1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1994-1999 Linus Torvalds
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
23 #include <linux/swap.h>
24 #include <linux/mman.h>
25 #include <linux/pagemap.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/error-injection.h>
29 #include <linux/hash.h>
30 #include <linux/writeback.h>
31 #include <linux/backing-dev.h>
32 #include <linux/pagevec.h>
33 #include <linux/blkdev.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/cleancache.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include <linux/page_idle.h>
45 #include <asm/pgalloc.h>
46 #include <asm/tlbflush.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/filemap.h>
53 * FIXME: remove all knowledge of the buffer layer from the core VM
55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
63 * Shared mappings now work. 15.8.1995 Bruno.
65 * finished 'unifying' the page and buffer cache and SMP-threaded the
66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
74 * ->i_mmap_rwsem (truncate_pagecache)
75 * ->private_lock (__free_pte->__set_page_dirty_buffers)
76 * ->swap_lock (exclusive_swap_page, others)
80 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
84 * ->page_table_lock or pte_lock (various, mainly in memory.c)
85 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
88 * ->lock_page (access_process_vm)
90 * ->i_mutex (generic_perform_write)
91 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
94 * sb_lock (fs/fs-writeback.c)
95 * ->i_pages lock (__sync_single_inode)
98 * ->anon_vma.lock (vma_adjust)
101 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
103 * ->page_table_lock or pte_lock
104 * ->swap_lock (try_to_unmap_one)
105 * ->private_lock (try_to_unmap_one)
106 * ->i_pages lock (try_to_unmap_one)
107 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
108 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
109 * ->private_lock (page_remove_rmap->set_page_dirty)
110 * ->i_pages lock (page_remove_rmap->set_page_dirty)
111 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
112 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
113 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
114 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
115 * ->inode->i_lock (zap_pte_range->set_page_dirty)
116 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
119 * ->tasklist_lock (memory_failure, collect_procs_ao)
122 static void page_cache_delete(struct address_space
*mapping
,
123 struct page
*page
, void *shadow
)
125 XA_STATE(xas
, &mapping
->i_pages
, page
->index
);
128 mapping_set_update(&xas
, mapping
);
130 /* hugetlb pages are represented by a single entry in the xarray */
131 if (!PageHuge(page
)) {
132 xas_set_order(&xas
, page
->index
, compound_order(page
));
133 nr
= compound_nr(page
);
136 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
137 VM_BUG_ON_PAGE(PageTail(page
), page
);
138 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
140 xas_store(&xas
, shadow
);
141 xas_init_marks(&xas
);
143 page
->mapping
= NULL
;
144 /* Leave page->index set: truncation lookup relies upon it */
145 mapping
->nrpages
-= nr
;
148 static void unaccount_page_cache_page(struct address_space
*mapping
,
154 * if we're uptodate, flush out into the cleancache, otherwise
155 * invalidate any existing cleancache entries. We can't leave
156 * stale data around in the cleancache once our page is gone
158 if (PageUptodate(page
) && PageMappedToDisk(page
))
159 cleancache_put_page(page
);
161 cleancache_invalidate_page(mapping
, page
);
163 VM_BUG_ON_PAGE(PageTail(page
), page
);
164 VM_BUG_ON_PAGE(page_mapped(page
), page
);
165 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
168 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
169 current
->comm
, page_to_pfn(page
));
170 dump_page(page
, "still mapped when deleted");
172 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
174 mapcount
= page_mapcount(page
);
175 if (mapping_exiting(mapping
) &&
176 page_count(page
) >= mapcount
+ 2) {
178 * All vmas have already been torn down, so it's
179 * a good bet that actually the page is unmapped,
180 * and we'd prefer not to leak it: if we're wrong,
181 * some other bad page check should catch it later.
183 page_mapcount_reset(page
);
184 page_ref_sub(page
, mapcount
);
188 /* hugetlb pages do not participate in page cache accounting. */
192 nr
= thp_nr_pages(page
);
194 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, -nr
);
195 if (PageSwapBacked(page
)) {
196 __mod_lruvec_page_state(page
, NR_SHMEM
, -nr
);
197 if (PageTransHuge(page
))
198 __mod_lruvec_page_state(page
, NR_SHMEM_THPS
, -nr
);
199 } else if (PageTransHuge(page
)) {
200 __mod_lruvec_page_state(page
, NR_FILE_THPS
, -nr
);
201 filemap_nr_thps_dec(mapping
);
205 * At this point page must be either written or cleaned by
206 * truncate. Dirty page here signals a bug and loss of
209 * This fixes dirty accounting after removing the page entirely
210 * but leaves PageDirty set: it has no effect for truncated
211 * page and anyway will be cleared before returning page into
214 if (WARN_ON_ONCE(PageDirty(page
)))
215 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
219 * Delete a page from the page cache and free it. Caller has to make
220 * sure the page is locked and that nobody else uses it - or that usage
221 * is safe. The caller must hold the i_pages lock.
223 void __delete_from_page_cache(struct page
*page
, void *shadow
)
225 struct address_space
*mapping
= page
->mapping
;
227 trace_mm_filemap_delete_from_page_cache(page
);
229 unaccount_page_cache_page(mapping
, page
);
230 page_cache_delete(mapping
, page
, shadow
);
233 static void page_cache_free_page(struct address_space
*mapping
,
236 void (*freepage
)(struct page
*);
238 freepage
= mapping
->a_ops
->freepage
;
242 if (PageTransHuge(page
) && !PageHuge(page
)) {
243 page_ref_sub(page
, thp_nr_pages(page
));
244 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
251 * delete_from_page_cache - delete page from page cache
252 * @page: the page which the kernel is trying to remove from page cache
254 * This must be called only on pages that have been verified to be in the page
255 * cache and locked. It will never put the page into the free list, the caller
256 * has a reference on the page.
258 void delete_from_page_cache(struct page
*page
)
260 struct address_space
*mapping
= page_mapping(page
);
263 BUG_ON(!PageLocked(page
));
264 xa_lock_irqsave(&mapping
->i_pages
, flags
);
265 __delete_from_page_cache(page
, NULL
);
266 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
268 page_cache_free_page(mapping
, page
);
270 EXPORT_SYMBOL(delete_from_page_cache
);
273 * page_cache_delete_batch - delete several pages from page cache
274 * @mapping: the mapping to which pages belong
275 * @pvec: pagevec with pages to delete
277 * The function walks over mapping->i_pages and removes pages passed in @pvec
278 * from the mapping. The function expects @pvec to be sorted by page index
279 * and is optimised for it to be dense.
280 * It tolerates holes in @pvec (mapping entries at those indices are not
281 * modified). The function expects only THP head pages to be present in the
284 * The function expects the i_pages lock to be held.
286 static void page_cache_delete_batch(struct address_space
*mapping
,
287 struct pagevec
*pvec
)
289 XA_STATE(xas
, &mapping
->i_pages
, pvec
->pages
[0]->index
);
294 mapping_set_update(&xas
, mapping
);
295 xas_for_each(&xas
, page
, ULONG_MAX
) {
296 if (i
>= pagevec_count(pvec
))
299 /* A swap/dax/shadow entry got inserted? Skip it. */
300 if (xa_is_value(page
))
303 * A page got inserted in our range? Skip it. We have our
304 * pages locked so they are protected from being removed.
305 * If we see a page whose index is higher than ours, it
306 * means our page has been removed, which shouldn't be
307 * possible because we're holding the PageLock.
309 if (page
!= pvec
->pages
[i
]) {
310 VM_BUG_ON_PAGE(page
->index
> pvec
->pages
[i
]->index
,
315 WARN_ON_ONCE(!PageLocked(page
));
317 if (page
->index
== xas
.xa_index
)
318 page
->mapping
= NULL
;
319 /* Leave page->index set: truncation lookup relies on it */
322 * Move to the next page in the vector if this is a regular
323 * page or the index is of the last sub-page of this compound
326 if (page
->index
+ compound_nr(page
) - 1 == xas
.xa_index
)
328 xas_store(&xas
, NULL
);
331 mapping
->nrpages
-= total_pages
;
334 void delete_from_page_cache_batch(struct address_space
*mapping
,
335 struct pagevec
*pvec
)
340 if (!pagevec_count(pvec
))
343 xa_lock_irqsave(&mapping
->i_pages
, flags
);
344 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
345 trace_mm_filemap_delete_from_page_cache(pvec
->pages
[i
]);
347 unaccount_page_cache_page(mapping
, pvec
->pages
[i
]);
349 page_cache_delete_batch(mapping
, pvec
);
350 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
352 for (i
= 0; i
< pagevec_count(pvec
); i
++)
353 page_cache_free_page(mapping
, pvec
->pages
[i
]);
356 int filemap_check_errors(struct address_space
*mapping
)
359 /* Check for outstanding write errors */
360 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
361 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
363 if (test_bit(AS_EIO
, &mapping
->flags
) &&
364 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
368 EXPORT_SYMBOL(filemap_check_errors
);
370 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
372 /* Check for outstanding write errors */
373 if (test_bit(AS_EIO
, &mapping
->flags
))
375 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
381 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
382 * @mapping: address space structure to write
383 * @start: offset in bytes where the range starts
384 * @end: offset in bytes where the range ends (inclusive)
385 * @sync_mode: enable synchronous operation
387 * Start writeback against all of a mapping's dirty pages that lie
388 * within the byte offsets <start, end> inclusive.
390 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
391 * opposed to a regular memory cleansing writeback. The difference between
392 * these two operations is that if a dirty page/buffer is encountered, it must
393 * be waited upon, and not just skipped over.
395 * Return: %0 on success, negative error code otherwise.
397 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
398 loff_t end
, int sync_mode
)
401 struct writeback_control wbc
= {
402 .sync_mode
= sync_mode
,
403 .nr_to_write
= LONG_MAX
,
404 .range_start
= start
,
408 if (!mapping_can_writeback(mapping
) ||
409 !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
412 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
413 ret
= do_writepages(mapping
, &wbc
);
414 wbc_detach_inode(&wbc
);
418 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
421 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
424 int filemap_fdatawrite(struct address_space
*mapping
)
426 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
428 EXPORT_SYMBOL(filemap_fdatawrite
);
430 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
433 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
435 EXPORT_SYMBOL(filemap_fdatawrite_range
);
438 * filemap_flush - mostly a non-blocking flush
439 * @mapping: target address_space
441 * This is a mostly non-blocking flush. Not suitable for data-integrity
442 * purposes - I/O may not be started against all dirty pages.
444 * Return: %0 on success, negative error code otherwise.
446 int filemap_flush(struct address_space
*mapping
)
448 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
450 EXPORT_SYMBOL(filemap_flush
);
453 * filemap_range_has_page - check if a page exists in range.
454 * @mapping: address space within which to check
455 * @start_byte: offset in bytes where the range starts
456 * @end_byte: offset in bytes where the range ends (inclusive)
458 * Find at least one page in the range supplied, usually used to check if
459 * direct writing in this range will trigger a writeback.
461 * Return: %true if at least one page exists in the specified range,
464 bool filemap_range_has_page(struct address_space
*mapping
,
465 loff_t start_byte
, loff_t end_byte
)
468 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
469 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
471 if (end_byte
< start_byte
)
476 page
= xas_find(&xas
, max
);
477 if (xas_retry(&xas
, page
))
479 /* Shadow entries don't count */
480 if (xa_is_value(page
))
483 * We don't need to try to pin this page; we're about to
484 * release the RCU lock anyway. It is enough to know that
485 * there was a page here recently.
493 EXPORT_SYMBOL(filemap_range_has_page
);
495 static void __filemap_fdatawait_range(struct address_space
*mapping
,
496 loff_t start_byte
, loff_t end_byte
)
498 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
499 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
503 if (end_byte
< start_byte
)
507 while (index
<= end
) {
510 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
,
511 end
, PAGECACHE_TAG_WRITEBACK
);
515 for (i
= 0; i
< nr_pages
; i
++) {
516 struct page
*page
= pvec
.pages
[i
];
518 wait_on_page_writeback(page
);
519 ClearPageError(page
);
521 pagevec_release(&pvec
);
527 * filemap_fdatawait_range - wait for writeback to complete
528 * @mapping: address space structure to wait for
529 * @start_byte: offset in bytes where the range starts
530 * @end_byte: offset in bytes where the range ends (inclusive)
532 * Walk the list of under-writeback pages of the given address space
533 * in the given range and wait for all of them. Check error status of
534 * the address space and return it.
536 * Since the error status of the address space is cleared by this function,
537 * callers are responsible for checking the return value and handling and/or
538 * reporting the error.
540 * Return: error status of the address space.
