4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
14 #include <linux/dax.h>
16 #include <linux/uaccess.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/cpuset.h>
33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
34 #include <linux/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/rmap.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/filemap.h>
44 * FIXME: remove all knowledge of the buffer layer from the core VM
46 #include <linux/buffer_head.h> /* for try_to_free_buffers */
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
54 * Shared mappings now work. 15.8.1995 Bruno.
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
65 * ->i_mmap_rwsem (truncate_pagecache)
66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
71 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
79 * ->lock_page (access_process_vm)
81 * ->i_mutex (generic_perform_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
85 * sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
99 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
104 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
110 * ->tasklist_lock (memory_failure, collect_procs_ao)
113 static int page_cache_tree_insert(struct address_space
*mapping
,
114 struct page
*page
, void **shadowp
)
116 struct radix_tree_node
*node
;
120 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
127 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
128 if (!radix_tree_exceptional_entry(p
))
131 mapping
->nrexceptional
--;
132 if (!dax_mapping(mapping
)) {
136 workingset_node_shadows_dec(node
);
138 /* DAX can replace empty locked entry with a hole */
140 dax_radix_locked_entry(0, RADIX_DAX_EMPTY
));
141 /* DAX accounts exceptional entries as normal pages */
143 workingset_node_pages_dec(node
);
144 /* Wakeup waiters for exceptional entry lock */
145 dax_wake_mapping_entry_waiter(mapping
, page
->index
, p
,
149 radix_tree_replace_slot(slot
, page
);
152 workingset_node_pages_inc(node
);
154 * Don't track node that contains actual pages.
156 * Avoid acquiring the list_lru lock if already
157 * untracked. The list_empty() test is safe as
158 * node->private_list is protected by
159 * mapping->tree_lock.
161 if (!list_empty(&node
->private_list
))
162 list_lru_del(&workingset_shadow_nodes
,
163 &node
->private_list
);
168 static void page_cache_tree_delete(struct address_space
*mapping
,
169 struct page
*page
, void *shadow
)
171 int i
, nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
173 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
174 VM_BUG_ON_PAGE(PageTail(page
), page
);
175 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
177 for (i
= 0; i
< nr
; i
++) {
178 struct radix_tree_node
*node
;
181 __radix_tree_lookup(&mapping
->page_tree
, page
->index
+ i
,
184 radix_tree_clear_tags(&mapping
->page_tree
, node
, slot
);
187 VM_BUG_ON_PAGE(nr
!= 1, page
);
189 * We need a node to properly account shadow
190 * entries. Don't plant any without. XXX
195 radix_tree_replace_slot(slot
, shadow
);
200 workingset_node_pages_dec(node
);
202 workingset_node_shadows_inc(node
);
204 if (__radix_tree_delete_node(&mapping
->page_tree
, node
))
208 * Track node that only contains shadow entries. DAX mappings
209 * contain no shadow entries and may contain other exceptional
210 * entries so skip those.
212 * Avoid acquiring the list_lru lock if already tracked.
213 * The list_empty() test is safe as node->private_list is
214 * protected by mapping->tree_lock.
216 if (!dax_mapping(mapping
) && !workingset_node_pages(node
) &&
217 list_empty(&node
->private_list
)) {
218 node
->private_data
= mapping
;
219 list_lru_add(&workingset_shadow_nodes
,
220 &node
->private_list
);
225 mapping
->nrexceptional
+= nr
;
227 * Make sure the nrexceptional update is committed before
228 * the nrpages update so that final truncate racing
229 * with reclaim does not see both counters 0 at the
230 * same time and miss a shadow entry.
234 mapping
->nrpages
-= nr
;
238 * Delete a page from the page cache and free it. Caller has to make
239 * sure the page is locked and that nobody else uses it - or that usage
240 * is safe. The caller must hold the mapping's tree_lock.
242 void __delete_from_page_cache(struct page
*page
, void *shadow
)
244 struct address_space
*mapping
= page
->mapping
;
245 int nr
= hpage_nr_pages(page
);
247 trace_mm_filemap_delete_from_page_cache(page
);
249 * if we're uptodate, flush out into the cleancache, otherwise
250 * invalidate any existing cleancache entries. We can't leave
251 * stale data around in the cleancache once our page is gone
253 if (PageUptodate(page
) && PageMappedToDisk(page
))
254 cleancache_put_page(page
);
256 cleancache_invalidate_page(mapping
, page
);
258 VM_BUG_ON_PAGE(PageTail(page
), page
);
259 VM_BUG_ON_PAGE(page_mapped(page
), page
);
260 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
263 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
264 current
->comm
, page_to_pfn(page
));
265 dump_page(page
, "still mapped when deleted");
267 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
269 mapcount
= page_mapcount(page
);
270 if (mapping_exiting(mapping
) &&
271 page_count(page
) >= mapcount
+ 2) {
273 * All vmas have already been torn down, so it's
274 * a good bet that actually the page is unmapped,
275 * and we'd prefer not to leak it: if we're wrong,
276 * some other bad page check should catch it later.
278 page_mapcount_reset(page
);
279 page_ref_sub(page
, mapcount
);
283 page_cache_tree_delete(mapping
, page
, shadow
);
285 page
->mapping
= NULL
;
286 /* Leave page->index set: truncation lookup relies upon it */
288 /* hugetlb pages do not participate in page cache accounting. */
290 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
291 if (PageSwapBacked(page
)) {
292 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
293 if (PageTransHuge(page
))
294 __dec_node_page_state(page
, NR_SHMEM_THPS
);
296 VM_BUG_ON_PAGE(PageTransHuge(page
) && !PageHuge(page
), page
);
300 * At this point page must be either written or cleaned by truncate.
301 * Dirty page here signals a bug and loss of unwritten data.
303 * This fixes dirty accounting after removing the page entirely but
304 * leaves PageDirty set: it has no effect for truncated page and
305 * anyway will be cleared before returning page into buddy allocator.
307 if (WARN_ON_ONCE(PageDirty(page
)))
308 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
312 * delete_from_page_cache - delete page from page cache
313 * @page: the page which the kernel is trying to remove from page cache
315 * This must be called only on pages that have been verified to be in the page
316 * cache and locked. It will never put the page into the free list, the caller
317 * has a reference on the page.
319 void delete_from_page_cache(struct page
*page
)
321 struct address_space
*mapping
= page_mapping(page
);
323 void (*freepage
)(struct page
*);
325 BUG_ON(!PageLocked(page
));
327 freepage
= mapping
->a_ops
->freepage
;
329 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
330 __delete_from_page_cache(page
, NULL
);
331 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
336 if (PageTransHuge(page
) && !PageHuge(page
)) {
337 page_ref_sub(page
, HPAGE_PMD_NR
);
338 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
343 EXPORT_SYMBOL(delete_from_page_cache
);
345 int filemap_check_errors(struct address_space
*mapping
)
348 /* Check for outstanding write errors */
349 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
350 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
352 if (test_bit(AS_EIO
, &mapping
->flags
) &&
353 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
357 EXPORT_SYMBOL(filemap_check_errors
);
360 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
361 * @mapping: address space structure to write
362 * @start: offset in bytes where the range starts
363 * @end: offset in bytes where the range ends (inclusive)
364 * @sync_mode: enable synchronous operation
366 * Start writeback against all of a mapping's dirty pages that lie
367 * within the byte offsets <start, end> inclusive.
369 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
370 * opposed to a regular memory cleansing writeback. The difference between
371 * these two operations is that if a dirty page/buffer is encountered, it must
372 * be waited upon, and not just skipped over.
374 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
375 loff_t end
, int sync_mode
)
378 struct writeback_control wbc
= {
379 .sync_mode
= sync_mode
,
380 .nr_to_write
= LONG_MAX
,
381 .range_start
= start
,
385 if (!mapping_cap_writeback_dirty(mapping
))
388 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
389 ret
= do_writepages(mapping
, &wbc
);
390 wbc_detach_inode(&wbc
);
394 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
397 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
400 int filemap_fdatawrite(struct address_space
*mapping
)
402 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
404 EXPORT_SYMBOL(filemap_fdatawrite
);
406 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
409 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
411 EXPORT_SYMBOL(filemap_fdatawrite_range
);
414 * filemap_flush - mostly a non-blocking flush
415 * @mapping: target address_space
417 * This is a mostly non-blocking flush. Not suitable for data-integrity
418 * purposes - I/O may not be started against all dirty pages.
