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
))
444 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
445 ret
= do_writepages(mapping
, &wbc
);
446 wbc_detach_inode(&wbc
);
450 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
453 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
456 int filemap_fdatawrite(struct address_space
*mapping
)
458 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
460 EXPORT_SYMBOL(filemap_fdatawrite
);
462 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
465 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
467 EXPORT_SYMBOL(filemap_fdatawrite_range
);
470 * filemap_flush - mostly a non-blocking flush
471 * @mapping: target address_space
473 * This is a mostly non-blocking flush. Not suitable for data-integrity
474 * purposes - I/O may not be started against all dirty pages.
476 int filemap_flush(struct address_space
*mapping
)
478 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
480 EXPORT_SYMBOL(filemap_flush
);
483 * filemap_range_has_page - check if a page exists in range.
484 * @mapping: address space within which to check
485 * @start_byte: offset in bytes where the range starts
486 * @end_byte: offset in bytes where the range ends (inclusive)
488 * Find at least one page in the range supplied, usually used to check if
489 * direct writing in this range will trigger a writeback.
491 bool filemap_range_has_page(struct address_space
*mapping
,
492 loff_t start_byte
, loff_t end_byte
)
494 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
495 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
498 if (end_byte
< start_byte
)
501 if (mapping
->nrpages
== 0)
504 if (!find_get_pages_range(mapping
, &index
, end
, 1, &page
))
509 EXPORT_SYMBOL(filemap_range_has_page
);
511 static void __filemap_fdatawait_range(struct address_space
*mapping
,
512 loff_t start_byte
, loff_t end_byte
)
514 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
515 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
519 if (end_byte
< start_byte
)
523 while (index
<= end
) {
526 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
,
527 end
, PAGECACHE_TAG_WRITEBACK
);
531 for (i
= 0; i
< nr_pages
; i
++) {
532 struct page
*page
= pvec
.pages
[i
];
534 wait_on_page_writeback(page
);
535 ClearPageError(page
);
537 pagevec_release(&pvec
);
543 * filemap_fdatawait_range - wait for writeback to complete
544 * @mapping: address space structure to wait for
545 * @start_byte: offset in bytes where the range starts
546 * @end_byte: offset in bytes where the range ends (inclusive)
548 * Walk the list of under-writeback pages of the given address space
549 * in the given range and wait for all of them. Check error status of
550 * the address space and return it.
552 * Since the error status of the address space is cleared by this function,
553 * callers are responsible for checking the return value and handling and/or
554 * reporting the error.
556 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
559 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
560 return filemap_check_errors(mapping
);
562 EXPORT_SYMBOL(filemap_fdatawait_range
);
565 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
566 * @mapping: address space structure to wait for
567 * @start_byte: offset in bytes where the range starts
568 * @end_byte: offset in bytes where the range ends (inclusive)
570 * Walk the list of under-writeback pages of the given address space in the
571 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
572 * this function does not clear error status of the address space.
574 * Use this function if callers don't handle errors themselves. Expected
575 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
578 int filemap_fdatawait_range_keep_errors(struct address_space
*mapping
,
579 loff_t start_byte
, loff_t end_byte
)
581 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
582 return filemap_check_and_keep_errors(mapping
);
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors
);
587 * file_fdatawait_range - wait for writeback to complete
588 * @file: file pointing to address space structure to wait for
589 * @start_byte: offset in bytes where the range starts
590 * @end_byte: offset in bytes where the range ends (inclusive)
592 * Walk the list of under-writeback pages of the address space that file
593 * refers to, in the given range and wait for all of them. Check error
594 * status of the address space vs. the file->f_wb_err cursor and return it.
596 * Since the error status of the file is advanced by this function,
597 * callers are responsible for checking the return value and handling and/or
598 * reporting the error.
600 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
602 struct address_space
*mapping
= file
->f_mapping
;
604 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
605 return file_check_and_advance_wb_err(file
);
607 EXPORT_SYMBOL(file_fdatawait_range
);
610 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
611 * @mapping: address space structure to wait for
613 * Walk the list of under-writeback pages of the given address space
614 * and wait for all of them. Unlike filemap_fdatawait(), this function
615 * does not clear error status of the address space.
617 * Use this function if callers don't handle errors themselves. Expected
618 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
621 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
623 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
624 return filemap_check_and_keep_errors(mapping
);
626 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
628 static bool mapping_needs_writeback(struct address_space
*mapping
)
630 return (!dax_mapping(mapping
) && mapping
->nrpages
) ||
631 (dax_mapping(mapping
) && mapping
->nrexceptional
);
634 int filemap_write_and_wait(struct address_space
*mapping
)
638 if (mapping_needs_writeback(mapping
)) {
639 err
= filemap_fdatawrite(mapping
);
641 * Even if the above returned error, the pages may be
642 * written partially (e.g. -ENOSPC), so we wait for it.
643 * But the -EIO is special case, it may indicate the worst
644 * thing (e.g. bug) happened, so we avoid waiting for it.
647 int err2
= filemap_fdatawait(mapping
);
651 /* Clear any previously stored errors */
652 filemap_check_errors(mapping
);
655 err
= filemap_check_errors(mapping
);
659 EXPORT_SYMBOL(filemap_write_and_wait
);
662 * filemap_write_and_wait_range - write out & wait on a file range
663 * @mapping: the address_space for the pages
664 * @lstart: offset in bytes where the range starts
665 * @lend: offset in bytes where the range ends (inclusive)
667 * Write out and wait upon file offsets lstart->lend, inclusive.
669 * Note that @lend is inclusive (describes the last byte to be written) so
670 * that this function can be used to write to the very end-of-file (end = -1).
672 int filemap_write_and_wait_range(struct address_space
*mapping
,
673 loff_t lstart
, loff_t lend
)
677 if (mapping_needs_writeback(mapping
)) {
678 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
680 /* See comment of filemap_write_and_wait() */
682 int err2
= filemap_fdatawait_range(mapping
,
687 /* Clear any previously stored errors */
688 filemap_check_errors(mapping
);
691 err
= filemap_check_errors(mapping
);
695 EXPORT_SYMBOL(filemap_write_and_wait_range
);
697 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
699 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
701 trace_filemap_set_wb_err(mapping
, eseq
);
703 EXPORT_SYMBOL(__filemap_set_wb_err
);
706 * file_check_and_advance_wb_err - report wb error (if any) that was previously
707 * and advance wb_err to current one
708 * @file: struct file on which the error is being reported
710 * When userland calls fsync (or something like nfsd does the equivalent), we
711 * want to report any writeback errors that occurred since the last fsync (or
712 * since the file was opened if there haven't been any).
714 * Grab the wb_err from the mapping. If it matches what we have in the file,
715 * then just quickly return 0. The file is all caught up.
717 * If it doesn't match, then take the mapping value, set the "seen" flag in
718 * it and try to swap it into place. If it works, or another task beat us
719 * to it with the new value, then update the f_wb_err and return the error
720 * portion. The error at this point must be reported via proper channels
721 * (a'la fsync, or NFS COMMIT operation, etc.).
723 * While we handle mapping->wb_err with atomic operations, the f_wb_err
724 * value is protected by the f_lock since we must ensure that it reflects
725 * the latest value swapped in for this file descriptor.
727 int file_check_and_advance_wb_err(struct file
*file
)
730 errseq_t old
= READ_ONCE(file
->f_wb_err
);
731 struct address_space
*mapping
= file
->f_mapping
;
733 /* Locklessly handle the common case where nothing has changed */
734 if (errseq_check(&mapping
->wb_err
, old
)) {
735 /* Something changed, must use slow path */
736 spin_lock(&file
->f_lock
);
737 old
= file
->f_wb_err
;
738 err
= errseq_check_and_advance(&mapping
->wb_err
,
740 trace_file_check_and_advance_wb_err(file
, old
);
741 spin_unlock(&file
->f_lock
);
745 * We're mostly using this function as a drop in replacement for
746 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
747 * that the legacy code would have had on these flags.
749 clear_bit(AS_EIO
, &mapping
->flags
);
750 clear_bit(AS_ENOSPC
, &mapping
->flags
);
753 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
756 * file_write_and_wait_range - write out & wait on a file range
757 * @file: file pointing to address_space with pages
758 * @lstart: offset in bytes where the range starts
759 * @lend: offset in bytes where the range ends (inclusive)
761 * Write out and wait upon file offsets lstart->lend, inclusive.
763 * Note that @lend is inclusive (describes the last byte to be written) so
764 * that this function can be used to write to the very end-of-file (end = -1).
766 * After writing out and waiting on the data, we check and advance the
767 * f_wb_err cursor to the latest value, and return any errors detected there.
