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/shmem_fs.h>
39 #include <linux/rmap.h>
42 #define CREATE_TRACE_POINTS
43 #include <trace/events/filemap.h>
46 * FIXME: remove all knowledge of the buffer layer from the core VM
48 #include <linux/buffer_head.h> /* for try_to_free_buffers */
53 * Shared mappings implemented 30.11.1994. It's not fully working yet,
56 * Shared mappings now work. 15.8.1995 Bruno.
58 * finished 'unifying' the page and buffer cache and SMP-threaded the
59 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
61 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
67 * ->i_mmap_rwsem (truncate_pagecache)
68 * ->private_lock (__free_pte->__set_page_dirty_buffers)
69 * ->swap_lock (exclusive_swap_page, others)
70 * ->mapping->tree_lock
73 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
77 * ->page_table_lock or pte_lock (various, mainly in memory.c)
78 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
81 * ->lock_page (access_process_vm)
83 * ->i_mutex (generic_perform_write)
84 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
87 * sb_lock (fs/fs-writeback.c)
88 * ->mapping->tree_lock (__sync_single_inode)
91 * ->anon_vma.lock (vma_adjust)
94 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
96 * ->page_table_lock or pte_lock
97 * ->swap_lock (try_to_unmap_one)
98 * ->private_lock (try_to_unmap_one)
99 * ->tree_lock (try_to_unmap_one)
100 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
101 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
102 * ->private_lock (page_remove_rmap->set_page_dirty)
103 * ->tree_lock (page_remove_rmap->set_page_dirty)
104 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
105 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
106 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
107 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
108 * ->inode->i_lock (zap_pte_range->set_page_dirty)
109 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
112 * ->tasklist_lock (memory_failure, collect_procs_ao)
115 static int page_cache_tree_insert(struct address_space
*mapping
,
116 struct page
*page
, void **shadowp
)
118 struct radix_tree_node
*node
;
122 error
= __radix_tree_create(&mapping
->page_tree
, page
->index
, 0,
129 p
= radix_tree_deref_slot_protected(slot
, &mapping
->tree_lock
);
130 if (!radix_tree_exceptional_entry(p
))
133 mapping
->nrexceptional
--;
137 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, page
,
138 workingset_lookup_update(mapping
));
143 static void page_cache_tree_delete(struct address_space
*mapping
,
144 struct page
*page
, void *shadow
)
148 /* hugetlb pages are represented by one entry in the radix tree */
149 nr
= PageHuge(page
) ? 1 : hpage_nr_pages(page
);
151 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
152 VM_BUG_ON_PAGE(PageTail(page
), page
);
153 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
155 for (i
= 0; i
< nr
; i
++) {
156 struct radix_tree_node
*node
;
159 __radix_tree_lookup(&mapping
->page_tree
, page
->index
+ i
,
162 VM_BUG_ON_PAGE(!node
&& nr
!= 1, page
);
164 radix_tree_clear_tags(&mapping
->page_tree
, node
, slot
);
165 __radix_tree_replace(&mapping
->page_tree
, node
, slot
, shadow
,
166 workingset_lookup_update(mapping
));
169 page
->mapping
= NULL
;
170 /* Leave page->index set: truncation lookup relies upon it */
173 mapping
->nrexceptional
+= nr
;
175 * Make sure the nrexceptional update is committed before
176 * the nrpages update so that final truncate racing
177 * with reclaim does not see both counters 0 at the
178 * same time and miss a shadow entry.
182 mapping
->nrpages
-= nr
;
185 static void unaccount_page_cache_page(struct address_space
*mapping
,
191 * if we're uptodate, flush out into the cleancache, otherwise
192 * invalidate any existing cleancache entries. We can't leave
193 * stale data around in the cleancache once our page is gone
195 if (PageUptodate(page
) && PageMappedToDisk(page
))
196 cleancache_put_page(page
);
198 cleancache_invalidate_page(mapping
, page
);
200 VM_BUG_ON_PAGE(PageTail(page
), page
);
201 VM_BUG_ON_PAGE(page_mapped(page
), page
);
202 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
205 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
206 current
->comm
, page_to_pfn(page
));
207 dump_page(page
, "still mapped when deleted");
209 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
211 mapcount
= page_mapcount(page
);
212 if (mapping_exiting(mapping
) &&
213 page_count(page
) >= mapcount
+ 2) {
215 * All vmas have already been torn down, so it's
216 * a good bet that actually the page is unmapped,
217 * and we'd prefer not to leak it: if we're wrong,
218 * some other bad page check should catch it later.
220 page_mapcount_reset(page
);
221 page_ref_sub(page
, mapcount
);
225 /* hugetlb pages do not participate in page cache accounting. */
229 nr
= hpage_nr_pages(page
);
231 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
232 if (PageSwapBacked(page
)) {
233 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
234 if (PageTransHuge(page
))
235 __dec_node_page_state(page
, NR_SHMEM_THPS
);
237 VM_BUG_ON_PAGE(PageTransHuge(page
), page
);
241 * At this point page must be either written or cleaned by
242 * truncate. Dirty page here signals a bug and loss of
245 * This fixes dirty accounting after removing the page entirely
246 * but leaves PageDirty set: it has no effect for truncated
247 * page and anyway will be cleared before returning page into
250 if (WARN_ON_ONCE(PageDirty(page
)))
251 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
255 * Delete a page from the page cache and free it. Caller has to make
256 * sure the page is locked and that nobody else uses it - or that usage
257 * is safe. The caller must hold the mapping's tree_lock.
259 void __delete_from_page_cache(struct page
*page
, void *shadow
)
261 struct address_space
*mapping
= page
->mapping
;
263 trace_mm_filemap_delete_from_page_cache(page
);
265 unaccount_page_cache_page(mapping
, page
);
266 page_cache_tree_delete(mapping
, page
, shadow
);
269 static void page_cache_free_page(struct address_space
*mapping
,
272 void (*freepage
)(struct page
*);
274 freepage
= mapping
->a_ops
->freepage
;
278 if (PageTransHuge(page
) && !PageHuge(page
)) {
279 page_ref_sub(page
, HPAGE_PMD_NR
);
280 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
287 * delete_from_page_cache - delete page from page cache
288 * @page: the page which the kernel is trying to remove from page cache
290 * This must be called only on pages that have been verified to be in the page
291 * cache and locked. It will never put the page into the free list, the caller
292 * has a reference on the page.
294 void delete_from_page_cache(struct page
*page
)
296 struct address_space
*mapping
= page_mapping(page
);
299 BUG_ON(!PageLocked(page
));
300 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
301 __delete_from_page_cache(page
, NULL
);
302 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
304 page_cache_free_page(mapping
, page
);
306 EXPORT_SYMBOL(delete_from_page_cache
);
309 * page_cache_tree_delete_batch - delete several pages from page cache
310 * @mapping: the mapping to which pages belong
311 * @pvec: pagevec with pages to delete
313 * The function walks over mapping->page_tree and removes pages passed in @pvec
314 * from the radix tree. The function expects @pvec to be sorted by page index.
315 * It tolerates holes in @pvec (radix tree entries at those indices are not
316 * modified). The function expects only THP head pages to be present in the
317 * @pvec and takes care to delete all corresponding tail pages from the radix
320 * The function expects mapping->tree_lock to be held.
323 page_cache_tree_delete_batch(struct address_space
*mapping
,
324 struct pagevec
*pvec
)
326 struct radix_tree_iter iter
;
329 int i
= 0, tail_pages
= 0;
333 start
= pvec
->pages
[0]->index
;
334 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
335 if (i
>= pagevec_count(pvec
) && !tail_pages
)
337 page
= radix_tree_deref_slot_protected(slot
,
338 &mapping
->tree_lock
);
339 if (radix_tree_exceptional_entry(page
))
343 * Some page got inserted in our range? Skip it. We
344 * have our pages locked so they are protected from
347 if (page
!= pvec
->pages
[i
])
349 WARN_ON_ONCE(!PageLocked(page
));
350 if (PageTransHuge(page
) && !PageHuge(page
))
351 tail_pages
= HPAGE_PMD_NR
- 1;
352 page
->mapping
= NULL
;
354 * Leave page->index set: truncation lookup relies
361 radix_tree_clear_tags(&mapping
->page_tree
, iter
.node
, slot
);
362 __radix_tree_replace(&mapping
->page_tree
, iter
.node
, slot
, NULL
,
363 workingset_lookup_update(mapping
));
366 mapping
->nrpages
-= total_pages
;
369 void delete_from_page_cache_batch(struct address_space
*mapping
,
370 struct pagevec
*pvec
)
375 if (!pagevec_count(pvec
))
378 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
379 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
380 trace_mm_filemap_delete_from_page_cache(pvec
->pages
[i
]);
382 unaccount_page_cache_page(mapping
, pvec
->pages
[i
]);
384 page_cache_tree_delete_batch(mapping
, pvec
);
385 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
387 for (i
= 0; i
< pagevec_count(pvec
); i
++)
388 page_cache_free_page(mapping
, pvec
->pages
[i
]);
391 int filemap_check_errors(struct address_space
*mapping
)
394 /* Check for outstanding write errors */
395 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
396 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
398 if (test_bit(AS_EIO
, &mapping
->flags
) &&
399 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
403 EXPORT_SYMBOL(filemap_check_errors
);
405 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
407 /* Check for outstanding write errors */
408 if (test_bit(AS_EIO
, &mapping
->flags
))
410 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
416 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
417 * @mapping: address space structure to write
418 * @start: offset in bytes where the range starts
419 * @end: offset in bytes where the range ends (inclusive)
420 * @sync_mode: enable synchronous operation
422 * Start writeback against all of a mapping's dirty pages that lie
423 * within the byte offsets <start, end> inclusive.