542 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
545 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
546 return filemap_check_errors(mapping
);
548 EXPORT_SYMBOL(filemap_fdatawait_range
);
551 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
552 * @mapping: address space structure to wait for
553 * @start_byte: offset in bytes where the range starts
554 * @end_byte: offset in bytes where the range ends (inclusive)
556 * Walk the list of under-writeback pages of the given address space in the
557 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
558 * this function does not clear error status of the address space.
560 * Use this function if callers don't handle errors themselves. Expected
561 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
564 int filemap_fdatawait_range_keep_errors(struct address_space
*mapping
,
565 loff_t start_byte
, loff_t end_byte
)
567 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
568 return filemap_check_and_keep_errors(mapping
);
570 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors
);
573 * file_fdatawait_range - wait for writeback to complete
574 * @file: file pointing to address space structure to wait for
575 * @start_byte: offset in bytes where the range starts
576 * @end_byte: offset in bytes where the range ends (inclusive)
578 * Walk the list of under-writeback pages of the address space that file
579 * refers to, in the given range and wait for all of them. Check error
580 * status of the address space vs. the file->f_wb_err cursor and return it.
582 * Since the error status of the file is advanced by this function,
583 * callers are responsible for checking the return value and handling and/or
584 * reporting the error.
586 * Return: error status of the address space vs. the file->f_wb_err cursor.
588 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
590 struct address_space
*mapping
= file
->f_mapping
;
592 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
593 return file_check_and_advance_wb_err(file
);
595 EXPORT_SYMBOL(file_fdatawait_range
);
598 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
599 * @mapping: address space structure to wait for
601 * Walk the list of under-writeback pages of the given address space
602 * and wait for all of them. Unlike filemap_fdatawait(), this function
603 * does not clear error status of the address space.
605 * Use this function if callers don't handle errors themselves. Expected
606 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
609 * Return: error status of the address space.
611 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
613 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
614 return filemap_check_and_keep_errors(mapping
);
616 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
618 /* Returns true if writeback might be needed or already in progress. */
619 static bool mapping_needs_writeback(struct address_space
*mapping
)
621 return mapping
->nrpages
;
625 * filemap_range_needs_writeback - check if range potentially needs writeback
626 * @mapping: address space within which to check
627 * @start_byte: offset in bytes where the range starts
628 * @end_byte: offset in bytes where the range ends (inclusive)
630 * Find at least one page in the range supplied, usually used to check if
631 * direct writing in this range will trigger a writeback. Used by O_DIRECT
632 * read/write with IOCB_NOWAIT, to see if the caller needs to do
633 * filemap_write_and_wait_range() before proceeding.
635 * Return: %true if the caller should do filemap_write_and_wait_range() before
636 * doing O_DIRECT to a page in this range, %false otherwise.
638 bool filemap_range_needs_writeback(struct address_space
*mapping
,
639 loff_t start_byte
, loff_t end_byte
)
641 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
642 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
645 if (!mapping_needs_writeback(mapping
))
647 if (!mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
648 !mapping_tagged(mapping
, PAGECACHE_TAG_WRITEBACK
))
650 if (end_byte
< start_byte
)
654 xas_for_each(&xas
, page
, max
) {
655 if (xas_retry(&xas
, page
))
657 if (xa_is_value(page
))
659 if (PageDirty(page
) || PageLocked(page
) || PageWriteback(page
))
665 EXPORT_SYMBOL_GPL(filemap_range_needs_writeback
);
668 * filemap_write_and_wait_range - write out & wait on a file range
669 * @mapping: the address_space for the pages
670 * @lstart: offset in bytes where the range starts
671 * @lend: offset in bytes where the range ends (inclusive)
673 * Write out and wait upon file offsets lstart->lend, inclusive.
675 * Note that @lend is inclusive (describes the last byte to be written) so
676 * that this function can be used to write to the very end-of-file (end = -1).
678 * Return: error status of the address space.
680 int filemap_write_and_wait_range(struct address_space
*mapping
,
681 loff_t lstart
, loff_t lend
)
685 if (mapping_needs_writeback(mapping
)) {
686 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
689 * Even if the above returned error, the pages may be
690 * written partially (e.g. -ENOSPC), so we wait for it.
691 * But the -EIO is special case, it may indicate the worst
692 * thing (e.g. bug) happened, so we avoid waiting for it.
695 int err2
= filemap_fdatawait_range(mapping
,
700 /* Clear any previously stored errors */
701 filemap_check_errors(mapping
);
704 err
= filemap_check_errors(mapping
);
708 EXPORT_SYMBOL(filemap_write_and_wait_range
);
710 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
712 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
714 trace_filemap_set_wb_err(mapping
, eseq
);
716 EXPORT_SYMBOL(__filemap_set_wb_err
);
719 * file_check_and_advance_wb_err - report wb error (if any) that was previously
720 * and advance wb_err to current one
721 * @file: struct file on which the error is being reported
723 * When userland calls fsync (or something like nfsd does the equivalent), we
724 * want to report any writeback errors that occurred since the last fsync (or
725 * since the file was opened if there haven't been any).
727 * Grab the wb_err from the mapping. If it matches what we have in the file,
728 * then just quickly return 0. The file is all caught up.
730 * If it doesn't match, then take the mapping value, set the "seen" flag in
731 * it and try to swap it into place. If it works, or another task beat us
732 * to it with the new value, then update the f_wb_err and return the error
733 * portion. The error at this point must be reported via proper channels
734 * (a'la fsync, or NFS COMMIT operation, etc.).
736 * While we handle mapping->wb_err with atomic operations, the f_wb_err
737 * value is protected by the f_lock since we must ensure that it reflects
738 * the latest value swapped in for this file descriptor.
740 * Return: %0 on success, negative error code otherwise.
742 int file_check_and_advance_wb_err(struct file
*file
)
745 errseq_t old
= READ_ONCE(file
->f_wb_err
);
746 struct address_space
*mapping
= file
->f_mapping
;
748 /* Locklessly handle the common case where nothing has changed */
749 if (errseq_check(&mapping
->wb_err
, old
)) {
750 /* Something changed, must use slow path */
751 spin_lock(&file
->f_lock
);
752 old
= file
->f_wb_err
;
753 err
= errseq_check_and_advance(&mapping
->wb_err
,
755 trace_file_check_and_advance_wb_err(file
, old
);
756 spin_unlock(&file
->f_lock
);
760 * We're mostly using this function as a drop in replacement for
761 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
762 * that the legacy code would have had on these flags.
764 clear_bit(AS_EIO
, &mapping
->flags
);
765 clear_bit(AS_ENOSPC
, &mapping
->flags
);
768 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
771 * file_write_and_wait_range - write out & wait on a file range
772 * @file: file pointing to address_space with pages
773 * @lstart: offset in bytes where the range starts
774 * @lend: offset in bytes where the range ends (inclusive)
776 * Write out and wait upon file offsets lstart->lend, inclusive.
778 * Note that @lend is inclusive (describes the last byte to be written) so
779 * that this function can be used to write to the very end-of-file (end = -1).
781 * After writing out and waiting on the data, we check and advance the
782 * f_wb_err cursor to the latest value, and return any errors detected there.
784 * Return: %0 on success, negative error code otherwise.
786 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
789 struct address_space
*mapping
= file
->f_mapping
;
791 if (mapping_needs_writeback(mapping
)) {
792 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
794 /* See comment of filemap_write_and_wait() */
796 __filemap_fdatawait_range(mapping
, lstart
, lend
);
798 err2
= file_check_and_advance_wb_err(file
);
803 EXPORT_SYMBOL(file_write_and_wait_range
);
806 * replace_page_cache_page - replace a pagecache page with a new one
807 * @old: page to be replaced
808 * @new: page to replace with
810 * This function replaces a page in the pagecache with a new one. On
811 * success it acquires the pagecache reference for the new page and
812 * drops it for the old page. Both the old and new pages must be
813 * locked. This function does not add the new page to the LRU, the
814 * caller must do that.
816 * The remove + add is atomic. This function cannot fail.
818 void replace_page_cache_page(struct page
*old
, struct page
*new)
820 struct address_space
*mapping
= old
->mapping
;
821 void (*freepage
)(struct page
*) = mapping
->a_ops
->freepage
;
822 pgoff_t offset
= old
->index
;
823 XA_STATE(xas
, &mapping
->i_pages
, offset
);
826 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
827 VM_BUG_ON_PAGE(!PageLocked(new), new);
828 VM_BUG_ON_PAGE(new->mapping
, new);
831 new->mapping
= mapping
;
834 mem_cgroup_migrate(old
, new);
836 xas_lock_irqsave(&xas
, flags
);
837 xas_store(&xas
, new);
840 /* hugetlb pages do not participate in page cache accounting. */
842 __dec_lruvec_page_state(old
, NR_FILE_PAGES
);
844 __inc_lruvec_page_state(new, NR_FILE_PAGES
);
845 if (PageSwapBacked(old
))
846 __dec_lruvec_page_state(old
, NR_SHMEM
);
847 if (PageSwapBacked(new))
848 __inc_lruvec_page_state(new, NR_SHMEM
);
849 xas_unlock_irqrestore(&xas
, flags
);
854 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
856 noinline
int __add_to_page_cache_locked(struct page
*page
,
857 struct address_space
*mapping
,
858 pgoff_t offset
, gfp_t gfp
,
861 XA_STATE(xas
, &mapping
->i_pages
, offset
);
862 int huge
= PageHuge(page
);
864 bool charged
= false;
866 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
867 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
868 mapping_set_update(&xas
, mapping
);
871 page
->mapping
= mapping
;
872 page
->index
= offset
;
875 error
= mem_cgroup_charge(page
, current
->mm
, gfp
);
881 gfp
&= GFP_RECLAIM_MASK
;
884 unsigned int order
= xa_get_order(xas
.xa
, xas
.xa_index
);
885 void *entry
, *old
= NULL
;
887 if (order
> thp_order(page
))
888 xas_split_alloc(&xas
, xa_load(xas
.xa
, xas
.xa_index
),
891 xas_for_each_conflict(&xas
, entry
) {
893 if (!xa_is_value(entry
)) {
894 xas_set_err(&xas
, -EEXIST
);
902 /* entry may have been split before we acquired lock */
903 order
= xa_get_order(xas
.xa
, xas
.xa_index
);
904 if (order
> thp_order(page
)) {
905 xas_split(&xas
, old
, order
);
910 xas_store(&xas
, page
);
916 /* hugetlb pages do not participate in page cache accounting */
918 __inc_lruvec_page_state(page
, NR_FILE_PAGES
);
920 xas_unlock_irq(&xas
);
921 } while (xas_nomem(&xas
, gfp
));
923 if (xas_error(&xas
)) {
924 error
= xas_error(&xas
);
926 mem_cgroup_uncharge(page
);
930 trace_mm_filemap_add_to_page_cache(page
);
933 page
->mapping
= NULL
;
934 /* Leave page->index set: truncation relies upon it */
938 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked
, ERRNO
);
941 * add_to_page_cache_locked - add a locked page to the pagecache
943 * @mapping: the page's address_space
944 * @offset: page index
945 * @gfp_mask: page allocation mode
947 * This function is used to add a page to the pagecache. It must be locked.
948 * This function does not add the page to the LRU. The caller must do that.
950 * Return: %0 on success, negative error code otherwise.
952 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
953 pgoff_t offset
, gfp_t gfp_mask
)
955 return __add_to_page_cache_locked(page
, mapping
, offset
,
958 EXPORT_SYMBOL(add_to_page_cache_locked
);
960 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
961 pgoff_t offset
, gfp_t gfp_mask
)
966 __SetPageLocked(page
);
967 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
970 __ClearPageLocked(page
);
973 * The page might have been evicted from cache only
974 * recently, in which case it should be activated like
975 * any other repeatedly accessed page.
976 * The exception is pages getting rewritten; evicting other
977 * data from the working set, only to cache data that will
978 * get overwritten with something else, is a waste of memory.
980 WARN_ON_ONCE(PageActive(page
));
981 if (!(gfp_mask
& __GFP_WRITE
) && shadow
)
982 workingset_refault(page
, shadow
);
987 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
990 struct page
*__page_cache_alloc(gfp_t gfp
)
995 if (cpuset_do_page_mem_spread()) {
996 unsigned int cpuset_mems_cookie
;
998 cpuset_mems_cookie
= read_mems_allowed_begin();
999 n
= cpuset_mem_spread_node();
1000 page
= __alloc_pages_node(n
, gfp
, 0);
1001 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
1005 return alloc_pages(gfp
, 0);
1007 EXPORT_SYMBOL(__page_cache_alloc
);
1011 * In order to wait for pages to become available there must be
1012 * waitqueues associated with pages. By using a hash table of
1013 * waitqueues where the bucket discipline is to maintain all
1014 * waiters on the same queue and wake all when any of the pages
1015 * become available, and for the woken contexts to check to be
1016 * sure the appropriate page became available, this saves space
1017 * at a cost of "thundering herd" phenomena during rare hash
1020 #define PAGE_WAIT_TABLE_BITS 8
1021 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1022 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
1024 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
1026 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
1029 void __init
pagecache_init(void)
1033 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
1034 init_waitqueue_head(&page_wait_table
[i
]);
1036 page_writeback_init();
1040 * The page wait code treats the "wait->flags" somewhat unusually, because
1041 * we have multiple different kinds of waits, not just the usual "exclusive"
1046 * (a) no special bits set:
1048 * We're just waiting for the bit to be released, and when a waker
1049 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1050 * and remove it from the wait queue.
