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
);
224 if (PageTransHuge(page
))
225 __dec_zone_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 static 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
))
290 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
291 * @mapping: address space structure to write
292 * @start: offset in bytes where the range starts
293 * @end: offset in bytes where the range ends (inclusive)
294 * @sync_mode: enable synchronous operation
296 * Start writeback against all of a mapping's dirty pages that lie
297 * within the byte offsets <start, end> inclusive.
299 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
300 * opposed to a regular memory cleansing writeback. The difference between
301 * these two operations is that if a dirty page/buffer is encountered, it must
302 * be waited upon, and not just skipped over.
304 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
305 loff_t end
, int sync_mode
)
308 struct writeback_control wbc
= {
309 .sync_mode
= sync_mode
,
310 .nr_to_write
= LONG_MAX
,
311 .range_start
= start
,
315 if (!mapping_cap_writeback_dirty(mapping
))
318 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
319 ret
= do_writepages(mapping
, &wbc
);
320 wbc_detach_inode(&wbc
);
324 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
327 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
330 int filemap_fdatawrite(struct address_space
*mapping
)
332 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
334 EXPORT_SYMBOL(filemap_fdatawrite
);
336 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
339 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
341 EXPORT_SYMBOL(filemap_fdatawrite_range
);
344 * filemap_flush - mostly a non-blocking flush
345 * @mapping: target address_space
347 * This is a mostly non-blocking flush. Not suitable for data-integrity
348 * purposes - I/O may not be started against all dirty pages.
350 int filemap_flush(struct address_space
*mapping
)
352 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
354 EXPORT_SYMBOL(filemap_flush
);
356 static int __filemap_fdatawait_range(struct address_space
*mapping
,
357 loff_t start_byte
, loff_t end_byte
)
359 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
360 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
365 if (end_byte
< start_byte
)
368 pagevec_init(&pvec
, 0);
369 while ((index
<= end
) &&
370 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
371 PAGECACHE_TAG_WRITEBACK
,
372 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
375 for (i
= 0; i
< nr_pages
; i
++) {
376 struct page
*page
= pvec
.pages
[i
];
378 /* until radix tree lookup accepts end_index */
379 if (page
->index
> end
)
382 wait_on_page_writeback(page
);
383 if (TestClearPageError(page
))
386 pagevec_release(&pvec
);
394 * filemap_fdatawait_range - wait for writeback to complete
395 * @mapping: address space structure to wait for
396 * @start_byte: offset in bytes where the range starts
397 * @end_byte: offset in bytes where the range ends (inclusive)
399 * Walk the list of under-writeback pages of the given address space
400 * in the given range and wait for all of them. Check error status of
401 * the address space and return it.
403 * Since the error status of the address space is cleared by this function,
404 * callers are responsible for checking the return value and handling and/or
405 * reporting the error.
407 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
412 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
413 ret2
= filemap_check_errors(mapping
);
419 EXPORT_SYMBOL(filemap_fdatawait_range
);
422 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
423 * @mapping: address space structure to wait for
425 * Walk the list of under-writeback pages of the given address space
426 * and wait for all of them. Unlike filemap_fdatawait(), this function
427 * does not clear error status of the address space.
429 * Use this function if callers don't handle errors themselves. Expected
430 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
433 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
435 loff_t i_size
= i_size_read(mapping
->host
);
440 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
444 * filemap_fdatawait - wait for all under-writeback pages to complete
445 * @mapping: address space structure to wait for
447 * Walk the list of under-writeback pages of the given address space
448 * and wait for all of them. Check error status of the address space
451 * Since the error status of the address space is cleared by this function,
452 * callers are responsible for checking the return value and handling and/or
453 * reporting the error.
455 int filemap_fdatawait(struct address_space
*mapping
)
457 loff_t i_size
= i_size_read(mapping
->host
);
462 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
464 EXPORT_SYMBOL(filemap_fdatawait
);
466 int filemap_write_and_wait(struct address_space
*mapping
)
470 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
471 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
472 err
= filemap_fdatawrite(mapping
);
474 * Even if the above returned error, the pages may be
475 * written partially (e.g. -ENOSPC), so we wait for it.
476 * But the -EIO is special case, it may indicate the worst
477 * thing (e.g. bug) happened, so we avoid waiting for it.
480 int err2
= filemap_fdatawait(mapping
);
485 err
= filemap_check_errors(mapping
);
489 EXPORT_SYMBOL(filemap_write_and_wait
);
492 * filemap_write_and_wait_range - write out & wait on a file range
493 * @mapping: the address_space for the pages
494 * @lstart: offset in bytes where the range starts
495 * @lend: offset in bytes where the range ends (inclusive)
497 * Write out and wait upon file offsets lstart->lend, inclusive.
499 * Note that `lend' is inclusive (describes the last byte to be written) so
500 * that this function can be used to write to the very end-of-file (end = -1).
502 int filemap_write_and_wait_range(struct address_space
*mapping
,
503 loff_t lstart
, loff_t lend
)
507 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
508 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
509 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
511 /* See comment of filemap_write_and_wait() */
513 int err2
= filemap_fdatawait_range(mapping
,
519 err
= filemap_check_errors(mapping
);
523 EXPORT_SYMBOL(filemap_write_and_wait_range
);
526 * replace_page_cache_page - replace a pagecache page with a new one
527 * @old: page to be replaced
528 * @new: page to replace with
529 * @gfp_mask: allocation mode
531 * This function replaces a page in the pagecache with a new one. On
532 * success it acquires the pagecache reference for the new page and
533 * drops it for the old page. Both the old and new pages must be
534 * locked. This function does not add the new page to the LRU, the
535 * caller must do that.
537 * The remove + add is atomic. The only way this function can fail is
538 * memory allocation failure.
540 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
544 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
545 VM_BUG_ON_PAGE(!PageLocked(new), new);
546 VM_BUG_ON_PAGE(new->mapping
, new);
548 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
550 struct address_space
*mapping
= old
->mapping
;
551 void (*freepage
)(struct page
*);
554 pgoff_t offset
= old
->index
;
555 freepage
= mapping
->a_ops
->freepage
;
558 new->mapping
= mapping
;
561 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
562 __delete_from_page_cache(old
, NULL
);
563 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
568 * hugetlb pages do not participate in page cache accounting.
571 __inc_zone_page_state(new, NR_FILE_PAGES
);
572 if (PageSwapBacked(new))
573 __inc_zone_page_state(new, NR_SHMEM
);
574 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
575 mem_cgroup_migrate(old
, new);
576 radix_tree_preload_end();
584 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
586 static int page_cache_tree_insert(struct address_space
*mapping
,
587 struct page
*page
, void **shadowp
)
589 struct radix_tree_node
*node
;
593 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
600 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
601 if (!radix_tree_exceptional_entry(p
))
604 mapping
->nrexceptional
--;
605 if (!dax_mapping(mapping
)) {
609 workingset_node_shadows_dec(node
);
611 /* DAX can replace empty locked entry with a hole */
613 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY
|
614 RADIX_DAX_ENTRY_LOCK
));
615 /* DAX accounts exceptional entries as normal pages */
617 workingset_node_pages_dec(node
);
618 /* Wakeup waiters for exceptional entry lock */
619 dax_wake_mapping_entry_waiter(mapping
, page
->index
,
623 radix_tree_replace_slot(slot
, page
);
626 workingset_node_pages_inc(node
);
628 * Don't track node that contains actual pages.