425 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
426 * opposed to a regular memory cleansing writeback. The difference between
427 * these two operations is that if a dirty page/buffer is encountered, it must
428 * be waited upon, and not just skipped over.
430 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
431 loff_t end
, int sync_mode
)
434 struct writeback_control wbc
= {
435 .sync_mode
= sync_mode
,
436 .nr_to_write
= LONG_MAX
,
437 .range_start
= start
,
441 if (!mapping_cap_writeback_dirty(mapping
) ||
442 !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
445 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
446 ret
= do_writepages(mapping
, &wbc
);
447 wbc_detach_inode(&wbc
);
451 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
454 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
457 int filemap_fdatawrite(struct address_space
*mapping
)
459 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
461 EXPORT_SYMBOL(filemap_fdatawrite
);
463 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
466 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
468 EXPORT_SYMBOL(filemap_fdatawrite_range
);
471 * filemap_flush - mostly a non-blocking flush
472 * @mapping: target address_space
474 * This is a mostly non-blocking flush. Not suitable for data-integrity
475 * purposes - I/O may not be started against all dirty pages.
477 int filemap_flush(struct address_space
*mapping
)
479 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
481 EXPORT_SYMBOL(filemap_flush
);
484 * filemap_range_has_page - check if a page exists in range.
485 * @mapping: address space within which to check
486 * @start_byte: offset in bytes where the range starts
487 * @end_byte: offset in bytes where the range ends (inclusive)
489 * Find at least one page in the range supplied, usually used to check if
490 * direct writing in this range will trigger a writeback.
492 bool filemap_range_has_page(struct address_space
*mapping
,
493 loff_t start_byte
, loff_t end_byte
)
495 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
496 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
499 if (end_byte
< start_byte
)
502 if (mapping
->nrpages
== 0)
505 if (!find_get_pages_range(mapping
, &index
, end
, 1, &page
))
510 EXPORT_SYMBOL(filemap_range_has_page
);
512 static void __filemap_fdatawait_range(struct address_space
*mapping
,
513 loff_t start_byte
, loff_t end_byte
)
515 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
516 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
520 if (end_byte
< start_byte
)
524 while (index
<= end
) {
527 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
,
528 end
, PAGECACHE_TAG_WRITEBACK
);
532 for (i
= 0; i
< nr_pages
; i
++) {
533 struct page
*page
= pvec
.pages
[i
];
535 wait_on_page_writeback(page
);
536 ClearPageError(page
);
538 pagevec_release(&pvec
);
544 * filemap_fdatawait_range - wait for writeback to complete
545 * @mapping: address space structure to wait for
546 * @start_byte: offset in bytes where the range starts
547 * @end_byte: offset in bytes where the range ends (inclusive)
549 * Walk the list of under-writeback pages of the given address space
550 * in the given range and wait for all of them. Check error status of
551 * the address space and return it.
553 * Since the error status of the address space is cleared by this function,
554 * callers are responsible for checking the return value and handling and/or
555 * reporting the error.
557 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
560 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
561 return filemap_check_errors(mapping
);
563 EXPORT_SYMBOL(filemap_fdatawait_range
);
566 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
567 * @mapping: address space structure to wait for
568 * @start_byte: offset in bytes where the range starts
569 * @end_byte: offset in bytes where the range ends (inclusive)
571 * Walk the list of under-writeback pages of the given address space in the
572 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
573 * this function does not clear error status of the address space.
575 * Use this function if callers don't handle errors themselves. Expected
576 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
579 int filemap_fdatawait_range_keep_errors(struct address_space
*mapping
,
580 loff_t start_byte
, loff_t end_byte
)
582 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
583 return filemap_check_and_keep_errors(mapping
);
585 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors
);
588 * file_fdatawait_range - wait for writeback to complete
589 * @file: file pointing to address space structure to wait for
590 * @start_byte: offset in bytes where the range starts
591 * @end_byte: offset in bytes where the range ends (inclusive)
593 * Walk the list of under-writeback pages of the address space that file
594 * refers to, in the given range and wait for all of them. Check error
595 * status of the address space vs. the file->f_wb_err cursor and return it.
597 * Since the error status of the file is advanced by this function,
598 * callers are responsible for checking the return value and handling and/or
599 * reporting the error.
601 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
603 struct address_space
*mapping
= file
->f_mapping
;
605 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
606 return file_check_and_advance_wb_err(file
);
608 EXPORT_SYMBOL(file_fdatawait_range
);
611 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
612 * @mapping: address space structure to wait for
614 * Walk the list of under-writeback pages of the given address space
615 * and wait for all of them. Unlike filemap_fdatawait(), this function
616 * does not clear error status of the address space.
618 * Use this function if callers don't handle errors themselves. Expected
619 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
622 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
624 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
625 return filemap_check_and_keep_errors(mapping
);
627 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
629 static bool mapping_needs_writeback(struct address_space
*mapping
)
631 return (!dax_mapping(mapping
) && mapping
->nrpages
) ||
632 (dax_mapping(mapping
) && mapping
->nrexceptional
);
635 int filemap_write_and_wait(struct address_space
*mapping
)
639 if (mapping_needs_writeback(mapping
)) {
640 err
= filemap_fdatawrite(mapping
);
642 * Even if the above returned error, the pages may be
643 * written partially (e.g. -ENOSPC), so we wait for it.
644 * But the -EIO is special case, it may indicate the worst
645 * thing (e.g. bug) happened, so we avoid waiting for it.
648 int err2
= filemap_fdatawait(mapping
);
652 /* Clear any previously stored errors */
653 filemap_check_errors(mapping
);
656 err
= filemap_check_errors(mapping
);
660 EXPORT_SYMBOL(filemap_write_and_wait
);
663 * filemap_write_and_wait_range - write out & wait on a file range
664 * @mapping: the address_space for the pages
665 * @lstart: offset in bytes where the range starts
666 * @lend: offset in bytes where the range ends (inclusive)
668 * Write out and wait upon file offsets lstart->lend, inclusive.
670 * Note that @lend is inclusive (describes the last byte to be written) so
671 * that this function can be used to write to the very end-of-file (end = -1).
673 int filemap_write_and_wait_range(struct address_space
*mapping
,
674 loff_t lstart
, loff_t lend
)
678 if (mapping_needs_writeback(mapping
)) {
679 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
681 /* See comment of filemap_write_and_wait() */
683 int err2
= filemap_fdatawait_range(mapping
,
688 /* Clear any previously stored errors */
689 filemap_check_errors(mapping
);
692 err
= filemap_check_errors(mapping
);
696 EXPORT_SYMBOL(filemap_write_and_wait_range
);
698 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
700 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
702 trace_filemap_set_wb_err(mapping
, eseq
);
704 EXPORT_SYMBOL(__filemap_set_wb_err
);
707 * file_check_and_advance_wb_err - report wb error (if any) that was previously
708 * and advance wb_err to current one
709 * @file: struct file on which the error is being reported
711 * When userland calls fsync (or something like nfsd does the equivalent), we
712 * want to report any writeback errors that occurred since the last fsync (or
713 * since the file was opened if there haven't been any).
715 * Grab the wb_err from the mapping. If it matches what we have in the file,
716 * then just quickly return 0. The file is all caught up.
718 * If it doesn't match, then take the mapping value, set the "seen" flag in
719 * it and try to swap it into place. If it works, or another task beat us
720 * to it with the new value, then update the f_wb_err and return the error
721 * portion. The error at this point must be reported via proper channels
722 * (a'la fsync, or NFS COMMIT operation, etc.).
724 * While we handle mapping->wb_err with atomic operations, the f_wb_err
725 * value is protected by the f_lock since we must ensure that it reflects
726 * the latest value swapped in for this file descriptor.
728 int file_check_and_advance_wb_err(struct file
*file
)
731 errseq_t old
= READ_ONCE(file
->f_wb_err
);
732 struct address_space
*mapping
= file
->f_mapping
;
734 /* Locklessly handle the common case where nothing has changed */
735 if (errseq_check(&mapping
->wb_err
, old
)) {
736 /* Something changed, must use slow path */
737 spin_lock(&file
->f_lock
);
738 old
= file
->f_wb_err
;
739 err
= errseq_check_and_advance(&mapping
->wb_err
,
741 trace_file_check_and_advance_wb_err(file
, old
);
742 spin_unlock(&file
->f_lock
);
746 * We're mostly using this function as a drop in replacement for
747 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
748 * that the legacy code would have had on these flags.
750 clear_bit(AS_EIO
, &mapping
->flags
);
751 clear_bit(AS_ENOSPC
, &mapping
->flags
);
754 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
757 * file_write_and_wait_range - write out & wait on a file range
758 * @file: file pointing to address_space with pages
759 * @lstart: offset in bytes where the range starts
760 * @lend: offset in bytes where the range ends (inclusive)
762 * Write out and wait upon file offsets lstart->lend, inclusive.
764 * Note that @lend is inclusive (describes the last byte to be written) so
765 * that this function can be used to write to the very end-of-file (end = -1).
767 * After writing out and waiting on the data, we check and advance the
768 * f_wb_err cursor to the latest value, and return any errors detected there.
770 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
773 struct address_space
*mapping
= file
->f_mapping
;
775 if (mapping_needs_writeback(mapping
)) {
776 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
778 /* See comment of filemap_write_and_wait() */
780 __filemap_fdatawait_range(mapping
, lstart
, lend
);
782 err2
= file_check_and_advance_wb_err(file
);
787 EXPORT_SYMBOL(file_write_and_wait_range
);
790 * replace_page_cache_page - replace a pagecache page with a new one
791 * @old: page to be replaced
792 * @new: page to replace with
793 * @gfp_mask: allocation mode
795 * This function replaces a page in the pagecache with a new one. On
796 * success it acquires the pagecache reference for the new page and
797 * drops it for the old page. Both the old and new pages must be
798 * locked. This function does not add the new page to the LRU, the
799 * caller must do that.