1052 * Simple and straightforward.
1054 * (b) WQ_FLAG_EXCLUSIVE:
1056 * The waiter is waiting to get the lock, and only one waiter should
1057 * be woken up to avoid any thundering herd behavior. We'll set the
1058 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1060 * This is the traditional exclusive wait.
1062 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1064 * The waiter is waiting to get the bit, and additionally wants the
1065 * lock to be transferred to it for fair lock behavior. If the lock
1066 * cannot be taken, we stop walking the wait queue without waking
1069 * This is the "fair lock handoff" case, and in addition to setting
1070 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1071 * that it now has the lock.
1073 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
1076 struct wait_page_key
*key
= arg
;
1077 struct wait_page_queue
*wait_page
1078 = container_of(wait
, struct wait_page_queue
, wait
);
1080 if (!wake_page_match(wait_page
, key
))
1084 * If it's a lock handoff wait, we get the bit for it, and
1085 * stop walking (and do not wake it up) if we can't.
1087 flags
= wait
->flags
;
1088 if (flags
& WQ_FLAG_EXCLUSIVE
) {
1089 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
1091 if (flags
& WQ_FLAG_CUSTOM
) {
1092 if (test_and_set_bit(key
->bit_nr
, &key
->page
->flags
))
1094 flags
|= WQ_FLAG_DONE
;
1099 * We are holding the wait-queue lock, but the waiter that
1100 * is waiting for this will be checking the flags without
1103 * So update the flags atomically, and wake up the waiter
1104 * afterwards to avoid any races. This store-release pairs
1105 * with the load-acquire in wait_on_page_bit_common().
1107 smp_store_release(&wait
->flags
, flags
| WQ_FLAG_WOKEN
);
1108 wake_up_state(wait
->private, mode
);
1111 * Ok, we have successfully done what we're waiting for,
1112 * and we can unconditionally remove the wait entry.
1114 * Note that this pairs with the "finish_wait()" in the
1115 * waiter, and has to be the absolute last thing we do.
1116 * After this list_del_init(&wait->entry) the wait entry
1117 * might be de-allocated and the process might even have
1120 list_del_init_careful(&wait
->entry
);
1121 return (flags
& WQ_FLAG_EXCLUSIVE
) != 0;
1124 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
1126 wait_queue_head_t
*q
= page_waitqueue(page
);
1127 struct wait_page_key key
;
1128 unsigned long flags
;
1129 wait_queue_entry_t bookmark
;
1132 key
.bit_nr
= bit_nr
;
1136 bookmark
.private = NULL
;
1137 bookmark
.func
= NULL
;
1138 INIT_LIST_HEAD(&bookmark
.entry
);
1140 spin_lock_irqsave(&q
->lock
, flags
);
1141 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1143 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1145 * Take a breather from holding the lock,
1146 * allow pages that finish wake up asynchronously
1147 * to acquire the lock and remove themselves
1150 spin_unlock_irqrestore(&q
->lock
, flags
);
1152 spin_lock_irqsave(&q
->lock
, flags
);
1153 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1157 * It is possible for other pages to have collided on the waitqueue
1158 * hash, so in that case check for a page match. That prevents a long-
1161 * It is still possible to miss a case here, when we woke page waiters
1162 * and removed them from the waitqueue, but there are still other
1165 if (!waitqueue_active(q
) || !key
.page_match
) {
1166 ClearPageWaiters(page
);
1168 * It's possible to miss clearing Waiters here, when we woke
1169 * our page waiters, but the hashed waitqueue has waiters for
1170 * other pages on it.
1172 * That's okay, it's a rare case. The next waker will clear it.
1175 spin_unlock_irqrestore(&q
->lock
, flags
);
1178 static void wake_up_page(struct page
*page
, int bit
)
1180 if (!PageWaiters(page
))
1182 wake_up_page_bit(page
, bit
);
1186 * A choice of three behaviors for wait_on_page_bit_common():
1189 EXCLUSIVE
, /* Hold ref to page and take the bit when woken, like
1190 * __lock_page() waiting on then setting PG_locked.
1192 SHARED
, /* Hold ref to page and check the bit when woken, like
1193 * wait_on_page_writeback() waiting on PG_writeback.
1195 DROP
, /* Drop ref to page before wait, no check when woken,
1196 * like put_and_wait_on_page_locked() on PG_locked.
1201 * Attempt to check (or get) the page bit, and mark us done
1204 static inline bool trylock_page_bit_common(struct page
*page
, int bit_nr
,
1205 struct wait_queue_entry
*wait
)
1207 if (wait
->flags
& WQ_FLAG_EXCLUSIVE
) {
1208 if (test_and_set_bit(bit_nr
, &page
->flags
))
1210 } else if (test_bit(bit_nr
, &page
->flags
))
1213 wait
->flags
|= WQ_FLAG_WOKEN
| WQ_FLAG_DONE
;
1217 /* How many times do we accept lock stealing from under a waiter? */
1218 int sysctl_page_lock_unfairness
= 5;
1220 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
1221 struct page
*page
, int bit_nr
, int state
, enum behavior behavior
)
1223 int unfairness
= sysctl_page_lock_unfairness
;
1224 struct wait_page_queue wait_page
;
1225 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1226 bool thrashing
= false;
1227 bool delayacct
= false;
1228 unsigned long pflags
;
1230 if (bit_nr
== PG_locked
&&
1231 !PageUptodate(page
) && PageWorkingset(page
)) {
1232 if (!PageSwapBacked(page
)) {
1233 delayacct_thrashing_start();
1236 psi_memstall_enter(&pflags
);
1241 wait
->func
= wake_page_function
;
1242 wait_page
.page
= page
;
1243 wait_page
.bit_nr
= bit_nr
;
1247 if (behavior
== EXCLUSIVE
) {
1248 wait
->flags
= WQ_FLAG_EXCLUSIVE
;
1249 if (--unfairness
< 0)
1250 wait
->flags
|= WQ_FLAG_CUSTOM
;
1254 * Do one last check whether we can get the
1255 * page bit synchronously.
1257 * Do the SetPageWaiters() marking before that
1258 * to let any waker we _just_ missed know they
1259 * need to wake us up (otherwise they'll never
1260 * even go to the slow case that looks at the
1261 * page queue), and add ourselves to the wait
1262 * queue if we need to sleep.
1264 * This part needs to be done under the queue
1265 * lock to avoid races.
1267 spin_lock_irq(&q
->lock
);
1268 SetPageWaiters(page
);
1269 if (!trylock_page_bit_common(page
, bit_nr
, wait
))
1270 __add_wait_queue_entry_tail(q
, wait
);
1271 spin_unlock_irq(&q
->lock
);
1274 * From now on, all the logic will be based on
1275 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1276 * see whether the page bit testing has already
1277 * been done by the wake function.
1279 * We can drop our reference to the page.
1281 if (behavior
== DROP
)
1285 * Note that until the "finish_wait()", or until
1286 * we see the WQ_FLAG_WOKEN flag, we need to
1287 * be very careful with the 'wait->flags', because
1288 * we may race with a waker that sets them.
1293 set_current_state(state
);
1295 /* Loop until we've been woken or interrupted */
1296 flags
= smp_load_acquire(&wait
->flags
);
1297 if (!(flags
& WQ_FLAG_WOKEN
)) {
1298 if (signal_pending_state(state
, current
))
1305 /* If we were non-exclusive, we're done */
1306 if (behavior
!= EXCLUSIVE
)
1309 /* If the waker got the lock for us, we're done */
1310 if (flags
& WQ_FLAG_DONE
)
1314 * Otherwise, if we're getting the lock, we need to
1315 * try to get it ourselves.
1317 * And if that fails, we'll have to retry this all.
1319 if (unlikely(test_and_set_bit(bit_nr
, &page
->flags
)))
1322 wait
->flags
|= WQ_FLAG_DONE
;
1327 * If a signal happened, this 'finish_wait()' may remove the last
1328 * waiter from the wait-queues, but the PageWaiters bit will remain
1329 * set. That's ok. The next wakeup will take care of it, and trying
1330 * to do it here would be difficult and prone to races.
1332 finish_wait(q
, wait
);
1336 delayacct_thrashing_end();
1337 psi_memstall_leave(&pflags
);
1341 * NOTE! The wait->flags weren't stable until we've done the
1342 * 'finish_wait()', and we could have exited the loop above due
1343 * to a signal, and had a wakeup event happen after the signal
1344 * test but before the 'finish_wait()'.
1346 * So only after the finish_wait() can we reliably determine
1347 * if we got woken up or not, so we can now figure out the final
1348 * return value based on that state without races.
1350 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1351 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1353 if (behavior
== EXCLUSIVE
)
1354 return wait
->flags
& WQ_FLAG_DONE
? 0 : -EINTR
;
1356 return wait
->flags
& WQ_FLAG_WOKEN
? 0 : -EINTR
;
1359 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1361 wait_queue_head_t
*q
= page_waitqueue(page
);
1362 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, SHARED
);
1364 EXPORT_SYMBOL(wait_on_page_bit
);
1366 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1368 wait_queue_head_t
*q
= page_waitqueue(page
);
1369 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, SHARED
);
1371 EXPORT_SYMBOL(wait_on_page_bit_killable
);
1374 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1375 * @page: The page to wait for.
1376 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1378 * The caller should hold a reference on @page. They expect the page to
1379 * become unlocked relatively soon, but do not wish to hold up migration
1380 * (for example) by holding the reference while waiting for the page to
1381 * come unlocked. After this function returns, the caller should not
1382 * dereference @page.
1384 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1386 int put_and_wait_on_page_locked(struct page
*page
, int state
)
1388 wait_queue_head_t
*q
;
1390 page
= compound_head(page
);
1391 q
= page_waitqueue(page
);
1392 return wait_on_page_bit_common(q
, page
, PG_locked
, state
, DROP
);
1396 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1397 * @page: Page defining the wait queue of interest
1398 * @waiter: Waiter to add to the queue
1400 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1402 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1404 wait_queue_head_t
*q
= page_waitqueue(page
);
1405 unsigned long flags
;
1407 spin_lock_irqsave(&q
->lock
, flags
);
1408 __add_wait_queue_entry_tail(q
, waiter
);
1409 SetPageWaiters(page
);
1410 spin_unlock_irqrestore(&q
->lock
, flags
);
1412 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1414 #ifndef clear_bit_unlock_is_negative_byte
1417 * PG_waiters is the high bit in the same byte as PG_lock.
1419 * On x86 (and on many other architectures), we can clear PG_lock and
1420 * test the sign bit at the same time. But if the architecture does
1421 * not support that special operation, we just do this all by hand
1424 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1425 * being cleared, but a memory barrier should be unnecessary since it is
1426 * in the same byte as PG_locked.
1428 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1430 clear_bit_unlock(nr
, mem
);
1431 /* smp_mb__after_atomic(); */
1432 return test_bit(PG_waiters
, mem
);
1438 * unlock_page - unlock a locked page
1441 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1442 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1443 * mechanism between PageLocked pages and PageWriteback pages is shared.
1444 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1446 * Note that this depends on PG_waiters being the sign bit in the byte
1447 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1448 * clear the PG_locked bit and test PG_waiters at the same time fairly
1449 * portably (architectures that do LL/SC can test any bit, while x86 can
1450 * test the sign bit).
1452 void unlock_page(struct page
*page
)
1454 BUILD_BUG_ON(PG_waiters
!= 7);
1455 page
= compound_head(page
);
1456 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1457 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1458 wake_up_page_bit(page
, PG_locked
);
1460 EXPORT_SYMBOL(unlock_page
);
1463 * end_page_private_2 - Clear PG_private_2 and release any waiters
1466 * Clear the PG_private_2 bit on a page and wake up any sleepers waiting for
1467 * this. The page ref held for PG_private_2 being set is released.
