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 atomic_sub(mapcount
, &page
->_count
);
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
);
268 page_cache_release(page
);
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_CACHE_SHIFT
;
356 pgoff_t end
= end_byte
>> PAGE_CACHE_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();
575 page_cache_release(old
);
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
,
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);
654 page_cache_get(page
);
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);
678 page_cache_release(page
);
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.
717 if (shadow
&& workingset_refault(shadow
)) {
719 workingset_activation(page
);
721 ClearPageActive(page
);
726 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
729 struct page
*__page_cache_alloc(gfp_t gfp
)
734 if (cpuset_do_page_mem_spread()) {
735 unsigned int cpuset_mems_cookie
;
737 cpuset_mems_cookie
= read_mems_allowed_begin();
738 n
= cpuset_mem_spread_node();
739 page
= __alloc_pages_node(n
, gfp
, 0);
740 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
744 return alloc_pages(gfp
, 0);
746 EXPORT_SYMBOL(__page_cache_alloc
);
750 * In order to wait for pages to become available there must be
751 * waitqueues associated with pages. By using a hash table of
752 * waitqueues where the bucket discipline is to maintain all
753 * waiters on the same queue and wake all when any of the pages
754 * become available, and for the woken contexts to check to be
755 * sure the appropriate page became available, this saves space
756 * at a cost of "thundering herd" phenomena during rare hash
759 wait_queue_head_t
*page_waitqueue(struct page
*page
)
761 const struct zone
*zone
= page_zone(page
);
763 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
765 EXPORT_SYMBOL(page_waitqueue
);
767 void wait_on_page_bit(struct page
*page
, int bit_nr
)
769 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
771 if (test_bit(bit_nr
, &page
->flags
))
772 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
773 TASK_UNINTERRUPTIBLE
);
775 EXPORT_SYMBOL(wait_on_page_bit
);
777 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
779 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
781 if (!test_bit(bit_nr
, &page
->flags
))
784 return __wait_on_bit(page_waitqueue(page
), &wait
,
785 bit_wait_io
, TASK_KILLABLE
);
788 int wait_on_page_bit_killable_timeout(struct page
*page
,
789 int bit_nr
, unsigned long timeout
)
791 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
793 wait
.key
.timeout
= jiffies
+ timeout
;
794 if (!test_bit(bit_nr
, &page
->flags
))
796 return __wait_on_bit(page_waitqueue(page
), &wait
,
797 bit_wait_io_timeout
, TASK_KILLABLE
);
799 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
802 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
803 * @page: Page defining the wait queue of interest
804 * @waiter: Waiter to add to the queue
806 * Add an arbitrary @waiter to the wait queue for the nominated @page.
808 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
810 wait_queue_head_t
*q
= page_waitqueue(page
);
813 spin_lock_irqsave(&q
->lock
, flags
);
814 __add_wait_queue(q
, waiter
);
815 spin_unlock_irqrestore(&q
->lock
, flags
);
817 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
820 * unlock_page - unlock a locked page
823 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
824 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
825 * mechanism between PageLocked pages and PageWriteback pages is shared.
826 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
828 * The mb is necessary to enforce ordering between the clear_bit and the read
829 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
831 void unlock_page(struct page
*page
)
833 page
= compound_head(page
);
834 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
835 clear_bit_unlock(PG_locked
, &page
->flags
);
836 smp_mb__after_atomic();
837 wake_up_page(page
, PG_locked
);
839 EXPORT_SYMBOL(unlock_page
);
842 * end_page_writeback - end writeback against a page
845 void end_page_writeback(struct page
*page
)
848 * TestClearPageReclaim could be used here but it is an atomic
849 * operation and overkill in this particular case. Failing to
850 * shuffle a page marked for immediate reclaim is too mild to
851 * justify taking an atomic operation penalty at the end of
852 * ever page writeback.
854 if (PageReclaim(page
)) {
855 ClearPageReclaim(page
);
856 rotate_reclaimable_page(page
);
859 if (!test_clear_page_writeback(page
))
862 smp_mb__after_atomic();
863 wake_up_page(page
, PG_writeback
);
865 EXPORT_SYMBOL(end_page_writeback
);
868 * After completing I/O on a page, call this routine to update the page
869 * flags appropriately
871 void page_endio(struct page
*page
, int rw
, int err
)
875 SetPageUptodate(page
);
877 ClearPageUptodate(page
);
881 } else { /* rw == WRITE */
885 mapping_set_error(page
->mapping
, err
);
887 end_page_writeback(page
);
890 EXPORT_SYMBOL_GPL(page_endio
);
893 * __lock_page - get a lock on the page, assuming we need to sleep to get it
894 * @page: the page to lock
896 void __lock_page(struct page
*page
)
898 struct page
*page_head
= compound_head(page
);
899 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
901 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
902 TASK_UNINTERRUPTIBLE
);
904 EXPORT_SYMBOL(__lock_page
);
906 int __lock_page_killable(struct page
*page
)
908 struct page
*page_head
= compound_head(page
);
909 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
911 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
912 bit_wait_io
, TASK_KILLABLE
);
914 EXPORT_SYMBOL_GPL(__lock_page_killable
);
918 * 1 - page is locked; mmap_sem is still held.
919 * 0 - page is not locked.
920 * mmap_sem has been released (up_read()), unless flags had both
921 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
922 * which case mmap_sem is still held.
924 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
925 * with the page locked and the mmap_sem unperturbed.
927 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
930 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
932 * CAUTION! In this case, mmap_sem is not released
933 * even though return 0.
935 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
938 up_read(&mm
->mmap_sem
);
939 if (flags
& FAULT_FLAG_KILLABLE
)
940 wait_on_page_locked_killable(page
);
942 wait_on_page_locked(page
);
945 if (flags
& FAULT_FLAG_KILLABLE
) {
948 ret
= __lock_page_killable(page
);
950 up_read(&mm
->mmap_sem
);
960 * page_cache_next_hole - find the next hole (not-present entry)
963 * @max_scan: maximum range to search
965 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
966 * lowest indexed hole.
968 * Returns: the index of the hole if found, otherwise returns an index
969 * outside of the set specified (in which case 'return - index >=
970 * max_scan' will be true). In rare cases of index wrap-around, 0 will
973 * page_cache_next_hole may be called under rcu_read_lock. However,
974 * like radix_tree_gang_lookup, this will not atomically search a
975 * snapshot of the tree at a single point in time. For example, if a
976 * hole is created at index 5, then subsequently a hole is created at
977 * index 10, page_cache_next_hole covering both indexes may return 10
978 * if called under rcu_read_lock.
