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/sched/signal.h>
17 #include <linux/uaccess.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/gfp.h>
22 #include <linux/swap.h>
23 #include <linux/mman.h>
24 #include <linux/pagemap.h>
25 #include <linux/file.h>
26 #include <linux/uio.h>
27 #include <linux/hash.h>
28 #include <linux/writeback.h>
29 #include <linux/backing-dev.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/security.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/hugetlb.h>
36 #include <linux/memcontrol.h>
37 #include <linux/cleancache.h>
38 #include <linux/rmap.h>
39 #include <linux/delayacct.h>
40 #include <linux/psi.h>
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/filemap.h>
47 * FIXME: remove all knowledge of the buffer layer from the core VM
49 #include <linux/buffer_head.h> /* for try_to_free_buffers */
54 * Shared mappings implemented 30.11.1994. It's not fully working yet,
57 * Shared mappings now work. 15.8.1995 Bruno.
59 * finished 'unifying' the page and buffer cache and SMP-threaded the
60 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
62 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
68 * ->i_mmap_rwsem (truncate_pagecache)
69 * ->private_lock (__free_pte->__set_page_dirty_buffers)
70 * ->swap_lock (exclusive_swap_page, others)
71 * ->mapping->tree_lock
74 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
78 * ->page_table_lock or pte_lock (various, mainly in memory.c)
79 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
82 * ->lock_page (access_process_vm)
84 * ->i_mutex (generic_perform_write)
85 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
88 * sb_lock (fs/fs-writeback.c)
89 * ->mapping->tree_lock (__sync_single_inode)
92 * ->anon_vma.lock (vma_adjust)
95 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
97 * ->page_table_lock or pte_lock
98 * ->swap_lock (try_to_unmap_one)
99 * ->private_lock (try_to_unmap_one)
100 * ->tree_lock (try_to_unmap_one)
101 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
102 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
103 * ->private_lock (page_remove_rmap->set_page_dirty)
104 * ->tree_lock (page_remove_rmap->set_page_dirty)
105 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
106 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
107 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
108 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
109 * ->inode->i_lock (zap_pte_range->set_page_dirty)
110 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
113 * ->tasklist_lock (memory_failure, collect_procs_ao)
116 static int page_cache_tree_insert(struct address_space
*mapping
,
117 struct page
*page
, void **shadowp
)
119 struct radix_tree_node
*node
;
123 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
130 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
131 if (!radix_tree_exceptional_entry(p
))
134 mapping
->nrexceptional
--;
138 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, page
,
139 workingset_update_node
, mapping
);
144 static void page_cache_tree_delete(struct address_space
*mapping
,
145 struct page
*page
, void *shadow
)
149 /* hugetlb pages are represented by one entry in the radix tree */
150 nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
152 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
153 VM_BUG_ON_PAGE(PageTail(page
), page
);
154 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
156 for (i
= 0; i
< nr
; i
++) {
157 struct radix_tree_node
*node
;
160 __radix_tree_lookup(&mapping
->page_tree
, page
->index
+ i
,
163 VM_BUG_ON_PAGE(!node
&& nr
!= 1, page
);
165 radix_tree_clear_tags(&mapping
->page_tree
, node
, slot
);
166 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, shadow
,
167 workingset_update_node
, mapping
);
171 mapping
->nrexceptional
+= nr
;
173 * Make sure the nrexceptional update is committed before
174 * the nrpages update so that final truncate racing
175 * with reclaim does not see both counters 0 at the
176 * same time and miss a shadow entry.
180 mapping
->nrpages
-= nr
;
184 * Delete a page from the page cache and free it. Caller has to make
185 * sure the page is locked and that nobody else uses it - or that usage
186 * is safe. The caller must hold the mapping's tree_lock.
188 void __delete_from_page_cache(struct page
*page
, void *shadow
)
190 struct address_space
*mapping
= page
->mapping
;
191 int nr
= hpage_nr_pages(page
);
193 trace_mm_filemap_delete_from_page_cache(page
);
195 * if we're uptodate, flush out into the cleancache, otherwise
196 * invalidate any existing cleancache entries. We can't leave
197 * stale data around in the cleancache once our page is gone
199 if (PageUptodate(page
) && PageMappedToDisk(page
))
200 cleancache_put_page(page
);
202 cleancache_invalidate_page(mapping
, page
);
204 VM_BUG_ON_PAGE(PageTail(page
), page
);
205 VM_BUG_ON_PAGE(page_mapped(page
), page
);
206 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
209 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
210 current
->comm
, page_to_pfn(page
));
211 dump_page(page
, "still mapped when deleted");
213 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
215 mapcount
= page_mapcount(page
);
216 if (mapping_exiting(mapping
) &&
217 page_count(page
) >= mapcount
+ 2) {
219 * All vmas have already been torn down, so it's
220 * a good bet that actually the page is unmapped,
221 * and we'd prefer not to leak it: if we're wrong,
222 * some other bad page check should catch it later.
224 page_mapcount_reset(page
);
225 page_ref_sub(page
, mapcount
);
229 page_cache_tree_delete(mapping
, page
, shadow
);
231 page
->mapping
= NULL
;
232 /* Leave page->index set: truncation lookup relies upon it */
234 /* hugetlb pages do not participate in page cache accounting. */
238 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
239 if (PageSwapBacked(page
)) {
240 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
241 if (PageTransHuge(page
))
242 __dec_node_page_state(page
, NR_SHMEM_THPS
);
244 VM_BUG_ON_PAGE(PageTransHuge(page
), page
);
248 * At this point page must be either written or cleaned by truncate.
249 * Dirty page here signals a bug and loss of unwritten data.
251 * This fixes dirty accounting after removing the page entirely but
252 * leaves PageDirty set: it has no effect for truncated page and
253 * anyway will be cleared before returning page into buddy allocator.
255 if (WARN_ON_ONCE(PageDirty(page
)))
256 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
260 * delete_from_page_cache - delete page from page cache
261 * @page: the page which the kernel is trying to remove from page cache
263 * This must be called only on pages that have been verified to be in the page
264 * cache and locked. It will never put the page into the free list, the caller
265 * has a reference on the page.
267 void delete_from_page_cache(struct page
*page
)
269 struct address_space
*mapping
= page_mapping(page
);
271 void (*freepage
)(struct page
*);
273 BUG_ON(!PageLocked(page
));
275 freepage
= mapping
->a_ops
->freepage
;
277 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
278 __delete_from_page_cache(page
, NULL
);
279 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
284 if (PageTransHuge(page
) && !PageHuge(page
)) {
285 page_ref_sub(page
, HPAGE_PMD_NR
);
286 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
291 EXPORT_SYMBOL(delete_from_page_cache
);
293 int filemap_check_errors(struct address_space
*mapping
)
296 /* Check for outstanding write errors */
297 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
298 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
300 if (test_bit(AS_EIO
, &mapping
->flags
) &&
301 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
305 EXPORT_SYMBOL(filemap_check_errors
);
307 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
309 /* Check for outstanding write errors */
310 if (test_bit(AS_EIO
, &mapping
->flags
))
312 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
318 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
319 * @mapping: address space structure to write
320 * @start: offset in bytes where the range starts
321 * @end: offset in bytes where the range ends (inclusive)
322 * @sync_mode: enable synchronous operation
324 * Start writeback against all of a mapping's dirty pages that lie
325 * within the byte offsets <start, end> inclusive.
327 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
328 * opposed to a regular memory cleansing writeback. The difference between
329 * these two operations is that if a dirty page/buffer is encountered, it must
330 * be waited upon, and not just skipped over.
332 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
333 loff_t end
, int sync_mode
)
336 struct writeback_control wbc
= {
337 .sync_mode
= sync_mode
,
338 .nr_to_write
= LONG_MAX
,
339 .range_start
= start
,
343 if (!mapping_cap_writeback_dirty(mapping
))
346 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
347 ret
= do_writepages(mapping
, &wbc
);
348 wbc_detach_inode(&wbc
);
352 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
355 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
358 int filemap_fdatawrite(struct address_space
*mapping
)
360 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
362 EXPORT_SYMBOL(filemap_fdatawrite
);
364 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
367 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
369 EXPORT_SYMBOL(filemap_fdatawrite_range
);
372 * filemap_flush - mostly a non-blocking flush
373 * @mapping: target address_space
375 * This is a mostly non-blocking flush. Not suitable for data-integrity
376 * purposes - I/O may not be started against all dirty pages.
378 int filemap_flush(struct address_space
*mapping
)
380 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
382 EXPORT_SYMBOL(filemap_flush
);
385 * filemap_range_has_page - check if a page exists in range.
386 * @mapping: address space within which to check
387 * @start_byte: offset in bytes where the range starts
388 * @end_byte: offset in bytes where the range ends (inclusive)
390 * Find at least one page in the range supplied, usually used to check if
391 * direct writing in this range will trigger a writeback.