769 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
772 struct address_space
*mapping
= file
->f_mapping
;
774 if (mapping_needs_writeback(mapping
)) {
775 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
777 /* See comment of filemap_write_and_wait() */
779 __filemap_fdatawait_range(mapping
, lstart
, lend
);
781 err2
= file_check_and_advance_wb_err(file
);
786 EXPORT_SYMBOL(file_write_and_wait_range
);
789 * replace_page_cache_page - replace a pagecache page with a new one
790 * @old: page to be replaced
791 * @new: page to replace with
792 * @gfp_mask: allocation mode
794 * This function replaces a page in the pagecache with a new one. On
795 * success it acquires the pagecache reference for the new page and
796 * drops it for the old page. Both the old and new pages must be
797 * locked. This function does not add the new page to the LRU, the
798 * caller must do that.
800 * The remove + add is atomic. The only way this function can fail is
801 * memory allocation failure.
803 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
807 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
808 VM_BUG_ON_PAGE(!PageLocked(new), new);
809 VM_BUG_ON_PAGE(new->mapping
, new);
811 error
= radix_tree_preload(gfp_mask
& GFP_RECLAIM_MASK
);
813 struct address_space
*mapping
= old
->mapping
;
814 void (*freepage
)(struct page
*);
817 pgoff_t offset
= old
->index
;
818 freepage
= mapping
->a_ops
->freepage
;
821 new->mapping
= mapping
;
824 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
825 __delete_from_page_cache(old
, NULL
);
826 error
= page_cache_tree_insert(mapping
, new, NULL
);
830 * hugetlb pages do not participate in page cache accounting.
833 __inc_node_page_state(new, NR_FILE_PAGES
);
834 if (PageSwapBacked(new))
835 __inc_node_page_state(new, NR_SHMEM
);
836 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
837 mem_cgroup_migrate(old
, new);
838 radix_tree_preload_end();
846 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
848 static int __add_to_page_cache_locked(struct page
*page
,
849 struct address_space
*mapping
,
850 pgoff_t offset
, gfp_t gfp_mask
,
853 int huge
= PageHuge(page
);
854 struct mem_cgroup
*memcg
;
857 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
858 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
861 error
= mem_cgroup_try_charge(page
, current
->mm
,
862 gfp_mask
, &memcg
, false);
867 error
= radix_tree_maybe_preload(gfp_mask
& GFP_RECLAIM_MASK
);
870 mem_cgroup_cancel_charge(page
, memcg
, false);
875 page
->mapping
= mapping
;
876 page
->index
= offset
;
878 spin_lock_irq(&mapping
->tree_lock
);
879 error
= page_cache_tree_insert(mapping
, page
, shadowp
);
880 radix_tree_preload_end();
884 /* hugetlb pages do not participate in page cache accounting. */
886 __inc_node_page_state(page
, NR_FILE_PAGES
);
887 spin_unlock_irq(&mapping
->tree_lock
);
889 mem_cgroup_commit_charge(page
, memcg
, false, false);
890 trace_mm_filemap_add_to_page_cache(page
);
893 page
->mapping
= NULL
;
894 /* Leave page->index set: truncation relies upon it */
895 spin_unlock_irq(&mapping
->tree_lock
);
897 mem_cgroup_cancel_charge(page
, memcg
, false);
903 * add_to_page_cache_locked - add a locked page to the pagecache
905 * @mapping: the page's address_space
906 * @offset: page index
907 * @gfp_mask: page allocation mode
909 * This function is used to add a page to the pagecache. It must be locked.
910 * This function does not add the page to the LRU. The caller must do that.
912 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
913 pgoff_t offset
, gfp_t gfp_mask
)
915 return __add_to_page_cache_locked(page
, mapping
, offset
,
918 EXPORT_SYMBOL(add_to_page_cache_locked
);
920 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
921 pgoff_t offset
, gfp_t gfp_mask
)
926 __SetPageLocked(page
);
927 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
930 __ClearPageLocked(page
);
933 * The page might have been evicted from cache only
934 * recently, in which case it should be activated like
935 * any other repeatedly accessed page.
936 * The exception is pages getting rewritten; evicting other
937 * data from the working set, only to cache data that will
938 * get overwritten with something else, is a waste of memory.
940 if (!(gfp_mask
& __GFP_WRITE
) &&
941 shadow
&& workingset_refault(shadow
)) {
943 workingset_activation(page
);
945 ClearPageActive(page
);
950 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
953 struct page
*__page_cache_alloc(gfp_t gfp
)
958 if (cpuset_do_page_mem_spread()) {
959 unsigned int cpuset_mems_cookie
;
961 cpuset_mems_cookie
= read_mems_allowed_begin();
962 n
= cpuset_mem_spread_node();
963 page
= __alloc_pages_node(n
, gfp
, 0);
964 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
968 return alloc_pages(gfp
, 0);
970 EXPORT_SYMBOL(__page_cache_alloc
);
974 * In order to wait for pages to become available there must be
975 * waitqueues associated with pages. By using a hash table of
976 * waitqueues where the bucket discipline is to maintain all
977 * waiters on the same queue and wake all when any of the pages
978 * become available, and for the woken contexts to check to be
979 * sure the appropriate page became available, this saves space
980 * at a cost of "thundering herd" phenomena during rare hash
983 #define PAGE_WAIT_TABLE_BITS 8
984 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
985 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
987 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
989 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
992 void __init
pagecache_init(void)
996 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
997 init_waitqueue_head(&page_wait_table
[i
]);
999 page_writeback_init();
1002 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1003 struct wait_page_key
{
1009 struct wait_page_queue
{
1012 wait_queue_entry_t wait
;
1015 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
1017 struct wait_page_key
*key
= arg
;
1018 struct wait_page_queue
*wait_page
1019 = container_of(wait
, struct wait_page_queue
, wait
);
1021 if (wait_page
->page
!= key
->page
)
1023 key
->page_match
= 1;
1025 if (wait_page
->bit_nr
!= key
->bit_nr
)
1028 /* Stop walking if it's locked */
1029 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
1032 return autoremove_wake_function(wait
, mode
, sync
, key
);
1035 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
1037 wait_queue_head_t
*q
= page_waitqueue(page
);
1038 struct wait_page_key key
;
1039 unsigned long flags
;
1040 wait_queue_entry_t bookmark
;
1043 key
.bit_nr
= bit_nr
;
1047 bookmark
.private = NULL
;
1048 bookmark
.func
= NULL
;
1049 INIT_LIST_HEAD(&bookmark
.entry
);
1051 spin_lock_irqsave(&q
->lock
, flags
);
1052 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1054 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1056 * Take a breather from holding the lock,
1057 * allow pages that finish wake up asynchronously
1058 * to acquire the lock and remove themselves
1061 spin_unlock_irqrestore(&q
->lock
, flags
);
1063 spin_lock_irqsave(&q
->lock
, flags
);
1064 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1068 * It is possible for other pages to have collided on the waitqueue
1069 * hash, so in that case check for a page match. That prevents a long-
1072 * It is still possible to miss a case here, when we woke page waiters
1073 * and removed them from the waitqueue, but there are still other
1076 if (!waitqueue_active(q
) || !key
.page_match
) {
1077 ClearPageWaiters(page
);
1079 * It's possible to miss clearing Waiters here, when we woke
1080 * our page waiters, but the hashed waitqueue has waiters for
1081 * other pages on it.
1083 * That's okay, it's a rare case. The next waker will clear it.
1086 spin_unlock_irqrestore(&q
->lock
, flags
);
1089 static void wake_up_page(struct page
*page
, int bit
)
1091 if (!PageWaiters(page
))
1093 wake_up_page_bit(page
, bit
);
1096 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
1097 struct page
*page
, int bit_nr
, int state
, bool lock
)
1099 struct wait_page_queue wait_page
;
1100 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1104 wait
->flags
= lock
? WQ_FLAG_EXCLUSIVE
: 0;
1105 wait
->func
= wake_page_function
;
1106 wait_page
.page
= page
;
1107 wait_page
.bit_nr
= bit_nr
;
1110 spin_lock_irq(&q
->lock
);
1112 if (likely(list_empty(&wait
->entry
))) {
1113 __add_wait_queue_entry_tail(q
, wait
);
1114 SetPageWaiters(page
);
1117 set_current_state(state
);
1119 spin_unlock_irq(&q
->lock
);
1121 if (likely(test_bit(bit_nr
, &page
->flags
))) {
1126 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
1129 if (!test_bit(bit_nr
, &page
->flags
))
1133 if (unlikely(signal_pending_state(state
, current
))) {
1139 finish_wait(q
, wait
);
1142 * A signal could leave PageWaiters set. Clearing it here if
1143 * !waitqueue_active would be possible (by open-coding finish_wait),
1144 * but still fail to catch it in the case of wait hash collision. We
1145 * already can fail to clear wait hash collision cases, so don't
1146 * bother with signals either.
