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 */
147 mapping
->nrexceptional
+= nr
;
149 * Make sure the nrexceptional update is committed before
150 * the nrpages update so that final truncate racing
151 * with reclaim does not see both counters 0 at the
152 * same time and miss a shadow entry.
156 mapping
->nrpages
-= nr
;
159 static void unaccount_page_cache_page(struct address_space
*mapping
,
165 * if we're uptodate, flush out into the cleancache, otherwise
166 * invalidate any existing cleancache entries. We can't leave
167 * stale data around in the cleancache once our page is gone
169 if (PageUptodate(page
) && PageMappedToDisk(page
))
170 cleancache_put_page(page
);
172 cleancache_invalidate_page(mapping
, page
);
174 VM_BUG_ON_PAGE(PageTail(page
), page
);
175 VM_BUG_ON_PAGE(page_mapped(page
), page
);
176 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
179 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
180 current
->comm
, page_to_pfn(page
));
181 dump_page(page
, "still mapped when deleted");
183 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
185 mapcount
= page_mapcount(page
);
186 if (mapping_exiting(mapping
) &&
187 page_count(page
) >= mapcount
+ 2) {
189 * All vmas have already been torn down, so it's
190 * a good bet that actually the page is unmapped,
191 * and we'd prefer not to leak it: if we're wrong,
192 * some other bad page check should catch it later.
194 page_mapcount_reset(page
);
195 page_ref_sub(page
, mapcount
);
199 /* hugetlb pages do not participate in page cache accounting. */
203 nr
= thp_nr_pages(page
);
205 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, -nr
);
206 if (PageSwapBacked(page
)) {
207 __mod_lruvec_page_state(page
, NR_SHMEM
, -nr
);
208 if (PageTransHuge(page
))
209 __mod_lruvec_page_state(page
, NR_SHMEM_THPS
, -nr
);
210 } else if (PageTransHuge(page
)) {
211 __mod_lruvec_page_state(page
, NR_FILE_THPS
, -nr
);
212 filemap_nr_thps_dec(mapping
);
216 * At this point page must be either written or cleaned by
217 * truncate. Dirty page here signals a bug and loss of
220 * This fixes dirty accounting after removing the page entirely
221 * but leaves PageDirty set: it has no effect for truncated
222 * page and anyway will be cleared before returning page into
225 if (WARN_ON_ONCE(PageDirty(page
)))
226 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
230 * Delete a page from the page cache and free it. Caller has to make
231 * sure the page is locked and that nobody else uses it - or that usage
232 * is safe. The caller must hold the i_pages lock.
234 void __delete_from_page_cache(struct page
*page
, void *shadow
)
236 struct address_space
*mapping
= page
->mapping
;
238 trace_mm_filemap_delete_from_page_cache(page
);
240 unaccount_page_cache_page(mapping
, page
);
241 page_cache_delete(mapping
, page
, shadow
);
244 static void page_cache_free_page(struct address_space
*mapping
,
247 void (*freepage
)(struct page
*);
249 freepage
= mapping
->a_ops
->freepage
;
253 if (PageTransHuge(page
) && !PageHuge(page
)) {
254 page_ref_sub(page
, thp_nr_pages(page
));
255 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
262 * delete_from_page_cache - delete page from page cache
263 * @page: the page which the kernel is trying to remove from page cache
265 * This must be called only on pages that have been verified to be in the page
266 * cache and locked. It will never put the page into the free list, the caller
267 * has a reference on the page.
269 void delete_from_page_cache(struct page
*page
)
271 struct address_space
*mapping
= page_mapping(page
);
274 BUG_ON(!PageLocked(page
));
275 xa_lock_irqsave(&mapping
->i_pages
, flags
);
276 __delete_from_page_cache(page
, NULL
);
277 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
279 page_cache_free_page(mapping
, page
);
281 EXPORT_SYMBOL(delete_from_page_cache
);
284 * page_cache_delete_batch - delete several pages from page cache
285 * @mapping: the mapping to which pages belong
286 * @pvec: pagevec with pages to delete
288 * The function walks over mapping->i_pages and removes pages passed in @pvec
289 * from the mapping. The function expects @pvec to be sorted by page index
290 * and is optimised for it to be dense.
291 * It tolerates holes in @pvec (mapping entries at those indices are not
292 * modified). The function expects only THP head pages to be present in the
295 * The function expects the i_pages lock to be held.
297 static void page_cache_delete_batch(struct address_space
*mapping
,
298 struct pagevec
*pvec
)
300 XA_STATE(xas
, &mapping
->i_pages
, pvec
->pages
[0]->index
);
305 mapping_set_update(&xas
, mapping
);
306 xas_for_each(&xas
, page
, ULONG_MAX
) {
307 if (i
>= pagevec_count(pvec
))
310 /* A swap/dax/shadow entry got inserted? Skip it. */
311 if (xa_is_value(page
))
314 * A page got inserted in our range? Skip it. We have our
315 * pages locked so they are protected from being removed.
316 * If we see a page whose index is higher than ours, it
317 * means our page has been removed, which shouldn't be
318 * possible because we're holding the PageLock.
320 if (page
!= pvec
->pages
[i
]) {
321 VM_BUG_ON_PAGE(page
->index
> pvec
->pages
[i
]->index
,
326 WARN_ON_ONCE(!PageLocked(page
));
328 if (page
->index
== xas
.xa_index
)
329 page
->mapping
= NULL
;
330 /* Leave page->index set: truncation lookup relies on it */
333 * Move to the next page in the vector if this is a regular
334 * page or the index is of the last sub-page of this compound
337 if (page
->index
+ compound_nr(page
) - 1 == xas
.xa_index
)
339 xas_store(&xas
, NULL
);
342 mapping
->nrpages
-= total_pages
;
345 void delete_from_page_cache_batch(struct address_space
*mapping
,
346 struct pagevec
*pvec
)
351 if (!pagevec_count(pvec
))
354 xa_lock_irqsave(&mapping
->i_pages
, flags
);
355 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
356 trace_mm_filemap_delete_from_page_cache(pvec
->pages
[i
]);
358 unaccount_page_cache_page(mapping
, pvec
->pages
[i
]);
360 page_cache_delete_batch(mapping
, pvec
);
361 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
363 for (i
= 0; i
< pagevec_count(pvec
); i
++)
364 page_cache_free_page(mapping
, pvec
->pages
[i
]);
367 int filemap_check_errors(struct address_space
*mapping
)
370 /* Check for outstanding write errors */
371 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
372 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
374 if (test_bit(AS_EIO
, &mapping
->flags
) &&
375 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
379 EXPORT_SYMBOL(filemap_check_errors
);
381 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
383 /* Check for outstanding write errors */
384 if (test_bit(AS_EIO
, &mapping
->flags
))
386 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
392 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
393 * @mapping: address space structure to write
394 * @start: offset in bytes where the range starts
395 * @end: offset in bytes where the range ends (inclusive)
396 * @sync_mode: enable synchronous operation
398 * Start writeback against all of a mapping's dirty pages that lie
399 * within the byte offsets <start, end> inclusive.
401 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
402 * opposed to a regular memory cleansing writeback. The difference between
403 * these two operations is that if a dirty page/buffer is encountered, it must
404 * be waited upon, and not just skipped over.
406 * Return: %0 on success, negative error code otherwise.
408 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
409 loff_t end
, int sync_mode
)
412 struct writeback_control wbc
= {
413 .sync_mode
= sync_mode
,
414 .nr_to_write
= LONG_MAX
,
415 .range_start
= start
,
419 if (!mapping_can_writeback(mapping
) ||
420 !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
423 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
424 ret
= do_writepages(mapping
, &wbc
);
425 wbc_detach_inode(&wbc
);
429 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
432 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
435 int filemap_fdatawrite(struct address_space
*mapping
)
437 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
439 EXPORT_SYMBOL(filemap_fdatawrite
);
441 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
444 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
446 EXPORT_SYMBOL(filemap_fdatawrite_range
);
449 * filemap_flush - mostly a non-blocking flush
450 * @mapping: target address_space
452 * This is a mostly non-blocking flush. Not suitable for data-integrity
453 * purposes - I/O may not be started against all dirty pages.
455 * Return: %0 on success, negative error code otherwise.
457 int filemap_flush(struct address_space
*mapping
)
459 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
461 EXPORT_SYMBOL(filemap_flush
);
464 * filemap_range_has_page - check if a page exists in range.
465 * @mapping: address space within which to check
466 * @start_byte: offset in bytes where the range starts
467 * @end_byte: offset in bytes where the range ends (inclusive)
469 * Find at least one page in the range supplied, usually used to check if
470 * direct writing in this range will trigger a writeback.
472 * Return: %true if at least one page exists in the specified range,
475 bool filemap_range_has_page(struct address_space
*mapping
,
476 loff_t start_byte
, loff_t end_byte
)
479 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
480 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
482 if (end_byte
< start_byte
)
487 page
= xas_find(&xas
, max
);
488 if (xas_retry(&xas
, page
))
490 /* Shadow entries don't count */
491 if (xa_is_value(page
))
494 * We don't need to try to pin this page; we're about to
495 * release the RCU lock anyway. It is enough to know that
496 * there was a page here recently.
504 EXPORT_SYMBOL(filemap_range_has_page
);
506 static void __filemap_fdatawait_range(struct address_space
*mapping
,
507 loff_t start_byte
, loff_t end_byte
)
509 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
510 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
514 if (end_byte
< start_byte
)
518 while (index
<= end
) {
521 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
,
522 end
, PAGECACHE_TAG_WRITEBACK
);
526 for (i
= 0; i
< nr_pages
; i
++) {
527 struct page
*page
= pvec
.pages
[i
];
529 wait_on_page_writeback(page
);
530 ClearPageError(page
);
532 pagevec_release(&pvec
);
538 * filemap_fdatawait_range - wait for writeback to complete
539 * @mapping: address space structure to wait for
540 * @start_byte: offset in bytes where the range starts
541 * @end_byte: offset in bytes where the range ends (inclusive)
543 * Walk the list of under-writeback pages of the given address space
544 * in the given range and wait for all of them. Check error status of
545 * the address space and return it.
547 * Since the error status of the address space is cleared by this function,
548 * callers are responsible for checking the return value and handling and/or
549 * reporting the error.
551 * Return: error status of the address space.
553 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
556 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
557 return filemap_check_errors(mapping
);
559 EXPORT_SYMBOL(filemap_fdatawait_range
);
562 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
563 * @mapping: address space structure to wait for
564 * @start_byte: offset in bytes where the range starts
565 * @end_byte: offset in bytes where the range ends (inclusive)
567 * Walk the list of under-writeback pages of the given address space in the
568 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
569 * this function does not clear error status of the address space.
571 * Use this function if callers don't handle errors themselves. Expected
572 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
575 int filemap_fdatawait_range_keep_errors(struct address_space
*mapping
,
576 loff_t start_byte
, loff_t end_byte
)
578 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
579 return filemap_check_and_keep_errors(mapping
);
581 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors
);
584 * file_fdatawait_range - wait for writeback to complete
585 * @file: file pointing to address space structure to wait for
586 * @start_byte: offset in bytes where the range starts
587 * @end_byte: offset in bytes where the range ends (inclusive)
589 * Walk the list of under-writeback pages of the address space that file
590 * refers to, in the given range and wait for all of them. Check error
591 * status of the address space vs. the file->f_wb_err cursor and return it.
593 * Since the error status of the file is advanced by this function,
594 * callers are responsible for checking the return value and handling and/or
595 * reporting the error.
597 * Return: error status of the address space vs. the file->f_wb_err cursor.
599 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
601 struct address_space
*mapping
= file
->f_mapping
;
603 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
604 return file_check_and_advance_wb_err(file
);
606 EXPORT_SYMBOL(file_fdatawait_range
);
609 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
610 * @mapping: address space structure to wait for
612 * Walk the list of under-writeback pages of the given address space
613 * and wait for all of them. Unlike filemap_fdatawait(), this function
614 * does not clear error status of the address space.
