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/swapops.h>
25 #include <linux/syscalls.h>
26 #include <linux/mman.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/uio.h>
30 #include <linux/error-injection.h>
31 #include <linux/hash.h>
32 #include <linux/writeback.h>
33 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h>
35 #include <linux/security.h>
36 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.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 <linux/migrate.h>
46 #include <linux/pipe_fs_i.h>
47 #include <linux/splice.h>
48 #include <asm/pgalloc.h>
49 #include <asm/tlbflush.h>
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/filemap.h>
56 * FIXME: remove all knowledge of the buffer layer from the core VM
58 #include <linux/buffer_head.h> /* for try_to_free_buffers */
65 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * Shared mappings now work. 15.8.1995 Bruno.
70 * finished 'unifying' the page and buffer cache and SMP-threaded the
71 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
73 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
79 * ->i_mmap_rwsem (truncate_pagecache)
80 * ->private_lock (__free_pte->block_dirty_folio)
81 * ->swap_lock (exclusive_swap_page, others)
85 * ->invalidate_lock (acquired by fs in truncate path)
86 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
90 * ->page_table_lock or pte_lock (various, mainly in memory.c)
91 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
94 * ->invalidate_lock (filemap_fault)
95 * ->lock_page (filemap_fault, access_process_vm)
97 * ->i_rwsem (generic_perform_write)
98 * ->mmap_lock (fault_in_readable->do_page_fault)
101 * sb_lock (fs/fs-writeback.c)
102 * ->i_pages lock (__sync_single_inode)
105 * ->anon_vma.lock (vma_merge)
108 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
110 * ->page_table_lock or pte_lock
111 * ->swap_lock (try_to_unmap_one)
112 * ->private_lock (try_to_unmap_one)
113 * ->i_pages lock (try_to_unmap_one)
114 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
115 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
116 * ->private_lock (page_remove_rmap->set_page_dirty)
117 * ->i_pages lock (page_remove_rmap->set_page_dirty)
118 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
119 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
120 * ->memcg->move_lock (page_remove_rmap->folio_memcg_lock)
121 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
122 * ->inode->i_lock (zap_pte_range->set_page_dirty)
123 * ->private_lock (zap_pte_range->block_dirty_folio)
126 static void page_cache_delete(struct address_space
*mapping
,
127 struct folio
*folio
, void *shadow
)
129 XA_STATE(xas
, &mapping
->i_pages
, folio
->index
);
132 mapping_set_update(&xas
, mapping
);
134 /* hugetlb pages are represented by a single entry in the xarray */
135 if (!folio_test_hugetlb(folio
)) {
136 xas_set_order(&xas
, folio
->index
, folio_order(folio
));
137 nr
= folio_nr_pages(folio
);
140 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
142 xas_store(&xas
, shadow
);
143 xas_init_marks(&xas
);
145 folio
->mapping
= NULL
;
146 /* Leave page->index set: truncation lookup relies upon it */
147 mapping
->nrpages
-= nr
;
150 static void filemap_unaccount_folio(struct address_space
*mapping
,
155 VM_BUG_ON_FOLIO(folio_mapped(folio
), folio
);
156 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(folio_mapped(folio
))) {
157 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
158 current
->comm
, folio_pfn(folio
));
159 dump_page(&folio
->page
, "still mapped when deleted");
161 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
163 if (mapping_exiting(mapping
) && !folio_test_large(folio
)) {
164 int mapcount
= page_mapcount(&folio
->page
);
166 if (folio_ref_count(folio
) >= mapcount
+ 2) {
168 * All vmas have already been torn down, so it's
169 * a good bet that actually the page is unmapped
170 * and we'd rather not leak it: if we're wrong,
171 * another bad page check should catch it later.
173 page_mapcount_reset(&folio
->page
);
174 folio_ref_sub(folio
, mapcount
);
179 /* hugetlb folios do not participate in page cache accounting. */
180 if (folio_test_hugetlb(folio
))
183 nr
= folio_nr_pages(folio
);
185 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, -nr
);
186 if (folio_test_swapbacked(folio
)) {
187 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, -nr
);
188 if (folio_test_pmd_mappable(folio
))
189 __lruvec_stat_mod_folio(folio
, NR_SHMEM_THPS
, -nr
);
190 } else if (folio_test_pmd_mappable(folio
)) {
191 __lruvec_stat_mod_folio(folio
, NR_FILE_THPS
, -nr
);
192 filemap_nr_thps_dec(mapping
);
196 * At this point folio must be either written or cleaned by
197 * truncate. Dirty folio here signals a bug and loss of
198 * unwritten data - on ordinary filesystems.
200 * But it's harmless on in-memory filesystems like tmpfs; and can
201 * occur when a driver which did get_user_pages() sets page dirty
202 * before putting it, while the inode is being finally evicted.
204 * Below fixes dirty accounting after removing the folio entirely
205 * but leaves the dirty flag set: it has no effect for truncated
206 * folio and anyway will be cleared before returning folio to
209 if (WARN_ON_ONCE(folio_test_dirty(folio
) &&
210 mapping_can_writeback(mapping
)))
211 folio_account_cleaned(folio
, inode_to_wb(mapping
->host
));
215 * Delete a page from the page cache and free it. Caller has to make
216 * sure the page is locked and that nobody else uses it - or that usage
217 * is safe. The caller must hold the i_pages lock.
219 void __filemap_remove_folio(struct folio
*folio
, void *shadow
)
221 struct address_space
*mapping
= folio
->mapping
;
223 trace_mm_filemap_delete_from_page_cache(folio
);
224 filemap_unaccount_folio(mapping
, folio
);
225 page_cache_delete(mapping
, folio
, shadow
);
228 void filemap_free_folio(struct address_space
*mapping
, struct folio
*folio
)
230 void (*free_folio
)(struct folio
*);
233 free_folio
= mapping
->a_ops
->free_folio
;
237 if (folio_test_large(folio
) && !folio_test_hugetlb(folio
))
238 refs
= folio_nr_pages(folio
);
239 folio_put_refs(folio
, refs
);
243 * filemap_remove_folio - Remove folio from page cache.
246 * This must be called only on folios that are locked and have been
247 * verified to be in the page cache. It will never put the folio into
248 * the free list because the caller has a reference on the page.
