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>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/filemap.h>
45 * FIXME: remove all knowledge of the buffer layer from the core VM
47 #include <linux/buffer_head.h> /* for try_to_free_buffers */
52 * Shared mappings implemented 30.11.1994. It's not fully working yet,
55 * Shared mappings now work. 15.8.1995 Bruno.
57 * finished 'unifying' the page and buffer cache and SMP-threaded the
58 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
60 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
66 * ->i_mmap_rwsem (truncate_pagecache)
67 * ->private_lock (__free_pte->__set_page_dirty_buffers)
68 * ->swap_lock (exclusive_swap_page, others)
69 * ->mapping->tree_lock
72 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
76 * ->page_table_lock or pte_lock (various, mainly in memory.c)
77 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
80 * ->lock_page (access_process_vm)
82 * ->i_mutex (generic_perform_write)
83 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
86 * sb_lock (fs/fs-writeback.c)
87 * ->mapping->tree_lock (__sync_single_inode)
90 * ->anon_vma.lock (vma_adjust)
93 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
95 * ->page_table_lock or pte_lock
96 * ->swap_lock (try_to_unmap_one)
97 * ->private_lock (try_to_unmap_one)
98 * ->tree_lock (try_to_unmap_one)
99 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
100 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
101 * ->private_lock (page_remove_rmap->set_page_dirty)
102 * ->tree_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
104 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
105 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
106 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
107 * ->inode->i_lock (zap_pte_range->set_page_dirty)
108 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
111 * ->tasklist_lock (memory_failure, collect_procs_ao)
114 static int page_cache_tree_insert(struct address_space
*mapping
,
115 struct page
*page
, void **shadowp
)
117 struct radix_tree_node
*node
;
121 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
128 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
129 if (!radix_tree_exceptional_entry(p
))
132 mapping
->nrexceptional
--;
133 if (!dax_mapping(mapping
)) {
137 /* DAX can replace empty locked entry with a hole */
139 dax_radix_locked_entry(0, RADIX_DAX_EMPTY
));
140 /* Wakeup waiters for exceptional entry lock */
141 dax_wake_mapping_entry_waiter(mapping
, page
->index
, p
,
145 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, page
,
146 workingset_update_node
, mapping
);
151 static void page_cache_tree_delete(struct address_space
*mapping
,
152 struct page
*page
, void *shadow
)
156 /* hugetlb pages are represented by one entry in the radix tree */
157 nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
159 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
160 VM_BUG_ON_PAGE(PageTail(page
), page
);
161 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
163 for (i
= 0; i
< nr
; i
++) {
164 struct radix_tree_node
*node
;
167 __radix_tree_lookup(&mapping
->page_tree
, page
->index
+ i
,
170 VM_BUG_ON_PAGE(!node
&& nr
!= 1, page
);
172 radix_tree_clear_tags(&mapping
->page_tree
, node
, slot
);
173 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, shadow
,
174 workingset_update_node
, mapping
);
178 mapping
->nrexceptional
+= nr
;
180 * Make sure the nrexceptional update is committed before
181 * the nrpages update so that final truncate racing
182 * with reclaim does not see both counters 0 at the
183 * same time and miss a shadow entry.
187 mapping
->nrpages
-= nr
;
191 * Delete a page from the page cache and free it. Caller has to make
192 * sure the page is locked and that nobody else uses it - or that usage
193 * is safe. The caller must hold the mapping's tree_lock.
195 void __delete_from_page_cache(struct page
*page
, void *shadow
)
197 struct address_space
*mapping
= page
->mapping
;
198 int nr
= hpage_nr_pages(page
);
200 trace_mm_filemap_delete_from_page_cache(page
);
202 * if we're uptodate, flush out into the cleancache, otherwise
203 * invalidate any existing cleancache entries. We can't leave
204 * stale data around in the cleancache once our page is gone
206 if (PageUptodate(page
) && PageMappedToDisk(page
))
207 cleancache_put_page(page
);
209 cleancache_invalidate_page(mapping
, page
);
211 VM_BUG_ON_PAGE(PageTail(page
), page
);
212 VM_BUG_ON_PAGE(page_mapped(page
), page
);
213 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
216 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
217 current
->comm
, page_to_pfn(page
));
218 dump_page(page
, "still mapped when deleted");
220 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
222 mapcount
= page_mapcount(page
);
223 if (mapping_exiting(mapping
) &&
224 page_count(page
) >= mapcount
+ 2) {
226 * All vmas have already been torn down, so it's
227 * a good bet that actually the page is unmapped,
228 * and we'd prefer not to leak it: if we're wrong,
229 * some other bad page check should catch it later.
231 page_mapcount_reset(page
);
232 page_ref_sub(page
, mapcount
);
236 page_cache_tree_delete(mapping
, page
, shadow
);
238 page
->mapping
= NULL
;
239 /* Leave page->index set: truncation lookup relies upon it */
241 /* hugetlb pages do not participate in page cache accounting. */
245 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
246 if (PageSwapBacked(page
)) {
247 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
248 if (PageTransHuge(page
))
249 __dec_node_page_state(page
, NR_SHMEM_THPS
);
251 VM_BUG_ON_PAGE(PageTransHuge(page
), page
);
255 * At this point page must be either written or cleaned by truncate.
256 * Dirty page here signals a bug and loss of unwritten data.
258 * This fixes dirty accounting after removing the page entirely but
259 * leaves PageDirty set: it has no effect for truncated page and
260 * anyway will be cleared before returning page into buddy allocator.
262 if (WARN_ON_ONCE(PageDirty(page
)))
263 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
267 * delete_from_page_cache - delete page from page cache
268 * @page: the page which the kernel is trying to remove from page cache
270 * This must be called only on pages that have been verified to be in the page
271 * cache and locked. It will never put the page into the free list, the caller
272 * has a reference on the page.
274 void delete_from_page_cache(struct page
*page
)
276 struct address_space
*mapping
= page_mapping(page
);
278 void (*freepage
)(struct page
*);
280 BUG_ON(!PageLocked(page
));
282 freepage
= mapping
->a_ops
->freepage
;
284 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
285 __delete_from_page_cache(page
, NULL
);
286 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
291 if (PageTransHuge(page
) && !PageHuge(page
)) {
292 page_ref_sub(page
, HPAGE_PMD_NR
);
293 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
298 EXPORT_SYMBOL(delete_from_page_cache
);
300 int filemap_check_errors(struct address_space
*mapping
)
303 /* Check for outstanding write errors */
304 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
305 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
307 if (test_bit(AS_EIO
, &mapping
->flags
) &&
308 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
312 EXPORT_SYMBOL(filemap_check_errors
);
314 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
316 /* Check for outstanding write errors */
317 if (test_bit(AS_EIO
, &mapping
->flags
))
319 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
325 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
326 * @mapping: address space structure to write
327 * @start: offset in bytes where the range starts
328 * @end: offset in bytes where the range ends (inclusive)
329 * @sync_mode: enable synchronous operation
331 * Start writeback against all of a mapping's dirty pages that lie
332 * within the byte offsets <start, end> inclusive.
334 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
335 * opposed to a regular memory cleansing writeback. The difference between
336 * these two operations is that if a dirty page/buffer is encountered, it must
337 * be waited upon, and not just skipped over.
339 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
340 loff_t end
, int sync_mode
)
343 struct writeback_control wbc
= {
344 .sync_mode
= sync_mode
,
345 .nr_to_write
= LONG_MAX
,
346 .range_start
= start
,
350 if (!mapping_cap_writeback_dirty(mapping
))
353 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
354 ret
= do_writepages(mapping
, &wbc
);
355 wbc_detach_inode(&wbc
);
359 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
362 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
365 int filemap_fdatawrite(struct address_space
*mapping
)
367 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
369 EXPORT_SYMBOL(filemap_fdatawrite
);
371 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
374 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
376 EXPORT_SYMBOL(filemap_fdatawrite_range
);
379 * filemap_flush - mostly a non-blocking flush
380 * @mapping: target address_space
382 * This is a mostly non-blocking flush. Not suitable for data-integrity
383 * purposes - I/O may not be started against all dirty pages.
385 int filemap_flush(struct address_space
*mapping
)
387 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
389 EXPORT_SYMBOL(filemap_flush
);
392 * filemap_range_has_page - check if a page exists in range.
393 * @mapping: address space within which to check
394 * @start_byte: offset in bytes where the range starts
395 * @end_byte: offset in bytes where the range ends (inclusive)
397 * Find at least one page in the range supplied, usually used to check if
398 * direct writing in this range will trigger a writeback.