801 * The remove + add is atomic. The only way this function can fail is
802 * memory allocation failure.
804 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
808 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
809 VM_BUG_ON_PAGE(!PageLocked(new), new);
810 VM_BUG_ON_PAGE(new->mapping
, new);
812 error
= radix_tree_preload(gfp_mask
& GFP_RECLAIM_MASK
);
814 struct address_space
*mapping
= old
->mapping
;
815 void (*freepage
)(struct page
*);
818 pgoff_t offset
= old
->index
;
819 freepage
= mapping
->a_ops
->freepage
;
822 new->mapping
= mapping
;
825 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
826 __delete_from_page_cache(old
, NULL
);
827 error
= page_cache_tree_insert(mapping
, new, NULL
);
831 * hugetlb pages do not participate in page cache accounting.
834 __inc_node_page_state(new, NR_FILE_PAGES
);
835 if (PageSwapBacked(new))
836 __inc_node_page_state(new, NR_SHMEM
);
837 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
838 mem_cgroup_migrate(old
, new);
839 radix_tree_preload_end();
847 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
849 static int __add_to_page_cache_locked(struct page
*page
,
850 struct address_space
*mapping
,
851 pgoff_t offset
, gfp_t gfp_mask
,
854 int huge
= PageHuge(page
);
855 struct mem_cgroup
*memcg
;
858 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
859 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
862 error
= mem_cgroup_try_charge(page
, current
->mm
,
863 gfp_mask
, &memcg
, false);
868 error
= radix_tree_maybe_preload(gfp_mask
& GFP_RECLAIM_MASK
);
871 mem_cgroup_cancel_charge(page
, memcg
, false);
876 page
->mapping
= mapping
;
877 page
->index
= offset
;
879 spin_lock_irq(&mapping
->tree_lock
);
880 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
881 radix_tree_preload_end();
885 /* hugetlb pages do not participate in page cache accounting. */
887 __inc_node_page_state(page
, NR_FILE_PAGES
);
888 spin_unlock_irq(&mapping
->tree_lock
);
890 mem_cgroup_commit_charge(page
, memcg
, false, false);
891 trace_mm_filemap_add_to_page_cache(page
);
894 page
->mapping
= NULL
;
895 /* Leave page->index set: truncation relies upon it */
896 spin_unlock_irq(&mapping
->tree_lock
);
898 mem_cgroup_cancel_charge(page
, memcg
, false);
904 * add_to_page_cache_locked - add a locked page to the pagecache
906 * @mapping: the page's address_space
907 * @offset: page index
908 * @gfp_mask: page allocation mode
910 * This function is used to add a page to the pagecache. It must be locked.
911 * This function does not add the page to the LRU. The caller must do that.
913 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
914 pgoff_t offset
, gfp_t gfp_mask
)
916 return __add_to_page_cache_locked(page
, mapping
, offset
,
919 EXPORT_SYMBOL(add_to_page_cache_locked
);
921 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
922 pgoff_t offset
, gfp_t gfp_mask
)
927 __SetPageLocked(page
);
928 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
931 __ClearPageLocked(page
);
934 * The page might have been evicted from cache only
935 * recently, in which case it should be activated like
936 * any other repeatedly accessed page.
937 * The exception is pages getting rewritten; evicting other
938 * data from the working set, only to cache data that will
939 * get overwritten with something else, is a waste of memory.
941 if (!(gfp_mask
& __GFP_WRITE
) &&
942 shadow
&& workingset_refault(shadow
)) {
944 workingset_activation(page
);
946 ClearPageActive(page
);
951 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
954 struct page
*__page_cache_alloc(gfp_t gfp
)
959 if (cpuset_do_page_mem_spread()) {
960 unsigned int cpuset_mems_cookie
;
962 cpuset_mems_cookie
= read_mems_allowed_begin();
963 n
= cpuset_mem_spread_node();
964 page
= __alloc_pages_node(n
, gfp
, 0);
965 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
969 return alloc_pages(gfp
, 0);
971 EXPORT_SYMBOL(__page_cache_alloc
);
975 * In order to wait for pages to become available there must be
976 * waitqueues associated with pages. By using a hash table of
977 * waitqueues where the bucket discipline is to maintain all
978 * waiters on the same queue and wake all when any of the pages
979 * become available, and for the woken contexts to check to be
980 * sure the appropriate page became available, this saves space
981 * at a cost of "thundering herd" phenomena during rare hash
984 #define PAGE_WAIT_TABLE_BITS 8
985 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
986 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
988 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
990 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
993 void __init
pagecache_init(void)
997 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
998 init_waitqueue_head(&page_wait_table
[i
]);
1000 page_writeback_init();
1003 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1004 struct wait_page_key
{
1010 struct wait_page_queue
{
1013 wait_queue_entry_t wait
;
1016 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
1018 struct wait_page_key
*key
= arg
;
1019 struct wait_page_queue
*wait_page
1020 = container_of(wait
, struct wait_page_queue
, wait
);
1022 if (wait_page
->page
!= key
->page
)
1024 key
->page_match
= 1;
1026 if (wait_page
->bit_nr
!= key
->bit_nr
)
1029 /* Stop walking if it's locked */
1030 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
1033 return autoremove_wake_function(wait
, mode
, sync
, key
);
1036 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
1038 wait_queue_head_t
*q
= page_waitqueue(page
);
1039 struct wait_page_key key
;
1040 unsigned long flags
;
1041 wait_queue_entry_t bookmark
;
1044 key
.bit_nr
= bit_nr
;
1048 bookmark
.private = NULL
;
1049 bookmark
.func
= NULL
;
1050 INIT_LIST_HEAD(&bookmark
.entry
);
1052 spin_lock_irqsave(&q
->lock
, flags
);
1053 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1055 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1057 * Take a breather from holding the lock,
1058 * allow pages that finish wake up asynchronously
1059 * to acquire the lock and remove themselves
1062 spin_unlock_irqrestore(&q
->lock
, flags
);
1064 spin_lock_irqsave(&q
->lock
, flags
);
1065 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1069 * It is possible for other pages to have collided on the waitqueue
1070 * hash, so in that case check for a page match. That prevents a long-
1073 * It is still possible to miss a case here, when we woke page waiters
1074 * and removed them from the waitqueue, but there are still other
1077 if (!waitqueue_active(q
) || !key
.page_match
) {
1078 ClearPageWaiters(page
);
1080 * It's possible to miss clearing Waiters here, when we woke
1081 * our page waiters, but the hashed waitqueue has waiters for
1082 * other pages on it.
1084 * That's okay, it's a rare case. The next waker will clear it.
1087 spin_unlock_irqrestore(&q
->lock
, flags
);
1090 static void wake_up_page(struct page
*page
, int bit
)
1092 if (!PageWaiters(page
))
1094 wake_up_page_bit(page
, bit
);
1097 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
1098 struct page
*page
, int bit_nr
, int state
, bool lock
)
1100 struct wait_page_queue wait_page
;
1101 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1105 wait
->flags
= lock
? WQ_FLAG_EXCLUSIVE
: 0;
1106 wait
->func
= wake_page_function
;
1107 wait_page
.page
= page
;
1108 wait_page
.bit_nr
= bit_nr
;
1111 spin_lock_irq(&q
->lock
);
1113 if (likely(list_empty(&wait
->entry
))) {
1114 __add_wait_queue_entry_tail(q
, wait
);
1115 SetPageWaiters(page
);
1118 set_current_state(state
);
1120 spin_unlock_irq(&q
->lock
);
1122 if (likely(test_bit(bit_nr
, &page
->flags
))) {
1127 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
1130 if (!test_bit(bit_nr
, &page
->flags
))
1134 if (unlikely(signal_pending_state(state
, current
))) {
1140 finish_wait(q
, wait
);
1143 * A signal could leave PageWaiters set. Clearing it here if
1144 * !waitqueue_active would be possible (by open-coding finish_wait),
1145 * but still fail to catch it in the case of wait hash collision. We
1146 * already can fail to clear wait hash collision cases, so don't
1147 * bother with signals either.
1153 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1155 wait_queue_head_t
*q
= page_waitqueue(page
);
1156 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, false);
1158 EXPORT_SYMBOL(wait_on_page_bit
);
1160 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1162 wait_queue_head_t
*q
= page_waitqueue(page
);
1163 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, false);
1165 EXPORT_SYMBOL(wait_on_page_bit_killable
);
1168 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1169 * @page: Page defining the wait queue of interest
1170 * @waiter: Waiter to add to the queue
1172 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1174 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1176 wait_queue_head_t
*q
= page_waitqueue(page
);
1177 unsigned long flags
;
1179 spin_lock_irqsave(&q
->lock
, flags
);
1180 __add_wait_queue_entry_tail(q
, waiter
);
1181 SetPageWaiters(page
);
1182 spin_unlock_irqrestore(&q
->lock
, flags
);
1184 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1186 #ifndef clear_bit_unlock_is_negative_byte
1189 * PG_waiters is the high bit in the same byte as PG_lock.
1191 * On x86 (and on many other architectures), we can clear PG_lock and
1192 * test the sign bit at the same time. But if the architecture does
1193 * not support that special operation, we just do this all by hand
1196 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1197 * being cleared, but a memory barrier should be unneccssary since it is
1198 * in the same byte as PG_locked.
1200 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1202 clear_bit_unlock(nr
, mem
);
1203 /* smp_mb__after_atomic(); */
1204 return test_bit(PG_waiters
, mem
);
1210 * unlock_page - unlock a locked page
1213 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1214 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1215 * mechanism between PageLocked pages and PageWriteback pages is shared.