250 void filemap_remove_folio(struct folio
*folio
)
252 struct address_space
*mapping
= folio
->mapping
;
254 BUG_ON(!folio_test_locked(folio
));
255 spin_lock(&mapping
->host
->i_lock
);
256 xa_lock_irq(&mapping
->i_pages
);
257 __filemap_remove_folio(folio
, NULL
);
258 xa_unlock_irq(&mapping
->i_pages
);
259 if (mapping_shrinkable(mapping
))
260 inode_add_lru(mapping
->host
);
261 spin_unlock(&mapping
->host
->i_lock
);
263 filemap_free_folio(mapping
, folio
);
267 * page_cache_delete_batch - delete several folios from page cache
268 * @mapping: the mapping to which folios belong
269 * @fbatch: batch of folios to delete
271 * The function walks over mapping->i_pages and removes folios passed in
272 * @fbatch from the mapping. The function expects @fbatch to be sorted
273 * by page index and is optimised for it to be dense.
274 * It tolerates holes in @fbatch (mapping entries at those indices are not
277 * The function expects the i_pages lock to be held.
279 static void page_cache_delete_batch(struct address_space
*mapping
,
280 struct folio_batch
*fbatch
)
282 XA_STATE(xas
, &mapping
->i_pages
, fbatch
->folios
[0]->index
);
283 long total_pages
= 0;
287 mapping_set_update(&xas
, mapping
);
288 xas_for_each(&xas
, folio
, ULONG_MAX
) {
289 if (i
>= folio_batch_count(fbatch
))
292 /* A swap/dax/shadow entry got inserted? Skip it. */
293 if (xa_is_value(folio
))
296 * A page got inserted in our range? Skip it. We have our
297 * pages locked so they are protected from being removed.
298 * If we see a page whose index is higher than ours, it
299 * means our page has been removed, which shouldn't be
300 * possible because we're holding the PageLock.
302 if (folio
!= fbatch
->folios
[i
]) {
303 VM_BUG_ON_FOLIO(folio
->index
>
304 fbatch
->folios
[i
]->index
, folio
);
308 WARN_ON_ONCE(!folio_test_locked(folio
));
310 folio
->mapping
= NULL
;
311 /* Leave folio->index set: truncation lookup relies on it */
314 xas_store(&xas
, NULL
);
315 total_pages
+= folio_nr_pages(folio
);
317 mapping
->nrpages
-= total_pages
;
320 void delete_from_page_cache_batch(struct address_space
*mapping
,
321 struct folio_batch
*fbatch
)
325 if (!folio_batch_count(fbatch
))
328 spin_lock(&mapping
->host
->i_lock
);
329 xa_lock_irq(&mapping
->i_pages
);
330 for (i
= 0; i
< folio_batch_count(fbatch
); i
++) {
331 struct folio
*folio
= fbatch
->folios
[i
];
333 trace_mm_filemap_delete_from_page_cache(folio
);
334 filemap_unaccount_folio(mapping
, folio
);
336 page_cache_delete_batch(mapping
, fbatch
);
337 xa_unlock_irq(&mapping
->i_pages
);
338 if (mapping_shrinkable(mapping
))
339 inode_add_lru(mapping
->host
);
340 spin_unlock(&mapping
->host
->i_lock
);
342 for (i
= 0; i
< folio_batch_count(fbatch
); i
++)
343 filemap_free_folio(mapping
, fbatch
->folios
[i
]);
346 int filemap_check_errors(struct address_space
*mapping
)
349 /* Check for outstanding write errors */
350 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
351 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
353 if (test_bit(AS_EIO
, &mapping
->flags
) &&
354 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
358 EXPORT_SYMBOL(filemap_check_errors
);
360 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
362 /* Check for outstanding write errors */
363 if (test_bit(AS_EIO
, &mapping
->flags
))
365 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
371 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
372 * @mapping: address space structure to write
373 * @wbc: the writeback_control controlling the writeout
375 * Call writepages on the mapping using the provided wbc to control the
378 * Return: %0 on success, negative error code otherwise.
380 int filemap_fdatawrite_wbc(struct address_space
*mapping
,
381 struct writeback_control
*wbc
)
385 if (!mapping_can_writeback(mapping
) ||
386 !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
389 wbc_attach_fdatawrite_inode(wbc
, mapping
->host
);
390 ret
= do_writepages(mapping
, wbc
);
391 wbc_detach_inode(wbc
);
394 EXPORT_SYMBOL(filemap_fdatawrite_wbc
);
397 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
398 * @mapping: address space structure to write
399 * @start: offset in bytes where the range starts
400 * @end: offset in bytes where the range ends (inclusive)
401 * @sync_mode: enable synchronous operation
403 * Start writeback against all of a mapping's dirty pages that lie
404 * within the byte offsets <start, end> inclusive.
406 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
407 * opposed to a regular memory cleansing writeback. The difference between
408 * these two operations is that if a dirty page/buffer is encountered, it must
409 * be waited upon, and not just skipped over.
411 * Return: %0 on success, negative error code otherwise.
413 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
414 loff_t end
, int sync_mode
)
416 struct writeback_control wbc
= {
417 .sync_mode
= sync_mode
,
418 .nr_to_write
= LONG_MAX
,
419 .range_start
= start
,
423 return filemap_fdatawrite_wbc(mapping
, &wbc
);
426 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
429 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
432 int filemap_fdatawrite(struct address_space
*mapping
)
434 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
436 EXPORT_SYMBOL(filemap_fdatawrite
);
438 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
441 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
443 EXPORT_SYMBOL(filemap_fdatawrite_range
);
446 * filemap_flush - mostly a non-blocking flush
447 * @mapping: target address_space
449 * This is a mostly non-blocking flush. Not suitable for data-integrity
450 * purposes - I/O may not be started against all dirty pages.
452 * Return: %0 on success, negative error code otherwise.
454 int filemap_flush(struct address_space
*mapping
)
456 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
458 EXPORT_SYMBOL(filemap_flush
);
461 * filemap_range_has_page - check if a page exists in range.
462 * @mapping: address space within which to check
463 * @start_byte: offset in bytes where the range starts
464 * @end_byte: offset in bytes where the range ends (inclusive)
466 * Find at least one page in the range supplied, usually used to check if
467 * direct writing in this range will trigger a writeback.
469 * Return: %true if at least one page exists in the specified range,
472 bool filemap_range_has_page(struct address_space
*mapping
,
473 loff_t start_byte
, loff_t end_byte
)
476 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
477 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
479 if (end_byte
< start_byte
)
484 folio
= xas_find(&xas
, max
);
485 if (xas_retry(&xas
, folio
))
487 /* Shadow entries don't count */
488 if (xa_is_value(folio
))
491 * We don't need to try to pin this page; we're about to
492 * release the RCU lock anyway. It is enough to know that
493 * there was a page here recently.