616 * Use this function if callers don't handle errors themselves. Expected
617 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
620 * Return: error status of the address space.
622 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
624 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
625 return filemap_check_and_keep_errors(mapping
);
627 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
629 /* Returns true if writeback might be needed or already in progress. */
630 static bool mapping_needs_writeback(struct address_space
*mapping
)
632 if (dax_mapping(mapping
))
633 return mapping
->nrexceptional
;
635 return mapping
->nrpages
;
639 * filemap_write_and_wait_range - write out & wait on a file range
640 * @mapping: the address_space for the pages
641 * @lstart: offset in bytes where the range starts
642 * @lend: offset in bytes where the range ends (inclusive)
644 * Write out and wait upon file offsets lstart->lend, inclusive.
646 * Note that @lend is inclusive (describes the last byte to be written) so
647 * that this function can be used to write to the very end-of-file (end = -1).
649 * Return: error status of the address space.
651 int filemap_write_and_wait_range(struct address_space
*mapping
,
652 loff_t lstart
, loff_t lend
)
656 if (mapping_needs_writeback(mapping
)) {
657 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
660 * Even if the above returned error, the pages may be
661 * written partially (e.g. -ENOSPC), so we wait for it.
662 * But the -EIO is special case, it may indicate the worst
663 * thing (e.g. bug) happened, so we avoid waiting for it.
666 int err2
= filemap_fdatawait_range(mapping
,
671 /* Clear any previously stored errors */
672 filemap_check_errors(mapping
);
675 err
= filemap_check_errors(mapping
);
679 EXPORT_SYMBOL(filemap_write_and_wait_range
);
681 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
683 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
685 trace_filemap_set_wb_err(mapping
, eseq
);
687 EXPORT_SYMBOL(__filemap_set_wb_err
);
690 * file_check_and_advance_wb_err - report wb error (if any) that was previously
691 * and advance wb_err to current one
692 * @file: struct file on which the error is being reported
694 * When userland calls fsync (or something like nfsd does the equivalent), we
695 * want to report any writeback errors that occurred since the last fsync (or
696 * since the file was opened if there haven't been any).
698 * Grab the wb_err from the mapping. If it matches what we have in the file,
699 * then just quickly return 0. The file is all caught up.
701 * If it doesn't match, then take the mapping value, set the "seen" flag in
702 * it and try to swap it into place. If it works, or another task beat us
703 * to it with the new value, then update the f_wb_err and return the error
704 * portion. The error at this point must be reported via proper channels
705 * (a'la fsync, or NFS COMMIT operation, etc.).
707 * While we handle mapping->wb_err with atomic operations, the f_wb_err
708 * value is protected by the f_lock since we must ensure that it reflects
709 * the latest value swapped in for this file descriptor.
711 * Return: %0 on success, negative error code otherwise.
713 int file_check_and_advance_wb_err(struct file
*file
)
716 errseq_t old
= READ_ONCE(file
->f_wb_err
);
717 struct address_space
*mapping
= file
->f_mapping
;
719 /* Locklessly handle the common case where nothing has changed */
720 if (errseq_check(&mapping
->wb_err
, old
)) {
721 /* Something changed, must use slow path */
722 spin_lock(&file
->f_lock
);
723 old
= file
->f_wb_err
;
724 err
= errseq_check_and_advance(&mapping
->wb_err
,
726 trace_file_check_and_advance_wb_err(file
, old
);
727 spin_unlock(&file
->f_lock
);
731 * We're mostly using this function as a drop in replacement for
732 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
733 * that the legacy code would have had on these flags.
735 clear_bit(AS_EIO
, &mapping
->flags
);
736 clear_bit(AS_ENOSPC
, &mapping
->flags
);
739 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
742 * file_write_and_wait_range - write out & wait on a file range
743 * @file: file pointing to address_space with pages
744 * @lstart: offset in bytes where the range starts
745 * @lend: offset in bytes where the range ends (inclusive)
747 * Write out and wait upon file offsets lstart->lend, inclusive.
749 * Note that @lend is inclusive (describes the last byte to be written) so
750 * that this function can be used to write to the very end-of-file (end = -1).
752 * After writing out and waiting on the data, we check and advance the
753 * f_wb_err cursor to the latest value, and return any errors detected there.
755 * Return: %0 on success, negative error code otherwise.
757 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
760 struct address_space
*mapping
= file
->f_mapping
;
762 if (mapping_needs_writeback(mapping
)) {
763 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
765 /* See comment of filemap_write_and_wait() */
767 __filemap_fdatawait_range(mapping
, lstart
, lend
);
769 err2
= file_check_and_advance_wb_err(file
);
774 EXPORT_SYMBOL(file_write_and_wait_range
);
777 * replace_page_cache_page - replace a pagecache page with a new one
778 * @old: page to be replaced
779 * @new: page to replace with
781 * This function replaces a page in the pagecache with a new one. On
782 * success it acquires the pagecache reference for the new page and
783 * drops it for the old page. Both the old and new pages must be
784 * locked. This function does not add the new page to the LRU, the
785 * caller must do that.
787 * The remove + add is atomic. This function cannot fail.
789 void replace_page_cache_page(struct page
*old
, struct page
*new)
791 struct address_space
*mapping
= old
->mapping
;
792 void (*freepage
)(struct page
*) = mapping
->a_ops
->freepage
;
793 pgoff_t offset
= old
->index
;
794 XA_STATE(xas
, &mapping
->i_pages
, offset
);
797 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
798 VM_BUG_ON_PAGE(!PageLocked(new), new);
799 VM_BUG_ON_PAGE(new->mapping
, new);
802 new->mapping
= mapping
;
805 mem_cgroup_migrate(old
, new);
807 xas_lock_irqsave(&xas
, flags
);
808 xas_store(&xas
, new);
811 /* hugetlb pages do not participate in page cache accounting. */
813 __dec_lruvec_page_state(old
, NR_FILE_PAGES
);
815 __inc_lruvec_page_state(new, NR_FILE_PAGES
);
816 if (PageSwapBacked(old
))
817 __dec_lruvec_page_state(old
, NR_SHMEM
);
818 if (PageSwapBacked(new))
819 __inc_lruvec_page_state(new, NR_SHMEM
);
820 xas_unlock_irqrestore(&xas
, flags
);
825 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
827 noinline
int __add_to_page_cache_locked(struct page
*page
,
828 struct address_space
*mapping
,
829 pgoff_t offset
, gfp_t gfp
,
832 XA_STATE(xas
, &mapping
->i_pages
, offset
);
833 int huge
= PageHuge(page
);
835 bool charged
= false;
837 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
838 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
839 mapping_set_update(&xas
, mapping
);
842 page
->mapping
= mapping
;
843 page
->index
= offset
;
846 error
= mem_cgroup_charge(page
, current
->mm
, gfp
);
852 gfp
&= GFP_RECLAIM_MASK
;
855 unsigned int order
= xa_get_order(xas
.xa
, xas
.xa_index
);
856 void *entry
, *old
= NULL
;
858 if (order
> thp_order(page
))
859 xas_split_alloc(&xas
, xa_load(xas
.xa
, xas
.xa_index
),
862 xas_for_each_conflict(&xas
, entry
) {
864 if (!xa_is_value(entry
)) {
865 xas_set_err(&xas
, -EEXIST
);
873 /* entry may have been split before we acquired lock */
874 order
= xa_get_order(xas
.xa
, xas
.xa_index
);
875 if (order
> thp_order(page
)) {
876 xas_split(&xas
, old
, order
);
881 xas_store(&xas
, page
);
886 mapping
->nrexceptional
--;
889 /* hugetlb pages do not participate in page cache accounting */
891 __inc_lruvec_page_state(page
, NR_FILE_PAGES
);
893 xas_unlock_irq(&xas
);
894 } while (xas_nomem(&xas
, gfp
));
896 if (xas_error(&xas
)) {
897 error
= xas_error(&xas
);
899 mem_cgroup_uncharge(page
);
903 trace_mm_filemap_add_to_page_cache(page
);
906 page
->mapping
= NULL
;
907 /* Leave page->index set: truncation relies upon it */
911 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked
, ERRNO
);
914 * add_to_page_cache_locked - add a locked page to the pagecache
916 * @mapping: the page's address_space
917 * @offset: page index
918 * @gfp_mask: page allocation mode
920 * This function is used to add a page to the pagecache. It must be locked.
921 * This function does not add the page to the LRU. The caller must do that.
923 * Return: %0 on success, negative error code otherwise.
925 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
926 pgoff_t offset
, gfp_t gfp_mask
)
928 return __add_to_page_cache_locked(page
, mapping
, offset
,
931 EXPORT_SYMBOL(add_to_page_cache_locked
);
933 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
934 pgoff_t offset
, gfp_t gfp_mask
)
939 __SetPageLocked(page
);
940 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
943 __ClearPageLocked(page
);
946 * The page might have been evicted from cache only
947 * recently, in which case it should be activated like
948 * any other repeatedly accessed page.
949 * The exception is pages getting rewritten; evicting other
950 * data from the working set, only to cache data that will
951 * get overwritten with something else, is a waste of memory.
953 WARN_ON_ONCE(PageActive(page
));
954 if (!(gfp_mask
& __GFP_WRITE
) && shadow
)
955 workingset_refault(page
, shadow
);
960 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
963 struct page
*__page_cache_alloc(gfp_t gfp
)
968 if (cpuset_do_page_mem_spread()) {
969 unsigned int cpuset_mems_cookie
;
971 cpuset_mems_cookie
= read_mems_allowed_begin();
972 n
= cpuset_mem_spread_node();
973 page
= __alloc_pages_node(n
, gfp
, 0);
974 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
978 return alloc_pages(gfp
, 0);
980 EXPORT_SYMBOL(__page_cache_alloc
);
984 * In order to wait for pages to become available there must be
985 * waitqueues associated with pages. By using a hash table of
986 * waitqueues where the bucket discipline is to maintain all
987 * waiters on the same queue and wake all when any of the pages
988 * become available, and for the woken contexts to check to be
989 * sure the appropriate page became available, this saves space
990 * at a cost of "thundering herd" phenomena during rare hash
993 #define PAGE_WAIT_TABLE_BITS 8
994 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
995 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
997 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
999 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
1002 void __init
pagecache_init(void)
1006 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
1007 init_waitqueue_head(&page_wait_table
[i
]);
1009 page_writeback_init();
1013 * The page wait code treats the "wait->flags" somewhat unusually, because
1014 * we have multiple different kinds of waits, not just the usual "exclusive"
1019 * (a) no special bits set:
1021 * We're just waiting for the bit to be released, and when a waker
1022 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1023 * and remove it from the wait queue.
1025 * Simple and straightforward.
1027 * (b) WQ_FLAG_EXCLUSIVE:
1029 * The waiter is waiting to get the lock, and only one waiter should
1030 * be woken up to avoid any thundering herd behavior. We'll set the
1031 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1033 * This is the traditional exclusive wait.
1035 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1037 * The waiter is waiting to get the bit, and additionally wants the
1038 * lock to be transferred to it for fair lock behavior. If the lock
1039 * cannot be taken, we stop walking the wait queue without waking
1042 * This is the "fair lock handoff" case, and in addition to setting
1043 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1044 * that it now has the lock.
1046 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
1049 struct wait_page_key
*key
= arg
;
1050 struct wait_page_queue
*wait_page
1051 = container_of(wait
, struct wait_page_queue
, wait
);
1053 if (!wake_page_match(wait_page
, key
))
1057 * If it's a lock handoff wait, we get the bit for it, and
1058 * stop walking (and do not wake it up) if we can't.
1060 flags
= wait
->flags
;
1061 if (flags
& WQ_FLAG_EXCLUSIVE
) {
1062 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
1064 if (flags
& WQ_FLAG_CUSTOM
) {
1065 if (test_and_set_bit(key
->bit_nr
, &key
->page
->flags
))
1067 flags
|= WQ_FLAG_DONE
;
1072 * We are holding the wait-queue lock, but the waiter that
1073 * is waiting for this will be checking the flags without
1076 * So update the flags atomically, and wake up the waiter
1077 * afterwards to avoid any races. This store-release pairs
1078 * with the load-acquire in wait_on_page_bit_common().