630 * Avoid acquiring the list_lru lock if already
631 * untracked. The list_empty() test is safe as
632 * node->private_list is protected by
633 * mapping->tree_lock.
635 if (!list_empty(&node
->private_list
))
636 list_lru_del(&workingset_shadow_nodes
,
637 &node
->private_list
);
642 static int __add_to_page_cache_locked(struct page
*page
,
643 struct address_space
*mapping
,
644 pgoff_t offset
, gfp_t gfp_mask
,
647 int huge
= PageHuge(page
);
648 struct mem_cgroup
*memcg
;
651 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
652 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
655 error
= mem_cgroup_try_charge(page
, current
->mm
,
656 gfp_mask
, &memcg
, false);
661 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
664 mem_cgroup_cancel_charge(page
, memcg
, false);
669 page
->mapping
= mapping
;
670 page
->index
= offset
;
672 spin_lock_irq(&mapping
->tree_lock
);
673 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
674 radix_tree_preload_end();
678 /* hugetlb pages do not participate in page cache accounting. */
680 __inc_zone_page_state(page
, NR_FILE_PAGES
);
681 spin_unlock_irq(&mapping
->tree_lock
);
683 mem_cgroup_commit_charge(page
, memcg
, false, false);
684 trace_mm_filemap_add_to_page_cache(page
);
687 page
->mapping
= NULL
;
688 /* Leave page->index set: truncation relies upon it */
689 spin_unlock_irq(&mapping
->tree_lock
);
691 mem_cgroup_cancel_charge(page
, memcg
, false);
697 * add_to_page_cache_locked - add a locked page to the pagecache
699 * @mapping: the page's address_space
700 * @offset: page index
701 * @gfp_mask: page allocation mode
703 * This function is used to add a page to the pagecache. It must be locked.
704 * This function does not add the page to the LRU. The caller must do that.
706 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
707 pgoff_t offset
, gfp_t gfp_mask
)
709 return __add_to_page_cache_locked(page
, mapping
, offset
,
712 EXPORT_SYMBOL(add_to_page_cache_locked
);
714 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
715 pgoff_t offset
, gfp_t gfp_mask
)
720 __SetPageLocked(page
);
721 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
724 __ClearPageLocked(page
);
727 * The page might have been evicted from cache only
728 * recently, in which case it should be activated like
729 * any other repeatedly accessed page.
730 * The exception is pages getting rewritten; evicting other
731 * data from the working set, only to cache data that will
732 * get overwritten with something else, is a waste of memory.
734 if (!(gfp_mask
& __GFP_WRITE
) &&
735 shadow
&& workingset_refault(shadow
)) {
737 workingset_activation(page
);
739 ClearPageActive(page
);
744 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
747 struct page
*__page_cache_alloc(gfp_t gfp
)
752 if (cpuset_do_page_mem_spread()) {
753 unsigned int cpuset_mems_cookie
;
755 cpuset_mems_cookie
= read_mems_allowed_begin();
756 n
= cpuset_mem_spread_node();
757 page
= __alloc_pages_node(n
, gfp
, 0);
758 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
762 return alloc_pages(gfp
, 0);
764 EXPORT_SYMBOL(__page_cache_alloc
);
768 * In order to wait for pages to become available there must be
769 * waitqueues associated with pages. By using a hash table of
770 * waitqueues where the bucket discipline is to maintain all
771 * waiters on the same queue and wake all when any of the pages
772 * become available, and for the woken contexts to check to be
773 * sure the appropriate page became available, this saves space
774 * at a cost of "thundering herd" phenomena during rare hash
777 wait_queue_head_t
*page_waitqueue(struct page
*page
)
779 const struct zone
*zone
= page_zone(page
);
781 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
783 EXPORT_SYMBOL(page_waitqueue
);
785 void wait_on_page_bit(struct page
*page
, int bit_nr
)
787 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
789 if (test_bit(bit_nr
, &page
->flags
))
790 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
791 TASK_UNINTERRUPTIBLE
);
793 EXPORT_SYMBOL(wait_on_page_bit
);
795 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
797 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
799 if (!test_bit(bit_nr
, &page
->flags
))
802 return __wait_on_bit(page_waitqueue(page
), &wait
,
803 bit_wait_io
, TASK_KILLABLE
);
806 int wait_on_page_bit_killable_timeout(struct page
*page
,
807 int bit_nr
, unsigned long timeout
)
809 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
811 wait
.key
.timeout
= jiffies
+ timeout
;
812 if (!test_bit(bit_nr
, &page
->flags
))
814 return __wait_on_bit(page_waitqueue(page
), &wait
,
815 bit_wait_io_timeout
, TASK_KILLABLE
);
817 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
820 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
821 * @page: Page defining the wait queue of interest
822 * @waiter: Waiter to add to the queue
824 * Add an arbitrary @waiter to the wait queue for the nominated @page.
826 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
828 wait_queue_head_t
*q
= page_waitqueue(page
);
831 spin_lock_irqsave(&q
->lock
, flags
);
832 __add_wait_queue(q
, waiter
);
833 spin_unlock_irqrestore(&q
->lock
, flags
);
835 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
838 * unlock_page - unlock a locked page
841 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
842 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
843 * mechanism between PageLocked pages and PageWriteback pages is shared.
844 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
846 * The mb is necessary to enforce ordering between the clear_bit and the read
847 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
849 void unlock_page(struct page
*page
)
851 page
= compound_head(page
);
852 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
853 clear_bit_unlock(PG_locked
, &page
->flags
);
854 smp_mb__after_atomic();
855 wake_up_page(page
, PG_locked
);
857 EXPORT_SYMBOL(unlock_page
);
860 * end_page_writeback - end writeback against a page
863 void end_page_writeback(struct page
*page
)
866 * TestClearPageReclaim could be used here but it is an atomic
867 * operation and overkill in this particular case. Failing to
868 * shuffle a page marked for immediate reclaim is too mild to
869 * justify taking an atomic operation penalty at the end of
870 * ever page writeback.
872 if (PageReclaim(page
)) {
873 ClearPageReclaim(page
);
874 rotate_reclaimable_page(page
);
877 if (!test_clear_page_writeback(page
))
880 smp_mb__after_atomic();
881 wake_up_page(page
, PG_writeback
);
883 EXPORT_SYMBOL(end_page_writeback
);
886 * After completing I/O on a page, call this routine to update the page
887 * flags appropriately
889 void page_endio(struct page
*page
, int rw
, int err
)
893 SetPageUptodate(page
);
895 ClearPageUptodate(page
);
899 } else { /* rw == WRITE */
903 mapping_set_error(page
->mapping
, err
);
905 end_page_writeback(page
);
908 EXPORT_SYMBOL_GPL(page_endio
);
911 * __lock_page - get a lock on the page, assuming we need to sleep to get it
912 * @page: the page to lock
914 void __lock_page(struct page
*page
)
916 struct page
*page_head
= compound_head(page
);
917 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
919 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
920 TASK_UNINTERRUPTIBLE
);
922 EXPORT_SYMBOL(__lock_page
);
924 int __lock_page_killable(struct page
*page
)
926 struct page
*page_head
= compound_head(page
);
927 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
929 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
930 bit_wait_io
, TASK_KILLABLE
);
932 EXPORT_SYMBOL_GPL(__lock_page_killable
);
936 * 1 - page is locked; mmap_sem is still held.
937 * 0 - page is not locked.