420 int filemap_flush(struct address_space
*mapping
)
422 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
424 EXPORT_SYMBOL(filemap_flush
);
426 static int __filemap_fdatawait_range(struct address_space
*mapping
,
427 loff_t start_byte
, loff_t end_byte
)
429 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
430 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
435 if (end_byte
< start_byte
)
438 pagevec_init(&pvec
, 0);
439 while ((index
<= end
) &&
440 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
441 PAGECACHE_TAG_WRITEBACK
,
442 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
445 for (i
= 0; i
< nr_pages
; i
++) {
446 struct page
*page
= pvec
.pages
[i
];
448 /* until radix tree lookup accepts end_index */
449 if (page
->index
> end
)
452 wait_on_page_writeback(page
);
453 if (TestClearPageError(page
))
456 pagevec_release(&pvec
);
464 * filemap_fdatawait_range - wait for writeback to complete
465 * @mapping: address space structure to wait for
466 * @start_byte: offset in bytes where the range starts
467 * @end_byte: offset in bytes where the range ends (inclusive)
469 * Walk the list of under-writeback pages of the given address space
470 * in the given range and wait for all of them. Check error status of
471 * the address space and return it.
473 * Since the error status of the address space is cleared by this function,
474 * callers are responsible for checking the return value and handling and/or
475 * reporting the error.
477 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
482 ret
= __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
483 ret2
= filemap_check_errors(mapping
);
489 EXPORT_SYMBOL(filemap_fdatawait_range
);
492 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
493 * @mapping: address space structure to wait for
495 * Walk the list of under-writeback pages of the given address space
496 * and wait for all of them. Unlike filemap_fdatawait(), this function
497 * does not clear error status of the address space.
499 * Use this function if callers don't handle errors themselves. Expected
500 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
503 void filemap_fdatawait_keep_errors(struct address_space
*mapping
)
505 loff_t i_size
= i_size_read(mapping
->host
);
510 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
514 * filemap_fdatawait - wait for all under-writeback pages to complete
515 * @mapping: address space structure to wait for
517 * Walk the list of under-writeback pages of the given address space
518 * and wait for all of them. Check error status of the address space
521 * Since the error status of the address space is cleared by this function,
522 * callers are responsible for checking the return value and handling and/or
523 * reporting the error.
525 int filemap_fdatawait(struct address_space
*mapping
)
527 loff_t i_size
= i_size_read(mapping
->host
);
532 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
534 EXPORT_SYMBOL(filemap_fdatawait
);
536 int filemap_write_and_wait(struct address_space
*mapping
)
540 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
541 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
542 err
= filemap_fdatawrite(mapping
);
544 * Even if the above returned error, the pages may be
545 * written partially (e.g. -ENOSPC), so we wait for it.
546 * But the -EIO is special case, it may indicate the worst
547 * thing (e.g. bug) happened, so we avoid waiting for it.
550 int err2
= filemap_fdatawait(mapping
);
555 err
= filemap_check_errors(mapping
);
559 EXPORT_SYMBOL(filemap_write_and_wait
);
562 * filemap_write_and_wait_range - write out & wait on a file range
563 * @mapping: the address_space for the pages
564 * @lstart: offset in bytes where the range starts
565 * @lend: offset in bytes where the range ends (inclusive)
567 * Write out and wait upon file offsets lstart->lend, inclusive.
569 * Note that `lend' is inclusive (describes the last byte to be written) so
570 * that this function can be used to write to the very end-of-file (end = -1).
572 int filemap_write_and_wait_range(struct address_space
*mapping
,
573 loff_t lstart
, loff_t lend
)
577 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
578 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
579 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
581 /* See comment of filemap_write_and_wait() */
583 int err2
= filemap_fdatawait_range(mapping
,
589 err
= filemap_check_errors(mapping
);
593 EXPORT_SYMBOL(filemap_write_and_wait_range
);
596 * replace_page_cache_page - replace a pagecache page with a new one
597 * @old: page to be replaced
598 * @new: page to replace with
599 * @gfp_mask: allocation mode
601 * This function replaces a page in the pagecache with a new one. On
602 * success it acquires the pagecache reference for the new page and
603 * drops it for the old page. Both the old and new pages must be
604 * locked. This function does not add the new page to the LRU, the
605 * caller must do that.
607 * The remove + add is atomic. The only way this function can fail is
608 * memory allocation failure.
610 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
614 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
615 VM_BUG_ON_PAGE(!PageLocked(new), new);
616 VM_BUG_ON_PAGE(new->mapping
, new);
618 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
620 struct address_space
*mapping
= old
->mapping
;
621 void (*freepage
)(struct page
*);
624 pgoff_t offset
= old
->index
;
625 freepage
= mapping
->a_ops
->freepage
;
628 new->mapping
= mapping
;
631 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
632 __delete_from_page_cache(old
, NULL
);
633 error
= page_cache_tree_insert(mapping
, new, NULL
);
637 * hugetlb pages do not participate in page cache accounting.
640 __inc_node_page_state(new, NR_FILE_PAGES
);
641 if (PageSwapBacked(new))
642 __inc_node_page_state(new, NR_SHMEM
);
643 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
644 mem_cgroup_migrate(old
, new);
645 radix_tree_preload_end();
653 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
655 static int __add_to_page_cache_locked(struct page
*page
,
656 struct address_space
*mapping
,
657 pgoff_t offset
, gfp_t gfp_mask
,
660 int huge
= PageHuge(page
);
661 struct mem_cgroup
*memcg
;
664 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
665 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
668 error
= mem_cgroup_try_charge(page
, current
->mm
,
669 gfp_mask
, &memcg
, false);
674 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
677 mem_cgroup_cancel_charge(page
, memcg
, false);
682 page
->mapping
= mapping
;
683 page
->index
= offset
;
685 spin_lock_irq(&mapping
->tree_lock
);
686 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
687 radix_tree_preload_end();
691 /* hugetlb pages do not participate in page cache accounting. */
693 __inc_node_page_state(page
, NR_FILE_PAGES
);
694 spin_unlock_irq(&mapping
->tree_lock
);
696 mem_cgroup_commit_charge(page
, memcg
, false, false);
697 trace_mm_filemap_add_to_page_cache(page
);
700 page
->mapping
= NULL
;
701 /* Leave page->index set: truncation relies upon it */
702 spin_unlock_irq(&mapping
->tree_lock
);
704 mem_cgroup_cancel_charge(page
, memcg
, false);
710 * add_to_page_cache_locked - add a locked page to the pagecache
712 * @mapping: the page's address_space
713 * @offset: page index
714 * @gfp_mask: page allocation mode
716 * This function is used to add a page to the pagecache. It must be locked.
717 * This function does not add the page to the LRU. The caller must do that.
719 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
720 pgoff_t offset
, gfp_t gfp_mask
)
722 return __add_to_page_cache_locked(page
, mapping
, offset
,
725 EXPORT_SYMBOL(add_to_page_cache_locked
);
727 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
728 pgoff_t offset
, gfp_t gfp_mask
)
733 __SetPageLocked(page
);
734 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
737 __ClearPageLocked(page
);
740 * The page might have been evicted from cache only
741 * recently, in which case it should be activated like
742 * any other repeatedly accessed page.
743 * The exception is pages getting rewritten; evicting other
744 * data from the working set, only to cache data that will
745 * get overwritten with something else, is a waste of memory.