1152 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1154 wait_queue_head_t
*q
= page_waitqueue(page
);
1155 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, false);
1157 EXPORT_SYMBOL(wait_on_page_bit
);
1159 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1161 wait_queue_head_t
*q
= page_waitqueue(page
);
1162 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, false);
1164 EXPORT_SYMBOL(wait_on_page_bit_killable
);
1167 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1168 * @page: Page defining the wait queue of interest
1169 * @waiter: Waiter to add to the queue
1171 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1173 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1175 wait_queue_head_t
*q
= page_waitqueue(page
);
1176 unsigned long flags
;
1178 spin_lock_irqsave(&q
->lock
, flags
);
1179 __add_wait_queue_entry_tail(q
, waiter
);
1180 SetPageWaiters(page
);
1181 spin_unlock_irqrestore(&q
->lock
, flags
);
1183 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1185 #ifndef clear_bit_unlock_is_negative_byte
1188 * PG_waiters is the high bit in the same byte as PG_lock.
1190 * On x86 (and on many other architectures), we can clear PG_lock and
1191 * test the sign bit at the same time. But if the architecture does
1192 * not support that special operation, we just do this all by hand
1195 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1196 * being cleared, but a memory barrier should be unneccssary since it is
1197 * in the same byte as PG_locked.
1199 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1201 clear_bit_unlock(nr
, mem
);
1202 /* smp_mb__after_atomic(); */
1203 return test_bit(PG_waiters
, mem
);
1209 * unlock_page - unlock a locked page
1212 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1213 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1214 * mechanism between PageLocked pages and PageWriteback pages is shared.
1215 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1217 * Note that this depends on PG_waiters being the sign bit in the byte
1218 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1219 * clear the PG_locked bit and test PG_waiters at the same time fairly
1220 * portably (architectures that do LL/SC can test any bit, while x86 can
1221 * test the sign bit).
1223 void unlock_page(struct page
*page
)
1225 BUILD_BUG_ON(PG_waiters
!= 7);
1226 page
= compound_head(page
);
1227 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1228 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1229 wake_up_page_bit(page
, PG_locked
);
1231 EXPORT_SYMBOL(unlock_page
);
1234 * end_page_writeback - end writeback against a page
1237 void end_page_writeback(struct page
*page
)
1240 * TestClearPageReclaim could be used here but it is an atomic
1241 * operation and overkill in this particular case. Failing to
1242 * shuffle a page marked for immediate reclaim is too mild to
1243 * justify taking an atomic operation penalty at the end of
1244 * ever page writeback.
1246 if (PageReclaim(page
)) {
1247 ClearPageReclaim(page
);
1248 rotate_reclaimable_page(page
);
1251 if (!test_clear_page_writeback(page
))
1254 smp_mb__after_atomic();
1255 wake_up_page(page
, PG_writeback
);
1257 EXPORT_SYMBOL(end_page_writeback
);
1260 * After completing I/O on a page, call this routine to update the page
1261 * flags appropriately
1263 void page_endio(struct page
*page
, bool is_write
, int err
)
1267 SetPageUptodate(page
);
1269 ClearPageUptodate(page
);
1275 struct address_space
*mapping
;
1278 mapping
= page_mapping(page
);
1280 mapping_set_error(mapping
, err
);
1282 end_page_writeback(page
);
1285 EXPORT_SYMBOL_GPL(page_endio
);
1288 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1289 * @__page: the page to lock
1291 void __lock_page(struct page
*__page
)
1293 struct page
*page
= compound_head(__page
);
1294 wait_queue_head_t
*q
= page_waitqueue(page
);
1295 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, true);
1297 EXPORT_SYMBOL(__lock_page
);
1299 int __lock_page_killable(struct page
*__page
)
1301 struct page
*page
= compound_head(__page
);
1302 wait_queue_head_t
*q
= page_waitqueue(page
);
1303 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
, true);
1305 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1309 * 1 - page is locked; mmap_sem is still held.
1310 * 0 - page is not locked.
1311 * mmap_sem has been released (up_read()), unless flags had both
1312 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1313 * which case mmap_sem is still held.
1315 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1316 * with the page locked and the mmap_sem unperturbed.
1318 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1321 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1323 * CAUTION! In this case, mmap_sem is not released
1324 * even though return 0.
1326 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1329 up_read(&mm
->mmap_sem
);
1330 if (flags
& FAULT_FLAG_KILLABLE
)
1331 wait_on_page_locked_killable(page
);
1333 wait_on_page_locked(page
);
1336 if (flags
& FAULT_FLAG_KILLABLE
) {
1339 ret
= __lock_page_killable(page
);
1341 up_read(&mm
->mmap_sem
);
1351 * page_cache_next_hole - find the next hole (not-present entry)
1354 * @max_scan: maximum range to search
1356 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1357 * lowest indexed hole.
1359 * Returns: the index of the hole if found, otherwise returns an index
1360 * outside of the set specified (in which case 'return - index >=
1361 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1364 * page_cache_next_hole may be called under rcu_read_lock. However,
1365 * like radix_tree_gang_lookup, this will not atomically search a
1366 * snapshot of the tree at a single point in time. For example, if a
1367 * hole is created at index 5, then subsequently a hole is created at
1368 * index 10, page_cache_next_hole covering both indexes may return 10
1369 * if called under rcu_read_lock.
1371 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
1372 pgoff_t index
, unsigned long max_scan
)
1376 for (i
= 0; i
< max_scan
; i
++) {
1379 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1380 if (!page
|| radix_tree_exceptional_entry(page
))
1389 EXPORT_SYMBOL(page_cache_next_hole
);
1392 * page_cache_prev_hole - find the prev hole (not-present entry)
1395 * @max_scan: maximum range to search
1397 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1400 * Returns: the index of the hole if found, otherwise returns an index
1401 * outside of the set specified (in which case 'index - return >=
1402 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1405 * page_cache_prev_hole may be called under rcu_read_lock. However,
1406 * like radix_tree_gang_lookup, this will not atomically search a
1407 * snapshot of the tree at a single point in time. For example, if a
1408 * hole is created at index 10, then subsequently a hole is created at
1409 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1410 * called under rcu_read_lock.
1412 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
1413 pgoff_t index
, unsigned long max_scan
)
1417 for (i
= 0; i
< max_scan
; i
++) {
1420 page
= radix_tree_lookup(&mapping
->page_tree
, index
);
1421 if (!page
|| radix_tree_exceptional_entry(page
))
1424 if (index
== ULONG_MAX
)
1430 EXPORT_SYMBOL(page_cache_prev_hole
);
1433 * find_get_entry - find and get a page cache entry
1434 * @mapping: the address_space to search
1435 * @offset: the page cache index
1437 * Looks up the page cache slot at @mapping & @offset. If there is a
1438 * page cache page, it is returned with an increased refcount.
1440 * If the slot holds a shadow entry of a previously evicted page, or a
1441 * swap entry from shmem/tmpfs, it is returned.
1443 * Otherwise, %NULL is returned.
1445 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1448 struct page
*head
, *page
;
1453 pagep
= radix_tree_lookup_slot(&mapping
->page_tree
, offset
);
1455 page
= radix_tree_deref_slot(pagep
);
1456 if (unlikely(!page
))
1458 if (radix_tree_exception(page
)) {
1459 if (radix_tree_deref_retry(page
))
1462 * A shadow entry of a recently evicted page,
1463 * or a swap entry from shmem/tmpfs. Return
1464 * it without attempting to raise page count.
1469 head
= compound_head(page
);
1470 if (!page_cache_get_speculative(head
))
1473 /* The page was split under us? */
1474 if (compound_head(page
) != head
) {
1480 * Has the page moved?
1481 * This is part of the lockless pagecache protocol. See
1482 * include/linux/pagemap.h for details.
1484 if (unlikely(page
!= *pagep
)) {
1494 EXPORT_SYMBOL(find_get_entry
);
1497 * find_lock_entry - locate, pin and lock a page cache entry
1498 * @mapping: the address_space to search
1499 * @offset: the page cache index
1501 * Looks up the page cache slot at @mapping & @offset. If there is a
1502 * page cache page, it is returned locked and with an increased
1505 * If the slot holds a shadow entry of a previously evicted page, or a
1506 * swap entry from shmem/tmpfs, it is returned.