938 * mmap_sem has been released (up_read()), unless flags had both
939 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
940 * which case mmap_sem is still held.
942 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
943 * with the page locked and the mmap_sem unperturbed.
945 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
948 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
950 * CAUTION! In this case, mmap_sem is not released
951 * even though return 0.
953 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
956 up_read(&mm
->mmap_sem
);
957 if (flags
& FAULT_FLAG_KILLABLE
)
958 wait_on_page_locked_killable(page
);
960 wait_on_page_locked(page
);
963 if (flags
& FAULT_FLAG_KILLABLE
) {
966 ret
= __lock_page_killable(page
);
968 up_read(&mm
->mmap_sem
);
978 * page_cache_next_hole - find the next hole (not-present entry)
981 * @max_scan: maximum range to search
983 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
984 * lowest indexed hole.
986 * Returns: the index of the hole if found, otherwise returns an index
987 * outside of the set specified (in which case 'return - index >=
988 * max_scan' will be true). In rare cases of index wrap-around, 0 will
991 * page_cache_next_hole may be called under rcu_read_lock. However,
992 * like radix_tree_gang_lookup, this will not atomically search a
993 * snapshot of the tree at a single point in time. For example, if a
994 * hole is created at index 5, then subsequently a hole is created at
995 * index 10, page_cache_next_hole covering both indexes may return 10
996 * if called under rcu_read_lock.
998 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
999 pgoff_t index
, unsigned long max_scan
)
1003 for (i
= 0; i
< max_scan
; i
++) {
1006 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1007 if (!page
|| radix_tree_exceptional_entry(page
))
1016 EXPORT_SYMBOL(page_cache_next_hole
);
1019 * page_cache_prev_hole - find the prev hole (not-present entry)
1022 * @max_scan: maximum range to search
1024 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1027 * Returns: the index of the hole if found, otherwise returns an index
1028 * outside of the set specified (in which case 'index - return >=
1029 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1032 * page_cache_prev_hole may be called under rcu_read_lock. However,
1033 * like radix_tree_gang_lookup, this will not atomically search a
1034 * snapshot of the tree at a single point in time. For example, if a
1035 * hole is created at index 10, then subsequently a hole is created at
1036 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1037 * called under rcu_read_lock.
1039 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1040 pgoff_t index
, unsigned long max_scan
)
1044 for (i
= 0; i
< max_scan
; i
++) {
1047 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1048 if (!page
|| radix_tree_exceptional_entry(page
))
1051 if (index
== ULONG_MAX
)
1057 EXPORT_SYMBOL(page_cache_prev_hole
);
1060 * find_get_entry - find and get a page cache entry
1061 * @mapping: the address_space to search
1062 * @offset: the page cache index
1064 * Looks up the page cache slot at @mapping & @offset. If there is a
1065 * page cache page, it is returned with an increased refcount.
1067 * If the slot holds a shadow entry of a previously evicted page, or a
1068 * swap entry from shmem/tmpfs, it is returned.
1070 * Otherwise, %NULL is returned.
1072 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1075 struct page
*head
, *page
;
1080 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1082 page
= radix_tree_deref_slot(pagep
);
1083 if (unlikely(!page
))
1085 if (radix_tree_exception(page
)) {
1086 if (radix_tree_deref_retry(page
))
1089 * A shadow entry of a recently evicted page,
1090 * or a swap entry from shmem/tmpfs. Return
1091 * it without attempting to raise page count.
1096 head
= compound_head(page
);
1097 if (!page_cache_get_speculative(head
))
1100 /* The page was split under us? */
1101 if (compound_head(page
) != head
) {
1107 * Has the page moved?
1108 * This is part of the lockless pagecache protocol. See
1109 * include/linux/pagemap.h for details.
1111 if (unlikely(page
!= *pagep
)) {
1121 EXPORT_SYMBOL(find_get_entry
);
1124 * find_lock_entry - locate, pin and lock a page cache entry
1125 * @mapping: the address_space to search
1126 * @offset: the page cache index
1128 * Looks up the page cache slot at @mapping & @offset. If there is a
1129 * page cache page, it is returned locked and with an increased
1132 * If the slot holds a shadow entry of a previously evicted page, or a
1133 * swap entry from shmem/tmpfs, it is returned.
1135 * Otherwise, %NULL is returned.
1137 * find_lock_entry() may sleep.
1139 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1144 page
= find_get_entry(mapping
, offset
);
1145 if (page
&& !radix_tree_exception(page
)) {
1147 /* Has the page been truncated? */
1148 if (unlikely(page_mapping(page
) != mapping
)) {
1153 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1157 EXPORT_SYMBOL(find_lock_entry
);
1160 * pagecache_get_page - find and get a page reference
1161 * @mapping: the address_space to search
1162 * @offset: the page index
1163 * @fgp_flags: PCG flags
1164 * @gfp_mask: gfp mask to use for the page cache data page allocation
1166 * Looks up the page cache slot at @mapping & @offset.
1168 * PCG flags modify how the page is returned.
1170 * FGP_ACCESSED: the page will be marked accessed
1171 * FGP_LOCK: Page is return locked
1172 * FGP_CREAT: If page is not present then a new page is allocated using
1173 * @gfp_mask and added to the page cache and the VM's LRU
1174 * list. The page is returned locked and with an increased
1175 * refcount. Otherwise, %NULL is returned.
1177 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1178 * if the GFP flags specified for FGP_CREAT are atomic.
1180 * If there is a page cache page, it is returned with an increased refcount.
1182 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1183 int fgp_flags
, gfp_t gfp_mask
)
1188 page
= find_get_entry(mapping
, offset
);
1189 if (radix_tree_exceptional_entry(page
))
1194 if (fgp_flags
& FGP_LOCK
) {
1195 if (fgp_flags
& FGP_NOWAIT
) {
1196 if (!trylock_page(page
)) {
1204 /* Has the page been truncated? */
1205 if (unlikely(page
->mapping
!= mapping
)) {
1210 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1213 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1214 mark_page_accessed(page
);
1217 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1219 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1220 gfp_mask
|= __GFP_WRITE
;
1221 if (fgp_flags
& FGP_NOFS
)
1222 gfp_mask
&= ~__GFP_FS
;
1224 page
= __page_cache_alloc(gfp_mask
);
1228 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1229 fgp_flags
|= FGP_LOCK
;
1231 /* Init accessed so avoid atomic mark_page_accessed later */
1232 if (fgp_flags
& FGP_ACCESSED
)
1233 __SetPageReferenced(page
);
1235 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1236 gfp_mask
& GFP_RECLAIM_MASK
);
1237 if (unlikely(err
)) {
1247 EXPORT_SYMBOL(pagecache_get_page
);
1250 * find_get_entries - gang pagecache lookup
1251 * @mapping: The address_space to search
1252 * @start: The starting page cache index
1253 * @nr_entries: The maximum number of entries
1254 * @entries: Where the resulting entries are placed
1255 * @indices: The cache indices corresponding to the entries in @entries
1257 * find_get_entries() will search for and return a group of up to
1258 * @nr_entries entries in the mapping. The entries are placed at
1259 * @entries. find_get_entries() takes a reference against any actual
1262 * The search returns a group of mapping-contiguous page cache entries
1263 * with ascending indexes. There may be holes in the indices due to
1264 * not-present pages.
1266 * Any shadow entries of evicted pages, or swap entries from
1267 * shmem/tmpfs, are included in the returned array.
