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
;
122 VM_BUG_ON(!PageLocked(page
));
124 __radix_tree_lookup(&mapping
->page_tree
, page
->index
, &node
, &slot
);
127 mapping
->nrexceptional
++;
129 * Make sure the nrexceptional update is committed before
130 * the nrpages update so that final truncate racing
131 * with reclaim does not see both counters 0 at the
132 * same time and miss a shadow entry.
139 /* Clear direct pointer tags in root node */
140 mapping
->page_tree
.gfp_mask
&= __GFP_BITS_MASK
;
141 radix_tree_replace_slot(slot
, shadow
);
145 /* Clear tree tags for the removed page */
147 offset
= index
& RADIX_TREE_MAP_MASK
;
148 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
149 if (test_bit(offset
, node
->tags
[tag
]))
150 radix_tree_tag_clear(&mapping
->page_tree
, index
, tag
);
153 /* Delete page, swap shadow entry */
154 radix_tree_replace_slot(slot
, shadow
);
155 workingset_node_pages_dec(node
);
157 workingset_node_shadows_inc(node
);
159 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
163 * Track node that only contains shadow entries.
165 * Avoid acquiring the list_lru lock if already tracked. The
166 * list_empty() test is safe as node->private_list is
167 * protected by mapping->tree_lock.
169 if (!workingset_node_pages(node
) &&
170 list_empty(&node
->private_list
)) {
171 node
->private_data
= mapping
;
172 list_lru_add(&workingset_shadow_nodes
, &node
->private_list
);
177 * Delete a page from the page cache and free it. Caller has to make
178 * sure the page is locked and that nobody else uses it - or that usage
179 * is safe. The caller must hold the mapping's tree_lock.
181 void __delete_from_page_cache(struct page
*page
, void *shadow
)
183 struct address_space
*mapping
= page
->mapping
;
185 trace_mm_filemap_delete_from_page_cache(page
);
187 * if we're uptodate, flush out into the cleancache, otherwise
188 * invalidate any existing cleancache entries. We can't leave
189 * stale data around in the cleancache once our page is gone
191 if (PageUptodate(page
) && PageMappedToDisk(page
))
192 cleancache_put_page(page
);
194 cleancache_invalidate_page(mapping
, page
);
196 VM_BUG_ON_PAGE(page_mapped(page
), page
);
197 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
200 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
201 current
->comm
, page_to_pfn(page
));
202 dump_page(page
, "still mapped when deleted");
204 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
206 mapcount
= page_mapcount(page
);
207 if (mapping_exiting(mapping
) &&
208 page_count(page
) >= mapcount
+ 2) {
210 * All vmas have already been torn down, so it's
211 * a good bet that actually the page is unmapped,
212 * and we'd prefer not to leak it: if we're wrong,
213 * some other bad page check should catch it later.
215 page_mapcount_reset(page
);
216 page_ref_sub(page
, mapcount
);
220 page_cache_tree_delete(mapping
, page
, shadow
);
222 page
->mapping
= NULL
;
223 /* Leave page->index set: truncation lookup relies upon it */
225 /* hugetlb pages do not participate in page cache accounting. */
227 __dec_zone_page_state(page
, NR_FILE_PAGES
);
228 if (PageSwapBacked(page
))
229 __dec_zone_page_state(page
, NR_SHMEM
);
232 * At this point page must be either written or cleaned by truncate.
233 * Dirty page here signals a bug and loss of unwritten data.
235 * This fixes dirty accounting after removing the page entirely but
236 * leaves PageDirty set: it has no effect for truncated page and
237 * anyway will be cleared before returning page into buddy allocator.
239 if (WARN_ON_ONCE(PageDirty(page
)))
240 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
244 * delete_from_page_cache - delete page from page cache
245 * @page: the page which the kernel is trying to remove from page cache
247 * This must be called only on pages that have been verified to be in the page
248 * cache and locked. It will never put the page into the free list, the caller
249 * has a reference on the page.
251 void delete_from_page_cache(struct page
*page
)
253 struct address_space
*mapping
= page
->mapping
;
256 void (*freepage
)(struct page
*);
258 BUG_ON(!PageLocked(page
));
260 freepage
= mapping
->a_ops
->freepage
;
262 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
263 __delete_from_page_cache(page
, NULL
);
264 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
270 EXPORT_SYMBOL(delete_from_page_cache
);
272 static int filemap_check_errors(struct address_space
*mapping
)
275 /* Check for outstanding write errors */
276 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
277 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
279 if (test_bit(AS_EIO
, &mapping
->flags
) &&
280 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
286 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
287 * @mapping: address space structure to write
288 * @start: offset in bytes where the range starts
289 * @end: offset in bytes where the range ends (inclusive)
290 * @sync_mode: enable synchronous operation
292 * Start writeback against all of a mapping's dirty pages that lie
293 * within the byte offsets <start, end> inclusive.
295 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
296 * opposed to a regular memory cleansing writeback. The difference between
297 * these two operations is that if a dirty page/buffer is encountered, it must
298 * be waited upon, and not just skipped over.
300 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
301 loff_t end
, int sync_mode
)
304 struct writeback_control wbc
= {
305 .sync_mode
= sync_mode
,
306 .nr_to_write
= LONG_MAX
,
307 .range_start
= start
,
311 if (!mapping_cap_writeback_dirty(mapping
))
314 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
315 ret
= do_writepages(mapping
, &wbc
);
316 wbc_detach_inode(&wbc
);
320 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
323 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
326 int filemap_fdatawrite(struct address_space
*mapping
)
328 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
330 EXPORT_SYMBOL(filemap_fdatawrite
);
332 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
335 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
337 EXPORT_SYMBOL(filemap_fdatawrite_range
);
340 * filemap_flush - mostly a non-blocking flush
341 * @mapping: target address_space
343 * This is a mostly non-blocking flush. Not suitable for data-integrity
344 * purposes - I/O may not be started against all dirty pages.
346 int filemap_flush(struct address_space
*mapping
)
348 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
350 EXPORT_SYMBOL(filemap_flush
);
352 static int __filemap_fdatawait_range(struct address_space
*mapping
,
353 loff_t start_byte
, loff_t end_byte
)
355 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
356 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
361 if (end_byte
< start_byte
)
364 pagevec_init(&pvec
, 0);
365 while ((index
<= end
) &&
366 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
367 PAGECACHE_TAG_WRITEBACK
,
368 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
371 for (i
= 0; i
< nr_pages
; i
++) {
372 struct page
*page
= pvec
.pages
[i
];
374 /* until radix tree lookup accepts end_index */
375 if (page
->index
> end
)
378 wait_on_page_writeback(page
);
379 if (TestClearPageError(page
))
382 pagevec_release(&pvec
);
390 * filemap_fdatawait_range - wait for writeback to complete
391 * @mapping: address space structure to wait for
392 * @start_byte: offset in bytes where the range starts
393 * @end_byte: offset in bytes where the range ends (inclusive)
395 * Walk the list of under-writeback pages of the given address space
396 * in the given range and wait for all of them. Check error status of
397 * the address space and return it.
399 * Since the error status of the address space is cleared by this function,
400 * callers are responsible for checking the return value and handling and/or
401 * reporting the error.
403 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
408 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
409 ret2
= filemap_check_errors(mapping
);
415 EXPORT_SYMBOL(filemap_fdatawait_range
);
418 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
419 * @mapping: address space structure to wait for
421 * Walk the list of under-writeback pages of the given address space
422 * and wait for all of them. Unlike filemap_fdatawait(), this function
423 * does not clear error status of the address space.