1508 * Otherwise, %NULL is returned.
1510 * find_lock_entry() may sleep.
1512 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1517 page
= find_get_entry(mapping
, offset
);
1518 if (page
&& !radix_tree_exception(page
)) {
1520 /* Has the page been truncated? */
1521 if (unlikely(page_mapping(page
) != mapping
)) {
1526 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1530 EXPORT_SYMBOL(find_lock_entry
);
1533 * pagecache_get_page - find and get a page reference
1534 * @mapping: the address_space to search
1535 * @offset: the page index
1536 * @fgp_flags: PCG flags
1537 * @gfp_mask: gfp mask to use for the page cache data page allocation
1539 * Looks up the page cache slot at @mapping & @offset.
1541 * PCG flags modify how the page is returned.
1543 * @fgp_flags can be:
1545 * - FGP_ACCESSED: the page will be marked accessed
1546 * - FGP_LOCK: Page is return locked
1547 * - FGP_CREAT: If page is not present then a new page is allocated using
1548 * @gfp_mask and added to the page cache and the VM's LRU
1549 * list. The page is returned locked and with an increased
1550 * refcount. Otherwise, NULL is returned.
1552 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1553 * if the GFP flags specified for FGP_CREAT are atomic.
1555 * If there is a page cache page, it is returned with an increased refcount.
1557 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1558 int fgp_flags
, gfp_t gfp_mask
)
1563 page
= find_get_entry(mapping
, offset
);
1564 if (radix_tree_exceptional_entry(page
))
1569 if (fgp_flags
& FGP_LOCK
) {
1570 if (fgp_flags
& FGP_NOWAIT
) {
1571 if (!trylock_page(page
)) {
1579 /* Has the page been truncated? */
1580 if (unlikely(page
->mapping
!= mapping
)) {
1585 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1588 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1589 mark_page_accessed(page
);
1592 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1594 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1595 gfp_mask
|= __GFP_WRITE
;
1596 if (fgp_flags
& FGP_NOFS
)
1597 gfp_mask
&= ~__GFP_FS
;
1599 page
= __page_cache_alloc(gfp_mask
);
1603 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1604 fgp_flags
|= FGP_LOCK
;
1606 /* Init accessed so avoid atomic mark_page_accessed later */
1607 if (fgp_flags
& FGP_ACCESSED
)
1608 __SetPageReferenced(page
);
1610 err
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
1611 if (unlikely(err
)) {
1621 EXPORT_SYMBOL(pagecache_get_page
);
1624 * find_get_entries - gang pagecache lookup
1625 * @mapping: The address_space to search
1626 * @start: The starting page cache index
1627 * @nr_entries: The maximum number of entries
1628 * @entries: Where the resulting entries are placed
1629 * @indices: The cache indices corresponding to the entries in @entries
1631 * find_get_entries() will search for and return a group of up to
1632 * @nr_entries entries in the mapping. The entries are placed at
1633 * @entries. find_get_entries() takes a reference against any actual
1636 * The search returns a group of mapping-contiguous page cache entries
1637 * with ascending indexes. There may be holes in the indices due to
1638 * not-present pages.
1640 * Any shadow entries of evicted pages, or swap entries from
1641 * shmem/tmpfs, are included in the returned array.
1643 * find_get_entries() returns the number of pages and shadow entries
1646 unsigned find_get_entries(struct address_space
*mapping
,
1647 pgoff_t start
, unsigned int nr_entries
,
1648 struct page
**entries
, pgoff_t
*indices
)
1651 unsigned int ret
= 0;
1652 struct radix_tree_iter iter
;
1658 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1659 struct page
*head
, *page
;
1661 page
= radix_tree_deref_slot(slot
);
1662 if (unlikely(!page
))
1664 if (radix_tree_exception(page
)) {
1665 if (radix_tree_deref_retry(page
)) {
1666 slot
= radix_tree_iter_retry(&iter
);
1670 * A shadow entry of a recently evicted page, a swap
1671 * entry from shmem/tmpfs or a DAX entry. Return it
1672 * without attempting to raise page count.
1677 head
= compound_head(page
);
1678 if (!page_cache_get_speculative(head
))
1681 /* The page was split under us? */
1682 if (compound_head(page
) != head
) {
1687 /* Has the page moved? */
1688 if (unlikely(page
!= *slot
)) {
1693 indices
[ret
] = iter
.index
;
1694 entries
[ret
] = page
;
1695 if (++ret
== nr_entries
)
1703 * find_get_pages_range - gang pagecache lookup
1704 * @mapping: The address_space to search
1705 * @start: The starting page index
1706 * @end: The final page index (inclusive)
1707 * @nr_pages: The maximum number of pages
1708 * @pages: Where the resulting pages are placed
1710 * find_get_pages_range() will search for and return a group of up to @nr_pages
1711 * pages in the mapping starting at index @start and up to index @end
1712 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1713 * a reference against the returned pages.
1715 * The search returns a group of mapping-contiguous pages with ascending
1716 * indexes. There may be holes in the indices due to not-present pages.
1717 * We also update @start to index the next page for the traversal.
1719 * find_get_pages_range() returns the number of pages which were found. If this
1720 * number is smaller than @nr_pages, the end of specified range has been
1723 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
1724 pgoff_t end
, unsigned int nr_pages
,
1725 struct page
**pages
)
1727 struct radix_tree_iter iter
;
1731 if (unlikely(!nr_pages
))
1735 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, *start
) {
1736 struct page
*head
, *page
;
1738 if (iter
.index
> end
)
1741 page
= radix_tree_deref_slot(slot
);
1742 if (unlikely(!page
))
1745 if (radix_tree_exception(page
)) {
1746 if (radix_tree_deref_retry(page
)) {
1747 slot
= radix_tree_iter_retry(&iter
);
1751 * A shadow entry of a recently evicted page,
1752 * or a swap entry from shmem/tmpfs. Skip
1758 head
= compound_head(page
);
1759 if (!page_cache_get_speculative(head
))
1762 /* The page was split under us? */
1763 if (compound_head(page
) != head
) {
1768 /* Has the page moved? */
1769 if (unlikely(page
!= *slot
)) {
1775 if (++ret
== nr_pages
) {
1776 *start
= pages
[ret
- 1]->index
+ 1;
1782 * We come here when there is no page beyond @end. We take care to not
1783 * overflow the index @start as it confuses some of the callers. This
1784 * breaks the iteration when there is page at index -1 but that is
1785 * already broken anyway.
1787 if (end
== (pgoff_t
)-1)
1788 *start
= (pgoff_t
)-1;
1798 * find_get_pages_contig - gang contiguous pagecache lookup
1799 * @mapping: The address_space to search
1800 * @index: The starting page index
1801 * @nr_pages: The maximum number of pages
1802 * @pages: Where the resulting pages are placed
1804 * find_get_pages_contig() works exactly like find_get_pages(), except
1805 * that the returned number of pages are guaranteed to be contiguous.
1807 * find_get_pages_contig() returns the number of pages which were found.
1809 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1810 unsigned int nr_pages
, struct page
**pages
)
1812 struct radix_tree_iter iter
;
1814 unsigned int ret
= 0;
1816 if (unlikely(!nr_pages
))
1820 radix_tree_for_each_contig(slot
, &mapping
->page_tree
, &iter
, index
) {
1821 struct page
*head
, *page
;
1823 page
= radix_tree_deref_slot(slot
);
1824 /* The hole, there no reason to continue */
1825 if (unlikely(!page
))
1828 if (radix_tree_exception(page
)) {
1829 if (radix_tree_deref_retry(page
)) {
1830 slot
= radix_tree_iter_retry(&iter
);
1834 * A shadow entry of a recently evicted page,
1835 * or a swap entry from shmem/tmpfs. Stop
1836 * looking for contiguous pages.
1841 head
= compound_head(page
);
1842 if (!page_cache_get_speculative(head
))
1845 /* The page was split under us? */
1846 if (compound_head(page
) != head
) {
1851 /* Has the page moved? */
1852 if (unlikely(page
!= *slot
)) {
1858 * must check mapping and index after taking the ref.
1859 * otherwise we can get both false positives and false
1860 * negatives, which is just confusing to the caller.
1862 if (page
->mapping
== NULL
|| page_to_pgoff(page
) != iter
.index
) {
1868 if (++ret
== nr_pages
)
1874 EXPORT_SYMBOL(find_get_pages_contig
);
1877 * find_get_pages_range_tag - find and return pages in given range matching @tag
1878 * @mapping: the address_space to search
1879 * @index: the starting page index
1880 * @end: The final page index (inclusive)
1881 * @tag: the tag index
1882 * @nr_pages: the maximum number of pages
1883 * @pages: where the resulting pages are placed
1885 * Like find_get_pages, except we only return pages which are tagged with
1886 * @tag. We update @index to index the next page for the traversal.
