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/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/shmem_fs.h>
38 #include <linux/rmap.h>
39 #include <linux/delayacct.h>
40 #include <linux/psi.h>
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/filemap.h>
47 * FIXME: remove all knowledge of the buffer layer from the core VM
49 #include <linux/buffer_head.h> /* for try_to_free_buffers */
54 * Shared mappings implemented 30.11.1994. It's not fully working yet,
57 * Shared mappings now work. 15.8.1995 Bruno.
59 * finished 'unifying' the page and buffer cache and SMP-threaded the
60 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
62 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
68 * ->i_mmap_rwsem (truncate_pagecache)
69 * ->private_lock (__free_pte->__set_page_dirty_buffers)
70 * ->swap_lock (exclusive_swap_page, others)
74 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
78 * ->page_table_lock or pte_lock (various, mainly in memory.c)
79 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
82 * ->lock_page (access_process_vm)
84 * ->i_mutex (generic_perform_write)
85 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
88 * sb_lock (fs/fs-writeback.c)
89 * ->i_pages lock (__sync_single_inode)
92 * ->anon_vma.lock (vma_adjust)
95 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
97 * ->page_table_lock or pte_lock
98 * ->swap_lock (try_to_unmap_one)
99 * ->private_lock (try_to_unmap_one)
100 * ->i_pages lock (try_to_unmap_one)
101 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
102 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
103 * ->private_lock (page_remove_rmap->set_page_dirty)
104 * ->i_pages lock (page_remove_rmap->set_page_dirty)
105 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
106 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
107 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
108 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
109 * ->inode->i_lock (zap_pte_range->set_page_dirty)
110 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
113 * ->tasklist_lock (memory_failure, collect_procs_ao)
116 static void page_cache_delete(struct address_space
*mapping
,
117 struct page
*page
, void *shadow
)
119 XA_STATE(xas
, &mapping
->i_pages
, page
->index
);
122 mapping_set_update(&xas
, mapping
);
124 /* hugetlb pages are represented by a single entry in the xarray */
125 if (!PageHuge(page
)) {
126 xas_set_order(&xas
, page
->index
, compound_order(page
));
127 nr
= 1U << compound_order(page
);
130 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
131 VM_BUG_ON_PAGE(PageTail(page
), page
);
132 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
134 xas_store(&xas
, shadow
);
135 xas_init_marks(&xas
);
137 page
->mapping
= NULL
;
138 /* Leave page->index set: truncation lookup relies upon it */
141 mapping
->nrexceptional
+= nr
;
143 * Make sure the nrexceptional update is committed before
144 * the nrpages update so that final truncate racing
145 * with reclaim does not see both counters 0 at the
146 * same time and miss a shadow entry.
150 mapping
->nrpages
-= nr
;
153 static void unaccount_page_cache_page(struct address_space
*mapping
,
159 * if we're uptodate, flush out into the cleancache, otherwise
160 * invalidate any existing cleancache entries. We can't leave
161 * stale data around in the cleancache once our page is gone
163 if (PageUptodate(page
) && PageMappedToDisk(page
))
164 cleancache_put_page(page
);
166 cleancache_invalidate_page(mapping
, page
);
168 VM_BUG_ON_PAGE(PageTail(page
), page
);
169 VM_BUG_ON_PAGE(page_mapped(page
), page
);
170 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
173 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
174 current
->comm
, page_to_pfn(page
));
175 dump_page(page
, "still mapped when deleted");
177 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
179 mapcount
= page_mapcount(page
);
180 if (mapping_exiting(mapping
) &&
181 page_count(page
) >= mapcount
+ 2) {
183 * All vmas have already been torn down, so it's
184 * a good bet that actually the page is unmapped,
185 * and we'd prefer not to leak it: if we're wrong,
186 * some other bad page check should catch it later.
188 page_mapcount_reset(page
);
189 page_ref_sub(page
, mapcount
);
193 /* hugetlb pages do not participate in page cache accounting. */
197 nr
= hpage_nr_pages(page
);
199 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, -nr
);
200 if (PageSwapBacked(page
)) {
201 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, -nr
);
202 if (PageTransHuge(page
))
203 __dec_node_page_state(page
, NR_SHMEM_THPS
);
205 VM_BUG_ON_PAGE(PageTransHuge(page
), page
);
209 * At this point page must be either written or cleaned by
210 * truncate. Dirty page here signals a bug and loss of
213 * This fixes dirty accounting after removing the page entirely
214 * but leaves PageDirty set: it has no effect for truncated
215 * page and anyway will be cleared before returning page into
218 if (WARN_ON_ONCE(PageDirty(page
)))
219 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
223 * Delete a page from the page cache and free it. Caller has to make
224 * sure the page is locked and that nobody else uses it - or that usage
225 * is safe. The caller must hold the i_pages lock.
227 void __delete_from_page_cache(struct page
*page
, void *shadow
)
229 struct address_space
*mapping
= page
->mapping
;
231 trace_mm_filemap_delete_from_page_cache(page
);
233 unaccount_page_cache_page(mapping
, page
);
234 page_cache_delete(mapping
, page
, shadow
);
237 static void page_cache_free_page(struct address_space
*mapping
,
240 void (*freepage
)(struct page
*);
242 freepage
= mapping
->a_ops
->freepage
;
246 if (PageTransHuge(page
) && !PageHuge(page
)) {
247 page_ref_sub(page
, HPAGE_PMD_NR
);
248 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
255 * delete_from_page_cache - delete page from page cache
256 * @page: the page which the kernel is trying to remove from page cache
258 * This must be called only on pages that have been verified to be in the page
259 * cache and locked. It will never put the page into the free list, the caller
260 * has a reference on the page.
262 void delete_from_page_cache(struct page
*page
)
264 struct address_space
*mapping
= page_mapping(page
);
267 BUG_ON(!PageLocked(page
));
268 xa_lock_irqsave(&mapping
->i_pages
, flags
);
269 __delete_from_page_cache(page
, NULL
);
270 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
272 page_cache_free_page(mapping
, page
);
274 EXPORT_SYMBOL(delete_from_page_cache
);
277 * page_cache_delete_batch - delete several pages from page cache
278 * @mapping: the mapping to which pages belong
279 * @pvec: pagevec with pages to delete
281 * The function walks over mapping->i_pages and removes pages passed in @pvec
282 * from the mapping. The function expects @pvec to be sorted by page index.
283 * It tolerates holes in @pvec (mapping entries at those indices are not
284 * modified). The function expects only THP head pages to be present in the
285 * @pvec and takes care to delete all corresponding tail pages from the
288 * The function expects the i_pages lock to be held.
290 static void page_cache_delete_batch(struct address_space
*mapping
,
291 struct pagevec
*pvec
)
293 XA_STATE(xas
, &mapping
->i_pages
, pvec
->pages
[0]->index
);
295 int i
= 0, tail_pages
= 0;
298 mapping_set_update(&xas
, mapping
);
299 xas_for_each(&xas
, page
, ULONG_MAX
) {
300 if (i
>= pagevec_count(pvec
) && !tail_pages
)
302 if (xa_is_value(page
))
306 * Some page got inserted in our range? Skip it. We
307 * have our pages locked so they are protected from
310 if (page
!= pvec
->pages
[i
]) {
311 VM_BUG_ON_PAGE(page
->index
>
312 pvec
->pages
[i
]->index
, page
);
315 WARN_ON_ONCE(!PageLocked(page
));
316 if (PageTransHuge(page
) && !PageHuge(page
))
317 tail_pages
= HPAGE_PMD_NR
- 1;
318 page
->mapping
= NULL
;
320 * Leave page->index set: truncation lookup relies
325 VM_BUG_ON_PAGE(page
->index
+ HPAGE_PMD_NR
- tail_pages
326 != pvec
->pages
[i
]->index
, page
);
329 xas_store(&xas
, NULL
);
332 mapping
->nrpages
-= total_pages
;
335 void delete_from_page_cache_batch(struct address_space
*mapping
,
336 struct pagevec
*pvec
)
341 if (!pagevec_count(pvec
))
344 xa_lock_irqsave(&mapping
->i_pages
, flags
);
345 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
346 trace_mm_filemap_delete_from_page_cache(pvec
->pages
[i
]);
348 unaccount_page_cache_page(mapping
, pvec
->pages
[i
]);
350 page_cache_delete_batch(mapping
, pvec
);
351 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
353 for (i
= 0; i
< pagevec_count(pvec
); i
++)
354 page_cache_free_page(mapping
, pvec
->pages
[i
]);
357 int filemap_check_errors(struct address_space
*mapping
)
360 /* Check for outstanding write errors */
361 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
362 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
364 if (test_bit(AS_EIO
, &mapping
->flags
) &&
365 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
369 EXPORT_SYMBOL(filemap_check_errors
);
371 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
373 /* Check for outstanding write errors */
374 if (test_bit(AS_EIO
, &mapping
->flags
))
376 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
382 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
383 * @mapping: address space structure to write
384 * @start: offset in bytes where the range starts
385 * @end: offset in bytes where the range ends (inclusive)
386 * @sync_mode: enable synchronous operation
388 * Start writeback against all of a mapping's dirty pages that lie
389 * within the byte offsets <start, end> inclusive.
391 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
392 * opposed to a regular memory cleansing writeback. The difference between
393 * these two operations is that if a dirty page/buffer is encountered, it must
394 * be waited upon, and not just skipped over.
396 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
397 loff_t end
, int sync_mode
)
400 struct writeback_control wbc
= {
401 .sync_mode
= sync_mode
,
402 .nr_to_write
= LONG_MAX
,
403 .range_start
= start
,
407 if (!mapping_cap_writeback_dirty(mapping
))
410 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
411 ret
= do_writepages(mapping
, &wbc
);
412 wbc_detach_inode(&wbc
);
416 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
419 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
422 int filemap_fdatawrite(struct address_space
*mapping
)
424 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
426 EXPORT_SYMBOL(filemap_fdatawrite
);
428 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
431 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
433 EXPORT_SYMBOL(filemap_fdatawrite_range
);
436 * filemap_flush - mostly a non-blocking flush
437 * @mapping: target address_space
439 * This is a mostly non-blocking flush. Not suitable for data-integrity
440 * purposes - I/O may not be started against all dirty pages.