1469 * This is, for example, used when a netfs page is being written to a local
1470 * disk cache, thereby allowing writes to the cache for the same page to be
1473 void end_page_private_2(struct page
*page
)
1475 page
= compound_head(page
);
1476 VM_BUG_ON_PAGE(!PagePrivate2(page
), page
);
1477 clear_bit_unlock(PG_private_2
, &page
->flags
);
1478 wake_up_page_bit(page
, PG_private_2
);
1481 EXPORT_SYMBOL(end_page_private_2
);
1484 * wait_on_page_private_2 - Wait for PG_private_2 to be cleared on a page
1485 * @page: The page to wait on
1487 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page.
1489 void wait_on_page_private_2(struct page
*page
)
1491 page
= compound_head(page
);
1492 while (PagePrivate2(page
))
1493 wait_on_page_bit(page
, PG_private_2
);
1495 EXPORT_SYMBOL(wait_on_page_private_2
);
1498 * wait_on_page_private_2_killable - Wait for PG_private_2 to be cleared on a page
1499 * @page: The page to wait on
1501 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page or until a
1502 * fatal signal is received by the calling task.
1505 * - 0 if successful.
1506 * - -EINTR if a fatal signal was encountered.
1508 int wait_on_page_private_2_killable(struct page
*page
)
1512 page
= compound_head(page
);
1513 while (PagePrivate2(page
)) {
1514 ret
= wait_on_page_bit_killable(page
, PG_private_2
);
1521 EXPORT_SYMBOL(wait_on_page_private_2_killable
);
1524 * end_page_writeback - end writeback against a page
1527 void end_page_writeback(struct page
*page
)
1530 * TestClearPageReclaim could be used here but it is an atomic
1531 * operation and overkill in this particular case. Failing to
1532 * shuffle a page marked for immediate reclaim is too mild to
1533 * justify taking an atomic operation penalty at the end of
1534 * ever page writeback.
1536 if (PageReclaim(page
)) {
1537 ClearPageReclaim(page
);
1538 rotate_reclaimable_page(page
);
1542 * Writeback does not hold a page reference of its own, relying
1543 * on truncation to wait for the clearing of PG_writeback.
1544 * But here we must make sure that the page is not freed and
1545 * reused before the wake_up_page().
1548 if (!test_clear_page_writeback(page
))
1551 smp_mb__after_atomic();
1552 wake_up_page(page
, PG_writeback
);
1555 EXPORT_SYMBOL(end_page_writeback
);
1558 * After completing I/O on a page, call this routine to update the page
1559 * flags appropriately
1561 void page_endio(struct page
*page
, bool is_write
, int err
)
1565 SetPageUptodate(page
);
1567 ClearPageUptodate(page
);
1573 struct address_space
*mapping
;
1576 mapping
= page_mapping(page
);
1578 mapping_set_error(mapping
, err
);
1580 end_page_writeback(page
);
1583 EXPORT_SYMBOL_GPL(page_endio
);
1586 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1587 * @__page: the page to lock
1589 void __lock_page(struct page
*__page
)
1591 struct page
*page
= compound_head(__page
);
1592 wait_queue_head_t
*q
= page_waitqueue(page
);
1593 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
,
1596 EXPORT_SYMBOL(__lock_page
);
1598 int __lock_page_killable(struct page
*__page
)
1600 struct page
*page
= compound_head(__page
);
1601 wait_queue_head_t
*q
= page_waitqueue(page
);
1602 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
,
1605 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1607 int __lock_page_async(struct page
*page
, struct wait_page_queue
*wait
)
1609 struct wait_queue_head
*q
= page_waitqueue(page
);
1613 wait
->bit_nr
= PG_locked
;
1615 spin_lock_irq(&q
->lock
);
1616 __add_wait_queue_entry_tail(q
, &wait
->wait
);
1617 SetPageWaiters(page
);
1618 ret
= !trylock_page(page
);
1620 * If we were successful now, we know we're still on the
1621 * waitqueue as we're still under the lock. This means it's
1622 * safe to remove and return success, we know the callback
1623 * isn't going to trigger.
1626 __remove_wait_queue(q
, &wait
->wait
);
1629 spin_unlock_irq(&q
->lock
);
1635 * 1 - page is locked; mmap_lock is still held.
1636 * 0 - page is not locked.
1637 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1638 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1639 * which case mmap_lock is still held.
1641 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1642 * with the page locked and the mmap_lock unperturbed.
1644 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1647 if (fault_flag_allow_retry_first(flags
)) {
1649 * CAUTION! In this case, mmap_lock is not released
1650 * even though return 0.
1652 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1655 mmap_read_unlock(mm
);
1656 if (flags
& FAULT_FLAG_KILLABLE
)
1657 wait_on_page_locked_killable(page
);
1659 wait_on_page_locked(page
);
1662 if (flags
& FAULT_FLAG_KILLABLE
) {
1665 ret
= __lock_page_killable(page
);
1667 mmap_read_unlock(mm
);
1678 * page_cache_next_miss() - Find the next gap in the page cache.
1679 * @mapping: Mapping.
1681 * @max_scan: Maximum range to search.
1683 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1684 * gap with the lowest index.
1686 * This function may be called under the rcu_read_lock. However, this will
1687 * not atomically search a snapshot of the cache at a single point in time.
1688 * For example, if a gap is created at index 5, then subsequently a gap is
1689 * created at index 10, page_cache_next_miss covering both indices may
1690 * return 10 if called under the rcu_read_lock.
1692 * Return: The index of the gap if found, otherwise an index outside the
1693 * range specified (in which case 'return - index >= max_scan' will be true).
1694 * In the rare case of index wrap-around, 0 will be returned.
1696 pgoff_t
page_cache_next_miss(struct address_space
*mapping
,
1697 pgoff_t index
, unsigned long max_scan
)
1699 XA_STATE(xas
, &mapping
->i_pages
, index
);
1701 while (max_scan
--) {
1702 void *entry
= xas_next(&xas
);
1703 if (!entry
|| xa_is_value(entry
))
1705 if (xas
.xa_index
== 0)
1709 return xas
.xa_index
;
1711 EXPORT_SYMBOL(page_cache_next_miss
);
1714 * page_cache_prev_miss() - Find the previous gap in the page cache.
1715 * @mapping: Mapping.
1717 * @max_scan: Maximum range to search.
1719 * Search the range [max(index - max_scan + 1, 0), index] for the
1720 * gap with the highest index.
1722 * This function may be called under the rcu_read_lock. However, this will
1723 * not atomically search a snapshot of the cache at a single point in time.
1724 * For example, if a gap is created at index 10, then subsequently a gap is
1725 * created at index 5, page_cache_prev_miss() covering both indices may
1726 * return 5 if called under the rcu_read_lock.
1728 * Return: The index of the gap if found, otherwise an index outside the
1729 * range specified (in which case 'index - return >= max_scan' will be true).
1730 * In the rare case of wrap-around, ULONG_MAX will be returned.
1732 pgoff_t
page_cache_prev_miss(struct address_space
*mapping
,
1733 pgoff_t index
, unsigned long max_scan
)
1735 XA_STATE(xas
, &mapping
->i_pages
, index
);
1737 while (max_scan
--) {
1738 void *entry
= xas_prev(&xas
);
1739 if (!entry
|| xa_is_value(entry
))
1741 if (xas
.xa_index
== ULONG_MAX
)
1745 return xas
.xa_index
;
1747 EXPORT_SYMBOL(page_cache_prev_miss
);
1750 * mapping_get_entry - Get a page cache entry.
1751 * @mapping: the address_space to search
1752 * @index: The page cache index.
1754 * Looks up the page cache slot at @mapping & @index. If there is a
1755 * page cache page, the head page is returned with an increased refcount.
1757 * If the slot holds a shadow entry of a previously evicted page, or a
1758 * swap entry from shmem/tmpfs, it is returned.
1760 * Return: The head page or shadow entry, %NULL if nothing is found.
1762 static struct page
*mapping_get_entry(struct address_space
*mapping
,
1765 XA_STATE(xas
, &mapping
->i_pages
, index
);
1771 page
= xas_load(&xas
);
1772 if (xas_retry(&xas
, page
))
1775 * A shadow entry of a recently evicted page, or a swap entry from
1776 * shmem/tmpfs. Return it without attempting to raise page count.
1778 if (!page
|| xa_is_value(page
))
1781 if (!page_cache_get_speculative(page
))
1785 * Has the page moved or been split?
1786 * This is part of the lockless pagecache protocol. See
1787 * include/linux/pagemap.h for details.
1789 if (unlikely(page
!= xas_reload(&xas
))) {
1800 * pagecache_get_page - Find and get a reference to a page.
1801 * @mapping: The address_space to search.
1802 * @index: The page index.
1803 * @fgp_flags: %FGP flags modify how the page is returned.
1804 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1806 * Looks up the page cache entry at @mapping & @index.
1808 * @fgp_flags can be zero or more of these flags:
1810 * * %FGP_ACCESSED - The page will be marked accessed.
1811 * * %FGP_LOCK - The page is returned locked.
1812 * * %FGP_HEAD - If the page is present and a THP, return the head page
1813 * rather than the exact page specified by the index.
1814 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1815 * instead of allocating a new page to replace it.
1816 * * %FGP_CREAT - If no page is present then a new page is allocated using
1817 * @gfp_mask and added to the page cache and the VM's LRU list.
1818 * The page is returned locked and with an increased refcount.
1819 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1820 * page is already in cache. If the page was allocated, unlock it before
1821 * returning so the caller can do the same dance.
1822 * * %FGP_WRITE - The page will be written
1823 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1824 * * %FGP_NOWAIT - Don't get blocked by page lock
1826 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1827 * if the %GFP flags specified for %FGP_CREAT are atomic.
1829 * If there is a page cache page, it is returned with an increased refcount.
1831 * Return: The found page or %NULL otherwise.
1833 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t index
,
1834 int fgp_flags
, gfp_t gfp_mask
)
1839 page
= mapping_get_entry(mapping
, index
);
1840 if (xa_is_value(page
)) {
1841 if (fgp_flags
& FGP_ENTRY
)
1848 if (fgp_flags
& FGP_LOCK
) {
1849 if (fgp_flags
& FGP_NOWAIT
) {
1850 if (!trylock_page(page
)) {
1858 /* Has the page been truncated? */
1859 if (unlikely(page
->mapping
!= mapping
)) {
1864 VM_BUG_ON_PAGE(!thp_contains(page
, index
), page
);
1867 if (fgp_flags
& FGP_ACCESSED
)
1868 mark_page_accessed(page
);
1869 else if (fgp_flags
& FGP_WRITE
) {
1870 /* Clear idle flag for buffer write */
1871 if (page_is_idle(page
))
1872 clear_page_idle(page
);
1874 if (!(fgp_flags
& FGP_HEAD
))
1875 page
= find_subpage(page
, index
);
1878 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1880 if ((fgp_flags
& FGP_WRITE
) && mapping_can_writeback(mapping
))
1881 gfp_mask
|= __GFP_WRITE
;
1882 if (fgp_flags
& FGP_NOFS
)
1883 gfp_mask
&= ~__GFP_FS
;
1885 page
= __page_cache_alloc(gfp_mask
);
1889 if (WARN_ON_ONCE(!(fgp_flags
& (FGP_LOCK
| FGP_FOR_MMAP
))))
1890 fgp_flags
|= FGP_LOCK
;
1892 /* Init accessed so avoid atomic mark_page_accessed later */
1893 if (fgp_flags
& FGP_ACCESSED
)
1894 __SetPageReferenced(page
);
1896 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp_mask
);
1897 if (unlikely(err
)) {
1905 * add_to_page_cache_lru locks the page, and for mmap we expect
1908 if (page
&& (fgp_flags
& FGP_FOR_MMAP
))
1914 EXPORT_SYMBOL(pagecache_get_page
);
1916 static inline struct page
*find_get_entry(struct xa_state
*xas
, pgoff_t max
,
1922 if (mark
== XA_PRESENT
)
1923 page
= xas_find(xas
, max
);
1925 page
= xas_find_marked(xas
, max
, mark
);
1927 if (xas_retry(xas
, page
))
1930 * A shadow entry of a recently evicted page, a swap
1931 * entry from shmem/tmpfs or a DAX entry. Return it
1932 * without attempting to raise page count.
1934 if (!page
|| xa_is_value(page
))
1937 if (!page_cache_get_speculative(page
))
1940 /* Has the page moved or been split? */
1941 if (unlikely(page
!= xas_reload(xas
))) {
1953 * find_get_entries - gang pagecache lookup
1954 * @mapping: The address_space to search
1955 * @start: The starting page cache index
1956 * @end: The final page index (inclusive).
1957 * @pvec: Where the resulting entries are placed.
1958 * @indices: The cache indices corresponding to the entries in @entries
1960 * find_get_entries() will search for and return a batch of entries in
1961 * the mapping. The entries are placed in @pvec. find_get_entries()
1962 * takes a reference on any actual pages it returns.