499 return folio
!= NULL
;
501 EXPORT_SYMBOL(filemap_range_has_page
);
503 static void __filemap_fdatawait_range(struct address_space
*mapping
,
504 loff_t start_byte
, loff_t end_byte
)
506 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
507 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
508 struct folio_batch fbatch
;
511 folio_batch_init(&fbatch
);
513 while (index
<= end
) {
516 nr_folios
= filemap_get_folios_tag(mapping
, &index
, end
,
517 PAGECACHE_TAG_WRITEBACK
, &fbatch
);
522 for (i
= 0; i
< nr_folios
; i
++) {
523 struct folio
*folio
= fbatch
.folios
[i
];
525 folio_wait_writeback(folio
);
526 folio_clear_error(folio
);
528 folio_batch_release(&fbatch
);
534 * filemap_fdatawait_range - wait for writeback to complete
535 * @mapping: address space structure to wait for
536 * @start_byte: offset in bytes where the range starts
537 * @end_byte: offset in bytes where the range ends (inclusive)
539 * Walk the list of under-writeback pages of the given address space
540 * in the given range and wait for all of them. Check error status of
541 * the address space and return it.
543 * Since the error status of the address space is cleared by this function,
544 * callers are responsible for checking the return value and handling and/or
545 * reporting the error.
547 * Return: error status of the address space.
549 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
552 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
553 return filemap_check_errors(mapping
);
555 EXPORT_SYMBOL(filemap_fdatawait_range
);
558 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
559 * @mapping: address space structure to wait for
560 * @start_byte: offset in bytes where the range starts
561 * @end_byte: offset in bytes where the range ends (inclusive)
563 * Walk the list of under-writeback pages of the given address space in the
564 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
565 * this function does not clear error status of the address space.
567 * Use this function if callers don't handle errors themselves. Expected
568 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
571 int filemap_fdatawait_range_keep_errors(struct address_space
*mapping
,
572 loff_t start_byte
, loff_t end_byte
)
574 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
575 return filemap_check_and_keep_errors(mapping
);
577 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors
);
580 * file_fdatawait_range - wait for writeback to complete
581 * @file: file pointing to address space structure to wait for
582 * @start_byte: offset in bytes where the range starts
583 * @end_byte: offset in bytes where the range ends (inclusive)
585 * Walk the list of under-writeback pages of the address space that file
586 * refers to, in the given range and wait for all of them. Check error
587 * status of the address space vs. the file->f_wb_err cursor and return it.
589 * Since the error status of the file is advanced by this function,
590 * callers are responsible for checking the return value and handling and/or
591 * reporting the error.
593 * Return: error status of the address space vs. the file->f_wb_err cursor.
595 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
597 struct address_space
*mapping
= file
->f_mapping
;
599 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
600 return file_check_and_advance_wb_err(file
);
602 EXPORT_SYMBOL(file_fdatawait_range
);
605 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
606 * @mapping: address space structure to wait for
608 * Walk the list of under-writeback pages of the given address space
609 * and wait for all of them. Unlike filemap_fdatawait(), this function
610 * does not clear error status of the address space.
612 * Use this function if callers don't handle errors themselves. Expected
613 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
616 * Return: error status of the address space.
618 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
620 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
621 return filemap_check_and_keep_errors(mapping
);
623 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
625 /* Returns true if writeback might be needed or already in progress. */
626 static bool mapping_needs_writeback(struct address_space
*mapping
)
628 return mapping
->nrpages
;
631 bool filemap_range_has_writeback(struct address_space
*mapping
,
632 loff_t start_byte
, loff_t end_byte
)
634 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
635 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
638 if (end_byte
< start_byte
)
642 xas_for_each(&xas
, folio
, max
) {
643 if (xas_retry(&xas
, folio
))
645 if (xa_is_value(folio
))
647 if (folio_test_dirty(folio
) || folio_test_locked(folio
) ||
648 folio_test_writeback(folio
))
652 return folio
!= NULL
;
654 EXPORT_SYMBOL_GPL(filemap_range_has_writeback
);
657 * filemap_write_and_wait_range - write out & wait on a file range
658 * @mapping: the address_space for the pages
659 * @lstart: offset in bytes where the range starts
660 * @lend: offset in bytes where the range ends (inclusive)
662 * Write out and wait upon file offsets lstart->lend, inclusive.
664 * Note that @lend is inclusive (describes the last byte to be written) so
665 * that this function can be used to write to the very end-of-file (end = -1).
667 * Return: error status of the address space.
669 int filemap_write_and_wait_range(struct address_space
*mapping
,
670 loff_t lstart
, loff_t lend
)
677 if (mapping_needs_writeback(mapping
)) {
678 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
681 * Even if the above returned error, the pages may be
682 * written partially (e.g. -ENOSPC), so we wait for it.
683 * But the -EIO is special case, it may indicate the worst
684 * thing (e.g. bug) happened, so we avoid waiting for it.
687 __filemap_fdatawait_range(mapping
, lstart
, lend
);
689 err2
= filemap_check_errors(mapping
);
694 EXPORT_SYMBOL(filemap_write_and_wait_range
);
696 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
698 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
700 trace_filemap_set_wb_err(mapping
, eseq
);
702 EXPORT_SYMBOL(__filemap_set_wb_err
);
705 * file_check_and_advance_wb_err - report wb error (if any) that was previously
706 * and advance wb_err to current one
707 * @file: struct file on which the error is being reported
709 * When userland calls fsync (or something like nfsd does the equivalent), we
710 * want to report any writeback errors that occurred since the last fsync (or
711 * since the file was opened if there haven't been any).
713 * Grab the wb_err from the mapping. If it matches what we have in the file,
714 * then just quickly return 0. The file is all caught up.
716 * If it doesn't match, then take the mapping value, set the "seen" flag in
717 * it and try to swap it into place. If it works, or another task beat us
718 * to it with the new value, then update the f_wb_err and return the error
719 * portion. The error at this point must be reported via proper channels
720 * (a'la fsync, or NFS COMMIT operation, etc.).
722 * While we handle mapping->wb_err with atomic operations, the f_wb_err
723 * value is protected by the f_lock since we must ensure that it reflects
724 * the latest value swapped in for this file descriptor.
726 * Return: %0 on success, negative error code otherwise.