1888 unsigned find_get_pages_range_tag(struct address_space
*mapping
, pgoff_t
*index
,
1889 pgoff_t end
, int tag
, unsigned int nr_pages
,
1890 struct page
**pages
)
1892 struct radix_tree_iter iter
;
1896 if (unlikely(!nr_pages
))
1900 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1901 &iter
, *index
, tag
) {
1902 struct page
*head
, *page
;
1904 if (iter
.index
> end
)
1907 page
= radix_tree_deref_slot(slot
);
1908 if (unlikely(!page
))
1911 if (radix_tree_exception(page
)) {
1912 if (radix_tree_deref_retry(page
)) {
1913 slot
= radix_tree_iter_retry(&iter
);
1917 * A shadow entry of a recently evicted page.
1919 * Those entries should never be tagged, but
1920 * this tree walk is lockless and the tags are
1921 * looked up in bulk, one radix tree node at a
1922 * time, so there is a sizable window for page
1923 * reclaim to evict a page we saw tagged.
1930 head
= compound_head(page
);
1931 if (!page_cache_get_speculative(head
))
1934 /* The page was split under us? */
1935 if (compound_head(page
) != head
) {
1940 /* Has the page moved? */
1941 if (unlikely(page
!= *slot
)) {
1947 if (++ret
== nr_pages
) {
1948 *index
= pages
[ret
- 1]->index
+ 1;
1954 * We come here when we got at @end. We take care to not overflow the
1955 * index @index as it confuses some of the callers. This breaks the
1956 * iteration when there is page at index -1 but that is already broken
1959 if (end
== (pgoff_t
)-1)
1960 *index
= (pgoff_t
)-1;
1968 EXPORT_SYMBOL(find_get_pages_range_tag
);
1971 * find_get_entries_tag - find and return entries that match @tag
1972 * @mapping: the address_space to search
1973 * @start: the starting page cache index
1974 * @tag: the tag index
1975 * @nr_entries: the maximum number of entries
1976 * @entries: where the resulting entries are placed
1977 * @indices: the cache indices corresponding to the entries in @entries
1979 * Like find_get_entries, except we only return entries which are tagged with
1982 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1983 int tag
, unsigned int nr_entries
,
1984 struct page
**entries
, pgoff_t
*indices
)
1987 unsigned int ret
= 0;
1988 struct radix_tree_iter iter
;
1994 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
,
1995 &iter
, start
, tag
) {
1996 struct page
*head
, *page
;
1998 page
= radix_tree_deref_slot(slot
);
1999 if (unlikely(!page
))
2001 if (radix_tree_exception(page
)) {
2002 if (radix_tree_deref_retry(page
)) {
2003 slot
= radix_tree_iter_retry(&iter
);
2008 * A shadow entry of a recently evicted page, a swap
2009 * entry from shmem/tmpfs or a DAX entry. Return it
2010 * without attempting to raise page count.
2015 head
= compound_head(page
);
2016 if (!page_cache_get_speculative(head
))
2019 /* The page was split under us? */
2020 if (compound_head(page
) != head
) {
2025 /* Has the page moved? */
2026 if (unlikely(page
!= *slot
)) {
2031 indices
[ret
] = iter
.index
;
2032 entries
[ret
] = page
;
2033 if (++ret
== nr_entries
)
2039 EXPORT_SYMBOL(find_get_entries_tag
);
2042 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2043 * a _large_ part of the i/o request. Imagine the worst scenario:
2045 * ---R__________________________________________B__________
2046 * ^ reading here ^ bad block(assume 4k)
2048 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2049 * => failing the whole request => read(R) => read(R+1) =>
2050 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2051 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2052 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2054 * It is going insane. Fix it by quickly scaling down the readahead size.
2056 static void shrink_readahead_size_eio(struct file
*filp
,
2057 struct file_ra_state
*ra
)
2063 * generic_file_buffered_read - generic file read routine
2064 * @iocb: the iocb to read
2065 * @iter: data destination
2066 * @written: already copied
2068 * This is a generic file read routine, and uses the
2069 * mapping->a_ops->readpage() function for the actual low-level stuff.
2071 * This is really ugly. But the goto's actually try to clarify some
2072 * of the logic when it comes to error handling etc.
2074 static ssize_t
generic_file_buffered_read(struct kiocb
*iocb
,
2075 struct iov_iter
*iter
, ssize_t written
)
2077 struct file
*filp
= iocb
->ki_filp
;
2078 struct address_space
*mapping
= filp
->f_mapping
;
2079 struct inode
*inode
= mapping
->host
;
2080 struct file_ra_state
*ra
= &filp
->f_ra
;
2081 loff_t
*ppos
= &iocb
->ki_pos
;
2085 unsigned long offset
; /* offset into pagecache page */
2086 unsigned int prev_offset
;
2089 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
2091 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2093 index
= *ppos
>> PAGE_SHIFT
;
2094 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
2095 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
2096 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
2097 offset
= *ppos
& ~PAGE_MASK
;
2103 unsigned long nr
, ret
;
2107 if (fatal_signal_pending(current
)) {
2112 page
= find_get_page(mapping
, index
);
2114 if (iocb
->ki_flags
& IOCB_NOWAIT
)
2116 page_cache_sync_readahead(mapping
,
2118 index
, last_index
- index
);
2119 page
= find_get_page(mapping
, index
);
2120 if (unlikely(page
== NULL
))
2121 goto no_cached_page
;
2123 if (PageReadahead(page
)) {
2124 page_cache_async_readahead(mapping
,
2126 index
, last_index
- index
);
2128 if (!PageUptodate(page
)) {
2129 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2135 * See comment in do_read_cache_page on why
2136 * wait_on_page_locked is used to avoid unnecessarily
2137 * serialisations and why it's safe.
2139 error
= wait_on_page_locked_killable(page
);
2140 if (unlikely(error
))
2141 goto readpage_error
;
2142 if (PageUptodate(page
))
2145 if (inode
->i_blkbits
== PAGE_SHIFT
||
2146 !mapping
->a_ops
->is_partially_uptodate
)
2147 goto page_not_up_to_date
;
2148 /* pipes can't handle partially uptodate pages */
2149 if (unlikely(iter
->type
& ITER_PIPE
))
2150 goto page_not_up_to_date
;
2151 if (!trylock_page(page
))
2152 goto page_not_up_to_date
;
2153 /* Did it get truncated before we got the lock? */
2155 goto page_not_up_to_date_locked
;
2156 if (!mapping
->a_ops
->is_partially_uptodate(page
,
2157 offset
, iter
->count
))
2158 goto page_not_up_to_date_locked
;
2163 * i_size must be checked after we know the page is Uptodate.
2165 * Checking i_size after the check allows us to calculate
2166 * the correct value for "nr", which means the zero-filled
2167 * part of the page is not copied back to userspace (unless
2168 * another truncate extends the file - this is desired though).
2171 isize
= i_size_read(inode
);
2172 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2173 if (unlikely(!isize
|| index
> end_index
)) {
2178 /* nr is the maximum number of bytes to copy from this page */
2180 if (index
== end_index
) {
2181 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2189 /* If users can be writing to this page using arbitrary
2190 * virtual addresses, take care about potential aliasing
2191 * before reading the page on the kernel side.
2193 if (mapping_writably_mapped(mapping
))
2194 flush_dcache_page(page
);
2197 * When a sequential read accesses a page several times,
2198 * only mark it as accessed the first time.
2200 if (prev_index
!= index
|| offset
!= prev_offset
)
2201 mark_page_accessed(page
);
2205 * Ok, we have the page, and it's up-to-date, so
2206 * now we can copy it to user space...
2209 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
2211 index
+= offset
>> PAGE_SHIFT
;
2212 offset
&= ~PAGE_MASK
;
2213 prev_offset
= offset
;
2217 if (!iov_iter_count(iter
))
2225 page_not_up_to_date
:
2226 /* Get exclusive access to the page ... */
2227 error
= lock_page_killable(page
);
2228 if (unlikely(error
))
2229 goto readpage_error
;
2231 page_not_up_to_date_locked
:
2232 /* Did it get truncated before we got the lock? */
2233 if (!page
->mapping
) {
2239 /* Did somebody else fill it already? */
2240 if (PageUptodate(page
)) {
2247 * A previous I/O error may have been due to temporary
2248 * failures, eg. multipath errors.