442 int filemap_flush(struct address_space
*mapping
)
444 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
446 EXPORT_SYMBOL(filemap_flush
);
449 * filemap_range_has_page - check if a page exists in range.
450 * @mapping: address space within which to check
451 * @start_byte: offset in bytes where the range starts
452 * @end_byte: offset in bytes where the range ends (inclusive)
454 * Find at least one page in the range supplied, usually used to check if
455 * direct writing in this range will trigger a writeback.
457 bool filemap_range_has_page(struct address_space
*mapping
,
458 loff_t start_byte
, loff_t end_byte
)
461 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
462 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
464 if (end_byte
< start_byte
)
469 page
= xas_find(&xas
, max
);
470 if (xas_retry(&xas
, page
))
472 /* Shadow entries don't count */
473 if (xa_is_value(page
))
476 * We don't need to try to pin this page; we're about to
477 * release the RCU lock anyway. It is enough to know that
478 * there was a page here recently.
486 EXPORT_SYMBOL(filemap_range_has_page
);
488 static void __filemap_fdatawait_range(struct address_space
*mapping
,
489 loff_t start_byte
, loff_t end_byte
)
491 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
492 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
496 if (end_byte
< start_byte
)
500 while (index
<= end
) {
503 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
,
504 end
, PAGECACHE_TAG_WRITEBACK
);
508 for (i
= 0; i
< nr_pages
; i
++) {
509 struct page
*page
= pvec
.pages
[i
];
511 wait_on_page_writeback(page
);
512 ClearPageError(page
);
514 pagevec_release(&pvec
);
520 * filemap_fdatawait_range - wait for writeback to complete
521 * @mapping: address space structure to wait for
522 * @start_byte: offset in bytes where the range starts
523 * @end_byte: offset in bytes where the range ends (inclusive)
525 * Walk the list of under-writeback pages of the given address space
526 * in the given range and wait for all of them. Check error status of
527 * the address space and return it.
529 * Since the error status of the address space is cleared by this function,
530 * callers are responsible for checking the return value and handling and/or
531 * reporting the error.
533 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
536 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
537 return filemap_check_errors(mapping
);
539 EXPORT_SYMBOL(filemap_fdatawait_range
);
542 * file_fdatawait_range - wait for writeback to complete
543 * @file: file pointing to address space structure to wait for
544 * @start_byte: offset in bytes where the range starts
545 * @end_byte: offset in bytes where the range ends (inclusive)
547 * Walk the list of under-writeback pages of the address space that file
548 * refers to, in the given range and wait for all of them. Check error
549 * status of the address space vs. the file->f_wb_err cursor and return it.
551 * Since the error status of the file is advanced by this function,
552 * callers are responsible for checking the return value and handling and/or
553 * reporting the error.
555 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
557 struct address_space
*mapping
= file
->f_mapping
;
559 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
560 return file_check_and_advance_wb_err(file
);
562 EXPORT_SYMBOL(file_fdatawait_range
);
565 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
566 * @mapping: address space structure to wait for
568 * Walk the list of under-writeback pages of the given address space
569 * and wait for all of them. Unlike filemap_fdatawait(), this function
570 * does not clear error status of the address space.
572 * Use this function if callers don't handle errors themselves. Expected
573 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
576 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
578 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
579 return filemap_check_and_keep_errors(mapping
);
581 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
583 static bool mapping_needs_writeback(struct address_space
*mapping
)
585 return (!dax_mapping(mapping
) && mapping
->nrpages
) ||
586 (dax_mapping(mapping
) && mapping
->nrexceptional
);
589 int filemap_write_and_wait(struct address_space
*mapping
)
593 if (mapping_needs_writeback(mapping
)) {
594 err
= filemap_fdatawrite(mapping
);
596 * Even if the above returned error, the pages may be
597 * written partially (e.g. -ENOSPC), so we wait for it.
598 * But the -EIO is special case, it may indicate the worst
599 * thing (e.g. bug) happened, so we avoid waiting for it.
602 int err2
= filemap_fdatawait(mapping
);
606 /* Clear any previously stored errors */
607 filemap_check_errors(mapping
);
610 err
= filemap_check_errors(mapping
);
614 EXPORT_SYMBOL(filemap_write_and_wait
);
617 * filemap_write_and_wait_range - write out & wait on a file range
618 * @mapping: the address_space for the pages
619 * @lstart: offset in bytes where the range starts
620 * @lend: offset in bytes where the range ends (inclusive)
622 * Write out and wait upon file offsets lstart->lend, inclusive.
624 * Note that @lend is inclusive (describes the last byte to be written) so
625 * that this function can be used to write to the very end-of-file (end = -1).
627 int filemap_write_and_wait_range(struct address_space
*mapping
,
628 loff_t lstart
, loff_t lend
)
632 if (mapping_needs_writeback(mapping
)) {
633 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
635 /* See comment of filemap_write_and_wait() */
637 int err2
= filemap_fdatawait_range(mapping
,
642 /* Clear any previously stored errors */
643 filemap_check_errors(mapping
);
646 err
= filemap_check_errors(mapping
);
650 EXPORT_SYMBOL(filemap_write_and_wait_range
);
652 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
654 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
656 trace_filemap_set_wb_err(mapping
, eseq
);
658 EXPORT_SYMBOL(__filemap_set_wb_err
);
661 * file_check_and_advance_wb_err - report wb error (if any) that was previously
662 * and advance wb_err to current one
663 * @file: struct file on which the error is being reported
665 * When userland calls fsync (or something like nfsd does the equivalent), we
666 * want to report any writeback errors that occurred since the last fsync (or
667 * since the file was opened if there haven't been any).
669 * Grab the wb_err from the mapping. If it matches what we have in the file,
670 * then just quickly return 0. The file is all caught up.
672 * If it doesn't match, then take the mapping value, set the "seen" flag in
673 * it and try to swap it into place. If it works, or another task beat us
674 * to it with the new value, then update the f_wb_err and return the error
675 * portion. The error at this point must be reported via proper channels
676 * (a'la fsync, or NFS COMMIT operation, etc.).
678 * While we handle mapping->wb_err with atomic operations, the f_wb_err
679 * value is protected by the f_lock since we must ensure that it reflects
680 * the latest value swapped in for this file descriptor.
682 int file_check_and_advance_wb_err(struct file
*file
)
685 errseq_t old
= READ_ONCE(file
->f_wb_err
);
686 struct address_space
*mapping
= file
->f_mapping
;
688 /* Locklessly handle the common case where nothing has changed */
689 if (errseq_check(&mapping
->wb_err
, old
)) {
690 /* Something changed, must use slow path */
691 spin_lock(&file
->f_lock
);
692 old
= file
->f_wb_err
;
693 err
= errseq_check_and_advance(&mapping
->wb_err
,
695 trace_file_check_and_advance_wb_err(file
, old
);
696 spin_unlock(&file
->f_lock
);
700 * We're mostly using this function as a drop in replacement for
701 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
702 * that the legacy code would have had on these flags.
704 clear_bit(AS_EIO
, &mapping
->flags
);
705 clear_bit(AS_ENOSPC
, &mapping
->flags
);
708 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
711 * file_write_and_wait_range - write out & wait on a file range
712 * @file: file pointing to address_space with pages
713 * @lstart: offset in bytes where the range starts
714 * @lend: offset in bytes where the range ends (inclusive)
716 * Write out and wait upon file offsets lstart->lend, inclusive.
718 * Note that @lend is inclusive (describes the last byte to be written) so
719 * that this function can be used to write to the very end-of-file (end = -1).
721 * After writing out and waiting on the data, we check and advance the
722 * f_wb_err cursor to the latest value, and return any errors detected there.
724 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
727 struct address_space
*mapping
= file
->f_mapping
;
729 if (mapping_needs_writeback(mapping
)) {
730 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
732 /* See comment of filemap_write_and_wait() */
734 __filemap_fdatawait_range(mapping
, lstart
, lend
);
736 err2
= file_check_and_advance_wb_err(file
);
741 EXPORT_SYMBOL(file_write_and_wait_range
);
744 * replace_page_cache_page - replace a pagecache page with a new one
745 * @old: page to be replaced
746 * @new: page to replace with
747 * @gfp_mask: allocation mode
749 * This function replaces a page in the pagecache with a new one. On
750 * success it acquires the pagecache reference for the new page and
751 * drops it for the old page. Both the old and new pages must be
752 * locked. This function does not add the new page to the LRU, the
753 * caller must do that.
755 * The remove + add is atomic. This function cannot fail.