1080 smp_store_release(&wait
->flags
, flags
| WQ_FLAG_WOKEN
);
1081 wake_up_state(wait
->private, mode
);
1084 * Ok, we have successfully done what we're waiting for,
1085 * and we can unconditionally remove the wait entry.
1087 * Note that this pairs with the "finish_wait()" in the
1088 * waiter, and has to be the absolute last thing we do.
1089 * After this list_del_init(&wait->entry) the wait entry
1090 * might be de-allocated and the process might even have
1093 list_del_init_careful(&wait
->entry
);
1094 return (flags
& WQ_FLAG_EXCLUSIVE
) != 0;
1097 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
1099 wait_queue_head_t
*q
= page_waitqueue(page
);
1100 struct wait_page_key key
;
1101 unsigned long flags
;
1102 wait_queue_entry_t bookmark
;
1105 key
.bit_nr
= bit_nr
;
1109 bookmark
.private = NULL
;
1110 bookmark
.func
= NULL
;
1111 INIT_LIST_HEAD(&bookmark
.entry
);
1113 spin_lock_irqsave(&q
->lock
, flags
);
1114 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1116 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1118 * Take a breather from holding the lock,
1119 * allow pages that finish wake up asynchronously
1120 * to acquire the lock and remove themselves
1123 spin_unlock_irqrestore(&q
->lock
, flags
);
1125 spin_lock_irqsave(&q
->lock
, flags
);
1126 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1130 * It is possible for other pages to have collided on the waitqueue
1131 * hash, so in that case check for a page match. That prevents a long-
1134 * It is still possible to miss a case here, when we woke page waiters
1135 * and removed them from the waitqueue, but there are still other
1138 if (!waitqueue_active(q
) || !key
.page_match
) {
1139 ClearPageWaiters(page
);
1141 * It's possible to miss clearing Waiters here, when we woke
1142 * our page waiters, but the hashed waitqueue has waiters for
1143 * other pages on it.
1145 * That's okay, it's a rare case. The next waker will clear it.
1148 spin_unlock_irqrestore(&q
->lock
, flags
);
1151 static void wake_up_page(struct page
*page
, int bit
)
1153 if (!PageWaiters(page
))
1155 wake_up_page_bit(page
, bit
);
1159 * A choice of three behaviors for wait_on_page_bit_common():
1162 EXCLUSIVE
, /* Hold ref to page and take the bit when woken, like
1163 * __lock_page() waiting on then setting PG_locked.
1165 SHARED
, /* Hold ref to page and check the bit when woken, like
1166 * wait_on_page_writeback() waiting on PG_writeback.
1168 DROP
, /* Drop ref to page before wait, no check when woken,
1169 * like put_and_wait_on_page_locked() on PG_locked.
1174 * Attempt to check (or get) the page bit, and mark us done
1177 static inline bool trylock_page_bit_common(struct page
*page
, int bit_nr
,
1178 struct wait_queue_entry
*wait
)
1180 if (wait
->flags
& WQ_FLAG_EXCLUSIVE
) {
1181 if (test_and_set_bit(bit_nr
, &page
->flags
))
1183 } else if (test_bit(bit_nr
, &page
->flags
))
1186 wait
->flags
|= WQ_FLAG_WOKEN
| WQ_FLAG_DONE
;
1190 /* How many times do we accept lock stealing from under a waiter? */
1191 int sysctl_page_lock_unfairness
= 5;
1193 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
1194 struct page
*page
, int bit_nr
, int state
, enum behavior behavior
)
1196 int unfairness
= sysctl_page_lock_unfairness
;
1197 struct wait_page_queue wait_page
;
1198 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1199 bool thrashing
= false;
1200 bool delayacct
= false;
1201 unsigned long pflags
;
1203 if (bit_nr
== PG_locked
&&
1204 !PageUptodate(page
) && PageWorkingset(page
)) {
1205 if (!PageSwapBacked(page
)) {
1206 delayacct_thrashing_start();
1209 psi_memstall_enter(&pflags
);
1214 wait
->func
= wake_page_function
;
1215 wait_page
.page
= page
;
1216 wait_page
.bit_nr
= bit_nr
;
1220 if (behavior
== EXCLUSIVE
) {
1221 wait
->flags
= WQ_FLAG_EXCLUSIVE
;
1222 if (--unfairness
< 0)
1223 wait
->flags
|= WQ_FLAG_CUSTOM
;
1227 * Do one last check whether we can get the
1228 * page bit synchronously.
1230 * Do the SetPageWaiters() marking before that
1231 * to let any waker we _just_ missed know they
1232 * need to wake us up (otherwise they'll never
1233 * even go to the slow case that looks at the
1234 * page queue), and add ourselves to the wait
1235 * queue if we need to sleep.
1237 * This part needs to be done under the queue
1238 * lock to avoid races.
1240 spin_lock_irq(&q
->lock
);
1241 SetPageWaiters(page
);
1242 if (!trylock_page_bit_common(page
, bit_nr
, wait
))
1243 __add_wait_queue_entry_tail(q
, wait
);
1244 spin_unlock_irq(&q
->lock
);
1247 * From now on, all the logic will be based on
1248 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1249 * see whether the page bit testing has already
1250 * been done by the wake function.
1252 * We can drop our reference to the page.
1254 if (behavior
== DROP
)
1258 * Note that until the "finish_wait()", or until
1259 * we see the WQ_FLAG_WOKEN flag, we need to
1260 * be very careful with the 'wait->flags', because
1261 * we may race with a waker that sets them.
1266 set_current_state(state
);
1268 /* Loop until we've been woken or interrupted */
1269 flags
= smp_load_acquire(&wait
->flags
);
1270 if (!(flags
& WQ_FLAG_WOKEN
)) {
1271 if (signal_pending_state(state
, current
))
1278 /* If we were non-exclusive, we're done */
1279 if (behavior
!= EXCLUSIVE
)
1282 /* If the waker got the lock for us, we're done */
1283 if (flags
& WQ_FLAG_DONE
)
1287 * Otherwise, if we're getting the lock, we need to
1288 * try to get it ourselves.
1290 * And if that fails, we'll have to retry this all.
1292 if (unlikely(test_and_set_bit(bit_nr
, &page
->flags
)))
1295 wait
->flags
|= WQ_FLAG_DONE
;
1300 * If a signal happened, this 'finish_wait()' may remove the last
1301 * waiter from the wait-queues, but the PageWaiters bit will remain
1302 * set. That's ok. The next wakeup will take care of it, and trying
1303 * to do it here would be difficult and prone to races.
1305 finish_wait(q
, wait
);
1309 delayacct_thrashing_end();
1310 psi_memstall_leave(&pflags
);
1314 * NOTE! The wait->flags weren't stable until we've done the
1315 * 'finish_wait()', and we could have exited the loop above due
1316 * to a signal, and had a wakeup event happen after the signal
1317 * test but before the 'finish_wait()'.
1319 * So only after the finish_wait() can we reliably determine
1320 * if we got woken up or not, so we can now figure out the final
1321 * return value based on that state without races.
1323 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1324 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1326 if (behavior
== EXCLUSIVE
)
1327 return wait
->flags
& WQ_FLAG_DONE
? 0 : -EINTR
;
1329 return wait
->flags
& WQ_FLAG_WOKEN
? 0 : -EINTR
;
1332 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1334 wait_queue_head_t
*q
= page_waitqueue(page
);
1335 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, SHARED
);
1337 EXPORT_SYMBOL(wait_on_page_bit
);
1339 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1341 wait_queue_head_t
*q
= page_waitqueue(page
);
1342 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, SHARED
);
1344 EXPORT_SYMBOL(wait_on_page_bit_killable
);
1347 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1348 * @page: The page to wait for.
1349 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1351 * The caller should hold a reference on @page. They expect the page to
1352 * become unlocked relatively soon, but do not wish to hold up migration
1353 * (for example) by holding the reference while waiting for the page to
1354 * come unlocked. After this function returns, the caller should not
1355 * dereference @page.
1357 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1359 int put_and_wait_on_page_locked(struct page
*page
, int state
)
1361 wait_queue_head_t
*q
;
1363 page
= compound_head(page
);
1364 q
= page_waitqueue(page
);
1365 return wait_on_page_bit_common(q
, page
, PG_locked
, state
, DROP
);
1369 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1370 * @page: Page defining the wait queue of interest
1371 * @waiter: Waiter to add to the queue
1373 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1375 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1377 wait_queue_head_t
*q
= page_waitqueue(page
);
1378 unsigned long flags
;
1380 spin_lock_irqsave(&q
->lock
, flags
);
1381 __add_wait_queue_entry_tail(q
, waiter
);
1382 SetPageWaiters(page
);
1383 spin_unlock_irqrestore(&q
->lock
, flags
);
1385 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1387 #ifndef clear_bit_unlock_is_negative_byte
1390 * PG_waiters is the high bit in the same byte as PG_lock.
1392 * On x86 (and on many other architectures), we can clear PG_lock and
1393 * test the sign bit at the same time. But if the architecture does
1394 * not support that special operation, we just do this all by hand
1397 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1398 * being cleared, but a memory barrier should be unnecessary since it is
1399 * in the same byte as PG_locked.
1401 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1403 clear_bit_unlock(nr
, mem
);
1404 /* smp_mb__after_atomic(); */
1405 return test_bit(PG_waiters
, mem
);
1411 * unlock_page - unlock a locked page
1414 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1415 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1416 * mechanism between PageLocked pages and PageWriteback pages is shared.
1417 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1419 * Note that this depends on PG_waiters being the sign bit in the byte
1420 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1421 * clear the PG_locked bit and test PG_waiters at the same time fairly
1422 * portably (architectures that do LL/SC can test any bit, while x86 can
1423 * test the sign bit).
1425 void unlock_page(struct page
*page
)
1427 BUILD_BUG_ON(PG_waiters
!= 7);
1428 page
= compound_head(page
);
1429 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1430 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1431 wake_up_page_bit(page
, PG_locked
);
1433 EXPORT_SYMBOL(unlock_page
);
1436 * end_page_writeback - end writeback against a page
1439 void end_page_writeback(struct page
*page
)
1442 * TestClearPageReclaim could be used here but it is an atomic
1443 * operation and overkill in this particular case. Failing to
1444 * shuffle a page marked for immediate reclaim is too mild to
1445 * justify taking an atomic operation penalty at the end of
1446 * ever page writeback.
1448 if (PageReclaim(page
)) {
1449 ClearPageReclaim(page
);
1450 rotate_reclaimable_page(page
);
1454 * Writeback does not hold a page reference of its own, relying
1455 * on truncation to wait for the clearing of PG_writeback.
1456 * But here we must make sure that the page is not freed and
1457 * reused before the wake_up_page().
1460 if (!test_clear_page_writeback(page
))
1463 smp_mb__after_atomic();
1464 wake_up_page(page
, PG_writeback
);
1467 EXPORT_SYMBOL(end_page_writeback
);
1470 * After completing I/O on a page, call this routine to update the page
1471 * flags appropriately
1473 void page_endio(struct page
*page
, bool is_write
, int err
)
1477 SetPageUptodate(page
);
1479 ClearPageUptodate(page
);
1485 struct address_space
*mapping
;
1488 mapping
= page_mapping(page
);
1490 mapping_set_error(mapping
, err
);
1492 end_page_writeback(page
);
1495 EXPORT_SYMBOL_GPL(page_endio
);
1498 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1499 * @__page: the page to lock
1501 void __lock_page(struct page
*__page
)
1503 struct page
*page
= compound_head(__page
);
1504 wait_queue_head_t
*q
= page_waitqueue(page
);
1505 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
,
1508 EXPORT_SYMBOL(__lock_page
);
1510 int __lock_page_killable(struct page
*__page
)
1512 struct page
*page
= compound_head(__page
);
1513 wait_queue_head_t
*q
= page_waitqueue(page
);
1514 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
,
1517 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1519 int __lock_page_async(struct page
*page
, struct wait_page_queue
*wait
)
1521 struct wait_queue_head
*q
= page_waitqueue(page
);
1525 wait
->bit_nr
= PG_locked
;
1527 spin_lock_irq(&q
->lock
);
1528 __add_wait_queue_entry_tail(q
, &wait
->wait
);
1529 SetPageWaiters(page
);
1530 ret
= !trylock_page(page
);
1532 * If we were successful now, we know we're still on the
1533 * waitqueue as we're still under the lock. This means it's
1534 * safe to remove and return success, we know the callback
1535 * isn't going to trigger.