1216 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1218 * Note that this depends on PG_waiters being the sign bit in the byte
1219 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1220 * clear the PG_locked bit and test PG_waiters at the same time fairly
1221 * portably (architectures that do LL/SC can test any bit, while x86 can
1222 * test the sign bit).
1224 void unlock_page(struct page
*page
)
1226 BUILD_BUG_ON(PG_waiters
!= 7);
1227 page
= compound_head(page
);
1228 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1229 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1230 wake_up_page_bit(page
, PG_locked
);
1232 EXPORT_SYMBOL(unlock_page
);
1235 * end_page_writeback - end writeback against a page
1238 void end_page_writeback(struct page
*page
)
1241 * TestClearPageReclaim could be used here but it is an atomic
1242 * operation and overkill in this particular case. Failing to
1243 * shuffle a page marked for immediate reclaim is too mild to
1244 * justify taking an atomic operation penalty at the end of
1245 * ever page writeback.
1247 if (PageReclaim(page
)) {
1248 ClearPageReclaim(page
);
1249 rotate_reclaimable_page(page
);
1252 if (!test_clear_page_writeback(page
))
1255 smp_mb__after_atomic();
1256 wake_up_page(page
, PG_writeback
);
1258 EXPORT_SYMBOL(end_page_writeback
);
1261 * After completing I/O on a page, call this routine to update the page
1262 * flags appropriately
1264 void page_endio(struct page
*page
, bool is_write
, int err
)
1268 SetPageUptodate(page
);
1270 ClearPageUptodate(page
);
1276 struct address_space
*mapping
;
1279 mapping
= page_mapping(page
);
1281 mapping_set_error(mapping
, err
);
1283 end_page_writeback(page
);
1286 EXPORT_SYMBOL_GPL(page_endio
);
1289 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1290 * @__page: the page to lock
1292 void __lock_page(struct page
*__page
)
1294 struct page
*page
= compound_head(__page
);
1295 wait_queue_head_t
*q
= page_waitqueue(page
);
1296 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, true);
1298 EXPORT_SYMBOL(__lock_page
);
1300 int __lock_page_killable(struct page
*__page
)
1302 struct page
*page
= compound_head(__page
);
1303 wait_queue_head_t
*q
= page_waitqueue(page
);
1304 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
, true);
1306 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1310 * 1 - page is locked; mmap_sem is still held.
1311 * 0 - page is not locked.
1312 * mmap_sem has been released (up_read()), unless flags had both
1313 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1314 * which case mmap_sem is still held.
1316 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1317 * with the page locked and the mmap_sem unperturbed.
1319 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1322 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1324 * CAUTION! In this case, mmap_sem is not released
1325 * even though return 0.
1327 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1330 up_read(&mm
->mmap_sem
);
1331 if (flags
& FAULT_FLAG_KILLABLE
)
1332 wait_on_page_locked_killable(page
);
1334 wait_on_page_locked(page
);
1337 if (flags
& FAULT_FLAG_KILLABLE
) {
1340 ret
= __lock_page_killable(page
);
1342 up_read(&mm
->mmap_sem
);
1352 * page_cache_next_hole - find the next hole (not-present entry)
1355 * @max_scan: maximum range to search
1357 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1358 * lowest indexed hole.
1360 * Returns: the index of the hole if found, otherwise returns an index
1361 * outside of the set specified (in which case 'return - index >=
1362 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1365 * page_cache_next_hole may be called under rcu_read_lock. However,
1366 * like radix_tree_gang_lookup, this will not atomically search a
1367 * snapshot of the tree at a single point in time. For example, if a
1368 * hole is created at index 5, then subsequently a hole is created at
1369 * index 10, page_cache_next_hole covering both indexes may return 10
1370 * if called under rcu_read_lock.
1372 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1373 pgoff_t index
, unsigned long max_scan
)
1377 for (i
= 0; i
< max_scan
; i
++) {
1380 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1381 if (!page
|| radix_tree_exceptional_entry(page
))
1390 EXPORT_SYMBOL(page_cache_next_hole
);
1393 * page_cache_prev_hole - find the prev hole (not-present entry)
1396 * @max_scan: maximum range to search
1398 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1401 * Returns: the index of the hole if found, otherwise returns an index
1402 * outside of the set specified (in which case 'index - return >=
1403 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1406 * page_cache_prev_hole may be called under rcu_read_lock. However,
1407 * like radix_tree_gang_lookup, this will not atomically search a
1408 * snapshot of the tree at a single point in time. For example, if a
1409 * hole is created at index 10, then subsequently a hole is created at
1410 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1411 * called under rcu_read_lock.
1413 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1414 pgoff_t index
, unsigned long max_scan
)
1418 for (i
= 0; i
< max_scan
; i
++) {
1421 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1422 if (!page
|| radix_tree_exceptional_entry(page
))
1425 if (index
== ULONG_MAX
)
1431 EXPORT_SYMBOL(page_cache_prev_hole
);
1434 * find_get_entry - find and get a page cache entry
1435 * @mapping: the address_space to search
1436 * @offset: the page cache index
1438 * Looks up the page cache slot at @mapping & @offset. If there is a
1439 * page cache page, it is returned with an increased refcount.
1441 * If the slot holds a shadow entry of a previously evicted page, or a
1442 * swap entry from shmem/tmpfs, it is returned.
1444 * Otherwise, %NULL is returned.
1446 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1449 struct page
*head
, *page
;
1454 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1456 page
= radix_tree_deref_slot(pagep
);
1457 if (unlikely(!page
))
1459 if (radix_tree_exception(page
)) {
1460 if (radix_tree_deref_retry(page
))
1463 * A shadow entry of a recently evicted page,
1464 * or a swap entry from shmem/tmpfs. Return
1465 * it without attempting to raise page count.
1470 head
= compound_head(page
);
1471 if (!page_cache_get_speculative(head
))
1474 /* The page was split under us? */
1475 if (compound_head(page
) != head
) {
1481 * Has the page moved?
1482 * This is part of the lockless pagecache protocol. See
1483 * include/linux/pagemap.h for details.
1485 if (unlikely(page
!= *pagep
)) {
1495 EXPORT_SYMBOL(find_get_entry
);
1498 * find_lock_entry - locate, pin and lock a page cache entry
1499 * @mapping: the address_space to search
1500 * @offset: the page cache index
1502 * Looks up the page cache slot at @mapping & @offset. If there is a
1503 * page cache page, it is returned locked and with an increased
1506 * If the slot holds a shadow entry of a previously evicted page, or a
1507 * swap entry from shmem/tmpfs, it is returned.
1509 * Otherwise, %NULL is returned.
1511 * find_lock_entry() may sleep.
1513 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1518 page
= find_get_entry(mapping
, offset
);
1519 if (page
&& !radix_tree_exception(page
)) {
1521 /* Has the page been truncated? */
1522 if (unlikely(page_mapping(page
) != mapping
)) {
1527 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1531 EXPORT_SYMBOL(find_lock_entry
);
1534 * pagecache_get_page - find and get a page reference
1535 * @mapping: the address_space to search
1536 * @offset: the page index
1537 * @fgp_flags: PCG flags
1538 * @gfp_mask: gfp mask to use for the page cache data page allocation
1540 * Looks up the page cache slot at @mapping & @offset.
1542 * PCG flags modify how the page is returned.
1544 * @fgp_flags can be:
1546 * - FGP_ACCESSED: the page will be marked accessed
1547 * - FGP_LOCK: Page is return locked
1548 * - FGP_CREAT: If page is not present then a new page is allocated using
1549 * @gfp_mask and added to the page cache and the VM's LRU
1550 * list. The page is returned locked and with an increased
1551 * refcount. Otherwise, NULL is returned.
1553 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1554 * if the GFP flags specified for FGP_CREAT are atomic.
1556 * If there is a page cache page, it is returned with an increased refcount.
1558 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1559 int fgp_flags
, gfp_t gfp_mask
)
1564 page
= find_get_entry(mapping
, offset
);
1565 if (radix_tree_exceptional_entry(page
))
1570 if (fgp_flags
& FGP_LOCK
) {
1571 if (fgp_flags
& FGP_NOWAIT
) {
1572 if (!trylock_page(page
)) {
1580 /* Has the page been truncated? */
1581 if (unlikely(page
->mapping
!= mapping
)) {
1586 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1589 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1590 mark_page_accessed(page
);
1593 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1595 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1596 gfp_mask
|= __GFP_WRITE
;
1597 if (fgp_flags
& FGP_NOFS
)
1598 gfp_mask
&= ~__GFP_FS
;
1600 page
= __page_cache_alloc(gfp_mask
);
1604 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1605 fgp_flags
|= FGP_LOCK
;
1607 /* Init accessed so avoid atomic mark_page_accessed later */
1608 if (fgp_flags
& FGP_ACCESSED
)
1609 __SetPageReferenced(page
);
1611 err
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
1612 if (unlikely(err
)) {
1622 EXPORT_SYMBOL(pagecache_get_page
);
1625 * find_get_entries - gang pagecache lookup
1626 * @mapping: The address_space to search
1627 * @start: The starting page cache index
1628 * @nr_entries: The maximum number of entries
1629 * @entries: Where the resulting entries are placed
1630 * @indices: The cache indices corresponding to the entries in @entries
1632 * find_get_entries() will search for and return a group of up to
1633 * @nr_entries entries in the mapping. The entries are placed at
1634 * @entries. find_get_entries() takes a reference against any actual
1637 * The search returns a group of mapping-contiguous page cache entries
1638 * with ascending indexes. There may be holes in the indices due to
1639 * not-present pages.
1641 * Any shadow entries of evicted pages, or swap entries from
1642 * shmem/tmpfs, are included in the returned array.
