait_on_page_read(page
);
2398 if (PageUptodate(page
))
2402 * Page is not up to date and may be locked due one of the following
2403 * case a: Page is being filled and the page lock is held
2404 * case b: Read/write error clearing the page uptodate status
2405 * case c: Truncation in progress (page locked)
2406 * case d: Reclaim in progress
2408 * Case a, the page will be up to date when the page is unlocked.
2409 * There is no need to serialise on the page lock here as the page
2410 * is pinned so the lock gives no additional protection. Even if the
2411 * the page is truncated, the data is still valid if PageUptodate as
2412 * it's a race vs truncate race.
2413 * Case b, the page will not be up to date
2414 * Case c, the page may be truncated but in itself, the data may still
2415 * be valid after IO completes as it's a read vs truncate race. The
2416 * operation must restart if the page is not uptodate on unlock but
2417 * otherwise serialising on page lock to stabilise the mapping gives
2418 * no additional guarantees to the caller as the page lock is
2419 * released before return.
2420 * Case d, similar to truncation. If reclaim holds the page lock, it
2421 * will be a race with remove_mapping that determines if the mapping
2422 * is valid on unlock but otherwise the data is valid and there is
2423 * no need to serialise with page lock.
2425 * As the page lock gives no additional guarantee, we optimistically
2426 * wait on the page to be unlocked and check if it's up to date and
2427 * use the page if it is. Otherwise, the page lock is required to
2428 * distinguish between the different cases. The motivation is that we
2429 * avoid spurious serialisations and wakeups when multiple processes
2430 * wait on the same page for IO to complete.
2432 wait_on_page_locked(page
);
2433 if (PageUptodate(page
))
2436 /* Distinguish between all the cases under the safety of the lock */
2439 /* Case c or d, restart the operation */
2440 if (!page
->mapping
) {
2446 /* Someone else locked and filled the page in a very small window */
2447 if (PageUptodate(page
)) {
2454 mark_page_accessed(page
);
2459 * read_cache_page - read into page cache, fill it if needed
2460 * @mapping: the page's address_space
2461 * @index: the page index
2462 * @filler: function to perform the read
2463 * @data: first arg to filler(data, page) function, often left as NULL
2465 * Read into the page cache. If a page already exists, and PageUptodate() is
2466 * not set, try to fill the page and wait for it to become unlocked.
2468 * If the page does not get brought uptodate, return -EIO.
2470 struct page
*read_cache_page(struct address_space
*mapping
,
2472 int (*filler
)(void *, struct page
*),
2475 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2477 EXPORT_SYMBOL(read_cache_page
);
2480 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2481 * @mapping: the page's address_space
2482 * @index: the page index
2483 * @gfp: the page allocator flags to use if allocating
2485 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2486 * any new page allocations done using the specified allocation flags.
2488 * If the page does not get brought uptodate, return -EIO.
2490 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2494 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2496 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2498 EXPORT_SYMBOL(read_cache_page_gfp
);
2501 * Performs necessary checks before doing a write
2503 * Can adjust writing position or amount of bytes to write.
2504 * Returns appropriate error code that caller should return or
2505 * zero in case that write should be allowed.
2507 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2509 struct file
*file
= iocb
->ki_filp
;
2510 struct inode
*inode
= file
->f_mapping
->host
;
2511 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2514 if (!iov_iter_count(from
))
2517 /* FIXME: this is for backwards compatibility with 2.4 */
2518 if (iocb
->ki_flags
& IOCB_APPEND
)
2519 iocb
->ki_pos
= i_size_read(inode
);
2523 if (limit
!= RLIM_INFINITY
) {
2524 if (iocb
->ki_pos
>= limit
) {
2525 send_sig(SIGXFSZ
, current
, 0);
2528 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2534 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2535 !(file
->f_flags
& O_LARGEFILE
))) {
2536 if (pos
>= MAX_NON_LFS
)
2538 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2542 * Are we about to exceed the fs block limit ?
2544 * If we have written data it becomes a short write. If we have
2545 * exceeded without writing data we send a signal and return EFBIG.
2546 * Linus frestrict idea will clean these up nicely..