400 bool filemap_range_has_page(struct address_space
*mapping
,
401 loff_t start_byte
, loff_t end_byte
)
403 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
404 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
408 if (end_byte
< start_byte
)
411 if (mapping
->nrpages
== 0)
414 pagevec_init(&pvec
, 0);
415 if (!pagevec_lookup(&pvec
, mapping
, index
, 1))
417 ret
= (pvec
.pages
[0]->index
<= end
);
418 pagevec_release(&pvec
);
421 EXPORT_SYMBOL(filemap_range_has_page
);
423 static void __filemap_fdatawait_range(struct address_space
*mapping
,
424 loff_t start_byte
, loff_t end_byte
)
426 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
427 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
431 if (end_byte
< start_byte
)
434 pagevec_init(&pvec
, 0);
435 while ((index
<= end
) &&
436 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
437 PAGECACHE_TAG_WRITEBACK
,
438 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
441 for (i
= 0; i
< nr_pages
; i
++) {
442 struct page
*page
= pvec
.pages
[i
];
444 /* until radix tree lookup accepts end_index */
445 if (page
->index
> end
)
448 wait_on_page_writeback(page
);
449 ClearPageError(page
);
451 pagevec_release(&pvec
);
457 * filemap_fdatawait_range - wait for writeback to complete
458 * @mapping: address space structure to wait for
459 * @start_byte: offset in bytes where the range starts
460 * @end_byte: offset in bytes where the range ends (inclusive)
462 * Walk the list of under-writeback pages of the given address space
463 * in the given range and wait for all of them. Check error status of
464 * the address space and return it.
466 * Since the error status of the address space is cleared by this function,
467 * callers are responsible for checking the return value and handling and/or
468 * reporting the error.
470 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
473 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
474 return filemap_check_errors(mapping
);
476 EXPORT_SYMBOL(filemap_fdatawait_range
);
479 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
480 * @mapping: address space structure to wait for
482 * Walk the list of under-writeback pages of the given address space
483 * and wait for all of them. Unlike filemap_fdatawait(), this function
484 * does not clear error status of the address space.
486 * Use this function if callers don't handle errors themselves. Expected
487 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
490 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
492 loff_t i_size
= i_size_read(mapping
->host
);
497 __filemap_fdatawait_range(mapping
, 0, i_size
- 1);
498 return filemap_check_and_keep_errors(mapping
);
500 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
503 * filemap_fdatawait - wait for all under-writeback pages to complete
504 * @mapping: address space structure to wait for
506 * Walk the list of under-writeback pages of the given address space
507 * and wait for all of them. Check error status of the address space
510 * Since the error status of the address space is cleared by this function,
511 * callers are responsible for checking the return value and handling and/or
512 * reporting the error.
514 int filemap_fdatawait(struct address_space
*mapping
)
516 loff_t i_size
= i_size_read(mapping
->host
);
521 return filemap_fdatawait_range(mapping
, 0, i_size
- 1);
523 EXPORT_SYMBOL(filemap_fdatawait
);
525 int filemap_write_and_wait(struct address_space
*mapping
)
529 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
530 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
531 err
= filemap_fdatawrite(mapping
);
533 * Even if the above returned error, the pages may be
534 * written partially (e.g. -ENOSPC), so we wait for it.
535 * But the -EIO is special case, it may indicate the worst
536 * thing (e.g. bug) happened, so we avoid waiting for it.
539 int err2
= filemap_fdatawait(mapping
);
543 /* Clear any previously stored errors */
544 filemap_check_errors(mapping
);
547 err
= filemap_check_errors(mapping
);
551 EXPORT_SYMBOL(filemap_write_and_wait
);
554 * filemap_write_and_wait_range - write out & wait on a file range
555 * @mapping: the address_space for the pages
556 * @lstart: offset in bytes where the range starts
557 * @lend: offset in bytes where the range ends (inclusive)
559 * Write out and wait upon file offsets lstart->lend, inclusive.
561 * Note that @lend is inclusive (describes the last byte to be written) so
562 * that this function can be used to write to the very end-of-file (end = -1).
564 int filemap_write_and_wait_range(struct address_space
*mapping
,
565 loff_t lstart
, loff_t lend
)
569 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
570 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
571 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
573 /* See comment of filemap_write_and_wait() */
575 int err2
= filemap_fdatawait_range(mapping
,
580 /* Clear any previously stored errors */
581 filemap_check_errors(mapping
);
584 err
= filemap_check_errors(mapping
);
588 EXPORT_SYMBOL(filemap_write_and_wait_range
);
590 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
592 errseq_t eseq
= __errseq_set(&mapping
->wb_err
, err
);
594 trace_filemap_set_wb_err(mapping
, eseq
);
596 EXPORT_SYMBOL(__filemap_set_wb_err
);
599 * file_check_and_advance_wb_err - report wb error (if any) that was previously
600 * and advance wb_err to current one
601 * @file: struct file on which the error is being reported
603 * When userland calls fsync (or something like nfsd does the equivalent), we
604 * want to report any writeback errors that occurred since the last fsync (or
605 * since the file was opened if there haven't been any).
607 * Grab the wb_err from the mapping. If it matches what we have in the file,
608 * then just quickly return 0. The file is all caught up.
610 * If it doesn't match, then take the mapping value, set the "seen" flag in
611 * it and try to swap it into place. If it works, or another task beat us
612 * to it with the new value, then update the f_wb_err and return the error
613 * portion. The error at this point must be reported via proper channels
614 * (a'la fsync, or NFS COMMIT operation, etc.).
616 * While we handle mapping->wb_err with atomic operations, the f_wb_err
617 * value is protected by the f_lock since we must ensure that it reflects
618 * the latest value swapped in for this file descriptor.
620 int file_check_and_advance_wb_err(struct file
*file
)
623 errseq_t old
= READ_ONCE(file
->f_wb_err
);
624 struct address_space
*mapping
= file
->f_mapping
;
626 /* Locklessly handle the common case where nothing has changed */
627 if (errseq_check(&mapping
->wb_err
, old
)) {
628 /* Something changed, must use slow path */
629 spin_lock(&file
->f_lock
);
630 old
= file
->f_wb_err
;
631 err
= errseq_check_and_advance(&mapping
->wb_err
,
633 trace_file_check_and_advance_wb_err(file
, old
);
634 spin_unlock(&file
->f_lock
);
638 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
641 * file_write_and_wait_range - write out & wait on a file range
642 * @file: file pointing to address_space with pages
643 * @lstart: offset in bytes where the range starts
644 * @lend: offset in bytes where the range ends (inclusive)
646 * Write out and wait upon file offsets lstart->lend, inclusive.
648 * Note that @lend is inclusive (describes the last byte to be written) so
649 * that this function can be used to write to the very end-of-file (end = -1).
651 * After writing out and waiting on the data, we check and advance the
652 * f_wb_err cursor to the latest value, and return any errors detected there.
654 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
657 struct address_space
*mapping
= file
->f_mapping
;
659 if ((!dax_mapping(mapping
) && mapping
->nrpages
) ||
660 (dax_mapping(mapping
) && mapping
->nrexceptional
)) {
661 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
663 /* See comment of filemap_write_and_wait() */
665 __filemap_fdatawait_range(mapping
, lstart
, lend
);
667 err2
= file_check_and_advance_wb_err(file
);
672 EXPORT_SYMBOL(file_write_and_wait_range
);
675 * replace_page_cache_page - replace a pagecache page with a new one
676 * @old: page to be replaced
677 * @new: page to replace with
678 * @gfp_mask: allocation mode
680 * This function replaces a page in the pagecache with a new one. On
681 * success it acquires the pagecache reference for the new page and
682 * drops it for the old page. Both the old and new pages must be
683 * locked. This function does not add the new page to the LRU, the
684 * caller must do that.
686 * The remove + add is atomic. The only way this function can fail is
687 * memory allocation failure.
689 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
693 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
694 VM_BUG_ON_PAGE(!PageLocked(new), new);
695 VM_BUG_ON_PAGE(new->mapping
, new);
697 error
= radix_tree_preload(gfp_mask
& ~__GFP_HIGHMEM
);
699 struct address_space
*mapping
= old
->mapping
;
700 void (*freepage
)(struct page
*);
703 pgoff_t offset
= old
->index
;
704 freepage
= mapping
->a_ops
->freepage
;
707 new->mapping
= mapping
;
710 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
711 __delete_from_page_cache(old
, NULL
);
712 error
= page_cache_tree_insert(mapping
, new, NULL
);
716 * hugetlb pages do not participate in page cache accounting.
719 __inc_node_page_state(new, NR_FILE_PAGES
);
720 if (PageSwapBacked(new))
721 __inc_node_page_state(new, NR_SHMEM
);
722 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
723 mem_cgroup_migrate(old
, new);
724 radix_tree_preload_end();
732 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
734 static int __add_to_page_cache_locked(struct page
*page
,
735 struct address_space
*mapping
,
736 pgoff_t offset
, gfp_t gfp_mask
,
739 int huge
= PageHuge(page
);
740 struct mem_cgroup
*memcg
;
743 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
744 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
747 error
= mem_cgroup_try_charge(page
, current
->mm
,
748 gfp_mask
, &memcg
, false);
753 error
= radix_tree_maybe_preload(gfp_mask
& ~__GFP_HIGHMEM
);
756 mem_cgroup_cancel_charge(page
, memcg
, false);
761 page
->mapping
= mapping
;
762 page
->index
= offset
;
764 spin_lock_irq(&mapping
->tree_lock
);
765 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
766 radix_tree_preload_end();
770 /* hugetlb pages do not participate in page cache accounting. */
772 __inc_node_page_state(page
, NR_FILE_PAGES
);
773 spin_unlock_irq(&mapping
->tree_lock
);
775 mem_cgroup_commit_charge(page
, memcg
, false, false);
776 trace_mm_filemap_add_to_page_cache(page
);
779 page
->mapping
= NULL
;
780 /* Leave page->index set: truncation relies upon it */
781 spin_unlock_irq(&mapping
->tree_lock
);
783 mem_cgroup_cancel_charge(page
, memcg
, false);
789 * add_to_page_cache_locked - add a locked page to the pagecache
791 * @mapping: the page's address_space
792 * @offset: page index
793 * @gfp_mask: page allocation mode
795 * This function is used to add a page to the pagecache. It must be locked.