1538 __remove_wait_queue(q
, &wait
->wait
);
1541 spin_unlock_irq(&q
->lock
);
1547 * 1 - page is locked; mmap_lock is still held.
1548 * 0 - page is not locked.
1549 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1550 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1551 * which case mmap_lock is still held.
1553 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1554 * with the page locked and the mmap_lock unperturbed.
1556 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1559 if (fault_flag_allow_retry_first(flags
)) {
1561 * CAUTION! In this case, mmap_lock is not released
1562 * even though return 0.
1564 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1567 mmap_read_unlock(mm
);
1568 if (flags
& FAULT_FLAG_KILLABLE
)
1569 wait_on_page_locked_killable(page
);
1571 wait_on_page_locked(page
);
1574 if (flags
& FAULT_FLAG_KILLABLE
) {
1577 ret
= __lock_page_killable(page
);
1579 mmap_read_unlock(mm
);
1590 * page_cache_next_miss() - Find the next gap in the page cache.
1591 * @mapping: Mapping.
1593 * @max_scan: Maximum range to search.
1595 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1596 * gap with the lowest index.
1598 * This function may be called under the rcu_read_lock. However, this will
1599 * not atomically search a snapshot of the cache at a single point in time.
1600 * For example, if a gap is created at index 5, then subsequently a gap is
1601 * created at index 10, page_cache_next_miss covering both indices may
1602 * return 10 if called under the rcu_read_lock.
1604 * Return: The index of the gap if found, otherwise an index outside the
1605 * range specified (in which case 'return - index >= max_scan' will be true).
1606 * In the rare case of index wrap-around, 0 will be returned.
1608 pgoff_t
page_cache_next_miss(struct address_space
*mapping
,
1609 pgoff_t index
, unsigned long max_scan
)
1611 XA_STATE(xas
, &mapping
->i_pages
, index
);
1613 while (max_scan
--) {
1614 void *entry
= xas_next(&xas
);
1615 if (!entry
|| xa_is_value(entry
))
1617 if (xas
.xa_index
== 0)
1621 return xas
.xa_index
;
1623 EXPORT_SYMBOL(page_cache_next_miss
);
1626 * page_cache_prev_miss() - Find the previous gap in the page cache.
1627 * @mapping: Mapping.
1629 * @max_scan: Maximum range to search.
1631 * Search the range [max(index - max_scan + 1, 0), index] for the
1632 * gap with the highest index.
1634 * This function may be called under the rcu_read_lock. However, this will
1635 * not atomically search a snapshot of the cache at a single point in time.
1636 * For example, if a gap is created at index 10, then subsequently a gap is
1637 * created at index 5, page_cache_prev_miss() covering both indices may
1638 * return 5 if called under the rcu_read_lock.
1640 * Return: The index of the gap if found, otherwise an index outside the
1641 * range specified (in which case 'index - return >= max_scan' will be true).
1642 * In the rare case of wrap-around, ULONG_MAX will be returned.
1644 pgoff_t
page_cache_prev_miss(struct address_space
*mapping
,
1645 pgoff_t index
, unsigned long max_scan
)
1647 XA_STATE(xas
, &mapping
->i_pages
, index
);
1649 while (max_scan
--) {
1650 void *entry
= xas_prev(&xas
);
1651 if (!entry
|| xa_is_value(entry
))
1653 if (xas
.xa_index
== ULONG_MAX
)
1657 return xas
.xa_index
;
1659 EXPORT_SYMBOL(page_cache_prev_miss
);
1662 * mapping_get_entry - Get a page cache entry.
1663 * @mapping: the address_space to search
1664 * @index: The page cache index.
1666 * Looks up the page cache slot at @mapping & @offset. If there is a
1667 * page cache page, the head page is returned with an increased refcount.
1669 * If the slot holds a shadow entry of a previously evicted page, or a
1670 * swap entry from shmem/tmpfs, it is returned.
1672 * Return: The head page or shadow entry, %NULL if nothing is found.
1674 static struct page
*mapping_get_entry(struct address_space
*mapping
,
1677 XA_STATE(xas
, &mapping
->i_pages
, index
);
1683 page
= xas_load(&xas
);
1684 if (xas_retry(&xas
, page
))
1687 * A shadow entry of a recently evicted page, or a swap entry from
1688 * shmem/tmpfs. Return it without attempting to raise page count.
1690 if (!page
|| xa_is_value(page
))
1693 if (!page_cache_get_speculative(page
))
1697 * Has the page moved or been split?
1698 * This is part of the lockless pagecache protocol. See
1699 * include/linux/pagemap.h for details.
1701 if (unlikely(page
!= xas_reload(&xas
))) {
1712 * pagecache_get_page - Find and get a reference to a page.
1713 * @mapping: The address_space to search.
1714 * @index: The page index.
1715 * @fgp_flags: %FGP flags modify how the page is returned.
1716 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1718 * Looks up the page cache entry at @mapping & @index.
1720 * @fgp_flags can be zero or more of these flags:
1722 * * %FGP_ACCESSED - The page will be marked accessed.
1723 * * %FGP_LOCK - The page is returned locked.
1724 * * %FGP_HEAD - If the page is present and a THP, return the head page
1725 * rather than the exact page specified by the index.
1726 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1727 * instead of allocating a new page to replace it.
1728 * * %FGP_CREAT - If no page is present then a new page is allocated using
1729 * @gfp_mask and added to the page cache and the VM's LRU list.
1730 * The page is returned locked and with an increased refcount.
1731 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1732 * page is already in cache. If the page was allocated, unlock it before
1733 * returning so the caller can do the same dance.
1734 * * %FGP_WRITE - The page will be written
1735 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1736 * * %FGP_NOWAIT - Don't get blocked by page lock
1738 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1739 * if the %GFP flags specified for %FGP_CREAT are atomic.
1741 * If there is a page cache page, it is returned with an increased refcount.
1743 * Return: The found page or %NULL otherwise.
1745 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t index
,
1746 int fgp_flags
, gfp_t gfp_mask
)
1751 page
= mapping_get_entry(mapping
, index
);
1752 if (xa_is_value(page
)) {
1753 if (fgp_flags
& FGP_ENTRY
)
1760 if (fgp_flags
& FGP_LOCK
) {
1761 if (fgp_flags
& FGP_NOWAIT
) {
1762 if (!trylock_page(page
)) {
1770 /* Has the page been truncated? */
1771 if (unlikely(page
->mapping
!= mapping
)) {
1776 VM_BUG_ON_PAGE(!thp_contains(page
, index
), page
);
1779 if (fgp_flags
& FGP_ACCESSED
)
1780 mark_page_accessed(page
);
1781 else if (fgp_flags
& FGP_WRITE
) {
1782 /* Clear idle flag for buffer write */
1783 if (page_is_idle(page
))
1784 clear_page_idle(page
);
1786 if (!(fgp_flags
& FGP_HEAD
))
1787 page
= find_subpage(page
, index
);
1790 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1792 if ((fgp_flags
& FGP_WRITE
) && mapping_can_writeback(mapping
))
1793 gfp_mask
|= __GFP_WRITE
;
1794 if (fgp_flags
& FGP_NOFS
)
1795 gfp_mask
&= ~__GFP_FS
;
1797 page
= __page_cache_alloc(gfp_mask
);
1801 if (WARN_ON_ONCE(!(fgp_flags
& (FGP_LOCK
| FGP_FOR_MMAP
))))
1802 fgp_flags
|= FGP_LOCK
;
1804 /* Init accessed so avoid atomic mark_page_accessed later */
1805 if (fgp_flags
& FGP_ACCESSED
)
1806 __SetPageReferenced(page
);
1808 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp_mask
);
1809 if (unlikely(err
)) {
1817 * add_to_page_cache_lru locks the page, and for mmap we expect
1820 if (page
&& (fgp_flags
& FGP_FOR_MMAP
))
1826 EXPORT_SYMBOL(pagecache_get_page
);
1828 static inline struct page
*find_get_entry(struct xa_state
*xas
, pgoff_t max
,
1834 if (mark
== XA_PRESENT
)
1835 page
= xas_find(xas
, max
);
1837 page
= xas_find_marked(xas
, max
, mark
);
1839 if (xas_retry(xas
, page
))
1842 * A shadow entry of a recently evicted page, a swap
1843 * entry from shmem/tmpfs or a DAX entry. Return it
1844 * without attempting to raise page count.
1846 if (!page
|| xa_is_value(page
))
1849 if (!page_cache_get_speculative(page
))
1852 /* Has the page moved or been split? */
1853 if (unlikely(page
!= xas_reload(xas
))) {
1865 * find_get_entries - gang pagecache lookup
1866 * @mapping: The address_space to search
1867 * @start: The starting page cache index
1868 * @end: The final page index (inclusive).
1869 * @pvec: Where the resulting entries are placed.
1870 * @indices: The cache indices corresponding to the entries in @entries
1872 * find_get_entries() will search for and return a batch of entries in
1873 * the mapping. The entries are placed in @pvec. find_get_entries()
1874 * takes a reference on any actual pages it returns.
1876 * The search returns a group of mapping-contiguous page cache entries
1877 * with ascending indexes. There may be holes in the indices due to
1878 * not-present pages.
1880 * Any shadow entries of evicted pages, or swap entries from
1881 * shmem/tmpfs, are included in the returned array.
1883 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1884 * stops at that page: the caller is likely to have a better way to handle
1885 * the compound page as a whole, and then skip its extent, than repeatedly
1886 * calling find_get_entries() to return all its tails.
1888 * Return: the number of pages and shadow entries which were found.
1890 unsigned find_get_entries(struct address_space
*mapping
, pgoff_t start
,
1891 pgoff_t end
, struct pagevec
*pvec
, pgoff_t
*indices
)
1893 XA_STATE(xas
, &mapping
->i_pages
, start
);
1895 unsigned int ret
= 0;
1896 unsigned nr_entries
= PAGEVEC_SIZE
;
1899 while ((page
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
1901 * Terminate early on finding a THP, to allow the caller to
1902 * handle it all at once; but continue if this is hugetlbfs.
1904 if (!xa_is_value(page
) && PageTransHuge(page
) &&
1906 page
= find_subpage(page
, xas
.xa_index
);
1907 nr_entries
= ret
+ 1;
1910 indices
[ret
] = xas
.xa_index
;
1911 pvec
->pages
[ret
] = page
;
1912 if (++ret
== nr_entries
)
1922 * find_lock_entries - Find a batch of pagecache entries.
1923 * @mapping: The address_space to search.
1924 * @start: The starting page cache index.
1925 * @end: The final page index (inclusive).
1926 * @pvec: Where the resulting entries are placed.
1927 * @indices: The cache indices of the entries in @pvec.
1929 * find_lock_entries() will return a batch of entries from @mapping.
1930 * Swap, shadow and DAX entries are included. Pages are returned
1931 * locked and with an incremented refcount. Pages which are locked by
1932 * somebody else or under writeback are skipped. Only the head page of
1933 * a THP is returned. Pages which are partially outside the range are
1936 * The entries have ascending indexes. The indices may not be consecutive
1937 * due to not-present entries, THP pages, pages which could not be locked
1938 * or pages under writeback.
1940 * Return: The number of entries which were found.