980 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
981 pgoff_t index
, unsigned long max_scan
)
985 for (i
= 0; i
< max_scan
; i
++) {
988 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
989 if (!page
|| radix_tree_exceptional_entry(page
))
998 EXPORT_SYMBOL(page_cache_next_hole
);
1001 * page_cache_prev_hole - find the prev hole (not-present entry)
1004 * @max_scan: maximum range to search
1006 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1009 * Returns: the index of the hole if found, otherwise returns an index
1010 * outside of the set specified (in which case 'index - return >=
1011 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1014 * page_cache_prev_hole may be called under rcu_read_lock. However,
1015 * like radix_tree_gang_lookup, this will not atomically search a
1016 * snapshot of the tree at a single point in time. For example, if a
1017 * hole is created at index 10, then subsequently a hole is created at
1018 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1019 * called under rcu_read_lock.
1021 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1022 pgoff_t index
, unsigned long max_scan
)
1026 for (i
= 0; i
< max_scan
; i
++) {
1029 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1030 if (!page
|| radix_tree_exceptional_entry(page
))
1033 if (index
== ULONG_MAX
)
1039 EXPORT_SYMBOL(page_cache_prev_hole
);
1042 * find_get_entry - find and get a page cache entry
1043 * @mapping: the address_space to search
1044 * @offset: the page cache index
1046 * Looks up the page cache slot at @mapping & @offset. If there is a
1047 * page cache page, it is returned with an increased refcount.
1049 * If the slot holds a shadow entry of a previously evicted page, or a
1050 * swap entry from shmem/tmpfs, it is returned.
1052 * Otherwise, %NULL is returned.
1054 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1062 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1064 page
= radix_tree_deref_slot(pagep
);
1065 if (unlikely(!page
))
1067 if (radix_tree_exception(page
)) {
1068 if (radix_tree_deref_retry(page
))
1071 * A shadow entry of a recently evicted page,
1072 * or a swap entry from shmem/tmpfs. Return
1073 * it without attempting to raise page count.
1077 if (!page_cache_get_speculative(page
))
1081 * Has the page moved?
1082 * This is part of the lockless pagecache protocol. See
1083 * include/linux/pagemap.h for details.
1085 if (unlikely(page
!= *pagep
)) {
1086 page_cache_release(page
);
1095 EXPORT_SYMBOL(find_get_entry
);
1098 * find_lock_entry - locate, pin and lock a page cache entry
1099 * @mapping: the address_space to search
1100 * @offset: the page cache index
1102 * Looks up the page cache slot at @mapping & @offset. If there is a
1103 * page cache page, it is returned locked and with an increased
1106 * If the slot holds a shadow entry of a previously evicted page, or a
1107 * swap entry from shmem/tmpfs, it is returned.
1109 * Otherwise, %NULL is returned.
1111 * find_lock_entry() may sleep.
1113 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1118 page
= find_get_entry(mapping
, offset
);
1119 if (page
&& !radix_tree_exception(page
)) {
1121 /* Has the page been truncated? */
1122 if (unlikely(page
->mapping
!= mapping
)) {
1124 page_cache_release(page
);
1127 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1131 EXPORT_SYMBOL(find_lock_entry
);
1134 * pagecache_get_page - find and get a page reference
1135 * @mapping: the address_space to search
1136 * @offset: the page index
1137 * @fgp_flags: PCG flags
1138 * @gfp_mask: gfp mask to use for the page cache data page allocation
1140 * Looks up the page cache slot at @mapping & @offset.
1142 * PCG flags modify how the page is returned.
1144 * FGP_ACCESSED: the page will be marked accessed
1145 * FGP_LOCK: Page is return locked
1146 * FGP_CREAT: If page is not present then a new page is allocated using
1147 * @gfp_mask and added to the page cache and the VM's LRU
1148 * list. The page is returned locked and with an increased
1149 * refcount. Otherwise, %NULL is returned.
1151 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1152 * if the GFP flags specified for FGP_CREAT are atomic.
1154 * If there is a page cache page, it is returned with an increased refcount.
1156 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1157 int fgp_flags
, gfp_t gfp_mask
)
1162 page
= find_get_entry(mapping
, offset
);
1163 if (radix_tree_exceptional_entry(page
))
1168 if (fgp_flags
& FGP_LOCK
) {
1169 if (fgp_flags
& FGP_NOWAIT
) {
1170 if (!trylock_page(page
)) {
1171 page_cache_release(page
);
1178 /* Has the page been truncated? */
1179 if (unlikely(page
->mapping
!= mapping
)) {
1181 page_cache_release(page
);
1184 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1187 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1188 mark_page_accessed(page
);
1191 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1193 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1194 gfp_mask
|= __GFP_WRITE
;
1195 if (fgp_flags
& FGP_NOFS
)
1196 gfp_mask
&= ~__GFP_FS
;
1198 page
= __page_cache_alloc(gfp_mask
);
1202 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1203 fgp_flags
|= FGP_LOCK
;
1205 /* Init accessed so avoid atomic mark_page_accessed later */
1206 if (fgp_flags
& FGP_ACCESSED
)
1207 __SetPageReferenced(page
);
1209 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1210 gfp_mask
& GFP_RECLAIM_MASK
);
1211 if (unlikely(err
)) {
1212 page_cache_release(page
);
1221 EXPORT_SYMBOL(pagecache_get_page
);
1224 * find_get_entries - gang pagecache lookup
1225 * @mapping: The address_space to search
1226 * @start: The starting page cache index
1227 * @nr_entries: The maximum number of entries
1228 * @entries: Where the resulting entries are placed
1229 * @indices: The cache indices corresponding to the entries in @entries
1231 * find_get_entries() will search for and return a group of up to
1232 * @nr_entries entries in the mapping. The entries are placed at
1233 * @entries. find_get_entries() takes a reference against any actual
1236 * The search returns a group of mapping-contiguous page cache entries
1237 * with ascending indexes. There may be holes in the indices due to
1238 * not-present pages.
1240 * Any shadow entries of evicted pages, or swap entries from
1241 * shmem/tmpfs, are included in the returned array.