1269 * find_get_entries() returns the number of pages and shadow entries
1272 unsigned find_get_entries(struct address_space
*mapping
,
1273 pgoff_t start
, unsigned int nr_entries
,
1274 struct page
**entries
, pgoff_t
*indices
)
1277 unsigned int ret
= 0;
1278 struct radix_tree_iter iter
;
1284 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1285 struct page
*head
, *page
;
1287 page
= radix_tree_deref_slot(slot
);
1288 if (unlikely(!page
))
1290 if (radix_tree_exception(page
)) {
1291 if (radix_tree_deref_retry(page
)) {
1292 slot
= radix_tree_iter_retry(&iter
);
1296 * A shadow entry of a recently evicted page, a swap
1297 * entry from shmem/tmpfs or a DAX entry. Return it
1298 * without attempting to raise page count.
1303 head
= compound_head(page
);
1304 if (!page_cache_get_speculative(head
))
1307 /* The page was split under us? */
1308 if (compound_head(page
) != head
) {
1313 /* Has the page moved? */
1314 if (unlikely(page
!= *slot
)) {
1319 indices
[ret
] = iter
.index
;
1320 entries
[ret
] = page
;
1321 if (++ret
== nr_entries
)
1329 * find_get_pages - gang pagecache lookup
1330 * @mapping: The address_space to search
1331 * @start: The starting page index
1332 * @nr_pages: The maximum number of pages
1333 * @pages: Where the resulting pages are placed
1335 * find_get_pages() will search for and return a group of up to
1336 * @nr_pages pages in the mapping. The pages are placed at @pages.
1337 * find_get_pages() takes a reference against the returned pages.
1339 * The search returns a group of mapping-contiguous pages with ascending
1340 * indexes. There may be holes in the indices due to not-present pages.
1342 * find_get_pages() returns the number of pages which were found.
1344 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1345 unsigned int nr_pages
, struct page
**pages
)
1347 struct radix_tree_iter iter
;
1351 if (unlikely(!nr_pages
))
1355 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1356 struct page
*head
, *page
;
1358 page
= radix_tree_deref_slot(slot
);
1359 if (unlikely(!page
))
1362 if (radix_tree_exception(page
)) {
1363 if (radix_tree_deref_retry(page
)) {
1364 slot
= radix_tree_iter_retry(&iter
);
1368 * A shadow entry of a recently evicted page,
1369 * or a swap entry from shmem/tmpfs. Skip
1375 head
= compound_head(page
);
1376 if (!page_cache_get_speculative(head
))
1379 /* The page was split under us? */
1380 if (compound_head(page
) != head
) {
1385 /* Has the page moved? */
1386 if (unlikely(page
!= *slot
)) {
1392 if (++ret
== nr_pages
)
1401 * find_get_pages_contig - gang contiguous pagecache lookup
1402 * @mapping: The address_space to search
1403 * @index: The starting page index
1404 * @nr_pages: The maximum number of pages
1405 * @pages: Where the resulting pages are placed
1407 * find_get_pages_contig() works exactly like find_get_pages(), except
1408 * that the returned number of pages are guaranteed to be contiguous.
1410 * find_get_pages_contig() returns the number of pages which were found.
1412 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1413 unsigned int nr_pages
, struct page
**pages
)
1415 struct radix_tree_iter iter
;
1417 unsigned int ret
= 0;
1419 if (unlikely(!nr_pages
))
1423 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1424 struct page
*head
, *page
;
1426 page
= radix_tree_deref_slot(slot
);
1427 /* The hole, there no reason to continue */
1428 if (unlikely(!page
))
1431 if (radix_tree_exception(page
)) {
1432 if (radix_tree_deref_retry(page
)) {
1433 slot
= radix_tree_iter_retry(&iter
);
1437 * A shadow entry of a recently evicted page,
1438 * or a swap entry from shmem/tmpfs. Stop
1439 * looking for contiguous pages.
1444 head
= compound_head(page
);
1445 if (!page_cache_get_speculative(head
))
1448 /* The page was split under us? */
1449 if (compound_head(page
) != head
) {
1454 /* Has the page moved? */
1455 if (unlikely(page
!= *slot
)) {
1461 * must check mapping and index after taking the ref.
1462 * otherwise we can get both false positives and false
1463 * negatives, which is just confusing to the caller.
1465 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1471 if (++ret
== nr_pages
)
1477 EXPORT_SYMBOL(find_get_pages_contig
);
1480 * find_get_pages_tag - find and return pages that match @tag
1481 * @mapping: the address_space to search
1482 * @index: the starting page index
1483 * @tag: the tag index
1484 * @nr_pages: the maximum number of pages
1485 * @pages: where the resulting pages are placed
1487 * Like find_get_pages, except we only return pages which are tagged with
1488 * @tag. We update @index to index the next page for the traversal.
1490 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1491 int tag
, unsigned int nr_pages
, struct page
**pages
)
1493 struct radix_tree_iter iter
;
1497 if (unlikely(!nr_pages
))
1501 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1502 &iter
, *index
, tag
) {
1503 struct page
*head
, *page
;
1505 page
= radix_tree_deref_slot(slot
);
1506 if (unlikely(!page
))
1509 if (radix_tree_exception(page
)) {
1510 if (radix_tree_deref_retry(page
)) {
1511 slot
= radix_tree_iter_retry(&iter
);
1515 * A shadow entry of a recently evicted page.
1517 * Those entries should never be tagged, but
1518 * this tree walk is lockless and the tags are
1519 * looked up in bulk, one radix tree node at a
1520 * time, so there is a sizable window for page
1521 * reclaim to evict a page we saw tagged.
1528 head
= compound_head(page
);
1529 if (!page_cache_get_speculative(head
))
1532 /* The page was split under us? */
1533 if (compound_head(page
) != head
) {
1538 /* Has the page moved? */
1539 if (unlikely(page
!= *slot
)) {
1545 if (++ret
== nr_pages
)
1552 *index
= pages
[ret
- 1]->index
+ 1;
1556 EXPORT_SYMBOL(find_get_pages_tag
);
1559 * find_get_entries_tag - find and return entries that match @tag
1560 * @mapping: the address_space to search
1561 * @start: the starting page cache index
1562 * @tag: the tag index
1563 * @nr_entries: the maximum number of entries
1564 * @entries: where the resulting entries are placed
1565 * @indices: the cache indices corresponding to the entries in @entries
1567 * Like find_get_entries, except we only return entries which are tagged with
1570 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1571 int tag
, unsigned int nr_entries
,
1572 struct page
**entries
, pgoff_t
*indices
)
1575 unsigned int ret
= 0;
1576 struct radix_tree_iter iter
;
1582 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1583 &iter
, start
, tag
) {
1584 struct page
*head
, *page
;
1586 page
= radix_tree_deref_slot(slot
);
1587 if (unlikely(!page
))
1589 if (radix_tree_exception(page
)) {
1590 if (radix_tree_deref_retry(page
)) {
1591 slot
= radix_tree_iter_retry(&iter
);
1596 * A shadow entry of a recently evicted page, a swap
1597 * entry from shmem/tmpfs or a DAX entry. Return it
1598 * without attempting to raise page count.