393 bool filemap_range_has_page(struct address_space
*mapping
,
394 loff_t start_byte
, loff_t end_byte
)
396 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
397 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
400 if (end_byte
< start_byte
)
403 if (mapping
->nrpages
== 0)
406 if (!find_get_pages_range(mapping
, &index
, end
, 1, &page
))
411 EXPORT_SYMBOL(filemap_range_has_page
);
413 static void __filemap_fdatawait_range(struct address_space
*mapping
,
414 loff_t start_byte
, loff_t end_byte
)
416 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
417 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
421 if (end_byte
< start_byte
)
424 pagevec_init(&pvec
, 0);
425 while ((index
<= end
) &&
426 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
427 PAGECACHE_TAG_WRITEBACK
,
428 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
431 for (i
= 0; i
< nr_pages
; i
++) {
432 struct page
*page
= pvec
.pages
[i
];
434 /* until radix tree lookup accepts end_index */
435 if (page
->index
> end
)
438 wait_on_page_writeback(page
);
439 ClearPageError(page
);
441 pagevec_release(&pvec
);
447 * filemap_fdatawait_range - wait for writeback to complete
448 * @mapping: address space structure to wait for
449 * @start_byte: offset in bytes where the range starts
450 * @end_byte: offset in bytes where the range ends (inclusive)
452 * Walk the list of under-writeback pages of the given address space
453 * in the given range and wait for all of them. Check error status of
454 * the address space and return it.
456 * Since the error status of the address space is cleared by this function,
457 * callers are responsible for checking the return value and handling and/or
458 * reporting the error.
460 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
463 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
464 return filemap_check_errors(mapping
);
466 EXPORT_SYMBOL(filemap_fdatawait_range
);
469 * file_fdatawait_range - wait for writeback to complete
470 * @file: file pointing to address space structure to wait for
471 * @start_byte: offset in bytes where the range starts
472 * @end_byte: offset in bytes where the range ends (inclusive)
474 * Walk the list of under-writeback pages of the address space that file
475 * refers to, in the given range and wait for all of them. Check error
476 * status of the address space vs. the file->f_wb_err cursor and return it.
478 * Since the error status of the file is advanced by this function,
479 * callers are responsible for checking the return value and handling and/or
480 * reporting the error.
482 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
484 struct address_space
*mapping
= file
->f_mapping
;
486 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
487 return file_check_and_advance_wb_err(file
);
489 EXPORT_SYMBOL(file_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 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
505 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
506 return filemap_check_and_keep_errors(mapping
);
508 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
510 static bool mapping_needs_writeback(struct address_space
*mapping
)
512 return (!dax_mapping(mapping
) && mapping
->nrpages
) ||
513 (dax_mapping(mapping
) && mapping
->nrexceptional
);
516 int filemap_write_and_wait(struct address_space
*mapping
)
520 if (mapping_needs_writeback(mapping
)) {
521 err
= filemap_fdatawrite(mapping
);
523 * Even if the above returned error, the pages may be
524 * written partially (e.g. -ENOSPC), so we wait for it.
525 * But the -EIO is special case, it may indicate the worst
526 * thing (e.g. bug) happened, so we avoid waiting for it.
529 int err2
= filemap_fdatawait(mapping
);
533 /* Clear any previously stored errors */
534 filemap_check_errors(mapping
);
537 err
= filemap_check_errors(mapping
);
541 EXPORT_SYMBOL(filemap_write_and_wait
);
544 * filemap_write_and_wait_range - write out & wait on a file range
545 * @mapping: the address_space for the pages
546 * @lstart: offset in bytes where the range starts
547 * @lend: offset in bytes where the range ends (inclusive)
549 * Write out and wait upon file offsets lstart->lend, inclusive.
551 * Note that @lend is inclusive (describes the last byte to be written) so
552 * that this function can be used to write to the very end-of-file (end = -1).
554 int filemap_write_and_wait_range(struct address_space
*mapping
,
555 loff_t lstart
, loff_t lend
)
559 if (mapping_needs_writeback(mapping
)) {
560 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
562 /* See comment of filemap_write_and_wait() */
564 int err2
= filemap_fdatawait_range(mapping
,
569 /* Clear any previously stored errors */
570 filemap_check_errors(mapping
);
573 err
= filemap_check_errors(mapping
);
577 EXPORT_SYMBOL(filemap_write_and_wait_range
);
579 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
581 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
583 trace_filemap_set_wb_err(mapping
, eseq
);
585 EXPORT_SYMBOL(__filemap_set_wb_err
);
588 * file_check_and_advance_wb_err - report wb error (if any) that was previously
589 * and advance wb_err to current one
590 * @file: struct file on which the error is being reported
592 * When userland calls fsync (or something like nfsd does the equivalent), we
593 * want to report any writeback errors that occurred since the last fsync (or
594 * since the file was opened if there haven't been any).
596 * Grab the wb_err from the mapping. If it matches what we have in the file,
597 * then just quickly return 0. The file is all caught up.
599 * If it doesn't match, then take the mapping value, set the "seen" flag in
600 * it and try to swap it into place. If it works, or another task beat us
601 * to it with the new value, then update the f_wb_err and return the error
602 * portion. The error at this point must be reported via proper channels
603 * (a'la fsync, or NFS COMMIT operation, etc.).
605 * While we handle mapping->wb_err with atomic operations, the f_wb_err
606 * value is protected by the f_lock since we must ensure that it reflects
607 * the latest value swapped in for this file descriptor.
609 int file_check_and_advance_wb_err(struct file
*file
)
612 errseq_t old
= READ_ONCE(file
->f_wb_err
);
613 struct address_space
*mapping
= file
->f_mapping
;
615 /* Locklessly handle the common case where nothing has changed */
616 if (errseq_check(&mapping
->wb_err
, old
)) {
617 /* Something changed, must use slow path */
618 spin_lock(&file
->f_lock
);
619 old
= file
->f_wb_err
;
620 err
= errseq_check_and_advance(&mapping
->wb_err
,
622 trace_file_check_and_advance_wb_err(file
, old
);
623 spin_unlock(&file
->f_lock
);
627 * We're mostly using this function as a drop in replacement for
628 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
629 * that the legacy code would have had on these flags.
631 clear_bit(AS_EIO
, &mapping
->flags
);
632 clear_bit(AS_ENOSPC
, &mapping
->flags
);
635 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
638 * file_write_and_wait_range - write out & wait on a file range
639 * @file: file pointing to address_space with pages
640 * @lstart: offset in bytes where the range starts
641 * @lend: offset in bytes where the range ends (inclusive)
643 * Write out and wait upon file offsets lstart->lend, inclusive.
645 * Note that @lend is inclusive (describes the last byte to be written) so
646 * that this function can be used to write to the very end-of-file (end = -1).
648 * After writing out and waiting on the data, we check and advance the
649 * f_wb_err cursor to the latest value, and return any errors detected there.
651 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
654 struct address_space
*mapping
= file
->f_mapping
;
656 if (mapping_needs_writeback(mapping
)) {
657 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
659 /* See comment of filemap_write_and_wait() */
661 __filemap_fdatawait_range(mapping
, lstart
, lend
);
663 err2
= file_check_and_advance_wb_err(file
);
668 EXPORT_SYMBOL(file_write_and_wait_range
);
671 * replace_page_cache_page - replace a pagecache page with a new one
672 * @old: page to be replaced
673 * @new: page to replace with
674 * @gfp_mask: allocation mode
676 * This function replaces a page in the pagecache with a new one. On
677 * success it acquires the pagecache reference for the new page and
678 * drops it for the old page. Both the old and new pages must be
679 * locked. This function does not add the new page to the LRU, the
680 * caller must do that.
682 * The remove + add is atomic. The only way this function can fail is
683 * memory allocation failure.
685 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
689 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
690 VM_BUG_ON_PAGE(!PageLocked(new), new);
691 VM_BUG_ON_PAGE(new->mapping
, new);
693 error
= radix_tree_preload(gfp_mask
& GFP_RECLAIM_MASK
);
695 struct address_space
*mapping
= old
->mapping
;
696 void (*freepage
)(struct page
*);
699 pgoff_t offset
= old
->index
;
700 freepage
= mapping
->a_ops
->freepage
;
703 new->mapping
= mapping
;
706 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
707 __delete_from_page_cache(old
, NULL
);
708 error
= page_cache_tree_insert(mapping
, new, NULL
);
712 * hugetlb pages do not participate in page cache accounting.
715 __inc_node_page_state(new, NR_FILE_PAGES
);
716 if (PageSwapBacked(new))
717 __inc_node_page_state(new, NR_SHMEM
);
718 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
719 mem_cgroup_migrate(old
, new);
720 radix_tree_preload_end();
728 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
730 static int __add_to_page_cache_locked(struct page
*page
,
731 struct address_space
*mapping
,
732 pgoff_t offset
, gfp_t gfp_mask
,
735 int huge
= PageHuge(page
);
736 struct mem_cgroup
*memcg
;
739 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
740 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
743 error
= mem_cgroup_try_charge(page
, current
->mm
,
744 gfp_mask
, &memcg
, false);
749 error
= radix_tree_maybe_preload(gfp_mask
& GFP_RECLAIM_MASK
);
752 mem_cgroup_cancel_charge(page
, memcg
, false);
757 page
->mapping
= mapping
;
758 page
->index
= offset
;
760 spin_lock_irq(&mapping
->tree_lock
);
761 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
762 radix_tree_preload_end();
766 /* hugetlb pages do not participate in page cache accounting. */
768 __inc_node_page_state(page
, NR_FILE_PAGES
);
769 spin_unlock_irq(&mapping
->tree_lock
);
771 mem_cgroup_commit_charge(page
, memcg
, false, false);
772 trace_mm_filemap_add_to_page_cache(page
);
775 page
->mapping
= NULL
;
776 /* Leave page->index set: truncation relies upon it */
777 spin_unlock_irq(&mapping
->tree_lock
);
779 mem_cgroup_cancel_charge(page
, memcg
, false);
785 * add_to_page_cache_locked - add a locked page to the pagecache
787 * @mapping: the page's address_space
788 * @offset: page index
789 * @gfp_mask: page allocation mode
791 * This function is used to add a page to the pagecache. It must be locked.