757 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
759 struct address_space
*mapping
= old
->mapping
;
760 void (*freepage
)(struct page
*) = mapping
->a_ops
->freepage
;
761 pgoff_t offset
= old
->index
;
762 XA_STATE(xas
, &mapping
->i_pages
, offset
);
765 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
766 VM_BUG_ON_PAGE(!PageLocked(new), new);
767 VM_BUG_ON_PAGE(new->mapping
, new);
770 new->mapping
= mapping
;
773 xas_lock_irqsave(&xas
, flags
);
774 xas_store(&xas
, new);
777 /* hugetlb pages do not participate in page cache accounting. */
779 __dec_node_page_state(new, NR_FILE_PAGES
);
781 __inc_node_page_state(new, NR_FILE_PAGES
);
782 if (PageSwapBacked(old
))
783 __dec_node_page_state(new, NR_SHMEM
);
784 if (PageSwapBacked(new))
785 __inc_node_page_state(new, NR_SHMEM
);
786 xas_unlock_irqrestore(&xas
, flags
);
787 mem_cgroup_migrate(old
, new);
794 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
796 static int __add_to_page_cache_locked(struct page
*page
,
797 struct address_space
*mapping
,
798 pgoff_t offset
, gfp_t gfp_mask
,
801 XA_STATE(xas
, &mapping
->i_pages
, offset
);
802 int huge
= PageHuge(page
);
803 struct mem_cgroup
*memcg
;
807 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
808 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
809 mapping_set_update(&xas
, mapping
);
812 error
= mem_cgroup_try_charge(page
, current
->mm
,
813 gfp_mask
, &memcg
, false);
819 page
->mapping
= mapping
;
820 page
->index
= offset
;
824 old
= xas_load(&xas
);
825 if (old
&& !xa_is_value(old
))
826 xas_set_err(&xas
, -EEXIST
);
827 xas_store(&xas
, page
);
831 if (xa_is_value(old
)) {
832 mapping
->nrexceptional
--;
838 /* hugetlb pages do not participate in page cache accounting */
840 __inc_node_page_state(page
, NR_FILE_PAGES
);
842 xas_unlock_irq(&xas
);
843 } while (xas_nomem(&xas
, gfp_mask
& GFP_RECLAIM_MASK
));
849 mem_cgroup_commit_charge(page
, memcg
, false, false);
850 trace_mm_filemap_add_to_page_cache(page
);
853 page
->mapping
= NULL
;
854 /* Leave page->index set: truncation relies upon it */
856 mem_cgroup_cancel_charge(page
, memcg
, false);
858 return xas_error(&xas
);
862 * add_to_page_cache_locked - add a locked page to the pagecache
864 * @mapping: the page's address_space
865 * @offset: page index
866 * @gfp_mask: page allocation mode
868 * This function is used to add a page to the pagecache. It must be locked.
869 * This function does not add the page to the LRU. The caller must do that.
871 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
872 pgoff_t offset
, gfp_t gfp_mask
)
874 return __add_to_page_cache_locked(page
, mapping
, offset
,
877 EXPORT_SYMBOL(add_to_page_cache_locked
);
879 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
880 pgoff_t offset
, gfp_t gfp_mask
)
885 __SetPageLocked(page
);
886 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
889 __ClearPageLocked(page
);
892 * The page might have been evicted from cache only
893 * recently, in which case it should be activated like
894 * any other repeatedly accessed page.
895 * The exception is pages getting rewritten; evicting other
896 * data from the working set, only to cache data that will
897 * get overwritten with something else, is a waste of memory.
899 WARN_ON_ONCE(PageActive(page
));
900 if (!(gfp_mask
& __GFP_WRITE
) && shadow
)
901 workingset_refault(page
, shadow
);
906 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
909 struct page
*__page_cache_alloc(gfp_t gfp
)
914 if (cpuset_do_page_mem_spread()) {
915 unsigned int cpuset_mems_cookie
;
917 cpuset_mems_cookie
= read_mems_allowed_begin();
918 n
= cpuset_mem_spread_node();
919 page
= __alloc_pages_node(n
, gfp
, 0);
920 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
924 return alloc_pages(gfp
, 0);
926 EXPORT_SYMBOL(__page_cache_alloc
);
930 * In order to wait for pages to become available there must be
931 * waitqueues associated with pages. By using a hash table of
932 * waitqueues where the bucket discipline is to maintain all
933 * waiters on the same queue and wake all when any of the pages
934 * become available, and for the woken contexts to check to be
935 * sure the appropriate page became available, this saves space
936 * at a cost of "thundering herd" phenomena during rare hash
939 #define PAGE_WAIT_TABLE_BITS 8
940 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
941 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
943 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
945 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
948 void __init
pagecache_init(void)
952 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
953 init_waitqueue_head(&page_wait_table
[i
]);
955 page_writeback_init();
958 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
959 struct wait_page_key
{
965 struct wait_page_queue
{
968 wait_queue_entry_t wait
;
971 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
973 struct wait_page_key
*key
= arg
;
974 struct wait_page_queue
*wait_page
975 = container_of(wait
, struct wait_page_queue
, wait
);
977 if (wait_page
->page
!= key
->page
)
981 if (wait_page
->bit_nr
!= key
->bit_nr
)
984 /* Stop walking if it's locked */
985 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
988 return autoremove_wake_function(wait
, mode
, sync
, key
);
991 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
993 wait_queue_head_t
*q
= page_waitqueue(page
);
994 struct wait_page_key key
;
996 wait_queue_entry_t bookmark
;
1003 bookmark
.private = NULL
;
1004 bookmark
.func
= NULL
;
1005 INIT_LIST_HEAD(&bookmark
.entry
);
1007 spin_lock_irqsave(&q
->lock
, flags
);
1008 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1010 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1012 * Take a breather from holding the lock,
1013 * allow pages that finish wake up asynchronously
1014 * to acquire the lock and remove themselves
1017 spin_unlock_irqrestore(&q
->lock
, flags
);
1019 spin_lock_irqsave(&q
->lock
, flags
);
1020 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1024 * It is possible for other pages to have collided on the waitqueue
1025 * hash, so in that case check for a page match. That prevents a long-
1028 * It is still possible to miss a case here, when we woke page waiters
1029 * and removed them from the waitqueue, but there are still other
1032 if (!waitqueue_active(q
) || !key
.page_match
) {
1033 ClearPageWaiters(page
);
1035 * It's possible to miss clearing Waiters here, when we woke
1036 * our page waiters, but the hashed waitqueue has waiters for
1037 * other pages on it.
1039 * That's okay, it's a rare case. The next waker will clear it.
1042 spin_unlock_irqrestore(&q
->lock
, flags
);
1045 static void wake_up_page(struct page
*page
, int bit
)
1047 if (!PageWaiters(page
))
1049 wake_up_page_bit(page
, bit
);
1052 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
1053 struct page
*page
, int bit_nr
, int state
, bool lock
)
1055 struct wait_page_queue wait_page
;
1056 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1057 bool thrashing
= false;
1058 unsigned long pflags
;
1061 if (bit_nr
== PG_locked
&&
1062 !PageUptodate(page
) && PageWorkingset(page
)) {
1063 if (!PageSwapBacked(page
))
1064 delayacct_thrashing_start();
1065 psi_memstall_enter(&pflags
);
1070 wait
->flags
= lock
? WQ_FLAG_EXCLUSIVE
: 0;
1071 wait
->func
= wake_page_function
;
1072 wait_page
.page
= page
;
1073 wait_page
.bit_nr
= bit_nr
;
1076 spin_lock_irq(&q
->lock
);
1078 if (likely(list_empty(&wait
->entry
))) {
1079 __add_wait_queue_entry_tail(q
, wait
);
1080 SetPageWaiters(page
);
1083 set_current_state(state
);
1085 spin_unlock_irq(&q
->lock
);
1087 if (likely(test_bit(bit_nr
, &page
->flags
))) {
1092 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
1095 if (!test_bit(bit_nr
, &page
->flags
))
1099 if (unlikely(signal_pending_state(state
, current
))) {
1105 finish_wait(q
, wait
);
1108 if (!PageSwapBacked(page
))
1109 delayacct_thrashing_end();
1110 psi_memstall_leave(&pflags
);
1114 * A signal could leave PageWaiters set. Clearing it here if
1115 * !waitqueue_active would be possible (by open-coding finish_wait),
1116 * but still fail to catch it in the case of wait hash collision. We
1117 * already can fail to clear wait hash collision cases, so don't
1118 * bother with signals either.
1124 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1126 wait_queue_head_t
*q
= page_waitqueue(page
);
1127 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, false);
1129 EXPORT_SYMBOL(wait_on_page_bit
);
1131 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1133 wait_queue_head_t
*q
= page_waitqueue(page
);
1134 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, false);
1136 EXPORT_SYMBOL(wait_on_page_bit_killable
);
1139 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1140 * @page: Page defining the wait queue of interest
1141 * @waiter: Waiter to add to the queue
1143 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1145 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1147 wait_queue_head_t
*q
= page_waitqueue(page
);
1148 unsigned long flags
;
1150 spin_lock_irqsave(&q
->lock
, flags
);
1151 __add_wait_queue_entry_tail(q
, waiter
);
1152 SetPageWaiters(page
);
1153 spin_unlock_irqrestore(&q
->lock
, flags
);
1155 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1157 #ifndef clear_bit_unlock_is_negative_byte
1160 * PG_waiters is the high bit in the same byte as PG_lock.
1162 * On x86 (and on many other architectures), we can clear PG_lock and
1163 * test the sign bit at the same time. But if the architecture does
1164 * not support that special operation, we just do this all by hand
1167 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1168 * being cleared, but a memory barrier should be unneccssary since it is
1169 * in the same byte as PG_locked.
1171 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1173 clear_bit_unlock(nr
, mem
);
1174 /* smp_mb__after_atomic(); */
1175 return test_bit(PG_waiters
, mem
);
1181 * unlock_page - unlock a locked page
1184 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1185 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1186 * mechanism between PageLocked pages and PageWriteback pages is shared.
1187 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1189 * Note that this depends on PG_waiters being the sign bit in the byte
1190 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1191 * clear the PG_locked bit and test PG_waiters at the same time fairly
1192 * portably (architectures that do LL/SC can test any bit, while x86 can
1193 * test the sign bit).
1195 void unlock_page(struct page
*page
)
1197 BUILD_BUG_ON(PG_waiters
!= 7);
1198 page
= compound_head(page
);
1199 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1200 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1201 wake_up_page_bit(page
, PG_locked
);
1203 EXPORT_SYMBOL(unlock_page
);
1206 * end_page_writeback - end writeback against a page
1209 void end_page_writeback(struct page
*page
)
1212 * TestClearPageReclaim could be used here but it is an atomic
1213 * operation and overkill in this particular case. Failing to
1214 * shuffle a page marked for immediate reclaim is too mild to
1215 * justify taking an atomic operation penalty at the end of
1216 * ever page writeback.