796 * This function does not add the page to the LRU. The caller must do that.
798 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
799 pgoff_t offset
, gfp_t gfp_mask
)
801 return __add_to_page_cache_locked(page
, mapping
, offset
,
804 EXPORT_SYMBOL(add_to_page_cache_locked
);
806 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
807 pgoff_t offset
, gfp_t gfp_mask
)
812 __SetPageLocked(page
);
813 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
816 __ClearPageLocked(page
);
819 * The page might have been evicted from cache only
820 * recently, in which case it should be activated like
821 * any other repeatedly accessed page.
822 * The exception is pages getting rewritten; evicting other
823 * data from the working set, only to cache data that will
824 * get overwritten with something else, is a waste of memory.
826 if (!(gfp_mask
& __GFP_WRITE
) &&
827 shadow
&& workingset_refault(shadow
)) {
829 workingset_activation(page
);
831 ClearPageActive(page
);
836 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
839 struct page
*__page_cache_alloc(gfp_t gfp
)
844 if (cpuset_do_page_mem_spread()) {
845 unsigned int cpuset_mems_cookie
;
847 cpuset_mems_cookie
= read_mems_allowed_begin();
848 n
= cpuset_mem_spread_node();
849 page
= __alloc_pages_node(n
, gfp
, 0);
850 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
854 return alloc_pages(gfp
, 0);
856 EXPORT_SYMBOL(__page_cache_alloc
);
860 * In order to wait for pages to become available there must be
861 * waitqueues associated with pages. By using a hash table of
862 * waitqueues where the bucket discipline is to maintain all
863 * waiters on the same queue and wake all when any of the pages
864 * become available, and for the woken contexts to check to be
865 * sure the appropriate page became available, this saves space
866 * at a cost of "thundering herd" phenomena during rare hash
869 #define PAGE_WAIT_TABLE_BITS 8
870 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
871 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
873 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
875 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
878 void __init
pagecache_init(void)
882 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
883 init_waitqueue_head(&page_wait_table
[i
]);
885 page_writeback_init();
888 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
889 struct wait_page_key
{
895 struct wait_page_queue
{
898 wait_queue_entry_t wait
;
901 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
903 struct wait_page_key
*key
= arg
;
904 struct wait_page_queue
*wait_page
905 = container_of(wait
, struct wait_page_queue
, wait
);
907 if (wait_page
->page
!= key
->page
)
911 if (wait_page
->bit_nr
!= key
->bit_nr
)
914 /* Stop walking if it's locked */
915 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
918 return autoremove_wake_function(wait
, mode
, sync
, key
);
921 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
923 wait_queue_head_t
*q
= page_waitqueue(page
);
924 struct wait_page_key key
;
931 spin_lock_irqsave(&q
->lock
, flags
);
932 __wake_up_locked_key(q
, TASK_NORMAL
, &key
);
934 * It is possible for other pages to have collided on the waitqueue
935 * hash, so in that case check for a page match. That prevents a long-
938 * It is still possible to miss a case here, when we woke page waiters
939 * and removed them from the waitqueue, but there are still other
942 if (!waitqueue_active(q
) || !key
.page_match
) {
943 ClearPageWaiters(page
);
945 * It's possible to miss clearing Waiters here, when we woke
946 * our page waiters, but the hashed waitqueue has waiters for
949 * That's okay, it's a rare case. The next waker will clear it.
952 spin_unlock_irqrestore(&q
->lock
, flags
);
955 static void wake_up_page(struct page
*page
, int bit
)
957 if (!PageWaiters(page
))
959 wake_up_page_bit(page
, bit
);
962 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
963 struct page
*page
, int bit_nr
, int state
, bool lock
)
965 struct wait_page_queue wait_page
;
966 wait_queue_entry_t
*wait
= &wait_page
.wait
;
970 wait
->flags
= lock
? WQ_FLAG_EXCLUSIVE
: 0;
971 wait
->func
= wake_page_function
;
972 wait_page
.page
= page
;
973 wait_page
.bit_nr
= bit_nr
;
976 spin_lock_irq(&q
->lock
);
978 if (likely(list_empty(&wait
->entry
))) {
979 __add_wait_queue_entry_tail(q
, wait
);
980 SetPageWaiters(page
);
983 set_current_state(state
);
985 spin_unlock_irq(&q
->lock
);
987 if (likely(test_bit(bit_nr
, &page
->flags
))) {
989 if (unlikely(signal_pending_state(state
, current
))) {
996 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
999 if (!test_bit(bit_nr
, &page
->flags
))
1004 finish_wait(q
, wait
);
1007 * A signal could leave PageWaiters set. Clearing it here if
1008 * !waitqueue_active would be possible (by open-coding finish_wait),
1009 * but still fail to catch it in the case of wait hash collision. We
1010 * already can fail to clear wait hash collision cases, so don't
1011 * bother with signals either.
1017 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1019 wait_queue_head_t
*q
= page_waitqueue(page
);
1020 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, false);
1022 EXPORT_SYMBOL(wait_on_page_bit
);
1024 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1026 wait_queue_head_t
*q
= page_waitqueue(page
);
1027 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, false);
1031 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1032 * @page: Page defining the wait queue of interest
1033 * @waiter: Waiter to add to the queue
1035 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1037 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1039 wait_queue_head_t
*q
= page_waitqueue(page
);
1040 unsigned long flags
;
1042 spin_lock_irqsave(&q
->lock
, flags
);
1043 __add_wait_queue(q
, waiter
);
1044 SetPageWaiters(page
);
1045 spin_unlock_irqrestore(&q
->lock
, flags
);
1047 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1049 #ifndef clear_bit_unlock_is_negative_byte
1052 * PG_waiters is the high bit in the same byte as PG_lock.
1054 * On x86 (and on many other architectures), we can clear PG_lock and
1055 * test the sign bit at the same time. But if the architecture does
1056 * not support that special operation, we just do this all by hand
1059 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1060 * being cleared, but a memory barrier should be unneccssary since it is
1061 * in the same byte as PG_locked.
1063 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1065 clear_bit_unlock(nr
, mem
);
1066 /* smp_mb__after_atomic(); */
1067 return test_bit(PG_waiters
, mem
);
1073 * unlock_page - unlock a locked page
1076 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1077 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1078 * mechanism between PageLocked pages and PageWriteback pages is shared.
1079 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1081 * Note that this depends on PG_waiters being the sign bit in the byte
1082 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1083 * clear the PG_locked bit and test PG_waiters at the same time fairly
1084 * portably (architectures that do LL/SC can test any bit, while x86 can
1085 * test the sign bit).
1087 void unlock_page(struct page
*page
)
1089 BUILD_BUG_ON(PG_waiters
!= 7);
1090 page
= compound_head(page
);
1091 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1092 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1093 wake_up_page_bit(page
, PG_locked
);
1095 EXPORT_SYMBOL(unlock_page
);
1098 * end_page_writeback - end writeback against a page
1101 void end_page_writeback(struct page
*page
)
1104 * TestClearPageReclaim could be used here but it is an atomic
1105 * operation and overkill in this particular case. Failing to
1106 * shuffle a page marked for immediate reclaim is too mild to
1107 * justify taking an atomic operation penalty at the end of
1108 * ever page writeback.
1110 if (PageReclaim(page
)) {
1111 ClearPageReclaim(page
);
1112 rotate_reclaimable_page(page
);
1115 if (!test_clear_page_writeback(page
))
1118 smp_mb__after_atomic();
1119 wake_up_page(page
, PG_writeback
);
1121 EXPORT_SYMBOL(end_page_writeback
);
1124 * After completing I/O on a page, call this routine to update the page
1125 * flags appropriately
1127 void page_endio(struct page
*page
, bool is_write
, int err
)
1131 SetPageUptodate(page
);
1133 ClearPageUptodate(page
);
1139 struct address_space
*mapping
;
1142 mapping
= page_mapping(page
);
1144 mapping_set_error(mapping
, err
);
1146 end_page_writeback(page
);
1149 EXPORT_SYMBOL_GPL(page_endio
);
1152 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1153 * @__page: the page to lock
1155 void __lock_page(struct page
*__page
)
1157 struct page
*page
= compound_head(__page
);
1158 wait_queue_head_t
*q
= page_waitqueue(page
);
1159 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, true);
1161 EXPORT_SYMBOL(__lock_page
);
1163 int __lock_page_killable(struct page
*__page
)
1165 struct page
*page
= compound_head(__page
);
1166 wait_queue_head_t
*q
= page_waitqueue(page
);
1167 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
, true);
1169 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1173 * 1 - page is locked; mmap_sem is still held.