425 * Use this function if callers don't handle errors themselves. Expected
426 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
429 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
431 loff_t i_size
= i_size_read(mapping
->host
);
436 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
440 * filemap_fdatawait - wait for all under-writeback pages to complete
441 * @mapping: address space structure to wait for
443 * Walk the list of under-writeback pages of the given address space
444 * and wait for all of them. Check error status of the address space
447 * Since the error status of the address space is cleared by this function,
448 * callers are responsible for checking the return value and handling and/or
449 * reporting the error.
451 int filemap_fdatawait(struct address_space
*mapping
)
453 loff_t i_size
= i_size_read(mapping
->host
);
458 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
460 EXPORT_SYMBOL(filemap_fdatawait
);
462 int filemap_write_and_wait(struct address_space
*mapping
)
466 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
467 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
468 err
= filemap_fdatawrite(mapping
);
470 * Even if the above returned error, the pages may be
471 * written partially (e.g. -ENOSPC), so we wait for it.
472 * But the -EIO is special case, it may indicate the worst
473 * thing (e.g. bug) happened, so we avoid waiting for it.
476 int err2
= filemap_fdatawait(mapping
);
481 err
= filemap_check_errors(mapping
);
485 EXPORT_SYMBOL(filemap_write_and_wait
);
488 * filemap_write_and_wait_range - write out & wait on a file range
489 * @mapping: the address_space for the pages
490 * @lstart: offset in bytes where the range starts
491 * @lend: offset in bytes where the range ends (inclusive)
493 * Write out and wait upon file offsets lstart->lend, inclusive.
495 * Note that `lend' is inclusive (describes the last byte to be written) so
496 * that this function can be used to write to the very end-of-file (end = -1).
498 int filemap_write_and_wait_range(struct address_space
*mapping
,
499 loff_t lstart
, loff_t lend
)
503 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
504 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
505 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
507 /* See comment of filemap_write_and_wait() */
509 int err2
= filemap_fdatawait_range(mapping
,
515 err
= filemap_check_errors(mapping
);
519 EXPORT_SYMBOL(filemap_write_and_wait_range
);
522 * replace_page_cache_page - replace a pagecache page with a new one
523 * @old: page to be replaced
524 * @new: page to replace with
525 * @gfp_mask: allocation mode
527 * This function replaces a page in the pagecache with a new one. On
528 * success it acquires the pagecache reference for the new page and
529 * drops it for the old page. Both the old and new pages must be
530 * locked. This function does not add the new page to the LRU, the
531 * caller must do that.
533 * The remove + add is atomic. The only way this function can fail is
534 * memory allocation failure.
536 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
540 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
541 VM_BUG_ON_PAGE(!PageLocked(new), new);
542 VM_BUG_ON_PAGE(new->mapping
, new);
544 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
546 struct address_space
*mapping
= old
->mapping
;
547 void (*freepage
)(struct page
*);
550 pgoff_t offset
= old
->index
;
551 freepage
= mapping
->a_ops
->freepage
;
554 new->mapping
= mapping
;
557 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
558 __delete_from_page_cache(old
, NULL
);
559 error
= radix_tree_insert(&mapping
->page_tree
, offset
, new);
564 * hugetlb pages do not participate in page cache accounting.
567 __inc_zone_page_state(new, NR_FILE_PAGES
);
568 if (PageSwapBacked(new))
569 __inc_zone_page_state(new, NR_SHMEM
);
570 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
571 mem_cgroup_migrate(old
, new);
572 radix_tree_preload_end();
580 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
582 static int page_cache_tree_insert(struct address_space
*mapping
,
583 struct page
*page
, void **shadowp
)
585 struct radix_tree_node
*node
;
589 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
596 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
597 if (!radix_tree_exceptional_entry(p
))
600 if (WARN_ON(dax_mapping(mapping
)))
605 mapping
->nrexceptional
--;
607 workingset_node_shadows_dec(node
);
609 radix_tree_replace_slot(slot
, page
);
612 workingset_node_pages_inc(node
);
614 * Don't track node that contains actual pages.
616 * Avoid acquiring the list_lru lock if already
617 * untracked. The list_empty() test is safe as
618 * node->private_list is protected by
619 * mapping->tree_lock.
621 if (!list_empty(&node
->private_list
))
622 list_lru_del(&workingset_shadow_nodes
,
623 &node
->private_list
);
628 static int __add_to_page_cache_locked(struct page
*page
,
629 struct address_space
*mapping
,
630 pgoff_t offset
, gfp_t gfp_mask
,
633 int huge
= PageHuge(page
);
634 struct mem_cgroup
*memcg
;
637 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
638 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
641 error
= mem_cgroup_try_charge(page
, current
->mm
,
642 gfp_mask
, &memcg
, false);
647 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
650 mem_cgroup_cancel_charge(page
, memcg
, false);
655 page
->mapping
= mapping
;
656 page
->index
= offset
;
658 spin_lock_irq(&mapping
->tree_lock
);
659 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
660 radix_tree_preload_end();
664 /* hugetlb pages do not participate in page cache accounting. */
666 __inc_zone_page_state(page
, NR_FILE_PAGES
);
667 spin_unlock_irq(&mapping
->tree_lock
);
669 mem_cgroup_commit_charge(page
, memcg
, false, false);
670 trace_mm_filemap_add_to_page_cache(page
);
673 page
->mapping
= NULL
;
674 /* Leave page->index set: truncation relies upon it */
675 spin_unlock_irq(&mapping
->tree_lock
);
677 mem_cgroup_cancel_charge(page
, memcg
, false);
683 * add_to_page_cache_locked - add a locked page to the pagecache
685 * @mapping: the page's address_space
686 * @offset: page index
687 * @gfp_mask: page allocation mode
689 * This function is used to add a page to the pagecache. It must be locked.
690 * This function does not add the page to the LRU. The caller must do that.
692 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
693 pgoff_t offset
, gfp_t gfp_mask
)
695 return __add_to_page_cache_locked(page
, mapping
, offset
,
698 EXPORT_SYMBOL(add_to_page_cache_locked
);
700 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
701 pgoff_t offset
, gfp_t gfp_mask
)
706 __SetPageLocked(page
);
707 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
710 __ClearPageLocked(page
);
713 * The page might have been evicted from cache only
714 * recently, in which case it should be activated like
715 * any other repeatedly accessed page.
716 * The exception is pages getting rewritten; evicting other
717 * data from the working set, only to cache data that will
718 * get overwritten with something else, is a waste of memory.
720 if (!(gfp_mask
& __GFP_WRITE
) &&
721 shadow
&& workingset_refault(shadow
)) {
723 workingset_activation(page
);
725 ClearPageActive(page
);
730 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
733 struct page
*__page_cache_alloc(gfp_t gfp
)
738 if (cpuset_do_page_mem_spread()) {
739 unsigned int cpuset_mems_cookie
;
741 cpuset_mems_cookie
= read_mems_allowed_begin();
742 n
= cpuset_mem_spread_node();
743 page
= __alloc_pages_node(n
, gfp
, 0);
744 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
748 return alloc_pages(gfp
, 0);
750 EXPORT_SYMBOL(__page_cache_alloc
);
754 * In order to wait for pages to become available there must be
755 * waitqueues associated with pages. By using a hash table of
756 * waitqueues where the bucket discipline is to maintain all
757 * waiters on the same queue and wake all when any of the pages
758 * become available, and for the woken contexts to check to be
759 * sure the appropriate page became available, this saves space
760 * at a cost of "thundering herd" phenomena during rare hash
763 wait_queue_head_t
*page_waitqueue(struct page
*page
)
765 const struct zone
*zone
= page_zone(page
);
767 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
769 EXPORT_SYMBOL(page_waitqueue
);
771 void wait_on_page_bit(struct page
*page
, int bit_nr
)
773 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
775 if (test_bit(bit_nr
, &page
->flags
))
776 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
777 TASK_UNINTERRUPTIBLE
);
779 EXPORT_SYMBOL(wait_on_page_bit
);
781 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
783 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
785 if (!test_bit(bit_nr
, &page
->flags
))
788 return __wait_on_bit(page_waitqueue(page
), &wait
,
789 bit_wait_io
, TASK_KILLABLE
);
792 int wait_on_page_bit_killable_timeout(struct page
*page
,
793 int bit_nr
, unsigned long timeout
)
795 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
797 wait
.key
.timeout
= jiffies
+ timeout
;
798 if (!test_bit(bit_nr
, &page
->flags
))
800 return __wait_on_bit(page_waitqueue(page
), &wait
,
801 bit_wait_io_timeout
, TASK_KILLABLE
);
803 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
806 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
807 * @page: Page defining the wait queue of interest
808 * @waiter: Waiter to add to the queue
810 * Add an arbitrary @waiter to the wait queue for the nominated @page.