1218 if (PageReclaim(page
)) {
1219 ClearPageReclaim(page
);
1220 rotate_reclaimable_page(page
);
1223 if (!test_clear_page_writeback(page
))
1226 smp_mb__after_atomic();
1227 wake_up_page(page
, PG_writeback
);
1229 EXPORT_SYMBOL(end_page_writeback
);
1232 * After completing I/O on a page, call this routine to update the page
1233 * flags appropriately
1235 void page_endio(struct page
*page
, bool is_write
, int err
)
1239 SetPageUptodate(page
);
1241 ClearPageUptodate(page
);
1247 struct address_space
*mapping
;
1250 mapping
= page_mapping(page
);
1252 mapping_set_error(mapping
, err
);
1254 end_page_writeback(page
);
1257 EXPORT_SYMBOL_GPL(page_endio
);
1260 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1261 * @__page: the page to lock
1263 void __lock_page(struct page
*__page
)
1265 struct page
*page
= compound_head(__page
);
1266 wait_queue_head_t
*q
= page_waitqueue(page
);
1267 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, true);
1269 EXPORT_SYMBOL(__lock_page
);
1271 int __lock_page_killable(struct page
*__page
)
1273 struct page
*page
= compound_head(__page
);
1274 wait_queue_head_t
*q
= page_waitqueue(page
);
1275 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
, true);
1277 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1281 * 1 - page is locked; mmap_sem is still held.
1282 * 0 - page is not locked.
1283 * mmap_sem has been released (up_read()), unless flags had both
1284 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1285 * which case mmap_sem is still held.
1287 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1288 * with the page locked and the mmap_sem unperturbed.
1290 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1293 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1295 * CAUTION! In this case, mmap_sem is not released
1296 * even though return 0.
1298 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1301 up_read(&mm
->mmap_sem
);
1302 if (flags
& FAULT_FLAG_KILLABLE
)
1303 wait_on_page_locked_killable(page
);
1305 wait_on_page_locked(page
);
1308 if (flags
& FAULT_FLAG_KILLABLE
) {
1311 ret
= __lock_page_killable(page
);
1313 up_read(&mm
->mmap_sem
);
1323 * page_cache_next_miss() - Find the next gap in the page cache.
1324 * @mapping: Mapping.
1326 * @max_scan: Maximum range to search.
1328 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1329 * gap with the lowest index.
1331 * This function may be called under the rcu_read_lock. However, this will
1332 * not atomically search a snapshot of the cache at a single point in time.
1333 * For example, if a gap is created at index 5, then subsequently a gap is
1334 * created at index 10, page_cache_next_miss covering both indices may
1335 * return 10 if called under the rcu_read_lock.
1337 * Return: The index of the gap if found, otherwise an index outside the
1338 * range specified (in which case 'return - index >= max_scan' will be true).
1339 * In the rare case of index wrap-around, 0 will be returned.
1341 pgoff_t
page_cache_next_miss(struct address_space
*mapping
,
1342 pgoff_t index
, unsigned long max_scan
)
1344 XA_STATE(xas
, &mapping
->i_pages
, index
);
1346 while (max_scan
--) {
1347 void *entry
= xas_next(&xas
);
1348 if (!entry
|| xa_is_value(entry
))
1350 if (xas
.xa_index
== 0)
1354 return xas
.xa_index
;
1356 EXPORT_SYMBOL(page_cache_next_miss
);
1359 * page_cache_prev_miss() - Find the next gap in the page cache.
1360 * @mapping: Mapping.
1362 * @max_scan: Maximum range to search.
1364 * Search the range [max(index - max_scan + 1, 0), index] for the
1365 * gap with the highest index.
1367 * This function may be called under the rcu_read_lock. However, this will
1368 * not atomically search a snapshot of the cache at a single point in time.
1369 * For example, if a gap is created at index 10, then subsequently a gap is
1370 * created at index 5, page_cache_prev_miss() covering both indices may
1371 * return 5 if called under the rcu_read_lock.
1373 * Return: The index of the gap if found, otherwise an index outside the
1374 * range specified (in which case 'index - return >= max_scan' will be true).
1375 * In the rare case of wrap-around, ULONG_MAX will be returned.
1377 pgoff_t
page_cache_prev_miss(struct address_space
*mapping
,
1378 pgoff_t index
, unsigned long max_scan
)
1380 XA_STATE(xas
, &mapping
->i_pages
, index
);
1382 while (max_scan
--) {
1383 void *entry
= xas_prev(&xas
);
1384 if (!entry
|| xa_is_value(entry
))
1386 if (xas
.xa_index
== ULONG_MAX
)
1390 return xas
.xa_index
;
1392 EXPORT_SYMBOL(page_cache_prev_miss
);
1395 * find_get_entry - find and get a page cache entry
1396 * @mapping: the address_space to search
1397 * @offset: the page cache index
1399 * Looks up the page cache slot at @mapping & @offset. If there is a
1400 * page cache page, it is returned with an increased refcount.
1402 * If the slot holds a shadow entry of a previously evicted page, or a
1403 * swap entry from shmem/tmpfs, it is returned.
1405 * Otherwise, %NULL is returned.
1407 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1409 XA_STATE(xas
, &mapping
->i_pages
, offset
);
1410 struct page
*head
, *page
;
1415 page
= xas_load(&xas
);
1416 if (xas_retry(&xas
, page
))
1419 * A shadow entry of a recently evicted page, or a swap entry from
1420 * shmem/tmpfs. Return it without attempting to raise page count.
1422 if (!page
|| xa_is_value(page
))
1425 head
= compound_head(page
);
1426 if (!page_cache_get_speculative(head
))
1429 /* The page was split under us? */
1430 if (compound_head(page
) != head
) {
1436 * Has the page moved?
1437 * This is part of the lockless pagecache protocol. See
1438 * include/linux/pagemap.h for details.
1440 if (unlikely(page
!= xas_reload(&xas
))) {
1449 EXPORT_SYMBOL(find_get_entry
);
1452 * find_lock_entry - locate, pin and lock a page cache entry
1453 * @mapping: the address_space to search
1454 * @offset: the page cache index
1456 * Looks up the page cache slot at @mapping & @offset. If there is a
1457 * page cache page, it is returned locked and with an increased
1460 * If the slot holds a shadow entry of a previously evicted page, or a
1461 * swap entry from shmem/tmpfs, it is returned.
1463 * Otherwise, %NULL is returned.
1465 * find_lock_entry() may sleep.
1467 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1472 page
= find_get_entry(mapping
, offset
);
1473 if (page
&& !xa_is_value(page
)) {
1475 /* Has the page been truncated? */
1476 if (unlikely(page_mapping(page
) != mapping
)) {
1481 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1485 EXPORT_SYMBOL(find_lock_entry
);
1488 * pagecache_get_page - find and get a page reference
1489 * @mapping: the address_space to search
1490 * @offset: the page index
1491 * @fgp_flags: PCG flags
1492 * @gfp_mask: gfp mask to use for the page cache data page allocation
1494 * Looks up the page cache slot at @mapping & @offset.
1496 * PCG flags modify how the page is returned.
1498 * @fgp_flags can be:
1500 * - FGP_ACCESSED: the page will be marked accessed
1501 * - FGP_LOCK: Page is return locked
1502 * - FGP_CREAT: If page is not present then a new page is allocated using
1503 * @gfp_mask and added to the page cache and the VM's LRU
1504 * list. The page is returned locked and with an increased
1505 * refcount. Otherwise, NULL is returned.
1507 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1508 * if the GFP flags specified for FGP_CREAT are atomic.
1510 * If there is a page cache page, it is returned with an increased refcount.
1512 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
1513 int fgp_flags
, gfp_t gfp_mask
)
1518 page
= find_get_entry(mapping
, offset
);
1519 if (xa_is_value(page
))
1524 if (fgp_flags
& FGP_LOCK
) {
1525 if (fgp_flags
& FGP_NOWAIT
) {
1526 if (!trylock_page(page
)) {
1534 /* Has the page been truncated? */
1535 if (unlikely(page
->mapping
!= mapping
)) {
1540 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
1543 if (page
&& (fgp_flags
& FGP_ACCESSED
))
1544 mark_page_accessed(page
);
1547 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1549 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1550 gfp_mask
|= __GFP_WRITE
;
1551 if (fgp_flags
& FGP_NOFS
)
1552 gfp_mask
&= ~__GFP_FS
;
1554 page
= __page_cache_alloc(gfp_mask
);
1558 if (WARN_ON_ONCE(!(fgp_flags
& FGP_LOCK
)))
1559 fgp_flags
|= FGP_LOCK
;
1561 /* Init accessed so avoid atomic mark_page_accessed later */
1562 if (fgp_flags
& FGP_ACCESSED
)
1563 __SetPageReferenced(page
);
1565 err
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
1566 if (unlikely(err
)) {
1576 EXPORT_SYMBOL(pagecache_get_page
);
1579 * find_get_entries - gang pagecache lookup
1580 * @mapping: The address_space to search
1581 * @start: The starting page cache index
1582 * @nr_entries: The maximum number of entries
1583 * @entries: Where the resulting entries are placed
1584 * @indices: The cache indices corresponding to the entries in @entries
1586 * find_get_entries() will search for and return a group of up to
1587 * @nr_entries entries in the mapping. The entries are placed at
1588 * @entries. find_get_entries() takes a reference against any actual
1591 * The search returns a group of mapping-contiguous page cache entries
1592 * with ascending indexes. There may be holes in the indices due to
1593 * not-present pages.
1595 * Any shadow entries of evicted pages, or swap entries from
1596 * shmem/tmpfs, are included in the returned array.