1644 * find_get_entries() returns the number of pages and shadow entries
1647 unsigned find_get_entries(struct address_space
*mapping
,
1648 pgoff_t start
, unsigned int nr_entries
,
1649 struct page
**entries
, pgoff_t
*indices
)
1652 unsigned int ret
= 0;
1653 struct radix_tree_iter iter
;
1659 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1660 struct page
*head
, *page
;
1662 page
= radix_tree_deref_slot(slot
);
1663 if (unlikely(!page
))
1665 if (radix_tree_exception(page
)) {
1666 if (radix_tree_deref_retry(page
)) {
1667 slot
= radix_tree_iter_retry(&iter
);
1671 * A shadow entry of a recently evicted page, a swap
1672 * entry from shmem/tmpfs or a DAX entry. Return it
1673 * without attempting to raise page count.
1678 head
= compound_head(page
);
1679 if (!page_cache_get_speculative(head
))
1682 /* The page was split under us? */
1683 if (compound_head(page
) != head
) {
1688 /* Has the page moved? */
1689 if (unlikely(page
!= *slot
)) {
1694 indices
[ret
] = iter
.index
;
1695 entries
[ret
] = page
;
1696 if (++ret
== nr_entries
)
1704 * find_get_pages_range - gang pagecache lookup
1705 * @mapping: The address_space to search
1706 * @start: The starting page index
1707 * @end: The final page index (inclusive)
1708 * @nr_pages: The maximum number of pages
1709 * @pages: Where the resulting pages are placed
1711 * find_get_pages_range() will search for and return a group of up to @nr_pages
1712 * pages in the mapping starting at index @start and up to index @end
1713 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1714 * a reference against the returned pages.
1716 * The search returns a group of mapping-contiguous pages with ascending
1717 * indexes. There may be holes in the indices due to not-present pages.
1718 * We also update @start to index the next page for the traversal.
1720 * find_get_pages_range() returns the number of pages which were found. If this
1721 * number is smaller than @nr_pages, the end of specified range has been
1724 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
1725 pgoff_t end
, unsigned int nr_pages
,
1726 struct page
**pages
)
1728 struct radix_tree_iter iter
;
1732 if (unlikely(!nr_pages
))
1736 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, *start
) {
1737 struct page
*head
, *page
;
1739 if (iter
.index
> end
)
1742 page
= radix_tree_deref_slot(slot
);
1743 if (unlikely(!page
))
1746 if (radix_tree_exception(page
)) {
1747 if (radix_tree_deref_retry(page
)) {
1748 slot
= radix_tree_iter_retry(&iter
);
1752 * A shadow entry of a recently evicted page,
1753 * or a swap entry from shmem/tmpfs. Skip
1759 head
= compound_head(page
);
1760 if (!page_cache_get_speculative(head
))
1763 /* The page was split under us? */
1764 if (compound_head(page
) != head
) {
1769 /* Has the page moved? */
1770 if (unlikely(page
!= *slot
)) {
1776 if (++ret
== nr_pages
) {
1777 *start
= pages
[ret
- 1]->index
+ 1;
1783 * We come here when there is no page beyond @end. We take care to not
1784 * overflow the index @start as it confuses some of the callers. This
1785 * breaks the iteration when there is page at index -1 but that is
1786 * already broken anyway.
1788 if (end
== (pgoff_t
)-1)
1789 *start
= (pgoff_t
)-1;
1799 * find_get_pages_contig - gang contiguous pagecache lookup
1800 * @mapping: The address_space to search
1801 * @index: The starting page index
1802 * @nr_pages: The maximum number of pages
1803 * @pages: Where the resulting pages are placed
1805 * find_get_pages_contig() works exactly like find_get_pages(), except
1806 * that the returned number of pages are guaranteed to be contiguous.
1808 * find_get_pages_contig() returns the number of pages which were found.
1810 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1811 unsigned int nr_pages
, struct page
**pages
)
1813 struct radix_tree_iter iter
;
1815 unsigned int ret
= 0;
1817 if (unlikely(!nr_pages
))
1821 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1822 struct page
*head
, *page
;
1824 page
= radix_tree_deref_slot(slot
);
1825 /* The hole, there no reason to continue */
1826 if (unlikely(!page
))
1829 if (radix_tree_exception(page
)) {
1830 if (radix_tree_deref_retry(page
)) {
1831 slot
= radix_tree_iter_retry(&iter
);
1835 * A shadow entry of a recently evicted page,
1836 * or a swap entry from shmem/tmpfs. Stop
1837 * looking for contiguous pages.
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
)) {
1859 * must check mapping and index after taking the ref.
1860 * otherwise we can get both false positives and false
1861 * negatives, which is just confusing to the caller.
1863 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1869 if (++ret
== nr_pages
)
1875 EXPORT_SYMBOL(find_get_pages_contig
);
1878 * find_get_pages_range_tag - find and return pages in given range matching @tag
1879 * @mapping: the address_space to search
1880 * @index: the starting page index
1881 * @end: The final page index (inclusive)
1882 * @tag: the tag index
1883 * @nr_pages: the maximum number of pages
1884 * @pages: where the resulting pages are placed
1886 * Like find_get_pages, except we only return pages which are tagged with
1887 * @tag. We update @index to index the next page for the traversal.
1889 unsigned find_get_pages_range_tag(struct address_space
*mapping
, pgoff_t
*index
,
1890 pgoff_t end
, int tag
, unsigned int nr_pages
,
1891 struct page
**pages
)
1893 struct radix_tree_iter iter
;
1897 if (unlikely(!nr_pages
))
1901 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1902 &iter
, *index
, tag
) {
1903 struct page
*head
, *page
;
1905 if (iter
.index
> end
)
1908 page
= radix_tree_deref_slot(slot
);
1909 if (unlikely(!page
))
1912 if (radix_tree_exception(page
)) {
1913 if (radix_tree_deref_retry(page
)) {
1914 slot
= radix_tree_iter_retry(&iter
);
1918 * A shadow entry of a recently evicted page.
1920 * Those entries should never be tagged, but
1921 * this tree walk is lockless and the tags are
1922 * looked up in bulk, one radix tree node at a
1923 * time, so there is a sizable window for page
1924 * reclaim to evict a page we saw tagged.
1931 head
= compound_head(page
);
1932 if (!page_cache_get_speculative(head
))
1935 /* The page was split under us? */
1936 if (compound_head(page
) != head
) {
1941 /* Has the page moved? */
1942 if (unlikely(page
!= *slot
)) {
1948 if (++ret
== nr_pages
) {
1949 *index
= pages
[ret
- 1]->index
+ 1;
1955 * We come here when we got at @end. We take care to not overflow the
1956 * index @index as it confuses some of the callers. This breaks the
1957 * iteration when there is page at index -1 but that is already broken
1960 if (end
== (pgoff_t
)-1)
1961 *index
= (pgoff_t
)-1;
1969 EXPORT_SYMBOL(find_get_pages_range_tag
);
1972 * find_get_entries_tag - find and return entries that match @tag
1973 * @mapping: the address_space to search
1974 * @start: the starting page cache index
1975 * @tag: the tag index
1976 * @nr_entries: the maximum number of entries
1977 * @entries: where the resulting entries are placed
1978 * @indices: the cache indices corresponding to the entries in @entries
1980 * Like find_get_entries, except we only return entries which are tagged with
1983 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1984 int tag
, unsigned int nr_entries
,
1985 struct page
**entries
, pgoff_t
*indices
)
1988 unsigned int ret
= 0;
1989 struct radix_tree_iter iter
;
1995 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1996 &iter
, start
, tag
) {
1997 struct page
*head
, *page
;
1999 page
= radix_tree_deref_slot(slot
);
2000 if (unlikely(!page
))
2002 if (radix_tree_exception(page
)) {
2003 if (radix_tree_deref_retry(page
)) {
2004 slot
= radix_tree_iter_retry(&iter
);
2009 * A shadow entry of a recently evicted page, a swap
2010 * entry from shmem/tmpfs or a DAX entry. Return it
2011 * without attempting to raise page count.
2016 head
= compound_head(page
);
2017 if (!page_cache_get_speculative(head
))
2020 /* The page was split under us? */
2021 if (compound_head(page
) != head
) {
2026 /* Has the page moved? */
2027 if (unlikely(page
!= *slot
)) {
2032 indices
[ret
] = iter
.index
;
2033 entries
[ret
] = page
;
2034 if (++ret
== nr_entries
)
2040 EXPORT_SYMBOL(find_get_entries_tag
);
2043 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2044 * a _large_ part of the i/o request. Imagine the worst scenario:
2046 * ---R__________________________________________B__________
2047 * ^ reading here ^ bad block(assume 4k)
2049 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2050 * => failing the whole request => read(R) => read(R+1) =>
2051 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2052 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2053 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2055 * It is going insane. Fix it by quickly scaling down the readahead size.
2057 static void shrink_readahead_size_eio(struct file
*filp
,
2058 struct file_ra_state
*ra
)
2064 * generic_file_buffered_read - generic file read routine
2065 * @iocb: the iocb to read
2066 * @iter: data destination
2067 * @written: already copied
2069 * This is a generic file read routine, and uses the
2070 * mapping->a_ops->readpage() function for the actual low-level stuff.
2072 * This is really ugly. But the goto's actually try to clarify some
2073 * of the logic when it comes to error handling etc.