1603 head
= compound_head(page
);
1604 if (!page_cache_get_speculative(head
))
1607 /* The page was split under us? */
1608 if (compound_head(page
) != head
) {
1613 /* Has the page moved? */
1614 if (unlikely(page
!= *slot
)) {
1619 indices
[ret
] = iter
.index
;
1620 entries
[ret
] = page
;
1621 if (++ret
== nr_entries
)
1627 EXPORT_SYMBOL(find_get_entries_tag
);
1630 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1631 * a _large_ part of the i/o request. Imagine the worst scenario:
1633 * ---R__________________________________________B__________
1634 * ^ reading here ^ bad block(assume 4k)
1636 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1637 * => failing the whole request => read(R) => read(R+1) =>
1638 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1639 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1640 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1642 * It is going insane. Fix it by quickly scaling down the readahead size.
1644 static void shrink_readahead_size_eio(struct file
*filp
,
1645 struct file_ra_state
*ra
)
1651 * do_generic_file_read - generic file read routine
1652 * @filp: the file to read
1653 * @ppos: current file position
1654 * @iter: data destination
1655 * @written: already copied
1657 * This is a generic file read routine, and uses the
1658 * mapping->a_ops->readpage() function for the actual low-level stuff.
1660 * This is really ugly. But the goto's actually try to clarify some
1661 * of the logic when it comes to error handling etc.
1663 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1664 struct iov_iter
*iter
, ssize_t written
)
1666 struct address_space
*mapping
= filp
->f_mapping
;
1667 struct inode
*inode
= mapping
->host
;
1668 struct file_ra_state
*ra
= &filp
->f_ra
;
1672 unsigned long offset
; /* offset into pagecache page */
1673 unsigned int prev_offset
;
1676 index
= *ppos
>> PAGE_SHIFT
;
1677 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1678 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1679 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1680 offset
= *ppos
& ~PAGE_MASK
;
1686 unsigned long nr
, ret
;
1690 page
= find_get_page(mapping
, index
);
1692 page_cache_sync_readahead(mapping
,
1694 index
, last_index
- index
);
1695 page
= find_get_page(mapping
, index
);
1696 if (unlikely(page
== NULL
))
1697 goto no_cached_page
;
1699 if (PageReadahead(page
)) {
1700 page_cache_async_readahead(mapping
,
1702 index
, last_index
- index
);
1704 if (!PageUptodate(page
)) {
1706 * See comment in do_read_cache_page on why
1707 * wait_on_page_locked is used to avoid unnecessarily
1708 * serialisations and why it's safe.
1710 wait_on_page_locked_killable(page
);
1711 if (PageUptodate(page
))
1714 if (inode
->i_blkbits
== PAGE_SHIFT
||
1715 !mapping
->a_ops
->is_partially_uptodate
)
1716 goto page_not_up_to_date
;
1717 if (!trylock_page(page
))
1718 goto page_not_up_to_date
;
1719 /* Did it get truncated before we got the lock? */
1721 goto page_not_up_to_date_locked
;
1722 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1723 offset
, iter
->count
))
1724 goto page_not_up_to_date_locked
;
1729 * i_size must be checked after we know the page is Uptodate.
1731 * Checking i_size after the check allows us to calculate
1732 * the correct value for "nr", which means the zero-filled
1733 * part of the page is not copied back to userspace (unless
1734 * another truncate extends the file - this is desired though).
1737 isize
= i_size_read(inode
);
1738 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1739 if (unlikely(!isize
|| index
> end_index
)) {
1744 /* nr is the maximum number of bytes to copy from this page */
1746 if (index
== end_index
) {
1747 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1755 /* If users can be writing to this page using arbitrary
1756 * virtual addresses, take care about potential aliasing
1757 * before reading the page on the kernel side.
1759 if (mapping_writably_mapped(mapping
))
1760 flush_dcache_page(page
);
1763 * When a sequential read accesses a page several times,
1764 * only mark it as accessed the first time.
1766 if (prev_index
!= index
|| offset
!= prev_offset
)
1767 mark_page_accessed(page
);
1771 * Ok, we have the page, and it's up-to-date, so
1772 * now we can copy it to user space...
1775 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1777 index
+= offset
>> PAGE_SHIFT
;
1778 offset
&= ~PAGE_MASK
;
1779 prev_offset
= offset
;
1783 if (!iov_iter_count(iter
))
1791 page_not_up_to_date
:
1792 /* Get exclusive access to the page ... */
1793 error
= lock_page_killable(page
);
1794 if (unlikely(error
))
1795 goto readpage_error
;
1797 page_not_up_to_date_locked
:
1798 /* Did it get truncated before we got the lock? */
1799 if (!page
->mapping
) {
1805 /* Did somebody else fill it already? */
1806 if (PageUptodate(page
)) {
1813 * A previous I/O error may have been due to temporary
1814 * failures, eg. multipath errors.
1815 * PG_error will be set again if readpage fails.
1817 ClearPageError(page
);
1818 /* Start the actual read. The read will unlock the page. */
1819 error
= mapping
->a_ops
->readpage(filp
, page
);
1821 if (unlikely(error
)) {
1822 if (error
== AOP_TRUNCATED_PAGE
) {
1827 goto readpage_error
;
1830 if (!PageUptodate(page
)) {
1831 error
= lock_page_killable(page
);
1832 if (unlikely(error
))
1833 goto readpage_error
;
1834 if (!PageUptodate(page
)) {
1835 if (page
->mapping
== NULL
) {
1837 * invalidate_mapping_pages got it
1844 shrink_readahead_size_eio(filp
, ra
);
1846 goto readpage_error
;
1854 /* UHHUH! A synchronous read error occurred. Report it */
1860 * Ok, it wasn't cached, so we need to create a new
1863 page
= page_cache_alloc_cold(mapping
);
1868 error
= add_to_page_cache_lru(page
, mapping
, index
,
1869 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1872 if (error
== -EEXIST
) {
1882 ra
->prev_pos
= prev_index
;
1883 ra
->prev_pos
<<= PAGE_SHIFT
;
1884 ra
->prev_pos
|= prev_offset
;
1886 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
1887 file_accessed(filp
);
1888 return written
? written
: error
;
1892 * generic_file_read_iter - generic filesystem read routine
1893 * @iocb: kernel I/O control block
1894 * @iter: destination for the data read
1896 * This is the "read_iter()" routine for all filesystems
1897 * that can use the page cache directly.
1900 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1902 struct file
*file
= iocb
->ki_filp
;
1904 size_t count
= iov_iter_count(iter
);
1907 goto out
; /* skip atime */
1909 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1910 struct address_space
*mapping
= file
->f_mapping
;
1911 struct inode
*inode
= mapping
->host
;
1914 size
= i_size_read(inode
);
1915 retval
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
1916 iocb
->ki_pos
+ count
- 1);
1918 struct iov_iter data
= *iter
;
1919 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
);
1923 iocb
->ki_pos
+= retval
;
1924 iov_iter_advance(iter
, retval
);
1928 * Btrfs can have a short DIO read if we encounter
1929 * compressed extents, so if there was an error, or if
1930 * we've already read everything we wanted to, or if
1931 * there was a short read because we hit EOF, go ahead
1932 * and return. Otherwise fallthrough to buffered io for
1933 * the rest of the read. Buffered reads will not work for
1934 * DAX files, so don't bother trying.
1936 if (retval
< 0 || !iov_iter_count(iter
) || iocb
->ki_pos
>= size
||
1938 file_accessed(file
);
1943 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
1947 EXPORT_SYMBOL(generic_file_read_iter
);
1951 * page_cache_read - adds requested page to the page cache if not already there
1952 * @file: file to read
1953 * @offset: page index
1954 * @gfp_mask: memory allocation flags
1956 * This adds the requested page to the page cache if it isn't already there,
1957 * and schedules an I/O to read in its contents from disk.