1598 * find_get_entries() returns the number of pages and shadow entries
1601 unsigned find_get_entries(struct address_space
*mapping
,
1602 pgoff_t start
, unsigned int nr_entries
,
1603 struct page
**entries
, pgoff_t
*indices
)
1605 XA_STATE(xas
, &mapping
->i_pages
, start
);
1607 unsigned int ret
= 0;
1613 xas_for_each(&xas
, page
, ULONG_MAX
) {
1615 if (xas_retry(&xas
, page
))
1618 * A shadow entry of a recently evicted page, a swap
1619 * entry from shmem/tmpfs or a DAX entry. Return it
1620 * without attempting to raise page count.
1622 if (xa_is_value(page
))
1625 head
= compound_head(page
);
1626 if (!page_cache_get_speculative(head
))
1629 /* The page was split under us? */
1630 if (compound_head(page
) != head
)
1633 /* Has the page moved? */
1634 if (unlikely(page
!= xas_reload(&xas
)))
1638 indices
[ret
] = xas
.xa_index
;
1639 entries
[ret
] = page
;
1640 if (++ret
== nr_entries
)
1653 * find_get_pages_range - gang pagecache lookup
1654 * @mapping: The address_space to search
1655 * @start: The starting page index
1656 * @end: The final page index (inclusive)
1657 * @nr_pages: The maximum number of pages
1658 * @pages: Where the resulting pages are placed
1660 * find_get_pages_range() will search for and return a group of up to @nr_pages
1661 * pages in the mapping starting at index @start and up to index @end
1662 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1663 * a reference against the returned pages.
1665 * The search returns a group of mapping-contiguous pages with ascending
1666 * indexes. There may be holes in the indices due to not-present pages.
1667 * We also update @start to index the next page for the traversal.
1669 * find_get_pages_range() returns the number of pages which were found. If this
1670 * number is smaller than @nr_pages, the end of specified range has been
1673 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
1674 pgoff_t end
, unsigned int nr_pages
,
1675 struct page
**pages
)
1677 XA_STATE(xas
, &mapping
->i_pages
, *start
);
1681 if (unlikely(!nr_pages
))
1685 xas_for_each(&xas
, page
, end
) {
1687 if (xas_retry(&xas
, page
))
1689 /* Skip over shadow, swap and DAX entries */
1690 if (xa_is_value(page
))
1693 head
= compound_head(page
);
1694 if (!page_cache_get_speculative(head
))
1697 /* The page was split under us? */
1698 if (compound_head(page
) != head
)
1701 /* Has the page moved? */
1702 if (unlikely(page
!= xas_reload(&xas
)))
1706 if (++ret
== nr_pages
) {
1707 *start
= page
->index
+ 1;
1718 * We come here when there is no page beyond @end. We take care to not
1719 * overflow the index @start as it confuses some of the callers. This
1720 * breaks the iteration when there is a page at index -1 but that is
1721 * already broken anyway.
1723 if (end
== (pgoff_t
)-1)
1724 *start
= (pgoff_t
)-1;
1734 * find_get_pages_contig - gang contiguous pagecache lookup
1735 * @mapping: The address_space to search
1736 * @index: The starting page index
1737 * @nr_pages: The maximum number of pages
1738 * @pages: Where the resulting pages are placed
1740 * find_get_pages_contig() works exactly like find_get_pages(), except
1741 * that the returned number of pages are guaranteed to be contiguous.
1743 * find_get_pages_contig() returns the number of pages which were found.
1745 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1746 unsigned int nr_pages
, struct page
**pages
)
1748 XA_STATE(xas
, &mapping
->i_pages
, index
);
1750 unsigned int ret
= 0;
1752 if (unlikely(!nr_pages
))
1756 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
1758 if (xas_retry(&xas
, page
))
1761 * If the entry has been swapped out, we can stop looking.
1762 * No current caller is looking for DAX entries.
1764 if (xa_is_value(page
))
1767 head
= compound_head(page
);
1768 if (!page_cache_get_speculative(head
))
1771 /* The page was split under us? */
1772 if (compound_head(page
) != head
)
1775 /* Has the page moved? */
1776 if (unlikely(page
!= xas_reload(&xas
)))
1780 * must check mapping and index after taking the ref.
1781 * otherwise we can get both false positives and false
1782 * negatives, which is just confusing to the caller.
1784 if (!page
->mapping
|| page_to_pgoff(page
) != xas
.xa_index
) {
1790 if (++ret
== nr_pages
)
1801 EXPORT_SYMBOL(find_get_pages_contig
);
1804 * find_get_pages_range_tag - find and return pages in given range matching @tag
1805 * @mapping: the address_space to search
1806 * @index: the starting page index
1807 * @end: The final page index (inclusive)
1808 * @tag: the tag index
1809 * @nr_pages: the maximum number of pages
1810 * @pages: where the resulting pages are placed
1812 * Like find_get_pages, except we only return pages which are tagged with
1813 * @tag. We update @index to index the next page for the traversal.
1815 unsigned find_get_pages_range_tag(struct address_space
*mapping
, pgoff_t
*index
,
1816 pgoff_t end
, xa_mark_t tag
, unsigned int nr_pages
,
1817 struct page
**pages
)
1819 XA_STATE(xas
, &mapping
->i_pages
, *index
);
1823 if (unlikely(!nr_pages
))
1827 xas_for_each_marked(&xas
, page
, end
, tag
) {
1829 if (xas_retry(&xas
, page
))
1832 * Shadow entries should never be tagged, but this iteration
1833 * is lockless so there is a window for page reclaim to evict
1834 * a page we saw tagged. Skip over it.
1836 if (xa_is_value(page
))
1839 head
= compound_head(page
);
1840 if (!page_cache_get_speculative(head
))
1843 /* The page was split under us? */
1844 if (compound_head(page
) != head
)
1847 /* Has the page moved? */
1848 if (unlikely(page
!= xas_reload(&xas
)))
1852 if (++ret
== nr_pages
) {
1853 *index
= page
->index
+ 1;
1864 * We come here when we got to @end. We take care to not overflow the
1865 * index @index as it confuses some of the callers. This breaks the
1866 * iteration when there is a page at index -1 but that is already
1869 if (end
== (pgoff_t
)-1)
1870 *index
= (pgoff_t
)-1;
1878 EXPORT_SYMBOL(find_get_pages_range_tag
);
1881 * find_get_entries_tag - find and return entries that match @tag
1882 * @mapping: the address_space to search
1883 * @start: the starting page cache index
1884 * @tag: the tag index
1885 * @nr_entries: the maximum number of entries
1886 * @entries: where the resulting entries are placed
1887 * @indices: the cache indices corresponding to the entries in @entries
1889 * Like find_get_entries, except we only return entries which are tagged with
1892 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
1893 xa_mark_t tag
, unsigned int nr_entries
,
1894 struct page
**entries
, pgoff_t
*indices
)
1896 XA_STATE(xas
, &mapping
->i_pages
, start
);
1898 unsigned int ret
= 0;
1904 xas_for_each_marked(&xas
, page
, ULONG_MAX
, tag
) {
1906 if (xas_retry(&xas
, page
))
1909 * A shadow entry of a recently evicted page, a swap
1910 * entry from shmem/tmpfs or a DAX entry. Return it
1911 * without attempting to raise page count.
1913 if (xa_is_value(page
))
1916 head
= compound_head(page
);
1917 if (!page_cache_get_speculative(head
))
1920 /* The page was split under us? */
1921 if (compound_head(page
) != head
)
1924 /* Has the page moved? */
1925 if (unlikely(page
!= xas_reload(&xas
)))
1929 indices
[ret
] = xas
.xa_index
;
1930 entries
[ret
] = page
;
1931 if (++ret
== nr_entries
)
1942 EXPORT_SYMBOL(find_get_entries_tag
);
1945 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1946 * a _large_ part of the i/o request. Imagine the worst scenario:
1948 * ---R__________________________________________B__________
1949 * ^ reading here ^ bad block(assume 4k)
1951 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1952 * => failing the whole request => read(R) => read(R+1) =>
1953 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1954 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1955 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1957 * It is going insane. Fix it by quickly scaling down the readahead size.
1959 static void shrink_readahead_size_eio(struct file
*filp
,
1960 struct file_ra_state
*ra
)
1966 * generic_file_buffered_read - generic file read routine
1967 * @iocb: the iocb to read
1968 * @iter: data destination
1969 * @written: already copied
1971 * This is a generic file read routine, and uses the
1972 * mapping->a_ops->readpage() function for the actual low-level stuff.
1974 * This is really ugly. But the goto's actually try to clarify some
1975 * of the logic when it comes to error handling etc.
1977 static ssize_t
generic_file_buffered_read(struct kiocb
*iocb
,
1978 struct iov_iter
*iter
, ssize_t written
)
1980 struct file
*filp
= iocb
->ki_filp
;
1981 struct address_space
*mapping
= filp
->f_mapping
;
1982 struct inode
*inode
= mapping
->host
;
1983 struct file_ra_state
*ra
= &filp
->f_ra
;
1984 loff_t
*ppos
= &iocb
->ki_pos
;
1988 unsigned long offset
; /* offset into pagecache page */
1989 unsigned int prev_offset
;
1992 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
1994 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
1996 index
= *ppos
>> PAGE_SHIFT
;
1997 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
1998 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
1999 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
2000 offset
= *ppos
& ~PAGE_MASK
;
2006 unsigned long nr
, ret
;
2010 if (fatal_signal_pending(current
)) {
2015 page
= find_get_page(mapping
, index
);
2017 if (iocb
->ki_flags
& IOCB_NOWAIT
)
2019 page_cache_sync_readahead(mapping
,
2021 index
, last_index
- index
);
2022 page
= find_get_page(mapping
, index
);
2023 if (unlikely(page
== NULL
))
2024 goto no_cached_page
;
2026 if (PageReadahead(page
)) {
2027 page_cache_async_readahead(mapping
,
2029 index
, last_index
- index
);
2031 if (!PageUptodate(page
)) {
2032 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2038 * See comment in do_read_cache_page on why
2039 * wait_on_page_locked is used to avoid unnecessarily
2040 * serialisations and why it's safe.