1174 * 0 - page is not locked.
1175 * mmap_sem has been released (up_read()), unless flags had both
1176 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1177 * which case mmap_sem is still held.
1179 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1180 * with the page locked and the mmap_sem unperturbed.
1182 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1185 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1187 * CAUTION! In this case, mmap_sem is not released
1188 * even though return 0.
1190 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1193 up_read(&mm
->mmap_sem
);
1194 if (flags
& FAULT_FLAG_KILLABLE
)
1195 wait_on_page_locked_killable(page
);
1197 wait_on_page_locked(page
);
1200 if (flags
& FAULT_FLAG_KILLABLE
) {
1203 ret
= __lock_page_killable(page
);
1205 up_read(&mm
->mmap_sem
);
1215 * page_cache_next_hole - find the next hole (not-present entry)
1218 * @max_scan: maximum range to search
1220 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1221 * lowest indexed hole.
1223 * Returns: the index of the hole if found, otherwise returns an index
1224 * outside of the set specified (in which case 'return - index >=
1225 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1228 * page_cache_next_hole may be called under rcu_read_lock. However,
1229 * like radix_tree_gang_lookup, this will not atomically search a
1230 * snapshot of the tree at a single point in time. For example, if a
1231 * hole is created at index 5, then subsequently a hole is created at
1232 * index 10, page_cache_next_hole covering both indexes may return 10
1233 * if called under rcu_read_lock.
1235 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1236 pgoff_t index
, unsigned long max_scan
)
1240 for (i
= 0; i
< max_scan
; i
++) {
1243 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1244 if (!page
|| radix_tree_exceptional_entry(page
))
1253 EXPORT_SYMBOL(page_cache_next_hole
);
1256 * page_cache_prev_hole - find the prev hole (not-present entry)
1259 * @max_scan: maximum range to search
1261 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1264 * Returns: the index of the hole if found, otherwise returns an index
1265 * outside of the set specified (in which case 'index - return >=
1266 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1269 * page_cache_prev_hole may be called under rcu_read_lock. However,
1270 * like radix_tree_gang_lookup, this will not atomically search a
1271 * snapshot of the tree at a single point in time. For example, if a
1272 * hole is created at index 10, then subsequently a hole is created at
1273 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1274 * called under rcu_read_lock.
1276 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1277 pgoff_t index
, unsigned long max_scan
)
1281 for (i
= 0; i
< max_scan
; i
++) {
1284 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1285 if (!page
|| radix_tree_exceptional_entry(page
))
1288 if (index
== ULONG_MAX
)
1294 EXPORT_SYMBOL(page_cache_prev_hole
);
1297 * find_get_entry - find and get a page cache entry
1298 * @mapping: the address_space to search
1299 * @offset: the page cache index
1301 * Looks up the page cache slot at @mapping & @offset. If there is a
1302 * page cache page, it is returned with an increased refcount.
1304 * If the slot holds a shadow entry of a previously evicted page, or a
1305 * swap entry from shmem/tmpfs, it is returned.
1307 * Otherwise, %NULL is returned.
1309 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1312 struct page
*head
, *page
;
1317 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1319 page
= radix_tree_deref_slot(pagep
);
1320 if (unlikely(!page
))
1322 if (radix_tree_exception(page
)) {
1323 if (radix_tree_deref_retry(page
))
1326 * A shadow entry of a recently evicted page,
1327 * or a swap entry from shmem/tmpfs. Return
1328 * it without attempting to raise page count.
1333 head
= compound_head(page
);
1334 if (!page_cache_get_speculative(head
))
1337 /* The page was split under us? */
1338 if (compound_head(page
) != head
) {
1344 * Has the page moved?
1345 * This is part of the lockless pagecache protocol. See
1346 * include/linux/pagemap.h for details.
1348 if (unlikely(page
!= *pagep
)) {
1358 EXPORT_SYMBOL(find_get_entry
);
1361 * find_lock_entry - locate, pin and lock a page cache entry
1362 * @mapping: the address_space to search
1363 * @offset: the page cache index
1365 * Looks up the page cache slot at @mapping & @offset. If there is a
1366 * page cache page, it is returned locked and with an increased
1369 * If the slot holds a shadow entry of a previously evicted page, or a
1370 * swap entry from shmem/tmpfs, it is returned.
1372 * Otherwise, %NULL is returned.
1374 * find_lock_entry() may sleep.
1376 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1381 page
= find_get_entry(mapping
, offset
);
1382 if (page
&& !radix_tree_exception(page
)) {
1384 /* Has the page been truncated? */
1385 if (unlikely(page_mapping(page
) != mapping
)) {
1390 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1394 EXPORT_SYMBOL(find_lock_entry
);
1397 * pagecache_get_page - find and get a page reference
1398 * @mapping: the address_space to search
1399 * @offset: the page index
1400 * @fgp_flags: PCG flags
1401 * @gfp_mask: gfp mask to use for the page cache data page allocation
1403 * Looks up the page cache slot at @mapping & @offset.
1405 * PCG flags modify how the page is returned.
1407 * @fgp_flags can be:
1409 * - FGP_ACCESSED: the page will be marked accessed
1410 * - FGP_LOCK: Page is return locked
1411 * - FGP_CREAT: If page is not present then a new page is allocated using
1412 * @gfp_mask and added to the page cache and the VM's LRU
1413 * list. The page is returned locked and with an increased
1414 * refcount. Otherwise, NULL is returned.
1416 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1417 * if the GFP flags specified for FGP_CREAT are atomic.
1419 * If there is a page cache page, it is returned with an increased refcount.
1421 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1422 int fgp_flags
, gfp_t gfp_mask
)
1427 page
= find_get_entry(mapping
, offset
);
1428 if (radix_tree_exceptional_entry(page
))
1433 if (fgp_flags
& FGP_LOCK
) {
1434 if (fgp_flags
& FGP_NOWAIT
) {
1435 if (!trylock_page(page
)) {
1443 /* Has the page been truncated? */
1444 if (unlikely(page
->mapping
!= mapping
)) {
1449 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1452 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1453 mark_page_accessed(page
);
1456 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1458 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1459 gfp_mask
|= __GFP_WRITE
;
1460 if (fgp_flags
& FGP_NOFS
)
1461 gfp_mask
&= ~__GFP_FS
;
1463 page
= __page_cache_alloc(gfp_mask
);
1467 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1468 fgp_flags
|= FGP_LOCK
;
1470 /* Init accessed so avoid atomic mark_page_accessed later */
1471 if (fgp_flags
& FGP_ACCESSED
)
1472 __SetPageReferenced(page
);
1474 err
= add_to_page_cache_lru(page
, mapping
, offset
,
1475 gfp_mask
& GFP_RECLAIM_MASK
);
1476 if (unlikely(err
)) {
1486 EXPORT_SYMBOL(pagecache_get_page
);
1489 * find_get_entries - gang pagecache lookup
1490 * @mapping: The address_space to search
1491 * @start: The starting page cache index
1492 * @nr_entries: The maximum number of entries
1493 * @entries: Where the resulting entries are placed
1494 * @indices: The cache indices corresponding to the entries in @entries
1496 * find_get_entries() will search for and return a group of up to
1497 * @nr_entries entries in the mapping. The entries are placed at
1498 * @entries. find_get_entries() takes a reference against any actual
1501 * The search returns a group of mapping-contiguous page cache entries
1502 * with ascending indexes. There may be holes in the indices due to
1503 * not-present pages.
1505 * Any shadow entries of evicted pages, or swap entries from
1506 * shmem/tmpfs, are included in the returned array.
1508 * find_get_entries() returns the number of pages and shadow entries
1511 unsigned find_get_entries(struct address_space
*mapping
,
1512 pgoff_t start
, unsigned int nr_entries
,
1513 struct page
**entries
, pgoff_t
*indices
)
1516 unsigned int ret
= 0;
1517 struct radix_tree_iter iter
;
1523 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1524 struct page
*head
, *page
;
1526 page
= radix_tree_deref_slot(slot
);
1527 if (unlikely(!page
))
1529 if (radix_tree_exception(page
)) {
1530 if (radix_tree_deref_retry(page
)) {
1531 slot
= radix_tree_iter_retry(&iter
);
1535 * A shadow entry of a recently evicted page, a swap
1536 * entry from shmem/tmpfs or a DAX entry. Return it
1537 * without attempting to raise page count.
1542 head
= compound_head(page
);
1543 if (!page_cache_get_speculative(head
))
1546 /* The page was split under us? */
1547 if (compound_head(page
) != head
) {
1552 /* Has the page moved? */
1553 if (unlikely(page
!= *slot
)) {
1558 indices
[ret
] = iter
.index
;
1559 entries
[ret
] = page
;
1560 if (++ret
== nr_entries
)
1568 * find_get_pages - gang pagecache lookup
1569 * @mapping: The address_space to search
1570 * @start: The starting page index
1571 * @nr_pages: The maximum number of pages
1572 * @pages: Where the resulting pages are placed
1574 * find_get_pages() will search for and return a group of up to
1575 * @nr_pages pages in the mapping. The pages are placed at @pages.