2075 static ssize_t
generic_file_buffered_read(struct kiocb
*iocb
,
2076 struct iov_iter
*iter
, ssize_t written
)
2078 struct file
*filp
= iocb
->ki_filp
;
2079 struct address_space
*mapping
= filp
->f_mapping
;
2080 struct inode
*inode
= mapping
->host
;
2081 struct file_ra_state
*ra
= &filp
->f_ra
;
2082 loff_t
*ppos
= &iocb
->ki_pos
;
2086 unsigned long offset
; /* offset into pagecache page */
2087 unsigned int prev_offset
;
2090 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
2092 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2094 index
= *ppos
>> PAGE_SHIFT
;
2095 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
2096 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
2097 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
2098 offset
= *ppos
& ~PAGE_MASK
;
2104 unsigned long nr
, ret
;
2108 if (fatal_signal_pending(current
)) {
2113 page
= find_get_page(mapping
, index
);
2115 if (iocb
->ki_flags
& IOCB_NOWAIT
)
2117 page_cache_sync_readahead(mapping
,
2119 index
, last_index
- index
);
2120 page
= find_get_page(mapping
, index
);
2121 if (unlikely(page
== NULL
))
2122 goto no_cached_page
;
2124 if (PageReadahead(page
)) {
2125 page_cache_async_readahead(mapping
,
2127 index
, last_index
- index
);
2129 if (!PageUptodate(page
)) {
2130 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2136 * See comment in do_read_cache_page on why
2137 * wait_on_page_locked is used to avoid unnecessarily
2138 * serialisations and why it's safe.
2140 error
= wait_on_page_locked_killable(page
);
2141 if (unlikely(error
))
2142 goto readpage_error
;
2143 if (PageUptodate(page
))
2146 if (inode
->i_blkbits
== PAGE_SHIFT
||
2147 !mapping
->a_ops
->is_partially_uptodate
)
2148 goto page_not_up_to_date
;
2149 /* pipes can't handle partially uptodate pages */
2150 if (unlikely(iter
->type
& ITER_PIPE
))
2151 goto page_not_up_to_date
;
2152 if (!trylock_page(page
))
2153 goto page_not_up_to_date
;
2154 /* Did it get truncated before we got the lock? */
2156 goto page_not_up_to_date_locked
;
2157 if (!mapping
->a_ops
->is_partially_uptodate(page
,
2158 offset
, iter
->count
))
2159 goto page_not_up_to_date_locked
;
2164 * i_size must be checked after we know the page is Uptodate.
2166 * Checking i_size after the check allows us to calculate
2167 * the correct value for "nr", which means the zero-filled
2168 * part of the page is not copied back to userspace (unless
2169 * another truncate extends the file - this is desired though).
2172 isize
= i_size_read(inode
);
2173 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2174 if (unlikely(!isize
|| index
> end_index
)) {
2179 /* nr is the maximum number of bytes to copy from this page */
2181 if (index
== end_index
) {
2182 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2190 /* If users can be writing to this page using arbitrary
2191 * virtual addresses, take care about potential aliasing
2192 * before reading the page on the kernel side.
2194 if (mapping_writably_mapped(mapping
))
2195 flush_dcache_page(page
);
2198 * When a sequential read accesses a page several times,
2199 * only mark it as accessed the first time.
2201 if (prev_index
!= index
|| offset
!= prev_offset
)
2202 mark_page_accessed(page
);
2206 * Ok, we have the page, and it's up-to-date, so
2207 * now we can copy it to user space...
2210 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
2212 index
+= offset
>> PAGE_SHIFT
;
2213 offset
&= ~PAGE_MASK
;
2214 prev_offset
= offset
;
2218 if (!iov_iter_count(iter
))
2226 page_not_up_to_date
:
2227 /* Get exclusive access to the page ... */
2228 error
= lock_page_killable(page
);
2229 if (unlikely(error
))
2230 goto readpage_error
;
2232 page_not_up_to_date_locked
:
2233 /* Did it get truncated before we got the lock? */
2234 if (!page
->mapping
) {
2240 /* Did somebody else fill it already? */
2241 if (PageUptodate(page
)) {
2248 * A previous I/O error may have been due to temporary
2249 * failures, eg. multipath errors.
2250 * PG_error will be set again if readpage fails.
2252 ClearPageError(page
);
2253 /* Start the actual read. The read will unlock the page. */
2254 error
= mapping
->a_ops
->readpage(filp
, page
);
2256 if (unlikely(error
)) {
2257 if (error
== AOP_TRUNCATED_PAGE
) {
2262 goto readpage_error
;
2265 if (!PageUptodate(page
)) {
2266 error
= lock_page_killable(page
);
2267 if (unlikely(error
))
2268 goto readpage_error
;
2269 if (!PageUptodate(page
)) {
2270 if (page
->mapping
== NULL
) {
2272 * invalidate_mapping_pages got it
2279 shrink_readahead_size_eio(filp
, ra
);
2281 goto readpage_error
;
2289 /* UHHUH! A synchronous read error occurred. Report it */
2295 * Ok, it wasn't cached, so we need to create a new
2298 page
= page_cache_alloc(mapping
);
2303 error
= add_to_page_cache_lru(page
, mapping
, index
,
2304 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2307 if (error
== -EEXIST
) {
2319 ra
->prev_pos
= prev_index
;
2320 ra
->prev_pos
<<= PAGE_SHIFT
;
2321 ra
->prev_pos
|= prev_offset
;
2323 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2324 file_accessed(filp
);
2325 return written
? written
: error
;
2329 * generic_file_read_iter - generic filesystem read routine
2330 * @iocb: kernel I/O control block
2331 * @iter: destination for the data read
2333 * This is the "read_iter()" routine for all filesystems
2334 * that can use the page cache directly.
2337 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2339 size_t count
= iov_iter_count(iter
);
2343 goto out
; /* skip atime */
2345 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2346 struct file
*file
= iocb
->ki_filp
;
2347 struct address_space
*mapping
= file
->f_mapping
;
2348 struct inode
*inode
= mapping
->host
;
2351 size
= i_size_read(inode
);
2352 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2353 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2354 iocb
->ki_pos
+ count
- 1))
2357 retval
= filemap_write_and_wait_range(mapping
,
2359 iocb
->ki_pos
+ count
- 1);
2364 file_accessed(file
);
2366 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2368 iocb
->ki_pos
+= retval
;
2371 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2374 * Btrfs can have a short DIO read if we encounter
2375 * compressed extents, so if there was an error, or if
2376 * we've already read everything we wanted to, or if
2377 * there was a short read because we hit EOF, go ahead
2378 * and return. Otherwise fallthrough to buffered io for
2379 * the rest of the read. Buffered reads will not work for
2380 * DAX files, so don't bother trying.
2382 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2387 retval
= generic_file_buffered_read(iocb
, iter
, retval
);
2391 EXPORT_SYMBOL(generic_file_read_iter
);
2395 * page_cache_read - adds requested page to the page cache if not already there
2396 * @file: file to read
2397 * @offset: page index
2398 * @gfp_mask: memory allocation flags
2400 * This adds the requested page to the page cache if it isn't already there,
2401 * and schedules an I/O to read in its contents from disk.
2403 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
2405 struct address_space
*mapping
= file
->f_mapping
;
2410 page
= __page_cache_alloc(gfp_mask
);
2414 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
2416 ret
= mapping
->a_ops
->readpage(file
, page
);
2417 else if (ret
== -EEXIST
)
2418 ret
= 0; /* losing race to add is OK */
2422 } while (ret
== AOP_TRUNCATED_PAGE
);
2427 #define MMAP_LOTSAMISS (100)
2430 * Synchronous readahead happens when we don't even find
2431 * a page in the page cache at all.
2433 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2434 struct file_ra_state
*ra
,
2438 struct address_space
*mapping
= file
->f_mapping
;
2440 /* If we don't want any read-ahead, don't bother */
2441 if (vma
->vm_flags
& VM_RAND_READ
)
2446 if (vma
->vm_flags
& VM_SEQ_READ
) {
2447 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2452 /* Avoid banging the cache line if not needed */
2453 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2457 * Do we miss much more than hit in this file? If so,
2458 * stop bothering with read-ahead. It will only hurt.
2460 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2466 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2467 ra
->size
= ra
->ra_pages
;
2468 ra
->async_size
= ra
->ra_pages
/ 4;
2469 ra_submit(ra
, mapping
, file
);
2473 * Asynchronous readahead happens when we find the page and PG_readahead,
2474 * so we want to possibly extend the readahead further..
2476 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2477 struct file_ra_state
*ra
,
2482 struct address_space
*mapping
= file
->f_mapping
;
2484 /* If we don't want any read-ahead, don't bother */
2485 if (vma
->vm_flags
& VM_RAND_READ
)
2487 if (ra
->mmap_miss
> 0)
2489 if (PageReadahead(page
))
2490 page_cache_async_readahead(mapping
, ra
, file
,
2491 page
, offset
, ra
->ra_pages
);
2495 * filemap_fault - read in file data for page fault handling
2496 * @vmf: struct vm_fault containing details of the fault
2498 * filemap_fault() is invoked via the vma operations vector for a
2499 * mapped memory region to read in file data during a page fault.
2501 * The goto's are kind of ugly, but this streamlines the normal case of having
2502 * it in the page cache, and handles the special cases reasonably without
2503 * having a lot of duplicated code.
2505 * vma->vm_mm->mmap_sem must be held on entry.
2507 * If our return value has VM_FAULT_RETRY set, it's because
2508 * lock_page_or_retry() returned 0.
2509 * The mmap_sem has usually been released in this case.
2510 * See __lock_page_or_retry() for the exception.
2512 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2513 * has not been released.
2515 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2517 int filemap_fault(struct vm_fault
*vmf
)
2520 struct file
*file
= vmf
->vma
->vm_file
;
2521 struct address_space
*mapping
= file
->f_mapping
;
2522 struct file_ra_state
*ra
= &file
->f_ra
;
2523 struct inode
*inode
= mapping
->host
;
2524 pgoff_t offset
= vmf
->pgoff
;
2529 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2530 if (unlikely(offset
>= max_off
))
2531 return VM_FAULT_SIGBUS
;
2534 * Do we have something in the page cache already?