2042 error
= wait_on_page_locked_killable(page
);
2043 if (unlikely(error
))
2044 goto readpage_error
;
2045 if (PageUptodate(page
))
2048 if (inode
->i_blkbits
== PAGE_SHIFT
||
2049 !mapping
->a_ops
->is_partially_uptodate
)
2050 goto page_not_up_to_date
;
2051 /* pipes can't handle partially uptodate pages */
2052 if (unlikely(iter
->type
& ITER_PIPE
))
2053 goto page_not_up_to_date
;
2054 if (!trylock_page(page
))
2055 goto page_not_up_to_date
;
2056 /* Did it get truncated before we got the lock? */
2058 goto page_not_up_to_date_locked
;
2059 if (!mapping
->a_ops
->is_partially_uptodate(page
,
2060 offset
, iter
->count
))
2061 goto page_not_up_to_date_locked
;
2066 * i_size must be checked after we know the page is Uptodate.
2068 * Checking i_size after the check allows us to calculate
2069 * the correct value for "nr", which means the zero-filled
2070 * part of the page is not copied back to userspace (unless
2071 * another truncate extends the file - this is desired though).
2074 isize
= i_size_read(inode
);
2075 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2076 if (unlikely(!isize
|| index
> end_index
)) {
2081 /* nr is the maximum number of bytes to copy from this page */
2083 if (index
== end_index
) {
2084 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2092 /* If users can be writing to this page using arbitrary
2093 * virtual addresses, take care about potential aliasing
2094 * before reading the page on the kernel side.
2096 if (mapping_writably_mapped(mapping
))
2097 flush_dcache_page(page
);
2100 * When a sequential read accesses a page several times,
2101 * only mark it as accessed the first time.
2103 if (prev_index
!= index
|| offset
!= prev_offset
)
2104 mark_page_accessed(page
);
2108 * Ok, we have the page, and it's up-to-date, so
2109 * now we can copy it to user space...
2112 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
2114 index
+= offset
>> PAGE_SHIFT
;
2115 offset
&= ~PAGE_MASK
;
2116 prev_offset
= offset
;
2120 if (!iov_iter_count(iter
))
2128 page_not_up_to_date
:
2129 /* Get exclusive access to the page ... */
2130 error
= lock_page_killable(page
);
2131 if (unlikely(error
))
2132 goto readpage_error
;
2134 page_not_up_to_date_locked
:
2135 /* Did it get truncated before we got the lock? */
2136 if (!page
->mapping
) {
2142 /* Did somebody else fill it already? */
2143 if (PageUptodate(page
)) {
2150 * A previous I/O error may have been due to temporary
2151 * failures, eg. multipath errors.
2152 * PG_error will be set again if readpage fails.
2154 ClearPageError(page
);
2155 /* Start the actual read. The read will unlock the page. */
2156 error
= mapping
->a_ops
->readpage(filp
, page
);
2158 if (unlikely(error
)) {
2159 if (error
== AOP_TRUNCATED_PAGE
) {
2164 goto readpage_error
;
2167 if (!PageUptodate(page
)) {
2168 error
= lock_page_killable(page
);
2169 if (unlikely(error
))
2170 goto readpage_error
;
2171 if (!PageUptodate(page
)) {
2172 if (page
->mapping
== NULL
) {
2174 * invalidate_mapping_pages got it
2181 shrink_readahead_size_eio(filp
, ra
);
2183 goto readpage_error
;
2191 /* UHHUH! A synchronous read error occurred. Report it */
2197 * Ok, it wasn't cached, so we need to create a new
2200 page
= page_cache_alloc(mapping
);
2205 error
= add_to_page_cache_lru(page
, mapping
, index
,
2206 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2209 if (error
== -EEXIST
) {
2221 ra
->prev_pos
= prev_index
;
2222 ra
->prev_pos
<<= PAGE_SHIFT
;
2223 ra
->prev_pos
|= prev_offset
;
2225 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2226 file_accessed(filp
);
2227 return written
? written
: error
;
2231 * generic_file_read_iter - generic filesystem read routine
2232 * @iocb: kernel I/O control block
2233 * @iter: destination for the data read
2235 * This is the "read_iter()" routine for all filesystems
2236 * that can use the page cache directly.
2239 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2241 size_t count
= iov_iter_count(iter
);
2245 goto out
; /* skip atime */
2247 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2248 struct file
*file
= iocb
->ki_filp
;
2249 struct address_space
*mapping
= file
->f_mapping
;
2250 struct inode
*inode
= mapping
->host
;
2253 size
= i_size_read(inode
);
2254 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2255 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2256 iocb
->ki_pos
+ count
- 1))
2259 retval
= filemap_write_and_wait_range(mapping
,
2261 iocb
->ki_pos
+ count
- 1);
2266 file_accessed(file
);
2268 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2270 iocb
->ki_pos
+= retval
;
2273 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2276 * Btrfs can have a short DIO read if we encounter
2277 * compressed extents, so if there was an error, or if
2278 * we've already read everything we wanted to, or if
2279 * there was a short read because we hit EOF, go ahead
2280 * and return. Otherwise fallthrough to buffered io for
2281 * the rest of the read. Buffered reads will not work for
2282 * DAX files, so don't bother trying.
2284 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2289 retval
= generic_file_buffered_read(iocb
, iter
, retval
);
2293 EXPORT_SYMBOL(generic_file_read_iter
);
2297 * page_cache_read - adds requested page to the page cache if not already there
2298 * @file: file to read
2299 * @offset: page index
2300 * @gfp_mask: memory allocation flags
2302 * This adds the requested page to the page cache if it isn't already there,
2303 * and schedules an I/O to read in its contents from disk.
2305 static int page_cache_read(struct file
*file
, pgoff_t offset
, gfp_t gfp_mask
)
2307 struct address_space
*mapping
= file
->f_mapping
;
2312 page
= __page_cache_alloc(gfp_mask
);
2316 ret
= add_to_page_cache_lru(page
, mapping
, offset
, gfp_mask
);
2318 ret
= mapping
->a_ops
->readpage(file
, page
);
2319 else if (ret
== -EEXIST
)
2320 ret
= 0; /* losing race to add is OK */
2324 } while (ret
== AOP_TRUNCATED_PAGE
);
2329 #define MMAP_LOTSAMISS (100)
2332 * Synchronous readahead happens when we don't even find
2333 * a page in the page cache at all.
2335 static void do_sync_mmap_readahead(struct vm_area_struct
*vma
,
2336 struct file_ra_state
*ra
,
2340 struct address_space
*mapping
= file
->f_mapping
;
2342 /* If we don't want any read-ahead, don't bother */
2343 if (vma
->vm_flags
& VM_RAND_READ
)
2348 if (vma
->vm_flags
& VM_SEQ_READ
) {
2349 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2354 /* Avoid banging the cache line if not needed */
2355 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2359 * Do we miss much more than hit in this file? If so,
2360 * stop bothering with read-ahead. It will only hurt.
2362 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2368 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2369 ra
->size
= ra
->ra_pages
;
2370 ra
->async_size
= ra
->ra_pages
/ 4;
2371 ra_submit(ra
, mapping
, file
);
2375 * Asynchronous readahead happens when we find the page and PG_readahead,
2376 * so we want to possibly extend the readahead further..
2378 static void do_async_mmap_readahead(struct vm_area_struct
*vma
,
2379 struct file_ra_state
*ra
,
2384 struct address_space
*mapping
= file
->f_mapping
;
2386 /* If we don't want any read-ahead, don't bother */
2387 if (vma
->vm_flags
& VM_RAND_READ
)
2389 if (ra
->mmap_miss
> 0)
2391 if (PageReadahead(page
))
2392 page_cache_async_readahead(mapping
, ra
, file
,
2393 page
, offset
, ra
->ra_pages
);
2397 * filemap_fault - read in file data for page fault handling
2398 * @vmf: struct vm_fault containing details of the fault
2400 * filemap_fault() is invoked via the vma operations vector for a
2401 * mapped memory region to read in file data during a page fault.
2403 * The goto's are kind of ugly, but this streamlines the normal case of having
2404 * it in the page cache, and handles the special cases reasonably without
2405 * having a lot of duplicated code.
2407 * vma->vm_mm->mmap_sem must be held on entry.
2409 * If our return value has VM_FAULT_RETRY set, it's because
2410 * lock_page_or_retry() returned 0.
2411 * The mmap_sem has usually been released in this case.
2412 * See __lock_page_or_retry() for the exception.
2414 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2415 * has not been released.
2417 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2419 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
2422 struct file
*file
= vmf
->vma
->vm_file
;
2423 struct address_space
*mapping
= file
->f_mapping
;
2424 struct file_ra_state
*ra
= &file
->f_ra
;
2425 struct inode
*inode
= mapping
->host
;
2426 pgoff_t offset
= vmf
->pgoff
;
2431 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2432 if (unlikely(offset
>= max_off
))
2433 return VM_FAULT_SIGBUS
;
2436 * Do we have something in the page cache already?
2438 page
= find_get_page(mapping
, offset
);
2439 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2441 * We found the page, so try async readahead before
2442 * waiting for the lock.