2548 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2551 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2552 return iov_iter_count(from
);
2554 EXPORT_SYMBOL(generic_write_checks
);
2556 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2557 loff_t pos
, unsigned len
, unsigned flags
,
2558 struct page
**pagep
, void **fsdata
)
2560 const struct address_space_operations
*aops
= mapping
->a_ops
;
2562 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2565 EXPORT_SYMBOL(pagecache_write_begin
);
2567 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2568 loff_t pos
, unsigned len
, unsigned copied
,
2569 struct page
*page
, void *fsdata
)
2571 const struct address_space_operations
*aops
= mapping
->a_ops
;
2573 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2575 EXPORT_SYMBOL(pagecache_write_end
);
2578 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2580 struct file
*file
= iocb
->ki_filp
;
2581 struct address_space
*mapping
= file
->f_mapping
;
2582 struct inode
*inode
= mapping
->host
;
2583 loff_t pos
= iocb
->ki_pos
;
2587 struct iov_iter data
;
2589 write_len
= iov_iter_count(from
);
2590 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2592 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2597 * After a write we want buffered reads to be sure to go to disk to get
2598 * the new data. We invalidate clean cached page from the region we're
2599 * about to write. We do this *before* the write so that we can return
2600 * without clobbering -EIOCBQUEUED from ->direct_IO().
2602 if (mapping
->nrpages
) {
2603 written
= invalidate_inode_pages2_range(mapping
,
2604 pos
>> PAGE_SHIFT
, end
);
2606 * If a page can not be invalidated, return 0 to fall back
2607 * to buffered write.
2610 if (written
== -EBUSY
)
2617 written
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2620 * Finally, try again to invalidate clean pages which might have been
2621 * cached by non-direct readahead, or faulted in by get_user_pages()
2622 * if the source of the write was an mmap'ed region of the file
2623 * we're writing. Either one is a pretty crazy thing to do,
2624 * so we don't support it 100%. If this invalidation
2625 * fails, tough, the write still worked...
2627 if (mapping
->nrpages
) {
2628 invalidate_inode_pages2_range(mapping
,
2629 pos
>> PAGE_SHIFT
, end
);
2634 iov_iter_advance(from
, written
);
2635 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2636 i_size_write(inode
, pos
);
2637 mark_inode_dirty(inode
);
2644 EXPORT_SYMBOL(generic_file_direct_write
);
2647 * Find or create a page at the given pagecache position. Return the locked
2648 * page. This function is specifically for buffered writes.
2650 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2651 pgoff_t index
, unsigned flags
)
2654 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2656 if (flags
& AOP_FLAG_NOFS
)
2657 fgp_flags
|= FGP_NOFS
;
2659 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2660 mapping_gfp_mask(mapping
));
2662 wait_for_stable_page(page
);
2666 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2668 ssize_t
generic_perform_write(struct file
*file
,
2669 struct iov_iter
*i
, loff_t pos
)
2671 struct address_space
*mapping
= file
->f_mapping
;
2672 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2674 ssize_t written
= 0;
2675 unsigned int flags
= 0;
2678 * Copies from kernel address space cannot fail (NFSD is a big user).
2680 if (!iter_is_iovec(i
))
2681 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2685 unsigned long offset
; /* Offset into pagecache page */
2686 unsigned long bytes
; /* Bytes to write to page */
2687 size_t copied
; /* Bytes copied from user */
2690 offset
= (pos
& (PAGE_SIZE
- 1));
2691 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2696 * Bring in the user page that we will copy from _first_.
2697 * Otherwise there's a nasty deadlock on copying from the
2698 * same page as we're writing to, without it being marked
2701 * Not only is this an optimisation, but it is also required
2702 * to check that the address is actually valid, when atomic
2703 * usercopies are used, below.
2705 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2710 if (fatal_signal_pending(current
)) {
2715 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2717 if (unlikely(status
< 0))
2720 if (mapping_writably_mapped(mapping
))
2721 flush_dcache_page(page
);
2723 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2724 flush_dcache_page(page
);
2726 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2728 if (unlikely(status
< 0))
2734 iov_iter_advance(i
, copied
);
2735 if (unlikely(copied
== 0)) {
2737 * If we were unable to copy any data at all, we must
2738 * fall back to a single segment length write.
2740 * If we didn't fallback here, we could livelock
2741 * because not all segments in the iov can be copied at
2742 * once without a pagefault.