812 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
814 wait_queue_head_t
*q
= page_waitqueue(page
);
817 spin_lock_irqsave(&q
->lock
, flags
);
818 __add_wait_queue(q
, waiter
);
819 spin_unlock_irqrestore(&q
->lock
, flags
);
821 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
824 * unlock_page - unlock a locked page
827 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
828 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
829 * mechanism between PageLocked pages and PageWriteback pages is shared.
830 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
832 * The mb is necessary to enforce ordering between the clear_bit and the read
833 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
835 void unlock_page(struct page
*page
)
837 page
= compound_head(page
);
838 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
839 clear_bit_unlock(PG_locked
, &page
->flags
);
840 smp_mb__after_atomic();
841 wake_up_page(page
, PG_locked
);
843 EXPORT_SYMBOL(unlock_page
);
846 * end_page_writeback - end writeback against a page
849 void end_page_writeback(struct page
*page
)
852 * TestClearPageReclaim could be used here but it is an atomic
853 * operation and overkill in this particular case. Failing to
854 * shuffle a page marked for immediate reclaim is too mild to
855 * justify taking an atomic operation penalty at the end of
856 * ever page writeback.
858 if (PageReclaim(page
)) {
859 ClearPageReclaim(page
);
860 rotate_reclaimable_page(page
);
863 if (!test_clear_page_writeback(page
))
866 smp_mb__after_atomic();
867 wake_up_page(page
, PG_writeback
);
869 EXPORT_SYMBOL(end_page_writeback
);
872 * After completing I/O on a page, call this routine to update the page
873 * flags appropriately
875 void page_endio(struct page
*page
, int rw
, int err
)
879 SetPageUptodate(page
);
881 ClearPageUptodate(page
);
885 } else { /* rw == WRITE */
889 mapping_set_error(page
->mapping
, err
);
891 end_page_writeback(page
);
894 EXPORT_SYMBOL_GPL(page_endio
);
897 * __lock_page - get a lock on the page, assuming we need to sleep to get it
898 * @page: the page to lock
900 void __lock_page(struct page
*page
)
902 struct page
*page_head
= compound_head(page
);
903 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
905 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
906 TASK_UNINTERRUPTIBLE
);
908 EXPORT_SYMBOL(__lock_page
);
910 int __lock_page_killable(struct page
*page
)
912 struct page
*page_head
= compound_head(page
);
913 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
915 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
916 bit_wait_io
, TASK_KILLABLE
);
918 EXPORT_SYMBOL_GPL(__lock_page_killable
);
922 * 1 - page is locked; mmap_sem is still held.
923 * 0 - page is not locked.
924 * mmap_sem has been released (up_read()), unless flags had both
925 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
926 * which case mmap_sem is still held.
928 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
929 * with the page locked and the mmap_sem unperturbed.
931 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
934 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
936 * CAUTION! In this case, mmap_sem is not released
937 * even though return 0.
939 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
942 up_read(&mm
->mmap_sem
);
943 if (flags
& FAULT_FLAG_KILLABLE
)
944 wait_on_page_locked_killable(page
);
946 wait_on_page_locked(page
);
949 if (flags
& FAULT_FLAG_KILLABLE
) {
952 ret
= __lock_page_killable(page
);
954 up_read(&mm
->mmap_sem
);
964 * page_cache_next_hole - find the next hole (not-present entry)
967 * @max_scan: maximum range to search
969 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
970 * lowest indexed hole.
972 * Returns: the index of the hole if found, otherwise returns an index
973 * outside of the set specified (in which case 'return - index >=
974 * max_scan' will be true). In rare cases of index wrap-around, 0 will
977 * page_cache_next_hole may be called under rcu_read_lock. However,
978 * like radix_tree_gang_lookup, this will not atomically search a
979 * snapshot of the tree at a single point in time. For example, if a
980 * hole is created at index 5, then subsequently a hole is created at
981 * index 10, page_cache_next_hole covering both indexes may return 10
982 * if called under rcu_read_lock.
984 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
985 pgoff_t index
, unsigned long max_scan
)
989 for (i
= 0; i
< max_scan
; i
++) {
992 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
993 if (!page
|| radix_tree_exceptional_entry(page
))
1002 EXPORT_SYMBOL(page_cache_next_hole
);
1005 * page_cache_prev_hole - find the prev hole (not-present entry)
1008 * @max_scan: maximum range to search
1010 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1013 * Returns: the index of the hole if found, otherwise returns an index
1014 * outside of the set specified (in which case 'index - return >=
1015 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1018 * page_cache_prev_hole may be called under rcu_read_lock. However,
1019 * like radix_tree_gang_lookup, this will not atomically search a
1020 * snapshot of the tree at a single point in time. For example, if a
1021 * hole is created at index 10, then subsequently a hole is created at
1022 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1023 * called under rcu_read_lock.
1025 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1026 pgoff_t index
, unsigned long max_scan
)
1030 for (i
= 0; i
< max_scan
; i
++) {
1033 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1034 if (!page
|| radix_tree_exceptional_entry(page
))
1037 if (index
== ULONG_MAX
)
1043 EXPORT_SYMBOL(page_cache_prev_hole
);
1046 * find_get_entry - find and get a page cache entry
1047 * @mapping: the address_space to search
1048 * @offset: the page cache index
1050 * Looks up the page cache slot at @mapping & @offset. If there is a
1051 * page cache page, it is returned with an increased refcount.
1053 * If the slot holds a shadow entry of a previously evicted page, or a
1054 * swap entry from shmem/tmpfs, it is returned.
1056 * Otherwise, %NULL is returned.
1058 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1066 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1068 page
= radix_tree_deref_slot(pagep
);
1069 if (unlikely(!page
))
1071 if (radix_tree_exception(page
)) {
1072 if (radix_tree_deref_retry(page
))
1075 * A shadow entry of a recently evicted page,
1076 * or a swap entry from shmem/tmpfs. Return
1077 * it without attempting to raise page count.
1081 if (!page_cache_get_speculative(page
))
1085 * Has the page moved?
1086 * This is part of the lockless pagecache protocol. See
1087 * include/linux/pagemap.h for details.
1089 if (unlikely(page
!= *pagep
)) {
1099 EXPORT_SYMBOL(find_get_entry
);
1102 * find_lock_entry - locate, pin and lock a page cache entry
1103 * @mapping: the address_space to search
1104 * @offset: the page cache index
1106 * Looks up the page cache slot at @mapping & @offset. If there is a
1107 * page cache page, it is returned locked and with an increased
1110 * If the slot holds a shadow entry of a previously evicted page, or a
1111 * swap entry from shmem/tmpfs, it is returned.
1113 * Otherwise, %NULL is returned.
1115 * find_lock_entry() may sleep.