1964 * The search returns a group of mapping-contiguous page cache entries
1965 * with ascending indexes. There may be holes in the indices due to
1966 * not-present pages.
1968 * Any shadow entries of evicted pages, or swap entries from
1969 * shmem/tmpfs, are included in the returned array.
1971 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1972 * stops at that page: the caller is likely to have a better way to handle
1973 * the compound page as a whole, and then skip its extent, than repeatedly
1974 * calling find_get_entries() to return all its tails.
1976 * Return: the number of pages and shadow entries which were found.
1978 unsigned find_get_entries(struct address_space
*mapping
, pgoff_t start
,
1979 pgoff_t end
, struct pagevec
*pvec
, pgoff_t
*indices
)
1981 XA_STATE(xas
, &mapping
->i_pages
, start
);
1983 unsigned int ret
= 0;
1984 unsigned nr_entries
= PAGEVEC_SIZE
;
1987 while ((page
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
1989 * Terminate early on finding a THP, to allow the caller to
1990 * handle it all at once; but continue if this is hugetlbfs.
1992 if (!xa_is_value(page
) && PageTransHuge(page
) &&
1994 page
= find_subpage(page
, xas
.xa_index
);
1995 nr_entries
= ret
+ 1;
1998 indices
[ret
] = xas
.xa_index
;
1999 pvec
->pages
[ret
] = page
;
2000 if (++ret
== nr_entries
)
2010 * find_lock_entries - Find a batch of pagecache entries.
2011 * @mapping: The address_space to search.
2012 * @start: The starting page cache index.
2013 * @end: The final page index (inclusive).
2014 * @pvec: Where the resulting entries are placed.
2015 * @indices: The cache indices of the entries in @pvec.
2017 * find_lock_entries() will return a batch of entries from @mapping.
2018 * Swap, shadow and DAX entries are included. Pages are returned
2019 * locked and with an incremented refcount. Pages which are locked by
2020 * somebody else or under writeback are skipped. Only the head page of
2021 * a THP is returned. Pages which are partially outside the range are
2024 * The entries have ascending indexes. The indices may not be consecutive
2025 * due to not-present entries, THP pages, pages which could not be locked
2026 * or pages under writeback.
2028 * Return: The number of entries which were found.
2030 unsigned find_lock_entries(struct address_space
*mapping
, pgoff_t start
,
2031 pgoff_t end
, struct pagevec
*pvec
, pgoff_t
*indices
)
2033 XA_STATE(xas
, &mapping
->i_pages
, start
);
2037 while ((page
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
2038 if (!xa_is_value(page
)) {
2039 if (page
->index
< start
)
2041 VM_BUG_ON_PAGE(page
->index
!= xas
.xa_index
, page
);
2042 if (page
->index
+ thp_nr_pages(page
) - 1 > end
)
2044 if (!trylock_page(page
))
2046 if (page
->mapping
!= mapping
|| PageWriteback(page
))
2048 VM_BUG_ON_PAGE(!thp_contains(page
, xas
.xa_index
),
2051 indices
[pvec
->nr
] = xas
.xa_index
;
2052 if (!pagevec_add(pvec
, page
))
2060 if (!xa_is_value(page
) && PageTransHuge(page
)) {
2061 unsigned int nr_pages
= thp_nr_pages(page
);
2063 /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */
2064 xas_set(&xas
, page
->index
+ nr_pages
);
2065 if (xas
.xa_index
< nr_pages
)
2071 return pagevec_count(pvec
);
2075 * find_get_pages_range - gang pagecache lookup
2076 * @mapping: The address_space to search
2077 * @start: The starting page index
2078 * @end: The final page index (inclusive)
2079 * @nr_pages: The maximum number of pages
2080 * @pages: Where the resulting pages are placed
2082 * find_get_pages_range() will search for and return a group of up to @nr_pages
2083 * pages in the mapping starting at index @start and up to index @end
2084 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2085 * a reference against the returned pages.
2087 * The search returns a group of mapping-contiguous pages with ascending
2088 * indexes. There may be holes in the indices due to not-present pages.
2089 * We also update @start to index the next page for the traversal.
2091 * Return: the number of pages which were found. If this number is
2092 * smaller than @nr_pages, the end of specified range has been
2095 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
2096 pgoff_t end
, unsigned int nr_pages
,
2097 struct page
**pages
)
2099 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2103 if (unlikely(!nr_pages
))
2107 while ((page
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
2108 /* Skip over shadow, swap and DAX entries */
2109 if (xa_is_value(page
))
2112 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
2113 if (++ret
== nr_pages
) {
2114 *start
= xas
.xa_index
+ 1;
2120 * We come here when there is no page beyond @end. We take care to not
2121 * overflow the index @start as it confuses some of the callers. This
2122 * breaks the iteration when there is a page at index -1 but that is
2123 * already broken anyway.
2125 if (end
== (pgoff_t
)-1)
2126 *start
= (pgoff_t
)-1;
2136 * find_get_pages_contig - gang contiguous pagecache lookup
2137 * @mapping: The address_space to search
2138 * @index: The starting page index
2139 * @nr_pages: The maximum number of pages
2140 * @pages: Where the resulting pages are placed
2142 * find_get_pages_contig() works exactly like find_get_pages(), except
2143 * that the returned number of pages are guaranteed to be contiguous.
2145 * Return: the number of pages which were found.
2147 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
2148 unsigned int nr_pages
, struct page
**pages
)
2150 XA_STATE(xas
, &mapping
->i_pages
, index
);
2152 unsigned int ret
= 0;
2154 if (unlikely(!nr_pages
))
2158 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
2159 if (xas_retry(&xas
, page
))
2162 * If the entry has been swapped out, we can stop looking.
2163 * No current caller is looking for DAX entries.
2165 if (xa_is_value(page
))
2168 if (!page_cache_get_speculative(page
))
2171 /* Has the page moved or been split? */
2172 if (unlikely(page
!= xas_reload(&xas
)))
2175 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
2176 if (++ret
== nr_pages
)
2187 EXPORT_SYMBOL(find_get_pages_contig
);
2190 * find_get_pages_range_tag - Find and return head pages matching @tag.
2191 * @mapping: the address_space to search
2192 * @index: the starting page index
2193 * @end: The final page index (inclusive)
2194 * @tag: the tag index
2195 * @nr_pages: the maximum number of pages
2196 * @pages: where the resulting pages are placed
2198 * Like find_get_pages(), except we only return head pages which are tagged
2199 * with @tag. @index is updated to the index immediately after the last
2200 * page we return, ready for the next iteration.
2202 * Return: the number of pages which were found.
2204 unsigned find_get_pages_range_tag(struct address_space
*mapping
, pgoff_t
*index
,
2205 pgoff_t end
, xa_mark_t tag
, unsigned int nr_pages
,
2206 struct page
**pages
)
2208 XA_STATE(xas
, &mapping
->i_pages
, *index
);
2212 if (unlikely(!nr_pages
))
2216 while ((page
= find_get_entry(&xas
, end
, tag
))) {
2218 * Shadow entries should never be tagged, but this iteration
2219 * is lockless so there is a window for page reclaim to evict
2220 * a page we saw tagged. Skip over it.
2222 if (xa_is_value(page
))
2226 if (++ret
== nr_pages
) {
2227 *index
= page
->index
+ thp_nr_pages(page
);
2233 * We come here when we got to @end. We take care to not overflow the
2234 * index @index as it confuses some of the callers. This breaks the
2235 * iteration when there is a page at index -1 but that is already
2238 if (end
== (pgoff_t
)-1)
2239 *index
= (pgoff_t
)-1;
2247 EXPORT_SYMBOL(find_get_pages_range_tag
);
2250 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2251 * a _large_ part of the i/o request. Imagine the worst scenario:
2253 * ---R__________________________________________B__________
2254 * ^ reading here ^ bad block(assume 4k)
2256 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2257 * => failing the whole request => read(R) => read(R+1) =>
2258 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2259 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2260 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2262 * It is going insane. Fix it by quickly scaling down the readahead size.
2264 static void shrink_readahead_size_eio(struct file_ra_state
*ra
)
2270 * filemap_get_read_batch - Get a batch of pages for read
2272 * Get a batch of pages which represent a contiguous range of bytes
2273 * in the file. No tail pages will be returned. If @index is in the
2274 * middle of a THP, the entire THP will be returned. The last page in
2275 * the batch may have Readahead set or be not Uptodate so that the
2276 * caller can take the appropriate action.
2278 static void filemap_get_read_batch(struct address_space
*mapping
,
2279 pgoff_t index
, pgoff_t max
, struct pagevec
*pvec
)
2281 XA_STATE(xas
, &mapping
->i_pages
, index
);
2285 for (head
= xas_load(&xas
); head
; head
= xas_next(&xas
)) {
2286 if (xas_retry(&xas
, head
))
2288 if (xas
.xa_index
> max
|| xa_is_value(head
))
2290 if (!page_cache_get_speculative(head
))
2293 /* Has the page moved or been split? */
2294 if (unlikely(head
!= xas_reload(&xas
)))
2297 if (!pagevec_add(pvec
, head
))
2299 if (!PageUptodate(head
))
2301 if (PageReadahead(head
))
2303 xas
.xa_index
= head
->index
+ thp_nr_pages(head
) - 1;
2304 xas
.xa_offset
= (xas
.xa_index
>> xas
.xa_shift
) & XA_CHUNK_MASK
;
2314 static int filemap_read_page(struct file
*file
, struct address_space
*mapping
,
2320 * A previous I/O error may have been due to temporary failures,
2321 * eg. multipath errors. PG_error will be set again if readpage
2324 ClearPageError(page
);
2325 /* Start the actual read. The read will unlock the page. */
2326 error
= mapping
->a_ops
->readpage(file
, page
);
2330 error
= wait_on_page_locked_killable(page
);
2333 if (PageUptodate(page
))
2335 shrink_readahead_size_eio(&file
->f_ra
);
2339 static bool filemap_range_uptodate(struct address_space
*mapping
,
2340 loff_t pos
, struct iov_iter
*iter
, struct page
*page
)
2344 if (PageUptodate(page
))
2346 /* pipes can't handle partially uptodate pages */
2347 if (iov_iter_is_pipe(iter
))
2349 if (!mapping
->a_ops
->is_partially_uptodate
)
2351 if (mapping
->host
->i_blkbits
>= (PAGE_SHIFT
+ thp_order(page
)))
2354 count
= iter
->count
;
2355 if (page_offset(page
) > pos
) {
2356 count
-= page_offset(page
) - pos
;
2359 pos
-= page_offset(page
);
2362 return mapping
->a_ops
->is_partially_uptodate(page
, pos
, count
);
2365 static int filemap_update_page(struct kiocb
*iocb
,
2366 struct address_space
*mapping
, struct iov_iter
*iter
,
2371 if (!trylock_page(page
)) {
2372 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_NOIO
))
2374 if (!(iocb
->ki_flags
& IOCB_WAITQ
)) {
2375 put_and_wait_on_page_locked(page
, TASK_KILLABLE
);
2376 return AOP_TRUNCATED_PAGE
;
2378 error
= __lock_page_async(page
, iocb
->ki_waitq
);
2387 if (filemap_range_uptodate(mapping
, iocb
->ki_pos
, iter
, page
))
2391 if (iocb
->ki_flags
& (IOCB_NOIO
| IOCB_NOWAIT
| IOCB_WAITQ
))
2394 error
= filemap_read_page(iocb
->ki_filp
, mapping
, page
);
2395 if (error
== AOP_TRUNCATED_PAGE
)
2401 return AOP_TRUNCATED_PAGE
;
2407 static int filemap_create_page(struct file
*file
,
2408 struct address_space
*mapping
, pgoff_t index
,
2409 struct pagevec
*pvec
)
2414 page
= page_cache_alloc(mapping
);
2418 error
= add_to_page_cache_lru(page
, mapping
, index
,
2419 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2420 if (error
== -EEXIST
)
2421 error
= AOP_TRUNCATED_PAGE
;
2425 error
= filemap_read_page(file
, mapping
, page
);
2429 pagevec_add(pvec
, page
);
2436 static int filemap_readahead(struct kiocb
*iocb
, struct file
*file
,
2437 struct address_space
*mapping
, struct page
*page
,
2440 if (iocb
->ki_flags
& IOCB_NOIO
)
2442 page_cache_async_readahead(mapping
, &file
->f_ra
, file
, page
,
2443 page
->index
, last_index
- page
->index
);
2447 static int filemap_get_pages(struct kiocb
*iocb
, struct iov_iter
*iter
,
2448 struct pagevec
*pvec
)
2450 struct file
*filp
= iocb
->ki_filp
;
2451 struct address_space
*mapping
= filp
->f_mapping
;
2452 struct file_ra_state
*ra
= &filp
->f_ra
;
2453 pgoff_t index
= iocb
->ki_pos
>> PAGE_SHIFT
;
2458 last_index
= DIV_ROUND_UP(iocb
->ki_pos
+ iter
->count
, PAGE_SIZE
);
2460 if (fatal_signal_pending(current
))
2463 filemap_get_read_batch(mapping
, index
, last_index
, pvec
);
2464 if (!pagevec_count(pvec
)) {
2465 if (iocb
->ki_flags
& IOCB_NOIO
)
2467 page_cache_sync_readahead(mapping
, ra
, filp
, index
,
2468 last_index
- index
);
2469 filemap_get_read_batch(mapping
, index
, last_index
, pvec
);
2471 if (!pagevec_count(pvec
)) {
2472 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_WAITQ
))
2474 err
= filemap_create_page(filp
, mapping
,
2475 iocb
->ki_pos
>> PAGE_SHIFT
, pvec
);
2476 if (err
== AOP_TRUNCATED_PAGE
)
2481 page
= pvec
->pages
[pagevec_count(pvec
) - 1];
2482 if (PageReadahead(page
)) {
2483 err
= filemap_readahead(iocb
, filp
, mapping
, page
, last_index
);
2487 if (!PageUptodate(page
)) {
2488 if ((iocb
->ki_flags
& IOCB_WAITQ
) && pagevec_count(pvec
) > 1)
2489 iocb
->ki_flags
|= IOCB_NOWAIT
;
2490 err
= filemap_update_page(iocb
, mapping
, iter
, page
);
2499 if (likely(--pvec
->nr
))
2501 if (err
== AOP_TRUNCATED_PAGE
)
2507 * filemap_read - Read data from the page cache.