792 * This function does not add the page to the LRU. The caller must do that.
794 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
795 pgoff_t offset
, gfp_t gfp_mask
)
797 return __add_to_page_cache_locked(page
, mapping
, offset
,
800 EXPORT_SYMBOL(add_to_page_cache_locked
);
802 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
803 pgoff_t offset
, gfp_t gfp_mask
)
808 __SetPageLocked(page
);
809 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
812 __ClearPageLocked(page
);
815 * The page might have been evicted from cache only
816 * recently, in which case it should be activated like
817 * any other repeatedly accessed page.
818 * The exception is pages getting rewritten; evicting other
819 * data from the working set, only to cache data that will
820 * get overwritten with something else, is a waste of memory.
822 WARN_ON_ONCE(PageActive(page
));
823 if (!(gfp_mask
& __GFP_WRITE
) && shadow
)
824 workingset_refault(page
, shadow
);
829 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
832 struct page
*__page_cache_alloc(gfp_t gfp
)
837 if (cpuset_do_page_mem_spread()) {
838 unsigned int cpuset_mems_cookie
;
840 cpuset_mems_cookie
= read_mems_allowed_begin();
841 n
= cpuset_mem_spread_node();
842 page
= __alloc_pages_node(n
, gfp
, 0);
843 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
847 return alloc_pages(gfp
, 0);
849 EXPORT_SYMBOL(__page_cache_alloc
);
853 * In order to wait for pages to become available there must be
854 * waitqueues associated with pages. By using a hash table of
855 * waitqueues where the bucket discipline is to maintain all
856 * waiters on the same queue and wake all when any of the pages
857 * become available, and for the woken contexts to check to be
858 * sure the appropriate page became available, this saves space
859 * at a cost of "thundering herd" phenomena during rare hash
862 #define PAGE_WAIT_TABLE_BITS 8
863 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
864 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
866 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
868 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
871 void __init
pagecache_init(void)
875 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
876 init_waitqueue_head(&page_wait_table
[i
]);
878 page_writeback_init();
881 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
882 struct wait_page_key
{
888 struct wait_page_queue
{
891 wait_queue_entry_t wait
;
894 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
896 struct wait_page_key
*key
= arg
;
897 struct wait_page_queue
*wait_page
898 = container_of(wait
, struct wait_page_queue
, wait
);
900 if (wait_page
->page
!= key
->page
)
904 if (wait_page
->bit_nr
!= key
->bit_nr
)
907 /* Stop walking if it's locked */
908 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
911 return autoremove_wake_function(wait
, mode
, sync
, key
);
914 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
916 wait_queue_head_t
*q
= page_waitqueue(page
);
917 struct wait_page_key key
;
919 wait_queue_entry_t bookmark
;
926 bookmark
.private = NULL
;
927 bookmark
.func
= NULL
;
928 INIT_LIST_HEAD(&bookmark
.entry
);
930 spin_lock_irqsave(&q
->lock
, flags
);
931 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
933 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
935 * Take a breather from holding the lock,
936 * allow pages that finish wake up asynchronously
937 * to acquire the lock and remove themselves
940 spin_unlock_irqrestore(&q
->lock
, flags
);
942 spin_lock_irqsave(&q
->lock
, flags
);
943 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
947 * It is possible for other pages to have collided on the waitqueue
948 * hash, so in that case check for a page match. That prevents a long-
951 * It is still possible to miss a case here, when we woke page waiters
952 * and removed them from the waitqueue, but there are still other
955 if (!waitqueue_active(q
) || !key
.page_match
) {
956 ClearPageWaiters(page
);
958 * It's possible to miss clearing Waiters here, when we woke
959 * our page waiters, but the hashed waitqueue has waiters for
962 * That's okay, it's a rare case. The next waker will clear it.
965 spin_unlock_irqrestore(&q
->lock
, flags
);
968 static void wake_up_page(struct page
*page
, int bit
)
970 if (!PageWaiters(page
))
972 wake_up_page_bit(page
, bit
);
975 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
976 struct page
*page
, int bit_nr
, int state
, bool lock
)
978 struct wait_page_queue wait_page
;
979 wait_queue_entry_t
*wait
= &wait_page
.wait
;
980 bool thrashing
= false;
981 unsigned long pflags
;
984 if (bit_nr
== PG_locked
&&
985 !PageUptodate(page
) && PageWorkingset(page
)) {
986 if (!PageSwapBacked(page
))
987 delayacct_thrashing_start();
988 psi_memstall_enter(&pflags
);
993 wait
->flags
= lock
? WQ_FLAG_EXCLUSIVE
: 0;
994 wait
->func
= wake_page_function
;
995 wait_page
.page
= page
;
996 wait_page
.bit_nr
= bit_nr
;
999 spin_lock_irq(&q
->lock
);
1001 if (likely(list_empty(&wait
->entry
))) {
1002 __add_wait_queue_entry_tail(q
, wait
);
1003 SetPageWaiters(page
);
1006 set_current_state(state
);
1008 spin_unlock_irq(&q
->lock
);
1010 if (likely(test_bit(bit_nr
, &page
->flags
))) {
1015 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
1018 if (!test_bit(bit_nr
, &page
->flags
))
1022 if (unlikely(signal_pending_state(state
, current
))) {
1028 finish_wait(q
, wait
);
1031 if (!PageSwapBacked(page
))
1032 delayacct_thrashing_end();
1033 psi_memstall_leave(&pflags
);
1037 * A signal could leave PageWaiters set. Clearing it here if
1038 * !waitqueue_active would be possible (by open-coding finish_wait),
1039 * but still fail to catch it in the case of wait hash collision. We
1040 * already can fail to clear wait hash collision cases, so don't
1041 * bother with signals either.
1047 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1049 wait_queue_head_t
*q
= page_waitqueue(page
);
1050 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, false);
1052 EXPORT_SYMBOL(wait_on_page_bit
);
1054 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1056 wait_queue_head_t
*q
= page_waitqueue(page
);
1057 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, false);
1061 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1062 * @page: Page defining the wait queue of interest
1063 * @waiter: Waiter to add to the queue
1065 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1067 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1069 wait_queue_head_t
*q
= page_waitqueue(page
);
1070 unsigned long flags
;
1072 spin_lock_irqsave(&q
->lock
, flags
);
1073 __add_wait_queue_entry_tail(q
, waiter
);
1074 SetPageWaiters(page
);
1075 spin_unlock_irqrestore(&q
->lock
, flags
);
1077 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1079 #ifndef clear_bit_unlock_is_negative_byte
1082 * PG_waiters is the high bit in the same byte as PG_lock.
1084 * On x86 (and on many other architectures), we can clear PG_lock and
1085 * test the sign bit at the same time. But if the architecture does
1086 * not support that special operation, we just do this all by hand
1089 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1090 * being cleared, but a memory barrier should be unneccssary since it is
1091 * in the same byte as PG_locked.
1093 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1095 clear_bit_unlock(nr
, mem
);
1096 /* smp_mb__after_atomic(); */
1097 return test_bit(PG_waiters
, mem
);
1103 * unlock_page - unlock a locked page
1106 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1107 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1108 * mechanism between PageLocked pages and PageWriteback pages is shared.
1109 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1111 * Note that this depends on PG_waiters being the sign bit in the byte
1112 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1113 * clear the PG_locked bit and test PG_waiters at the same time fairly
1114 * portably (architectures that do LL/SC can test any bit, while x86 can
1115 * test the sign bit).
1117 void unlock_page(struct page
*page
)
1119 BUILD_BUG_ON(PG_waiters
!= 7);
1120 page
= compound_head(page
);
1121 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1122 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1123 wake_up_page_bit(page
, PG_locked
);
1125 EXPORT_SYMBOL(unlock_page
);
1128 * end_page_writeback - end writeback against a page
1131 void end_page_writeback(struct page
*page
)
1134 * TestClearPageReclaim could be used here but it is an atomic
1135 * operation and overkill in this particular case. Failing to
1136 * shuffle a page marked for immediate reclaim is too mild to
1137 * justify taking an atomic operation penalty at the end of
1138 * ever page writeback.
1140 if (PageReclaim(page
)) {
1141 ClearPageReclaim(page
);
1142 rotate_reclaimable_page(page
);
1145 if (!test_clear_page_writeback(page
))
1148 smp_mb__after_atomic();
1149 wake_up_page(page
, PG_writeback
);
1151 EXPORT_SYMBOL(end_page_writeback
);
1154 * After completing I/O on a page, call this routine to update the page
1155 * flags appropriately
1157 void page_endio(struct page
*page
, bool is_write
, int err
)
1161 SetPageUptodate(page
);
1163 ClearPageUptodate(page
);
1169 struct address_space
*mapping
;
1172 mapping
= page_mapping(page
);
1174 mapping_set_error(mapping
, err
);
1176 end_page_writeback(page
);
1179 EXPORT_SYMBOL_GPL(page_endio
);
1182 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1183 * @__page: the page to lock
1185 void __lock_page(struct page
*__page
)
1187 struct page
*page
= compound_head(__page
);
1188 wait_queue_head_t
*q
= page_waitqueue(page
);
1189 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, true);
1191 EXPORT_SYMBOL(__lock_page
);
1193 int __lock_page_killable(struct page
*__page
)
1195 struct page
*page
= compound_head(__page
);
1196 wait_queue_head_t
*q
= page_waitqueue(page
);
1197 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
, true);
1199 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1203 * 1 - page is locked; mmap_sem is still held.