747 if (!(gfp_mask
& __GFP_WRITE
) &&
748 shadow
&& workingset_refault(shadow
)) {
750 workingset_activation(page
);
752 ClearPageActive(page
);
757 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
760 struct page
*__page_cache_alloc(gfp_t gfp
)
765 if (cpuset_do_page_mem_spread()) {
766 unsigned int cpuset_mems_cookie
;
768 cpuset_mems_cookie
= read_mems_allowed_begin();
769 n
= cpuset_mem_spread_node();
770 page
= __alloc_pages_node(n
, gfp
, 0);
771 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
775 return alloc_pages(gfp
, 0);
777 EXPORT_SYMBOL(__page_cache_alloc
);
781 * In order to wait for pages to become available there must be
782 * waitqueues associated with pages. By using a hash table of
783 * waitqueues where the bucket discipline is to maintain all
784 * waiters on the same queue and wake all when any of the pages
785 * become available, and for the woken contexts to check to be
786 * sure the appropriate page became available, this saves space
787 * at a cost of "thundering herd" phenomena during rare hash
790 wait_queue_head_t
*page_waitqueue(struct page
*page
)
792 const struct zone
*zone
= page_zone(page
);
794 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
796 EXPORT_SYMBOL(page_waitqueue
);
798 void wait_on_page_bit(struct page
*page
, int bit_nr
)
800 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
802 if (test_bit(bit_nr
, &page
->flags
))
803 __wait_on_bit(page_waitqueue(page
), &wait
, bit_wait_io
,
804 TASK_UNINTERRUPTIBLE
);
806 EXPORT_SYMBOL(wait_on_page_bit
);
808 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
810 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
812 if (!test_bit(bit_nr
, &page
->flags
))
815 return __wait_on_bit(page_waitqueue(page
), &wait
,
816 bit_wait_io
, TASK_KILLABLE
);
819 int wait_on_page_bit_killable_timeout(struct page
*page
,
820 int bit_nr
, unsigned long timeout
)
822 DEFINE_WAIT_BIT(wait
, &page
->flags
, bit_nr
);
824 wait
.key
.timeout
= jiffies
+ timeout
;
825 if (!test_bit(bit_nr
, &page
->flags
))
827 return __wait_on_bit(page_waitqueue(page
), &wait
,
828 bit_wait_io_timeout
, TASK_KILLABLE
);
830 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout
);
833 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
834 * @page: Page defining the wait queue of interest
835 * @waiter: Waiter to add to the queue
837 * Add an arbitrary @waiter to the wait queue for the nominated @page.
839 void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
)
841 wait_queue_head_t
*q
= page_waitqueue(page
);
844 spin_lock_irqsave(&q
->lock
, flags
);
845 __add_wait_queue(q
, waiter
);
846 spin_unlock_irqrestore(&q
->lock
, flags
);
848 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
851 * unlock_page - unlock a locked page
854 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
855 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
856 * mechanism between PageLocked pages and PageWriteback pages is shared.
857 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
859 * The mb is necessary to enforce ordering between the clear_bit and the read
860 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
862 void unlock_page(struct page
*page
)
864 page
= compound_head(page
);
865 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
866 clear_bit_unlock(PG_locked
, &page
->flags
);
867 smp_mb__after_atomic();
868 wake_up_page(page
, PG_locked
);
870 EXPORT_SYMBOL(unlock_page
);
873 * end_page_writeback - end writeback against a page
876 void end_page_writeback(struct page
*page
)
879 * TestClearPageReclaim could be used here but it is an atomic
880 * operation and overkill in this particular case. Failing to
881 * shuffle a page marked for immediate reclaim is too mild to
882 * justify taking an atomic operation penalty at the end of
883 * ever page writeback.
885 if (PageReclaim(page
)) {
886 ClearPageReclaim(page
);
887 rotate_reclaimable_page(page
);
890 if (!test_clear_page_writeback(page
))
893 smp_mb__after_atomic();
894 wake_up_page(page
, PG_writeback
);
896 EXPORT_SYMBOL(end_page_writeback
);
899 * After completing I/O on a page, call this routine to update the page
900 * flags appropriately
902 void page_endio(struct page
*page
, bool is_write
, int err
)
906 SetPageUptodate(page
);
908 ClearPageUptodate(page
);
916 mapping_set_error(page
->mapping
, err
);
918 end_page_writeback(page
);
921 EXPORT_SYMBOL_GPL(page_endio
);
924 * __lock_page - get a lock on the page, assuming we need to sleep to get it
925 * @page: the page to lock
927 void __lock_page(struct page
*page
)
929 struct page
*page_head
= compound_head(page
);
930 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
932 __wait_on_bit_lock(page_waitqueue(page_head
), &wait
, bit_wait_io
,
933 TASK_UNINTERRUPTIBLE
);
935 EXPORT_SYMBOL(__lock_page
);
937 int __lock_page_killable(struct page
*page
)
939 struct page
*page_head
= compound_head(page
);
940 DEFINE_WAIT_BIT(wait
, &page_head
->flags
, PG_locked
);
942 return __wait_on_bit_lock(page_waitqueue(page_head
), &wait
,
943 bit_wait_io
, TASK_KILLABLE
);
945 EXPORT_SYMBOL_GPL(__lock_page_killable
);
949 * 1 - page is locked; mmap_sem is still held.
950 * 0 - page is not locked.
951 * mmap_sem has been released (up_read()), unless flags had both
952 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
953 * which case mmap_sem is still held.
955 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
956 * with the page locked and the mmap_sem unperturbed.
958 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
961 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
963 * CAUTION! In this case, mmap_sem is not released
964 * even though return 0.
966 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
969 up_read(&mm
->mmap_sem
);
970 if (flags
& FAULT_FLAG_KILLABLE
)
971 wait_on_page_locked_killable(page
);
973 wait_on_page_locked(page
);
976 if (flags
& FAULT_FLAG_KILLABLE
) {
979 ret
= __lock_page_killable(page
);
981 up_read(&mm
->mmap_sem
);
991 * page_cache_next_hole - find the next hole (not-present entry)
994 * @max_scan: maximum range to search
996 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
997 * lowest indexed hole.
999 * Returns: the index of the hole if found, otherwise returns an index
1000 * outside of the set specified (in which case 'return - index >=
1001 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1004 * page_cache_next_hole may be called under rcu_read_lock. However,
1005 * like radix_tree_gang_lookup, this will not atomically search a
1006 * snapshot of the tree at a single point in time. For example, if a
1007 * hole is created at index 5, then subsequently a hole is created at
1008 * index 10, page_cache_next_hole covering both indexes may return 10
1009 * if called under rcu_read_lock.
1011 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1012 pgoff_t index
, unsigned long max_scan
)
1016 for (i
= 0; i
< max_scan
; i
++) {
1019 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1020 if (!page
|| radix_tree_exceptional_entry(page
))
1029 EXPORT_SYMBOL(page_cache_next_hole
);
1032 * page_cache_prev_hole - find the prev hole (not-present entry)
1035 * @max_scan: maximum range to search
1037 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1040 * Returns: the index of the hole if found, otherwise returns an index
1041 * outside of the set specified (in which case 'index - return >=
1042 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1045 * page_cache_prev_hole may be called under rcu_read_lock. However,
1046 * like radix_tree_gang_lookup, this will not atomically search a
1047 * snapshot of the tree at a single point in time. For example, if a
1048 * hole is created at index 10, then subsequently a hole is created at
1049 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1050 * called under rcu_read_lock.
1052 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1053 pgoff_t index
, unsigned long max_scan
)
1057 for (i
= 0; i
< max_scan
; i
++) {
1060 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1061 if (!page
|| radix_tree_exceptional_entry(page
))
1064 if (index
== ULONG_MAX
)
1070 EXPORT_SYMBOL(page_cache_prev_hole
);
1073 * find_get_entry - find and get a page cache entry
1074 * @mapping: the address_space to search
1075 * @offset: the page cache index
1077 * Looks up the page cache slot at @mapping & @offset. If there is a
1078 * page cache page, it is returned with an increased refcount.
1080 * If the slot holds a shadow entry of a previously evicted page, or a
1081 * swap entry from shmem/tmpfs, it is returned.
1083 * Otherwise, %NULL is returned.
1085 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1088 struct page
*head
, *page
;
1093 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1095 page
= radix_tree_deref_slot(pagep
);
1096 if (unlikely(!page
))
1098 if (radix_tree_exception(page
)) {
1099 if (radix_tree_deref_retry(page
))
1102 * A shadow entry of a recently evicted page,
1103 * or a swap entry from shmem/tmpfs. Return
1104 * it without attempting to raise page count.
1109 head
= compound_head(page
);
1110 if (!page_cache_get_speculative(head
))
1113 /* The page was split under us? */
1114 if (compound_head(page
) != head
) {
1120 * Has the page moved?
1121 * This is part of the lockless pagecache protocol. See
1122 * include/linux/pagemap.h for details.