1942 unsigned find_lock_entries(struct address_space
*mapping
, pgoff_t start
,
1943 pgoff_t end
, struct pagevec
*pvec
, pgoff_t
*indices
)
1945 XA_STATE(xas
, &mapping
->i_pages
, start
);
1949 while ((page
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
1950 if (!xa_is_value(page
)) {
1951 if (page
->index
< start
)
1953 VM_BUG_ON_PAGE(page
->index
!= xas
.xa_index
, page
);
1954 if (page
->index
+ thp_nr_pages(page
) - 1 > end
)
1956 if (!trylock_page(page
))
1958 if (page
->mapping
!= mapping
|| PageWriteback(page
))
1960 VM_BUG_ON_PAGE(!thp_contains(page
, xas
.xa_index
),
1963 indices
[pvec
->nr
] = xas
.xa_index
;
1964 if (!pagevec_add(pvec
, page
))
1972 if (!xa_is_value(page
) && PageTransHuge(page
))
1973 xas_set(&xas
, page
->index
+ thp_nr_pages(page
));
1977 return pagevec_count(pvec
);
1981 * find_get_pages_range - gang pagecache lookup
1982 * @mapping: The address_space to search
1983 * @start: The starting page index
1984 * @end: The final page index (inclusive)
1985 * @nr_pages: The maximum number of pages
1986 * @pages: Where the resulting pages are placed
1988 * find_get_pages_range() will search for and return a group of up to @nr_pages
1989 * pages in the mapping starting at index @start and up to index @end
1990 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1991 * a reference against the returned pages.
1993 * The search returns a group of mapping-contiguous pages with ascending
1994 * indexes. There may be holes in the indices due to not-present pages.
1995 * We also update @start to index the next page for the traversal.
1997 * Return: the number of pages which were found. If this number is
1998 * smaller than @nr_pages, the end of specified range has been
2001 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
2002 pgoff_t end
, unsigned int nr_pages
,
2003 struct page
**pages
)
2005 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2009 if (unlikely(!nr_pages
))
2013 while ((page
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
2014 /* Skip over shadow, swap and DAX entries */
2015 if (xa_is_value(page
))
2018 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
2019 if (++ret
== nr_pages
) {
2020 *start
= xas
.xa_index
+ 1;
2026 * We come here when there is no page beyond @end. We take care to not
2027 * overflow the index @start as it confuses some of the callers. This
2028 * breaks the iteration when there is a page at index -1 but that is
2029 * already broken anyway.
2031 if (end
== (pgoff_t
)-1)
2032 *start
= (pgoff_t
)-1;
2042 * find_get_pages_contig - gang contiguous pagecache lookup
2043 * @mapping: The address_space to search
2044 * @index: The starting page index
2045 * @nr_pages: The maximum number of pages
2046 * @pages: Where the resulting pages are placed
2048 * find_get_pages_contig() works exactly like find_get_pages(), except
2049 * that the returned number of pages are guaranteed to be contiguous.
2051 * Return: the number of pages which were found.
2053 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
2054 unsigned int nr_pages
, struct page
**pages
)
2056 XA_STATE(xas
, &mapping
->i_pages
, index
);
2058 unsigned int ret
= 0;
2060 if (unlikely(!nr_pages
))
2064 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
2065 if (xas_retry(&xas
, page
))
2068 * If the entry has been swapped out, we can stop looking.
2069 * No current caller is looking for DAX entries.
2071 if (xa_is_value(page
))
2074 if (!page_cache_get_speculative(page
))
2077 /* Has the page moved or been split? */
2078 if (unlikely(page
!= xas_reload(&xas
)))
2081 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
2082 if (++ret
== nr_pages
)
2093 EXPORT_SYMBOL(find_get_pages_contig
);
2096 * find_get_pages_range_tag - Find and return head pages matching @tag.
2097 * @mapping: the address_space to search
2098 * @index: the starting page index
2099 * @end: The final page index (inclusive)
2100 * @tag: the tag index
2101 * @nr_pages: the maximum number of pages
2102 * @pages: where the resulting pages are placed
2104 * Like find_get_pages(), except we only return head pages which are tagged
2105 * with @tag. @index is updated to the index immediately after the last
2106 * page we return, ready for the next iteration.
2108 * Return: the number of pages which were found.
2110 unsigned find_get_pages_range_tag(struct address_space
*mapping
, pgoff_t
*index
,
2111 pgoff_t end
, xa_mark_t tag
, unsigned int nr_pages
,
2112 struct page
**pages
)
2114 XA_STATE(xas
, &mapping
->i_pages
, *index
);
2118 if (unlikely(!nr_pages
))
2122 while ((page
= find_get_entry(&xas
, end
, tag
))) {
2124 * Shadow entries should never be tagged, but this iteration
2125 * is lockless so there is a window for page reclaim to evict
2126 * a page we saw tagged. Skip over it.
2128 if (xa_is_value(page
))
2132 if (++ret
== nr_pages
) {
2133 *index
= page
->index
+ thp_nr_pages(page
);
2139 * We come here when we got to @end. We take care to not overflow the
2140 * index @index as it confuses some of the callers. This breaks the
2141 * iteration when there is a page at index -1 but that is already
2144 if (end
== (pgoff_t
)-1)
2145 *index
= (pgoff_t
)-1;
2153 EXPORT_SYMBOL(find_get_pages_range_tag
);
2156 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2157 * a _large_ part of the i/o request. Imagine the worst scenario:
2159 * ---R__________________________________________B__________
2160 * ^ reading here ^ bad block(assume 4k)
2162 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2163 * => failing the whole request => read(R) => read(R+1) =>
2164 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2165 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2166 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2168 * It is going insane. Fix it by quickly scaling down the readahead size.
2170 static void shrink_readahead_size_eio(struct file_ra_state
*ra
)
2176 * filemap_get_read_batch - Get a batch of pages for read
2178 * Get a batch of pages which represent a contiguous range of bytes
2179 * in the file. No tail pages will be returned. If @index is in the
2180 * middle of a THP, the entire THP will be returned. The last page in
2181 * the batch may have Readahead set or be not Uptodate so that the
2182 * caller can take the appropriate action.
2184 static void filemap_get_read_batch(struct address_space
*mapping
,
2185 pgoff_t index
, pgoff_t max
, struct pagevec
*pvec
)
2187 XA_STATE(xas
, &mapping
->i_pages
, index
);
2191 for (head
= xas_load(&xas
); head
; head
= xas_next(&xas
)) {
2192 if (xas_retry(&xas
, head
))
2194 if (xas
.xa_index
> max
|| xa_is_value(head
))
2196 if (!page_cache_get_speculative(head
))
2199 /* Has the page moved or been split? */
2200 if (unlikely(head
!= xas_reload(&xas
)))
2203 if (!pagevec_add(pvec
, head
))
2205 if (!PageUptodate(head
))
2207 if (PageReadahead(head
))
2209 xas
.xa_index
= head
->index
+ thp_nr_pages(head
) - 1;
2210 xas
.xa_offset
= (xas
.xa_index
>> xas
.xa_shift
) & XA_CHUNK_MASK
;
2220 static int filemap_read_page(struct file
*file
, struct address_space
*mapping
,
2226 * A previous I/O error may have been due to temporary failures,
2227 * eg. multipath errors. PG_error will be set again if readpage
2230 ClearPageError(page
);
2231 /* Start the actual read. The read will unlock the page. */
2232 error
= mapping
->a_ops
->readpage(file
, page
);
2236 error
= wait_on_page_locked_killable(page
);
2239 if (PageUptodate(page
))
2241 if (!page
->mapping
) /* page truncated */
2242 return AOP_TRUNCATED_PAGE
;
2243 shrink_readahead_size_eio(&file
->f_ra
);
2247 static bool filemap_range_uptodate(struct address_space
*mapping
,
2248 loff_t pos
, struct iov_iter
*iter
, struct page
*page
)
2252 if (PageUptodate(page
))
2254 /* pipes can't handle partially uptodate pages */
2255 if (iov_iter_is_pipe(iter
))
2257 if (!mapping
->a_ops
->is_partially_uptodate
)
2259 if (mapping
->host
->i_blkbits
>= (PAGE_SHIFT
+ thp_order(page
)))
2262 count
= iter
->count
;
2263 if (page_offset(page
) > pos
) {
2264 count
-= page_offset(page
) - pos
;
2267 pos
-= page_offset(page
);
2270 return mapping
->a_ops
->is_partially_uptodate(page
, pos
, count
);
2273 static int filemap_update_page(struct kiocb
*iocb
,
2274 struct address_space
*mapping
, struct iov_iter
*iter
,
2279 if (!trylock_page(page
)) {
2280 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_NOIO
))
2282 if (!(iocb
->ki_flags
& IOCB_WAITQ
)) {
2283 put_and_wait_on_page_locked(page
, TASK_KILLABLE
);
2284 return AOP_TRUNCATED_PAGE
;
2286 error
= __lock_page_async(page
, iocb
->ki_waitq
);
2295 if (filemap_range_uptodate(mapping
, iocb
->ki_pos
, iter
, page
))
2299 if (iocb
->ki_flags
& (IOCB_NOIO
| IOCB_NOWAIT
| IOCB_WAITQ
))
2302 error
= filemap_read_page(iocb
->ki_filp
, mapping
, page
);
2303 if (error
== AOP_TRUNCATED_PAGE
)
2309 return AOP_TRUNCATED_PAGE
;
2315 static int filemap_create_page(struct file
*file
,
2316 struct address_space
*mapping
, pgoff_t index
,
2317 struct pagevec
*pvec
)
2322 page
= page_cache_alloc(mapping
);
2326 error
= add_to_page_cache_lru(page
, mapping
, index
,
2327 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2328 if (error
== -EEXIST
)
2329 error
= AOP_TRUNCATED_PAGE
;
2333 error
= filemap_read_page(file
, mapping
, page
);
2337 pagevec_add(pvec
, page
);
2344 static int filemap_readahead(struct kiocb
*iocb
, struct file
*file
,
2345 struct address_space
*mapping
, struct page
*page
,
2348 if (iocb
->ki_flags
& IOCB_NOIO
)
2350 page_cache_async_readahead(mapping
, &file
->f_ra
, file
, page
,
2351 page
->index
, last_index
- page
->index
);
2355 static int filemap_get_pages(struct kiocb
*iocb
, struct iov_iter
*iter
,
2356 struct pagevec
*pvec
)
2358 struct file
*filp
= iocb
->ki_filp
;
2359 struct address_space
*mapping
= filp
->f_mapping
;
2360 struct file_ra_state
*ra
= &filp
->f_ra
;
2361 pgoff_t index
= iocb
->ki_pos
>> PAGE_SHIFT
;
2366 last_index
= DIV_ROUND_UP(iocb
->ki_pos
+ iter
->count
, PAGE_SIZE
);
2368 if (fatal_signal_pending(current
))
2371 filemap_get_read_batch(mapping
, index
, last_index
, pvec
);
2372 if (!pagevec_count(pvec
)) {
2373 if (iocb
->ki_flags
& IOCB_NOIO
)
2375 page_cache_sync_readahead(mapping
, ra
, filp
, index
,
2376 last_index
- index
);
2377 filemap_get_read_batch(mapping
, index
, last_index
, pvec
);
2379 if (!pagevec_count(pvec
)) {
2380 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_WAITQ
))
2382 err
= filemap_create_page(filp
, mapping
,
2383 iocb
->ki_pos
>> PAGE_SHIFT
, pvec
);
2384 if (err
== AOP_TRUNCATED_PAGE
)
2389 page
= pvec
->pages
[pagevec_count(pvec
) - 1];
2390 if (PageReadahead(page
)) {
2391 err
= filemap_readahead(iocb
, filp
, mapping
, page
, last_index
);
2395 if (!PageUptodate(page
)) {
2396 if ((iocb
->ki_flags
& IOCB_WAITQ
) && pagevec_count(pvec
) > 1)
2397 iocb
->ki_flags
|= IOCB_NOWAIT
;
2398 err
= filemap_update_page(iocb
, mapping
, iter
, page
);
2407 if (likely(--pvec
->nr
))
2409 if (err
== AOP_TRUNCATED_PAGE
)
2415 * filemap_read - Read data from the page cache.
2416 * @iocb: The iocb to read.
2417 * @iter: Destination for the data.