1204 * 0 - page is not locked.
1205 * mmap_sem has been released (up_read()), unless flags had both
1206 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1207 * which case mmap_sem is still held.
1209 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1210 * with the page locked and the mmap_sem unperturbed.
1212 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1215 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1217 * CAUTION! In this case, mmap_sem is not released
1218 * even though return 0.
1220 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1223 up_read(&mm
->mmap_sem
);
1224 if (flags
& FAULT_FLAG_KILLABLE
)
1225 wait_on_page_locked_killable(page
);
1227 wait_on_page_locked(page
);
1230 if (flags
& FAULT_FLAG_KILLABLE
) {
1233 ret
= __lock_page_killable(page
);
1235 up_read(&mm
->mmap_sem
);
1245 * page_cache_next_hole - find the next hole (not-present entry)
1248 * @max_scan: maximum range to search
1250 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1251 * lowest indexed hole.
1253 * Returns: the index of the hole if found, otherwise returns an index
1254 * outside of the set specified (in which case 'return - index >=
1255 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1258 * page_cache_next_hole may be called under rcu_read_lock. However,
1259 * like radix_tree_gang_lookup, this will not atomically search a
1260 * snapshot of the tree at a single point in time. For example, if a
1261 * hole is created at index 5, then subsequently a hole is created at
1262 * index 10, page_cache_next_hole covering both indexes may return 10
1263 * if called under rcu_read_lock.
1265 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1266 pgoff_t index
, unsigned long max_scan
)
1270 for (i
= 0; i
< max_scan
; i
++) {
1273 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1274 if (!page
|| radix_tree_exceptional_entry(page
))
1283 EXPORT_SYMBOL(page_cache_next_hole
);
1286 * page_cache_prev_hole - find the prev hole (not-present entry)
1289 * @max_scan: maximum range to search
1291 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1294 * Returns: the index of the hole if found, otherwise returns an index
1295 * outside of the set specified (in which case 'index - return >=
1296 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1299 * page_cache_prev_hole may be called under rcu_read_lock. However,
1300 * like radix_tree_gang_lookup, this will not atomically search a
1301 * snapshot of the tree at a single point in time. For example, if a
1302 * hole is created at index 10, then subsequently a hole is created at
1303 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1304 * called under rcu_read_lock.
1306 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1307 pgoff_t index
, unsigned long max_scan
)
1311 for (i
= 0; i
< max_scan
; i
++) {
1314 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1315 if (!page
|| radix_tree_exceptional_entry(page
))
1318 if (index
== ULONG_MAX
)
1324 EXPORT_SYMBOL(page_cache_prev_hole
);
1327 * find_get_entry - find and get a page cache entry
1328 * @mapping: the address_space to search
1329 * @offset: the page cache index
1331 * Looks up the page cache slot at @mapping & @offset. If there is a
1332 * page cache page, it is returned with an increased refcount.
1334 * If the slot holds a shadow entry of a previously evicted page, or a
1335 * swap entry from shmem/tmpfs, it is returned.
1337 * Otherwise, %NULL is returned.
1339 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1342 struct page
*head
, *page
;
1347 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1349 page
= radix_tree_deref_slot(pagep
);
1350 if (unlikely(!page
))
1352 if (radix_tree_exception(page
)) {
1353 if (radix_tree_deref_retry(page
))
1356 * A shadow entry of a recently evicted page,
1357 * or a swap entry from shmem/tmpfs. Return
1358 * it without attempting to raise page count.
1363 head
= compound_head(page
);
1364 if (!page_cache_get_speculative(head
))
1367 /* The page was split under us? */
1368 if (compound_head(page
) != head
) {
1374 * Has the page moved?
1375 * This is part of the lockless pagecache protocol. See
1376 * include/linux/pagemap.h for details.
1378 if (unlikely(page
!= *pagep
)) {
1388 EXPORT_SYMBOL(find_get_entry
);
1391 * find_lock_entry - locate, pin and lock a page cache entry
1392 * @mapping: the address_space to search
1393 * @offset: the page cache index
1395 * Looks up the page cache slot at @mapping & @offset. If there is a
1396 * page cache page, it is returned locked and with an increased
1399 * If the slot holds a shadow entry of a previously evicted page, or a
1400 * swap entry from shmem/tmpfs, it is returned.
1402 * Otherwise, %NULL is returned.
1404 * find_lock_entry() may sleep.
1406 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1411 page
= find_get_entry(mapping
, offset
);
1412 if (page
&& !radix_tree_exception(page
)) {
1414 /* Has the page been truncated? */
1415 if (unlikely(page_mapping(page
) != mapping
)) {
1420 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1424 EXPORT_SYMBOL(find_lock_entry
);
1427 * pagecache_get_page - find and get a page reference
1428 * @mapping: the address_space to search
1429 * @offset: the page index
1430 * @fgp_flags: PCG flags
1431 * @gfp_mask: gfp mask to use for the page cache data page allocation
1433 * Looks up the page cache slot at @mapping & @offset.
1435 * PCG flags modify how the page is returned.
1437 * @fgp_flags can be:
1439 * - FGP_ACCESSED: the page will be marked accessed
1440 * - FGP_LOCK: Page is return locked
1441 * - FGP_CREAT: If page is not present then a new page is allocated using
1442 * @gfp_mask and added to the page cache and the VM's LRU
1443 * list. The page is returned locked and with an increased
1444 * refcount. Otherwise, NULL is returned.
1446 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1447 * if the GFP flags specified for FGP_CREAT are atomic.
1449 * If there is a page cache page, it is returned with an increased refcount.
1451 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1452 int fgp_flags
, gfp_t gfp_mask
)
1457 page
= find_get_entry(mapping
, offset
);
1458 if (radix_tree_exceptional_entry(page
))
1463 if (fgp_flags
& FGP_LOCK
) {
1464 if (fgp_flags
& FGP_NOWAIT
) {
1465 if (!trylock_page(page
)) {
1473 /* Has the page been truncated? */
1474 if (unlikely(page
->mapping
!= mapping
)) {
1479 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1482 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1483 mark_page_accessed(page
);
1486 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1488 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1489 gfp_mask
|= __GFP_WRITE
;
1490 if (fgp_flags
& FGP_NOFS
)
1491 gfp_mask
&= ~__GFP_FS
;
1493 page
= __page_cache_alloc(gfp_mask
);
1497 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1498 fgp_flags
|= FGP_LOCK
;
1500 /* Init accessed so avoid atomic mark_page_accessed later */
1501 if (fgp_flags
& FGP_ACCESSED
)
1502 __SetPageReferenced(page
);
1504 err
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
1505 if (unlikely(err
)) {
1515 EXPORT_SYMBOL(pagecache_get_page
);
1518 * find_get_entries - gang pagecache lookup
1519 * @mapping: The address_space to search
1520 * @start: The starting page cache index
1521 * @nr_entries: The maximum number of entries
1522 * @entries: Where the resulting entries are placed
1523 * @indices: The cache indices corresponding to the entries in @entries
1525 * find_get_entries() will search for and return a group of up to
1526 * @nr_entries entries in the mapping. The entries are placed at
1527 * @entries. find_get_entries() takes a reference against any actual
1530 * The search returns a group of mapping-contiguous page cache entries
1531 * with ascending indexes. There may be holes in the indices due to
1532 * not-present pages.
1534 * Any shadow entries of evicted pages, or swap entries from
1535 * shmem/tmpfs, are included in the returned array.
1537 * find_get_entries() returns the number of pages and shadow entries
1540 unsigned find_get_entries(struct address_space
*mapping
,
1541 pgoff_t start
, unsigned int nr_entries
,
1542 struct page
**entries
, pgoff_t
*indices
)
1545 unsigned int ret
= 0;
1546 struct radix_tree_iter iter
;
1552 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1553 struct page
*head
, *page
;
1555 page
= radix_tree_deref_slot(slot
);
1556 if (unlikely(!page
))
1558 if (radix_tree_exception(page
)) {
1559 if (radix_tree_deref_retry(page
)) {
1560 slot
= radix_tree_iter_retry(&iter
);
1564 * A shadow entry of a recently evicted page, a swap
1565 * entry from shmem/tmpfs or a DAX entry. Return it
1566 * without attempting to raise page count.
1571 head
= compound_head(page
);
1572 if (!page_cache_get_speculative(head
))
1575 /* The page was split under us? */
1576 if (compound_head(page
) != head
) {
1581 /* Has the page moved? */
1582 if (unlikely(page
!= *slot
)) {
1587 indices
[ret
] = iter
.index
;
1588 entries
[ret
] = page
;
1589 if (++ret
== nr_entries
)
1597 * find_get_pages_range - gang pagecache lookup
1598 * @mapping: The address_space to search
1599 * @start: The starting page index
1600 * @end: The final page index (inclusive)
1601 * @nr_pages: The maximum number of pages
1602 * @pages: Where the resulting pages are placed
1604 * find_get_pages_range() will search for and return a group of up to @nr_pages
1605 * pages in the mapping starting at index @start and up to index @end
1606 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1607 * a reference against the returned pages.