728 int file_check_and_advance_wb_err(struct file
*file
)
731 errseq_t old
= READ_ONCE(file
->f_wb_err
);
732 struct address_space
*mapping
= file
->f_mapping
;
734 /* Locklessly handle the common case where nothing has changed */
735 if (errseq_check(&mapping
->wb_err
, old
)) {
736 /* Something changed, must use slow path */
737 spin_lock(&file
->f_lock
);
738 old
= file
->f_wb_err
;
739 err
= errseq_check_and_advance(&mapping
->wb_err
,
741 trace_file_check_and_advance_wb_err(file
, old
);
742 spin_unlock(&file
->f_lock
);
746 * We're mostly using this function as a drop in replacement for
747 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
748 * that the legacy code would have had on these flags.
750 clear_bit(AS_EIO
, &mapping
->flags
);
751 clear_bit(AS_ENOSPC
, &mapping
->flags
);
754 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
757 * file_write_and_wait_range - write out & wait on a file range
758 * @file: file pointing to address_space with pages
759 * @lstart: offset in bytes where the range starts
760 * @lend: offset in bytes where the range ends (inclusive)
762 * Write out and wait upon file offsets lstart->lend, inclusive.
764 * Note that @lend is inclusive (describes the last byte to be written) so
765 * that this function can be used to write to the very end-of-file (end = -1).
767 * After writing out and waiting on the data, we check and advance the
768 * f_wb_err cursor to the latest value, and return any errors detected there.
770 * Return: %0 on success, negative error code otherwise.
772 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
775 struct address_space
*mapping
= file
->f_mapping
;
780 if (mapping_needs_writeback(mapping
)) {
781 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
783 /* See comment of filemap_write_and_wait() */
785 __filemap_fdatawait_range(mapping
, lstart
, lend
);
787 err2
= file_check_and_advance_wb_err(file
);
792 EXPORT_SYMBOL(file_write_and_wait_range
);
795 * replace_page_cache_folio - replace a pagecache folio with a new one
796 * @old: folio to be replaced
797 * @new: folio to replace with
799 * This function replaces a folio in the pagecache with a new one. On
800 * success it acquires the pagecache reference for the new folio and
801 * drops it for the old folio. Both the old and new folios must be
802 * locked. This function does not add the new folio to the LRU, the
803 * caller must do that.
805 * The remove + add is atomic. This function cannot fail.
807 void replace_page_cache_folio(struct folio
*old
, struct folio
*new)
809 struct address_space
*mapping
= old
->mapping
;
810 void (*free_folio
)(struct folio
*) = mapping
->a_ops
->free_folio
;
811 pgoff_t offset
= old
->index
;
812 XA_STATE(xas
, &mapping
->i_pages
, offset
);
814 VM_BUG_ON_FOLIO(!folio_test_locked(old
), old
);
815 VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
816 VM_BUG_ON_FOLIO(new->mapping
, new);
819 new->mapping
= mapping
;
822 mem_cgroup_migrate(old
, new);
825 xas_store(&xas
, new);
828 /* hugetlb pages do not participate in page cache accounting. */
829 if (!folio_test_hugetlb(old
))
830 __lruvec_stat_sub_folio(old
, NR_FILE_PAGES
);
831 if (!folio_test_hugetlb(new))
832 __lruvec_stat_add_folio(new, NR_FILE_PAGES
);
833 if (folio_test_swapbacked(old
))
834 __lruvec_stat_sub_folio(old
, NR_SHMEM
);
835 if (folio_test_swapbacked(new))
836 __lruvec_stat_add_folio(new, NR_SHMEM
);
837 xas_unlock_irq(&xas
);
842 EXPORT_SYMBOL_GPL(replace_page_cache_folio
);
844 noinline
int __filemap_add_folio(struct address_space
*mapping
,
845 struct folio
*folio
, pgoff_t index
, gfp_t gfp
, void **shadowp
)
847 XA_STATE(xas
, &mapping
->i_pages
, index
);
848 int huge
= folio_test_hugetlb(folio
);
849 bool charged
= false;
852 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
853 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio
), folio
);
854 mapping_set_update(&xas
, mapping
);
857 int error
= mem_cgroup_charge(folio
, NULL
, gfp
);
858 VM_BUG_ON_FOLIO(index
& (folio_nr_pages(folio
) - 1), folio
);
862 xas_set_order(&xas
, index
, folio_order(folio
));
863 nr
= folio_nr_pages(folio
);
866 gfp
&= GFP_RECLAIM_MASK
;
867 folio_ref_add(folio
, nr
);
868 folio
->mapping
= mapping
;
869 folio
->index
= xas
.xa_index
;
872 unsigned int order
= xa_get_order(xas
.xa
, xas
.xa_index
);
873 void *entry
, *old
= NULL
;
875 if (order
> folio_order(folio
))
876 xas_split_alloc(&xas
, xa_load(xas
.xa
, xas
.xa_index
),
879 xas_for_each_conflict(&xas
, entry
) {
881 if (!xa_is_value(entry
)) {
882 xas_set_err(&xas
, -EEXIST
);
890 /* entry may have been split before we acquired lock */
891 order
= xa_get_order(xas
.xa
, xas
.xa_index
);
892 if (order
> folio_order(folio
)) {
893 /* How to handle large swap entries? */
894 BUG_ON(shmem_mapping(mapping
));
895 xas_split(&xas
, old
, order
);
900 xas_store(&xas
, folio
);
904 mapping
->nrpages
+= nr
;
906 /* hugetlb pages do not participate in page cache accounting */
908 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, nr
);
909 if (folio_test_pmd_mappable(folio
))
910 __lruvec_stat_mod_folio(folio
,
914 xas_unlock_irq(&xas
);
915 } while (xas_nomem(&xas
, gfp
));
920 trace_mm_filemap_add_to_page_cache(folio
);
924 mem_cgroup_uncharge(folio
);
925 folio
->mapping
= NULL
;
926 /* Leave page->index set: truncation relies upon it */
927 folio_put_refs(folio
, nr
);
928 return xas_error(&xas
);
930 ALLOW_ERROR_INJECTION(__filemap_add_folio
, ERRNO
);
932 int filemap_add_folio(struct address_space
*mapping
, struct folio
*folio
,
933 pgoff_t index
, gfp_t gfp
)
938 __folio_set_locked(folio
);
939 ret
= __filemap_add_folio(mapping
, folio
, index
, gfp
, &shadow
);
941 __folio_clear_locked(folio
);
944 * The folio might have been evicted from cache only
945 * recently, in which case it should be activated like
946 * any other repeatedly accessed folio.
947 * The exception is folios getting rewritten; evicting other
948 * data from the working set, only to cache data that will
949 * get overwritten with something else, is a waste of memory.