2249 * PG_error will be set again if readpage fails.
2251 ClearPageError(page
);
2252 /* Start the actual read. The read will unlock the page. */
2253 error
= mapping
->a_ops
->readpage(filp
, page
);
2255 if (unlikely(error
)) {
2256 if (error
== AOP_TRUNCATED_PAGE
) {
2261 goto readpage_error
;
2264 if (!PageUptodate(page
)) {
2265 error
= lock_page_killable(page
);
2266 if (unlikely(error
))
2267 goto readpage_error
;
2268 if (!PageUptodate(page
)) {
2269 if (page
->mapping
== NULL
) {
2271 * invalidate_mapping_pages got it
2278 shrink_readahead_size_eio(filp
, ra
);
2280 goto readpage_error
;
2288 /* UHHUH! A synchronous read error occurred. Report it */
2294 * Ok, it wasn't cached, so we need to create a new
2297 page
= page_cache_alloc(mapping
);
2302 error
= add_to_page_cache_lru(page
, mapping
, index
,
2303 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2306 if (error
== -EEXIST
) {
2318 ra
->prev_pos
= prev_index
;
2319 ra
->prev_pos
<<= PAGE_SHIFT
;
2320 ra
->prev_pos
|= prev_offset
;
2322 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2323 file_accessed(filp
);
2324 return written
? written
: error
;
2328 * generic_file_read_iter - generic filesystem read routine
2329 * @iocb: kernel I/O control block
2330 * @iter: destination for the data read
2332 * This is the "read_iter()" routine for all filesystems
2333 * that can use the page cache directly.
2336 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2338 size_t count
= iov_iter_count(iter
);
2342 goto out
; /* skip atime */
2344 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2345 struct file
*file
= iocb
->ki_filp
;
2346 struct address_space
*mapping
= file
->f_mapping
;
2347 struct inode
*inode
= mapping
->host
;
2350 size
= i_size_read(inode
);
2351 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2352 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2353 iocb
->ki_pos
+ count
- 1))
2356 retval
= filemap_write_and_wait_range(mapping
,
2358 iocb
->ki_pos
+ count
- 1);
2363 file_accessed(file
);
2365 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2367 iocb
->ki_pos
+= retval
;
2370 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2373 * Btrfs can have a short DIO read if we encounter
2374 * compressed extents, so if there was an error, or if
2375 * we've already read everything we wanted to, or if
2376 * there was a short read because we hit EOF, go ahead
2377 * and return. Otherwise fallthrough to buffered io for
2378 * the rest of the read. Buffered reads will not work for
2379 * DAX files, so don't bother trying.
2381 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2386 retval
= generic_file_buffered_read(iocb
, iter
, retval
);
2390 EXPORT_SYMBOL(generic_file_read_iter
);
2394 * page_cache_read - adds requested page to the page cache if not already there
2395 * @file: file to read
2396 * @offset: page index
2397 * @gfp_mask: memory allocation flags
2399 * This adds the requested page to the page cache if it isn't already there,
2400 * and schedules an I/O to read in its contents from disk.
2402 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
2404 struct address_space
*mapping
= file
->f_mapping
;
2409 page
= __page_cache_alloc(gfp_mask
);
2413 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
2415 ret
= mapping
->a_ops
->readpage(file
, page
);
2416 else if (ret
== -EEXIST
)
2417 ret
= 0; /* losing race to add is OK */
2421 } while (ret
== AOP_TRUNCATED_PAGE
);
2426 #define MMAP_LOTSAMISS (100)
2429 * Synchronous readahead happens when we don't even find
2430 * a page in the page cache at all.
2432 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2433 struct file_ra_state
*ra
,
2437 struct address_space
*mapping
= file
->f_mapping
;
2439 /* If we don't want any read-ahead, don't bother */
2440 if (vma
->vm_flags
& VM_RAND_READ
)
2445 if (vma
->vm_flags
& VM_SEQ_READ
) {
2446 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2451 /* Avoid banging the cache line if not needed */
2452 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2456 * Do we miss much more than hit in this file? If so,
2457 * stop bothering with read-ahead. It will only hurt.
2459 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2465 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2466 ra
->size
= ra
->ra_pages
;
2467 ra
->async_size
= ra
->ra_pages
/ 4;
2468 ra_submit(ra
, mapping
, file
);
2472 * Asynchronous readahead happens when we find the page and PG_readahead,
2473 * so we want to possibly extend the readahead further..
2475 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2476 struct file_ra_state
*ra
,
2481 struct address_space
*mapping
= file
->f_mapping
;
2483 /* If we don't want any read-ahead, don't bother */
2484 if (vma
->vm_flags
& VM_RAND_READ
)
2486 if (ra
->mmap_miss
> 0)
2488 if (PageReadahead(page
))
2489 page_cache_async_readahead(mapping
, ra
, file
,
2490 page
, offset
, ra
->ra_pages
);
2494 * filemap_fault - read in file data for page fault handling
2495 * @vmf: struct vm_fault containing details of the fault
2497 * filemap_fault() is invoked via the vma operations vector for a
2498 * mapped memory region to read in file data during a page fault.
2500 * The goto's are kind of ugly, but this streamlines the normal case of having
2501 * it in the page cache, and handles the special cases reasonably without
2502 * having a lot of duplicated code.
2504 * vma->vm_mm->mmap_sem must be held on entry.
2506 * If our return value has VM_FAULT_RETRY set, it's because
2507 * lock_page_or_retry() returned 0.
2508 * The mmap_sem has usually been released in this case.
2509 * See __lock_page_or_retry() for the exception.
2511 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2512 * has not been released.
2514 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2516 int filemap_fault(struct vm_fault
*vmf
)
2519 struct file
*file
= vmf
->vma
->vm_file
;
2520 struct address_space
*mapping
= file
->f_mapping
;
2521 struct file_ra_state
*ra
= &file
->f_ra
;
2522 struct inode
*inode
= mapping
->host
;
2523 pgoff_t offset
= vmf
->pgoff
;
2528 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2529 if (unlikely(offset
>= max_off
))
2530 return VM_FAULT_SIGBUS
;
2533 * Do we have something in the page cache already?
2535 page
= find_get_page(mapping
, offset
);
2536 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2538 * We found the page, so try async readahead before
2539 * waiting for the lock.
2541 do_async_mmap_readahead(vmf
->vma
, ra
, file
, page
, offset
);
2543 /* No page in the page cache at all */
2544 do_sync_mmap_readahead(vmf
->vma
, ra
, file
, offset
);
2545 count_vm_event(PGMAJFAULT
);
2546 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2547 ret
= VM_FAULT_MAJOR
;
2549 page
= find_get_page(mapping
, offset
);
2551 goto no_cached_page
;
2554 if (!lock_page_or_retry(page
, vmf
->vma
->vm_mm
, vmf
->flags
)) {
2556 return ret
| VM_FAULT_RETRY
;
2559 /* Did it get truncated? */
2560 if (unlikely(page
->mapping
!= mapping
)) {
2565 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2568 * We have a locked page in the page cache, now we need to check
2569 * that it's up-to-date. If not, it is going to be due to an error.
2571 if (unlikely(!PageUptodate(page
)))
2572 goto page_not_uptodate
;
2575 * Found the page and have a reference on it.
2576 * We must recheck i_size under page lock.
2578 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2579 if (unlikely(offset
>= max_off
)) {
2582 return VM_FAULT_SIGBUS
;
2586 return ret
| VM_FAULT_LOCKED
;
2590 * We're only likely to ever get here if MADV_RANDOM is in
2593 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2596 * The page we want has now been added to the page cache.
2597 * In the unlikely event that someone removed it in the
2598 * meantime, we'll just come back here and read it again.
2604 * An error return from page_cache_read can result if the
2605 * system is low on memory, or a problem occurs while trying
2608 if (error
== -ENOMEM
)
2609 return VM_FAULT_OOM
;
2610 return VM_FAULT_SIGBUS
;
2614 * Umm, take care of errors if the page isn't up-to-date.
2615 * Try to re-read it _once_. We do this synchronously,
2616 * because there really aren't any performance issues here
2617 * and we need to check for errors.