1124 if (unlikely(page
!= *pagep
)) {
1134 EXPORT_SYMBOL(find_get_entry
);
1137 * find_lock_entry - locate, pin and lock a page cache entry
1138 * @mapping: the address_space to search
1139 * @offset: the page cache index
1141 * Looks up the page cache slot at @mapping & @offset. If there is a
1142 * page cache page, it is returned locked and with an increased
1145 * If the slot holds a shadow entry of a previously evicted page, or a
1146 * swap entry from shmem/tmpfs, it is returned.
1148 * Otherwise, %NULL is returned.
1150 * find_lock_entry() may sleep.
1152 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1157 page
= find_get_entry(mapping
, offset
);
1158 if (page
&& !radix_tree_exception(page
)) {
1160 /* Has the page been truncated? */
1161 if (unlikely(page_mapping(page
) != mapping
)) {
1166 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1170 EXPORT_SYMBOL(find_lock_entry
);
1173 * pagecache_get_page - find and get a page reference
1174 * @mapping: the address_space to search
1175 * @offset: the page index
1176 * @fgp_flags: PCG flags
1177 * @gfp_mask: gfp mask to use for the page cache data page allocation
1179 * Looks up the page cache slot at @mapping & @offset.
1181 * PCG flags modify how the page is returned.
1183 * FGP_ACCESSED: the page will be marked accessed
1184 * FGP_LOCK: Page is return locked
1185 * FGP_CREAT: If page is not present then a new page is allocated using
1186 * @gfp_mask and added to the page cache and the VM's LRU
1187 * list. The page is returned locked and with an increased
1188 * refcount. Otherwise, %NULL is returned.
1190 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1191 * if the GFP flags specified for FGP_CREAT are atomic.
1193 * If there is a page cache page, it is returned with an increased refcount.
1195 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1196 int fgp_flags
, gfp_t gfp_mask
)
1201 page
= find_get_entry(mapping
, offset
);
1202 if (radix_tree_exceptional_entry(page
))
1207 if (fgp_flags
& FGP_LOCK
) {
1208 if (fgp_flags
& FGP_NOWAIT
) {
1209 if (!trylock_page(page
)) {
1217 /* Has the page been truncated? */
1218 if (unlikely(page
->mapping
!= mapping
)) {
1223 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1226 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1227 mark_page_accessed(page
);
1230 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1232 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1233 gfp_mask
|= __GFP_WRITE
;
1234 if (fgp_flags
& FGP_NOFS
)
1235 gfp_mask
&= ~__GFP_FS
;
1237 page
= __page_cache_alloc(gfp_mask
);
1241 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1242 fgp_flags
|= FGP_LOCK
;
1244 /* Init accessed so avoid atomic mark_page_accessed later */
1245 if (fgp_flags
& FGP_ACCESSED
)
1246 __SetPageReferenced(page
);
1248 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1249 gfp_mask
& GFP_RECLAIM_MASK
);
1250 if (unlikely(err
)) {
1260 EXPORT_SYMBOL(pagecache_get_page
);
1263 * find_get_entries - gang pagecache lookup
1264 * @mapping: The address_space to search
1265 * @start: The starting page cache index
1266 * @nr_entries: The maximum number of entries
1267 * @entries: Where the resulting entries are placed
1268 * @indices: The cache indices corresponding to the entries in @entries
1270 * find_get_entries() will search for and return a group of up to
1271 * @nr_entries entries in the mapping. The entries are placed at
1272 * @entries. find_get_entries() takes a reference against any actual
1275 * The search returns a group of mapping-contiguous page cache entries
1276 * with ascending indexes. There may be holes in the indices due to
1277 * not-present pages.
1279 * Any shadow entries of evicted pages, or swap entries from
1280 * shmem/tmpfs, are included in the returned array.
1282 * find_get_entries() returns the number of pages and shadow entries
1285 unsigned find_get_entries(struct address_space
*mapping
,
1286 pgoff_t start
, unsigned int nr_entries
,
1287 struct page
**entries
, pgoff_t
*indices
)
1290 unsigned int ret
= 0;
1291 struct radix_tree_iter iter
;
1297 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1298 struct page
*head
, *page
;
1300 page
= radix_tree_deref_slot(slot
);
1301 if (unlikely(!page
))
1303 if (radix_tree_exception(page
)) {
1304 if (radix_tree_deref_retry(page
)) {
1305 slot
= radix_tree_iter_retry(&iter
);
1309 * A shadow entry of a recently evicted page, a swap
1310 * entry from shmem/tmpfs or a DAX entry. Return it
1311 * without attempting to raise page count.
1316 head
= compound_head(page
);
1317 if (!page_cache_get_speculative(head
))
1320 /* The page was split under us? */
1321 if (compound_head(page
) != head
) {
1326 /* Has the page moved? */
1327 if (unlikely(page
!= *slot
)) {
1332 indices
[ret
] = iter
.index
;
1333 entries
[ret
] = page
;
1334 if (++ret
== nr_entries
)
1342 * find_get_pages - gang pagecache lookup
1343 * @mapping: The address_space to search
1344 * @start: The starting page index
1345 * @nr_pages: The maximum number of pages
1346 * @pages: Where the resulting pages are placed
1348 * find_get_pages() will search for and return a group of up to
1349 * @nr_pages pages in the mapping. The pages are placed at @pages.
1350 * find_get_pages() takes a reference against the returned pages.
1352 * The search returns a group of mapping-contiguous pages with ascending
1353 * indexes. There may be holes in the indices due to not-present pages.
1355 * find_get_pages() returns the number of pages which were found.
1357 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1358 unsigned int nr_pages
, struct page
**pages
)
1360 struct radix_tree_iter iter
;
1364 if (unlikely(!nr_pages
))
1368 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1369 struct page
*head
, *page
;
1371 page
= radix_tree_deref_slot(slot
);
1372 if (unlikely(!page
))
1375 if (radix_tree_exception(page
)) {
1376 if (radix_tree_deref_retry(page
)) {
1377 slot
= radix_tree_iter_retry(&iter
);
1381 * A shadow entry of a recently evicted page,
1382 * or a swap entry from shmem/tmpfs. Skip
1388 head
= compound_head(page
);
1389 if (!page_cache_get_speculative(head
))
1392 /* The page was split under us? */
1393 if (compound_head(page
) != head
) {
1398 /* Has the page moved? */
1399 if (unlikely(page
!= *slot
)) {
1405 if (++ret
== nr_pages
)
1414 * find_get_pages_contig - gang contiguous pagecache lookup
1415 * @mapping: The address_space to search
1416 * @index: The starting page index
1417 * @nr_pages: The maximum number of pages
1418 * @pages: Where the resulting pages are placed
1420 * find_get_pages_contig() works exactly like find_get_pages(), except
1421 * that the returned number of pages are guaranteed to be contiguous.
1423 * find_get_pages_contig() returns the number of pages which were found.
1425 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1426 unsigned int nr_pages
, struct page
**pages
)
1428 struct radix_tree_iter iter
;
1430 unsigned int ret
= 0;
1432 if (unlikely(!nr_pages
))
1436 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1437 struct page
*head
, *page
;
1439 page
= radix_tree_deref_slot(slot
);
1440 /* The hole, there no reason to continue */
1441 if (unlikely(!page
))
1444 if (radix_tree_exception(page
)) {
1445 if (radix_tree_deref_retry(page
)) {
1446 slot
= radix_tree_iter_retry(&iter
);
1450 * A shadow entry of a recently evicted page,
1451 * or a swap entry from shmem/tmpfs. Stop
1452 * looking for contiguous pages.
1457 head
= compound_head(page
);
1458 if (!page_cache_get_speculative(head
))
1461 /* The page was split under us? */
1462 if (compound_head(page
) != head
) {
1467 /* Has the page moved? */
1468 if (unlikely(page
!= *slot
)) {
1474 * must check mapping and index after taking the ref.
1475 * otherwise we can get both false positives and false
1476 * negatives, which is just confusing to the caller.
1478 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1484 if (++ret
== nr_pages
)
1490 EXPORT_SYMBOL(find_get_pages_contig
);
1493 * find_get_pages_tag - find and return pages that match @tag
1494 * @mapping: the address_space to search
1495 * @index: the starting page index
1496 * @tag: the tag index
1497 * @nr_pages: the maximum number of pages
1498 * @pages: where the resulting pages are placed
1500 * Like find_get_pages, except we only return pages which are tagged with
1501 * @tag. We update @index to index the next page for the traversal.