951 WARN_ON_ONCE(folio_test_active(folio
));
952 if (!(gfp
& __GFP_WRITE
) && shadow
)
953 workingset_refault(folio
, shadow
);
954 folio_add_lru(folio
);
958 EXPORT_SYMBOL_GPL(filemap_add_folio
);
961 struct folio
*filemap_alloc_folio(gfp_t gfp
, unsigned int order
)
966 if (cpuset_do_page_mem_spread()) {
967 unsigned int cpuset_mems_cookie
;
969 cpuset_mems_cookie
= read_mems_allowed_begin();
970 n
= cpuset_mem_spread_node();
971 folio
= __folio_alloc_node(gfp
, order
, n
);
972 } while (!folio
&& read_mems_allowed_retry(cpuset_mems_cookie
));
976 return folio_alloc(gfp
, order
);
978 EXPORT_SYMBOL(filemap_alloc_folio
);
982 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
984 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
986 * @mapping1: the first mapping to lock
987 * @mapping2: the second mapping to lock
989 void filemap_invalidate_lock_two(struct address_space
*mapping1
,
990 struct address_space
*mapping2
)
992 if (mapping1
> mapping2
)
993 swap(mapping1
, mapping2
);
995 down_write(&mapping1
->invalidate_lock
);
996 if (mapping2
&& mapping1
!= mapping2
)
997 down_write_nested(&mapping2
->invalidate_lock
, 1);
999 EXPORT_SYMBOL(filemap_invalidate_lock_two
);
1002 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1004 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1006 * @mapping1: the first mapping to unlock
1007 * @mapping2: the second mapping to unlock
1009 void filemap_invalidate_unlock_two(struct address_space
*mapping1
,
1010 struct address_space
*mapping2
)
1013 up_write(&mapping1
->invalidate_lock
);
1014 if (mapping2
&& mapping1
!= mapping2
)
1015 up_write(&mapping2
->invalidate_lock
);
1017 EXPORT_SYMBOL(filemap_invalidate_unlock_two
);
1020 * In order to wait for pages to become available there must be
1021 * waitqueues associated with pages. By using a hash table of
1022 * waitqueues where the bucket discipline is to maintain all
1023 * waiters on the same queue and wake all when any of the pages
1024 * become available, and for the woken contexts to check to be
1025 * sure the appropriate page became available, this saves space
1026 * at a cost of "thundering herd" phenomena during rare hash
1029 #define PAGE_WAIT_TABLE_BITS 8
1030 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1031 static wait_queue_head_t folio_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
1033 static wait_queue_head_t
*folio_waitqueue(struct folio
*folio
)
1035 return &folio_wait_table
[hash_ptr(folio
, PAGE_WAIT_TABLE_BITS
)];
1038 void __init
pagecache_init(void)
1042 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
1043 init_waitqueue_head(&folio_wait_table
[i
]);
1045 page_writeback_init();
1049 * The page wait code treats the "wait->flags" somewhat unusually, because
1050 * we have multiple different kinds of waits, not just the usual "exclusive"
1055 * (a) no special bits set:
1057 * We're just waiting for the bit to be released, and when a waker
1058 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1059 * and remove it from the wait queue.
1061 * Simple and straightforward.
1063 * (b) WQ_FLAG_EXCLUSIVE:
1065 * The waiter is waiting to get the lock, and only one waiter should
1066 * be woken up to avoid any thundering herd behavior. We'll set the
1067 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1069 * This is the traditional exclusive wait.
1071 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1073 * The waiter is waiting to get the bit, and additionally wants the
1074 * lock to be transferred to it for fair lock behavior. If the lock
1075 * cannot be taken, we stop walking the wait queue without waking
1078 * This is the "fair lock handoff" case, and in addition to setting
1079 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1080 * that it now has the lock.
1082 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
1085 struct wait_page_key
*key
= arg
;
1086 struct wait_page_queue
*wait_page
1087 = container_of(wait
, struct wait_page_queue
, wait
);
1089 if (!wake_page_match(wait_page
, key
))
1093 * If it's a lock handoff wait, we get the bit for it, and
1094 * stop walking (and do not wake it up) if we can't.
1096 flags
= wait
->flags
;
1097 if (flags
& WQ_FLAG_EXCLUSIVE
) {
1098 if (test_bit(key
->bit_nr
, &key
->folio
->flags
))
1100 if (flags
& WQ_FLAG_CUSTOM
) {
1101 if (test_and_set_bit(key
->bit_nr
, &key
->folio
->flags
))
1103 flags
|= WQ_FLAG_DONE
;
1108 * We are holding the wait-queue lock, but the waiter that
1109 * is waiting for this will be checking the flags without
1112 * So update the flags atomically, and wake up the waiter
1113 * afterwards to avoid any races. This store-release pairs
1114 * with the load-acquire in folio_wait_bit_common().
1116 smp_store_release(&wait
->flags
, flags
| WQ_FLAG_WOKEN
);
1117 wake_up_state(wait
->private, mode
);
1120 * Ok, we have successfully done what we're waiting for,
1121 * and we can unconditionally remove the wait entry.
1123 * Note that this pairs with the "finish_wait()" in the
1124 * waiter, and has to be the absolute last thing we do.
1125 * After this list_del_init(&wait->entry) the wait entry
1126 * might be de-allocated and the process might even have
1129 list_del_init_careful(&wait
->entry
);
1130 return (flags
& WQ_FLAG_EXCLUSIVE
) != 0;
1133 static void folio_wake_bit(struct folio
*folio
, int bit_nr
)
1135 wait_queue_head_t
*q
= folio_waitqueue(folio
);
1136 struct wait_page_key key
;
1137 unsigned long flags
;
1138 wait_queue_entry_t bookmark
;
1141 key
.bit_nr
= bit_nr
;
1145 bookmark
.private = NULL
;
1146 bookmark
.func
= NULL
;
1147 INIT_LIST_HEAD(&bookmark
.entry
);
1149 spin_lock_irqsave(&q
->lock
, flags
);
1150 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1152 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1154 * Take a breather from holding the lock,
1155 * allow pages that finish wake up asynchronously
1156 * to acquire the lock and remove themselves
1159 spin_unlock_irqrestore(&q
->lock
, flags
);
1161 spin_lock_irqsave(&q
->lock
, flags
);
1162 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1166 * It's possible to miss clearing waiters here, when we woke our page
1167 * waiters, but the hashed waitqueue has waiters for other pages on it.
1168 * That's okay, it's a rare case. The next waker will clear it.