1243 * find_get_entries() returns the number of pages and shadow entries
1246 unsigned find_get_entries(struct address_space
*mapping
,
1247 pgoff_t start
, unsigned int nr_entries
,
1248 struct page
**entries
, pgoff_t
*indices
)
1251 unsigned int ret
= 0;
1252 struct radix_tree_iter iter
;
1259 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1262 page
= radix_tree_deref_slot(slot
);
1263 if (unlikely(!page
))
1265 if (radix_tree_exception(page
)) {
1266 if (radix_tree_deref_retry(page
))
1269 * A shadow entry of a recently evicted page, a swap
1270 * entry from shmem/tmpfs or a DAX entry. Return it
1271 * without attempting to raise page count.
1275 if (!page_cache_get_speculative(page
))
1278 /* Has the page moved? */
1279 if (unlikely(page
!= *slot
)) {
1280 page_cache_release(page
);
1284 indices
[ret
] = iter
.index
;
1285 entries
[ret
] = page
;
1286 if (++ret
== nr_entries
)
1294 * find_get_pages - gang pagecache lookup
1295 * @mapping: The address_space to search
1296 * @start: The starting page index
1297 * @nr_pages: The maximum number of pages
1298 * @pages: Where the resulting pages are placed
1300 * find_get_pages() will search for and return a group of up to
1301 * @nr_pages pages in the mapping. The pages are placed at @pages.
1302 * find_get_pages() takes a reference against the returned pages.
1304 * The search returns a group of mapping-contiguous pages with ascending
1305 * indexes. There may be holes in the indices due to not-present pages.
1307 * find_get_pages() returns the number of pages which were found.
1309 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1310 unsigned int nr_pages
, struct page
**pages
)
1312 struct radix_tree_iter iter
;
1316 if (unlikely(!nr_pages
))
1321 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1324 page
= radix_tree_deref_slot(slot
);
1325 if (unlikely(!page
))
1328 if (radix_tree_exception(page
)) {
1329 if (radix_tree_deref_retry(page
)) {
1331 * Transient condition which can only trigger
1332 * when entry at index 0 moves out of or back
1333 * to root: none yet gotten, safe to restart.
1335 WARN_ON(iter
.index
);
1339 * A shadow entry of a recently evicted page,
1340 * or a swap entry from shmem/tmpfs. Skip
1346 if (!page_cache_get_speculative(page
))
1349 /* Has the page moved? */
1350 if (unlikely(page
!= *slot
)) {
1351 page_cache_release(page
);
1356 if (++ret
== nr_pages
)
1365 * find_get_pages_contig - gang contiguous pagecache lookup
1366 * @mapping: The address_space to search
1367 * @index: The starting page index
1368 * @nr_pages: The maximum number of pages
1369 * @pages: Where the resulting pages are placed
1371 * find_get_pages_contig() works exactly like find_get_pages(), except
1372 * that the returned number of pages are guaranteed to be contiguous.
1374 * find_get_pages_contig() returns the number of pages which were found.
1376 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1377 unsigned int nr_pages
, struct page
**pages
)
1379 struct radix_tree_iter iter
;
1381 unsigned int ret
= 0;
1383 if (unlikely(!nr_pages
))
1388 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1391 page
= radix_tree_deref_slot(slot
);
1392 /* The hole, there no reason to continue */
1393 if (unlikely(!page
))
1396 if (radix_tree_exception(page
)) {
1397 if (radix_tree_deref_retry(page
)) {
1399 * Transient condition which can only trigger
1400 * when entry at index 0 moves out of or back
1401 * to root: none yet gotten, safe to restart.
1406 * A shadow entry of a recently evicted page,
1407 * or a swap entry from shmem/tmpfs. Stop
1408 * looking for contiguous pages.
1413 if (!page_cache_get_speculative(page
))
1416 /* Has the page moved? */
1417 if (unlikely(page
!= *slot
)) {
1418 page_cache_release(page
);
1423 * must check mapping and index after taking the ref.
1424 * otherwise we can get both false positives and false
1425 * negatives, which is just confusing to the caller.
1427 if (page
->mapping
== NULL
|| page
->index
!= iter
.index
) {
1428 page_cache_release(page
);
1433 if (++ret
== nr_pages
)
1439 EXPORT_SYMBOL(find_get_pages_contig
);
1442 * find_get_pages_tag - find and return pages that match @tag
1443 * @mapping: the address_space to search
1444 * @index: the starting page index
1445 * @tag: the tag index
1446 * @nr_pages: the maximum number of pages
1447 * @pages: where the resulting pages are placed
1449 * Like find_get_pages, except we only return pages which are tagged with
1450 * @tag. We update @index to index the next page for the traversal.
1452 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1453 int tag
, unsigned int nr_pages
, struct page
**pages
)
1455 struct radix_tree_iter iter
;
1459 if (unlikely(!nr_pages
))
1464 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1465 &iter
, *index
, tag
) {
1468 page
= radix_tree_deref_slot(slot
);
1469 if (unlikely(!page
))
1472 if (radix_tree_exception(page
)) {
1473 if (radix_tree_deref_retry(page
)) {
1475 * Transient condition which can only trigger
1476 * when entry at index 0 moves out of or back
1477 * to root: none yet gotten, safe to restart.
1482 * A shadow entry of a recently evicted page.
1484 * Those entries should never be tagged, but
1485 * this tree walk is lockless and the tags are
1486 * looked up in bulk, one radix tree node at a
1487 * time, so there is a sizable window for page
1488 * reclaim to evict a page we saw tagged.
1495 if (!page_cache_get_speculative(page
))
1498 /* Has the page moved? */
1499 if (unlikely(page
!= *slot
)) {
1500 page_cache_release(page
);
1505 if (++ret
== nr_pages
)
1512 *index
= pages
[ret
- 1]->index
+ 1;
1516 EXPORT_SYMBOL(find_get_pages_tag
);
1519 * find_get_entries_tag - find and return entries that match @tag
1520 * @mapping: the address_space to search
1521 * @start: the starting page cache index
1522 * @tag: the tag index
1523 * @nr_entries: the maximum number of entries
1524 * @entries: where the resulting entries are placed
1525 * @indices: the cache indices corresponding to the entries in @entries
1527 * Like find_get_entries, except we only return entries which are tagged with
1530 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1531 int tag
, unsigned int nr_entries
,
1532 struct page
**entries
, pgoff_t
*indices
)
1535 unsigned int ret
= 0;
1536 struct radix_tree_iter iter
;
1543 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1544 &iter
, start
, tag
) {
1547 page
= radix_tree_deref_slot(slot
);
1548 if (unlikely(!page
))
1550 if (radix_tree_exception(page
)) {
1551 if (radix_tree_deref_retry(page
)) {
1553 * Transient condition which can only trigger
1554 * when entry at index 0 moves out of or back
1555 * to root: none yet gotten, safe to restart.