1503 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1504 int tag
, unsigned int nr_pages
, struct page
**pages
)
1506 struct radix_tree_iter iter
;
1510 if (unlikely(!nr_pages
))
1514 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1515 &iter
, *index
, tag
) {
1516 struct page
*head
, *page
;
1518 page
= radix_tree_deref_slot(slot
);
1519 if (unlikely(!page
))
1522 if (radix_tree_exception(page
)) {
1523 if (radix_tree_deref_retry(page
)) {
1524 slot
= radix_tree_iter_retry(&iter
);
1528 * A shadow entry of a recently evicted page.
1530 * Those entries should never be tagged, but
1531 * this tree walk is lockless and the tags are
1532 * looked up in bulk, one radix tree node at a
1533 * time, so there is a sizable window for page
1534 * reclaim to evict a page we saw tagged.
1541 head
= compound_head(page
);
1542 if (!page_cache_get_speculative(head
))
1545 /* The page was split under us? */
1546 if (compound_head(page
) != head
) {
1551 /* Has the page moved? */
1552 if (unlikely(page
!= *slot
)) {
1558 if (++ret
== nr_pages
)
1565 *index
= pages
[ret
- 1]->index
+ 1;
1569 EXPORT_SYMBOL(find_get_pages_tag
);
1572 * find_get_entries_tag - find and return entries that match @tag
1573 * @mapping: the address_space to search
1574 * @start: the starting page cache index
1575 * @tag: the tag index
1576 * @nr_entries: the maximum number of entries
1577 * @entries: where the resulting entries are placed
1578 * @indices: the cache indices corresponding to the entries in @entries
1580 * Like find_get_entries, except we only return entries which are tagged with
1583 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1584 int tag
, unsigned int nr_entries
,
1585 struct page
**entries
, pgoff_t
*indices
)
1588 unsigned int ret
= 0;
1589 struct radix_tree_iter iter
;
1595 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1596 &iter
, start
, tag
) {
1597 struct page
*head
, *page
;
1599 page
= radix_tree_deref_slot(slot
);
1600 if (unlikely(!page
))
1602 if (radix_tree_exception(page
)) {
1603 if (radix_tree_deref_retry(page
)) {
1604 slot
= radix_tree_iter_retry(&iter
);
1609 * A shadow entry of a recently evicted page, a swap
1610 * entry from shmem/tmpfs or a DAX entry. Return it
1611 * without attempting to raise page count.
1616 head
= compound_head(page
);
1617 if (!page_cache_get_speculative(head
))
1620 /* The page was split under us? */
1621 if (compound_head(page
) != head
) {
1626 /* Has the page moved? */
1627 if (unlikely(page
!= *slot
)) {
1632 indices
[ret
] = iter
.index
;
1633 entries
[ret
] = page
;
1634 if (++ret
== nr_entries
)
1640 EXPORT_SYMBOL(find_get_entries_tag
);
1643 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1644 * a _large_ part of the i/o request. Imagine the worst scenario:
1646 * ---R__________________________________________B__________
1647 * ^ reading here ^ bad block(assume 4k)
1649 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1650 * => failing the whole request => read(R) => read(R+1) =>
1651 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1652 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1653 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1655 * It is going insane. Fix it by quickly scaling down the readahead size.
1657 static void shrink_readahead_size_eio(struct file
*filp
,
1658 struct file_ra_state
*ra
)
1664 * do_generic_file_read - generic file read routine
1665 * @filp: the file to read
1666 * @ppos: current file position
1667 * @iter: data destination
1668 * @written: already copied
1670 * This is a generic file read routine, and uses the
1671 * mapping->a_ops->readpage() function for the actual low-level stuff.
1673 * This is really ugly. But the goto's actually try to clarify some
1674 * of the logic when it comes to error handling etc.
1676 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1677 struct iov_iter
*iter
, ssize_t written
)
1679 struct address_space
*mapping
= filp
->f_mapping
;
1680 struct inode
*inode
= mapping
->host
;
1681 struct file_ra_state
*ra
= &filp
->f_ra
;
1685 unsigned long offset
; /* offset into pagecache page */
1686 unsigned int prev_offset
;
1689 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
1691 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
1693 index
= *ppos
>> PAGE_SHIFT
;
1694 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1695 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1696 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1697 offset
= *ppos
& ~PAGE_MASK
;
1703 unsigned long nr
, ret
;
1707 page
= find_get_page(mapping
, index
);
1709 page_cache_sync_readahead(mapping
,
1711 index
, last_index
- index
);
1712 page
= find_get_page(mapping
, index
);
1713 if (unlikely(page
== NULL
))
1714 goto no_cached_page
;
1716 if (PageReadahead(page
)) {
1717 page_cache_async_readahead(mapping
,
1719 index
, last_index
- index
);
1721 if (!PageUptodate(page
)) {
1723 * See comment in do_read_cache_page on why
1724 * wait_on_page_locked is used to avoid unnecessarily
1725 * serialisations and why it's safe.
1727 error
= wait_on_page_locked_killable(page
);
1728 if (unlikely(error
))
1729 goto readpage_error
;
1730 if (PageUptodate(page
))
1733 if (inode
->i_blkbits
== PAGE_SHIFT
||
1734 !mapping
->a_ops
->is_partially_uptodate
)
1735 goto page_not_up_to_date
;
1736 if (!trylock_page(page
))
1737 goto page_not_up_to_date
;
1738 /* Did it get truncated before we got the lock? */
1740 goto page_not_up_to_date_locked
;
1741 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1742 offset
, iter
->count
))
1743 goto page_not_up_to_date_locked
;
1748 * i_size must be checked after we know the page is Uptodate.
1750 * Checking i_size after the check allows us to calculate
1751 * the correct value for "nr", which means the zero-filled
1752 * part of the page is not copied back to userspace (unless
1753 * another truncate extends the file - this is desired though).
1756 isize
= i_size_read(inode
);
1757 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1758 if (unlikely(!isize
|| index
> end_index
)) {
1763 /* nr is the maximum number of bytes to copy from this page */
1765 if (index
== end_index
) {
1766 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1774 /* If users can be writing to this page using arbitrary
1775 * virtual addresses, take care about potential aliasing
1776 * before reading the page on the kernel side.
1778 if (mapping_writably_mapped(mapping
))
1779 flush_dcache_page(page
);
1782 * When a sequential read accesses a page several times,
1783 * only mark it as accessed the first time.
1785 if (prev_index
!= index
|| offset
!= prev_offset
)
1786 mark_page_accessed(page
);
1790 * Ok, we have the page, and it's up-to-date, so
1791 * now we can copy it to user space...
1794 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
1796 index
+= offset
>> PAGE_SHIFT
;
1797 offset
&= ~PAGE_MASK
;
1798 prev_offset
= offset
;
1802 if (!iov_iter_count(iter
))
1810 page_not_up_to_date
:
1811 /* Get exclusive access to the page ... */
1812 error
= lock_page_killable(page
);
1813 if (unlikely(error
))
1814 goto readpage_error
;
1816 page_not_up_to_date_locked
:
1817 /* Did it get truncated before we got the lock? */
1818 if (!page
->mapping
) {
1824 /* Did somebody else fill it already? */
1825 if (PageUptodate(page
)) {
1832 * A previous I/O error may have been due to temporary
1833 * failures, eg. multipath errors.
1834 * PG_error will be set again if readpage fails.
1836 ClearPageError(page
);
1837 /* Start the actual read. The read will unlock the page. */
1838 error
= mapping
->a_ops
->readpage(filp
, page
);
1840 if (unlikely(error
)) {
1841 if (error
== AOP_TRUNCATED_PAGE
) {
1846 goto readpage_error
;
1849 if (!PageUptodate(page
)) {
1850 error
= lock_page_killable(page
);
1851 if (unlikely(error
))
1852 goto readpage_error
;
1853 if (!PageUptodate(page
)) {
1854 if (page
->mapping
== NULL
) {
1856 * invalidate_mapping_pages got it
1863 shrink_readahead_size_eio(filp
, ra
);
1865 goto readpage_error
;
1873 /* UHHUH! A synchronous read error occurred. Report it */
1879 * Ok, it wasn't cached, so we need to create a new
1882 page
= page_cache_alloc_cold(mapping
);
1887 error
= add_to_page_cache_lru(page
, mapping
, index
,
1888 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
1891 if (error
== -EEXIST
) {
1901 ra
->prev_pos
= prev_index
;
1902 ra
->prev_pos
<<= PAGE_SHIFT
;
1903 ra
->prev_pos
|= prev_offset
;
1905 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
1906 file_accessed(filp
);
1907 return written
? written
: error
;
1911 * generic_file_read_iter - generic filesystem read routine
1912 * @iocb: kernel I/O control block
1913 * @iter: destination for the data read
1915 * This is the "read_iter()" routine for all filesystems
1916 * that can use the page cache directly.