1170 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1171 * other), the flag may be cleared in the course of freeing the page;
1172 * but that is not required for correctness.
1174 if (!waitqueue_active(q
) || !key
.page_match
)
1175 folio_clear_waiters(folio
);
1177 spin_unlock_irqrestore(&q
->lock
, flags
);
1180 static void folio_wake(struct folio
*folio
, int bit
)
1182 if (!folio_test_waiters(folio
))
1184 folio_wake_bit(folio
, bit
);
1188 * A choice of three behaviors for folio_wait_bit_common():
1191 EXCLUSIVE
, /* Hold ref to page and take the bit when woken, like
1192 * __folio_lock() waiting on then setting PG_locked.
1194 SHARED
, /* Hold ref to page and check the bit when woken, like
1195 * folio_wait_writeback() waiting on PG_writeback.
1197 DROP
, /* Drop ref to page before wait, no check when woken,
1198 * like folio_put_wait_locked() on PG_locked.
1203 * Attempt to check (or get) the folio flag, and mark us done
1206 static inline bool folio_trylock_flag(struct folio
*folio
, int bit_nr
,
1207 struct wait_queue_entry
*wait
)
1209 if (wait
->flags
& WQ_FLAG_EXCLUSIVE
) {
1210 if (test_and_set_bit(bit_nr
, &folio
->flags
))
1212 } else if (test_bit(bit_nr
, &folio
->flags
))
1215 wait
->flags
|= WQ_FLAG_WOKEN
| WQ_FLAG_DONE
;
1219 /* How many times do we accept lock stealing from under a waiter? */
1220 int sysctl_page_lock_unfairness
= 5;
1222 static inline int folio_wait_bit_common(struct folio
*folio
, int bit_nr
,
1223 int state
, enum behavior behavior
)
1225 wait_queue_head_t
*q
= folio_waitqueue(folio
);
1226 int unfairness
= sysctl_page_lock_unfairness
;
1227 struct wait_page_queue wait_page
;
1228 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1229 bool thrashing
= false;
1230 unsigned long pflags
;
1233 if (bit_nr
== PG_locked
&&
1234 !folio_test_uptodate(folio
) && folio_test_workingset(folio
)) {
1235 delayacct_thrashing_start(&in_thrashing
);
1236 psi_memstall_enter(&pflags
);
1241 wait
->func
= wake_page_function
;
1242 wait_page
.folio
= folio
;
1243 wait_page
.bit_nr
= bit_nr
;
1247 if (behavior
== EXCLUSIVE
) {
1248 wait
->flags
= WQ_FLAG_EXCLUSIVE
;
1249 if (--unfairness
< 0)
1250 wait
->flags
|= WQ_FLAG_CUSTOM
;
1254 * Do one last check whether we can get the
1255 * page bit synchronously.
1257 * Do the folio_set_waiters() marking before that
1258 * to let any waker we _just_ missed know they
1259 * need to wake us up (otherwise they'll never
1260 * even go to the slow case that looks at the
1261 * page queue), and add ourselves to the wait
1262 * queue if we need to sleep.
1264 * This part needs to be done under the queue
1265 * lock to avoid races.
1267 spin_lock_irq(&q
->lock
);
1268 folio_set_waiters(folio
);
1269 if (!folio_trylock_flag(folio
, bit_nr
, wait
))
1270 __add_wait_queue_entry_tail(q
, wait
);
1271 spin_unlock_irq(&q
->lock
);
1274 * From now on, all the logic will be based on
1275 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1276 * see whether the page bit testing has already
1277 * been done by the wake function.
1279 * We can drop our reference to the folio.
1281 if (behavior
== DROP
)
1285 * Note that until the "finish_wait()", or until
1286 * we see the WQ_FLAG_WOKEN flag, we need to
1287 * be very careful with the 'wait->flags', because
1288 * we may race with a waker that sets them.
1293 set_current_state(state
);
1295 /* Loop until we've been woken or interrupted */
1296 flags
= smp_load_acquire(&wait
->flags
);
1297 if (!(flags
& WQ_FLAG_WOKEN
)) {
1298 if (signal_pending_state(state
, current
))
1305 /* If we were non-exclusive, we're done */
1306 if (behavior
!= EXCLUSIVE
)
1309 /* If the waker got the lock for us, we're done */
1310 if (flags
& WQ_FLAG_DONE
)
1314 * Otherwise, if we're getting the lock, we need to
1315 * try to get it ourselves.
1317 * And if that fails, we'll have to retry this all.
1319 if (unlikely(test_and_set_bit(bit_nr
, folio_flags(folio
, 0))))
1322 wait
->flags
|= WQ_FLAG_DONE
;
1327 * If a signal happened, this 'finish_wait()' may remove the last
1328 * waiter from the wait-queues, but the folio waiters bit will remain
1329 * set. That's ok. The next wakeup will take care of it, and trying
1330 * to do it here would be difficult and prone to races.
1332 finish_wait(q
, wait
);
1335 delayacct_thrashing_end(&in_thrashing
);
1336 psi_memstall_leave(&pflags
);
1340 * NOTE! The wait->flags weren't stable until we've done the
1341 * 'finish_wait()', and we could have exited the loop above due
1342 * to a signal, and had a wakeup event happen after the signal
1343 * test but before the 'finish_wait()'.
1345 * So only after the finish_wait() can we reliably determine
1346 * if we got woken up or not, so we can now figure out the final
1347 * return value based on that state without races.
1349 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1350 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1352 if (behavior
== EXCLUSIVE
)
1353 return wait
->flags
& WQ_FLAG_DONE
? 0 : -EINTR
;
1355 return wait
->flags
& WQ_FLAG_WOKEN
? 0 : -EINTR
;
1358 #ifdef CONFIG_MIGRATION
1360 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1361 * @entry: migration swap entry.
1362 * @ptl: already locked ptl. This function will drop the lock.
1364 * Wait for a migration entry referencing the given page to be removed. This is
1365 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1366 * this can be called without taking a reference on the page. Instead this
1367 * should be called while holding the ptl for the migration entry referencing
1370 * Returns after unlocking the ptl.