1959 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1961 struct address_space
*mapping
= file
->f_mapping
;
1966 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1970 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1972 ret
= mapping
->a_ops
->readpage(file
, page
);
1973 else if (ret
== -EEXIST
)
1974 ret
= 0; /* losing race to add is OK */
1978 } while (ret
== AOP_TRUNCATED_PAGE
);
1983 #define MMAP_LOTSAMISS (100)
1986 * Synchronous readahead happens when we don't even find
1987 * a page in the page cache at all.
1989 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1990 struct file_ra_state
*ra
,
1994 struct address_space
*mapping
= file
->f_mapping
;
1996 /* If we don't want any read-ahead, don't bother */
1997 if (vma
->vm_flags
& VM_RAND_READ
)
2002 if (vma
->vm_flags
& VM_SEQ_READ
) {
2003 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2008 /* Avoid banging the cache line if not needed */
2009 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2013 * Do we miss much more than hit in this file? If so,
2014 * stop bothering with read-ahead. It will only hurt.
2016 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2022 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2023 ra
->size
= ra
->ra_pages
;
2024 ra
->async_size
= ra
->ra_pages
/ 4;
2025 ra_submit(ra
, mapping
, file
);
2029 * Asynchronous readahead happens when we find the page and PG_readahead,
2030 * so we want to possibly extend the readahead further..
2032 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2033 struct file_ra_state
*ra
,
2038 struct address_space
*mapping
= file
->f_mapping
;
2040 /* If we don't want any read-ahead, don't bother */
2041 if (vma
->vm_flags
& VM_RAND_READ
)
2043 if (ra
->mmap_miss
> 0)
2045 if (PageReadahead(page
))
2046 page_cache_async_readahead(mapping
, ra
, file
,
2047 page
, offset
, ra
->ra_pages
);
2051 * filemap_fault - read in file data for page fault handling
2052 * @vma: vma in which the fault was taken
2053 * @vmf: struct vm_fault containing details of the fault
2055 * filemap_fault() is invoked via the vma operations vector for a
2056 * mapped memory region to read in file data during a page fault.
2058 * The goto's are kind of ugly, but this streamlines the normal case of having
2059 * it in the page cache, and handles the special cases reasonably without
2060 * having a lot of duplicated code.
2062 * vma->vm_mm->mmap_sem must be held on entry.
2064 * If our return value has VM_FAULT_RETRY set, it's because
2065 * lock_page_or_retry() returned 0.
2066 * The mmap_sem has usually been released in this case.
2067 * See __lock_page_or_retry() for the exception.
2069 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2070 * has not been released.
2072 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2074 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2077 struct file
*file
= vma
->vm_file
;
2078 struct address_space
*mapping
= file
->f_mapping
;
2079 struct file_ra_state
*ra
= &file
->f_ra
;
2080 struct inode
*inode
= mapping
->host
;
2081 pgoff_t offset
= vmf
->pgoff
;
2086 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2087 if (offset
>= size
>> PAGE_SHIFT
)
2088 return VM_FAULT_SIGBUS
;
2091 * Do we have something in the page cache already?
2093 page
= find_get_page(mapping
, offset
);
2094 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2096 * We found the page, so try async readahead before
2097 * waiting for the lock.
2099 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2101 /* No page in the page cache at all */
2102 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2103 count_vm_event(PGMAJFAULT
);
2104 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2105 ret
= VM_FAULT_MAJOR
;
2107 page
= find_get_page(mapping
, offset
);
2109 goto no_cached_page
;
2112 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2114 return ret
| VM_FAULT_RETRY
;
2117 /* Did it get truncated? */
2118 if (unlikely(page
->mapping
!= mapping
)) {
2123 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2126 * We have a locked page in the page cache, now we need to check
2127 * that it's up-to-date. If not, it is going to be due to an error.
2129 if (unlikely(!PageUptodate(page
)))
2130 goto page_not_uptodate
;
2133 * Found the page and have a reference on it.
2134 * We must recheck i_size under page lock.
2136 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2137 if (unlikely(offset
>= size
>> PAGE_SHIFT
)) {
2140 return VM_FAULT_SIGBUS
;
2144 return ret
| VM_FAULT_LOCKED
;
2148 * We're only likely to ever get here if MADV_RANDOM is in
2151 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2154 * The page we want has now been added to the page cache.
2155 * In the unlikely event that someone removed it in the
2156 * meantime, we'll just come back here and read it again.
2162 * An error return from page_cache_read can result if the
2163 * system is low on memory, or a problem occurs while trying
2166 if (error
== -ENOMEM
)
2167 return VM_FAULT_OOM
;
2168 return VM_FAULT_SIGBUS
;
2172 * Umm, take care of errors if the page isn't up-to-date.
2173 * Try to re-read it _once_. We do this synchronously,
2174 * because there really aren't any performance issues here
2175 * and we need to check for errors.
2177 ClearPageError(page
);
2178 error
= mapping
->a_ops
->readpage(file
, page
);
2180 wait_on_page_locked(page
);
2181 if (!PageUptodate(page
))
2186 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2189 /* Things didn't work out. Return zero to tell the mm layer so. */
2190 shrink_readahead_size_eio(file
, ra
);
2191 return VM_FAULT_SIGBUS
;
2193 EXPORT_SYMBOL(filemap_fault
);
2195 void filemap_map_pages(struct fault_env
*fe
,
2196 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2198 struct radix_tree_iter iter
;
2200 struct file
*file
= fe
->vma
->vm_file
;
2201 struct address_space
*mapping
= file
->f_mapping
;
2202 pgoff_t last_pgoff
= start_pgoff
;
2204 struct page
*head
, *page
;
2207 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2209 if (iter
.index
> end_pgoff
)
2212 page
= radix_tree_deref_slot(slot
);
2213 if (unlikely(!page
))
2215 if (radix_tree_exception(page
)) {
2216 if (radix_tree_deref_retry(page
)) {
2217 slot
= radix_tree_iter_retry(&iter
);
2223 head
= compound_head(page
);
2224 if (!page_cache_get_speculative(head
))
2227 /* The page was split under us? */
2228 if (compound_head(page
) != head
) {
2233 /* Has the page moved? */
2234 if (unlikely(page
!= *slot
)) {
2239 if (!PageUptodate(page
) ||
2240 PageReadahead(page
) ||
2243 if (!trylock_page(page
))
2246 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2249 size
= round_up(i_size_read(mapping
->host
), PAGE_SIZE
);
2250 if (page
->index
>= size
>> PAGE_SHIFT
)
2253 if (file
->f_ra
.mmap_miss
> 0)
2254 file
->f_ra
.mmap_miss
--;
2256 fe
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2258 fe
->pte
+= iter
.index
- last_pgoff
;
2259 last_pgoff
= iter
.index
;
2260 if (alloc_set_pte(fe
, NULL
, page
))
2269 /* Huge page is mapped? No need to proceed. */
2270 if (pmd_trans_huge(*fe
->pmd
))
2272 if (iter
.index
== end_pgoff
)
2277 EXPORT_SYMBOL(filemap_map_pages
);
2279 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2281 struct page
*page
= vmf
->page
;
2282 struct inode
*inode
= file_inode(vma
->vm_file
);
2283 int ret
= VM_FAULT_LOCKED
;
2285 sb_start_pagefault(inode
->i_sb
);
2286 file_update_time(vma
->vm_file
);
2288 if (page
->mapping
!= inode
->i_mapping
) {
2290 ret
= VM_FAULT_NOPAGE
;
2294 * We mark the page dirty already here so that when freeze is in
2295 * progress, we are guaranteed that writeback during freezing will
2296 * see the dirty page and writeprotect it again.