2418 * @already_read: Number of bytes already read by the caller.
2420 * Copies data from the page cache. If the data is not currently present,
2421 * uses the readahead and readpage address_space operations to fetch it.
2423 * Return: Total number of bytes copied, including those already read by
2424 * the caller. If an error happens before any bytes are copied, returns
2425 * a negative error number.
2427 ssize_t
filemap_read(struct kiocb
*iocb
, struct iov_iter
*iter
,
2428 ssize_t already_read
)
2430 struct file
*filp
= iocb
->ki_filp
;
2431 struct file_ra_state
*ra
= &filp
->f_ra
;
2432 struct address_space
*mapping
= filp
->f_mapping
;
2433 struct inode
*inode
= mapping
->host
;
2434 struct pagevec pvec
;
2436 bool writably_mapped
;
2437 loff_t isize
, end_offset
;
2439 if (unlikely(iocb
->ki_pos
>= inode
->i_sb
->s_maxbytes
))
2441 if (unlikely(!iov_iter_count(iter
)))
2444 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2445 pagevec_init(&pvec
);
2451 * If we've already successfully copied some data, then we
2452 * can no longer safely return -EIOCBQUEUED. Hence mark
2453 * an async read NOWAIT at that point.
2455 if ((iocb
->ki_flags
& IOCB_WAITQ
) && already_read
)
2456 iocb
->ki_flags
|= IOCB_NOWAIT
;
2458 error
= filemap_get_pages(iocb
, iter
, &pvec
);
2463 * i_size must be checked after we know the pages are Uptodate.
2465 * Checking i_size after the check allows us to calculate
2466 * the correct value for "nr", which means the zero-filled
2467 * part of the page is not copied back to userspace (unless
2468 * another truncate extends the file - this is desired though).
2470 isize
= i_size_read(inode
);
2471 if (unlikely(iocb
->ki_pos
>= isize
))
2473 end_offset
= min_t(loff_t
, isize
, iocb
->ki_pos
+ iter
->count
);
2476 * Once we start copying data, we don't want to be touching any
2477 * cachelines that might be contended:
2479 writably_mapped
= mapping_writably_mapped(mapping
);
2482 * When a sequential read accesses a page several times, only
2483 * mark it as accessed the first time.
2485 if (iocb
->ki_pos
>> PAGE_SHIFT
!=
2486 ra
->prev_pos
>> PAGE_SHIFT
)
2487 mark_page_accessed(pvec
.pages
[0]);
2489 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
2490 struct page
*page
= pvec
.pages
[i
];
2491 size_t page_size
= thp_size(page
);
2492 size_t offset
= iocb
->ki_pos
& (page_size
- 1);
2493 size_t bytes
= min_t(loff_t
, end_offset
- iocb
->ki_pos
,
2494 page_size
- offset
);
2497 if (end_offset
< page_offset(page
))
2500 mark_page_accessed(page
);
2502 * If users can be writing to this page using arbitrary
2503 * virtual addresses, take care about potential aliasing
2504 * before reading the page on the kernel side.
2506 if (writably_mapped
) {
2509 for (j
= 0; j
< thp_nr_pages(page
); j
++)
2510 flush_dcache_page(page
+ j
);
2513 copied
= copy_page_to_iter(page
, offset
, bytes
, iter
);
2515 already_read
+= copied
;
2516 iocb
->ki_pos
+= copied
;
2517 ra
->prev_pos
= iocb
->ki_pos
;
2519 if (copied
< bytes
) {
2525 for (i
= 0; i
< pagevec_count(&pvec
); i
++)
2526 put_page(pvec
.pages
[i
]);
2527 pagevec_reinit(&pvec
);
2528 } while (iov_iter_count(iter
) && iocb
->ki_pos
< isize
&& !error
);
2530 file_accessed(filp
);
2532 return already_read
? already_read
: error
;
2534 EXPORT_SYMBOL_GPL(filemap_read
);
2537 * generic_file_read_iter - generic filesystem read routine
2538 * @iocb: kernel I/O control block
2539 * @iter: destination for the data read
2541 * This is the "read_iter()" routine for all filesystems
2542 * that can use the page cache directly.
2544 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2545 * be returned when no data can be read without waiting for I/O requests
2546 * to complete; it doesn't prevent readahead.
2548 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2549 * requests shall be made for the read or for readahead. When no data
2550 * can be read, -EAGAIN shall be returned. When readahead would be
2551 * triggered, a partial, possibly empty read shall be returned.
2554 * * number of bytes copied, even for partial reads
2555 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2558 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2560 size_t count
= iov_iter_count(iter
);
2564 return 0; /* skip atime */
2566 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2567 struct file
*file
= iocb
->ki_filp
;
2568 struct address_space
*mapping
= file
->f_mapping
;
2569 struct inode
*inode
= mapping
->host
;
2572 size
= i_size_read(inode
);
2573 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2574 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2575 iocb
->ki_pos
+ count
- 1))
2578 retval
= filemap_write_and_wait_range(mapping
,
2580 iocb
->ki_pos
+ count
- 1);
2585 file_accessed(file
);
2587 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2589 iocb
->ki_pos
+= retval
;
2592 if (retval
!= -EIOCBQUEUED
)
2593 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2596 * Btrfs can have a short DIO read if we encounter
2597 * compressed extents, so if there was an error, or if
2598 * we've already read everything we wanted to, or if
2599 * there was a short read because we hit EOF, go ahead
2600 * and return. Otherwise fallthrough to buffered io for
2601 * the rest of the read. Buffered reads will not work for
2602 * DAX files, so don't bother trying.
2604 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2609 return filemap_read(iocb
, iter
, retval
);
2611 EXPORT_SYMBOL(generic_file_read_iter
);
2613 static inline loff_t
page_seek_hole_data(struct xa_state
*xas
,
2614 struct address_space
*mapping
, struct page
*page
,
2615 loff_t start
, loff_t end
, bool seek_data
)
2617 const struct address_space_operations
*ops
= mapping
->a_ops
;
2618 size_t offset
, bsz
= i_blocksize(mapping
->host
);
2620 if (xa_is_value(page
) || PageUptodate(page
))
2621 return seek_data
? start
: end
;
2622 if (!ops
->is_partially_uptodate
)
2623 return seek_data
? end
: start
;
2628 if (unlikely(page
->mapping
!= mapping
))
2631 offset
= offset_in_thp(page
, start
) & ~(bsz
- 1);
2634 if (ops
->is_partially_uptodate(page
, offset
, bsz
) == seek_data
)
2636 start
= (start
+ bsz
) & ~(bsz
- 1);
2638 } while (offset
< thp_size(page
));
2646 unsigned int seek_page_size(struct xa_state
*xas
, struct page
*page
)
2648 if (xa_is_value(page
))
2649 return PAGE_SIZE
<< xa_get_order(xas
->xa
, xas
->xa_index
);
2650 return thp_size(page
);
2654 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2655 * @mapping: Address space to search.
2656 * @start: First byte to consider.
2657 * @end: Limit of search (exclusive).
2658 * @whence: Either SEEK_HOLE or SEEK_DATA.
2660 * If the page cache knows which blocks contain holes and which blocks
2661 * contain data, your filesystem can use this function to implement
2662 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2663 * entirely memory-based such as tmpfs, and filesystems which support
2664 * unwritten extents.
2666 * Return: The requested offset on successs, or -ENXIO if @whence specifies
2667 * SEEK_DATA and there is no data after @start. There is an implicit hole
2668 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2669 * and @end contain data.
2671 loff_t
mapping_seek_hole_data(struct address_space
*mapping
, loff_t start
,
2672 loff_t end
, int whence
)
2674 XA_STATE(xas
, &mapping
->i_pages
, start
>> PAGE_SHIFT
);
2675 pgoff_t max
= (end
- 1) / PAGE_SIZE
;
2676 bool seek_data
= (whence
== SEEK_DATA
);
2683 while ((page
= find_get_entry(&xas
, max
, XA_PRESENT
))) {
2684 loff_t pos
= xas
.xa_index
* PAGE_SIZE
;
2692 pos
+= seek_page_size(&xas
, page
);
2693 start
= page_seek_hole_data(&xas
, mapping
, page
, start
, pos
,
2697 if (!xa_is_value(page
))
2708 if (!xa_is_value(page
))
2717 #define MMAP_LOTSAMISS (100)
2719 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2720 * @vmf - the vm_fault for this fault.
2721 * @page - the page to lock.
2722 * @fpin - the pointer to the file we may pin (or is already pinned).
2724 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2725 * It differs in that it actually returns the page locked if it returns 1 and 0
2726 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2727 * will point to the pinned file and needs to be fput()'ed at a later point.
2729 static int lock_page_maybe_drop_mmap(struct vm_fault
*vmf
, struct page
*page
,
2732 if (trylock_page(page
))
2736 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2737 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2738 * is supposed to work. We have way too many special cases..
2740 if (vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)
2743 *fpin
= maybe_unlock_mmap_for_io(vmf
, *fpin
);
2744 if (vmf
->flags
& FAULT_FLAG_KILLABLE
) {
2745 if (__lock_page_killable(page
)) {
2747 * We didn't have the right flags to drop the mmap_lock,
2748 * but all fault_handlers only check for fatal signals
2749 * if we return VM_FAULT_RETRY, so we need to drop the
2750 * mmap_lock here and return 0 if we don't have a fpin.
2753 mmap_read_unlock(vmf
->vma
->vm_mm
);
2763 * Synchronous readahead happens when we don't even find a page in the page
2764 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2765 * to drop the mmap sem we return the file that was pinned in order for us to do
2766 * that. If we didn't pin a file then we return NULL. The file that is
2767 * returned needs to be fput()'ed when we're done with it.
2769 static struct file
*do_sync_mmap_readahead(struct vm_fault
*vmf
)
2771 struct file
*file
= vmf
->vma
->vm_file
;
2772 struct file_ra_state
*ra
= &file
->f_ra
;
2773 struct address_space
*mapping
= file
->f_mapping
;
2774 DEFINE_READAHEAD(ractl
, file
, mapping
, vmf
->pgoff
);
2775 struct file
*fpin
= NULL
;
2776 unsigned int mmap_miss
;
2778 /* If we don't want any read-ahead, don't bother */
2779 if (vmf
->vma
->vm_flags
& VM_RAND_READ
)
2784 if (vmf
->vma
->vm_flags
& VM_SEQ_READ
) {
2785 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2786 page_cache_sync_ra(&ractl
, ra
, ra
->ra_pages
);
2790 /* Avoid banging the cache line if not needed */
2791 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
2792 if (mmap_miss
< MMAP_LOTSAMISS
* 10)
2793 WRITE_ONCE(ra
->mmap_miss
, ++mmap_miss
);
2796 * Do we miss much more than hit in this file? If so,
2797 * stop bothering with read-ahead. It will only hurt.
2799 if (mmap_miss
> MMAP_LOTSAMISS
)
2805 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2806 ra
->start
= max_t(long, 0, vmf
->pgoff
- ra
->ra_pages
/ 2);
2807 ra
->size
= ra
->ra_pages
;
2808 ra
->async_size
= ra
->ra_pages
/ 4;
2809 ractl
._index
= ra
->start
;
2810 do_page_cache_ra(&ractl
, ra
->size
, ra
->async_size
);
2815 * Asynchronous readahead happens when we find the page and PG_readahead,
2816 * so we want to possibly extend the readahead further. We return the file that
2817 * was pinned if we have to drop the mmap_lock in order to do IO.
2819 static struct file
*do_async_mmap_readahead(struct vm_fault
*vmf
,
2822 struct file
*file
= vmf
->vma
->vm_file
;
2823 struct file_ra_state
*ra
= &file
->f_ra
;
2824 struct address_space
*mapping
= file
->f_mapping
;
2825 struct file
*fpin
= NULL
;
2826 unsigned int mmap_miss
;
2827 pgoff_t offset
= vmf
->pgoff
;
2829 /* If we don't want any read-ahead, don't bother */
2830 if (vmf
->vma
->vm_flags
& VM_RAND_READ
|| !ra
->ra_pages
)
2832 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
2834 WRITE_ONCE(ra
->mmap_miss
, --mmap_miss
);
2835 if (PageReadahead(page
)) {
2836 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2837 page_cache_async_readahead(mapping
, ra
, file
,
2838 page
, offset
, ra
->ra_pages
);
2844 * filemap_fault - read in file data for page fault handling
2845 * @vmf: struct vm_fault containing details of the fault
2847 * filemap_fault() is invoked via the vma operations vector for a
2848 * mapped memory region to read in file data during a page fault.