1609 * The search returns a group of mapping-contiguous pages with ascending
1610 * indexes. There may be holes in the indices due to not-present pages.
1611 * We also update @start to index the next page for the traversal.
1613 * find_get_pages_range() returns the number of pages which were found. If this
1614 * number is smaller than @nr_pages, the end of specified range has been
1617 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
1618 pgoff_t end
, unsigned int nr_pages
,
1619 struct page
**pages
)
1621 struct radix_tree_iter iter
;
1625 if (unlikely(!nr_pages
))
1629 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, *start
) {
1630 struct page
*head
, *page
;
1632 if (iter
.index
> end
)
1635 page
= radix_tree_deref_slot(slot
);
1636 if (unlikely(!page
))
1639 if (radix_tree_exception(page
)) {
1640 if (radix_tree_deref_retry(page
)) {
1641 slot
= radix_tree_iter_retry(&iter
);
1645 * A shadow entry of a recently evicted page,
1646 * or a swap entry from shmem/tmpfs. Skip
1652 head
= compound_head(page
);
1653 if (!page_cache_get_speculative(head
))
1656 /* The page was split under us? */
1657 if (compound_head(page
) != head
) {
1662 /* Has the page moved? */
1663 if (unlikely(page
!= *slot
)) {
1669 if (++ret
== nr_pages
) {
1670 *start
= pages
[ret
- 1]->index
+ 1;
1676 * We come here when there is no page beyond @end. We take care to not
1677 * overflow the index @start as it confuses some of the callers. This
1678 * breaks the iteration when there is page at index -1 but that is
1679 * already broken anyway.
1681 if (end
== (pgoff_t
)-1)
1682 *start
= (pgoff_t
)-1;
1692 * find_get_pages_contig - gang contiguous pagecache lookup
1693 * @mapping: The address_space to search
1694 * @index: The starting page index
1695 * @nr_pages: The maximum number of pages
1696 * @pages: Where the resulting pages are placed
1698 * find_get_pages_contig() works exactly like find_get_pages(), except
1699 * that the returned number of pages are guaranteed to be contiguous.
1701 * find_get_pages_contig() returns the number of pages which were found.
1703 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1704 unsigned int nr_pages
, struct page
**pages
)
1706 struct radix_tree_iter iter
;
1708 unsigned int ret
= 0;
1710 if (unlikely(!nr_pages
))
1714 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1715 struct page
*head
, *page
;
1717 page
= radix_tree_deref_slot(slot
);
1718 /* The hole, there no reason to continue */
1719 if (unlikely(!page
))
1722 if (radix_tree_exception(page
)) {
1723 if (radix_tree_deref_retry(page
)) {
1724 slot
= radix_tree_iter_retry(&iter
);
1728 * A shadow entry of a recently evicted page,
1729 * or a swap entry from shmem/tmpfs. Stop
1730 * looking for contiguous pages.
1735 head
= compound_head(page
);
1736 if (!page_cache_get_speculative(head
))
1739 /* The page was split under us? */
1740 if (compound_head(page
) != head
) {
1745 /* Has the page moved? */
1746 if (unlikely(page
!= *slot
)) {
1752 * must check mapping and index after taking the ref.
1753 * otherwise we can get both false positives and false
1754 * negatives, which is just confusing to the caller.
1756 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1762 if (++ret
== nr_pages
)
1768 EXPORT_SYMBOL(find_get_pages_contig
);
1771 * find_get_pages_tag - find and return pages that match @tag
1772 * @mapping: the address_space to search
1773 * @index: the starting page index
1774 * @tag: the tag index
1775 * @nr_pages: the maximum number of pages
1776 * @pages: where the resulting pages are placed
1778 * Like find_get_pages, except we only return pages which are tagged with
1779 * @tag. We update @index to index the next page for the traversal.
1781 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1782 int tag
, unsigned int nr_pages
, struct page
**pages
)
1784 struct radix_tree_iter iter
;
1788 if (unlikely(!nr_pages
))
1792 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1793 &iter
, *index
, tag
) {
1794 struct page
*head
, *page
;
1796 page
= radix_tree_deref_slot(slot
);
1797 if (unlikely(!page
))
1800 if (radix_tree_exception(page
)) {
1801 if (radix_tree_deref_retry(page
)) {
1802 slot
= radix_tree_iter_retry(&iter
);
1806 * A shadow entry of a recently evicted page.
1808 * Those entries should never be tagged, but
1809 * this tree walk is lockless and the tags are
1810 * looked up in bulk, one radix tree node at a
1811 * time, so there is a sizable window for page
1812 * reclaim to evict a page we saw tagged.
1819 head
= compound_head(page
);
1820 if (!page_cache_get_speculative(head
))
1823 /* The page was split under us? */
1824 if (compound_head(page
) != head
) {
1829 /* Has the page moved? */
1830 if (unlikely(page
!= *slot
)) {
1836 if (++ret
== nr_pages
)
1843 *index
= pages
[ret
- 1]->index
+ 1;
1847 EXPORT_SYMBOL(find_get_pages_tag
);
1850 * find_get_entries_tag - find and return entries that match @tag
1851 * @mapping: the address_space to search
1852 * @start: the starting page cache index
1853 * @tag: the tag index
1854 * @nr_entries: the maximum number of entries
1855 * @entries: where the resulting entries are placed
1856 * @indices: the cache indices corresponding to the entries in @entries
1858 * Like find_get_entries, except we only return entries which are tagged with
1861 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1862 int tag
, unsigned int nr_entries
,
1863 struct page
**entries
, pgoff_t
*indices
)
1866 unsigned int ret
= 0;
1867 struct radix_tree_iter iter
;
1873 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1874 &iter
, start
, tag
) {
1875 struct page
*head
, *page
;
1877 page
= radix_tree_deref_slot(slot
);
1878 if (unlikely(!page
))
1880 if (radix_tree_exception(page
)) {
1881 if (radix_tree_deref_retry(page
)) {
1882 slot
= radix_tree_iter_retry(&iter
);
1887 * A shadow entry of a recently evicted page, a swap
1888 * entry from shmem/tmpfs or a DAX entry. Return it
1889 * without attempting to raise page count.
1894 head
= compound_head(page
);
1895 if (!page_cache_get_speculative(head
))
1898 /* The page was split under us? */
1899 if (compound_head(page
) != head
) {
1904 /* Has the page moved? */
1905 if (unlikely(page
!= *slot
)) {
1910 indices
[ret
] = iter
.index
;
1911 entries
[ret
] = page
;
1912 if (++ret
== nr_entries
)
1918 EXPORT_SYMBOL(find_get_entries_tag
);
1921 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1922 * a _large_ part of the i/o request. Imagine the worst scenario:
1924 * ---R__________________________________________B__________
1925 * ^ reading here ^ bad block(assume 4k)
1927 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1928 * => failing the whole request => read(R) => read(R+1) =>
1929 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1930 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1931 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1933 * It is going insane. Fix it by quickly scaling down the readahead size.
1935 static void shrink_readahead_size_eio(struct file
*filp
,
1936 struct file_ra_state
*ra
)
1942 * generic_file_buffered_read - generic file read routine
1943 * @iocb: the iocb to read
1944 * @iter: data destination
1945 * @written: already copied
1947 * This is a generic file read routine, and uses the
1948 * mapping->a_ops->readpage() function for the actual low-level stuff.
1950 * This is really ugly. But the goto's actually try to clarify some
1951 * of the logic when it comes to error handling etc.
1953 static ssize_t
generic_file_buffered_read(struct kiocb
*iocb
,
1954 struct iov_iter
*iter
, ssize_t written
)
1956 struct file
*filp
= iocb
->ki_filp
;
1957 struct address_space
*mapping
= filp
->f_mapping
;
1958 struct inode
*inode
= mapping
->host
;
1959 struct file_ra_state
*ra
= &filp
->f_ra
;
1960 loff_t
*ppos
= &iocb
->ki_pos
;
1964 unsigned long offset
; /* offset into pagecache page */
1965 unsigned int prev_offset
;
1968 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
1970 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
1972 index
= *ppos
>> PAGE_SHIFT
;
1973 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1974 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1975 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1976 offset
= *ppos
& ~PAGE_MASK
;
1982 unsigned long nr
, ret
;
1988 if (fatal_signal_pending(current
)) {
1993 page
= find_get_page(mapping
, index
);
1995 if (iocb
->ki_flags
& IOCB_NOWAIT
)
1997 mm_event_start(&event_ts
);
1998 page_cache_sync_readahead(mapping
,
2000 index
, last_index
- index
);
2001 page
= find_get_page(mapping
, index
);
2002 if (unlikely(page
== NULL
))
2003 goto no_cached_page
;
2005 if (PageReadahead(page
)) {
2006 page_cache_async_readahead(mapping
,
2008 index
, last_index
- index
);
2010 if (!PageUptodate(page
)) {
2011 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2017 * See comment in do_read_cache_page on why
2018 * wait_on_page_locked is used to avoid unnecessarily
2019 * serialisations and why it's safe.