1919 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
1921 struct file
*file
= iocb
->ki_filp
;
1923 size_t count
= iov_iter_count(iter
);
1926 goto out
; /* skip atime */
1928 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1929 struct address_space
*mapping
= file
->f_mapping
;
1930 struct inode
*inode
= mapping
->host
;
1931 struct iov_iter data
= *iter
;
1934 size
= i_size_read(inode
);
1935 retval
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
1936 iocb
->ki_pos
+ count
- 1);
1940 file_accessed(file
);
1942 retval
= mapping
->a_ops
->direct_IO(iocb
, &data
);
1944 iocb
->ki_pos
+= retval
;
1945 iov_iter_advance(iter
, retval
);
1949 * Btrfs can have a short DIO read if we encounter
1950 * compressed extents, so if there was an error, or if
1951 * we've already read everything we wanted to, or if
1952 * there was a short read because we hit EOF, go ahead
1953 * and return. Otherwise fallthrough to buffered io for
1954 * the rest of the read. Buffered reads will not work for
1955 * DAX files, so don't bother trying.
1957 if (retval
< 0 || !iov_iter_count(iter
) || iocb
->ki_pos
>= size
||
1962 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
1966 EXPORT_SYMBOL(generic_file_read_iter
);
1970 * page_cache_read - adds requested page to the page cache if not already there
1971 * @file: file to read
1972 * @offset: page index
1973 * @gfp_mask: memory allocation flags
1975 * This adds the requested page to the page cache if it isn't already there,
1976 * and schedules an I/O to read in its contents from disk.
1978 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
1980 struct address_space
*mapping
= file
->f_mapping
;
1985 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
1989 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
1991 ret
= mapping
->a_ops
->readpage(file
, page
);
1992 else if (ret
== -EEXIST
)
1993 ret
= 0; /* losing race to add is OK */
1997 } while (ret
== AOP_TRUNCATED_PAGE
);
2002 #define MMAP_LOTSAMISS (100)
2005 * Synchronous readahead happens when we don't even find
2006 * a page in the page cache at all.
2008 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2009 struct file_ra_state
*ra
,
2013 struct address_space
*mapping
= file
->f_mapping
;
2015 /* If we don't want any read-ahead, don't bother */
2016 if (vma
->vm_flags
& VM_RAND_READ
)
2021 if (vma
->vm_flags
& VM_SEQ_READ
) {
2022 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2027 /* Avoid banging the cache line if not needed */
2028 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2032 * Do we miss much more than hit in this file? If so,
2033 * stop bothering with read-ahead. It will only hurt.
2035 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2041 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2042 ra
->size
= ra
->ra_pages
;
2043 ra
->async_size
= ra
->ra_pages
/ 4;
2044 ra_submit(ra
, mapping
, file
);
2048 * Asynchronous readahead happens when we find the page and PG_readahead,
2049 * so we want to possibly extend the readahead further..
2051 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2052 struct file_ra_state
*ra
,
2057 struct address_space
*mapping
= file
->f_mapping
;
2059 /* If we don't want any read-ahead, don't bother */
2060 if (vma
->vm_flags
& VM_RAND_READ
)
2062 if (ra
->mmap_miss
> 0)
2064 if (PageReadahead(page
))
2065 page_cache_async_readahead(mapping
, ra
, file
,
2066 page
, offset
, ra
->ra_pages
);
2070 * filemap_fault - read in file data for page fault handling
2071 * @vma: vma in which the fault was taken
2072 * @vmf: struct vm_fault containing details of the fault
2074 * filemap_fault() is invoked via the vma operations vector for a
2075 * mapped memory region to read in file data during a page fault.
2077 * The goto's are kind of ugly, but this streamlines the normal case of having
2078 * it in the page cache, and handles the special cases reasonably without
2079 * having a lot of duplicated code.
2081 * vma->vm_mm->mmap_sem must be held on entry.
2083 * If our return value has VM_FAULT_RETRY set, it's because
2084 * lock_page_or_retry() returned 0.
2085 * The mmap_sem has usually been released in this case.
2086 * See __lock_page_or_retry() for the exception.
2088 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2089 * has not been released.
2091 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2093 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2096 struct file
*file
= vma
->vm_file
;
2097 struct address_space
*mapping
= file
->f_mapping
;
2098 struct file_ra_state
*ra
= &file
->f_ra
;
2099 struct inode
*inode
= mapping
->host
;
2100 pgoff_t offset
= vmf
->pgoff
;
2105 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2106 if (offset
>= size
>> PAGE_SHIFT
)
2107 return VM_FAULT_SIGBUS
;
2110 * Do we have something in the page cache already?
2112 page
= find_get_page(mapping
, offset
);
2113 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2115 * We found the page, so try async readahead before
2116 * waiting for the lock.
2118 do_async_mmap_readahead(vma
, ra
, file
, page
, offset
);
2120 /* No page in the page cache at all */
2121 do_sync_mmap_readahead(vma
, ra
, file
, offset
);
2122 count_vm_event(PGMAJFAULT
);
2123 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
2124 ret
= VM_FAULT_MAJOR
;
2126 page
= find_get_page(mapping
, offset
);
2128 goto no_cached_page
;
2131 if (!lock_page_or_retry(page
, vma
->vm_mm
, vmf
->flags
)) {
2133 return ret
| VM_FAULT_RETRY
;
2136 /* Did it get truncated? */
2137 if (unlikely(page
->mapping
!= mapping
)) {
2142 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2145 * We have a locked page in the page cache, now we need to check
2146 * that it's up-to-date. If not, it is going to be due to an error.
2148 if (unlikely(!PageUptodate(page
)))
2149 goto page_not_uptodate
;
2152 * Found the page and have a reference on it.
2153 * We must recheck i_size under page lock.
2155 size
= round_up(i_size_read(inode
), PAGE_SIZE
);
2156 if (unlikely(offset
>= size
>> PAGE_SHIFT
)) {
2159 return VM_FAULT_SIGBUS
;
2163 return ret
| VM_FAULT_LOCKED
;
2167 * We're only likely to ever get here if MADV_RANDOM is in
2170 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2173 * The page we want has now been added to the page cache.
2174 * In the unlikely event that someone removed it in the
2175 * meantime, we'll just come back here and read it again.
2181 * An error return from page_cache_read can result if the
2182 * system is low on memory, or a problem occurs while trying
2185 if (error
== -ENOMEM
)
2186 return VM_FAULT_OOM
;
2187 return VM_FAULT_SIGBUS
;
2191 * Umm, take care of errors if the page isn't up-to-date.
2192 * Try to re-read it _once_. We do this synchronously,
2193 * because there really aren't any performance issues here
2194 * and we need to check for errors.