1576 * find_get_pages() takes a reference against the returned pages.
1578 * The search returns a group of mapping-contiguous pages with ascending
1579 * indexes. There may be holes in the indices due to not-present pages.
1581 * find_get_pages() returns the number of pages which were found.
1583 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
1584 unsigned int nr_pages
, struct page
**pages
)
1586 struct radix_tree_iter iter
;
1590 if (unlikely(!nr_pages
))
1594 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1595 struct page
*head
, *page
;
1597 page
= radix_tree_deref_slot(slot
);
1598 if (unlikely(!page
))
1601 if (radix_tree_exception(page
)) {
1602 if (radix_tree_deref_retry(page
)) {
1603 slot
= radix_tree_iter_retry(&iter
);
1607 * A shadow entry of a recently evicted page,
1608 * or a swap entry from shmem/tmpfs. Skip
1614 head
= compound_head(page
);
1615 if (!page_cache_get_speculative(head
))
1618 /* The page was split under us? */
1619 if (compound_head(page
) != head
) {
1624 /* Has the page moved? */
1625 if (unlikely(page
!= *slot
)) {
1631 if (++ret
== nr_pages
)
1640 * find_get_pages_contig - gang contiguous pagecache lookup
1641 * @mapping: The address_space to search
1642 * @index: The starting page index
1643 * @nr_pages: The maximum number of pages
1644 * @pages: Where the resulting pages are placed
1646 * find_get_pages_contig() works exactly like find_get_pages(), except
1647 * that the returned number of pages are guaranteed to be contiguous.
1649 * find_get_pages_contig() returns the number of pages which were found.
1651 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1652 unsigned int nr_pages
, struct page
**pages
)
1654 struct radix_tree_iter iter
;
1656 unsigned int ret
= 0;
1658 if (unlikely(!nr_pages
))
1662 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1663 struct page
*head
, *page
;
1665 page
= radix_tree_deref_slot(slot
);
1666 /* The hole, there no reason to continue */
1667 if (unlikely(!page
))
1670 if (radix_tree_exception(page
)) {
1671 if (radix_tree_deref_retry(page
)) {
1672 slot
= radix_tree_iter_retry(&iter
);
1676 * A shadow entry of a recently evicted page,
1677 * or a swap entry from shmem/tmpfs. Stop
1678 * looking for contiguous pages.
1683 head
= compound_head(page
);
1684 if (!page_cache_get_speculative(head
))
1687 /* The page was split under us? */
1688 if (compound_head(page
) != head
) {
1693 /* Has the page moved? */
1694 if (unlikely(page
!= *slot
)) {
1700 * must check mapping and index after taking the ref.
1701 * otherwise we can get both false positives and false
1702 * negatives, which is just confusing to the caller.
1704 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1710 if (++ret
== nr_pages
)
1716 EXPORT_SYMBOL(find_get_pages_contig
);
1719 * find_get_pages_tag - find and return pages that match @tag
1720 * @mapping: the address_space to search
1721 * @index: the starting page index
1722 * @tag: the tag index
1723 * @nr_pages: the maximum number of pages
1724 * @pages: where the resulting pages are placed
1726 * Like find_get_pages, except we only return pages which are tagged with
1727 * @tag. We update @index to index the next page for the traversal.
1729 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
1730 int tag
, unsigned int nr_pages
, struct page
**pages
)
1732 struct radix_tree_iter iter
;
1736 if (unlikely(!nr_pages
))
1740 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1741 &iter
, *index
, tag
) {
1742 struct page
*head
, *page
;
1744 page
= radix_tree_deref_slot(slot
);
1745 if (unlikely(!page
))
1748 if (radix_tree_exception(page
)) {
1749 if (radix_tree_deref_retry(page
)) {
1750 slot
= radix_tree_iter_retry(&iter
);
1754 * A shadow entry of a recently evicted page.
1756 * Those entries should never be tagged, but
1757 * this tree walk is lockless and the tags are
1758 * looked up in bulk, one radix tree node at a
1759 * time, so there is a sizable window for page
1760 * reclaim to evict a page we saw tagged.
1767 head
= compound_head(page
);
1768 if (!page_cache_get_speculative(head
))
1771 /* The page was split under us? */
1772 if (compound_head(page
) != head
) {
1777 /* Has the page moved? */
1778 if (unlikely(page
!= *slot
)) {
1784 if (++ret
== nr_pages
)
1791 *index
= pages
[ret
- 1]->index
+ 1;
1795 EXPORT_SYMBOL(find_get_pages_tag
);
1798 * find_get_entries_tag - find and return entries that match @tag
1799 * @mapping: the address_space to search
1800 * @start: the starting page cache index
1801 * @tag: the tag index
1802 * @nr_entries: the maximum number of entries
1803 * @entries: where the resulting entries are placed
1804 * @indices: the cache indices corresponding to the entries in @entries
1806 * Like find_get_entries, except we only return entries which are tagged with
1809 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1810 int tag
, unsigned int nr_entries
,
1811 struct page
**entries
, pgoff_t
*indices
)
1814 unsigned int ret
= 0;
1815 struct radix_tree_iter iter
;
1821 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1822 &iter
, start
, tag
) {
1823 struct page
*head
, *page
;
1825 page
= radix_tree_deref_slot(slot
);
1826 if (unlikely(!page
))
1828 if (radix_tree_exception(page
)) {
1829 if (radix_tree_deref_retry(page
)) {
1830 slot
= radix_tree_iter_retry(&iter
);
1835 * A shadow entry of a recently evicted page, a swap
1836 * entry from shmem/tmpfs or a DAX entry. Return it
1837 * without attempting to raise page count.
1842 head
= compound_head(page
);
1843 if (!page_cache_get_speculative(head
))
1846 /* The page was split under us? */
1847 if (compound_head(page
) != head
) {
1852 /* Has the page moved? */
1853 if (unlikely(page
!= *slot
)) {
1858 indices
[ret
] = iter
.index
;
1859 entries
[ret
] = page
;
1860 if (++ret
== nr_entries
)
1866 EXPORT_SYMBOL(find_get_entries_tag
);
1869 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1870 * a _large_ part of the i/o request. Imagine the worst scenario:
1872 * ---R__________________________________________B__________
1873 * ^ reading here ^ bad block(assume 4k)
1875 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1876 * => failing the whole request => read(R) => read(R+1) =>
1877 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1878 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1879 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1881 * It is going insane. Fix it by quickly scaling down the readahead size.
1883 static void shrink_readahead_size_eio(struct file
*filp
,
1884 struct file_ra_state
*ra
)
1890 * do_generic_file_read - generic file read routine
1891 * @filp: the file to read
1892 * @ppos: current file position
1893 * @iter: data destination
1894 * @written: already copied
1896 * This is a generic file read routine, and uses the
1897 * mapping->a_ops->readpage() function for the actual low-level stuff.
1899 * This is really ugly. But the goto's actually try to clarify some
1900 * of the logic when it comes to error handling etc.
1902 static ssize_t
do_generic_file_read(struct file
*filp
, loff_t
*ppos
,
1903 struct iov_iter
*iter
, ssize_t written
)
1905 struct address_space
*mapping
= filp
->f_mapping
;
1906 struct inode
*inode
= mapping
->host
;
1907 struct file_ra_state
*ra
= &filp
->f_ra
;
1911 unsigned long offset
; /* offset into pagecache page */
1912 unsigned int prev_offset
;
1915 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
1917 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
1919 index
= *ppos
>> PAGE_SHIFT
;
1920 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1921 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1922 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
1923 offset
= *ppos
& ~PAGE_MASK
;
1929 unsigned long nr
, ret
;
1933 if (fatal_signal_pending(current
)) {
1938 page
= find_get_page(mapping
, index
);
1940 page_cache_sync_readahead(mapping
,
1942 index
, last_index
- index
);
1943 page
= find_get_page(mapping
, index
);
1944 if (unlikely(page
== NULL
))
1945 goto no_cached_page
;
1947 if (PageReadahead(page
)) {
1948 page_cache_async_readahead(mapping
,
1950 index
, last_index
- index
);
1952 if (!PageUptodate(page
)) {
1954 * See comment in do_read_cache_page on why
1955 * wait_on_page_locked is used to avoid unnecessarily
1956 * serialisations and why it's safe.
1958 error
= wait_on_page_locked_killable(page
);
1959 if (unlikely(error
))
1960 goto readpage_error
;
1961 if (PageUptodate(page
))
1964 if (inode
->i_blkbits
== PAGE_SHIFT
||
1965 !mapping
->a_ops
->is_partially_uptodate
)
1966 goto page_not_up_to_date
;
1967 /* pipes can't handle partially uptodate pages */
1968 if (unlikely(iter
->type
& ITER_PIPE
))
1969 goto page_not_up_to_date
;
1970 if (!trylock_page(page
))
1971 goto page_not_up_to_date
;
1972 /* Did it get truncated before we got the lock? */
1974 goto page_not_up_to_date_locked
;
1975 if (!mapping
->a_ops
->is_partially_uptodate(page
,
1976 offset
, iter
->count
))
1977 goto page_not_up_to_date_locked
;
1982 * i_size must be checked after we know the page is Uptodate.