2536 page
= find_get_page(mapping
, offset
);
2537 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2539 * We found the page, so try async readahead before
2540 * waiting for the lock.
2542 do_async_mmap_readahead(vmf
->vma
, ra
, file
, page
, offset
);
2544 /* No page in the page cache at all */
2545 do_sync_mmap_readahead(vmf
->vma
, ra
, file
, offset
);
2546 count_vm_event(PGMAJFAULT
);
2547 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2548 ret
= VM_FAULT_MAJOR
;
2550 page
= find_get_page(mapping
, offset
);
2552 goto no_cached_page
;
2555 if (!lock_page_or_retry(page
, vmf
->vma
->vm_mm
, vmf
->flags
)) {
2557 return ret
| VM_FAULT_RETRY
;
2560 /* Did it get truncated? */
2561 if (unlikely(page
->mapping
!= mapping
)) {
2566 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2569 * We have a locked page in the page cache, now we need to check
2570 * that it's up-to-date. If not, it is going to be due to an error.
2572 if (unlikely(!PageUptodate(page
)))
2573 goto page_not_uptodate
;
2576 * Found the page and have a reference on it.
2577 * We must recheck i_size under page lock.
2579 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2580 if (unlikely(offset
>= max_off
)) {
2583 return VM_FAULT_SIGBUS
;
2587 return ret
| VM_FAULT_LOCKED
;
2591 * We're only likely to ever get here if MADV_RANDOM is in
2594 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2597 * The page we want has now been added to the page cache.
2598 * In the unlikely event that someone removed it in the
2599 * meantime, we'll just come back here and read it again.
2605 * An error return from page_cache_read can result if the
2606 * system is low on memory, or a problem occurs while trying
2609 if (error
== -ENOMEM
)
2610 return VM_FAULT_OOM
;
2611 return VM_FAULT_SIGBUS
;
2615 * Umm, take care of errors if the page isn't up-to-date.
2616 * Try to re-read it _once_. We do this synchronously,
2617 * because there really aren't any performance issues here
2618 * and we need to check for errors.
2620 ClearPageError(page
);
2621 error
= mapping
->a_ops
->readpage(file
, page
);
2623 wait_on_page_locked(page
);
2624 if (!PageUptodate(page
))
2629 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2632 /* Things didn't work out. Return zero to tell the mm layer so. */
2633 shrink_readahead_size_eio(file
, ra
);
2634 return VM_FAULT_SIGBUS
;
2636 EXPORT_SYMBOL(filemap_fault
);
2638 void filemap_map_pages(struct vm_fault
*vmf
,
2639 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2641 struct radix_tree_iter iter
;
2643 struct file
*file
= vmf
->vma
->vm_file
;
2644 struct address_space
*mapping
= file
->f_mapping
;
2645 pgoff_t last_pgoff
= start_pgoff
;
2646 unsigned long max_idx
;
2647 struct page
*head
, *page
;
2650 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2652 if (iter
.index
> end_pgoff
)
2655 page
= radix_tree_deref_slot(slot
);
2656 if (unlikely(!page
))
2658 if (radix_tree_exception(page
)) {
2659 if (radix_tree_deref_retry(page
)) {
2660 slot
= radix_tree_iter_retry(&iter
);
2666 head
= compound_head(page
);
2667 if (!page_cache_get_speculative(head
))
2670 /* The page was split under us? */
2671 if (compound_head(page
) != head
) {
2676 /* Has the page moved? */
2677 if (unlikely(page
!= *slot
)) {
2682 if (!PageUptodate(page
) ||
2683 PageReadahead(page
) ||
2686 if (!trylock_page(page
))
2689 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2692 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2693 if (page
->index
>= max_idx
)
2696 if (file
->f_ra
.mmap_miss
> 0)
2697 file
->f_ra
.mmap_miss
--;
2699 vmf
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2701 vmf
->pte
+= iter
.index
- last_pgoff
;
2702 last_pgoff
= iter
.index
;
2703 if (alloc_set_pte(vmf
, NULL
, page
))
2712 /* Huge page is mapped? No need to proceed. */
2713 if (pmd_trans_huge(*vmf
->pmd
))
2715 if (iter
.index
== end_pgoff
)
2720 EXPORT_SYMBOL(filemap_map_pages
);
2722 int filemap_page_mkwrite(struct vm_fault
*vmf
)
2724 struct page
*page
= vmf
->page
;
2725 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2726 int ret
= VM_FAULT_LOCKED
;
2728 sb_start_pagefault(inode
->i_sb
);
2729 vma_file_update_time(vmf
->vma
);
2731 if (page
->mapping
!= inode
->i_mapping
) {
2733 ret
= VM_FAULT_NOPAGE
;
2737 * We mark the page dirty already here so that when freeze is in
2738 * progress, we are guaranteed that writeback during freezing will
2739 * see the dirty page and writeprotect it again.
2741 set_page_dirty(page
);
2742 wait_for_stable_page(page
);
2744 sb_end_pagefault(inode
->i_sb
);
2747 EXPORT_SYMBOL(filemap_page_mkwrite
);
2749 const struct vm_operations_struct generic_file_vm_ops
= {
2750 .fault
= filemap_fault
,
2751 .map_pages
= filemap_map_pages
,
2752 .page_mkwrite
= filemap_page_mkwrite
,
2755 /* This is used for a general mmap of a disk file */
2757 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2759 struct address_space
*mapping
= file
->f_mapping
;
2761 if (!mapping
->a_ops
->readpage
)
2763 file_accessed(file
);
2764 vma
->vm_ops
= &generic_file_vm_ops
;
2769 * This is for filesystems which do not implement ->writepage.
2771 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2773 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2775 return generic_file_mmap(file
, vma
);
2778 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2782 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2786 #endif /* CONFIG_MMU */
2788 EXPORT_SYMBOL(generic_file_mmap
);
2789 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2791 static struct page
*wait_on_page_read(struct page
*page
)
2793 if (!IS_ERR(page
)) {
2794 wait_on_page_locked(page
);
2795 if (!PageUptodate(page
)) {
2797 page
= ERR_PTR(-EIO
);
2803 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2805 int (*filler
)(void *, struct page
*),
2812 page
= find_get_page(mapping
, index
);
2814 page
= __page_cache_alloc(gfp
);
2816 return ERR_PTR(-ENOMEM
);
2817 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2818 if (unlikely(err
)) {
2822 /* Presumably ENOMEM for radix tree node */
2823 return ERR_PTR(err
);
2827 err
= filler(data
, page
);
2830 return ERR_PTR(err
);
2833 page
= wait_on_page_read(page
);
2838 if (PageUptodate(page
))
2842 * Page is not up to date and may be locked due one of the following
2843 * case a: Page is being filled and the page lock is held
2844 * case b: Read/write error clearing the page uptodate status
2845 * case c: Truncation in progress (page locked)
2846 * case d: Reclaim in progress
2848 * Case a, the page will be up to date when the page is unlocked.
2849 * There is no need to serialise on the page lock here as the page
2850 * is pinned so the lock gives no additional protection. Even if the
2851 * the page is truncated, the data is still valid if PageUptodate as
2852 * it's a race vs truncate race.
2853 * Case b, the page will not be up to date
2854 * Case c, the page may be truncated but in itself, the data may still
2855 * be valid after IO completes as it's a read vs truncate race. The
2856 * operation must restart if the page is not uptodate on unlock but
2857 * otherwise serialising on page lock to stabilise the mapping gives
2858 * no additional guarantees to the caller as the page lock is
2859 * released before return.
2860 * Case d, similar to truncation. If reclaim holds the page lock, it
2861 * will be a race with remove_mapping that determines if the mapping
2862 * is valid on unlock but otherwise the data is valid and there is
2863 * no need to serialise with page lock.
2865 * As the page lock gives no additional guarantee, we optimistically
2866 * wait on the page to be unlocked and check if it's up to date and
2867 * use the page if it is. Otherwise, the page lock is required to
2868 * distinguish between the different cases. The motivation is that we
2869 * avoid spurious serialisations and wakeups when multiple processes
2870 * wait on the same page for IO to complete.
2872 wait_on_page_locked(page
);
2873 if (PageUptodate(page
))
2876 /* Distinguish between all the cases under the safety of the lock */
2879 /* Case c or d, restart the operation */
2880 if (!page
->mapping
) {
2886 /* Someone else locked and filled the page in a very small window */
2887 if (PageUptodate(page
)) {
2894 mark_page_accessed(page
);
2899 * read_cache_page - read into page cache, fill it if needed
2900 * @mapping: the page's address_space
2901 * @index: the page index
2902 * @filler: function to perform the read
2903 * @data: first arg to filler(data, page) function, often left as NULL
2905 * Read into the page cache. If a page already exists, and PageUptodate() is
2906 * not set, try to fill the page and wait for it to become unlocked.
2908 * If the page does not get brought uptodate, return -EIO.
2910 struct page
*read_cache_page(struct address_space
*mapping
,
2912 int (*filler
)(void *, struct page
*),
2915 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2917 EXPORT_SYMBOL(read_cache_page
);
2920 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2921 * @mapping: the page's address_space
2922 * @index: the page index
2923 * @gfp: the page allocator flags to use if allocating
2925 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2926 * any new page allocations done using the specified allocation flags.
2928 * If the page does not get brought uptodate, return -EIO.
2930 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2934 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2936 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2938 EXPORT_SYMBOL(read_cache_page_gfp
);
2941 * Performs necessary checks before doing a write
2943 * Can adjust writing position or amount of bytes to write.
2944 * Returns appropriate error code that caller should return or
2945 * zero in case that write should be allowed.