2508 * @iocb: The iocb to read.
2509 * @iter: Destination for the data.
2510 * @already_read: Number of bytes already read by the caller.
2512 * Copies data from the page cache. If the data is not currently present,
2513 * uses the readahead and readpage address_space operations to fetch it.
2515 * Return: Total number of bytes copied, including those already read by
2516 * the caller. If an error happens before any bytes are copied, returns
2517 * a negative error number.
2519 ssize_t
filemap_read(struct kiocb
*iocb
, struct iov_iter
*iter
,
2520 ssize_t already_read
)
2522 struct file
*filp
= iocb
->ki_filp
;
2523 struct file_ra_state
*ra
= &filp
->f_ra
;
2524 struct address_space
*mapping
= filp
->f_mapping
;
2525 struct inode
*inode
= mapping
->host
;
2526 struct pagevec pvec
;
2528 bool writably_mapped
;
2529 loff_t isize
, end_offset
;
2531 if (unlikely(iocb
->ki_pos
>= inode
->i_sb
->s_maxbytes
))
2533 if (unlikely(!iov_iter_count(iter
)))
2536 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2537 pagevec_init(&pvec
);
2543 * If we've already successfully copied some data, then we
2544 * can no longer safely return -EIOCBQUEUED. Hence mark
2545 * an async read NOWAIT at that point.
2547 if ((iocb
->ki_flags
& IOCB_WAITQ
) && already_read
)
2548 iocb
->ki_flags
|= IOCB_NOWAIT
;
2550 error
= filemap_get_pages(iocb
, iter
, &pvec
);
2555 * i_size must be checked after we know the pages are Uptodate.
2557 * Checking i_size after the check allows us to calculate
2558 * the correct value for "nr", which means the zero-filled
2559 * part of the page is not copied back to userspace (unless
2560 * another truncate extends the file - this is desired though).
2562 isize
= i_size_read(inode
);
2563 if (unlikely(iocb
->ki_pos
>= isize
))
2565 end_offset
= min_t(loff_t
, isize
, iocb
->ki_pos
+ iter
->count
);
2568 * Once we start copying data, we don't want to be touching any
2569 * cachelines that might be contended:
2571 writably_mapped
= mapping_writably_mapped(mapping
);
2574 * When a sequential read accesses a page several times, only
2575 * mark it as accessed the first time.
2577 if (iocb
->ki_pos
>> PAGE_SHIFT
!=
2578 ra
->prev_pos
>> PAGE_SHIFT
)
2579 mark_page_accessed(pvec
.pages
[0]);
2581 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
2582 struct page
*page
= pvec
.pages
[i
];
2583 size_t page_size
= thp_size(page
);
2584 size_t offset
= iocb
->ki_pos
& (page_size
- 1);
2585 size_t bytes
= min_t(loff_t
, end_offset
- iocb
->ki_pos
,
2586 page_size
- offset
);
2589 if (end_offset
< page_offset(page
))
2592 mark_page_accessed(page
);
2594 * If users can be writing to this page using arbitrary
2595 * virtual addresses, take care about potential aliasing
2596 * before reading the page on the kernel side.
2598 if (writably_mapped
) {
2601 for (j
= 0; j
< thp_nr_pages(page
); j
++)
2602 flush_dcache_page(page
+ j
);
2605 copied
= copy_page_to_iter(page
, offset
, bytes
, iter
);
2607 already_read
+= copied
;
2608 iocb
->ki_pos
+= copied
;
2609 ra
->prev_pos
= iocb
->ki_pos
;
2611 if (copied
< bytes
) {
2617 for (i
= 0; i
< pagevec_count(&pvec
); i
++)
2618 put_page(pvec
.pages
[i
]);
2619 pagevec_reinit(&pvec
);
2620 } while (iov_iter_count(iter
) && iocb
->ki_pos
< isize
&& !error
);
2622 file_accessed(filp
);
2624 return already_read
? already_read
: error
;
2626 EXPORT_SYMBOL_GPL(filemap_read
);
2629 * generic_file_read_iter - generic filesystem read routine
2630 * @iocb: kernel I/O control block
2631 * @iter: destination for the data read
2633 * This is the "read_iter()" routine for all filesystems
2634 * that can use the page cache directly.
2636 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2637 * be returned when no data can be read without waiting for I/O requests
2638 * to complete; it doesn't prevent readahead.
2640 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2641 * requests shall be made for the read or for readahead. When no data
2642 * can be read, -EAGAIN shall be returned. When readahead would be
2643 * triggered, a partial, possibly empty read shall be returned.
2646 * * number of bytes copied, even for partial reads
2647 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2650 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2652 size_t count
= iov_iter_count(iter
);
2656 return 0; /* skip atime */
2658 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2659 struct file
*file
= iocb
->ki_filp
;
2660 struct address_space
*mapping
= file
->f_mapping
;
2661 struct inode
*inode
= mapping
->host
;
2664 size
= i_size_read(inode
);
2665 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2666 if (filemap_range_needs_writeback(mapping
, iocb
->ki_pos
,
2667 iocb
->ki_pos
+ count
- 1))
2670 retval
= filemap_write_and_wait_range(mapping
,
2672 iocb
->ki_pos
+ count
- 1);
2677 file_accessed(file
);
2679 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2681 iocb
->ki_pos
+= retval
;
2684 if (retval
!= -EIOCBQUEUED
)
2685 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2688 * Btrfs can have a short DIO read if we encounter
2689 * compressed extents, so if there was an error, or if
2690 * we've already read everything we wanted to, or if
2691 * there was a short read because we hit EOF, go ahead
2692 * and return. Otherwise fallthrough to buffered io for
2693 * the rest of the read. Buffered reads will not work for
2694 * DAX files, so don't bother trying.
2696 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2701 return filemap_read(iocb
, iter
, retval
);
2703 EXPORT_SYMBOL(generic_file_read_iter
);
2705 static inline loff_t
page_seek_hole_data(struct xa_state
*xas
,
2706 struct address_space
*mapping
, struct page
*page
,
2707 loff_t start
, loff_t end
, bool seek_data
)
2709 const struct address_space_operations
*ops
= mapping
->a_ops
;
2710 size_t offset
, bsz
= i_blocksize(mapping
->host
);
2712 if (xa_is_value(page
) || PageUptodate(page
))
2713 return seek_data
? start
: end
;
2714 if (!ops
->is_partially_uptodate
)
2715 return seek_data
? end
: start
;
2720 if (unlikely(page
->mapping
!= mapping
))
2723 offset
= offset_in_thp(page
, start
) & ~(bsz
- 1);
2726 if (ops
->is_partially_uptodate(page
, offset
, bsz
) == seek_data
)
2728 start
= (start
+ bsz
) & ~(bsz
- 1);
2730 } while (offset
< thp_size(page
));
2738 unsigned int seek_page_size(struct xa_state
*xas
, struct page
*page
)
2740 if (xa_is_value(page
))
2741 return PAGE_SIZE
<< xa_get_order(xas
->xa
, xas
->xa_index
);
2742 return thp_size(page
);
2746 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2747 * @mapping: Address space to search.
2748 * @start: First byte to consider.
2749 * @end: Limit of search (exclusive).
2750 * @whence: Either SEEK_HOLE or SEEK_DATA.
2752 * If the page cache knows which blocks contain holes and which blocks
2753 * contain data, your filesystem can use this function to implement
2754 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2755 * entirely memory-based such as tmpfs, and filesystems which support
2756 * unwritten extents.
2758 * Return: The requested offset on success, or -ENXIO if @whence specifies
2759 * SEEK_DATA and there is no data after @start. There is an implicit hole
2760 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2761 * and @end contain data.
2763 loff_t
mapping_seek_hole_data(struct address_space
*mapping
, loff_t start
,
2764 loff_t end
, int whence
)
2766 XA_STATE(xas
, &mapping
->i_pages
, start
>> PAGE_SHIFT
);
2767 pgoff_t max
= (end
- 1) >> PAGE_SHIFT
;
2768 bool seek_data
= (whence
== SEEK_DATA
);
2775 while ((page
= find_get_entry(&xas
, max
, XA_PRESENT
))) {
2776 loff_t pos
= (u64
)xas
.xa_index
<< PAGE_SHIFT
;
2777 unsigned int seek_size
;
2785 seek_size
= seek_page_size(&xas
, page
);
2786 pos
= round_up(pos
+ 1, seek_size
);
2787 start
= page_seek_hole_data(&xas
, mapping
, page
, start
, pos
,
2793 if (seek_size
> PAGE_SIZE
)
2794 xas_set(&xas
, pos
>> PAGE_SHIFT
);
2795 if (!xa_is_value(page
))
2802 if (page
&& !xa_is_value(page
))
2810 #define MMAP_LOTSAMISS (100)
2812 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2813 * @vmf - the vm_fault for this fault.
2814 * @page - the page to lock.
2815 * @fpin - the pointer to the file we may pin (or is already pinned).
2817 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2818 * It differs in that it actually returns the page locked if it returns 1 and 0
2819 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2820 * will point to the pinned file and needs to be fput()'ed at a later point.
2822 static int lock_page_maybe_drop_mmap(struct vm_fault
*vmf
, struct page
*page
,
2825 if (trylock_page(page
))
2829 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2830 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2831 * is supposed to work. We have way too many special cases..
2833 if (vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)
2836 *fpin
= maybe_unlock_mmap_for_io(vmf
, *fpin
);
2837 if (vmf
->flags
& FAULT_FLAG_KILLABLE
) {
2838 if (__lock_page_killable(page
)) {
2840 * We didn't have the right flags to drop the mmap_lock,
2841 * but all fault_handlers only check for fatal signals
2842 * if we return VM_FAULT_RETRY, so we need to drop the
2843 * mmap_lock here and return 0 if we don't have a fpin.
2846 mmap_read_unlock(vmf
->vma
->vm_mm
);
2856 * Synchronous readahead happens when we don't even find a page in the page
2857 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2858 * to drop the mmap sem we return the file that was pinned in order for us to do
2859 * that. If we didn't pin a file then we return NULL. The file that is
2860 * returned needs to be fput()'ed when we're done with it.
2862 static struct file
*do_sync_mmap_readahead(struct vm_fault
*vmf
)
2864 struct file
*file
= vmf
->vma
->vm_file
;
2865 struct file_ra_state
*ra
= &file
->f_ra
;
2866 struct address_space
*mapping
= file
->f_mapping
;
2867 DEFINE_READAHEAD(ractl
, file
, ra
, mapping
, vmf
->pgoff
);
2868 struct file
*fpin
= NULL
;
2869 unsigned int mmap_miss
;
2871 /* If we don't want any read-ahead, don't bother */
2872 if (vmf
->vma
->vm_flags
& VM_RAND_READ
)
2877 if (vmf
->vma
->vm_flags
& VM_SEQ_READ
) {
2878 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2879 page_cache_sync_ra(&ractl
, ra
->ra_pages
);
2883 /* Avoid banging the cache line if not needed */
2884 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
2885 if (mmap_miss
< MMAP_LOTSAMISS
* 10)
2886 WRITE_ONCE(ra
->mmap_miss
, ++mmap_miss
);
2889 * Do we miss much more than hit in this file? If so,
2890 * stop bothering with read-ahead. It will only hurt.