1984 * Checking i_size after the check allows us to calculate
1985 * the correct value for "nr", which means the zero-filled
1986 * part of the page is not copied back to userspace (unless
1987 * another truncate extends the file - this is desired though).
1990 isize
= i_size_read(inode
);
1991 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1992 if (unlikely(!isize
|| index
> end_index
)) {
1997 /* nr is the maximum number of bytes to copy from this page */
1999 if (index
== end_index
) {
2000 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2008 /* If users can be writing to this page using arbitrary
2009 * virtual addresses, take care about potential aliasing
2010 * before reading the page on the kernel side.
2012 if (mapping_writably_mapped(mapping
))
2013 flush_dcache_page(page
);
2016 * When a sequential read accesses a page several times,
2017 * only mark it as accessed the first time.
2019 if (prev_index
!= index
|| offset
!= prev_offset
)
2020 mark_page_accessed(page
);
2024 * Ok, we have the page, and it's up-to-date, so
2025 * now we can copy it to user space...
2028 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
2030 index
+= offset
>> PAGE_SHIFT
;
2031 offset
&= ~PAGE_MASK
;
2032 prev_offset
= offset
;
2036 if (!iov_iter_count(iter
))
2044 page_not_up_to_date
:
2045 /* Get exclusive access to the page ... */
2046 error
= lock_page_killable(page
);
2047 if (unlikely(error
))
2048 goto readpage_error
;
2050 page_not_up_to_date_locked
:
2051 /* Did it get truncated before we got the lock? */
2052 if (!page
->mapping
) {
2058 /* Did somebody else fill it already? */
2059 if (PageUptodate(page
)) {
2066 * A previous I/O error may have been due to temporary
2067 * failures, eg. multipath errors.
2068 * PG_error will be set again if readpage fails.
2070 ClearPageError(page
);
2071 /* Start the actual read. The read will unlock the page. */
2072 error
= mapping
->a_ops
->readpage(filp
, page
);
2074 if (unlikely(error
)) {
2075 if (error
== AOP_TRUNCATED_PAGE
) {
2080 goto readpage_error
;
2083 if (!PageUptodate(page
)) {
2084 error
= lock_page_killable(page
);
2085 if (unlikely(error
))
2086 goto readpage_error
;
2087 if (!PageUptodate(page
)) {
2088 if (page
->mapping
== NULL
) {
2090 * invalidate_mapping_pages got it
2097 shrink_readahead_size_eio(filp
, ra
);
2099 goto readpage_error
;
2107 /* UHHUH! A synchronous read error occurred. Report it */
2113 * Ok, it wasn't cached, so we need to create a new
2116 page
= page_cache_alloc_cold(mapping
);
2121 error
= add_to_page_cache_lru(page
, mapping
, index
,
2122 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2125 if (error
== -EEXIST
) {
2135 ra
->prev_pos
= prev_index
;
2136 ra
->prev_pos
<<= PAGE_SHIFT
;
2137 ra
->prev_pos
|= prev_offset
;
2139 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2140 file_accessed(filp
);
2141 return written
? written
: error
;
2145 * generic_file_read_iter - generic filesystem read routine
2146 * @iocb: kernel I/O control block
2147 * @iter: destination for the data read
2149 * This is the "read_iter()" routine for all filesystems
2150 * that can use the page cache directly.
2153 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2155 struct file
*file
= iocb
->ki_filp
;
2157 size_t count
= iov_iter_count(iter
);
2160 goto out
; /* skip atime */
2162 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2163 struct address_space
*mapping
= file
->f_mapping
;
2164 struct inode
*inode
= mapping
->host
;
2167 size
= i_size_read(inode
);
2168 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2169 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2170 iocb
->ki_pos
+ count
- 1))
2173 retval
= filemap_write_and_wait_range(mapping
,
2175 iocb
->ki_pos
+ count
- 1);
2180 file_accessed(file
);
2182 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2184 iocb
->ki_pos
+= retval
;
2187 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2190 * Btrfs can have a short DIO read if we encounter
2191 * compressed extents, so if there was an error, or if
2192 * we've already read everything we wanted to, or if
2193 * there was a short read because we hit EOF, go ahead
2194 * and return. Otherwise fallthrough to buffered io for
2195 * the rest of the read. Buffered reads will not work for
2196 * DAX files, so don't bother trying.
2198 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2203 retval
= do_generic_file_read(file
, &iocb
->ki_pos
, iter
, retval
);
2207 EXPORT_SYMBOL(generic_file_read_iter
);
2211 * page_cache_read - adds requested page to the page cache if not already there
2212 * @file: file to read
2213 * @offset: page index
2214 * @gfp_mask: memory allocation flags
2216 * This adds the requested page to the page cache if it isn't already there,
2217 * and schedules an I/O to read in its contents from disk.
2219 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
2221 struct address_space
*mapping
= file
->f_mapping
;
2226 page
= __page_cache_alloc(gfp_mask
|__GFP_COLD
);
2230 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
& GFP_KERNEL
);
2232 ret
= mapping
->a_ops
->readpage(file
, page
);
2233 else if (ret
== -EEXIST
)
2234 ret
= 0; /* losing race to add is OK */
2238 } while (ret
== AOP_TRUNCATED_PAGE
);
2243 #define MMAP_LOTSAMISS (100)
2246 * Synchronous readahead happens when we don't even find
2247 * a page in the page cache at all.
2249 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2250 struct file_ra_state
*ra
,
2254 struct address_space
*mapping
= file
->f_mapping
;
2256 /* If we don't want any read-ahead, don't bother */
2257 if (vma
->vm_flags
& VM_RAND_READ
)
2262 if (vma
->vm_flags
& VM_SEQ_READ
) {
2263 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2268 /* Avoid banging the cache line if not needed */
2269 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2273 * Do we miss much more than hit in this file? If so,
2274 * stop bothering with read-ahead. It will only hurt.
2276 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2282 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2283 ra
->size
= ra
->ra_pages
;
2284 ra
->async_size
= ra
->ra_pages
/ 4;
2285 ra_submit(ra
, mapping
, file
);
2289 * Asynchronous readahead happens when we find the page and PG_readahead,
2290 * so we want to possibly extend the readahead further..
2292 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2293 struct file_ra_state
*ra
,
2298 struct address_space
*mapping
= file
->f_mapping
;
2300 /* If we don't want any read-ahead, don't bother */
2301 if (vma
->vm_flags
& VM_RAND_READ
)
2303 if (ra
->mmap_miss
> 0)
2305 if (PageReadahead(page
))
2306 page_cache_async_readahead(mapping
, ra
, file
,
2307 page
, offset
, ra
->ra_pages
);
2311 * filemap_fault - read in file data for page fault handling
2312 * @vmf: struct vm_fault containing details of the fault
2314 * filemap_fault() is invoked via the vma operations vector for a
2315 * mapped memory region to read in file data during a page fault.
2317 * The goto's are kind of ugly, but this streamlines the normal case of having
2318 * it in the page cache, and handles the special cases reasonably without
2319 * having a lot of duplicated code.
2321 * vma->vm_mm->mmap_sem must be held on entry.
2323 * If our return value has VM_FAULT_RETRY set, it's because
2324 * lock_page_or_retry() returned 0.
2325 * The mmap_sem has usually been released in this case.
2326 * See __lock_page_or_retry() for the exception.
2328 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2329 * has not been released.
2331 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2333 int filemap_fault(struct vm_fault
*vmf
)
2336 struct file
*file
= vmf
->vma
->vm_file
;
2337 struct address_space
*mapping
= file
->f_mapping
;
2338 struct file_ra_state
*ra
= &file
->f_ra
;
2339 struct inode
*inode
= mapping
->host
;
2340 pgoff_t offset
= vmf
->pgoff
;
2345 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2346 if (unlikely(offset
>= max_off
))
2347 return VM_FAULT_SIGBUS
;
2350 * Do we have something in the page cache already?
2352 page
= find_get_page(mapping
, offset
);
2353 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2355 * We found the page, so try async readahead before
2356 * waiting for the lock.
2358 do_async_mmap_readahead(vmf
->vma
, ra
, file
, page
, offset
);
2360 /* No page in the page cache at all */
2361 do_sync_mmap_readahead(vmf
->vma
, ra
, file
, offset
);
2362 count_vm_event(PGMAJFAULT
);
2363 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2364 ret
= VM_FAULT_MAJOR
;
2366 page
= find_get_page(mapping
, offset
);
2368 goto no_cached_page
;
2371 if (!lock_page_or_retry(page
, vmf
->vma
->vm_mm
, vmf
->flags
)) {
2373 return ret
| VM_FAULT_RETRY
;
2376 /* Did it get truncated? */
2377 if (unlikely(page
->mapping
!= mapping
)) {
2382 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2385 * We have a locked page in the page cache, now we need to check
2386 * that it's up-to-date. If not, it is going to be due to an error.
2388 if (unlikely(!PageUptodate(page
)))
2389 goto page_not_uptodate
;
2392 * Found the page and have a reference on it.
2393 * We must recheck i_size under page lock.