2021 error
= wait_on_page_locked_killable(page
);
2022 if (unlikely(error
))
2023 goto readpage_error
;
2024 if (PageUptodate(page
))
2027 if (inode
->i_blkbits
== PAGE_SHIFT
||
2028 !mapping
->a_ops
->is_partially_uptodate
)
2029 goto page_not_up_to_date
;
2030 /* pipes can't handle partially uptodate pages */
2031 if (unlikely(iter
->type
& ITER_PIPE
))
2032 goto page_not_up_to_date
;
2033 if (!trylock_page(page
))
2034 goto page_not_up_to_date
;
2035 /* Did it get truncated before we got the lock? */
2037 goto page_not_up_to_date_locked
;
2038 if (!mapping
->a_ops
->is_partially_uptodate(page
,
2039 offset
, iter
->count
))
2040 goto page_not_up_to_date_locked
;
2045 mm_event_end(MM_READ_IO
, event_ts
);
2047 * i_size must be checked after we know the page is Uptodate.
2049 * Checking i_size after the check allows us to calculate
2050 * the correct value for "nr", which means the zero-filled
2051 * part of the page is not copied back to userspace (unless
2052 * another truncate extends the file - this is desired though).
2055 isize
= i_size_read(inode
);
2056 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2057 if (unlikely(!isize
|| index
> end_index
)) {
2062 /* nr is the maximum number of bytes to copy from this page */
2064 if (index
== end_index
) {
2065 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2073 /* If users can be writing to this page using arbitrary
2074 * virtual addresses, take care about potential aliasing
2075 * before reading the page on the kernel side.
2077 if (mapping_writably_mapped(mapping
))
2078 flush_dcache_page(page
);
2081 * When a sequential read accesses a page several times,
2082 * only mark it as accessed the first time.
2084 if (prev_index
!= index
|| offset
!= prev_offset
)
2085 mark_page_accessed(page
);
2089 * Ok, we have the page, and it's up-to-date, so
2090 * now we can copy it to user space...
2093 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
2095 index
+= offset
>> PAGE_SHIFT
;
2096 offset
&= ~PAGE_MASK
;
2097 prev_offset
= offset
;
2101 if (!iov_iter_count(iter
))
2109 page_not_up_to_date
:
2110 /* Get exclusive access to the page ... */
2111 error
= lock_page_killable(page
);
2112 if (unlikely(error
))
2113 goto readpage_error
;
2115 page_not_up_to_date_locked
:
2116 /* Did it get truncated before we got the lock? */
2117 if (!page
->mapping
) {
2123 /* Did somebody else fill it already? */
2124 if (PageUptodate(page
)) {
2131 * A previous I/O error may have been due to temporary
2132 * failures, eg. multipath errors.
2133 * PG_error will be set again if readpage fails.
2135 ClearPageError(page
);
2136 /* Start the actual read. The read will unlock the page. */
2137 error
= mapping
->a_ops
->readpage(filp
, page
);
2139 if (unlikely(error
)) {
2140 if (error
== AOP_TRUNCATED_PAGE
) {
2145 goto readpage_error
;
2148 if (!PageUptodate(page
)) {
2149 error
= lock_page_killable(page
);
2150 if (unlikely(error
))
2151 goto readpage_error
;
2152 if (!PageUptodate(page
)) {
2153 if (page
->mapping
== NULL
) {
2155 * invalidate_mapping_pages got it
2162 shrink_readahead_size_eio(filp
, ra
);
2164 goto readpage_error
;
2172 /* UHHUH! A synchronous read error occurred. Report it */
2178 * Ok, it wasn't cached, so we need to create a new
2181 page
= page_cache_alloc_cold(mapping
);
2186 error
= add_to_page_cache_lru(page
, mapping
, index
,
2187 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2190 if (error
== -EEXIST
) {
2202 ra
->prev_pos
= prev_index
;
2203 ra
->prev_pos
<<= PAGE_SHIFT
;
2204 ra
->prev_pos
|= prev_offset
;
2206 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2207 file_accessed(filp
);
2208 return written
? written
: error
;
2212 * generic_file_read_iter - generic filesystem read routine
2213 * @iocb: kernel I/O control block
2214 * @iter: destination for the data read
2216 * This is the "read_iter()" routine for all filesystems
2217 * that can use the page cache directly.
2220 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2222 size_t count
= iov_iter_count(iter
);
2226 goto out
; /* skip atime */
2228 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2229 struct file
*file
= iocb
->ki_filp
;
2230 struct address_space
*mapping
= file
->f_mapping
;
2231 struct inode
*inode
= mapping
->host
;
2234 size
= i_size_read(inode
);
2235 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2236 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2237 iocb
->ki_pos
+ count
- 1))
2240 retval
= filemap_write_and_wait_range(mapping
,
2242 iocb
->ki_pos
+ count
- 1);
2247 file_accessed(file
);
2249 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2251 iocb
->ki_pos
+= retval
;
2254 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2257 * Btrfs can have a short DIO read if we encounter
2258 * compressed extents, so if there was an error, or if
2259 * we've already read everything we wanted to, or if
2260 * there was a short read because we hit EOF, go ahead
2261 * and return. Otherwise fallthrough to buffered io for
2262 * the rest of the read. Buffered reads will not work for
2263 * DAX files, so don't bother trying.
2265 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2270 retval
= generic_file_buffered_read(iocb
, iter
, retval
);
2274 EXPORT_SYMBOL(generic_file_read_iter
);
2278 * page_cache_read - adds requested page to the page cache if not already there
2279 * @file: file to read
2280 * @offset: page index
2281 * @gfp_mask: memory allocation flags
2283 * This adds the requested page to the page cache if it isn't already there,
2284 * and schedules an I/O to read in its contents from disk.
2286 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
2288 struct address_space
*mapping
= file
->f_mapping
;
2293 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
2297 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
2299 ret
= mapping
->a_ops
->readpage(file
, page
);
2300 else if (ret
== -EEXIST
)
2301 ret
= 0; /* losing race to add is OK */
2305 } while (ret
== AOP_TRUNCATED_PAGE
);
2310 #define MMAP_LOTSAMISS (100)
2313 * Synchronous readahead happens when we don't even find
2314 * a page in the page cache at all.
2316 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2317 struct file_ra_state
*ra
,
2321 struct address_space
*mapping
= file
->f_mapping
;
2323 /* If we don't want any read-ahead, don't bother */
2324 if (vma
->vm_flags
& VM_RAND_READ
)
2329 if (vma
->vm_flags
& VM_SEQ_READ
) {
2330 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2335 /* Avoid banging the cache line if not needed */
2336 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2340 * Do we miss much more than hit in this file? If so,
2341 * stop bothering with read-ahead. It will only hurt.
2343 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2349 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2350 ra
->size
= ra
->ra_pages
;
2351 ra
->async_size
= ra
->ra_pages
/ 4;
2352 ra_submit(ra
, mapping
, file
);
2356 * Asynchronous readahead happens when we find the page and PG_readahead,
2357 * so we want to possibly extend the readahead further..
2359 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2360 struct file_ra_state
*ra
,
2365 struct address_space
*mapping
= file
->f_mapping
;
2367 /* If we don't want any read-ahead, don't bother */
2368 if (vma
->vm_flags
& VM_RAND_READ
)
2370 if (ra
->mmap_miss
> 0)
2372 if (PageReadahead(page
))
2373 page_cache_async_readahead(mapping
, ra
, file
,
2374 page
, offset
, ra
->ra_pages
);
2378 * filemap_fault - read in file data for page fault handling
2379 * @vmf: struct vm_fault containing details of the fault
2381 * filemap_fault() is invoked via the vma operations vector for a
2382 * mapped memory region to read in file data during a page fault.
2384 * The goto's are kind of ugly, but this streamlines the normal case of having
2385 * it in the page cache, and handles the special cases reasonably without
2386 * having a lot of duplicated code.
2388 * vma->vm_mm->mmap_sem must be held on entry.
2390 * If our return value has VM_FAULT_RETRY set, it's because
2391 * lock_page_or_retry() returned 0.
2392 * The mmap_sem has usually been released in this case.
2393 * See __lock_page_or_retry() for the exception.
2395 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2396 * has not been released.
2398 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2400 int filemap_fault(struct vm_fault
*vmf
)
2403 struct file
*file
= vmf
->vma
->vm_file
;
2404 struct address_space
*mapping
= file
->f_mapping
;
2405 struct file_ra_state
*ra
= &file
->f_ra
;
2406 struct inode
*inode
= mapping
->host
;
2407 pgoff_t offset
= vmf
->pgoff
;
2412 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2413 if (unlikely(offset
>= max_off
))
2414 return VM_FAULT_SIGBUS
;
2417 * Do we have something in the page cache already?
2419 page
= find_get_page(mapping
, offset
);
2420 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2422 * We found the page, so try async readahead before
2423 * waiting for the lock.
2425 do_async_mmap_readahead(vmf
->vma
, ra
, file
, page
, offset
);
2427 /* No page in the page cache at all */
2428 do_sync_mmap_readahead(vmf
->vma
, ra
, file
, offset
);
2429 count_vm_event(PGMAJFAULT
);
2430 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2431 ret
= VM_FAULT_MAJOR
;
2433 page
= find_get_page(mapping
, offset
);
2435 goto no_cached_page
;
2438 if (!lock_page_or_retry(page
, vmf
->vma
->vm_mm
, vmf
->flags
)) {
2440 return ret
| VM_FAULT_RETRY
;
2443 /* Did it get truncated? */
2444 if (unlikely(page
->mapping
!= mapping
)) {
2449 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2452 * We have a locked page in the page cache, now we need to check
2453 * that it's up-to-date. If not, it is going to be due to an error.