1561 * A shadow entry of a recently evicted page, a swap
1562 * entry from shmem/tmpfs or a DAX entry. Return it
1563 * without attempting to raise page count.
1567 if (!page_cache_get_speculative(page
))
1570 /* Has the page moved? */
1571 if (unlikely(page
!= *slot
)) {
1572 page_cache_release(page
);
1576 indices
[ret
] = iter
.index
;
1577 entries
[ret
] = page
;
1578 if (++ret
== nr_entries
)
1584 EXPORT_SYMBOL(find_get_entries_tag
);
1587 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1588 * a _large_ part of the i/o request. Imagine the worst scenario:
1590 * ---R__________________________________________B__________
1591 * ^ reading here ^ bad block(assume 4k)
1593 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1594 * => failing the whole request => read(R) => read(R+1) =>
1595 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1596 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1597 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1599 * It is going insane. Fix it by quickly scaling down the readahead size.
1601 static void shrink_readahead_size_eio(struct file
*filp
,
1602 struct file_ra_state
*ra
)
1608 * do_generic_file_read - generic file read routine
1609 * @filp: the file to read
1610 * @ppos: current file position
1611 * @iter: data destination
1612 * @written: already copied
1614 * This is a generic file read routine, and uses the
1615 * mapping->a_ops->readpage() function for the actual low-level stuff.
1617 * This is really ugly. But the goto's actually try to clarify some
1618 * of the logic when it comes to error handling etc.
1620 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1621 struct iov_iter
*iter
, ssize_t written
)
1623 struct address_space
*mapping
= filp
->f_mapping
;
1624 struct inode
*inode
= mapping
->host
;
1625 struct file_ra_state
*ra
= &filp
->f_ra
;
1629 unsigned long offset
; /* offset into pagecache page */
1630 unsigned int prev_offset
;
1633 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1634 prev_index
= ra
->prev_pos
>> PAGE_CACHE_SHIFT
;
1635 prev_offset
= ra
->prev_pos
& (PAGE_CACHE_SIZE
-1);
1636 last_index
= (*ppos
+ iter
->count
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
1637 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1643 unsigned long nr
, ret
;
1647 page
= find_get_page(mapping
, index
);
1649 page_cache_sync_readahead(mapping
,
1651 index
, last_index
- index
);
1652 page
= find_get_page(mapping
, index
);
1653 if (unlikely(page
== NULL
))
1654 goto no_cached_page
;
1656 if (PageReadahead(page
)) {
1657 page_cache_async_readahead(mapping
,
1659 index
, last_index
- index
);
1661 if (!PageUptodate(page
)) {
1663 * See comment in do_read_cache_page on why
1664 * wait_on_page_locked is used to avoid unnecessarily
1665 * serialisations and why it's safe.
1667 wait_on_page_locked_killable(page
);
1668 if (PageUptodate(page
))
1671 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
||
1672 !mapping
->a_ops
->is_partially_uptodate
)
1673 goto page_not_up_to_date
;
1674 if (!trylock_page(page
))
1675 goto page_not_up_to_date
;
1676 /* Did it get truncated before we got the lock? */
1678 goto page_not_up_to_date_locked
;
1679 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1680 offset
, iter
->count
))
1681 goto page_not_up_to_date_locked
;
1686 * i_size must be checked after we know the page is Uptodate.
1688 * Checking i_size after the check allows us to calculate
1689 * the correct value for "nr", which means the zero-filled
1690 * part of the page is not copied back to userspace (unless
1691 * another truncate extends the file - this is desired though).
1694 isize
= i_size_read(inode
);
1695 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1696 if (unlikely(!isize
|| index
> end_index
)) {
1697 page_cache_release(page
);
1701 /* nr is the maximum number of bytes to copy from this page */
1702 nr
= PAGE_CACHE_SIZE
;
1703 if (index
== end_index
) {
1704 nr
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1706 page_cache_release(page
);
1712 /* If users can be writing to this page using arbitrary
1713 * virtual addresses, take care about potential aliasing
1714 * before reading the page on the kernel side.
1716 if (mapping_writably_mapped(mapping
))
1717 flush_dcache_page(page
);
1720 * When a sequential read accesses a page several times,
1721 * only mark it as accessed the first time.
1723 if (prev_index
!= index
|| offset
!= prev_offset
)
1724 mark_page_accessed(page
);
1728 * Ok, we have the page, and it's up-to-date, so
1729 * now we can copy it to user space...
1732 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1734 index
+= offset
>> PAGE_CACHE_SHIFT
;
1735 offset
&= ~PAGE_CACHE_MASK
;
1736 prev_offset
= offset
;
1738 page_cache_release(page
);
1740 if (!iov_iter_count(iter
))
1748 page_not_up_to_date
:
1749 /* Get exclusive access to the page ... */
1750 error
= lock_page_killable(page
);
1751 if (unlikely(error
))
1752 goto readpage_error
;
1754 page_not_up_to_date_locked
:
1755 /* Did it get truncated before we got the lock? */
1756 if (!page
->mapping
) {
1758 page_cache_release(page
);
1762 /* Did somebody else fill it already? */
1763 if (PageUptodate(page
)) {
1770 * A previous I/O error may have been due to temporary
1771 * failures, eg. multipath errors.
1772 * PG_error will be set again if readpage fails.