2444 do_async_mmap_readahead(vmf
->vma
, ra
, file
, page
, offset
);
2446 /* No page in the page cache at all */
2447 do_sync_mmap_readahead(vmf
->vma
, ra
, file
, offset
);
2448 count_vm_event(PGMAJFAULT
);
2449 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2450 ret
= VM_FAULT_MAJOR
;
2452 page
= find_get_page(mapping
, offset
);
2454 goto no_cached_page
;
2457 if (!lock_page_or_retry(page
, vmf
->vma
->vm_mm
, vmf
->flags
)) {
2459 return ret
| VM_FAULT_RETRY
;
2462 /* Did it get truncated? */
2463 if (unlikely(page
->mapping
!= mapping
)) {
2468 VM_BUG_ON_PAGE(page
->index
!= offset
, page
);
2471 * We have a locked page in the page cache, now we need to check
2472 * that it's up-to-date. If not, it is going to be due to an error.
2474 if (unlikely(!PageUptodate(page
)))
2475 goto page_not_uptodate
;
2478 * Found the page and have a reference on it.
2479 * We must recheck i_size under page lock.
2481 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2482 if (unlikely(offset
>= max_off
)) {
2485 return VM_FAULT_SIGBUS
;
2489 return ret
| VM_FAULT_LOCKED
;
2493 * We're only likely to ever get here if MADV_RANDOM is in
2496 error
= page_cache_read(file
, offset
, vmf
->gfp_mask
);
2499 * The page we want has now been added to the page cache.
2500 * In the unlikely event that someone removed it in the
2501 * meantime, we'll just come back here and read it again.
2507 * An error return from page_cache_read can result if the
2508 * system is low on memory, or a problem occurs while trying
2511 return vmf_error(error
);
2515 * Umm, take care of errors if the page isn't up-to-date.
2516 * Try to re-read it _once_. We do this synchronously,
2517 * because there really aren't any performance issues here
2518 * and we need to check for errors.
2520 ClearPageError(page
);
2521 error
= mapping
->a_ops
->readpage(file
, page
);
2523 wait_on_page_locked(page
);
2524 if (!PageUptodate(page
))
2529 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2532 /* Things didn't work out. Return zero to tell the mm layer so. */
2533 shrink_readahead_size_eio(file
, ra
);
2534 return VM_FAULT_SIGBUS
;
2536 EXPORT_SYMBOL(filemap_fault
);
2538 void filemap_map_pages(struct vm_fault
*vmf
,
2539 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2541 struct file
*file
= vmf
->vma
->vm_file
;
2542 struct address_space
*mapping
= file
->f_mapping
;
2543 pgoff_t last_pgoff
= start_pgoff
;
2544 unsigned long max_idx
;
2545 XA_STATE(xas
, &mapping
->i_pages
, start_pgoff
);
2546 struct page
*head
, *page
;
2549 xas_for_each(&xas
, page
, end_pgoff
) {
2550 if (xas_retry(&xas
, page
))
2552 if (xa_is_value(page
))
2555 head
= compound_head(page
);
2556 if (!page_cache_get_speculative(head
))
2559 /* The page was split under us? */
2560 if (compound_head(page
) != head
)
2563 /* Has the page moved? */
2564 if (unlikely(page
!= xas_reload(&xas
)))
2567 if (!PageUptodate(page
) ||
2568 PageReadahead(page
) ||
2571 if (!trylock_page(page
))
2574 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2577 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2578 if (page
->index
>= max_idx
)
2581 if (file
->f_ra
.mmap_miss
> 0)
2582 file
->f_ra
.mmap_miss
--;
2584 vmf
->address
+= (xas
.xa_index
- last_pgoff
) << PAGE_SHIFT
;
2586 vmf
->pte
+= xas
.xa_index
- last_pgoff
;
2587 last_pgoff
= xas
.xa_index
;
2588 if (alloc_set_pte(vmf
, NULL
, page
))
2597 /* Huge page is mapped? No need to proceed. */
2598 if (pmd_trans_huge(*vmf
->pmd
))
2603 EXPORT_SYMBOL(filemap_map_pages
);
2605 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
2607 struct page
*page
= vmf
->page
;
2608 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2609 vm_fault_t ret
= VM_FAULT_LOCKED
;
2611 sb_start_pagefault(inode
->i_sb
);
2612 file_update_time(vmf
->vma
->vm_file
);
2614 if (page
->mapping
!= inode
->i_mapping
) {
2616 ret
= VM_FAULT_NOPAGE
;
2620 * We mark the page dirty already here so that when freeze is in
2621 * progress, we are guaranteed that writeback during freezing will
2622 * see the dirty page and writeprotect it again.
2624 set_page_dirty(page
);
2625 wait_for_stable_page(page
);
2627 sb_end_pagefault(inode
->i_sb
);
2631 const struct vm_operations_struct generic_file_vm_ops
= {
2632 .fault
= filemap_fault
,
2633 .map_pages
= filemap_map_pages
,
2634 .page_mkwrite
= filemap_page_mkwrite
,
2637 /* This is used for a general mmap of a disk file */
2639 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2641 struct address_space
*mapping
= file
->f_mapping
;
2643 if (!mapping
->a_ops
->readpage
)
2645 file_accessed(file
);
2646 vma
->vm_ops
= &generic_file_vm_ops
;
2651 * This is for filesystems which do not implement ->writepage.
2653 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2655 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2657 return generic_file_mmap(file
, vma
);
2660 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
2662 return VM_FAULT_SIGBUS
;
2664 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2668 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2672 #endif /* CONFIG_MMU */
2674 EXPORT_SYMBOL(filemap_page_mkwrite
);
2675 EXPORT_SYMBOL(generic_file_mmap
);
2676 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2678 static struct page
*wait_on_page_read(struct page
*page
)
2680 if (!IS_ERR(page
)) {
2681 wait_on_page_locked(page
);
2682 if (!PageUptodate(page
)) {
2684 page
= ERR_PTR(-EIO
);
2690 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2692 int (*filler
)(void *, struct page
*),
2699 page
= find_get_page(mapping
, index
);
2701 page
= __page_cache_alloc(gfp
);
2703 return ERR_PTR(-ENOMEM
);
2704 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2705 if (unlikely(err
)) {
2709 /* Presumably ENOMEM for xarray node */
2710 return ERR_PTR(err
);
2714 err
= filler(data
, page
);
2717 return ERR_PTR(err
);
2720 page
= wait_on_page_read(page
);
2725 if (PageUptodate(page
))
2729 * Page is not up to date and may be locked due one of the following
2730 * case a: Page is being filled and the page lock is held
2731 * case b: Read/write error clearing the page uptodate status
2732 * case c: Truncation in progress (page locked)
2733 * case d: Reclaim in progress
2735 * Case a, the page will be up to date when the page is unlocked.
2736 * There is no need to serialise on the page lock here as the page
2737 * is pinned so the lock gives no additional protection. Even if the
2738 * the page is truncated, the data is still valid if PageUptodate as
2739 * it's a race vs truncate race.
2740 * Case b, the page will not be up to date
2741 * Case c, the page may be truncated but in itself, the data may still
2742 * be valid after IO completes as it's a read vs truncate race. The
2743 * operation must restart if the page is not uptodate on unlock but
2744 * otherwise serialising on page lock to stabilise the mapping gives
2745 * no additional guarantees to the caller as the page lock is
2746 * released before return.
2747 * Case d, similar to truncation. If reclaim holds the page lock, it
2748 * will be a race with remove_mapping that determines if the mapping
2749 * is valid on unlock but otherwise the data is valid and there is
2750 * no need to serialise with page lock.
2752 * As the page lock gives no additional guarantee, we optimistically
2753 * wait on the page to be unlocked and check if it's up to date and
2754 * use the page if it is. Otherwise, the page lock is required to
2755 * distinguish between the different cases. The motivation is that we
2756 * avoid spurious serialisations and wakeups when multiple processes
2757 * wait on the same page for IO to complete.
2759 wait_on_page_locked(page
);
2760 if (PageUptodate(page
))
2763 /* Distinguish between all the cases under the safety of the lock */
2766 /* Case c or d, restart the operation */
2767 if (!page
->mapping
) {
2773 /* Someone else locked and filled the page in a very small window */
2774 if (PageUptodate(page
)) {
2781 mark_page_accessed(page
);
2786 * read_cache_page - read into page cache, fill it if needed
2787 * @mapping: the page's address_space
2788 * @index: the page index
2789 * @filler: function to perform the read
2790 * @data: first arg to filler(data, page) function, often left as NULL
2792 * Read into the page cache. If a page already exists, and PageUptodate() is
2793 * not set, try to fill the page and wait for it to become unlocked.
2795 * If the page does not get brought uptodate, return -EIO.
2797 struct page
*read_cache_page(struct address_space
*mapping
,
2799 int (*filler
)(void *, struct page
*),
2802 return do_read_cache_page(mapping
, index
, filler
, data
, mapping_gfp_mask(mapping
));
2804 EXPORT_SYMBOL(read_cache_page
);
2807 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2808 * @mapping: the page's address_space
2809 * @index: the page index
2810 * @gfp: the page allocator flags to use if allocating
2812 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2813 * any new page allocations done using the specified allocation flags.
2815 * If the page does not get brought uptodate, return -EIO.
2817 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2821 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
2823 return do_read_cache_page(mapping
, index
, filler
, NULL
, gfp
);
2825 EXPORT_SYMBOL(read_cache_page_gfp
);
2828 * Performs necessary checks before doing a write
2830 * Can adjust writing position or amount of bytes to write.
2831 * Returns appropriate error code that caller should return or
2832 * zero in case that write should be allowed.
2834 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2836 struct file
*file
= iocb
->ki_filp
;
2837 struct inode
*inode
= file
->f_mapping
->host
;
2838 unsigned long limit
= rlimit(RLIMIT_FSIZE
);
2841 if (!iov_iter_count(from
))
2844 /* FIXME: this is for backwards compatibility with 2.4 */
2845 if (iocb
->ki_flags
& IOCB_APPEND
)
2846 iocb
->ki_pos
= i_size_read(inode
);
2850 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
2853 if (limit
!= RLIM_INFINITY
) {
2854 if (iocb
->ki_pos
>= limit
) {
2855 send_sig(SIGXFSZ
, current
, 0);
2858 iov_iter_truncate(from
, limit
- (unsigned long)pos
);
2864 if (unlikely(pos
+ iov_iter_count(from
) > MAX_NON_LFS
&&
2865 !(file
->f_flags
& O_LARGEFILE
))) {
2866 if (pos
>= MAX_NON_LFS
)
2868 iov_iter_truncate(from
, MAX_NON_LFS
- (unsigned long)pos
);
2872 * Are we about to exceed the fs block limit ?