2196 ClearPageError(page
);
2197 error
= mapping
->a_ops
->readpage(file
, page
);
2199 wait_on_page_locked(page
);
2200 if (!PageUptodate(page
))
2205 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2208 /* Things didn't work out. Return zero to tell the mm layer so. */
2209 shrink_readahead_size_eio(file
, ra
);
2210 return VM_FAULT_SIGBUS
;
2212 EXPORT_SYMBOL(filemap_fault
);
2214 void filemap_map_pages(struct fault_env
*fe
,
2215 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2217 struct radix_tree_iter iter
;
2219 struct file
*file
= fe
->vma
->vm_file
;
2220 struct address_space
*mapping
= file
->f_mapping
;
2221 pgoff_t last_pgoff
= start_pgoff
;
2223 struct page
*head
, *page
;
2226 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2228 if (iter
.index
> end_pgoff
)
2231 page
= radix_tree_deref_slot(slot
);
2232 if (unlikely(!page
))
2234 if (radix_tree_exception(page
)) {
2235 if (radix_tree_deref_retry(page
)) {
2236 slot
= radix_tree_iter_retry(&iter
);
2242 head
= compound_head(page
);
2243 if (!page_cache_get_speculative(head
))
2246 /* The page was split under us? */
2247 if (compound_head(page
) != head
) {
2252 /* Has the page moved? */
2253 if (unlikely(page
!= *slot
)) {
2258 if (!PageUptodate(page
) ||
2259 PageReadahead(page
) ||
2262 if (!trylock_page(page
))
2265 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2268 size
= round_up(i_size_read(mapping
->host
), PAGE_SIZE
);
2269 if (page
->index
>= size
>> PAGE_SHIFT
)
2272 if (file
->f_ra
.mmap_miss
> 0)
2273 file
->f_ra
.mmap_miss
--;
2275 fe
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2277 fe
->pte
+= iter
.index
- last_pgoff
;
2278 last_pgoff
= iter
.index
;
2279 if (alloc_set_pte(fe
, NULL
, page
))
2288 /* Huge page is mapped? No need to proceed. */
2289 if (pmd_trans_huge(*fe
->pmd
))
2291 if (iter
.index
== end_pgoff
)
2296 EXPORT_SYMBOL(filemap_map_pages
);
2298 int filemap_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2300 struct page
*page
= vmf
->page
;
2301 struct inode
*inode
= file_inode(vma
->vm_file
);
2302 int ret
= VM_FAULT_LOCKED
;
2304 sb_start_pagefault(inode
->i_sb
);
2305 file_update_time(vma
->vm_file
);
2307 if (page
->mapping
!= inode
->i_mapping
) {
2309 ret
= VM_FAULT_NOPAGE
;
2313 * We mark the page dirty already here so that when freeze is in
2314 * progress, we are guaranteed that writeback during freezing will
2315 * see the dirty page and writeprotect it again.
2317 set_page_dirty(page
);
2318 wait_for_stable_page(page
);
2320 sb_end_pagefault(inode
->i_sb
);
2323 EXPORT_SYMBOL(filemap_page_mkwrite
);
2325 const struct vm_operations_struct generic_file_vm_ops
= {
2326 .fault
= filemap_fault
,
2327 .map_pages
= filemap_map_pages
,
2328 .page_mkwrite
= filemap_page_mkwrite
,
2331 /* This is used for a general mmap of a disk file */
2333 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2335 struct address_space
*mapping
= file
->f_mapping
;
2337 if (!mapping
->a_ops
->readpage
)
2339 file_accessed(file
);
2340 vma
->vm_ops
= &generic_file_vm_ops
;
2345 * This is for filesystems which do not implement ->writepage.
2347 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2349 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2351 return generic_file_mmap(file
, vma
);
2354 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2358 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2362 #endif /* CONFIG_MMU */
2364 EXPORT_SYMBOL(generic_file_mmap
);
2365 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2367 static struct page
*wait_on_page_read(struct page
*page
)
2369 if (!IS_ERR(page
)) {
2370 wait_on_page_locked(page
);
2371 if (!PageUptodate(page
)) {
2373 page
= ERR_PTR(-EIO
);
2379 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2381 int (*filler
)(void *, struct page
*),
2388 page
= find_get_page(mapping
, index
);
2390 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2392 return ERR_PTR(-ENOMEM
);
2393 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2394 if (unlikely(err
)) {
2398 /* Presumably ENOMEM for radix tree node */
2399 return ERR_PTR(err
);
2403 err
= filler(data
, page
);
2406 return ERR_PTR(err
);
2409 page
= wait_on_page_read(page
);
2414 if (PageUptodate(page
))
2418 * Page is not up to date and may be locked due one of the following
2419 * case a: Page is being filled and the page lock is held
2420 * case b: Read/write error clearing the page uptodate status
2421 * case c: Truncation in progress (page locked)
2422 * case d: Reclaim in progress
2424 * Case a, the page will be up to date when the page is unlocked.
2425 * There is no need to serialise on the page lock here as the page
2426 * is pinned so the lock gives no additional protection. Even if the
2427 * the page is truncated, the data is still valid if PageUptodate as
2428 * it's a race vs truncate race.
2429 * Case b, the page will not be up to date
2430 * Case c, the page may be truncated but in itself, the data may still
2431 * be valid after IO completes as it's a read vs truncate race. The
2432 * operation must restart if the page is not uptodate on unlock but
2433 * otherwise serialising on page lock to stabilise the mapping gives
2434 * no additional guarantees to the caller as the page lock is
2435 * released before return.
2436 * Case d, similar to truncation. If reclaim holds the page lock, it
2437 * will be a race with remove_mapping that determines if the mapping
2438 * is valid on unlock but otherwise the data is valid and there is
2439 * no need to serialise with page lock.
2441 * As the page lock gives no additional guarantee, we optimistically
2442 * wait on the page to be unlocked and check if it's up to date and
2443 * use the page if it is. Otherwise, the page lock is required to
2444 * distinguish between the different cases. The motivation is that we
2445 * avoid spurious serialisations and wakeups when multiple processes
2446 * wait on the same page for IO to complete.
2448 wait_on_page_locked(page
);
2449 if (PageUptodate(page
))
2452 /* Distinguish between all the cases under the safety of the lock */
2455 /* Case c or d, restart the operation */
2456 if (!page
->mapping
) {
2462 /* Someone else locked and filled the page in a very small window */
2463 if (PageUptodate(page
)) {
2470 mark_page_accessed(page
);
2475 * read_cache_page - read into page cache, fill it if needed
2476 * @mapping: the page's address_space
2477 * @index: the page index
2478 * @filler: function to perform the read
2479 * @data: first arg to filler(data, page) function, often left as NULL
2481 * Read into the page cache. If a page already exists, and PageUptodate() is
2482 * not set, try to fill the page and wait for it to become unlocked.
2484 * If the page does not get brought uptodate, return -EIO.
2486 struct page
*read_cache_page(struct address_space
*mapping
,
2488 int (*filler
)(void *, struct page
*),
2491 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2493 EXPORT_SYMBOL(read_cache_page
);
2496 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2497 * @mapping: the page's address_space
2498 * @index: the page index
2499 * @gfp: the page allocator flags to use if allocating
2501 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2502 * any new page allocations done using the specified allocation flags.
2504 * If the page does not get brought uptodate, return -EIO.
2506 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2510 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2512 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2514 EXPORT_SYMBOL(read_cache_page_gfp
);
2517 * Performs necessary checks before doing a write
2519 * Can adjust writing position or amount of bytes to write.
2520 * Returns appropriate error code that caller should return or
2521 * zero in case that write should be allowed.
2523 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2525 struct file
*file
= iocb
->ki_filp
;
2526 struct inode
*inode
= file
->f_mapping
->host
;
2527 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2530 if (!iov_iter_count(from
))
2533 /* FIXME: this is for backwards compatibility with 2.4 */
2534 if (iocb
->ki_flags
& IOCB_APPEND
)
2535 iocb
->ki_pos
= i_size_read(inode
);
2539 if (limit
!= RLIM_INFINITY
) {
2540 if (iocb
->ki_pos
>= limit
) {
2541 send_sig(SIGXFSZ
, current
, 0);
2544 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2550 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2551 !(file
->f_flags
& O_LARGEFILE
))) {
2552 if (pos
>= MAX_NON_LFS
)
2554 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2558 * Are we about to exceed the fs block limit ?
2560 * If we have written data it becomes a short write. If we have
2561 * exceeded without writing data we send a signal and return EFBIG.
2562 * Linus frestrict idea will clean these up nicely..