1774 ClearPageError(page
);
1775 /* Start the actual read. The read will unlock the page. */
1776 error
= mapping
->a_ops
->readpage(filp
, page
);
1778 if (unlikely(error
)) {
1779 if (error
== AOP_TRUNCATED_PAGE
) {
1780 page_cache_release(page
);
1784 goto readpage_error
;
1787 if (!PageUptodate(page
)) {
1788 error
= lock_page_killable(page
);
1789 if (unlikely(error
))
1790 goto readpage_error
;
1791 if (!PageUptodate(page
)) {
1792 if (page
->mapping
== NULL
) {
1794 * invalidate_mapping_pages got it
1797 page_cache_release(page
);
1801 shrink_readahead_size_eio(filp
, ra
);
1803 goto readpage_error
;
1811 /* UHHUH! A synchronous read error occurred. Report it */
1812 page_cache_release(page
);
1817 * Ok, it wasn't cached, so we need to create a new
1820 page
= page_cache_alloc_cold(mapping
);
1825 error
= add_to_page_cache_lru(page
, mapping
, index
,
1826 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1828 page_cache_release(page
);
1829 if (error
== -EEXIST
) {
1839 ra
->prev_pos
= prev_index
;
1840 ra
->prev_pos
<<= PAGE_CACHE_SHIFT
;
1841 ra
->prev_pos
|= prev_offset
;
1843 *ppos
= ((loff_t
)index
<< PAGE_CACHE_SHIFT
) + offset
;
1844 file_accessed(filp
);
1845 return written
? written
: error
;
1849 * generic_file_read_iter - generic filesystem read routine
1850 * @iocb: kernel I/O control block
1851 * @iter: destination for the data read
1853 * This is the "read_iter()" routine for all filesystems
1854 * that can use the page cache directly.
1857 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1859 struct file
*file
= iocb
->ki_filp
;
1861 loff_t
*ppos
= &iocb
->ki_pos
;
1864 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1865 struct address_space
*mapping
= file
->f_mapping
;
1866 struct inode
*inode
= mapping
->host
;
1867 size_t count
= iov_iter_count(iter
);
1871 goto out
; /* skip atime */
1872 size
= i_size_read(inode
);
1873 retval
= filemap_write_and_wait_range(mapping
, pos
,
1876 struct iov_iter data
= *iter
;
1877 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
1881 *ppos
= pos
+ retval
;
1882 iov_iter_advance(iter
, retval
);
1886 * Btrfs can have a short DIO read if we encounter
1887 * compressed extents, so if there was an error, or if
1888 * we've already read everything we wanted to, or if
1889 * there was a short read because we hit EOF, go ahead
1890 * and return. Otherwise fallthrough to buffered io for
1891 * the rest of the read. Buffered reads will not work for
1892 * DAX files, so don't bother trying.
1894 if (retval
< 0 || !iov_iter_count(iter
) || *ppos
>= size
||
1896 file_accessed(file
);
1901 retval
= do_generic_file_read(file
, ppos
, iter
, retval
);
1905 EXPORT_SYMBOL(generic_file_read_iter
);
1909 * page_cache_read - adds requested page to the page cache if not already there
1910 * @file: file to read
1911 * @offset: page index
1912 * @gfp_mask: memory allocation flags
1914 * This adds the requested page to the page cache if it isn't already there,
1915 * and schedules an I/O to read in its contents from disk.
1917 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1919 struct address_space
*mapping
= file
->f_mapping
;
1924 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1928 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1930 ret
= mapping
->a_ops
->readpage(file
, page
);
1931 else if (ret
== -EEXIST
)
1932 ret
= 0; /* losing race to add is OK */
1934 page_cache_release(page
);
1936 } while (ret
== AOP_TRUNCATED_PAGE
);
1941 #define MMAP_LOTSAMISS (100)
1944 * Synchronous readahead happens when we don't even find
1945 * a page in the page cache at all.
1947 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
1948 struct file_ra_state
*ra
,
1952 struct address_space
*mapping
= file
->f_mapping
;
1954 /* If we don't want any read-ahead, don't bother */
1955 if (vma
->vm_flags
& VM_RAND_READ
)
1960 if (vma
->vm_flags
& VM_SEQ_READ
) {
1961 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
1966 /* Avoid banging the cache line if not needed */
1967 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
1971 * Do we miss much more than hit in this file? If so,
1972 * stop bothering with read-ahead. It will only hurt.
1974 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
1980 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
1981 ra
->size
= ra
->ra_pages
;
1982 ra
->async_size
= ra
->ra_pages
/ 4;
1983 ra_submit(ra
, mapping
, file
);
1987 * Asynchronous readahead happens when we find the page and PG_readahead,
1988 * so we want to possibly extend the readahead further..
1990 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
1991 struct file_ra_state
*ra
,
1996 struct address_space
*mapping
= file
->f_mapping
;
1998 /* If we don't want any read-ahead, don't bother */
1999 if (vma
->vm_flags
& VM_RAND_READ
)
2001 if (ra
->mmap_miss
> 0)
2003 if (PageReadahead(page
))
2004 page_cache_async_readahead(mapping
, ra
, file
,
2005 page
, offset
, ra
->ra_pages
);
2009 * filemap_fault - read in file data for page fault handling
2010 * @vma: vma in which the fault was taken
2011 * @vmf: struct vm_fault containing details of the fault
2013 * filemap_fault() is invoked via the vma operations vector for a
2014 * mapped memory region to read in file data during a page fault.
2016 * The goto's are kind of ugly, but this streamlines the normal case of having
2017 * it in the page cache, and handles the special cases reasonably without
2018 * having a lot of duplicated code.
2020 * vma->vm_mm->mmap_sem must be held on entry.
2022 * If our return value has VM_FAULT_RETRY set, it's because
2023 * lock_page_or_retry() returned 0.
2024 * The mmap_sem has usually been released in this case.
2025 * See __lock_page_or_retry() for the exception.
2027 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2028 * has not been released.
2030 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2032 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2035 struct file
*file
= vma
->vm_file
;
2036 struct address_space
*mapping
= file
->f_mapping
;
2037 struct file_ra_state
*ra
= &file
->f_ra
;
2038 struct inode
*inode
= mapping
->host
;
2039 pgoff_t offset
= vmf
->pgoff
;
2044 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2045 if (offset
>= size
>> PAGE_CACHE_SHIFT
)
2046 return VM_FAULT_SIGBUS
;
2049 * Do we have something in the page cache already?
2051 page
= find_get_page(mapping
, offset
);
2052 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2054 * We found the page, so try async readahead before
2055 * waiting for the lock.
2057 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2059 /* No page in the page cache at all */
2060 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2061 count_vm_event(PGMAJFAULT
);
2062 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2063 ret
= VM_FAULT_MAJOR
;
2065 page
= find_get_page(mapping
, offset
);
2067 goto no_cached_page
;
2070 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2071 page_cache_release(page
);
2072 return ret
| VM_FAULT_RETRY
;
2075 /* Did it get truncated? */
2076 if (unlikely(page
->mapping
!= mapping
)) {
2081 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2084 * We have a locked page in the page cache, now we need to check
2085 * that it's up-to-date. If not, it is going to be due to an error.