2874 * If we have written data it becomes a short write. If we have
2875 * exceeded without writing data we send a signal and return EFBIG.
2876 * Linus frestrict idea will clean these up nicely..
2878 if (unlikely(pos
>= inode
->i_sb
->s_maxbytes
))
2881 iov_iter_truncate(from
, inode
->i_sb
->s_maxbytes
- pos
);
2882 return iov_iter_count(from
);
2884 EXPORT_SYMBOL(generic_write_checks
);
2886 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
2887 loff_t pos
, unsigned len
, unsigned flags
,
2888 struct page
**pagep
, void **fsdata
)
2890 const struct address_space_operations
*aops
= mapping
->a_ops
;
2892 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
2895 EXPORT_SYMBOL(pagecache_write_begin
);
2897 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
2898 loff_t pos
, unsigned len
, unsigned copied
,
2899 struct page
*page
, void *fsdata
)
2901 const struct address_space_operations
*aops
= mapping
->a_ops
;
2903 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2905 EXPORT_SYMBOL(pagecache_write_end
);
2908 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
2910 struct file
*file
= iocb
->ki_filp
;
2911 struct address_space
*mapping
= file
->f_mapping
;
2912 struct inode
*inode
= mapping
->host
;
2913 loff_t pos
= iocb
->ki_pos
;
2918 write_len
= iov_iter_count(from
);
2919 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
2921 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2922 /* If there are pages to writeback, return */
2923 if (filemap_range_has_page(inode
->i_mapping
, pos
,
2927 written
= filemap_write_and_wait_range(mapping
, pos
,
2928 pos
+ write_len
- 1);
2934 * After a write we want buffered reads to be sure to go to disk to get
2935 * the new data. We invalidate clean cached page from the region we're
2936 * about to write. We do this *before* the write so that we can return
2937 * without clobbering -EIOCBQUEUED from ->direct_IO().
2939 written
= invalidate_inode_pages2_range(mapping
,
2940 pos
>> PAGE_SHIFT
, end
);
2942 * If a page can not be invalidated, return 0 to fall back
2943 * to buffered write.
2946 if (written
== -EBUSY
)
2951 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
2954 * Finally, try again to invalidate clean pages which might have been
2955 * cached by non-direct readahead, or faulted in by get_user_pages()
2956 * if the source of the write was an mmap'ed region of the file
2957 * we're writing. Either one is a pretty crazy thing to do,
2958 * so we don't support it 100%. If this invalidation
2959 * fails, tough, the write still worked...
2961 * Most of the time we do not need this since dio_complete() will do
2962 * the invalidation for us. However there are some file systems that
2963 * do not end up with dio_complete() being called, so let's not break
2964 * them by removing it completely
2966 if (mapping
->nrpages
)
2967 invalidate_inode_pages2_range(mapping
,
2968 pos
>> PAGE_SHIFT
, end
);
2972 write_len
-= written
;
2973 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
2974 i_size_write(inode
, pos
);
2975 mark_inode_dirty(inode
);
2979 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
2983 EXPORT_SYMBOL(generic_file_direct_write
);
2986 * Find or create a page at the given pagecache position. Return the locked
2987 * page. This function is specifically for buffered writes.
2989 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
2990 pgoff_t index
, unsigned flags
)
2993 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
2995 if (flags
& AOP_FLAG_NOFS
)
2996 fgp_flags
|= FGP_NOFS
;
2998 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
2999 mapping_gfp_mask(mapping
));
3001 wait_for_stable_page(page
);
3005 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3007 ssize_t
generic_perform_write(struct file
*file
,
3008 struct iov_iter
*i
, loff_t pos
)
3010 struct address_space
*mapping
= file
->f_mapping
;
3011 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3013 ssize_t written
= 0;
3014 unsigned int flags
= 0;
3018 unsigned long offset
; /* Offset into pagecache page */
3019 unsigned long bytes
; /* Bytes to write to page */
3020 size_t copied
; /* Bytes copied from user */
3023 offset
= (pos
& (PAGE_SIZE
- 1));
3024 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3029 * Bring in the user page that we will copy from _first_.
3030 * Otherwise there's a nasty deadlock on copying from the
3031 * same page as we're writing to, without it being marked
3034 * Not only is this an optimisation, but it is also required
3035 * to check that the address is actually valid, when atomic
3036 * usercopies are used, below.
3038 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3043 if (fatal_signal_pending(current
)) {
3048 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3050 if (unlikely(status
< 0))
3053 if (mapping_writably_mapped(mapping
))
3054 flush_dcache_page(page
);
3056 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3057 flush_dcache_page(page
);
3059 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3061 if (unlikely(status
< 0))
3067 iov_iter_advance(i
, copied
);
3068 if (unlikely(copied
== 0)) {
3070 * If we were unable to copy any data at all, we must
3071 * fall back to a single segment length write.
3073 * If we didn't fallback here, we could livelock
3074 * because not all segments in the iov can be copied at
3075 * once without a pagefault.
3077 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3078 iov_iter_single_seg_count(i
));
3084 balance_dirty_pages_ratelimited(mapping
);
3085 } while (iov_iter_count(i
));
3087 return written
? written
: status
;
3089 EXPORT_SYMBOL(generic_perform_write
);
3092 * __generic_file_write_iter - write data to a file
3093 * @iocb: IO state structure (file, offset, etc.)
3094 * @from: iov_iter with data to write
3096 * This function does all the work needed for actually writing data to a
3097 * file. It does all basic checks, removes SUID from the file, updates
3098 * modification times and calls proper subroutines depending on whether we
3099 * do direct IO or a standard buffered write.
3101 * It expects i_mutex to be grabbed unless we work on a block device or similar
3102 * object which does not need locking at all.
3104 * This function does *not* take care of syncing data in case of O_SYNC write.
3105 * A caller has to handle it. This is mainly due to the fact that we want to
3106 * avoid syncing under i_mutex.
3108 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3110 struct file
*file
= iocb
->ki_filp
;
3111 struct address_space
* mapping
= file
->f_mapping
;
3112 struct inode
*inode
= mapping
->host
;
3113 ssize_t written
= 0;
3117 /* We can write back this queue in page reclaim */
3118 current
->backing_dev_info
= inode_to_bdi(inode
);
3119 err
= file_remove_privs(file
);
3123 err
= file_update_time(file
);
3127 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3128 loff_t pos
, endbyte
;
3130 written
= generic_file_direct_write(iocb
, from
);
3132 * If the write stopped short of completing, fall back to
3133 * buffered writes. Some filesystems do this for writes to
3134 * holes, for example. For DAX files, a buffered write will
3135 * not succeed (even if it did, DAX does not handle dirty
3136 * page-cache pages correctly).
3138 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3141 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3143 * If generic_perform_write() returned a synchronous error
3144 * then we want to return the number of bytes which were
3145 * direct-written, or the error code if that was zero. Note
3146 * that this differs from normal direct-io semantics, which
3147 * will return -EFOO even if some bytes were written.
3149 if (unlikely(status
< 0)) {
3154 * We need to ensure that the page cache pages are written to
3155 * disk and invalidated to preserve the expected O_DIRECT
3158 endbyte
= pos
+ status
- 1;
3159 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3161 iocb
->ki_pos
= endbyte
+ 1;
3163 invalidate_mapping_pages(mapping
,
3165 endbyte
>> PAGE_SHIFT
);
3168 * We don't know how much we wrote, so just return
3169 * the number of bytes which were direct-written
3173 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3174 if (likely(written
> 0))
3175 iocb
->ki_pos
+= written
;
3178 current
->backing_dev_info
= NULL
;
3179 return written
? written
: err
;
3181 EXPORT_SYMBOL(__generic_file_write_iter
);
3184 * generic_file_write_iter - write data to a file
3185 * @iocb: IO state structure
3186 * @from: iov_iter with data to write
3188 * This is a wrapper around __generic_file_write_iter() to be used by most
3189 * filesystems. It takes care of syncing the file in case of O_SYNC file
3190 * and acquires i_mutex as needed.
3192 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3194 struct file
*file
= iocb
->ki_filp
;
3195 struct inode
*inode
= file
->f_mapping
->host
;
3199 ret
= generic_write_checks(iocb
, from
);
3201 ret
= __generic_file_write_iter(iocb
, from
);
3202 inode_unlock(inode
);
3205 ret
= generic_write_sync(iocb
, ret
);
3208 EXPORT_SYMBOL(generic_file_write_iter
);
3211 * try_to_release_page() - release old fs-specific metadata on a page
3213 * @page: the page which the kernel is trying to free
3214 * @gfp_mask: memory allocation flags (and I/O mode)
3216 * The address_space is to try to release any data against the page
3217 * (presumably at page->private). If the release was successful, return '1'.
3218 * Otherwise return zero.
3220 * This may also be called if PG_fscache is set on a page, indicating that the
3221 * page is known to the local caching routines.
3223 * The @gfp_mask argument specifies whether I/O may be performed to release
3224 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3227 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3229 struct address_space
* const mapping
= page
->mapping
;
3231 BUG_ON(!PageLocked(page
));
3232 if (PageWriteback(page
))
3235 if (mapping
&& mapping
->a_ops
->releasepage
)
3236 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3237 return try_to_free_buffers(page
);
3240 EXPORT_SYMBOL(try_to_release_page
);