2850 * The goto's are kind of ugly, but this streamlines the normal case of having
2851 * it in the page cache, and handles the special cases reasonably without
2852 * having a lot of duplicated code.
2854 * vma->vm_mm->mmap_lock must be held on entry.
2856 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2857 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2859 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2860 * has not been released.
2862 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2864 * Return: bitwise-OR of %VM_FAULT_ codes.
2866 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
2869 struct file
*file
= vmf
->vma
->vm_file
;
2870 struct file
*fpin
= NULL
;
2871 struct address_space
*mapping
= file
->f_mapping
;
2872 struct file_ra_state
*ra
= &file
->f_ra
;
2873 struct inode
*inode
= mapping
->host
;
2874 pgoff_t offset
= vmf
->pgoff
;
2879 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2880 if (unlikely(offset
>= max_off
))
2881 return VM_FAULT_SIGBUS
;
2884 * Do we have something in the page cache already?
2886 page
= find_get_page(mapping
, offset
);
2887 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2889 * We found the page, so try async readahead before
2890 * waiting for the lock.
2892 fpin
= do_async_mmap_readahead(vmf
, page
);
2894 /* No page in the page cache at all */
2895 count_vm_event(PGMAJFAULT
);
2896 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2897 ret
= VM_FAULT_MAJOR
;
2898 fpin
= do_sync_mmap_readahead(vmf
);
2900 page
= pagecache_get_page(mapping
, offset
,
2901 FGP_CREAT
|FGP_FOR_MMAP
,
2906 return VM_FAULT_OOM
;
2910 if (!lock_page_maybe_drop_mmap(vmf
, page
, &fpin
))
2913 /* Did it get truncated? */
2914 if (unlikely(compound_head(page
)->mapping
!= mapping
)) {
2919 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
2922 * We have a locked page in the page cache, now we need to check
2923 * that it's up-to-date. If not, it is going to be due to an error.
2925 if (unlikely(!PageUptodate(page
)))
2926 goto page_not_uptodate
;
2929 * We've made it this far and we had to drop our mmap_lock, now is the
2930 * time to return to the upper layer and have it re-find the vma and
2939 * Found the page and have a reference on it.
2940 * We must recheck i_size under page lock.
2942 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2943 if (unlikely(offset
>= max_off
)) {
2946 return VM_FAULT_SIGBUS
;
2950 return ret
| VM_FAULT_LOCKED
;
2954 * Umm, take care of errors if the page isn't up-to-date.
2955 * Try to re-read it _once_. We do this synchronously,
2956 * because there really aren't any performance issues here
2957 * and we need to check for errors.
2959 ClearPageError(page
);
2960 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2961 error
= mapping
->a_ops
->readpage(file
, page
);
2963 wait_on_page_locked(page
);
2964 if (!PageUptodate(page
))
2971 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2974 shrink_readahead_size_eio(ra
);
2975 return VM_FAULT_SIGBUS
;
2979 * We dropped the mmap_lock, we need to return to the fault handler to
2980 * re-find the vma and come back and find our hopefully still populated
2987 return ret
| VM_FAULT_RETRY
;
2989 EXPORT_SYMBOL(filemap_fault
);
2991 static bool filemap_map_pmd(struct vm_fault
*vmf
, struct page
*page
)
2993 struct mm_struct
*mm
= vmf
->vma
->vm_mm
;
2995 /* Huge page is mapped? No need to proceed. */
2996 if (pmd_trans_huge(*vmf
->pmd
)) {
3002 if (pmd_none(*vmf
->pmd
) && PageTransHuge(page
)) {
3003 vm_fault_t ret
= do_set_pmd(vmf
, page
);
3005 /* The page is mapped successfully, reference consumed. */
3011 if (pmd_none(*vmf
->pmd
)) {
3012 vmf
->ptl
= pmd_lock(mm
, vmf
->pmd
);
3013 if (likely(pmd_none(*vmf
->pmd
))) {
3015 pmd_populate(mm
, vmf
->pmd
, vmf
->prealloc_pte
);
3016 vmf
->prealloc_pte
= NULL
;
3018 spin_unlock(vmf
->ptl
);
3021 /* See comment in handle_pte_fault() */
3022 if (pmd_devmap_trans_unstable(vmf
->pmd
)) {
3031 static struct page
*next_uptodate_page(struct page
*page
,
3032 struct address_space
*mapping
,
3033 struct xa_state
*xas
, pgoff_t end_pgoff
)
3035 unsigned long max_idx
;
3040 if (xas_retry(xas
, page
))
3042 if (xa_is_value(page
))
3044 if (PageLocked(page
))
3046 if (!page_cache_get_speculative(page
))
3048 /* Has the page moved or been split? */
3049 if (unlikely(page
!= xas_reload(xas
)))
3051 if (!PageUptodate(page
) || PageReadahead(page
))
3053 if (PageHWPoison(page
))
3055 if (!trylock_page(page
))
3057 if (page
->mapping
!= mapping
)
3059 if (!PageUptodate(page
))
3061 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
3062 if (xas
->xa_index
>= max_idx
)
3069 } while ((page
= xas_next_entry(xas
, end_pgoff
)) != NULL
);
3074 static inline struct page
*first_map_page(struct address_space
*mapping
,
3075 struct xa_state
*xas
,
3078 return next_uptodate_page(xas_find(xas
, end_pgoff
),
3079 mapping
, xas
, end_pgoff
);
3082 static inline struct page
*next_map_page(struct address_space
*mapping
,
3083 struct xa_state
*xas
,
3086 return next_uptodate_page(xas_next_entry(xas
, end_pgoff
),
3087 mapping
, xas
, end_pgoff
);
3090 vm_fault_t
filemap_map_pages(struct vm_fault
*vmf
,
3091 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
3093 struct vm_area_struct
*vma
= vmf
->vma
;
3094 struct file
*file
= vma
->vm_file
;
3095 struct address_space
*mapping
= file
->f_mapping
;
3096 pgoff_t last_pgoff
= start_pgoff
;
3098 XA_STATE(xas
, &mapping
->i_pages
, start_pgoff
);
3099 struct page
*head
, *page
;
3100 unsigned int mmap_miss
= READ_ONCE(file
->f_ra
.mmap_miss
);
3104 head
= first_map_page(mapping
, &xas
, end_pgoff
);
3108 if (filemap_map_pmd(vmf
, head
)) {
3109 ret
= VM_FAULT_NOPAGE
;
3113 addr
= vma
->vm_start
+ ((start_pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
3114 vmf
->pte
= pte_offset_map_lock(vma
->vm_mm
, vmf
->pmd
, addr
, &vmf
->ptl
);
3116 page
= find_subpage(head
, xas
.xa_index
);
3117 if (PageHWPoison(page
))
3123 addr
+= (xas
.xa_index
- last_pgoff
) << PAGE_SHIFT
;
3124 vmf
->pte
+= xas
.xa_index
- last_pgoff
;
3125 last_pgoff
= xas
.xa_index
;
3127 if (!pte_none(*vmf
->pte
))
3130 /* We're about to handle the fault */
3131 if (vmf
->address
== addr
)
3132 ret
= VM_FAULT_NOPAGE
;
3134 do_set_pte(vmf
, page
, addr
);
3135 /* no need to invalidate: a not-present page won't be cached */
3136 update_mmu_cache(vma
, addr
, vmf
->pte
);
3142 } while ((head
= next_map_page(mapping
, &xas
, end_pgoff
)) != NULL
);
3143 pte_unmap_unlock(vmf
->pte
, vmf
->ptl
);
3146 WRITE_ONCE(file
->f_ra
.mmap_miss
, mmap_miss
);
3149 EXPORT_SYMBOL(filemap_map_pages
);
3151 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3153 struct address_space
*mapping
= vmf
->vma
->vm_file
->f_mapping
;
3154 struct page
*page
= vmf
->page
;
3155 vm_fault_t ret
= VM_FAULT_LOCKED
;
3157 sb_start_pagefault(mapping
->host
->i_sb
);
3158 file_update_time(vmf
->vma
->vm_file
);
3160 if (page
->mapping
!= mapping
) {
3162 ret
= VM_FAULT_NOPAGE
;
3166 * We mark the page dirty already here so that when freeze is in
3167 * progress, we are guaranteed that writeback during freezing will
3168 * see the dirty page and writeprotect it again.
3170 set_page_dirty(page
);
3171 wait_for_stable_page(page
);
3173 sb_end_pagefault(mapping
->host
->i_sb
);
3177 const struct vm_operations_struct generic_file_vm_ops
= {
3178 .fault
= filemap_fault
,
3179 .map_pages
= filemap_map_pages
,
3180 .page_mkwrite
= filemap_page_mkwrite
,
3183 /* This is used for a general mmap of a disk file */
3185 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
3187 struct address_space
*mapping
= file
->f_mapping
;
3189 if (!mapping
->a_ops
->readpage
)
3191 file_accessed(file
);
3192 vma
->vm_ops
= &generic_file_vm_ops
;
3197 * This is for filesystems which do not implement ->writepage.
3199 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3201 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
3203 return generic_file_mmap(file
, vma
);
3206 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3208 return VM_FAULT_SIGBUS
;
3210 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
3214 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
3218 #endif /* CONFIG_MMU */
3220 EXPORT_SYMBOL(filemap_page_mkwrite
);
3221 EXPORT_SYMBOL(generic_file_mmap
);
3222 EXPORT_SYMBOL(generic_file_readonly_mmap
);
3224 static struct page
*wait_on_page_read(struct page
*page
)
3226 if (!IS_ERR(page
)) {
3227 wait_on_page_locked(page
);
3228 if (!PageUptodate(page
)) {
3230 page
= ERR_PTR(-EIO
);
3236 static struct page
*do_read_cache_page(struct address_space
*mapping
,
3238 int (*filler
)(void *, struct page
*),
3245 page
= find_get_page(mapping
, index
);
3247 page
= __page_cache_alloc(gfp
);
3249 return ERR_PTR(-ENOMEM
);
3250 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
3251 if (unlikely(err
)) {
3255 /* Presumably ENOMEM for xarray node */
3256 return ERR_PTR(err
);
3261 err
= filler(data
, page
);
3263 err
= mapping
->a_ops
->readpage(data
, page
);
3267 return ERR_PTR(err
);
3270 page
= wait_on_page_read(page
);
3275 if (PageUptodate(page
))
3279 * Page is not up to date and may be locked due to one of the following
3280 * case a: Page is being filled and the page lock is held
3281 * case b: Read/write error clearing the page uptodate status
3282 * case c: Truncation in progress (page locked)
3283 * case d: Reclaim in progress
3285 * Case a, the page will be up to date when the page is unlocked.
3286 * There is no need to serialise on the page lock here as the page
3287 * is pinned so the lock gives no additional protection. Even if the
3288 * page is truncated, the data is still valid if PageUptodate as
3289 * it's a race vs truncate race.
3290 * Case b, the page will not be up to date
3291 * Case c, the page may be truncated but in itself, the data may still
3292 * be valid after IO completes as it's a read vs truncate race. The
3293 * operation must restart if the page is not uptodate on unlock but
3294 * otherwise serialising on page lock to stabilise the mapping gives
3295 * no additional guarantees to the caller as the page lock is
3296 * released before return.
3297 * Case d, similar to truncation. If reclaim holds the page lock, it
3298 * will be a race with remove_mapping that determines if the mapping
3299 * is valid on unlock but otherwise the data is valid and there is
3300 * no need to serialise with page lock.