1117 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1122 page
= find_get_entry(mapping
, offset
);
1123 if (page
&& !radix_tree_exception(page
)) {
1125 /* Has the page been truncated? */
1126 if (unlikely(page
->mapping
!= mapping
)) {
1131 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1135 EXPORT_SYMBOL(find_lock_entry
);
1138 * pagecache_get_page - find and get a page reference
1139 * @mapping: the address_space to search
1140 * @offset: the page index
1141 * @fgp_flags: PCG flags
1142 * @gfp_mask: gfp mask to use for the page cache data page allocation
1144 * Looks up the page cache slot at @mapping & @offset.
1146 * PCG flags modify how the page is returned.
1148 * FGP_ACCESSED: the page will be marked accessed
1149 * FGP_LOCK: Page is return locked
1150 * FGP_CREAT: If page is not present then a new page is allocated using
1151 * @gfp_mask and added to the page cache and the VM's LRU
1152 * list. The page is returned locked and with an increased
1153 * refcount. Otherwise, %NULL is returned.
1155 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1156 * if the GFP flags specified for FGP_CREAT are atomic.
1158 * If there is a page cache page, it is returned with an increased refcount.
1160 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1161 int fgp_flags
, gfp_t gfp_mask
)
1166 page
= find_get_entry(mapping
, offset
);
1167 if (radix_tree_exceptional_entry(page
))
1172 if (fgp_flags
& FGP_LOCK
) {
1173 if (fgp_flags
& FGP_NOWAIT
) {
1174 if (!trylock_page(page
)) {
1182 /* Has the page been truncated? */
1183 if (unlikely(page
->mapping
!= mapping
)) {
1188 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1191 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1192 mark_page_accessed(page
);
1195 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1197 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1198 gfp_mask
|= __GFP_WRITE
;
1199 if (fgp_flags
& FGP_NOFS
)
1200 gfp_mask
&= ~__GFP_FS
;
1202 page
= __page_cache_alloc(gfp_mask
);
1206 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1207 fgp_flags
|= FGP_LOCK
;
1209 /* Init accessed so avoid atomic mark_page_accessed later */
1210 if (fgp_flags
& FGP_ACCESSED
)
1211 __SetPageReferenced(page
);
1213 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1214 gfp_mask
& GFP_RECLAIM_MASK
);
1215 if (unlikely(err
)) {
1225 EXPORT_SYMBOL(pagecache_get_page
);
1228 * find_get_entries - gang pagecache lookup
1229 * @mapping: The address_space to search
1230 * @start: The starting page cache index
1231 * @nr_entries: The maximum number of entries
1232 * @entries: Where the resulting entries are placed
1233 * @indices: The cache indices corresponding to the entries in @entries
1235 * find_get_entries() will search for and return a group of up to
1236 * @nr_entries entries in the mapping. The entries are placed at
1237 * @entries. find_get_entries() takes a reference against any actual
1240 * The search returns a group of mapping-contiguous page cache entries
1241 * with ascending indexes. There may be holes in the indices due to
1242 * not-present pages.
1244 * Any shadow entries of evicted pages, or swap entries from
1245 * shmem/tmpfs, are included in the returned array.
1247 * find_get_entries() returns the number of pages and shadow entries
1250 unsigned find_get_entries(struct address_space
*mapping
,
1251 pgoff_t start
, unsigned int nr_entries
,
1252 struct page
**entries
, pgoff_t
*indices
)
1255 unsigned int ret
= 0;
1256 struct radix_tree_iter iter
;
1262 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1265 page
= radix_tree_deref_slot(slot
);
1266 if (unlikely(!page
))
1268 if (radix_tree_exception(page
)) {
1269 if (radix_tree_deref_retry(page
)) {
1270 slot
= radix_tree_iter_retry(&iter
);
1274 * A shadow entry of a recently evicted page, a swap
1275 * entry from shmem/tmpfs or a DAX entry. Return it
1276 * without attempting to raise page count.
1280 if (!page_cache_get_speculative(page
))
1283 /* Has the page moved? */
1284 if (unlikely(page
!= *slot
)) {
1289 indices
[ret
] = iter
.index
;
1290 entries
[ret
] = page
;
1291 if (++ret
== nr_entries
)
1299 * find_get_pages - gang pagecache lookup
1300 * @mapping: The address_space to search
1301 * @start: The starting page index
1302 * @nr_pages: The maximum number of pages
1303 * @pages: Where the resulting pages are placed
1305 * find_get_pages() will search for and return a group of up to
1306 * @nr_pages pages in the mapping. The pages are placed at @pages.
1307 * find_get_pages() takes a reference against the returned pages.
1309 * The search returns a group of mapping-contiguous pages with ascending
1310 * indexes. There may be holes in the indices due to not-present pages.
1312 * find_get_pages() returns the number of pages which were found.
1314 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1315 unsigned int nr_pages
, struct page
**pages
)
1317 struct radix_tree_iter iter
;
1321 if (unlikely(!nr_pages
))
1325 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1328 page
= radix_tree_deref_slot(slot
);
1329 if (unlikely(!page
))
1332 if (radix_tree_exception(page
)) {
1333 if (radix_tree_deref_retry(page
)) {
1334 slot
= radix_tree_iter_retry(&iter
);
1338 * A shadow entry of a recently evicted page,
1339 * or a swap entry from shmem/tmpfs. Skip
1345 if (!page_cache_get_speculative(page
))
1348 /* Has the page moved? */
1349 if (unlikely(page
!= *slot
)) {
1355 if (++ret
== nr_pages
)
1364 * find_get_pages_contig - gang contiguous pagecache lookup
1365 * @mapping: The address_space to search
1366 * @index: The starting page index
1367 * @nr_pages: The maximum number of pages
1368 * @pages: Where the resulting pages are placed
1370 * find_get_pages_contig() works exactly like find_get_pages(), except
1371 * that the returned number of pages are guaranteed to be contiguous.
1373 * find_get_pages_contig() returns the number of pages which were found.
1375 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1376 unsigned int nr_pages
, struct page
**pages
)
1378 struct radix_tree_iter iter
;
1380 unsigned int ret
= 0;
1382 if (unlikely(!nr_pages
))
1386 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1389 page
= radix_tree_deref_slot(slot
);
1390 /* The hole, there no reason to continue */
1391 if (unlikely(!page
))
1394 if (radix_tree_exception(page
)) {
1395 if (radix_tree_deref_retry(page
)) {
1396 slot
= radix_tree_iter_retry(&iter
);
1400 * A shadow entry of a recently evicted page,
1401 * or a swap entry from shmem/tmpfs. Stop
1402 * looking for contiguous pages.
1407 if (!page_cache_get_speculative(page
))
1410 /* Has the page moved? */
1411 if (unlikely(page
!= *slot
)) {
1417 * must check mapping and index after taking the ref.
1418 * otherwise we can get both false positives and false
1419 * negatives, which is just confusing to the caller.
1421 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1427 if (++ret
== nr_pages
)
1433 EXPORT_SYMBOL(find_get_pages_contig
);
1436 * find_get_pages_tag - find and return pages that match @tag
1437 * @mapping: the address_space to search
1438 * @index: the starting page index
1439 * @tag: the tag index
1440 * @nr_pages: the maximum number of pages
1441 * @pages: where the resulting pages are placed
1443 * Like find_get_pages, except we only return pages which are tagged with
1444 * @tag. We update @index to index the next page for the traversal.
1446 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1447 int tag
, unsigned int nr_pages
, struct page
**pages
)
1449 struct radix_tree_iter iter
;
1453 if (unlikely(!nr_pages
))
1457 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1458 &iter
, *index
, tag
) {
1461 page
= radix_tree_deref_slot(slot
);
1462 if (unlikely(!page
))
1465 if (radix_tree_exception(page
)) {
1466 if (radix_tree_deref_retry(page
)) {
1467 slot
= radix_tree_iter_retry(&iter
);
1471 * A shadow entry of a recently evicted page.