1372 * This follows the same logic as folio_wait_bit_common() so see the comments
1375 void migration_entry_wait_on_locked(swp_entry_t entry
, spinlock_t
*ptl
)
1378 struct wait_page_queue wait_page
;
1379 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1380 bool thrashing
= false;
1381 unsigned long pflags
;
1383 wait_queue_head_t
*q
;
1384 struct folio
*folio
= page_folio(pfn_swap_entry_to_page(entry
));
1386 q
= folio_waitqueue(folio
);
1387 if (!folio_test_uptodate(folio
) && folio_test_workingset(folio
)) {
1388 delayacct_thrashing_start(&in_thrashing
);
1389 psi_memstall_enter(&pflags
);
1394 wait
->func
= wake_page_function
;
1395 wait_page
.folio
= folio
;
1396 wait_page
.bit_nr
= PG_locked
;
1399 spin_lock_irq(&q
->lock
);
1400 folio_set_waiters(folio
);
1401 if (!folio_trylock_flag(folio
, PG_locked
, wait
))
1402 __add_wait_queue_entry_tail(q
, wait
);
1403 spin_unlock_irq(&q
->lock
);
1406 * If a migration entry exists for the page the migration path must hold
1407 * a valid reference to the page, and it must take the ptl to remove the
1408 * migration entry. So the page is valid until the ptl is dropped.
1415 set_current_state(TASK_UNINTERRUPTIBLE
);
1417 /* Loop until we've been woken or interrupted */
1418 flags
= smp_load_acquire(&wait
->flags
);
1419 if (!(flags
& WQ_FLAG_WOKEN
)) {
1420 if (signal_pending_state(TASK_UNINTERRUPTIBLE
, current
))
1429 finish_wait(q
, wait
);
1432 delayacct_thrashing_end(&in_thrashing
);
1433 psi_memstall_leave(&pflags
);
1438 void folio_wait_bit(struct folio
*folio
, int bit_nr
)
1440 folio_wait_bit_common(folio
, bit_nr
, TASK_UNINTERRUPTIBLE
, SHARED
);
1442 EXPORT_SYMBOL(folio_wait_bit
);
1444 int folio_wait_bit_killable(struct folio
*folio
, int bit_nr
)
1446 return folio_wait_bit_common(folio
, bit_nr
, TASK_KILLABLE
, SHARED
);
1448 EXPORT_SYMBOL(folio_wait_bit_killable
);
1451 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1452 * @folio: The folio to wait for.
1453 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1455 * The caller should hold a reference on @folio. They expect the page to
1456 * become unlocked relatively soon, but do not wish to hold up migration
1457 * (for example) by holding the reference while waiting for the folio to
1458 * come unlocked. After this function returns, the caller should not
1459 * dereference @folio.
1461 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1463 static int folio_put_wait_locked(struct folio
*folio
, int state
)
1465 return folio_wait_bit_common(folio
, PG_locked
, state
, DROP
);
1469 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1470 * @folio: Folio defining the wait queue of interest
1471 * @waiter: Waiter to add to the queue
1473 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1475 void folio_add_wait_queue(struct folio
*folio
, wait_queue_entry_t
*waiter
)
1477 wait_queue_head_t
*q
= folio_waitqueue(folio
);
1478 unsigned long flags
;
1480 spin_lock_irqsave(&q
->lock
, flags
);
1481 __add_wait_queue_entry_tail(q
, waiter
);
1482 folio_set_waiters(folio
);
1483 spin_unlock_irqrestore(&q
->lock
, flags
);
1485 EXPORT_SYMBOL_GPL(folio_add_wait_queue
);
1487 #ifndef clear_bit_unlock_is_negative_byte
1490 * PG_waiters is the high bit in the same byte as PG_lock.
1492 * On x86 (and on many other architectures), we can clear PG_lock and
1493 * test the sign bit at the same time. But if the architecture does
1494 * not support that special operation, we just do this all by hand
1497 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1498 * being cleared, but a memory barrier should be unnecessary since it is
1499 * in the same byte as PG_locked.
1501 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1503 clear_bit_unlock(nr
, mem
);
1504 /* smp_mb__after_atomic(); */
1505 return test_bit(PG_waiters
, mem
);
1511 * folio_unlock - Unlock a locked folio.
1512 * @folio: The folio.
1514 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1516 * Context: May be called from interrupt or process context. May not be
1517 * called from NMI context.
1519 void folio_unlock(struct folio
*folio
)
1521 /* Bit 7 allows x86 to check the byte's sign bit */
1522 BUILD_BUG_ON(PG_waiters
!= 7);
1523 BUILD_BUG_ON(PG_locked
> 7);
1524 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
1525 if (clear_bit_unlock_is_negative_byte(PG_locked
, folio_flags(folio
, 0)))
1526 folio_wake_bit(folio
, PG_locked
);
1528 EXPORT_SYMBOL(folio_unlock
);
1531 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1532 * @folio: The folio.
1534 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1535 * it. The folio reference held for PG_private_2 being set is released.
1537 * This is, for example, used when a netfs folio is being written to a local
1538 * disk cache, thereby allowing writes to the cache for the same folio to be
1541 void folio_end_private_2(struct folio
*folio
)
1543 VM_BUG_ON_FOLIO(!folio_test_private_2(folio
), folio
);
1544 clear_bit_unlock(PG_private_2
, folio_flags(folio
, 0));
1545 folio_wake_bit(folio
, PG_private_2
);
1548 EXPORT_SYMBOL(folio_end_private_2
);
1551 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1552 * @folio: The folio to wait on.
1554 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1556 void folio_wait_private_2(struct folio
*folio
)
1558 while (folio_test_private_2(folio
))
1559 folio_wait_bit(folio
, PG_private_2
);
1561 EXPORT_SYMBOL(folio_wait_private_2
);
1564 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1565 * @folio: The folio to wait on.
1567 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1568 * fatal signal is received by the calling task.
1571 * - 0 if successful.
1572 * - -EINTR if a fatal signal was encountered.
1574 int folio_wait_private_2_killable(struct folio
*folio
)
1578 while (folio_test_private_2(folio
)) {
1579 ret
= folio_wait_bit_killable(folio
, PG_private_2
);
1586 EXPORT_SYMBOL(folio_wait_private_2_killable
);
1589 * folio_end_writeback - End writeback against a folio.
1590 * @folio: The folio.
1592 void folio_end_writeback(struct folio
*folio
)
1595 * folio_test_clear_reclaim() could be used here but it is an
1596 * atomic operation and overkill in this particular case. Failing
1597 * to shuffle a folio marked for immediate reclaim is too mild
1598 * a gain to justify taking an atomic operation penalty at the
1599 * end of every folio writeback.
1601 if (folio_test_reclaim(folio
)) {
1602 folio_clear_reclaim(folio
);
1603 folio_rotate_reclaimable(folio
);
1607 * Writeback does not hold a folio reference of its own, relying
1608 * on truncation to wait for the clearing of PG_writeback.