2947 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2949 struct file
*file
= iocb
->ki_filp
;
2950 struct inode
*inode
= file
->f_mapping
->host
;
2951 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2954 if (!iov_iter_count(from
))
2957 /* FIXME: this is for backwards compatibility with 2.4 */
2958 if (iocb
->ki_flags
& IOCB_APPEND
)
2959 iocb
->ki_pos
= i_size_read(inode
);
2963 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
2966 if (limit
!= RLIM_INFINITY
) {
2967 if (iocb
->ki_pos
>= limit
) {
2968 send_sig(SIGXFSZ
, current
, 0);
2971 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2977 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2978 !(file
->f_flags
& O_LARGEFILE
))) {
2979 if (pos
>= MAX_NON_LFS
)
2981 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2985 * Are we about to exceed the fs block limit ?
2987 * If we have written data it becomes a short write. If we have
2988 * exceeded without writing data we send a signal and return EFBIG.
2989 * Linus frestrict idea will clean these up nicely..
2991 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2994 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2995 return iov_iter_count(from
);
2997 EXPORT_SYMBOL(generic_write_checks
);
2999 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
3000 loff_t pos
, unsigned len
, unsigned flags
,
3001 struct page
**pagep
, void **fsdata
)
3003 const struct address_space_operations
*aops
= mapping
->a_ops
;
3005 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
3008 EXPORT_SYMBOL(pagecache_write_begin
);
3010 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
3011 loff_t pos
, unsigned len
, unsigned copied
,
3012 struct page
*page
, void *fsdata
)
3014 const struct address_space_operations
*aops
= mapping
->a_ops
;
3016 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
3018 EXPORT_SYMBOL(pagecache_write_end
);
3021 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3023 struct file
*file
= iocb
->ki_filp
;
3024 struct address_space
*mapping
= file
->f_mapping
;
3025 struct inode
*inode
= mapping
->host
;
3026 loff_t pos
= iocb
->ki_pos
;
3031 write_len
= iov_iter_count(from
);
3032 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
3034 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3035 /* If there are pages to writeback, return */
3036 if (filemap_range_has_page(inode
->i_mapping
, pos
,
3037 pos
+ iov_iter_count(from
)))
3040 written
= filemap_write_and_wait_range(mapping
, pos
,
3041 pos
+ write_len
- 1);
3047 * After a write we want buffered reads to be sure to go to disk to get
3048 * the new data. We invalidate clean cached page from the region we're
3049 * about to write. We do this *before* the write so that we can return
3050 * without clobbering -EIOCBQUEUED from ->direct_IO().
3052 written
= invalidate_inode_pages2_range(mapping
,
3053 pos
>> PAGE_SHIFT
, end
);
3055 * If a page can not be invalidated, return 0 to fall back
3056 * to buffered write.
3059 if (written
== -EBUSY
)
3064 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3067 * Finally, try again to invalidate clean pages which might have been
3068 * cached by non-direct readahead, or faulted in by get_user_pages()
3069 * if the source of the write was an mmap'ed region of the file
3070 * we're writing. Either one is a pretty crazy thing to do,
3071 * so we don't support it 100%. If this invalidation
3072 * fails, tough, the write still worked...
3074 * Most of the time we do not need this since dio_complete() will do
3075 * the invalidation for us. However there are some file systems that
3076 * do not end up with dio_complete() being called, so let's not break
3077 * them by removing it completely
3079 if (mapping
->nrpages
)
3080 invalidate_inode_pages2_range(mapping
,
3081 pos
>> PAGE_SHIFT
, end
);
3085 write_len
-= written
;
3086 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3087 i_size_write(inode
, pos
);
3088 mark_inode_dirty(inode
);
3092 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3096 EXPORT_SYMBOL(generic_file_direct_write
);
3099 * Find or create a page at the given pagecache position. Return the locked
3100 * page. This function is specifically for buffered writes.
3102 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
3103 pgoff_t index
, unsigned flags
)
3106 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
3108 if (flags
& AOP_FLAG_NOFS
)
3109 fgp_flags
|= FGP_NOFS
;
3111 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
3112 mapping_gfp_mask(mapping
));
3114 wait_for_stable_page(page
);
3118 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3120 ssize_t
generic_perform_write(struct file
*file
,
3121 struct iov_iter
*i
, loff_t pos
)
3123 struct address_space
*mapping
= file
->f_mapping
;
3124 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3126 ssize_t written
= 0;
3127 unsigned int flags
= 0;
3131 unsigned long offset
; /* Offset into pagecache page */
3132 unsigned long bytes
; /* Bytes to write to page */
3133 size_t copied
; /* Bytes copied from user */
3136 offset
= (pos
& (PAGE_SIZE
- 1));
3137 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3142 * Bring in the user page that we will copy from _first_.
3143 * Otherwise there's a nasty deadlock on copying from the
3144 * same page as we're writing to, without it being marked
3147 * Not only is this an optimisation, but it is also required
3148 * to check that the address is actually valid, when atomic
3149 * usercopies are used, below.
3151 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3156 if (fatal_signal_pending(current
)) {
3161 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3163 if (unlikely(status
< 0))
3166 if (mapping_writably_mapped(mapping
))
3167 flush_dcache_page(page
);
3169 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3170 flush_dcache_page(page
);
3172 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3174 if (unlikely(status
< 0))
3180 iov_iter_advance(i
, copied
);
3181 if (unlikely(copied
== 0)) {
3183 * If we were unable to copy any data at all, we must
3184 * fall back to a single segment length write.
3186 * If we didn't fallback here, we could livelock
3187 * because not all segments in the iov can be copied at
3188 * once without a pagefault.
3190 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3191 iov_iter_single_seg_count(i
));
3197 balance_dirty_pages_ratelimited(mapping
);
3198 } while (iov_iter_count(i
));
3200 return written
? written
: status
;
3202 EXPORT_SYMBOL(generic_perform_write
);
3205 * __generic_file_write_iter - write data to a file
3206 * @iocb: IO state structure (file, offset, etc.)
3207 * @from: iov_iter with data to write
3209 * This function does all the work needed for actually writing data to a
3210 * file. It does all basic checks, removes SUID from the file, updates
3211 * modification times and calls proper subroutines depending on whether we
3212 * do direct IO or a standard buffered write.
3214 * It expects i_mutex to be grabbed unless we work on a block device or similar
3215 * object which does not need locking at all.
3217 * This function does *not* take care of syncing data in case of O_SYNC write.
3218 * A caller has to handle it. This is mainly due to the fact that we want to
3219 * avoid syncing under i_mutex.
3221 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3223 struct file
*file
= iocb
->ki_filp
;
3224 struct address_space
* mapping
= file
->f_mapping
;
3225 struct inode
*inode
= mapping
->host
;
3226 ssize_t written
= 0;
3230 /* We can write back this queue in page reclaim */
3231 current
->backing_dev_info
= inode_to_bdi(inode
);
3232 err
= file_remove_privs(file
);
3236 err
= file_update_time(file
);
3240 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3241 loff_t pos
, endbyte
;
3243 written
= generic_file_direct_write(iocb
, from
);
3245 * If the write stopped short of completing, fall back to
3246 * buffered writes. Some filesystems do this for writes to
3247 * holes, for example. For DAX files, a buffered write will
3248 * not succeed (even if it did, DAX does not handle dirty
3249 * page-cache pages correctly).
3251 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3254 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3256 * If generic_perform_write() returned a synchronous error
3257 * then we want to return the number of bytes which were
3258 * direct-written, or the error code if that was zero. Note
3259 * that this differs from normal direct-io semantics, which
3260 * will return -EFOO even if some bytes were written.
3262 if (unlikely(status
< 0)) {
3267 * We need to ensure that the page cache pages are written to
3268 * disk and invalidated to preserve the expected O_DIRECT
3271 endbyte
= pos
+ status
- 1;
3272 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3274 iocb
->ki_pos
= endbyte
+ 1;
3276 invalidate_mapping_pages(mapping
,
3278 endbyte
>> PAGE_SHIFT
);
3281 * We don't know how much we wrote, so just return
3282 * the number of bytes which were direct-written
3286 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3287 if (likely(written
> 0))
3288 iocb
->ki_pos
+= written
;
3291 current
->backing_dev_info
= NULL
;
3292 return written
? written
: err
;
3294 EXPORT_SYMBOL(__generic_file_write_iter
);
3297 * generic_file_write_iter - write data to a file
3298 * @iocb: IO state structure
3299 * @from: iov_iter with data to write
3301 * This is a wrapper around __generic_file_write_iter() to be used by most
3302 * filesystems. It takes care of syncing the file in case of O_SYNC file
3303 * and acquires i_mutex as needed.
3305 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3307 struct file
*file
= iocb
->ki_filp
;
3308 struct inode
*inode
= file
->f_mapping
->host
;
3312 ret
= generic_write_checks(iocb
, from
);
3314 ret
= __generic_file_write_iter(iocb
, from
);
3315 inode_unlock(inode
);
3318 ret
= generic_write_sync(iocb
, ret
);
3321 EXPORT_SYMBOL(generic_file_write_iter
);
3324 * try_to_release_page() - release old fs-specific metadata on a page
3326 * @page: the page which the kernel is trying to free
3327 * @gfp_mask: memory allocation flags (and I/O mode)
3329 * The address_space is to try to release any data against the page
3330 * (presumably at page->private). If the release was successful, return '1'.
3331 * Otherwise return zero.
3333 * This may also be called if PG_fscache is set on a page, indicating that the
3334 * page is known to the local caching routines.
3336 * The @gfp_mask argument specifies whether I/O may be performed to release
3337 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3340 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3342 struct address_space
* const mapping
= page
->mapping
;
3344 BUG_ON(!PageLocked(page
));
3345 if (PageWriteback(page
))
3348 if (mapping
&& mapping
->a_ops
->releasepage
)
3349 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3350 return try_to_free_buffers(page
);
3353 EXPORT_SYMBOL(try_to_release_page
);