2892 if (mmap_miss
> MMAP_LOTSAMISS
)
2898 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2899 ra
->start
= max_t(long, 0, vmf
->pgoff
- ra
->ra_pages
/ 2);
2900 ra
->size
= ra
->ra_pages
;
2901 ra
->async_size
= ra
->ra_pages
/ 4;
2902 ractl
._index
= ra
->start
;
2903 do_page_cache_ra(&ractl
, ra
->size
, ra
->async_size
);
2908 * Asynchronous readahead happens when we find the page and PG_readahead,
2909 * so we want to possibly extend the readahead further. We return the file that
2910 * was pinned if we have to drop the mmap_lock in order to do IO.
2912 static struct file
*do_async_mmap_readahead(struct vm_fault
*vmf
,
2915 struct file
*file
= vmf
->vma
->vm_file
;
2916 struct file_ra_state
*ra
= &file
->f_ra
;
2917 struct address_space
*mapping
= file
->f_mapping
;
2918 struct file
*fpin
= NULL
;
2919 unsigned int mmap_miss
;
2920 pgoff_t offset
= vmf
->pgoff
;
2922 /* If we don't want any read-ahead, don't bother */
2923 if (vmf
->vma
->vm_flags
& VM_RAND_READ
|| !ra
->ra_pages
)
2925 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
2927 WRITE_ONCE(ra
->mmap_miss
, --mmap_miss
);
2928 if (PageReadahead(page
)) {
2929 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2930 page_cache_async_readahead(mapping
, ra
, file
,
2931 page
, offset
, ra
->ra_pages
);
2937 * filemap_fault - read in file data for page fault handling
2938 * @vmf: struct vm_fault containing details of the fault
2940 * filemap_fault() is invoked via the vma operations vector for a
2941 * mapped memory region to read in file data during a page fault.
2943 * The goto's are kind of ugly, but this streamlines the normal case of having
2944 * it in the page cache, and handles the special cases reasonably without
2945 * having a lot of duplicated code.
2947 * vma->vm_mm->mmap_lock must be held on entry.
2949 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2950 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2952 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2953 * has not been released.
2955 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2957 * Return: bitwise-OR of %VM_FAULT_ codes.
2959 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
2962 struct file
*file
= vmf
->vma
->vm_file
;
2963 struct file
*fpin
= NULL
;
2964 struct address_space
*mapping
= file
->f_mapping
;
2965 struct inode
*inode
= mapping
->host
;
2966 pgoff_t offset
= vmf
->pgoff
;
2971 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2972 if (unlikely(offset
>= max_off
))
2973 return VM_FAULT_SIGBUS
;
2976 * Do we have something in the page cache already?
2978 page
= find_get_page(mapping
, offset
);
2979 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2981 * We found the page, so try async readahead before
2982 * waiting for the lock.
2984 fpin
= do_async_mmap_readahead(vmf
, page
);
2986 /* No page in the page cache at all */
2987 count_vm_event(PGMAJFAULT
);
2988 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2989 ret
= VM_FAULT_MAJOR
;
2990 fpin
= do_sync_mmap_readahead(vmf
);
2992 page
= pagecache_get_page(mapping
, offset
,
2993 FGP_CREAT
|FGP_FOR_MMAP
,
2998 return VM_FAULT_OOM
;
3002 if (!lock_page_maybe_drop_mmap(vmf
, page
, &fpin
))
3005 /* Did it get truncated? */
3006 if (unlikely(compound_head(page
)->mapping
!= mapping
)) {
3011 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
3014 * We have a locked page in the page cache, now we need to check
3015 * that it's up-to-date. If not, it is going to be due to an error.
3017 if (unlikely(!PageUptodate(page
)))
3018 goto page_not_uptodate
;
3021 * We've made it this far and we had to drop our mmap_lock, now is the
3022 * time to return to the upper layer and have it re-find the vma and
3031 * Found the page and have a reference on it.
3032 * We must recheck i_size under page lock.
3034 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
3035 if (unlikely(offset
>= max_off
)) {
3038 return VM_FAULT_SIGBUS
;
3042 return ret
| VM_FAULT_LOCKED
;
3046 * Umm, take care of errors if the page isn't up-to-date.
3047 * Try to re-read it _once_. We do this synchronously,
3048 * because there really aren't any performance issues here
3049 * and we need to check for errors.
3051 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3052 error
= filemap_read_page(file
, mapping
, page
);
3057 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
3060 return VM_FAULT_SIGBUS
;
3064 * We dropped the mmap_lock, we need to return to the fault handler to
3065 * re-find the vma and come back and find our hopefully still populated
3072 return ret
| VM_FAULT_RETRY
;
3074 EXPORT_SYMBOL(filemap_fault
);
3076 static bool filemap_map_pmd(struct vm_fault
*vmf
, struct page
*page
)
3078 struct mm_struct
*mm
= vmf
->vma
->vm_mm
;
3080 /* Huge page is mapped? No need to proceed. */
3081 if (pmd_trans_huge(*vmf
->pmd
)) {
3087 if (pmd_none(*vmf
->pmd
) && PageTransHuge(page
)) {
3088 vm_fault_t ret
= do_set_pmd(vmf
, page
);
3090 /* The page is mapped successfully, reference consumed. */
3096 if (pmd_none(*vmf
->pmd
)) {
3097 vmf
->ptl
= pmd_lock(mm
, vmf
->pmd
);
3098 if (likely(pmd_none(*vmf
->pmd
))) {
3100 pmd_populate(mm
, vmf
->pmd
, vmf
->prealloc_pte
);
3101 vmf
->prealloc_pte
= NULL
;
3103 spin_unlock(vmf
->ptl
);
3106 /* See comment in handle_pte_fault() */
3107 if (pmd_devmap_trans_unstable(vmf
->pmd
)) {
3116 static struct page
*next_uptodate_page(struct page
*page
,
3117 struct address_space
*mapping
,
3118 struct xa_state
*xas
, pgoff_t end_pgoff
)
3120 unsigned long max_idx
;
3125 if (xas_retry(xas
, page
))
3127 if (xa_is_value(page
))
3129 if (PageLocked(page
))
3131 if (!page_cache_get_speculative(page
))
3133 /* Has the page moved or been split? */
3134 if (unlikely(page
!= xas_reload(xas
)))
3136 if (!PageUptodate(page
) || PageReadahead(page
))
3138 if (PageHWPoison(page
))
3140 if (!trylock_page(page
))
3142 if (page
->mapping
!= mapping
)
3144 if (!PageUptodate(page
))
3146 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
3147 if (xas
->xa_index
>= max_idx
)
3154 } while ((page
= xas_next_entry(xas
, end_pgoff
)) != NULL
);
3159 static inline struct page
*first_map_page(struct address_space
*mapping
,
3160 struct xa_state
*xas
,
3163 return next_uptodate_page(xas_find(xas
, end_pgoff
),
3164 mapping
, xas
, end_pgoff
);
3167 static inline struct page
*next_map_page(struct address_space
*mapping
,
3168 struct xa_state
*xas
,
3171 return next_uptodate_page(xas_next_entry(xas
, end_pgoff
),
3172 mapping
, xas
, end_pgoff
);
3175 vm_fault_t
filemap_map_pages(struct vm_fault
*vmf
,
3176 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
3178 struct vm_area_struct
*vma
= vmf
->vma
;
3179 struct file
*file
= vma
->vm_file
;
3180 struct address_space
*mapping
= file
->f_mapping
;
3181 pgoff_t last_pgoff
= start_pgoff
;
3183 XA_STATE(xas
, &mapping
->i_pages
, start_pgoff
);
3184 struct page
*head
, *page
;
3185 unsigned int mmap_miss
= READ_ONCE(file
->f_ra
.mmap_miss
);
3189 head
= first_map_page(mapping
, &xas
, end_pgoff
);
3193 if (filemap_map_pmd(vmf
, head
)) {
3194 ret
= VM_FAULT_NOPAGE
;
3198 addr
= vma
->vm_start
+ ((start_pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
3199 vmf
->pte
= pte_offset_map_lock(vma
->vm_mm
, vmf
->pmd
, addr
, &vmf
->ptl
);
3201 page
= find_subpage(head
, xas
.xa_index
);
3202 if (PageHWPoison(page
))
3208 addr
+= (xas
.xa_index
- last_pgoff
) << PAGE_SHIFT
;
3209 vmf
->pte
+= xas
.xa_index
- last_pgoff
;
3210 last_pgoff
= xas
.xa_index
;
3212 if (!pte_none(*vmf
->pte
))
3215 /* We're about to handle the fault */
3216 if (vmf
->address
== addr
)
3217 ret
= VM_FAULT_NOPAGE
;
3219 do_set_pte(vmf
, page
, addr
);
3220 /* no need to invalidate: a not-present page won't be cached */
3221 update_mmu_cache(vma
, addr
, vmf
->pte
);
3227 } while ((head
= next_map_page(mapping
, &xas
, end_pgoff
)) != NULL
);
3228 pte_unmap_unlock(vmf
->pte
, vmf
->ptl
);
3231 WRITE_ONCE(file
->f_ra
.mmap_miss
, mmap_miss
);
3234 EXPORT_SYMBOL(filemap_map_pages
);
3236 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3238 struct address_space
*mapping
= vmf
->vma
->vm_file
->f_mapping
;
3239 struct page
*page
= vmf
->page
;
3240 vm_fault_t ret
= VM_FAULT_LOCKED
;
3242 sb_start_pagefault(mapping
->host
->i_sb
);
3243 file_update_time(vmf
->vma
->vm_file
);
3245 if (page
->mapping
!= mapping
) {
3247 ret
= VM_FAULT_NOPAGE
;
3251 * We mark the page dirty already here so that when freeze is in
3252 * progress, we are guaranteed that writeback during freezing will
3253 * see the dirty page and writeprotect it again.
3255 set_page_dirty(page
);
3256 wait_for_stable_page(page
);
3258 sb_end_pagefault(mapping
->host
->i_sb
);
3262 const struct vm_operations_struct generic_file_vm_ops
= {
3263 .fault
= filemap_fault
,
3264 .map_pages
= filemap_map_pages
,
3265 .page_mkwrite
= filemap_page_mkwrite
,
3268 /* This is used for a general mmap of a disk file */
3270 int generic_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3272 struct address_space
*mapping
= file
->f_mapping
;
3274 if (!mapping
->a_ops
->readpage
)
3276 file_accessed(file
);
3277 vma
->vm_ops
= &generic_file_vm_ops
;
3282 * This is for filesystems which do not implement ->writepage.
3284 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3286 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
3288 return generic_file_mmap(file
, vma
);
3291 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3293 return VM_FAULT_SIGBUS
;
3295 int generic_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3299 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3303 #endif /* CONFIG_MMU */
3305 EXPORT_SYMBOL(filemap_page_mkwrite
);
3306 EXPORT_SYMBOL(generic_file_mmap
);
3307 EXPORT_SYMBOL(generic_file_readonly_mmap
);
3309 static struct page
*wait_on_page_read(struct page
*page
)
3311 if (!IS_ERR(page
)) {
3312 wait_on_page_locked(page
);
3313 if (!PageUptodate(page
)) {
3315 page
= ERR_PTR(-EIO
);
3321 static struct page
*do_read_cache_page(struct address_space
*mapping
,
3323 int (*filler
)(void *, struct page
*),
3330 page
= find_get_page(mapping
, index
);
3332 page
= __page_cache_alloc(gfp
);
3334 return ERR_PTR(-ENOMEM
);
3335 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
3336 if (unlikely(err
)) {
3340 /* Presumably ENOMEM for xarray node */
3341 return ERR_PTR(err
);
3346 err
= filler(data
, page
);
3348 err
= mapping
->a_ops
->readpage(data
, page
);
3352 return ERR_PTR(err
);
3355 page
= wait_on_page_read(page
);
3360 if (PageUptodate(page
))
3364 * Page is not up to date and may be locked due to one of the following
3365 * case a: Page is being filled and the page lock is held
3366 * case b: Read/write error clearing the page uptodate status
3367 * case c: Truncation in progress (page locked)
3368 * case d: Reclaim in progress
3370 * Case a, the page will be up to date when the page is unlocked.
3371 * There is no need to serialise on the page lock here as the page
3372 * is pinned so the lock gives no additional protection. Even if the
3373 * page is truncated, the data is still valid if PageUptodate as
3374 * it's a race vs truncate race.