2455 if (unlikely(!PageUptodate(page
)))
2456 goto page_not_uptodate
;
2459 * Found the page and have a reference on it.
2460 * We must recheck i_size under page lock.
2462 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2463 if (unlikely(offset
>= max_off
)) {
2466 return VM_FAULT_SIGBUS
;
2470 return ret
| VM_FAULT_LOCKED
;
2474 * We're only likely to ever get here if MADV_RANDOM is in
2477 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2480 * The page we want has now been added to the page cache.
2481 * In the unlikely event that someone removed it in the
2482 * meantime, we'll just come back here and read it again.
2488 * An error return from page_cache_read can result if the
2489 * system is low on memory, or a problem occurs while trying
2492 if (error
== -ENOMEM
)
2493 return VM_FAULT_OOM
;
2494 return VM_FAULT_SIGBUS
;
2498 * Umm, take care of errors if the page isn't up-to-date.
2499 * Try to re-read it _once_. We do this synchronously,
2500 * because there really aren't any performance issues here
2501 * and we need to check for errors.
2503 ClearPageError(page
);
2504 error
= mapping
->a_ops
->readpage(file
, page
);
2506 wait_on_page_locked(page
);
2507 if (!PageUptodate(page
))
2512 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2515 /* Things didn't work out. Return zero to tell the mm layer so. */
2516 shrink_readahead_size_eio(file
, ra
);
2517 return VM_FAULT_SIGBUS
;
2519 EXPORT_SYMBOL(filemap_fault
);
2521 void filemap_map_pages(struct vm_fault
*vmf
,
2522 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2524 struct radix_tree_iter iter
;
2526 struct file
*file
= vmf
->vma
->vm_file
;
2527 struct address_space
*mapping
= file
->f_mapping
;
2528 pgoff_t last_pgoff
= start_pgoff
;
2529 unsigned long max_idx
;
2530 struct page
*head
, *page
;
2533 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2535 if (iter
.index
> end_pgoff
)
2538 page
= radix_tree_deref_slot(slot
);
2539 if (unlikely(!page
))
2541 if (radix_tree_exception(page
)) {
2542 if (radix_tree_deref_retry(page
)) {
2543 slot
= radix_tree_iter_retry(&iter
);
2549 head
= compound_head(page
);
2550 if (!page_cache_get_speculative(head
))
2553 /* The page was split under us? */
2554 if (compound_head(page
) != head
) {
2559 /* Has the page moved? */
2560 if (unlikely(page
!= *slot
)) {
2565 if (!PageUptodate(page
) ||
2566 PageReadahead(page
) ||
2569 if (!trylock_page(page
))
2572 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2575 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2576 if (page
->index
>= max_idx
)
2579 if (file
->f_ra
.mmap_miss
> 0)
2580 file
->f_ra
.mmap_miss
--;
2582 vmf
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2584 vmf
->pte
+= iter
.index
- last_pgoff
;
2585 last_pgoff
= iter
.index
;
2586 if (alloc_set_pte(vmf
, NULL
, page
))
2595 /* Huge page is mapped? No need to proceed. */
2596 if (pmd_trans_huge(*vmf
->pmd
))
2598 if (iter
.index
== end_pgoff
)
2603 EXPORT_SYMBOL(filemap_map_pages
);
2605 int filemap_page_mkwrite(struct vm_fault
*vmf
)
2607 struct page
*page
= vmf
->page
;
2608 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2609 int ret
= VM_FAULT_LOCKED
;
2611 sb_start_pagefault(inode
->i_sb
);
2612 file_update_time(vmf
->vma
->vm_file
);
2614 if (page
->mapping
!= inode
->i_mapping
) {
2616 ret
= VM_FAULT_NOPAGE
;
2620 * We mark the page dirty already here so that when freeze is in
2621 * progress, we are guaranteed that writeback during freezing will
2622 * see the dirty page and writeprotect it again.
2624 set_page_dirty(page
);
2625 wait_for_stable_page(page
);
2627 sb_end_pagefault(inode
->i_sb
);
2630 EXPORT_SYMBOL(filemap_page_mkwrite
);
2632 const struct vm_operations_struct generic_file_vm_ops
= {
2633 .fault
= filemap_fault
,
2634 .map_pages
= filemap_map_pages
,
2635 .page_mkwrite
= filemap_page_mkwrite
,
2638 /* This is used for a general mmap of a disk file */
2640 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2642 struct address_space
*mapping
= file
->f_mapping
;
2644 if (!mapping
->a_ops
->readpage
)
2646 file_accessed(file
);
2647 vma
->vm_ops
= &generic_file_vm_ops
;
2652 * This is for filesystems which do not implement ->writepage.
2654 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2656 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2658 return generic_file_mmap(file
, vma
);
2661 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2665 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2669 #endif /* CONFIG_MMU */
2671 EXPORT_SYMBOL(generic_file_mmap
);
2672 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2674 static struct page
*wait_on_page_read(struct page
*page
)
2676 if (!IS_ERR(page
)) {
2677 wait_on_page_locked(page
);
2678 if (!PageUptodate(page
)) {
2680 page
= ERR_PTR(-EIO
);
2686 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2688 int (*filler
)(struct file
*, struct page
*),
2695 page
= find_get_page(mapping
, index
);
2697 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2699 return ERR_PTR(-ENOMEM
);
2700 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2701 if (unlikely(err
)) {
2705 /* Presumably ENOMEM for radix tree node */
2706 return ERR_PTR(err
);
2710 err
= filler(data
, page
);
2713 return ERR_PTR(err
);
2716 page
= wait_on_page_read(page
);
2721 if (PageUptodate(page
))
2725 * Page is not up to date and may be locked due one of the following
2726 * case a: Page is being filled and the page lock is held
2727 * case b: Read/write error clearing the page uptodate status
2728 * case c: Truncation in progress (page locked)
2729 * case d: Reclaim in progress
2731 * Case a, the page will be up to date when the page is unlocked.
2732 * There is no need to serialise on the page lock here as the page
2733 * is pinned so the lock gives no additional protection. Even if the
2734 * the page is truncated, the data is still valid if PageUptodate as
2735 * it's a race vs truncate race.
2736 * Case b, the page will not be up to date
2737 * Case c, the page may be truncated but in itself, the data may still
2738 * be valid after IO completes as it's a read vs truncate race. The
2739 * operation must restart if the page is not uptodate on unlock but
2740 * otherwise serialising on page lock to stabilise the mapping gives
2741 * no additional guarantees to the caller as the page lock is
2742 * released before return.
2743 * Case d, similar to truncation. If reclaim holds the page lock, it
2744 * will be a race with remove_mapping that determines if the mapping
2745 * is valid on unlock but otherwise the data is valid and there is
2746 * no need to serialise with page lock.
2748 * As the page lock gives no additional guarantee, we optimistically
2749 * wait on the page to be unlocked and check if it's up to date and
2750 * use the page if it is. Otherwise, the page lock is required to
2751 * distinguish between the different cases. The motivation is that we
2752 * avoid spurious serialisations and wakeups when multiple processes
2753 * wait on the same page for IO to complete.
2755 wait_on_page_locked(page
);
2756 if (PageUptodate(page
))
2759 /* Distinguish between all the cases under the safety of the lock */
2762 /* Case c or d, restart the operation */
2763 if (!page
->mapping
) {
2769 /* Someone else locked and filled the page in a very small window */
2770 if (PageUptodate(page
)) {
2777 mark_page_accessed(page
);
2782 * read_cache_page - read into page cache, fill it if needed
2783 * @mapping: the page's address_space
2784 * @index: the page index
2785 * @filler: function to perform the read
2786 * @data: first arg to filler(data, page) function, often left as NULL
2788 * Read into the page cache. If a page already exists, and PageUptodate() is
2789 * not set, try to fill the page and wait for it to become unlocked.
2791 * If the page does not get brought uptodate, return -EIO.
2793 struct page
*read_cache_page(struct address_space
*mapping
,
2795 int (*filler
)(struct file
*, struct page
*),
2798 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2800 EXPORT_SYMBOL(read_cache_page
);
2803 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2804 * @mapping: the page's address_space
2805 * @index: the page index
2806 * @gfp: the page allocator flags to use if allocating
2808 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2809 * any new page allocations done using the specified allocation flags.
2811 * If the page does not get brought uptodate, return -EIO.
2813 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2817 filler_t
*filler
= mapping
->a_ops
->readpage
;
2819 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2821 EXPORT_SYMBOL(read_cache_page_gfp
);
2824 * Performs necessary checks before doing a write
2826 * Can adjust writing position or amount of bytes to write.
2827 * Returns appropriate error code that caller should return or
2828 * zero in case that write should be allowed.
2830 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2832 struct file
*file
= iocb
->ki_filp
;
2833 struct inode
*inode
= file
->f_mapping
->host
;
2834 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2837 if (!iov_iter_count(from
))
2840 /* FIXME: this is for backwards compatibility with 2.4 */
2841 if (iocb
->ki_flags
& IOCB_APPEND
)
2842 iocb
->ki_pos
= i_size_read(inode
);
2846 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
2849 if (limit
!= RLIM_INFINITY
) {
2850 if (iocb
->ki_pos
>= limit
) {
2851 send_sig(SIGXFSZ
, current
, 0);
2854 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2860 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2861 !(file
->f_flags
& O_LARGEFILE
))) {
2862 if (pos
>= MAX_NON_LFS
)
2864 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2868 * Are we about to exceed the fs block limit ?