2087 if (unlikely(!PageUptodate(page
)))
2088 goto page_not_uptodate
;
2091 * Found the page and have a reference on it.
2092 * We must recheck i_size under page lock.
2094 size
= round_up(i_size_read(inode
), PAGE_CACHE_SIZE
);
2095 if (unlikely(offset
>= size
>> PAGE_CACHE_SHIFT
)) {
2097 page_cache_release(page
);
2098 return VM_FAULT_SIGBUS
;
2102 return ret
| VM_FAULT_LOCKED
;
2106 * We're only likely to ever get here if MADV_RANDOM is in
2109 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2112 * The page we want has now been added to the page cache.
2113 * In the unlikely event that someone removed it in the
2114 * meantime, we'll just come back here and read it again.
2120 * An error return from page_cache_read can result if the
2121 * system is low on memory, or a problem occurs while trying
2124 if (error
== -ENOMEM
)
2125 return VM_FAULT_OOM
;
2126 return VM_FAULT_SIGBUS
;
2130 * Umm, take care of errors if the page isn't up-to-date.
2131 * Try to re-read it _once_. We do this synchronously,
2132 * because there really aren't any performance issues here
2133 * and we need to check for errors.
2135 ClearPageError(page
);
2136 error
= mapping
->a_ops
->readpage(file
, page
);
2138 wait_on_page_locked(page
);
2139 if (!PageUptodate(page
))
2142 page_cache_release(page
);
2144 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2147 /* Things didn't work out. Return zero to tell the mm layer so. */
2148 shrink_readahead_size_eio(file
, ra
);
2149 return VM_FAULT_SIGBUS
;
2151 EXPORT_SYMBOL(filemap_fault
);
2153 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2155 struct radix_tree_iter iter
;
2157 struct file
*file
= vma
->vm_file
;
2158 struct address_space
*mapping
= file
->f_mapping
;
2161 unsigned long address
= (unsigned long) vmf
->virtual_address
;
2166 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, vmf
->pgoff
) {
2167 if (iter
.index
> vmf
->max_pgoff
)
2170 page
= radix_tree_deref_slot(slot
);
2171 if (unlikely(!page
))
2173 if (radix_tree_exception(page
)) {
2174 if (radix_tree_deref_retry(page
))
2180 if (!page_cache_get_speculative(page
))
2183 /* Has the page moved? */
2184 if (unlikely(page
!= *slot
)) {
2185 page_cache_release(page
);
2189 if (!PageUptodate(page
) ||
2190 PageReadahead(page
) ||
2193 if (!trylock_page(page
))
2196 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2199 size
= round_up(i_size_read(mapping
->host
), PAGE_CACHE_SIZE
);
2200 if (page
->index
>= size
>> PAGE_CACHE_SHIFT
)
2203 pte
= vmf
->pte
+ page
->index
- vmf
->pgoff
;
2204 if (!pte_none(*pte
))
2207 if (file
->f_ra
.mmap_miss
> 0)
2208 file
->f_ra
.mmap_miss
--;
2209 addr
= address
+ (page
->index
- vmf
->pgoff
) * PAGE_SIZE
;
2210 do_set_pte(vma
, addr
, page
, pte
, false, false);
2216 page_cache_release(page
);
2218 if (iter
.index
== vmf
->max_pgoff
)
2223 EXPORT_SYMBOL(filemap_map_pages
);
2225 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2227 struct page
*page
= vmf
->page
;
2228 struct inode
*inode
= file_inode(vma
->vm_file
);
2229 int ret
= VM_FAULT_LOCKED
;
2231 sb_start_pagefault(inode
->i_sb
);
2232 file_update_time(vma
->vm_file
);
2234 if (page
->mapping
!= inode
->i_mapping
) {
2236 ret
= VM_FAULT_NOPAGE
;
2240 * We mark the page dirty already here so that when freeze is in
2241 * progress, we are guaranteed that writeback during freezing will
2242 * see the dirty page and writeprotect it again.
2244 set_page_dirty(page
);
2245 wait_for_stable_page(page
);
2247 sb_end_pagefault(inode
->i_sb
);
2250 EXPORT_SYMBOL(filemap_page_mkwrite
);
2252 const struct vm_operations_struct generic_file_vm_ops
= {
2253 .fault
= filemap_fault
,
2254 .map_pages
= filemap_map_pages
,
2255 .page_mkwrite
= filemap_page_mkwrite
,
2258 /* This is used for a general mmap of a disk file */
2260 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2262 struct address_space
*mapping
= file
->f_mapping
;
2264 if (!mapping
->a_ops
->readpage
)
2266 file_accessed(file
);
2267 vma
->vm_ops
= &generic_file_vm_ops
;
2272 * This is for filesystems which do not implement ->writepage.
2274 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2276 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2278 return generic_file_mmap(file
, vma
);
2281 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2285 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2289 #endif /* CONFIG_MMU */
2291 EXPORT_SYMBOL(generic_file_mmap
);
2292 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2294 static struct page
*wait_on_page_read(struct page
*page
)
2296 if (!IS_ERR(page
)) {
2297 wait_on_page_locked(page
);
2298 if (!PageUptodate(page
)) {
2299 page_cache_release(page
);
2300 page
= ERR_PTR(-EIO
);
2306 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2308 int (*filler
)(void *, struct page
*),
2315 page
= find_get_page(mapping
, index
);
2317 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2319 return ERR_PTR(-ENOMEM
);
2320 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2321 if (unlikely(err
)) {
2322 page_cache_release(page
);
2325 /* Presumably ENOMEM for radix tree node */
2326 return ERR_PTR(err
);
2330 err
= filler(data
, page
);
2332 page_cache_release(page
);
2333 return ERR_PTR(err
);
2336 page
= wait_on_page_read(page
);
2341 if (PageUptodate(page
))
2345 * Page is not up to date and may be locked due one of the following
2346 * case a: Page is being filled and the page lock is held
2347 * case b: Read/write error clearing the page uptodate status
2348 * case c: Truncation in progress (page locked)
2349 * case d: Reclaim in progress
2351 * Case a, the page will be up to date when the page is unlocked.
2352 * There is no need to serialise on the page lock here as the page
2353 * is pinned so the lock gives no additional protection. Even if the
2354 * the page is truncated, the data is still valid if PageUptodate as
2355 * it's a race vs truncate race.