2298 set_page_dirty(page
);
2299 wait_for_stable_page(page
);
2301 sb_end_pagefault(inode
->i_sb
);
2304 EXPORT_SYMBOL(filemap_page_mkwrite
);
2306 const struct vm_operations_struct generic_file_vm_ops
= {
2307 .fault
= filemap_fault
,
2308 .map_pages
= filemap_map_pages
,
2309 .page_mkwrite
= filemap_page_mkwrite
,
2312 /* This is used for a general mmap of a disk file */
2314 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2316 struct address_space
*mapping
= file
->f_mapping
;
2318 if (!mapping
->a_ops
->readpage
)
2320 file_accessed(file
);
2321 vma
->vm_ops
= &generic_file_vm_ops
;
2326 * This is for filesystems which do not implement ->writepage.
2328 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2330 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2332 return generic_file_mmap(file
, vma
);
2335 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2339 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2343 #endif /* CONFIG_MMU */
2345 EXPORT_SYMBOL(generic_file_mmap
);
2346 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2348 static struct page
*wait_on_page_read(struct page
*page
)
2350 if (!IS_ERR(page
)) {
2351 wait_on_page_locked(page
);
2352 if (!PageUptodate(page
)) {
2354 page
= ERR_PTR(-EIO
);
2360 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2362 int (*filler
)(void *, struct page
*),
2369 page
= find_get_page(mapping
, index
);
2371 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2373 return ERR_PTR(-ENOMEM
);
2374 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2375 if (unlikely(err
)) {
2379 /* Presumably ENOMEM for radix tree node */
2380 return ERR_PTR(err
);
2384 err
= filler(data
, page
);
2387 return ERR_PTR(err
);
2390 page
= wait_on_page_read(page
);
2395 if (PageUptodate(page
))
2399 * Page is not up to date and may be locked due one of the following
2400 * case a: Page is being filled and the page lock is held
2401 * case b: Read/write error clearing the page uptodate status
2402 * case c: Truncation in progress (page locked)
2403 * case d: Reclaim in progress
2405 * Case a, the page will be up to date when the page is unlocked.
2406 * There is no need to serialise on the page lock here as the page
2407 * is pinned so the lock gives no additional protection. Even if the
2408 * the page is truncated, the data is still valid if PageUptodate as
2409 * it's a race vs truncate race.
2410 * Case b, the page will not be up to date
2411 * Case c, the page may be truncated but in itself, the data may still
2412 * be valid after IO completes as it's a read vs truncate race. The
2413 * operation must restart if the page is not uptodate on unlock but
2414 * otherwise serialising on page lock to stabilise the mapping gives
2415 * no additional guarantees to the caller as the page lock is
2416 * released before return.
2417 * Case d, similar to truncation. If reclaim holds the page lock, it
2418 * will be a race with remove_mapping that determines if the mapping
2419 * is valid on unlock but otherwise the data is valid and there is
2420 * no need to serialise with page lock.
2422 * As the page lock gives no additional guarantee, we optimistically
2423 * wait on the page to be unlocked and check if it's up to date and
2424 * use the page if it is. Otherwise, the page lock is required to
2425 * distinguish between the different cases. The motivation is that we
2426 * avoid spurious serialisations and wakeups when multiple processes
2427 * wait on the same page for IO to complete.
2429 wait_on_page_locked(page
);
2430 if (PageUptodate(page
))
2433 /* Distinguish between all the cases under the safety of the lock */
2436 /* Case c or d, restart the operation */
2437 if (!page
->mapping
) {
2443 /* Someone else locked and filled the page in a very small window */
2444 if (PageUptodate(page
)) {
2451 mark_page_accessed(page
);
2456 * read_cache_page - read into page cache, fill it if needed
2457 * @mapping: the page's address_space
2458 * @index: the page index
2459 * @filler: function to perform the read
2460 * @data: first arg to filler(data, page) function, often left as NULL
2462 * Read into the page cache. If a page already exists, and PageUptodate() is
2463 * not set, try to fill the page and wait for it to become unlocked.
2465 * If the page does not get brought uptodate, return -EIO.
2467 struct page
*read_cache_page(struct address_space
*mapping
,
2469 int (*filler
)(void *, struct page
*),
2472 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2474 EXPORT_SYMBOL(read_cache_page
);
2477 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2478 * @mapping: the page's address_space
2479 * @index: the page index
2480 * @gfp: the page allocator flags to use if allocating
2482 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2483 * any new page allocations done using the specified allocation flags.
2485 * If the page does not get brought uptodate, return -EIO.
2487 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2491 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2493 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2495 EXPORT_SYMBOL(read_cache_page_gfp
);
2498 * Performs necessary checks before doing a write
2500 * Can adjust writing position or amount of bytes to write.
2501 * Returns appropriate error code that caller should return or
2502 * zero in case that write should be allowed.
2504 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2506 struct file
*file
= iocb
->ki_filp
;
2507 struct inode
*inode
= file
->f_mapping
->host
;
2508 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2511 if (!iov_iter_count(from
))
2514 /* FIXME: this is for backwards compatibility with 2.4 */
2515 if (iocb
->ki_flags
& IOCB_APPEND
)
2516 iocb
->ki_pos
= i_size_read(inode
);
2520 if (limit
!= RLIM_INFINITY
) {
2521 if (iocb
->ki_pos
>= limit
) {
2522 send_sig(SIGXFSZ
, current
, 0);
2525 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2531 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2532 !(file
->f_flags
& O_LARGEFILE
))) {
2533 if (pos
>= MAX_NON_LFS
)
2535 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2539 * Are we about to exceed the fs block limit ?
2541 * If we have written data it becomes a short write. If we have
2542 * exceeded without writing data we send a signal and return EFBIG.
2543 * Linus frestrict idea will clean these up nicely..
2545 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2548 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2549 return iov_iter_count(from
);
2551 EXPORT_SYMBOL(generic_write_checks
);
2553 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2554 loff_t pos
, unsigned len
, unsigned flags
,
2555 struct page
**pagep
, void **fsdata
)
2557 const struct address_space_operations
*aops
= mapping
->a_ops
;
2559 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2562 EXPORT_SYMBOL(pagecache_write_begin
);
2564 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2565 loff_t pos
, unsigned len
, unsigned copied
,
2566 struct page
*page
, void *fsdata
)
2568 const struct address_space_operations
*aops
= mapping
->a_ops
;
2570 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2572 EXPORT_SYMBOL(pagecache_write_end
);
2575 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2577 struct file
*file
= iocb
->ki_filp
;
2578 struct address_space
*mapping
= file
->f_mapping
;
2579 struct inode
*inode
= mapping
->host
;
2580 loff_t pos
= iocb
->ki_pos
;
2584 struct iov_iter data
;
2586 write_len
= iov_iter_count(from
);
2587 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2589 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2594 * After a write we want buffered reads to be sure to go to disk to get
2595 * the new data. We invalidate clean cached page from the region we're
2596 * about to write. We do this *before* the write so that we can return
2597 * without clobbering -EIOCBQUEUED from ->direct_IO().