2395 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2396 if (unlikely(offset
>= max_off
)) {
2399 return VM_FAULT_SIGBUS
;
2403 return ret
| VM_FAULT_LOCKED
;
2407 * We're only likely to ever get here if MADV_RANDOM is in
2410 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2413 * The page we want has now been added to the page cache.
2414 * In the unlikely event that someone removed it in the
2415 * meantime, we'll just come back here and read it again.
2421 * An error return from page_cache_read can result if the
2422 * system is low on memory, or a problem occurs while trying
2425 if (error
== -ENOMEM
)
2426 return VM_FAULT_OOM
;
2427 return VM_FAULT_SIGBUS
;
2431 * Umm, take care of errors if the page isn't up-to-date.
2432 * Try to re-read it _once_. We do this synchronously,
2433 * because there really aren't any performance issues here
2434 * and we need to check for errors.
2436 ClearPageError(page
);
2437 error
= mapping
->a_ops
->readpage(file
, page
);
2439 wait_on_page_locked(page
);
2440 if (!PageUptodate(page
))
2445 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2448 /* Things didn't work out. Return zero to tell the mm layer so. */
2449 shrink_readahead_size_eio(file
, ra
);
2450 return VM_FAULT_SIGBUS
;
2452 EXPORT_SYMBOL(filemap_fault
);
2454 void filemap_map_pages(struct vm_fault
*vmf
,
2455 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2457 struct radix_tree_iter iter
;
2459 struct file
*file
= vmf
->vma
->vm_file
;
2460 struct address_space
*mapping
= file
->f_mapping
;
2461 pgoff_t last_pgoff
= start_pgoff
;
2462 unsigned long max_idx
;
2463 struct page
*head
, *page
;
2466 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2468 if (iter
.index
> end_pgoff
)
2471 page
= radix_tree_deref_slot(slot
);
2472 if (unlikely(!page
))
2474 if (radix_tree_exception(page
)) {
2475 if (radix_tree_deref_retry(page
)) {
2476 slot
= radix_tree_iter_retry(&iter
);
2482 head
= compound_head(page
);
2483 if (!page_cache_get_speculative(head
))
2486 /* The page was split under us? */
2487 if (compound_head(page
) != head
) {
2492 /* Has the page moved? */
2493 if (unlikely(page
!= *slot
)) {
2498 if (!PageUptodate(page
) ||
2499 PageReadahead(page
) ||
2502 if (!trylock_page(page
))
2505 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2508 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2509 if (page
->index
>= max_idx
)
2512 if (file
->f_ra
.mmap_miss
> 0)
2513 file
->f_ra
.mmap_miss
--;
2515 vmf
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2517 vmf
->pte
+= iter
.index
- last_pgoff
;
2518 last_pgoff
= iter
.index
;
2519 if (alloc_set_pte(vmf
, NULL
, page
))
2528 /* Huge page is mapped? No need to proceed. */
2529 if (pmd_trans_huge(*vmf
->pmd
))
2531 if (iter
.index
== end_pgoff
)
2536 EXPORT_SYMBOL(filemap_map_pages
);
2538 int filemap_page_mkwrite(struct vm_fault
*vmf
)
2540 struct page
*page
= vmf
->page
;
2541 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2542 int ret
= VM_FAULT_LOCKED
;
2544 sb_start_pagefault(inode
->i_sb
);
2545 file_update_time(vmf
->vma
->vm_file
);
2547 if (page
->mapping
!= inode
->i_mapping
) {
2549 ret
= VM_FAULT_NOPAGE
;
2553 * We mark the page dirty already here so that when freeze is in
2554 * progress, we are guaranteed that writeback during freezing will
2555 * see the dirty page and writeprotect it again.
2557 set_page_dirty(page
);
2558 wait_for_stable_page(page
);
2560 sb_end_pagefault(inode
->i_sb
);
2563 EXPORT_SYMBOL(filemap_page_mkwrite
);
2565 const struct vm_operations_struct generic_file_vm_ops
= {
2566 .fault
= filemap_fault
,
2567 .map_pages
= filemap_map_pages
,
2568 .page_mkwrite
= filemap_page_mkwrite
,
2571 /* This is used for a general mmap of a disk file */
2573 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2575 struct address_space
*mapping
= file
->f_mapping
;
2577 if (!mapping
->a_ops
->readpage
)
2579 file_accessed(file
);
2580 vma
->vm_ops
= &generic_file_vm_ops
;
2585 * This is for filesystems which do not implement ->writepage.
2587 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2589 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2591 return generic_file_mmap(file
, vma
);
2594 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2598 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2602 #endif /* CONFIG_MMU */
2604 EXPORT_SYMBOL(generic_file_mmap
);
2605 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2607 static struct page
*wait_on_page_read(struct page
*page
)
2609 if (!IS_ERR(page
)) {
2610 wait_on_page_locked(page
);
2611 if (!PageUptodate(page
)) {
2613 page
= ERR_PTR(-EIO
);
2619 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2621 int (*filler
)(void *, struct page
*),
2628 page
= find_get_page(mapping
, index
);
2630 page
= __page_cache_alloc(gfp
| __GFP_COLD
);
2632 return ERR_PTR(-ENOMEM
);
2633 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2634 if (unlikely(err
)) {
2638 /* Presumably ENOMEM for radix tree node */
2639 return ERR_PTR(err
);
2643 err
= filler(data
, page
);
2646 return ERR_PTR(err
);
2649 page
= wait_on_page_read(page
);
2654 if (PageUptodate(page
))
2658 * Page is not up to date and may be locked due one of the following
2659 * case a: Page is being filled and the page lock is held
2660 * case b: Read/write error clearing the page uptodate status
2661 * case c: Truncation in progress (page locked)
2662 * case d: Reclaim in progress
2664 * Case a, the page will be up to date when the page is unlocked.
2665 * There is no need to serialise on the page lock here as the page
2666 * is pinned so the lock gives no additional protection. Even if the
2667 * the page is truncated, the data is still valid if PageUptodate as
2668 * it's a race vs truncate race.
2669 * Case b, the page will not be up to date
2670 * Case c, the page may be truncated but in itself, the data may still
2671 * be valid after IO completes as it's a read vs truncate race. The
2672 * operation must restart if the page is not uptodate on unlock but
2673 * otherwise serialising on page lock to stabilise the mapping gives
2674 * no additional guarantees to the caller as the page lock is
2675 * released before return.
2676 * Case d, similar to truncation. If reclaim holds the page lock, it
2677 * will be a race with remove_mapping that determines if the mapping
2678 * is valid on unlock but otherwise the data is valid and there is
2679 * no need to serialise with page lock.
2681 * As the page lock gives no additional guarantee, we optimistically
2682 * wait on the page to be unlocked and check if it's up to date and
2683 * use the page if it is. Otherwise, the page lock is required to
2684 * distinguish between the different cases. The motivation is that we
2685 * avoid spurious serialisations and wakeups when multiple processes
2686 * wait on the same page for IO to complete.
2688 wait_on_page_locked(page
);
2689 if (PageUptodate(page
))
2692 /* Distinguish between all the cases under the safety of the lock */
2695 /* Case c or d, restart the operation */
2696 if (!page
->mapping
) {
2702 /* Someone else locked and filled the page in a very small window */
2703 if (PageUptodate(page
)) {
2710 mark_page_accessed(page
);
2715 * read_cache_page - read into page cache, fill it if needed
2716 * @mapping: the page's address_space
2717 * @index: the page index
2718 * @filler: function to perform the read
2719 * @data: first arg to filler(data, page) function, often left as NULL
2721 * Read into the page cache. If a page already exists, and PageUptodate() is
2722 * not set, try to fill the page and wait for it to become unlocked.
2724 * If the page does not get brought uptodate, return -EIO.
2726 struct page
*read_cache_page(struct address_space
*mapping
,
2728 int (*filler
)(void *, struct page
*),
2731 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2733 EXPORT_SYMBOL(read_cache_page
);
2736 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2737 * @mapping: the page's address_space
2738 * @index: the page index
2739 * @gfp: the page allocator flags to use if allocating
2741 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2742 * any new page allocations done using the specified allocation flags.
2744 * If the page does not get brought uptodate, return -EIO.
2746 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2750 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2752 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2754 EXPORT_SYMBOL(read_cache_page_gfp
);
2757 * Performs necessary checks before doing a write
2759 * Can adjust writing position or amount of bytes to write.
2760 * Returns appropriate error code that caller should return or
2761 * zero in case that write should be allowed.
2763 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2765 struct file
*file
= iocb
->ki_filp
;
2766 struct inode
*inode
= file
->f_mapping
->host
;
2767 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2770 if (!iov_iter_count(from
))
2773 /* FIXME: this is for backwards compatibility with 2.4 */
2774 if (iocb
->ki_flags
& IOCB_APPEND
)
2775 iocb
->ki_pos
= i_size_read(inode
);
2779 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
2782 if (limit
!= RLIM_INFINITY
) {
2783 if (iocb
->ki_pos
>= limit
) {
2784 send_sig(SIGXFSZ
, current
, 0);
2787 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2793 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2794 !(file
->f_flags
& O_LARGEFILE
))) {
2795 if (pos
>= MAX_NON_LFS
)
2797 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2801 * Are we about to exceed the fs block limit ?