2356 * Case b, the page will not be up to date
2357 * Case c, the page may be truncated but in itself, the data may still
2358 * be valid after IO completes as it's a read vs truncate race. The
2359 * operation must restart if the page is not uptodate on unlock but
2360 * otherwise serialising on page lock to stabilise the mapping gives
2361 * no additional guarantees to the caller as the page lock is
2362 * released before return.
2363 * Case d, similar to truncation. If reclaim holds the page lock, it
2364 * will be a race with remove_mapping that determines if the mapping
2365 * is valid on unlock but otherwise the data is valid and there is
2366 * no need to serialise with page lock.
2368 * As the page lock gives no additional guarantee, we optimistically
2369 * wait on the page to be unlocked and check if it's up to date and
2370 * use the page if it is. Otherwise, the page lock is required to
2371 * distinguish between the different cases. The motivation is that we
2372 * avoid spurious serialisations and wakeups when multiple processes
2373 * wait on the same page for IO to complete.
2375 wait_on_page_locked(page
);
2376 if (PageUptodate(page
))
2379 /* Distinguish between all the cases under the safety of the lock */
2382 /* Case c or d, restart the operation */
2383 if (!page
->mapping
) {
2385 page_cache_release(page
);
2389 /* Someone else locked and filled the page in a very small window */
2390 if (PageUptodate(page
)) {
2397 mark_page_accessed(page
);
2402 * read_cache_page - read into page cache, fill it if needed
2403 * @mapping: the page's address_space
2404 * @index: the page index
2405 * @filler: function to perform the read
2406 * @data: first arg to filler(data, page) function, often left as NULL
2408 * Read into the page cache. If a page already exists, and PageUptodate() is
2409 * not set, try to fill the page and wait for it to become unlocked.
2411 * If the page does not get brought uptodate, return -EIO.
2413 struct page
*read_cache_page(struct address_space
*mapping
,
2415 int (*filler
)(void *, struct page
*),
2418 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2420 EXPORT_SYMBOL(read_cache_page
);
2423 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2424 * @mapping: the page's address_space
2425 * @index: the page index
2426 * @gfp: the page allocator flags to use if allocating
2428 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2429 * any new page allocations done using the specified allocation flags.
2431 * If the page does not get brought uptodate, return -EIO.
2433 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2437 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2439 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2441 EXPORT_SYMBOL(read_cache_page_gfp
);
2444 * Performs necessary checks before doing a write
2446 * Can adjust writing position or amount of bytes to write.
2447 * Returns appropriate error code that caller should return or
2448 * zero in case that write should be allowed.
2450 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2452 struct file
*file
= iocb
->ki_filp
;
2453 struct inode
*inode
= file
->f_mapping
->host
;
2454 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2457 if (!iov_iter_count(from
))
2460 /* FIXME: this is for backwards compatibility with 2.4 */
2461 if (iocb
->ki_flags
& IOCB_APPEND
)
2462 iocb
->ki_pos
= i_size_read(inode
);
2466 if (limit
!= RLIM_INFINITY
) {
2467 if (iocb
->ki_pos
>= limit
) {
2468 send_sig(SIGXFSZ
, current
, 0);
2471 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2477 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2478 !(file
->f_flags
& O_LARGEFILE
))) {
2479 if (pos
>= MAX_NON_LFS
)
2481 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2485 * Are we about to exceed the fs block limit ?
2487 * If we have written data it becomes a short write. If we have
2488 * exceeded without writing data we send a signal and return EFBIG.
2489 * Linus frestrict idea will clean these up nicely..
2491 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2494 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2495 return iov_iter_count(from
);
2497 EXPORT_SYMBOL(generic_write_checks
);
2499 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2500 loff_t pos
, unsigned len
, unsigned flags
,
2501 struct page
**pagep
, void **fsdata
)
2503 const struct address_space_operations
*aops
= mapping
->a_ops
;
2505 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2508 EXPORT_SYMBOL(pagecache_write_begin
);
2510 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2511 loff_t pos
, unsigned len
, unsigned copied
,
2512 struct page
*page
, void *fsdata
)
2514 const struct address_space_operations
*aops
= mapping
->a_ops
;
2516 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2518 EXPORT_SYMBOL(pagecache_write_end
);
2521 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
, loff_t pos
)
2523 struct file
*file
= iocb
->ki_filp
;
2524 struct address_space
*mapping
= file
->f_mapping
;
2525 struct inode
*inode
= mapping
->host
;
2529 struct iov_iter data
;
2531 write_len
= iov_iter_count(from
);
2532 end
= (pos
+ write_len
- 1) >> PAGE_CACHE_SHIFT
;
2534 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2539 * After a write we want buffered reads to be sure to go to disk to get
2540 * the new data. We invalidate clean cached page from the region we're
2541 * about to write. We do this *before* the write so that we can return
2542 * without clobbering -EIOCBQUEUED from ->direct_IO().
2544 if (mapping
->nrpages
) {
2545 written
= invalidate_inode_pages2_range(mapping
,
2546 pos
>> PAGE_CACHE_SHIFT
, end
);
2548 * If a page can not be invalidated, return 0 to fall back
2549 * to buffered write.
2552 if (written
== -EBUSY
)
2559 written
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
2562 * Finally, try again to invalidate clean pages which might have been
2563 * cached by non-direct readahead, or faulted in by get_user_pages()
2564 * if the source of the write was an mmap'ed region of the file
2565 * we're writing. Either one is a pretty crazy thing to do,
2566 * so we don't support it 100%. If this invalidation
2567 * fails, tough, the write still worked...
2569 if (mapping
->nrpages
) {
2570 invalidate_inode_pages2_range(mapping
,
2571 pos
>> PAGE_CACHE_SHIFT
, end
);
2576 iov_iter_advance(from
, written
);
2577 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2578 i_size_write(inode
, pos
);
2579 mark_inode_dirty(inode
);
2586 EXPORT_SYMBOL(generic_file_direct_write
);
2589 * Find or create a page at the given pagecache position. Return the locked
2590 * page. This function is specifically for buffered writes.