3302 * As the page lock gives no additional guarantee, we optimistically
3303 * wait on the page to be unlocked and check if it's up to date and
3304 * use the page if it is. Otherwise, the page lock is required to
3305 * distinguish between the different cases. The motivation is that we
3306 * avoid spurious serialisations and wakeups when multiple processes
3307 * wait on the same page for IO to complete.
3309 wait_on_page_locked(page
);
3310 if (PageUptodate(page
))
3313 /* Distinguish between all the cases under the safety of the lock */
3316 /* Case c or d, restart the operation */
3317 if (!page
->mapping
) {
3323 /* Someone else locked and filled the page in a very small window */
3324 if (PageUptodate(page
)) {
3330 * A previous I/O error may have been due to temporary
3332 * Clear page error before actual read, PG_error will be
3333 * set again if read page fails.
3335 ClearPageError(page
);
3339 mark_page_accessed(page
);
3344 * read_cache_page - read into page cache, fill it if needed
3345 * @mapping: the page's address_space
3346 * @index: the page index
3347 * @filler: function to perform the read
3348 * @data: first arg to filler(data, page) function, often left as NULL
3350 * Read into the page cache. If a page already exists, and PageUptodate() is
3351 * not set, try to fill the page and wait for it to become unlocked.
3353 * If the page does not get brought uptodate, return -EIO.
3355 * Return: up to date page on success, ERR_PTR() on failure.
3357 struct page
*read_cache_page(struct address_space
*mapping
,
3359 int (*filler
)(void *, struct page
*),
3362 return do_read_cache_page(mapping
, index
, filler
, data
,
3363 mapping_gfp_mask(mapping
));
3365 EXPORT_SYMBOL(read_cache_page
);
3368 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3369 * @mapping: the page's address_space
3370 * @index: the page index
3371 * @gfp: the page allocator flags to use if allocating
3373 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3374 * any new page allocations done using the specified allocation flags.
3376 * If the page does not get brought uptodate, return -EIO.
3378 * Return: up to date page on success, ERR_PTR() on failure.
3380 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
3384 return do_read_cache_page(mapping
, index
, NULL
, NULL
, gfp
);
3386 EXPORT_SYMBOL(read_cache_page_gfp
);
3388 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
3389 loff_t pos
, unsigned len
, unsigned flags
,
3390 struct page
**pagep
, void **fsdata
)
3392 const struct address_space_operations
*aops
= mapping
->a_ops
;
3394 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
3397 EXPORT_SYMBOL(pagecache_write_begin
);
3399 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
3400 loff_t pos
, unsigned len
, unsigned copied
,
3401 struct page
*page
, void *fsdata
)
3403 const struct address_space_operations
*aops
= mapping
->a_ops
;
3405 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
3407 EXPORT_SYMBOL(pagecache_write_end
);
3410 * Warn about a page cache invalidation failure during a direct I/O write.
3412 void dio_warn_stale_pagecache(struct file
*filp
)
3414 static DEFINE_RATELIMIT_STATE(_rs
, 86400 * HZ
, DEFAULT_RATELIMIT_BURST
);
3418 errseq_set(&filp
->f_mapping
->wb_err
, -EIO
);
3419 if (__ratelimit(&_rs
)) {
3420 path
= file_path(filp
, pathname
, sizeof(pathname
));
3423 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3424 pr_crit("File: %s PID: %d Comm: %.20s\n", path
, current
->pid
,
3430 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3432 struct file
*file
= iocb
->ki_filp
;
3433 struct address_space
*mapping
= file
->f_mapping
;
3434 struct inode
*inode
= mapping
->host
;
3435 loff_t pos
= iocb
->ki_pos
;
3440 write_len
= iov_iter_count(from
);
3441 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
3443 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3444 /* If there are pages to writeback, return */
3445 if (filemap_range_has_page(file
->f_mapping
, pos
,
3446 pos
+ write_len
- 1))
3449 written
= filemap_write_and_wait_range(mapping
, pos
,
3450 pos
+ write_len
- 1);
3456 * After a write we want buffered reads to be sure to go to disk to get
3457 * the new data. We invalidate clean cached page from the region we're
3458 * about to write. We do this *before* the write so that we can return
3459 * without clobbering -EIOCBQUEUED from ->direct_IO().
3461 written
= invalidate_inode_pages2_range(mapping
,
3462 pos
>> PAGE_SHIFT
, end
);
3464 * If a page can not be invalidated, return 0 to fall back
3465 * to buffered write.
3468 if (written
== -EBUSY
)
3473 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3476 * Finally, try again to invalidate clean pages which might have been
3477 * cached by non-direct readahead, or faulted in by get_user_pages()
3478 * if the source of the write was an mmap'ed region of the file
3479 * we're writing. Either one is a pretty crazy thing to do,
3480 * so we don't support it 100%. If this invalidation
3481 * fails, tough, the write still worked...
3483 * Most of the time we do not need this since dio_complete() will do
3484 * the invalidation for us. However there are some file systems that
3485 * do not end up with dio_complete() being called, so let's not break
3486 * them by removing it completely.
3488 * Noticeable example is a blkdev_direct_IO().
3490 * Skip invalidation for async writes or if mapping has no pages.
3492 if (written
> 0 && mapping
->nrpages
&&
3493 invalidate_inode_pages2_range(mapping
, pos
>> PAGE_SHIFT
, end
))
3494 dio_warn_stale_pagecache(file
);
3498 write_len
-= written
;
3499 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3500 i_size_write(inode
, pos
);
3501 mark_inode_dirty(inode
);
3505 if (written
!= -EIOCBQUEUED
)
3506 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3510 EXPORT_SYMBOL(generic_file_direct_write
);
3513 * Find or create a page at the given pagecache position. Return the locked
3514 * page. This function is specifically for buffered writes.
3516 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
3517 pgoff_t index
, unsigned flags
)
3520 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
3522 if (flags
& AOP_FLAG_NOFS
)
3523 fgp_flags
|= FGP_NOFS
;
3525 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
3526 mapping_gfp_mask(mapping
));
3528 wait_for_stable_page(page
);
3532 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3534 ssize_t
generic_perform_write(struct file
*file
,
3535 struct iov_iter
*i
, loff_t pos
)
3537 struct address_space
*mapping
= file
->f_mapping
;
3538 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3540 ssize_t written
= 0;
3541 unsigned int flags
= 0;
3545 unsigned long offset
; /* Offset into pagecache page */
3546 unsigned long bytes
; /* Bytes to write to page */
3547 size_t copied
; /* Bytes copied from user */
3550 offset
= (pos
& (PAGE_SIZE
- 1));
3551 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3556 * Bring in the user page that we will copy from _first_.
3557 * Otherwise there's a nasty deadlock on copying from the
3558 * same page as we're writing to, without it being marked
3561 * Not only is this an optimisation, but it is also required
3562 * to check that the address is actually valid, when atomic
3563 * usercopies are used, below.
3565 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3570 if (fatal_signal_pending(current
)) {
3575 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3577 if (unlikely(status
< 0))
3580 if (mapping_writably_mapped(mapping
))
3581 flush_dcache_page(page
);
3583 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3584 flush_dcache_page(page
);
3586 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3588 if (unlikely(status
< 0))
3594 iov_iter_advance(i
, copied
);
3595 if (unlikely(copied
== 0)) {
3597 * If we were unable to copy any data at all, we must
3598 * fall back to a single segment length write.
3600 * If we didn't fallback here, we could livelock
3601 * because not all segments in the iov can be copied at
3602 * once without a pagefault.
3604 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3605 iov_iter_single_seg_count(i
));
3611 balance_dirty_pages_ratelimited(mapping
);
3612 } while (iov_iter_count(i
));
3614 return written
? written
: status
;
3616 EXPORT_SYMBOL(generic_perform_write
);
3619 * __generic_file_write_iter - write data to a file
3620 * @iocb: IO state structure (file, offset, etc.)
3621 * @from: iov_iter with data to write
3623 * This function does all the work needed for actually writing data to a
3624 * file. It does all basic checks, removes SUID from the file, updates
3625 * modification times and calls proper subroutines depending on whether we
3626 * do direct IO or a standard buffered write.
3628 * It expects i_mutex to be grabbed unless we work on a block device or similar
3629 * object which does not need locking at all.
3631 * This function does *not* take care of syncing data in case of O_SYNC write.
3632 * A caller has to handle it. This is mainly due to the fact that we want to
3633 * avoid syncing under i_mutex.
3636 * * number of bytes written, even for truncated writes
3637 * * negative error code if no data has been written at all
3639 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3641 struct file
*file
= iocb
->ki_filp
;
3642 struct address_space
* mapping
= file
->f_mapping
;
3643 struct inode
*inode
= mapping
->host
;
3644 ssize_t written
= 0;
3648 /* We can write back this queue in page reclaim */
3649 current
->backing_dev_info
= inode_to_bdi(inode
);
3650 err
= file_remove_privs(file
);
3654 err
= file_update_time(file
);
3658 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3659 loff_t pos
, endbyte
;
3661 written
= generic_file_direct_write(iocb
, from
);
3663 * If the write stopped short of completing, fall back to
3664 * buffered writes. Some filesystems do this for writes to
3665 * holes, for example. For DAX files, a buffered write will
3666 * not succeed (even if it did, DAX does not handle dirty
3667 * page-cache pages correctly).
3669 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3672 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3674 * If generic_perform_write() returned a synchronous error
3675 * then we want to return the number of bytes which were
3676 * direct-written, or the error code if that was zero. Note
3677 * that this differs from normal direct-io semantics, which
3678 * will return -EFOO even if some bytes were written.
3680 if (unlikely(status
< 0)) {
3685 * We need to ensure that the page cache pages are written to
3686 * disk and invalidated to preserve the expected O_DIRECT
3689 endbyte
= pos
+ status
- 1;
3690 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3692 iocb
->ki_pos
= endbyte
+ 1;
3694 invalidate_mapping_pages(mapping
,
3696 endbyte
>> PAGE_SHIFT
);
3699 * We don't know how much we wrote, so just return
3700 * the number of bytes which were direct-written
3704 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3705 if (likely(written
> 0))
3706 iocb
->ki_pos
+= written
;
3709 current
->backing_dev_info
= NULL
;
3710 return written
? written
: err
;
3712 EXPORT_SYMBOL(__generic_file_write_iter
);
3715 * generic_file_write_iter - write data to a file
3716 * @iocb: IO state structure
3717 * @from: iov_iter with data to write
3719 * This is a wrapper around __generic_file_write_iter() to be used by most
3720 * filesystems. It takes care of syncing the file in case of O_SYNC file
3721 * and acquires i_mutex as needed.
3723 * * negative error code if no data has been written at all of
3724 * vfs_fsync_range() failed for a synchronous write
3725 * * number of bytes written, even for truncated writes
3727 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3729 struct file
*file
= iocb
->ki_filp
;
3730 struct inode
*inode
= file
->f_mapping
->host
;
3734 ret
= generic_write_checks(iocb
, from
);
3736 ret
= __generic_file_write_iter(iocb
, from
);
3737 inode_unlock(inode
);
3740 ret
= generic_write_sync(iocb
, ret
);
3743 EXPORT_SYMBOL(generic_file_write_iter
);
3746 * try_to_release_page() - release old fs-specific metadata on a page
3748 * @page: the page which the kernel is trying to free
3749 * @gfp_mask: memory allocation flags (and I/O mode)
3751 * The address_space is to try to release any data against the page
3752 * (presumably at page->private).
3754 * This may also be called if PG_fscache is set on a page, indicating that the
3755 * page is known to the local caching routines.
3757 * The @gfp_mask argument specifies whether I/O may be performed to release
3758 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3760 * Return: %1 if the release was successful, otherwise return zero.
3762 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3764 struct address_space
* const mapping
= page
->mapping
;
3766 BUG_ON(!PageLocked(page
));
3767 if (PageWriteback(page
))
3770 if (mapping
&& mapping
->a_ops
->releasepage
)
3771 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3772 return try_to_free_buffers(page
);
3775 EXPORT_SYMBOL(try_to_release_page
);