1473 * Those entries should never be tagged, but
1474 * this tree walk is lockless and the tags are
1475 * looked up in bulk, one radix tree node at a
1476 * time, so there is a sizable window for page
1477 * reclaim to evict a page we saw tagged.
1484 if (!page_cache_get_speculative(page
))
1487 /* Has the page moved? */
1488 if (unlikely(page
!= *slot
)) {
1494 if (++ret
== nr_pages
)
1501 *index
= pages
[ret
- 1]->index
+ 1;
1505 EXPORT_SYMBOL(find_get_pages_tag
);
1508 * find_get_entries_tag - find and return entries that match @tag
1509 * @mapping: the address_space to search
1510 * @start: the starting page cache index
1511 * @tag: the tag index
1512 * @nr_entries: the maximum number of entries
1513 * @entries: where the resulting entries are placed
1514 * @indices: the cache indices corresponding to the entries in @entries
1516 * Like find_get_entries, except we only return entries which are tagged with
1519 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1520 int tag
, unsigned int nr_entries
,
1521 struct page
**entries
, pgoff_t
*indices
)
1524 unsigned int ret
= 0;
1525 struct radix_tree_iter iter
;
1531 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1532 &iter
, start
, tag
) {
1535 page
= radix_tree_deref_slot(slot
);
1536 if (unlikely(!page
))
1538 if (radix_tree_exception(page
)) {
1539 if (radix_tree_deref_retry(page
)) {
1540 slot
= radix_tree_iter_retry(&iter
);
1545 * A shadow entry of a recently evicted page, a swap
1546 * entry from shmem/tmpfs or a DAX entry. Return it
1547 * without attempting to raise page count.
1551 if (!page_cache_get_speculative(page
))
1554 /* Has the page moved? */
1555 if (unlikely(page
!= *slot
)) {
1560 indices
[ret
] = iter
.index
;
1561 entries
[ret
] = page
;
1562 if (++ret
== nr_entries
)
1568 EXPORT_SYMBOL(find_get_entries_tag
);
1571 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1572 * a _large_ part of the i/o request. Imagine the worst scenario:
1574 * ---R__________________________________________B__________
1575 * ^ reading here ^ bad block(assume 4k)
1577 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1578 * => failing the whole request => read(R) => read(R+1) =>
1579 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1580 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1581 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1583 * It is going insane. Fix it by quickly scaling down the readahead size.
1585 static void shrink_readahead_size_eio(struct file
*filp
,
1586 struct file_ra_state
*ra
)
1592 * do_generic_file_read - generic file read routine
1593 * @filp: the file to read
1594 * @ppos: current file position
1595 * @iter: data destination
1596 * @written: already copied
1598 * This is a generic file read routine, and uses the
1599 * mapping->a_ops->readpage() function for the actual low-level stuff.
1601 * This is really ugly. But the goto's actually try to clarify some
1602 * of the logic when it comes to error handling etc.
1604 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1605 struct iov_iter
*iter
, ssize_t written
)
1607 struct address_space
*mapping
= filp
->f_mapping
;
1608 struct inode
*inode
= mapping
->host
;
1609 struct file_ra_state
*ra
= &filp
->f_ra
;
1613 unsigned long offset
; /* offset into pagecache page */
1614 unsigned int prev_offset
;
1617 index
= *ppos
>> PAGE_SHIFT
;
1618 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1619 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1620 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1621 offset
= *ppos
& ~PAGE_MASK
;
1627 unsigned long nr
, ret
;
1631 page
= find_get_page(mapping
, index
);
1633 page_cache_sync_readahead(mapping
,
1635 index
, last_index
- index
);
1636 page
= find_get_page(mapping
, index
);
1637 if (unlikely(page
== NULL
))
1638 goto no_cached_page
;
1640 if (PageReadahead(page
)) {
1641 page_cache_async_readahead(mapping
,
1643 index
, last_index
- index
);
1645 if (!PageUptodate(page
)) {
1647 * See comment in do_read_cache_page on why
1648 * wait_on_page_locked is used to avoid unnecessarily
1649 * serialisations and why it's safe.
1651 wait_on_page_locked_killable(page
);
1652 if (PageUptodate(page
))
1655 if (inode
->i_blkbits
== PAGE_SHIFT
||
1656 !mapping
->a_ops
->is_partially_uptodate
)
1657 goto page_not_up_to_date
;
1658 if (!trylock_page(page
))
1659 goto page_not_up_to_date
;
1660 /* Did it get truncated before we got the lock? */
1662 goto page_not_up_to_date_locked
;
1663 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1664 offset
, iter
->count
))
1665 goto page_not_up_to_date_locked
;
1670 * i_size must be checked after we know the page is Uptodate.
1672 * Checking i_size after the check allows us to calculate
1673 * the correct value for "nr", which means the zero-filled
1674 * part of the page is not copied back to userspace (unless
1675 * another truncate extends the file - this is desired though).
1678 isize
= i_size_read(inode
);
1679 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1680 if (unlikely(!isize
|| index
> end_index
)) {
1685 /* nr is the maximum number of bytes to copy from this page */
1687 if (index
== end_index
) {
1688 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1696 /* If users can be writing to this page using arbitrary
1697 * virtual addresses, take care about potential aliasing
1698 * before reading the page on the kernel side.
1700 if (mapping_writably_mapped(mapping
))
1701 flush_dcache_page(page
);
1704 * When a sequential read accesses a page several times,
1705 * only mark it as accessed the first time.
1707 if (prev_index
!= index
|| offset
!= prev_offset
)
1708 mark_page_accessed(page
);
1712 * Ok, we have the page, and it's up-to-date, so
1713 * now we can copy it to user space...
1716 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1718 index
+= offset
>> PAGE_SHIFT
;
1719 offset
&= ~PAGE_MASK
;
1720 prev_offset
= offset
;
1724 if (!iov_iter_count(iter
))
1732 page_not_up_to_date
:
1733 /* Get exclusive access to the page ... */
1734 error
= lock_page_killable(page
);
1735 if (unlikely(error
))
1736 goto readpage_error
;
1738 page_not_up_to_date_locked
:
1739 /* Did it get truncated before we got the lock? */
1740 if (!page
->mapping
) {
1746 /* Did somebody else fill it already? */
1747 if (PageUptodate(page
)) {
1754 * A previous I/O error may have been due to temporary
1755 * failures, eg. multipath errors.
1756 * PG_error will be set again if readpage fails.
1758 ClearPageError(page
);
1759 /* Start the actual read. The read will unlock the page. */
1760 error
= mapping
->a_ops
->readpage(filp
, page
);
1762 if (unlikely(error
)) {
1763 if (error
== AOP_TRUNCATED_PAGE
) {
1768 goto readpage_error
;
1771 if (!PageUptodate(page
)) {
1772 error
= lock_page_killable(page
);
1773 if (unlikely(error
))
1774 goto readpage_error
;
1775 if (!PageUptodate(page
)) {
1776 if (page
->mapping
== NULL
) {
1778 * invalidate_mapping_pages got it
1785 shrink_readahead_size_eio(filp
, ra
);
1787 goto readpage_error
;
1795 /* UHHUH! A synchronous read error occurred. Report it */
1801 * Ok, it wasn't cached, so we need to create a new
1804 page
= page_cache_alloc_cold(mapping
);
1809 error
= add_to_page_cache_lru(page
, mapping
, index
,
1810 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1813 if (error
== -EEXIST
) {
1823 ra
->prev_pos
= prev_index
;
1824 ra
->prev_pos
<<= PAGE_SHIFT
;
1825 ra
->prev_pos
|= prev_offset
;
1827 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
1828 file_accessed(filp
);
1829 return written
? written
: error
;
1833 * generic_file_read_iter - generic filesystem read routine
1834 * @iocb: kernel I/O control block
1835 * @iter: destination for the data read
1837 * This is the "read_iter()" routine for all filesystems
1838 * that can use the page cache directly.