2744 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2745 iov_iter_single_seg_count(i
));
2751 balance_dirty_pages_ratelimited(mapping
);
2752 } while (iov_iter_count(i
));
2754 return written
? written
: status
;
2756 EXPORT_SYMBOL(generic_perform_write
);
2759 * __generic_file_write_iter - write data to a file
2760 * @iocb: IO state structure (file, offset, etc.)
2761 * @from: iov_iter with data to write
2763 * This function does all the work needed for actually writing data to a
2764 * file. It does all basic checks, removes SUID from the file, updates
2765 * modification times and calls proper subroutines depending on whether we
2766 * do direct IO or a standard buffered write.
2768 * It expects i_mutex to be grabbed unless we work on a block device or similar
2769 * object which does not need locking at all.
2771 * This function does *not* take care of syncing data in case of O_SYNC write.
2772 * A caller has to handle it. This is mainly due to the fact that we want to
2773 * avoid syncing under i_mutex.
2775 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2777 struct file
*file
= iocb
->ki_filp
;
2778 struct address_space
* mapping
= file
->f_mapping
;
2779 struct inode
*inode
= mapping
->host
;
2780 ssize_t written
= 0;
2784 /* We can write back this queue in page reclaim */
2785 current
->backing_dev_info
= inode_to_bdi(inode
);
2786 err
= file_remove_privs(file
);
2790 err
= file_update_time(file
);
2794 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2795 loff_t pos
, endbyte
;
2797 written
= generic_file_direct_write(iocb
, from
);
2799 * If the write stopped short of completing, fall back to
2800 * buffered writes. Some filesystems do this for writes to
2801 * holes, for example. For DAX files, a buffered write will
2802 * not succeed (even if it did, DAX does not handle dirty
2803 * page-cache pages correctly).
2805 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2808 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2810 * If generic_perform_write() returned a synchronous error
2811 * then we want to return the number of bytes which were
2812 * direct-written, or the error code if that was zero. Note
2813 * that this differs from normal direct-io semantics, which
2814 * will return -EFOO even if some bytes were written.
2816 if (unlikely(status
< 0)) {
2821 * We need to ensure that the page cache pages are written to
2822 * disk and invalidated to preserve the expected O_DIRECT
2825 endbyte
= pos
+ status
- 1;
2826 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2828 iocb
->ki_pos
= endbyte
+ 1;
2830 invalidate_mapping_pages(mapping
,
2832 endbyte
>> PAGE_SHIFT
);
2835 * We don't know how much we wrote, so just return
2836 * the number of bytes which were direct-written
2840 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2841 if (likely(written
> 0))
2842 iocb
->ki_pos
+= written
;
2845 current
->backing_dev_info
= NULL
;
2846 return written
? written
: err
;
2848 EXPORT_SYMBOL(__generic_file_write_iter
);
2851 * generic_file_write_iter - write data to a file
2852 * @iocb: IO state structure
2853 * @from: iov_iter with data to write
2855 * This is a wrapper around __generic_file_write_iter() to be used by most
2856 * filesystems. It takes care of syncing the file in case of O_SYNC file
2857 * and acquires i_mutex as needed.
2859 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2861 struct file
*file
= iocb
->ki_filp
;
2862 struct inode
*inode
= file
->f_mapping
->host
;
2866 ret
= generic_write_checks(iocb
, from
);
2868 ret
= __generic_file_write_iter(iocb
, from
);
2869 inode_unlock(inode
);
2872 ret
= generic_write_sync(iocb
, ret
);
2875 EXPORT_SYMBOL(generic_file_write_iter
);
2878 * try_to_release_page() - release old fs-specific metadata on a page
2880 * @page: the page which the kernel is trying to free
2881 * @gfp_mask: memory allocation flags (and I/O mode)
2883 * The address_space is to try to release any data against the page
2884 * (presumably at page->private). If the release was successful, return `1'.
2885 * Otherwise return zero.
2887 * This may also be called if PG_fscache is set on a page, indicating that the
2888 * page is known to the local caching routines.
2890 * The @gfp_mask argument specifies whether I/O may be performed to release
2891 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2894 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2896 struct address_space
* const mapping
= page
->mapping
;
2898 BUG_ON(!PageLocked(page
));
2899 if (PageWriteback(page
))
2902 if (mapping
&& mapping
->a_ops
->releasepage
)
2903 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2904 return try_to_free_buffers(page
);
2907 EXPORT_SYMBOL(try_to_release_page
);