2619 ClearPageError(page
);
2620 error
= mapping
->a_ops
->readpage(file
, page
);
2622 wait_on_page_locked(page
);
2623 if (!PageUptodate(page
))
2628 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2631 /* Things didn't work out. Return zero to tell the mm layer so. */
2632 shrink_readahead_size_eio(file
, ra
);
2633 return VM_FAULT_SIGBUS
;
2635 EXPORT_SYMBOL(filemap_fault
);
2637 void filemap_map_pages(struct vm_fault
*vmf
,
2638 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2640 struct radix_tree_iter iter
;
2642 struct file
*file
= vmf
->vma
->vm_file
;
2643 struct address_space
*mapping
= file
->f_mapping
;
2644 pgoff_t last_pgoff
= start_pgoff
;
2645 unsigned long max_idx
;
2646 struct page
*head
, *page
;
2649 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
,
2651 if (iter
.index
> end_pgoff
)
2654 page
= radix_tree_deref_slot(slot
);
2655 if (unlikely(!page
))
2657 if (radix_tree_exception(page
)) {
2658 if (radix_tree_deref_retry(page
)) {
2659 slot
= radix_tree_iter_retry(&iter
);
2665 head
= compound_head(page
);
2666 if (!page_cache_get_speculative(head
))
2669 /* The page was split under us? */
2670 if (compound_head(page
) != head
) {
2675 /* Has the page moved? */
2676 if (unlikely(page
!= *slot
)) {
2681 if (!PageUptodate(page
) ||
2682 PageReadahead(page
) ||
2685 if (!trylock_page(page
))
2688 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2691 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2692 if (page
->index
>= max_idx
)
2695 if (file
->f_ra
.mmap_miss
> 0)
2696 file
->f_ra
.mmap_miss
--;
2698 vmf
->address
+= (iter
.index
- last_pgoff
) << PAGE_SHIFT
;
2700 vmf
->pte
+= iter
.index
- last_pgoff
;
2701 last_pgoff
= iter
.index
;
2702 if (alloc_set_pte(vmf
, NULL
, page
))
2711 /* Huge page is mapped? No need to proceed. */
2712 if (pmd_trans_huge(*vmf
->pmd
))
2714 if (iter
.index
== end_pgoff
)
2719 EXPORT_SYMBOL(filemap_map_pages
);
2721 int filemap_page_mkwrite(struct vm_fault
*vmf
)
2723 struct page
*page
= vmf
->page
;
2724 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2725 int ret
= VM_FAULT_LOCKED
;
2727 sb_start_pagefault(inode
->i_sb
);
2728 vma_file_update_time(vmf
->vma
);
2730 if (page
->mapping
!= inode
->i_mapping
) {
2732 ret
= VM_FAULT_NOPAGE
;
2736 * We mark the page dirty already here so that when freeze is in
2737 * progress, we are guaranteed that writeback during freezing will
2738 * see the dirty page and writeprotect it again.
2740 set_page_dirty(page
);
2741 wait_for_stable_page(page
);
2743 sb_end_pagefault(inode
->i_sb
);
2746 EXPORT_SYMBOL(filemap_page_mkwrite
);
2748 const struct vm_operations_struct generic_file_vm_ops
= {
2749 .fault
= filemap_fault
,
2750 .map_pages
= filemap_map_pages
,
2751 .page_mkwrite
= filemap_page_mkwrite
,
2754 /* This is used for a general mmap of a disk file */
2756 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2758 struct address_space
*mapping
= file
->f_mapping
;
2760 if (!mapping
->a_ops
->readpage
)
2762 file_accessed(file
);
2763 vma
->vm_ops
= &generic_file_vm_ops
;
2768 * This is for filesystems which do not implement ->writepage.
2770 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2772 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2774 return generic_file_mmap(file
, vma
);
2777 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2781 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2785 #endif /* CONFIG_MMU */
2787 EXPORT_SYMBOL(generic_file_mmap
);
2788 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2790 static struct page
*wait_on_page_read(struct page
*page
)
2792 if (!IS_ERR(page
)) {
2793 wait_on_page_locked(page
);
2794 if (!PageUptodate(page
)) {
2796 page
= ERR_PTR(-EIO
);
2802 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2804 int (*filler
)(void *, struct page
*),
2811 page
= find_get_page(mapping
, index
);
2813 page
= __page_cache_alloc(gfp
);
2815 return ERR_PTR(-ENOMEM
);
2816 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2817 if (unlikely(err
)) {
2821 /* Presumably ENOMEM for radix tree node */
2822 return ERR_PTR(err
);
2826 err
= filler(data
, page
);
2829 return ERR_PTR(err
);
2832 page
= wait_on_page_read(page
);
2837 if (PageUptodate(page
))
2841 * Page is not up to date and may be locked due one of the following
2842 * case a: Page is being filled and the page lock is held
2843 * case b: Read/write error clearing the page uptodate status
2844 * case c: Truncation in progress (page locked)
2845 * case d: Reclaim in progress
2847 * Case a, the page will be up to date when the page is unlocked.
2848 * There is no need to serialise on the page lock here as the page
2849 * is pinned so the lock gives no additional protection. Even if the
2850 * the page is truncated, the data is still valid if PageUptodate as
2851 * it's a race vs truncate race.
2852 * Case b, the page will not be up to date
2853 * Case c, the page may be truncated but in itself, the data may still
2854 * be valid after IO completes as it's a read vs truncate race. The
2855 * operation must restart if the page is not uptodate on unlock but
2856 * otherwise serialising on page lock to stabilise the mapping gives
2857 * no additional guarantees to the caller as the page lock is
2858 * released before return.
2859 * Case d, similar to truncation. If reclaim holds the page lock, it
2860 * will be a race with remove_mapping that determines if the mapping
2861 * is valid on unlock but otherwise the data is valid and there is
2862 * no need to serialise with page lock.
2864 * As the page lock gives no additional guarantee, we optimistically
2865 * wait on the page to be unlocked and check if it's up to date and
2866 * use the page if it is. Otherwise, the page lock is required to
2867 * distinguish between the different cases. The motivation is that we
2868 * avoid spurious serialisations and wakeups when multiple processes
2869 * wait on the same page for IO to complete.
2871 wait_on_page_locked(page
);
2872 if (PageUptodate(page
))
2875 /* Distinguish between all the cases under the safety of the lock */
2878 /* Case c or d, restart the operation */
2879 if (!page
->mapping
) {
2885 /* Someone else locked and filled the page in a very small window */
2886 if (PageUptodate(page
)) {
2893 mark_page_accessed(page
);
2898 * read_cache_page - read into page cache, fill it if needed
2899 * @mapping: the page's address_space
2900 * @index: the page index
2901 * @filler: function to perform the read
2902 * @data: first arg to filler(data, page) function, often left as NULL
2904 * Read into the page cache. If a page already exists, and PageUptodate() is
2905 * not set, try to fill the page and wait for it to become unlocked.
2907 * If the page does not get brought uptodate, return -EIO.
2909 struct page
*read_cache_page(struct address_space
*mapping
,
2911 int (*filler
)(void *, struct page
*),
2914 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2916 EXPORT_SYMBOL(read_cache_page
);
2919 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2920 * @mapping: the page's address_space
2921 * @index: the page index
2922 * @gfp: the page allocator flags to use if allocating
2924 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2925 * any new page allocations done using the specified allocation flags.
2927 * If the page does not get brought uptodate, return -EIO.
2929 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2933 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2935 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2937 EXPORT_SYMBOL(read_cache_page_gfp
);
2940 * Performs necessary checks before doing a write
2942 * Can adjust writing position or amount of bytes to write.
2943 * Returns appropriate error code that caller should return or
2944 * zero in case that write should be allowed.
2946 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2948 struct file
*file
= iocb
->ki_filp
;
2949 struct inode
*inode
= file
->f_mapping
->host
;
2950 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2953 if (!iov_iter_count(from
))
2956 /* FIXME: this is for backwards compatibility with 2.4 */
2957 if (iocb
->ki_flags
& IOCB_APPEND
)
2958 iocb
->ki_pos
= i_size_read(inode
);
2962 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
2965 if (limit
!= RLIM_INFINITY
) {
2966 if (iocb
->ki_pos
>= limit
) {
2967 send_sig(SIGXFSZ
, current
, 0);
2970 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2976 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2977 !(file
->f_flags
& O_LARGEFILE
))) {
2978 if (pos
>= MAX_NON_LFS
)
2980 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2984 * Are we about to exceed the fs block limit ?
2986 * If we have written data it becomes a short write. If we have
2987 * exceeded without writing data we send a signal and return EFBIG.
2988 * Linus frestrict idea will clean these up nicely..