1841 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1843 struct file
*file
= iocb
->ki_filp
;
1845 size_t count
= iov_iter_count(iter
);
1848 goto out
; /* skip atime */
1850 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1851 struct address_space
*mapping
= file
->f_mapping
;
1852 struct inode
*inode
= mapping
->host
;
1855 size
= i_size_read(inode
);
1856 retval
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
1857 iocb
->ki_pos
+ count
- 1);
1859 struct iov_iter data
= *iter
;
1860 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
);
1864 iocb
->ki_pos
+= retval
;
1865 iov_iter_advance(iter
, retval
);
1869 * Btrfs can have a short DIO read if we encounter
1870 * compressed extents, so if there was an error, or if
1871 * we've already read everything we wanted to, or if
1872 * there was a short read because we hit EOF, go ahead
1873 * and return. Otherwise fallthrough to buffered io for
1874 * the rest of the read. Buffered reads will not work for
1875 * DAX files, so don't bother trying.
1877 if (retval
< 0 || !iov_iter_count(iter
) || iocb
->ki_pos
>= size
||
1879 file_accessed(file
);
1884 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
1888 EXPORT_SYMBOL(generic_file_read_iter
);
1892 * page_cache_read - adds requested page to the page cache if not already there
1893 * @file: file to read
1894 * @offset: page index
1895 * @gfp_mask: memory allocation flags
1897 * This adds the requested page to the page cache if it isn't already there,
1898 * and schedules an I/O to read in its contents from disk.
1900 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1902 struct address_space
*mapping
= file
->f_mapping
;
1907 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1911 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1913 ret
= mapping
->a_ops
->readpage(file
, page
);
1914 else if (ret
== -EEXIST
)
1915 ret
= 0; /* losing race to add is OK */
1919 } while (ret
== AOP_TRUNCATED_PAGE
);
1924 #define MMAP_LOTSAMISS (100)
1927 * Synchronous readahead happens when we don't even find
1928 * a page in the page cache at all.
1930 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1931 struct file_ra_state
*ra
,
1935 struct address_space
*mapping
= file
->f_mapping
;
1937 /* If we don't want any read-ahead, don't bother */
1938 if (vma
->vm_flags
& VM_RAND_READ
)
1943 if (vma
->vm_flags
& VM_SEQ_READ
) {
1944 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1949 /* Avoid banging the cache line if not needed */
1950 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1954 * Do we miss much more than hit in this file? If so,
1955 * stop bothering with read-ahead. It will only hurt.
1957 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1963 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
1964 ra
->size
= ra
->ra_pages
;
1965 ra
->async_size
= ra
->ra_pages
/ 4;
1966 ra_submit(ra
, mapping
, file
);
1970 * Asynchronous readahead happens when we find the page and PG_readahead,
1971 * so we want to possibly extend the readahead further..
1973 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1974 struct file_ra_state
*ra
,
1979 struct address_space
*mapping
= file
->f_mapping
;
1981 /* If we don't want any read-ahead, don't bother */
1982 if (vma
->vm_flags
& VM_RAND_READ
)
1984 if (ra
->mmap_miss
> 0)
1986 if (PageReadahead(page
))
1987 page_cache_async_readahead(mapping
, ra
, file
,
1988 page
, offset
, ra
->ra_pages
);
1992 * filemap_fault - read in file data for page fault handling
1993 * @vma: vma in which the fault was taken
1994 * @vmf: struct vm_fault containing details of the fault
1996 * filemap_fault() is invoked via the vma operations vector for a
1997 * mapped memory region to read in file data during a page fault.
1999 * The goto's are kind of ugly, but this streamlines the normal case of having
2000 * it in the page cache, and handles the special cases reasonably without
2001 * having a lot of duplicated code.
2003 * vma->vm_mm->mmap_sem must be held on entry.
2005 * If our return value has VM_FAULT_RETRY set, it's because
2006 * lock_page_or_retry() returned 0.
2007 * The mmap_sem has usually been released in this case.
2008 * See __lock_page_or_retry() for the exception.
2010 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2011 * has not been released.
2013 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2015 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2018 struct file
*file
= vma
->vm_file
;
2019 struct address_space
*mapping
= file
->f_mapping
;
2020 struct file_ra_state
*ra
= &file
->f_ra
;
2021 struct inode
*inode
= mapping
->host
;
2022 pgoff_t offset
= vmf
->pgoff
;
2027 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2028 if (offset
>= size
>> PAGE_SHIFT
)
2029 return VM_FAULT_SIGBUS
;
2032 * Do we have something in the page cache already?
2034 page
= find_get_page(mapping
, offset
);
2035 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2037 * We found the page, so try async readahead before
2038 * waiting for the lock.
2040 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2042 /* No page in the page cache at all */
2043 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2044 count_vm_event(PGMAJFAULT
);
2045 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2046 ret
= VM_FAULT_MAJOR
;
2048 page
= find_get_page(mapping
, offset
);
2050 goto no_cached_page
;
2053 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2055 return ret
| VM_FAULT_RETRY
;
2058 /* Did it get truncated? */
2059 if (unlikely(page
->mapping
!= mapping
)) {
2064 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2067 * We have a locked page in the page cache, now we need to check
2068 * that it's up-to-date. If not, it is going to be due to an error.
2070 if (unlikely(!PageUptodate(page
)))
2071 goto page_not_uptodate
;
2074 * Found the page and have a reference on it.
2075 * We must recheck i_size under page lock.
2077 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2078 if (unlikely(offset
>= size
>> PAGE_SHIFT
)) {
2081 return VM_FAULT_SIGBUS
;
2085 return ret
| VM_FAULT_LOCKED
;
2089 * We're only likely to ever get here if MADV_RANDOM is in
2092 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2095 * The page we want has now been added to the page cache.
2096 * In the unlikely event that someone removed it in the
2097 * meantime, we'll just come back here and read it again.
2103 * An error return from page_cache_read can result if the
2104 * system is low on memory, or a problem occurs while trying
2107 if (error
== -ENOMEM
)
2108 return VM_FAULT_OOM
;
2109 return VM_FAULT_SIGBUS
;
2113 * Umm, take care of errors if the page isn't up-to-date.
2114 * Try to re-read it _once_. We do this synchronously,
2115 * because there really aren't any performance issues here
2116 * and we need to check for errors.
2118 ClearPageError(page
);
2119 error
= mapping
->a_ops
->readpage(file
, page
);
2121 wait_on_page_locked(page
);
2122 if (!PageUptodate(page
))
2127 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2130 /* Things didn't work out. Return zero to tell the mm layer so. */
2131 shrink_readahead_size_eio(file
, ra
);
2132 return VM_FAULT_SIGBUS
;
2134 EXPORT_SYMBOL(filemap_fault
);
2136 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2138 struct radix_tree_iter iter
;
2140 struct file
*file
= vma
->vm_file
;
2141 struct address_space
*mapping
= file
->f_mapping
;
2144 unsigned long address
= (unsigned long) vmf
->virtual_address
;
2149 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
2150 if (iter
.index
> vmf
->max_pgoff
)
2153 page
= radix_tree_deref_slot(slot
);
2154 if (unlikely(!page
))
2156 if (radix_tree_exception(page
)) {
2157 if (radix_tree_deref_retry(page
)) {
2158 slot
= radix_tree_iter_retry(&iter
);
2164 if (!page_cache_get_speculative(page
))
2167 /* Has the page moved? */
2168 if (unlikely(page
!= *slot
)) {
2173 if (!PageUptodate(page
) ||
2174 PageReadahead(page
) ||
2177 if (!trylock_page(page
))
2180 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2183 size
= round_up(i_size_read(mapping
->host
), PAGE_SIZE
);
2184 if (page
->index
>= size
>> PAGE_SHIFT
)
2187 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2188 if (!pte_none(*pte
))
2191 if (file
->f_ra
.mmap_miss
> 0)
2192 file
->f_ra
.mmap_miss
--;
2193 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2194 do_set_pte(vma
, addr
, page
, pte
, false, false, true);
2202 if (iter
.index
== vmf
->max_pgoff
)
2207 EXPORT_SYMBOL(filemap_map_pages
);
2209 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2211 struct page
*page
= vmf
->page
;
2212 struct inode
*inode
= file_inode(vma
->vm_file
);
2213 int ret
= VM_FAULT_LOCKED
;
2215 sb_start_pagefault(inode
->i_sb
);
2216 file_update_time(vma
->vm_file
);
2218 if (page
->mapping
!= inode
->i_mapping
) {
2220 ret
= VM_FAULT_NOPAGE
;
2224 * We mark the page dirty already here so that when freeze is in
2225 * progress, we are guaranteed that writeback during freezing will
2226 * see the dirty page and writeprotect it again.