2803 * If we have written data it becomes a short write. If we have
2804 * exceeded without writing data we send a signal and return EFBIG.
2805 * Linus frestrict idea will clean these up nicely..
2807 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2810 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2811 return iov_iter_count(from
);
2813 EXPORT_SYMBOL(generic_write_checks
);
2815 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2816 loff_t pos
, unsigned len
, unsigned flags
,
2817 struct page
**pagep
, void **fsdata
)
2819 const struct address_space_operations
*aops
= mapping
->a_ops
;
2821 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2824 EXPORT_SYMBOL(pagecache_write_begin
);
2826 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2827 loff_t pos
, unsigned len
, unsigned copied
,
2828 struct page
*page
, void *fsdata
)
2830 const struct address_space_operations
*aops
= mapping
->a_ops
;
2832 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2834 EXPORT_SYMBOL(pagecache_write_end
);
2837 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2839 struct file
*file
= iocb
->ki_filp
;
2840 struct address_space
*mapping
= file
->f_mapping
;
2841 struct inode
*inode
= mapping
->host
;
2842 loff_t pos
= iocb
->ki_pos
;
2847 write_len
= iov_iter_count(from
);
2848 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2850 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2851 /* If there are pages to writeback, return */
2852 if (filemap_range_has_page(inode
->i_mapping
, pos
,
2853 pos
+ iov_iter_count(from
)))
2856 written
= filemap_write_and_wait_range(mapping
, pos
,
2857 pos
+ write_len
- 1);
2863 * After a write we want buffered reads to be sure to go to disk to get
2864 * the new data. We invalidate clean cached page from the region we're
2865 * about to write. We do this *before* the write so that we can return
2866 * without clobbering -EIOCBQUEUED from ->direct_IO().
2868 written
= invalidate_inode_pages2_range(mapping
,
2869 pos
>> PAGE_SHIFT
, end
);
2871 * If a page can not be invalidated, return 0 to fall back
2872 * to buffered write.
2875 if (written
== -EBUSY
)
2880 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
2883 * Finally, try again to invalidate clean pages which might have been
2884 * cached by non-direct readahead, or faulted in by get_user_pages()
2885 * if the source of the write was an mmap'ed region of the file
2886 * we're writing. Either one is a pretty crazy thing to do,
2887 * so we don't support it 100%. If this invalidation
2888 * fails, tough, the write still worked...
2890 invalidate_inode_pages2_range(mapping
,
2891 pos
>> PAGE_SHIFT
, end
);
2895 write_len
-= written
;
2896 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2897 i_size_write(inode
, pos
);
2898 mark_inode_dirty(inode
);
2902 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
2906 EXPORT_SYMBOL(generic_file_direct_write
);
2909 * Find or create a page at the given pagecache position. Return the locked
2910 * page. This function is specifically for buffered writes.
2912 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2913 pgoff_t index
, unsigned flags
)
2916 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2918 if (flags
& AOP_FLAG_NOFS
)
2919 fgp_flags
|= FGP_NOFS
;
2921 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2922 mapping_gfp_mask(mapping
));
2924 wait_for_stable_page(page
);
2928 EXPORT_SYMBOL(grab_cache_page_write_begin
);
2930 ssize_t
generic_perform_write(struct file
*file
,
2931 struct iov_iter
*i
, loff_t pos
)
2933 struct address_space
*mapping
= file
->f_mapping
;
2934 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
2936 ssize_t written
= 0;
2937 unsigned int flags
= 0;
2941 unsigned long offset
; /* Offset into pagecache page */
2942 unsigned long bytes
; /* Bytes to write to page */
2943 size_t copied
; /* Bytes copied from user */
2946 offset
= (pos
& (PAGE_SIZE
- 1));
2947 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
2952 * Bring in the user page that we will copy from _first_.
2953 * Otherwise there's a nasty deadlock on copying from the
2954 * same page as we're writing to, without it being marked
2957 * Not only is this an optimisation, but it is also required
2958 * to check that the address is actually valid, when atomic
2959 * usercopies are used, below.
2961 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
2966 if (fatal_signal_pending(current
)) {
2971 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
2973 if (unlikely(status
< 0))
2976 if (mapping_writably_mapped(mapping
))
2977 flush_dcache_page(page
);
2979 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
2980 flush_dcache_page(page
);
2982 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
2984 if (unlikely(status
< 0))
2990 iov_iter_advance(i
, copied
);
2991 if (unlikely(copied
== 0)) {
2993 * If we were unable to copy any data at all, we must
2994 * fall back to a single segment length write.
2996 * If we didn't fallback here, we could livelock
2997 * because not all segments in the iov can be copied at
2998 * once without a pagefault.
3000 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3001 iov_iter_single_seg_count(i
));
3007 balance_dirty_pages_ratelimited(mapping
);
3008 } while (iov_iter_count(i
));
3010 return written
? written
: status
;
3012 EXPORT_SYMBOL(generic_perform_write
);
3015 * __generic_file_write_iter - write data to a file
3016 * @iocb: IO state structure (file, offset, etc.)
3017 * @from: iov_iter with data to write
3019 * This function does all the work needed for actually writing data to a
3020 * file. It does all basic checks, removes SUID from the file, updates
3021 * modification times and calls proper subroutines depending on whether we
3022 * do direct IO or a standard buffered write.
3024 * It expects i_mutex to be grabbed unless we work on a block device or similar
3025 * object which does not need locking at all.
3027 * This function does *not* take care of syncing data in case of O_SYNC write.
3028 * A caller has to handle it. This is mainly due to the fact that we want to
3029 * avoid syncing under i_mutex.
3031 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3033 struct file
*file
= iocb
->ki_filp
;
3034 struct address_space
* mapping
= file
->f_mapping
;
3035 struct inode
*inode
= mapping
->host
;
3036 ssize_t written
= 0;
3040 /* We can write back this queue in page reclaim */
3041 current
->backing_dev_info
= inode_to_bdi(inode
);
3042 err
= file_remove_privs(file
);
3046 err
= file_update_time(file
);
3050 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3051 loff_t pos
, endbyte
;
3053 written
= generic_file_direct_write(iocb
, from
);
3055 * If the write stopped short of completing, fall back to
3056 * buffered writes. Some filesystems do this for writes to
3057 * holes, for example. For DAX files, a buffered write will
3058 * not succeed (even if it did, DAX does not handle dirty
3059 * page-cache pages correctly).
3061 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3064 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3066 * If generic_perform_write() returned a synchronous error
3067 * then we want to return the number of bytes which were
3068 * direct-written, or the error code if that was zero. Note
3069 * that this differs from normal direct-io semantics, which
3070 * will return -EFOO even if some bytes were written.
3072 if (unlikely(status
< 0)) {
3077 * We need to ensure that the page cache pages are written to
3078 * disk and invalidated to preserve the expected O_DIRECT
3081 endbyte
= pos
+ status
- 1;
3082 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3084 iocb
->ki_pos
= endbyte
+ 1;
3086 invalidate_mapping_pages(mapping
,
3088 endbyte
>> PAGE_SHIFT
);
3091 * We don't know how much we wrote, so just return
3092 * the number of bytes which were direct-written
3096 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3097 if (likely(written
> 0))
3098 iocb
->ki_pos
+= written
;
3101 current
->backing_dev_info
= NULL
;
3102 return written
? written
: err
;
3104 EXPORT_SYMBOL(__generic_file_write_iter
);
3107 * generic_file_write_iter - write data to a file
3108 * @iocb: IO state structure
3109 * @from: iov_iter with data to write
3111 * This is a wrapper around __generic_file_write_iter() to be used by most
3112 * filesystems. It takes care of syncing the file in case of O_SYNC file
3113 * and acquires i_mutex as needed.
3115 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3117 struct file
*file
= iocb
->ki_filp
;
3118 struct inode
*inode
= file
->f_mapping
->host
;
3122 ret
= generic_write_checks(iocb
, from
);
3124 ret
= __generic_file_write_iter(iocb
, from
);
3125 inode_unlock(inode
);
3128 ret
= generic_write_sync(iocb
, ret
);
3131 EXPORT_SYMBOL(generic_file_write_iter
);
3134 * try_to_release_page() - release old fs-specific metadata on a page
3136 * @page: the page which the kernel is trying to free
3137 * @gfp_mask: memory allocation flags (and I/O mode)
3139 * The address_space is to try to release any data against the page
3140 * (presumably at page->private). If the release was successful, return '1'.
3141 * Otherwise return zero.
3143 * This may also be called if PG_fscache is set on a page, indicating that the
3144 * page is known to the local caching routines.
3146 * The @gfp_mask argument specifies whether I/O may be performed to release
3147 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3150 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3152 struct address_space
* const mapping
= page
->mapping
;
3154 BUG_ON(!PageLocked(page
));
3155 if (PageWriteback(page
))
3158 if (mapping
&& mapping
->a_ops
->releasepage
)
3159 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3160 return try_to_free_buffers(page
);
3163 EXPORT_SYMBOL(try_to_release_page
);