2990 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2993 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2994 return iov_iter_count(from
);
2996 EXPORT_SYMBOL(generic_write_checks
);
2998 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2999 loff_t pos
, unsigned len
, unsigned flags
,
3000 struct page
**pagep
, void **fsdata
)
3002 const struct address_space_operations
*aops
= mapping
->a_ops
;
3004 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
3007 EXPORT_SYMBOL(pagecache_write_begin
);
3009 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
3010 loff_t pos
, unsigned len
, unsigned copied
,
3011 struct page
*page
, void *fsdata
)
3013 const struct address_space_operations
*aops
= mapping
->a_ops
;
3015 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
3017 EXPORT_SYMBOL(pagecache_write_end
);
3020 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3022 struct file
*file
= iocb
->ki_filp
;
3023 struct address_space
*mapping
= file
->f_mapping
;
3024 struct inode
*inode
= mapping
->host
;
3025 loff_t pos
= iocb
->ki_pos
;
3030 write_len
= iov_iter_count(from
);
3031 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
3033 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3034 /* If there are pages to writeback, return */
3035 if (filemap_range_has_page(inode
->i_mapping
, pos
,
3036 pos
+ iov_iter_count(from
)))
3039 written
= filemap_write_and_wait_range(mapping
, pos
,
3040 pos
+ write_len
- 1);
3046 * After a write we want buffered reads to be sure to go to disk to get
3047 * the new data. We invalidate clean cached page from the region we're
3048 * about to write. We do this *before* the write so that we can return
3049 * without clobbering -EIOCBQUEUED from ->direct_IO().
3051 written
= invalidate_inode_pages2_range(mapping
,
3052 pos
>> PAGE_SHIFT
, end
);
3054 * If a page can not be invalidated, return 0 to fall back
3055 * to buffered write.
3058 if (written
== -EBUSY
)
3063 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3066 * Finally, try again to invalidate clean pages which might have been
3067 * cached by non-direct readahead, or faulted in by get_user_pages()
3068 * if the source of the write was an mmap'ed region of the file
3069 * we're writing. Either one is a pretty crazy thing to do,
3070 * so we don't support it 100%. If this invalidation
3071 * fails, tough, the write still worked...
3073 * Most of the time we do not need this since dio_complete() will do
3074 * the invalidation for us. However there are some file systems that
3075 * do not end up with dio_complete() being called, so let's not break
3076 * them by removing it completely
3078 if (mapping
->nrpages
)
3079 invalidate_inode_pages2_range(mapping
,
3080 pos
>> PAGE_SHIFT
, end
);
3084 write_len
-= written
;
3085 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3086 i_size_write(inode
, pos
);
3087 mark_inode_dirty(inode
);
3091 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3095 EXPORT_SYMBOL(generic_file_direct_write
);
3098 * Find or create a page at the given pagecache position. Return the locked
3099 * page. This function is specifically for buffered writes.
3101 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
3102 pgoff_t index
, unsigned flags
)
3105 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
3107 if (flags
& AOP_FLAG_NOFS
)
3108 fgp_flags
|= FGP_NOFS
;
3110 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
3111 mapping_gfp_mask(mapping
));
3113 wait_for_stable_page(page
);
3117 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3119 ssize_t
generic_perform_write(struct file
*file
,
3120 struct iov_iter
*i
, loff_t pos
)
3122 struct address_space
*mapping
= file
->f_mapping
;
3123 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3125 ssize_t written
= 0;
3126 unsigned int flags
= 0;
3130 unsigned long offset
; /* Offset into pagecache page */
3131 unsigned long bytes
; /* Bytes to write to page */
3132 size_t copied
; /* Bytes copied from user */
3135 offset
= (pos
& (PAGE_SIZE
- 1));
3136 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3141 * Bring in the user page that we will copy from _first_.
3142 * Otherwise there's a nasty deadlock on copying from the
3143 * same page as we're writing to, without it being marked
3146 * Not only is this an optimisation, but it is also required
3147 * to check that the address is actually valid, when atomic
3148 * usercopies are used, below.
3150 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3155 if (fatal_signal_pending(current
)) {
3160 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3162 if (unlikely(status
< 0))
3165 if (mapping_writably_mapped(mapping
))
3166 flush_dcache_page(page
);
3168 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3169 flush_dcache_page(page
);
3171 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3173 if (unlikely(status
< 0))
3179 iov_iter_advance(i
, copied
);
3180 if (unlikely(copied
== 0)) {
3182 * If we were unable to copy any data at all, we must
3183 * fall back to a single segment length write.
3185 * If we didn't fallback here, we could livelock
3186 * because not all segments in the iov can be copied at
3187 * once without a pagefault.
3189 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3190 iov_iter_single_seg_count(i
));
3196 balance_dirty_pages_ratelimited(mapping
);
3197 } while (iov_iter_count(i
));
3199 return written
? written
: status
;
3201 EXPORT_SYMBOL(generic_perform_write
);
3204 * __generic_file_write_iter - write data to a file
3205 * @iocb: IO state structure (file, offset, etc.)
3206 * @from: iov_iter with data to write
3208 * This function does all the work needed for actually writing data to a
3209 * file. It does all basic checks, removes SUID from the file, updates
3210 * modification times and calls proper subroutines depending on whether we
3211 * do direct IO or a standard buffered write.
3213 * It expects i_mutex to be grabbed unless we work on a block device or similar
3214 * object which does not need locking at all.
3216 * This function does *not* take care of syncing data in case of O_SYNC write.
3217 * A caller has to handle it. This is mainly due to the fact that we want to
3218 * avoid syncing under i_mutex.
3220 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3222 struct file
*file
= iocb
->ki_filp
;
3223 struct address_space
* mapping
= file
->f_mapping
;
3224 struct inode
*inode
= mapping
->host
;
3225 ssize_t written
= 0;
3229 /* We can write back this queue in page reclaim */
3230 current
->backing_dev_info
= inode_to_bdi(inode
);
3231 err
= file_remove_privs(file
);
3235 err
= file_update_time(file
);
3239 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3240 loff_t pos
, endbyte
;
3242 written
= generic_file_direct_write(iocb
, from
);
3244 * If the write stopped short of completing, fall back to
3245 * buffered writes. Some filesystems do this for writes to
3246 * holes, for example. For DAX files, a buffered write will
3247 * not succeed (even if it did, DAX does not handle dirty
3248 * page-cache pages correctly).
3250 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3253 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3255 * If generic_perform_write() returned a synchronous error
3256 * then we want to return the number of bytes which were
3257 * direct-written, or the error code if that was zero. Note
3258 * that this differs from normal direct-io semantics, which
3259 * will return -EFOO even if some bytes were written.
3261 if (unlikely(status
< 0)) {
3266 * We need to ensure that the page cache pages are written to
3267 * disk and invalidated to preserve the expected O_DIRECT
3270 endbyte
= pos
+ status
- 1;
3271 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3273 iocb
->ki_pos
= endbyte
+ 1;
3275 invalidate_mapping_pages(mapping
,
3277 endbyte
>> PAGE_SHIFT
);
3280 * We don't know how much we wrote, so just return
3281 * the number of bytes which were direct-written
3285 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3286 if (likely(written
> 0))
3287 iocb
->ki_pos
+= written
;
3290 current
->backing_dev_info
= NULL
;
3291 return written
? written
: err
;
3293 EXPORT_SYMBOL(__generic_file_write_iter
);
3296 * generic_file_write_iter - write data to a file
3297 * @iocb: IO state structure
3298 * @from: iov_iter with data to write
3300 * This is a wrapper around __generic_file_write_iter() to be used by most
3301 * filesystems. It takes care of syncing the file in case of O_SYNC file
3302 * and acquires i_mutex as needed.
3304 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3306 struct file
*file
= iocb
->ki_filp
;
3307 struct inode
*inode
= file
->f_mapping
->host
;
3311 ret
= generic_write_checks(iocb
, from
);
3313 ret
= __generic_file_write_iter(iocb
, from
);
3314 inode_unlock(inode
);
3317 ret
= generic_write_sync(iocb
, ret
);
3320 EXPORT_SYMBOL(generic_file_write_iter
);
3323 * try_to_release_page() - release old fs-specific metadata on a page
3325 * @page: the page which the kernel is trying to free
3326 * @gfp_mask: memory allocation flags (and I/O mode)
3328 * The address_space is to try to release any data against the page
3329 * (presumably at page->private). If the release was successful, return '1'.
3330 * Otherwise return zero.
3332 * This may also be called if PG_fscache is set on a page, indicating that the
3333 * page is known to the local caching routines.
3335 * The @gfp_mask argument specifies whether I/O may be performed to release
3336 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3339 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3341 struct address_space
* const mapping
= page
->mapping
;
3343 BUG_ON(!PageLocked(page
));
3344 if (PageWriteback(page
))
3347 if (mapping
&& mapping
->a_ops
->releasepage
)
3348 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3349 return try_to_free_buffers(page
);
3352 EXPORT_SYMBOL(try_to_release_page
);