2228 set_page_dirty(page
);
2229 wait_for_stable_page(page
);
2231 sb_end_pagefault(inode
->i_sb
);
2234 EXPORT_SYMBOL(filemap_page_mkwrite
);
2236 const struct vm_operations_struct generic_file_vm_ops
= {
2237 .fault
= filemap_fault
,
2238 .map_pages
= filemap_map_pages
,
2239 .page_mkwrite
= filemap_page_mkwrite
,
2242 /* This is used for a general mmap of a disk file */
2244 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2246 struct address_space
*mapping
= file
->f_mapping
;
2248 if (!mapping
->a_ops
->readpage
)
2250 file_accessed(file
);
2251 vma
->vm_ops
= &generic_file_vm_ops
;
2256 * This is for filesystems which do not implement ->writepage.
2258 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2260 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2262 return generic_file_mmap(file
, vma
);
2265 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2269 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2273 #endif /* CONFIG_MMU */
2275 EXPORT_SYMBOL(generic_file_mmap
);
2276 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2278 static struct page
*wait_on_page_read(struct page
*page
)
2280 if (!IS_ERR(page
)) {
2281 wait_on_page_locked(page
);
2282 if (!PageUptodate(page
)) {
2284 page
= ERR_PTR(-EIO
);
2290 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2292 int (*filler
)(void *, struct page
*),
2299 page
= find_get_page(mapping
, index
);
2301 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2303 return ERR_PTR(-ENOMEM
);
2304 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2305 if (unlikely(err
)) {
2309 /* Presumably ENOMEM for radix tree node */
2310 return ERR_PTR(err
);
2314 err
= filler(data
, page
);
2317 return ERR_PTR(err
);
2320 page
= wait_on_page_read(page
);
2325 if (PageUptodate(page
))
2329 * Page is not up to date and may be locked due one of the following
2330 * case a: Page is being filled and the page lock is held
2331 * case b: Read/write error clearing the page uptodate status
2332 * case c: Truncation in progress (page locked)
2333 * case d: Reclaim in progress
2335 * Case a, the page will be up to date when the page is unlocked.
2336 * There is no need to serialise on the page lock here as the page
2337 * is pinned so the lock gives no additional protection. Even if the
2338 * the page is truncated, the data is still valid if PageUptodate as
2339 * it's a race vs truncate race.
2340 * Case b, the page will not be up to date
2341 * Case c, the page may be truncated but in itself, the data may still
2342 * be valid after IO completes as it's a read vs truncate race. The
2343 * operation must restart if the page is not uptodate on unlock but
2344 * otherwise serialising on page lock to stabilise the mapping gives
2345 * no additional guarantees to the caller as the page lock is
2346 * released before return.
2347 * Case d, similar to truncation. If reclaim holds the page lock, it
2348 * will be a race with remove_mapping that determines if the mapping
2349 * is valid on unlock but otherwise the data is valid and there is
2350 * no need to serialise with page lock.
2352 * As the page lock gives no additional guarantee, we optimistically
2353 * wait on the page to be unlocked and check if it's up to date and
2354 * use the page if it is. Otherwise, the page lock is required to
2355 * distinguish between the different cases. The motivation is that we
2356 * avoid spurious serialisations and wakeups when multiple processes
2357 * wait on the same page for IO to complete.
2359 wait_on_page_locked(page
);
2360 if (PageUptodate(page
))
2363 /* Distinguish between all the cases under the safety of the lock */
2366 /* Case c or d, restart the operation */
2367 if (!page
->mapping
) {
2373 /* Someone else locked and filled the page in a very small window */
2374 if (PageUptodate(page
)) {
2381 mark_page_accessed(page
);
2386 * read_cache_page - read into page cache, fill it if needed
2387 * @mapping: the page's address_space
2388 * @index: the page index
2389 * @filler: function to perform the read
2390 * @data: first arg to filler(data, page) function, often left as NULL
2392 * Read into the page cache. If a page already exists, and PageUptodate() is
2393 * not set, try to fill the page and wait for it to become unlocked.
2395 * If the page does not get brought uptodate, return -EIO.
2397 struct page
*read_cache_page(struct address_space
*mapping
,
2399 int (*filler
)(void *, struct page
*),
2402 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2404 EXPORT_SYMBOL(read_cache_page
);
2407 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2408 * @mapping: the page's address_space
2409 * @index: the page index
2410 * @gfp: the page allocator flags to use if allocating
2412 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2413 * any new page allocations done using the specified allocation flags.
2415 * If the page does not get brought uptodate, return -EIO.
2417 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2421 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2423 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2425 EXPORT_SYMBOL(read_cache_page_gfp
);
2428 * Performs necessary checks before doing a write
2430 * Can adjust writing position or amount of bytes to write.
2431 * Returns appropriate error code that caller should return or
2432 * zero in case that write should be allowed.
2434 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2436 struct file
*file
= iocb
->ki_filp
;
2437 struct inode
*inode
= file
->f_mapping
->host
;
2438 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2441 if (!iov_iter_count(from
))
2444 /* FIXME: this is for backwards compatibility with 2.4 */
2445 if (iocb
->ki_flags
& IOCB_APPEND
)
2446 iocb
->ki_pos
= i_size_read(inode
);
2450 if (limit
!= RLIM_INFINITY
) {
2451 if (iocb
->ki_pos
>= limit
) {
2452 send_sig(SIGXFSZ
, current
, 0);
2455 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2461 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2462 !(file
->f_flags
& O_LARGEFILE
))) {
2463 if (pos
>= MAX_NON_LFS
)
2465 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2469 * Are we about to exceed the fs block limit ?
2471 * If we have written data it becomes a short write. If we have
2472 * exceeded without writing data we send a signal and return EFBIG.
2473 * Linus frestrict idea will clean these up nicely..
2475 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2478 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2479 return iov_iter_count(from
);
2481 EXPORT_SYMBOL(generic_write_checks
);
2483 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2484 loff_t pos
, unsigned len
, unsigned flags
,
2485 struct page
**pagep
, void **fsdata
)
2487 const struct address_space_operations
*aops
= mapping
->a_ops
;
2489 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2492 EXPORT_SYMBOL(pagecache_write_begin
);
2494 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2495 loff_t pos
, unsigned len
, unsigned copied
,
2496 struct page
*page
, void *fsdata
)
2498 const struct address_space_operations
*aops
= mapping
->a_ops
;
2500 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2502 EXPORT_SYMBOL(pagecache_write_end
);
2505 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2507 struct file
*file
= iocb
->ki_filp
;
2508 struct address_space
*mapping
= file
->f_mapping
;
2509 struct inode
*inode
= mapping
->host
;
2510 loff_t pos
= iocb
->ki_pos
;
2514 struct iov_iter data
;
2516 write_len
= iov_iter_count(from
);
2517 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2519 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2524 * After a write we want buffered reads to be sure to go to disk to get
2525 * the new data. We invalidate clean cached page from the region we're
2526 * about to write. We do this *before* the write so that we can return
2527 * without clobbering -EIOCBQUEUED from ->direct_IO().