2599 if (mapping
->nrpages
) {
2600 written
= invalidate_inode_pages2_range(mapping
,
2601 pos
>> PAGE_SHIFT
, end
);
2603 * If a page can not be invalidated, return 0 to fall back
2604 * to buffered write.
2607 if (written
== -EBUSY
)
2614 written
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2617 * Finally, try again to invalidate clean pages which might have been
2618 * cached by non-direct readahead, or faulted in by get_user_pages()
2619 * if the source of the write was an mmap'ed region of the file
2620 * we're writing. Either one is a pretty crazy thing to do,
2621 * so we don't support it 100%. If this invalidation
2622 * fails, tough, the write still worked...
2624 if (mapping
->nrpages
) {
2625 invalidate_inode_pages2_range(mapping
,
2626 pos
>> PAGE_SHIFT
, end
);
2631 iov_iter_advance(from
, written
);
2632 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2633 i_size_write(inode
, pos
);
2634 mark_inode_dirty(inode
);
2641 EXPORT_SYMBOL(generic_file_direct_write
);
2644 * Find or create a page at the given pagecache position. Return the locked
2645 * page. This function is specifically for buffered writes.
2647 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2648 pgoff_t index
, unsigned flags
)
2651 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2653 if (flags
& AOP_FLAG_NOFS
)
2654 fgp_flags
|= FGP_NOFS
;
2656 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2657 mapping_gfp_mask(mapping
));
2659 wait_for_stable_page(page
);
2663 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2665 ssize_t
generic_perform_write(struct file
*file
,
2666 struct iov_iter
*i
, loff_t pos
)
2668 struct address_space
*mapping
= file
->f_mapping
;
2669 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2671 ssize_t written
= 0;
2672 unsigned int flags
= 0;
2675 * Copies from kernel address space cannot fail (NFSD is a big user).
2677 if (!iter_is_iovec(i
))
2678 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2682 unsigned long offset
; /* Offset into pagecache page */
2683 unsigned long bytes
; /* Bytes to write to page */
2684 size_t copied
; /* Bytes copied from user */
2687 offset
= (pos
& (PAGE_SIZE
- 1));
2688 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2693 * Bring in the user page that we will copy from _first_.
2694 * Otherwise there's a nasty deadlock on copying from the
2695 * same page as we're writing to, without it being marked
2698 * Not only is this an optimisation, but it is also required
2699 * to check that the address is actually valid, when atomic
2700 * usercopies are used, below.
2702 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2707 if (fatal_signal_pending(current
)) {
2712 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2714 if (unlikely(status
< 0))
2717 if (mapping_writably_mapped(mapping
))
2718 flush_dcache_page(page
);
2720 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2721 flush_dcache_page(page
);
2723 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2725 if (unlikely(status
< 0))
2731 iov_iter_advance(i
, copied
);
2732 if (unlikely(copied
== 0)) {
2734 * If we were unable to copy any data at all, we must
2735 * fall back to a single segment length write.
2737 * If we didn't fallback here, we could livelock
2738 * because not all segments in the iov can be copied at
2739 * once without a pagefault.
2741 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2742 iov_iter_single_seg_count(i
));
2748 balance_dirty_pages_ratelimited(mapping
);
2749 } while (iov_iter_count(i
));
2751 return written
? written
: status
;
2753 EXPORT_SYMBOL(generic_perform_write
);
2756 * __generic_file_write_iter - write data to a file
2757 * @iocb: IO state structure (file, offset, etc.)
2758 * @from: iov_iter with data to write
2760 * This function does all the work needed for actually writing data to a
2761 * file. It does all basic checks, removes SUID from the file, updates
2762 * modification times and calls proper subroutines depending on whether we
2763 * do direct IO or a standard buffered write.
2765 * It expects i_mutex to be grabbed unless we work on a block device or similar
2766 * object which does not need locking at all.
2768 * This function does *not* take care of syncing data in case of O_SYNC write.
2769 * A caller has to handle it. This is mainly due to the fact that we want to
2770 * avoid syncing under i_mutex.
2772 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2774 struct file
*file
= iocb
->ki_filp
;
2775 struct address_space
* mapping
= file
->f_mapping
;
2776 struct inode
*inode
= mapping
->host
;
2777 ssize_t written
= 0;
2781 /* We can write back this queue in page reclaim */
2782 current
->backing_dev_info
= inode_to_bdi(inode
);
2783 err
= file_remove_privs(file
);
2787 err
= file_update_time(file
);
2791 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2792 loff_t pos
, endbyte
;
2794 written
= generic_file_direct_write(iocb
, from
);
2796 * If the write stopped short of completing, fall back to
2797 * buffered writes. Some filesystems do this for writes to
2798 * holes, for example. For DAX files, a buffered write will
2799 * not succeed (even if it did, DAX does not handle dirty
2800 * page-cache pages correctly).
2802 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2805 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2807 * If generic_perform_write() returned a synchronous error
2808 * then we want to return the number of bytes which were
2809 * direct-written, or the error code if that was zero. Note
2810 * that this differs from normal direct-io semantics, which
2811 * will return -EFOO even if some bytes were written.
2813 if (unlikely(status
< 0)) {
2818 * We need to ensure that the page cache pages are written to
2819 * disk and invalidated to preserve the expected O_DIRECT
2822 endbyte
= pos
+ status
- 1;
2823 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2825 iocb
->ki_pos
= endbyte
+ 1;
2827 invalidate_mapping_pages(mapping
,
2829 endbyte
>> PAGE_SHIFT
);
2832 * We don't know how much we wrote, so just return
2833 * the number of bytes which were direct-written
2837 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2838 if (likely(written
> 0))
2839 iocb
->ki_pos
+= written
;
2842 current
->backing_dev_info
= NULL
;
2843 return written
? written
: err
;
2845 EXPORT_SYMBOL(__generic_file_write_iter
);
2848 * generic_file_write_iter - write data to a file
2849 * @iocb: IO state structure
2850 * @from: iov_iter with data to write
2852 * This is a wrapper around __generic_file_write_iter() to be used by most
2853 * filesystems. It takes care of syncing the file in case of O_SYNC file
2854 * and acquires i_mutex as needed.
2856 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2858 struct file
*file
= iocb
->ki_filp
;
2859 struct inode
*inode
= file
->f_mapping
->host
;
2863 ret
= generic_write_checks(iocb
, from
);
2865 ret
= __generic_file_write_iter(iocb
, from
);
2866 inode_unlock(inode
);
2869 ret
= generic_write_sync(iocb
, ret
);
2872 EXPORT_SYMBOL(generic_file_write_iter
);
2875 * try_to_release_page() - release old fs-specific metadata on a page
2877 * @page: the page which the kernel is trying to free
2878 * @gfp_mask: memory allocation flags (and I/O mode)
2880 * The address_space is to try to release any data against the page
2881 * (presumably at page->private). If the release was successful, return `1'.
2882 * Otherwise return zero.
2884 * This may also be called if PG_fscache is set on a page, indicating that the
2885 * page is known to the local caching routines.
2887 * The @gfp_mask argument specifies whether I/O may be performed to release
2888 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2891 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2893 struct address_space
* const mapping
= page
->mapping
;
2895 BUG_ON(!PageLocked(page
));
2896 if (PageWriteback(page
))
2899 if (mapping
&& mapping
->a_ops
->releasepage
)
2900 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2901 return try_to_free_buffers(page
);
2904 EXPORT_SYMBOL(try_to_release_page
);