2564 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2567 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2568 return iov_iter_count(from
);
2570 EXPORT_SYMBOL(generic_write_checks
);
2572 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2573 loff_t pos
, unsigned len
, unsigned flags
,
2574 struct page
**pagep
, void **fsdata
)
2576 const struct address_space_operations
*aops
= mapping
->a_ops
;
2578 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2581 EXPORT_SYMBOL(pagecache_write_begin
);
2583 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2584 loff_t pos
, unsigned len
, unsigned copied
,
2585 struct page
*page
, void *fsdata
)
2587 const struct address_space_operations
*aops
= mapping
->a_ops
;
2589 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2591 EXPORT_SYMBOL(pagecache_write_end
);
2594 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2596 struct file
*file
= iocb
->ki_filp
;
2597 struct address_space
*mapping
= file
->f_mapping
;
2598 struct inode
*inode
= mapping
->host
;
2599 loff_t pos
= iocb
->ki_pos
;
2603 struct iov_iter data
;
2605 write_len
= iov_iter_count(from
);
2606 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2608 written
= filemap_write_and_wait_range(mapping
, pos
, pos
+ write_len
- 1);
2613 * After a write we want buffered reads to be sure to go to disk to get
2614 * the new data. We invalidate clean cached page from the region we're
2615 * about to write. We do this *before* the write so that we can return
2616 * without clobbering -EIOCBQUEUED from ->direct_IO().
2618 if (mapping
->nrpages
) {
2619 written
= invalidate_inode_pages2_range(mapping
,
2620 pos
>> PAGE_SHIFT
, end
);
2622 * If a page can not be invalidated, return 0 to fall back
2623 * to buffered write.
2626 if (written
== -EBUSY
)
2633 written
= mapping
->a_ops
->direct_IO(iocb
, &data
);
2636 * Finally, try again to invalidate clean pages which might have been
2637 * cached by non-direct readahead, or faulted in by get_user_pages()
2638 * if the source of the write was an mmap'ed region of the file
2639 * we're writing. Either one is a pretty crazy thing to do,
2640 * so we don't support it 100%. If this invalidation
2641 * fails, tough, the write still worked...
2643 if (mapping
->nrpages
) {
2644 invalidate_inode_pages2_range(mapping
,
2645 pos
>> PAGE_SHIFT
, end
);
2650 iov_iter_advance(from
, written
);
2651 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2652 i_size_write(inode
, pos
);
2653 mark_inode_dirty(inode
);
2660 EXPORT_SYMBOL(generic_file_direct_write
);
2663 * Find or create a page at the given pagecache position. Return the locked
2664 * page. This function is specifically for buffered writes.
2666 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2667 pgoff_t index
, unsigned flags
)
2670 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2672 if (flags
& AOP_FLAG_NOFS
)
2673 fgp_flags
|= FGP_NOFS
;
2675 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2676 mapping_gfp_mask(mapping
));
2678 wait_for_stable_page(page
);
2682 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2684 ssize_t
generic_perform_write(struct file
*file
,
2685 struct iov_iter
*i
, loff_t pos
)
2687 struct address_space
*mapping
= file
->f_mapping
;
2688 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2690 ssize_t written
= 0;
2691 unsigned int flags
= 0;
2694 * Copies from kernel address space cannot fail (NFSD is a big user).
2696 if (!iter_is_iovec(i
))
2697 flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
2701 unsigned long offset
; /* Offset into pagecache page */
2702 unsigned long bytes
; /* Bytes to write to page */
2703 size_t copied
; /* Bytes copied from user */
2706 offset
= (pos
& (PAGE_SIZE
- 1));
2707 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2712 * Bring in the user page that we will copy from _first_.
2713 * Otherwise there's a nasty deadlock on copying from the
2714 * same page as we're writing to, without it being marked
2717 * Not only is this an optimisation, but it is also required
2718 * to check that the address is actually valid, when atomic
2719 * usercopies are used, below.
2721 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2726 if (fatal_signal_pending(current
)) {
2731 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2733 if (unlikely(status
< 0))
2736 if (mapping_writably_mapped(mapping
))
2737 flush_dcache_page(page
);
2739 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2740 flush_dcache_page(page
);
2742 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2744 if (unlikely(status
< 0))
2750 iov_iter_advance(i
, copied
);
2751 if (unlikely(copied
== 0)) {
2753 * If we were unable to copy any data at all, we must
2754 * fall back to a single segment length write.
2756 * If we didn't fallback here, we could livelock
2757 * because not all segments in the iov can be copied at
2758 * once without a pagefault.
2760 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2761 iov_iter_single_seg_count(i
));
2767 balance_dirty_pages_ratelimited(mapping
);
2768 } while (iov_iter_count(i
));
2770 return written
? written
: status
;
2772 EXPORT_SYMBOL(generic_perform_write
);
2775 * __generic_file_write_iter - write data to a file
2776 * @iocb: IO state structure (file, offset, etc.)
2777 * @from: iov_iter with data to write
2779 * This function does all the work needed for actually writing data to a
2780 * file. It does all basic checks, removes SUID from the file, updates
2781 * modification times and calls proper subroutines depending on whether we
2782 * do direct IO or a standard buffered write.
2784 * It expects i_mutex to be grabbed unless we work on a block device or similar
2785 * object which does not need locking at all.
2787 * This function does *not* take care of syncing data in case of O_SYNC write.
2788 * A caller has to handle it. This is mainly due to the fact that we want to
2789 * avoid syncing under i_mutex.
2791 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2793 struct file
*file
= iocb
->ki_filp
;
2794 struct address_space
* mapping
= file
->f_mapping
;
2795 struct inode
*inode
= mapping
->host
;
2796 ssize_t written
= 0;
2800 /* We can write back this queue in page reclaim */
2801 current
->backing_dev_info
= inode_to_bdi(inode
);
2802 err
= file_remove_privs(file
);
2806 err
= file_update_time(file
);
2810 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2811 loff_t pos
, endbyte
;
2813 written
= generic_file_direct_write(iocb
, from
);
2815 * If the write stopped short of completing, fall back to
2816 * buffered writes. Some filesystems do this for writes to
2817 * holes, for example. For DAX files, a buffered write will
2818 * not succeed (even if it did, DAX does not handle dirty
2819 * page-cache pages correctly).
2821 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
2824 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
2826 * If generic_perform_write() returned a synchronous error
2827 * then we want to return the number of bytes which were
2828 * direct-written, or the error code if that was zero. Note
2829 * that this differs from normal direct-io semantics, which
2830 * will return -EFOO even if some bytes were written.
2832 if (unlikely(status
< 0)) {
2837 * We need to ensure that the page cache pages are written to
2838 * disk and invalidated to preserve the expected O_DIRECT
2841 endbyte
= pos
+ status
- 1;
2842 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
2844 iocb
->ki_pos
= endbyte
+ 1;
2846 invalidate_mapping_pages(mapping
,
2848 endbyte
>> PAGE_SHIFT
);
2851 * We don't know how much we wrote, so just return
2852 * the number of bytes which were direct-written
2856 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
2857 if (likely(written
> 0))
2858 iocb
->ki_pos
+= written
;
2861 current
->backing_dev_info
= NULL
;
2862 return written
? written
: err
;
2864 EXPORT_SYMBOL(__generic_file_write_iter
);
2867 * generic_file_write_iter - write data to a file
2868 * @iocb: IO state structure
2869 * @from: iov_iter with data to write
2871 * This is a wrapper around __generic_file_write_iter() to be used by most
2872 * filesystems. It takes care of syncing the file in case of O_SYNC file
2873 * and acquires i_mutex as needed.
2875 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2877 struct file
*file
= iocb
->ki_filp
;
2878 struct inode
*inode
= file
->f_mapping
->host
;
2882 ret
= generic_write_checks(iocb
, from
);
2884 ret
= __generic_file_write_iter(iocb
, from
);
2885 inode_unlock(inode
);
2888 ret
= generic_write_sync(iocb
, ret
);
2891 EXPORT_SYMBOL(generic_file_write_iter
);
2894 * try_to_release_page() - release old fs-specific metadata on a page
2896 * @page: the page which the kernel is trying to free
2897 * @gfp_mask: memory allocation flags (and I/O mode)
2899 * The address_space is to try to release any data against the page
2900 * (presumably at page->private). If the release was successful, return `1'.
2901 * Otherwise return zero.
2903 * This may also be called if PG_fscache is set on a page, indicating that the
2904 * page is known to the local caching routines.
2906 * The @gfp_mask argument specifies whether I/O may be performed to release
2907 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2910 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
2912 struct address_space
* const mapping
= page
->mapping
;
2914 BUG_ON(!PageLocked(page
));
2915 if (PageWriteback(page
))
2918 if (mapping
&& mapping
->a_ops
->releasepage
)
2919 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
2920 return try_to_free_buffers(page
);
2923 EXPORT_SYMBOL(try_to_release_page
);