2529 if (mapping
->nrpages
) {
2530 written
= invalidate_inode_pages2_range(mapping
,
2531 pos
>> PAGE_SHIFT
, end
);
2533 * If a page can not be invalidated, return 0 to fall back
2534 * to buffered write.
2537 if (written
== -EBUSY
)
2544 written
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2547 * Finally, try again to invalidate clean pages which might have been
2548 * cached by non-direct readahead, or faulted in by get_user_pages()
2549 * if the source of the write was an mmap'ed region of the file
2550 * we're writing. Either one is a pretty crazy thing to do,
2551 * so we don't support it 100%. If this invalidation
2552 * fails, tough, the write still worked...
2554 if (mapping
->nrpages
) {
2555 invalidate_inode_pages2_range(mapping
,
2556 pos
>> PAGE_SHIFT
, end
);
2561 iov_iter_advance(from
, written
);
2562 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2563 i_size_write(inode
, pos
);
2564 mark_inode_dirty(inode
);
2571 EXPORT_SYMBOL(generic_file_direct_write
);
2574 * Find or create a page at the given pagecache position. Return the locked
2575 * page. This function is specifically for buffered writes.
2577 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2578 pgoff_t index
, unsigned flags
)
2581 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2583 if (flags
& AOP_FLAG_NOFS
)
2584 fgp_flags
|= FGP_NOFS
;
2586 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2587 mapping_gfp_mask(mapping
));
2589 wait_for_stable_page(page
);
2593 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2595 ssize_t
generic_perform_write(struct file
*file
,
2596 struct iov_iter
*i
, loff_t pos
)
2598 struct address_space
*mapping
= file
->f_mapping
;
2599 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2601 ssize_t written
= 0;
2602 unsigned int flags
= 0;
2605 * Copies from kernel address space cannot fail (NFSD is a big user).
2607 if (!iter_is_iovec(i
))
2608 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2612 unsigned long offset
; /* Offset into pagecache page */
2613 unsigned long bytes
; /* Bytes to write to page */
2614 size_t copied
; /* Bytes copied from user */
2617 offset
= (pos
& (PAGE_SIZE
- 1));
2618 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2623 * Bring in the user page that we will copy from _first_.
2624 * Otherwise there's a nasty deadlock on copying from the
2625 * same page as we're writing to, without it being marked
2628 * Not only is this an optimisation, but it is also required
2629 * to check that the address is actually valid, when atomic
2630 * usercopies are used, below.
2632 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2637 if (fatal_signal_pending(current
)) {
2642 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2644 if (unlikely(status
< 0))
2647 if (mapping_writably_mapped(mapping
))
2648 flush_dcache_page(page
);
2650 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2651 flush_dcache_page(page
);
2653 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2655 if (unlikely(status
< 0))
2661 iov_iter_advance(i
, copied
);
2662 if (unlikely(copied
== 0)) {
2664 * If we were unable to copy any data at all, we must
2665 * fall back to a single segment length write.
2667 * If we didn't fallback here, we could livelock
2668 * because not all segments in the iov can be copied at
2669 * once without a pagefault.
2671 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2672 iov_iter_single_seg_count(i
));
2678 balance_dirty_pages_ratelimited(mapping
);
2679 } while (iov_iter_count(i
));
2681 return written
? written
: status
;
2683 EXPORT_SYMBOL(generic_perform_write
);
2686 * __generic_file_write_iter - write data to a file
2687 * @iocb: IO state structure (file, offset, etc.)
2688 * @from: iov_iter with data to write
2690 * This function does all the work needed for actually writing data to a
2691 * file. It does all basic checks, removes SUID from the file, updates
2692 * modification times and calls proper subroutines depending on whether we
2693 * do direct IO or a standard buffered write.
2695 * It expects i_mutex to be grabbed unless we work on a block device or similar
2696 * object which does not need locking at all.
2698 * This function does *not* take care of syncing data in case of O_SYNC write.
2699 * A caller has to handle it. This is mainly due to the fact that we want to
2700 * avoid syncing under i_mutex.
2702 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2704 struct file
*file
= iocb
->ki_filp
;
2705 struct address_space
* mapping
= file
->f_mapping
;
2706 struct inode
*inode
= mapping
->host
;
2707 ssize_t written
= 0;
2711 /* We can write back this queue in page reclaim */
2712 current
->backing_dev_info
= inode_to_bdi(inode
);
2713 err
= file_remove_privs(file
);
2717 err
= file_update_time(file
);
2721 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2722 loff_t pos
, endbyte
;
2724 written
= generic_file_direct_write(iocb
, from
);
2726 * If the write stopped short of completing, fall back to
2727 * buffered writes. Some filesystems do this for writes to
2728 * holes, for example. For DAX files, a buffered write will
2729 * not succeed (even if it did, DAX does not handle dirty
2730 * page-cache pages correctly).
2732 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2735 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2737 * If generic_perform_write() returned a synchronous error
2738 * then we want to return the number of bytes which were
2739 * direct-written, or the error code if that was zero. Note
2740 * that this differs from normal direct-io semantics, which
2741 * will return -EFOO even if some bytes were written.
2743 if (unlikely(status
< 0)) {
2748 * We need to ensure that the page cache pages are written to
2749 * disk and invalidated to preserve the expected O_DIRECT
2752 endbyte
= pos
+ status
- 1;
2753 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2755 iocb
->ki_pos
= endbyte
+ 1;
2757 invalidate_mapping_pages(mapping
,
2759 endbyte
>> PAGE_SHIFT
);
2762 * We don't know how much we wrote, so just return
2763 * the number of bytes which were direct-written
2767 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2768 if (likely(written
> 0))
2769 iocb
->ki_pos
+= written
;
2772 current
->backing_dev_info
= NULL
;
2773 return written
? written
: err
;
2775 EXPORT_SYMBOL(__generic_file_write_iter
);
2778 * generic_file_write_iter - write data to a file
2779 * @iocb: IO state structure
2780 * @from: iov_iter with data to write
2782 * This is a wrapper around __generic_file_write_iter() to be used by most
2783 * filesystems. It takes care of syncing the file in case of O_SYNC file
2784 * and acquires i_mutex as needed.
2786 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2788 struct file
*file
= iocb
->ki_filp
;
2789 struct inode
*inode
= file
->f_mapping
->host
;
2793 ret
= generic_write_checks(iocb
, from
);
2795 ret
= __generic_file_write_iter(iocb
, from
);
2796 inode_unlock(inode
);
2799 ret
= generic_write_sync(iocb
, ret
);
2802 EXPORT_SYMBOL(generic_file_write_iter
);
2805 * try_to_release_page() - release old fs-specific metadata on a page
2807 * @page: the page which the kernel is trying to free
2808 * @gfp_mask: memory allocation flags (and I/O mode)
2810 * The address_space is to try to release any data against the page
2811 * (presumably at page->private). If the release was successful, return `1'.
2812 * Otherwise return zero.
2814 * This may also be called if PG_fscache is set on a page, indicating that the
2815 * page is known to the local caching routines.
2817 * The @gfp_mask argument specifies whether I/O may be performed to release
2818 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2821 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2823 struct address_space
* const mapping
= page
->mapping
;
2825 BUG_ON(!PageLocked(page
));
2826 if (PageWriteback(page
))
2829 if (mapping
&& mapping
->a_ops
->releasepage
)
2830 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2831 return try_to_free_buffers(page
);
2834 EXPORT_SYMBOL(try_to_release_page
);