3375 * Case b, the page will not be up to date
3376 * Case c, the page may be truncated but in itself, the data may still
3377 * be valid after IO completes as it's a read vs truncate race. The
3378 * operation must restart if the page is not uptodate on unlock but
3379 * otherwise serialising on page lock to stabilise the mapping gives
3380 * no additional guarantees to the caller as the page lock is
3381 * released before return.
3382 * Case d, similar to truncation. If reclaim holds the page lock, it
3383 * will be a race with remove_mapping that determines if the mapping
3384 * is valid on unlock but otherwise the data is valid and there is
3385 * no need to serialise with page lock.
3387 * As the page lock gives no additional guarantee, we optimistically
3388 * wait on the page to be unlocked and check if it's up to date and
3389 * use the page if it is. Otherwise, the page lock is required to
3390 * distinguish between the different cases. The motivation is that we
3391 * avoid spurious serialisations and wakeups when multiple processes
3392 * wait on the same page for IO to complete.
3394 wait_on_page_locked(page
);
3395 if (PageUptodate(page
))
3398 /* Distinguish between all the cases under the safety of the lock */
3401 /* Case c or d, restart the operation */
3402 if (!page
->mapping
) {
3408 /* Someone else locked and filled the page in a very small window */
3409 if (PageUptodate(page
)) {
3415 * A previous I/O error may have been due to temporary
3417 * Clear page error before actual read, PG_error will be
3418 * set again if read page fails.
3420 ClearPageError(page
);
3424 mark_page_accessed(page
);
3429 * read_cache_page - read into page cache, fill it if needed
3430 * @mapping: the page's address_space
3431 * @index: the page index
3432 * @filler: function to perform the read
3433 * @data: first arg to filler(data, page) function, often left as NULL
3435 * Read into the page cache. If a page already exists, and PageUptodate() is
3436 * not set, try to fill the page and wait for it to become unlocked.
3438 * If the page does not get brought uptodate, return -EIO.
3440 * Return: up to date page on success, ERR_PTR() on failure.
3442 struct page
*read_cache_page(struct address_space
*mapping
,
3444 int (*filler
)(void *, struct page
*),
3447 return do_read_cache_page(mapping
, index
, filler
, data
,
3448 mapping_gfp_mask(mapping
));
3450 EXPORT_SYMBOL(read_cache_page
);
3453 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3454 * @mapping: the page's address_space
3455 * @index: the page index
3456 * @gfp: the page allocator flags to use if allocating
3458 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3459 * any new page allocations done using the specified allocation flags.
3461 * If the page does not get brought uptodate, return -EIO.
3463 * Return: up to date page on success, ERR_PTR() on failure.
3465 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
3469 return do_read_cache_page(mapping
, index
, NULL
, NULL
, gfp
);
3471 EXPORT_SYMBOL(read_cache_page_gfp
);
3473 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
3474 loff_t pos
, unsigned len
, unsigned flags
,
3475 struct page
**pagep
, void **fsdata
)
3477 const struct address_space_operations
*aops
= mapping
->a_ops
;
3479 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
3482 EXPORT_SYMBOL(pagecache_write_begin
);
3484 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
3485 loff_t pos
, unsigned len
, unsigned copied
,
3486 struct page
*page
, void *fsdata
)
3488 const struct address_space_operations
*aops
= mapping
->a_ops
;
3490 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
3492 EXPORT_SYMBOL(pagecache_write_end
);
3495 * Warn about a page cache invalidation failure during a direct I/O write.
3497 void dio_warn_stale_pagecache(struct file
*filp
)
3499 static DEFINE_RATELIMIT_STATE(_rs
, 86400 * HZ
, DEFAULT_RATELIMIT_BURST
);
3503 errseq_set(&filp
->f_mapping
->wb_err
, -EIO
);
3504 if (__ratelimit(&_rs
)) {
3505 path
= file_path(filp
, pathname
, sizeof(pathname
));
3508 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3509 pr_crit("File: %s PID: %d Comm: %.20s\n", path
, current
->pid
,
3515 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3517 struct file
*file
= iocb
->ki_filp
;
3518 struct address_space
*mapping
= file
->f_mapping
;
3519 struct inode
*inode
= mapping
->host
;
3520 loff_t pos
= iocb
->ki_pos
;
3525 write_len
= iov_iter_count(from
);
3526 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
3528 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3529 /* If there are pages to writeback, return */
3530 if (filemap_range_has_page(file
->f_mapping
, pos
,
3531 pos
+ write_len
- 1))
3534 written
= filemap_write_and_wait_range(mapping
, pos
,
3535 pos
+ write_len
- 1);
3541 * After a write we want buffered reads to be sure to go to disk to get
3542 * the new data. We invalidate clean cached page from the region we're
3543 * about to write. We do this *before* the write so that we can return
3544 * without clobbering -EIOCBQUEUED from ->direct_IO().
3546 written
= invalidate_inode_pages2_range(mapping
,
3547 pos
>> PAGE_SHIFT
, end
);
3549 * If a page can not be invalidated, return 0 to fall back
3550 * to buffered write.
3553 if (written
== -EBUSY
)
3558 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3561 * Finally, try again to invalidate clean pages which might have been
3562 * cached by non-direct readahead, or faulted in by get_user_pages()
3563 * if the source of the write was an mmap'ed region of the file
3564 * we're writing. Either one is a pretty crazy thing to do,
3565 * so we don't support it 100%. If this invalidation
3566 * fails, tough, the write still worked...
3568 * Most of the time we do not need this since dio_complete() will do
3569 * the invalidation for us. However there are some file systems that
3570 * do not end up with dio_complete() being called, so let's not break
3571 * them by removing it completely.
3573 * Noticeable example is a blkdev_direct_IO().
3575 * Skip invalidation for async writes or if mapping has no pages.
3577 if (written
> 0 && mapping
->nrpages
&&
3578 invalidate_inode_pages2_range(mapping
, pos
>> PAGE_SHIFT
, end
))
3579 dio_warn_stale_pagecache(file
);
3583 write_len
-= written
;
3584 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3585 i_size_write(inode
, pos
);
3586 mark_inode_dirty(inode
);
3590 if (written
!= -EIOCBQUEUED
)
3591 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3595 EXPORT_SYMBOL(generic_file_direct_write
);
3598 * Find or create a page at the given pagecache position. Return the locked
3599 * page. This function is specifically for buffered writes.
3601 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
3602 pgoff_t index
, unsigned flags
)
3605 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
3607 if (flags
& AOP_FLAG_NOFS
)
3608 fgp_flags
|= FGP_NOFS
;
3610 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
3611 mapping_gfp_mask(mapping
));
3613 wait_for_stable_page(page
);
3617 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3619 ssize_t
generic_perform_write(struct file
*file
,
3620 struct iov_iter
*i
, loff_t pos
)
3622 struct address_space
*mapping
= file
->f_mapping
;
3623 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3625 ssize_t written
= 0;
3626 unsigned int flags
= 0;
3630 unsigned long offset
; /* Offset into pagecache page */
3631 unsigned long bytes
; /* Bytes to write to page */
3632 size_t copied
; /* Bytes copied from user */
3635 offset
= (pos
& (PAGE_SIZE
- 1));
3636 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3641 * Bring in the user page that we will copy from _first_.
3642 * Otherwise there's a nasty deadlock on copying from the
3643 * same page as we're writing to, without it being marked
3646 * Not only is this an optimisation, but it is also required
3647 * to check that the address is actually valid, when atomic
3648 * usercopies are used, below.
3650 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3655 if (fatal_signal_pending(current
)) {
3660 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3662 if (unlikely(status
< 0))
3665 if (mapping_writably_mapped(mapping
))
3666 flush_dcache_page(page
);
3668 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3669 flush_dcache_page(page
);
3671 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3673 if (unlikely(status
< 0))
3679 iov_iter_advance(i
, copied
);
3680 if (unlikely(copied
== 0)) {
3682 * If we were unable to copy any data at all, we must
3683 * fall back to a single segment length write.
3685 * If we didn't fallback here, we could livelock
3686 * because not all segments in the iov can be copied at
3687 * once without a pagefault.
3689 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3690 iov_iter_single_seg_count(i
));
3696 balance_dirty_pages_ratelimited(mapping
);
3697 } while (iov_iter_count(i
));
3699 return written
? written
: status
;
3701 EXPORT_SYMBOL(generic_perform_write
);
3704 * __generic_file_write_iter - write data to a file
3705 * @iocb: IO state structure (file, offset, etc.)
3706 * @from: iov_iter with data to write
3708 * This function does all the work needed for actually writing data to a
3709 * file. It does all basic checks, removes SUID from the file, updates
3710 * modification times and calls proper subroutines depending on whether we
3711 * do direct IO or a standard buffered write.
3713 * It expects i_mutex to be grabbed unless we work on a block device or similar
3714 * object which does not need locking at all.
3716 * This function does *not* take care of syncing data in case of O_SYNC write.
3717 * A caller has to handle it. This is mainly due to the fact that we want to
3718 * avoid syncing under i_mutex.
3721 * * number of bytes written, even for truncated writes
3722 * * negative error code if no data has been written at all
3724 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3726 struct file
*file
= iocb
->ki_filp
;
3727 struct address_space
*mapping
= file
->f_mapping
;
3728 struct inode
*inode
= mapping
->host
;
3729 ssize_t written
= 0;
3733 /* We can write back this queue in page reclaim */
3734 current
->backing_dev_info
= inode_to_bdi(inode
);
3735 err
= file_remove_privs(file
);
3739 err
= file_update_time(file
);
3743 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3744 loff_t pos
, endbyte
;
3746 written
= generic_file_direct_write(iocb
, from
);
3748 * If the write stopped short of completing, fall back to
3749 * buffered writes. Some filesystems do this for writes to
3750 * holes, for example. For DAX files, a buffered write will
3751 * not succeed (even if it did, DAX does not handle dirty
3752 * page-cache pages correctly).
3754 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3757 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3759 * If generic_perform_write() returned a synchronous error
3760 * then we want to return the number of bytes which were
3761 * direct-written, or the error code if that was zero. Note
3762 * that this differs from normal direct-io semantics, which
3763 * will return -EFOO even if some bytes were written.
3765 if (unlikely(status
< 0)) {
3770 * We need to ensure that the page cache pages are written to
3771 * disk and invalidated to preserve the expected O_DIRECT
3774 endbyte
= pos
+ status
- 1;
3775 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3777 iocb
->ki_pos
= endbyte
+ 1;
3779 invalidate_mapping_pages(mapping
,
3781 endbyte
>> PAGE_SHIFT
);
3784 * We don't know how much we wrote, so just return
3785 * the number of bytes which were direct-written
3789 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3790 if (likely(written
> 0))
3791 iocb
->ki_pos
+= written
;
3794 current
->backing_dev_info
= NULL
;
3795 return written
? written
: err
;
3797 EXPORT_SYMBOL(__generic_file_write_iter
);
3800 * generic_file_write_iter - write data to a file
3801 * @iocb: IO state structure
3802 * @from: iov_iter with data to write
3804 * This is a wrapper around __generic_file_write_iter() to be used by most
3805 * filesystems. It takes care of syncing the file in case of O_SYNC file
3806 * and acquires i_mutex as needed.
3808 * * negative error code if no data has been written at all of
3809 * vfs_fsync_range() failed for a synchronous write
3810 * * number of bytes written, even for truncated writes
3812 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3814 struct file
*file
= iocb
->ki_filp
;
3815 struct inode
*inode
= file
->f_mapping
->host
;
3819 ret
= generic_write_checks(iocb
, from
);
3821 ret
= __generic_file_write_iter(iocb
, from
);
3822 inode_unlock(inode
);
3825 ret
= generic_write_sync(iocb
, ret
);
3828 EXPORT_SYMBOL(generic_file_write_iter
);
3831 * try_to_release_page() - release old fs-specific metadata on a page
3833 * @page: the page which the kernel is trying to free
3834 * @gfp_mask: memory allocation flags (and I/O mode)
3836 * The address_space is to try to release any data against the page
3837 * (presumably at page->private).
3839 * This may also be called if PG_fscache is set on a page, indicating that the
3840 * page is known to the local caching routines.
3842 * The @gfp_mask argument specifies whether I/O may be performed to release
3843 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3845 * Return: %1 if the release was successful, otherwise return zero.
3847 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3849 struct address_space
* const mapping
= page
->mapping
;
3851 BUG_ON(!PageLocked(page
));
3852 if (PageWriteback(page
))
3855 if (mapping
&& mapping
->a_ops
->releasepage
)
3856 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3857 return try_to_free_buffers(page
);
3860 EXPORT_SYMBOL(try_to_release_page
);