2870 * If we have written data it becomes a short write. If we have
2871 * exceeded without writing data we send a signal and return EFBIG.
2872 * Linus frestrict idea will clean these up nicely..
2874 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2877 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2878 return iov_iter_count(from
);
2880 EXPORT_SYMBOL(generic_write_checks
);
2882 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2883 loff_t pos
, unsigned len
, unsigned flags
,
2884 struct page
**pagep
, void **fsdata
)
2886 const struct address_space_operations
*aops
= mapping
->a_ops
;
2888 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2891 EXPORT_SYMBOL(pagecache_write_begin
);
2893 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2894 loff_t pos
, unsigned len
, unsigned copied
,
2895 struct page
*page
, void *fsdata
)
2897 const struct address_space_operations
*aops
= mapping
->a_ops
;
2899 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2901 EXPORT_SYMBOL(pagecache_write_end
);
2904 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2906 struct file
*file
= iocb
->ki_filp
;
2907 struct address_space
*mapping
= file
->f_mapping
;
2908 struct inode
*inode
= mapping
->host
;
2909 loff_t pos
= iocb
->ki_pos
;
2914 write_len
= iov_iter_count(from
);
2915 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2917 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2918 /* If there are pages to writeback, return */
2919 if (filemap_range_has_page(inode
->i_mapping
, pos
,
2920 pos
+ iov_iter_count(from
)))
2923 written
= filemap_write_and_wait_range(mapping
, pos
,
2924 pos
+ write_len
- 1);
2930 * After a write we want buffered reads to be sure to go to disk to get
2931 * the new data. We invalidate clean cached page from the region we're
2932 * about to write. We do this *before* the write so that we can return
2933 * without clobbering -EIOCBQUEUED from ->direct_IO().
2935 written
= invalidate_inode_pages2_range(mapping
,
2936 pos
>> PAGE_SHIFT
, end
);
2938 * If a page can not be invalidated, return 0 to fall back
2939 * to buffered write.
2942 if (written
== -EBUSY
)
2947 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
2950 * Finally, try again to invalidate clean pages which might have been
2951 * cached by non-direct readahead, or faulted in by get_user_pages()
2952 * if the source of the write was an mmap'ed region of the file
2953 * we're writing. Either one is a pretty crazy thing to do,
2954 * so we don't support it 100%. If this invalidation
2955 * fails, tough, the write still worked...
2957 * Most of the time we do not need this since dio_complete() will do
2958 * the invalidation for us. However there are some file systems that
2959 * do not end up with dio_complete() being called, so let's not break
2960 * them by removing it completely
2962 if (mapping
->nrpages
)
2963 invalidate_inode_pages2_range(mapping
,
2964 pos
>> PAGE_SHIFT
, end
);
2968 write_len
-= written
;
2969 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2970 i_size_write(inode
, pos
);
2971 mark_inode_dirty(inode
);
2975 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
2979 EXPORT_SYMBOL(generic_file_direct_write
);
2982 * Find or create a page at the given pagecache position. Return the locked
2983 * page. This function is specifically for buffered writes.
2985 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2986 pgoff_t index
, unsigned flags
)
2989 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2991 if (flags
& AOP_FLAG_NOFS
)
2992 fgp_flags
|= FGP_NOFS
;
2994 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2995 mapping_gfp_mask(mapping
));
2997 wait_for_stable_page(page
);
3001 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3003 ssize_t
generic_perform_write(struct file
*file
,
3004 struct iov_iter
*i
, loff_t pos
)
3006 struct address_space
*mapping
= file
->f_mapping
;
3007 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3009 ssize_t written
= 0;
3010 unsigned int flags
= 0;
3014 unsigned long offset
; /* Offset into pagecache page */
3015 unsigned long bytes
; /* Bytes to write to page */
3016 size_t copied
; /* Bytes copied from user */
3019 offset
= (pos
& (PAGE_SIZE
- 1));
3020 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3025 * Bring in the user page that we will copy from _first_.
3026 * Otherwise there's a nasty deadlock on copying from the
3027 * same page as we're writing to, without it being marked
3030 * Not only is this an optimisation, but it is also required
3031 * to check that the address is actually valid, when atomic
3032 * usercopies are used, below.
3034 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3039 if (fatal_signal_pending(current
)) {
3044 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3046 if (unlikely(status
< 0))
3049 if (mapping_writably_mapped(mapping
))
3050 flush_dcache_page(page
);
3052 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3053 flush_dcache_page(page
);
3055 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3057 if (unlikely(status
< 0))
3063 iov_iter_advance(i
, copied
);
3064 if (unlikely(copied
== 0)) {
3066 * If we were unable to copy any data at all, we must
3067 * fall back to a single segment length write.
3069 * If we didn't fallback here, we could livelock
3070 * because not all segments in the iov can be copied at
3071 * once without a pagefault.
3073 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3074 iov_iter_single_seg_count(i
));
3080 balance_dirty_pages_ratelimited(mapping
);
3081 } while (iov_iter_count(i
));
3083 return written
? written
: status
;
3085 EXPORT_SYMBOL(generic_perform_write
);
3088 * __generic_file_write_iter - write data to a file
3089 * @iocb: IO state structure (file, offset, etc.)
3090 * @from: iov_iter with data to write
3092 * This function does all the work needed for actually writing data to a
3093 * file. It does all basic checks, removes SUID from the file, updates
3094 * modification times and calls proper subroutines depending on whether we
3095 * do direct IO or a standard buffered write.
3097 * It expects i_mutex to be grabbed unless we work on a block device or similar
3098 * object which does not need locking at all.
3100 * This function does *not* take care of syncing data in case of O_SYNC write.
3101 * A caller has to handle it. This is mainly due to the fact that we want to
3102 * avoid syncing under i_mutex.
3104 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3106 struct file
*file
= iocb
->ki_filp
;
3107 struct address_space
* mapping
= file
->f_mapping
;
3108 struct inode
*inode
= mapping
->host
;
3109 ssize_t written
= 0;
3113 /* We can write back this queue in page reclaim */
3114 current
->backing_dev_info
= inode_to_bdi(inode
);
3115 err
= file_remove_privs(file
);
3119 err
= file_update_time(file
);
3123 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3124 loff_t pos
, endbyte
;
3126 written
= generic_file_direct_write(iocb
, from
);
3128 * If the write stopped short of completing, fall back to
3129 * buffered writes. Some filesystems do this for writes to
3130 * holes, for example. For DAX files, a buffered write will
3131 * not succeed (even if it did, DAX does not handle dirty
3132 * page-cache pages correctly).
3134 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3137 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3139 * If generic_perform_write() returned a synchronous error
3140 * then we want to return the number of bytes which were
3141 * direct-written, or the error code if that was zero. Note
3142 * that this differs from normal direct-io semantics, which
3143 * will return -EFOO even if some bytes were written.
3145 if (unlikely(status
< 0)) {
3150 * We need to ensure that the page cache pages are written to
3151 * disk and invalidated to preserve the expected O_DIRECT
3154 endbyte
= pos
+ status
- 1;
3155 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3157 iocb
->ki_pos
= endbyte
+ 1;
3159 invalidate_mapping_pages(mapping
,
3161 endbyte
>> PAGE_SHIFT
);
3164 * We don't know how much we wrote, so just return
3165 * the number of bytes which were direct-written
3169 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3170 if (likely(written
> 0))
3171 iocb
->ki_pos
+= written
;
3174 current
->backing_dev_info
= NULL
;
3175 return written
? written
: err
;
3177 EXPORT_SYMBOL(__generic_file_write_iter
);
3180 * generic_file_write_iter - write data to a file
3181 * @iocb: IO state structure
3182 * @from: iov_iter with data to write
3184 * This is a wrapper around __generic_file_write_iter() to be used by most
3185 * filesystems. It takes care of syncing the file in case of O_SYNC file
3186 * and acquires i_mutex as needed.
3188 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3190 struct file
*file
= iocb
->ki_filp
;
3191 struct inode
*inode
= file
->f_mapping
->host
;
3195 ret
= generic_write_checks(iocb
, from
);
3197 ret
= __generic_file_write_iter(iocb
, from
);
3198 inode_unlock(inode
);
3201 ret
= generic_write_sync(iocb
, ret
);
3204 EXPORT_SYMBOL(generic_file_write_iter
);
3207 * try_to_release_page() - release old fs-specific metadata on a page
3209 * @page: the page which the kernel is trying to free
3210 * @gfp_mask: memory allocation flags (and I/O mode)
3212 * The address_space is to try to release any data against the page
3213 * (presumably at page->private). If the release was successful, return '1'.
3214 * Otherwise return zero.
3216 * This may also be called if PG_fscache is set on a page, indicating that the
3217 * page is known to the local caching routines.
3219 * The @gfp_mask argument specifies whether I/O may be performed to release
3220 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3223 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3225 struct address_space
* const mapping
= page
->mapping
;
3227 BUG_ON(!PageLocked(page
));
3228 if (PageWriteback(page
))
3231 if (mapping
&& mapping
->a_ops
->releasepage
)
3232 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3233 return try_to_free_buffers(page
);
3236 EXPORT_SYMBOL(try_to_release_page
);