2592 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2593 pgoff_t index
, unsigned flags
)
2596 int fgp_flags
= FGP_LOCK
|FGP_ACCESSED
|FGP_WRITE
|FGP_CREAT
;
2598 if (flags
& AOP_FLAG_NOFS
)
2599 fgp_flags
|= FGP_NOFS
;
2601 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2602 mapping_gfp_mask(mapping
));
2604 wait_for_stable_page(page
);
2608 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2610 ssize_t
generic_perform_write(struct file
*file
,
2611 struct iov_iter
*i
, loff_t pos
)
2613 struct address_space
*mapping
= file
->f_mapping
;
2614 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2616 ssize_t written
= 0;
2617 unsigned int flags
= 0;
2620 * Copies from kernel address space cannot fail (NFSD is a big user).
2622 if (!iter_is_iovec(i
))
2623 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2627 unsigned long offset
; /* Offset into pagecache page */
2628 unsigned long bytes
; /* Bytes to write to page */
2629 size_t copied
; /* Bytes copied from user */
2632 offset
= (pos
& (PAGE_CACHE_SIZE
- 1));
2633 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2638 * Bring in the user page that we will copy from _first_.
2639 * Otherwise there's a nasty deadlock on copying from the
2640 * same page as we're writing to, without it being marked
2643 * Not only is this an optimisation, but it is also required
2644 * to check that the address is actually valid, when atomic
2645 * usercopies are used, below.
2647 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2652 if (fatal_signal_pending(current
)) {
2657 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2659 if (unlikely(status
< 0))
2662 if (mapping_writably_mapped(mapping
))
2663 flush_dcache_page(page
);
2665 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2666 flush_dcache_page(page
);
2668 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2670 if (unlikely(status
< 0))
2676 iov_iter_advance(i
, copied
);
2677 if (unlikely(copied
== 0)) {
2679 * If we were unable to copy any data at all, we must
2680 * fall back to a single segment length write.
2682 * If we didn't fallback here, we could livelock
2683 * because not all segments in the iov can be copied at
2684 * once without a pagefault.
2686 bytes
= min_t(unsigned long, PAGE_CACHE_SIZE
- offset
,
2687 iov_iter_single_seg_count(i
));
2693 balance_dirty_pages_ratelimited(mapping
);
2694 } while (iov_iter_count(i
));
2696 return written
? written
: status
;
2698 EXPORT_SYMBOL(generic_perform_write
);
2701 * __generic_file_write_iter - write data to a file
2702 * @iocb: IO state structure (file, offset, etc.)
2703 * @from: iov_iter with data to write
2705 * This function does all the work needed for actually writing data to a
2706 * file. It does all basic checks, removes SUID from the file, updates
2707 * modification times and calls proper subroutines depending on whether we
2708 * do direct IO or a standard buffered write.
2710 * It expects i_mutex to be grabbed unless we work on a block device or similar
2711 * object which does not need locking at all.
2713 * This function does *not* take care of syncing data in case of O_SYNC write.
2714 * A caller has to handle it. This is mainly due to the fact that we want to
2715 * avoid syncing under i_mutex.
2717 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2719 struct file
*file
= iocb
->ki_filp
;
2720 struct address_space
* mapping
= file
->f_mapping
;
2721 struct inode
*inode
= mapping
->host
;
2722 ssize_t written
= 0;
2726 /* We can write back this queue in page reclaim */
2727 current
->backing_dev_info
= inode_to_bdi(inode
);
2728 err
= file_remove_privs(file
);
2732 err
= file_update_time(file
);
2736 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2737 loff_t pos
, endbyte
;
2739 written
= generic_file_direct_write(iocb
, from
, iocb
->ki_pos
);
2741 * If the write stopped short of completing, fall back to
2742 * buffered writes. Some filesystems do this for writes to
2743 * holes, for example. For DAX files, a buffered write will
2744 * not succeed (even if it did, DAX does not handle dirty
2745 * page-cache pages correctly).
2747 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2750 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2752 * If generic_perform_write() returned a synchronous error
2753 * then we want to return the number of bytes which were
2754 * direct-written, or the error code if that was zero. Note
2755 * that this differs from normal direct-io semantics, which
2756 * will return -EFOO even if some bytes were written.
2758 if (unlikely(status
< 0)) {
2763 * We need to ensure that the page cache pages are written to
2764 * disk and invalidated to preserve the expected O_DIRECT
2767 endbyte
= pos
+ status
- 1;
2768 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2770 iocb
->ki_pos
= endbyte
+ 1;
2772 invalidate_mapping_pages(mapping
,
2773 pos
>> PAGE_CACHE_SHIFT
,
2774 endbyte
>> PAGE_CACHE_SHIFT
);
2777 * We don't know how much we wrote, so just return
2778 * the number of bytes which were direct-written
2782 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2783 if (likely(written
> 0))
2784 iocb
->ki_pos
+= written
;
2787 current
->backing_dev_info
= NULL
;
2788 return written
? written
: err
;
2790 EXPORT_SYMBOL(__generic_file_write_iter
);
2793 * generic_file_write_iter - write data to a file
2794 * @iocb: IO state structure
2795 * @from: iov_iter with data to write
2797 * This is a wrapper around __generic_file_write_iter() to be used by most
2798 * filesystems. It takes care of syncing the file in case of O_SYNC file
2799 * and acquires i_mutex as needed.
2801 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2803 struct file
*file
= iocb
->ki_filp
;
2804 struct inode
*inode
= file
->f_mapping
->host
;
2808 ret
= generic_write_checks(iocb
, from
);
2810 ret
= __generic_file_write_iter(iocb
, from
);
2811 inode_unlock(inode
);
2816 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
2822 EXPORT_SYMBOL(generic_file_write_iter
);
2825 * try_to_release_page() - release old fs-specific metadata on a page
2827 * @page: the page which the kernel is trying to free
2828 * @gfp_mask: memory allocation flags (and I/O mode)
2830 * The address_space is to try to release any data against the page
2831 * (presumably at page->private). If the release was successful, return `1'.
2832 * Otherwise return zero.
2834 * This may also be called if PG_fscache is set on a page, indicating that the
2835 * page is known to the local caching routines.
2837 * The @gfp_mask argument specifies whether I/O may be performed to release
2838 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2841 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2843 struct address_space
* const mapping
= page
->mapping
;
2845 BUG_ON(!PageLocked(page
));
2846 if (PageWriteback(page
))
2849 if (mapping
&& mapping
->a_ops
->releasepage
)
2850 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2851 return try_to_free_buffers(page
);
2854 EXPORT_SYMBOL(try_to_release_page
);