1609 * But here we must make sure that the folio is not freed and
1610 * reused before the folio_wake().
1613 if (!__folio_end_writeback(folio
))
1616 smp_mb__after_atomic();
1617 folio_wake(folio
, PG_writeback
);
1618 acct_reclaim_writeback(folio
);
1621 EXPORT_SYMBOL(folio_end_writeback
);
1624 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1625 * @folio: The folio to lock
1627 void __folio_lock(struct folio
*folio
)
1629 folio_wait_bit_common(folio
, PG_locked
, TASK_UNINTERRUPTIBLE
,
1632 EXPORT_SYMBOL(__folio_lock
);
1634 int __folio_lock_killable(struct folio
*folio
)
1636 return folio_wait_bit_common(folio
, PG_locked
, TASK_KILLABLE
,
1639 EXPORT_SYMBOL_GPL(__folio_lock_killable
);
1641 static int __folio_lock_async(struct folio
*folio
, struct wait_page_queue
*wait
)
1643 struct wait_queue_head
*q
= folio_waitqueue(folio
);
1646 wait
->folio
= folio
;
1647 wait
->bit_nr
= PG_locked
;
1649 spin_lock_irq(&q
->lock
);
1650 __add_wait_queue_entry_tail(q
, &wait
->wait
);
1651 folio_set_waiters(folio
);
1652 ret
= !folio_trylock(folio
);
1654 * If we were successful now, we know we're still on the
1655 * waitqueue as we're still under the lock. This means it's
1656 * safe to remove and return success, we know the callback
1657 * isn't going to trigger.
1660 __remove_wait_queue(q
, &wait
->wait
);
1663 spin_unlock_irq(&q
->lock
);
1669 * 0 - folio is locked.
1670 * non-zero - folio is not locked.
1671 * mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
1672 * vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
1673 * FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
1675 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
1676 * with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
1678 vm_fault_t
__folio_lock_or_retry(struct folio
*folio
, struct vm_fault
*vmf
)
1680 unsigned int flags
= vmf
->flags
;
1682 if (fault_flag_allow_retry_first(flags
)) {
1684 * CAUTION! In this case, mmap_lock/per-VMA lock is not
1685 * released even though returning VM_FAULT_RETRY.
1687 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1688 return VM_FAULT_RETRY
;
1690 release_fault_lock(vmf
);
1691 if (flags
& FAULT_FLAG_KILLABLE
)
1692 folio_wait_locked_killable(folio
);
1694 folio_wait_locked(folio
);
1695 return VM_FAULT_RETRY
;
1697 if (flags
& FAULT_FLAG_KILLABLE
) {
1700 ret
= __folio_lock_killable(folio
);
1702 release_fault_lock(vmf
);
1703 return VM_FAULT_RETRY
;
1706 __folio_lock(folio
);
1713 * page_cache_next_miss() - Find the next gap in the page cache.
1714 * @mapping: Mapping.
1716 * @max_scan: Maximum range to search.
1718 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1719 * gap with the lowest index.
1721 * This function may be called under the rcu_read_lock. However, this will
1722 * not atomically search a snapshot of the cache at a single point in time.
1723 * For example, if a gap is created at index 5, then subsequently a gap is
1724 * created at index 10, page_cache_next_miss covering both indices may
1725 * return 10 if called under the rcu_read_lock.
1727 * Return: The index of the gap if found, otherwise an index outside the
1728 * range specified (in which case 'return - index >= max_scan' will be true).
1729 * In the rare case of index wrap-around, 0 will be returned.
1731 pgoff_t
page_cache_next_miss(struct address_space
*mapping
,
1732 pgoff_t index
, unsigned long max_scan
)
1734 XA_STATE(xas
, &mapping
->i_pages
, index
);
1736 while (max_scan
--) {
1737 void *entry
= xas_next(&xas
);
1738 if (!entry
|| xa_is_value(entry
))
1740 if (xas
.xa_index
== 0)
1744 return xas
.xa_index
;
1746 EXPORT_SYMBOL(page_cache_next_miss
);
1749 * page_cache_prev_miss() - Find the previous gap in the page cache.
1750 * @mapping: Mapping.
1752 * @max_scan: Maximum range to search.
1754 * Search the range [max(index - max_scan + 1, 0), index] for the
1755 * gap with the highest index.
1757 * This function may be called under the rcu_read_lock. However, this will
1758 * not atomically search a snapshot of the cache at a single point in time.
1759 * For example, if a gap is created at index 10, then subsequently a gap is
1760 * created at index 5, page_cache_prev_miss() covering both indices may
1761 * return 5 if called under the rcu_read_lock.
1763 * Return: The index of the gap if found, otherwise an index outside the
1764 * range specified (in which case 'index - return >= max_scan' will be true).
1765 * In the rare case of wrap-around, ULONG_MAX will be returned.
1767 pgoff_t
page_cache_prev_miss(struct address_space
*mapping
,
1768 pgoff_t index
, unsigned long max_scan
)
1770 XA_STATE(xas
, &mapping
->i_pages
, index
);
1772 while (max_scan
--) {
1773 void *entry
= xas_prev(&xas
);
1774 if (!entry
|| xa_is_value(entry
))
1776 if (xas
.xa_index
== ULONG_MAX
)
1780 return xas
.xa_index
;
1782 EXPORT_SYMBOL(page_cache_prev_miss
);
1785 * Lockless page cache protocol:
1786 * On the lookup side:
1787 * 1. Load the folio from i_pages
1788 * 2. Increment the refcount if it's not zero
1789 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1791 * On the removal side:
1792 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1793 * B. Remove the page from i_pages
1794 * C. Return the page to the page allocator
1796 * This means that any page may have its reference count temporarily
1797 * increased by a speculative page cache (or fast GUP) lookup as it can
1798 * be allocated by another user before the RCU grace period expires.
1799 * Because the refcount temporarily acquired here may end up being the
1800 * last refcount on the page, any page allocation must be freeable by
1805 * filemap_get_entry - Get a page cache entry.
1806 * @mapping: the address_space to search
1807 * @index: The page cache index.
1809 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1810 * it is returned with an increased refcount. If it is a shadow entry
1811 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1812 * it is returned without further action.
1814 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1816 void *filemap_get_entry(struct address_space
*mapping
, pgoff_t index
)
1818 XA_STATE(xas
, &mapping
->i_pages
, index
);
1819 struct folio
*folio
;
1824 folio
= xas_load(&xas
);
1825 if (xas_retry(&xas
, folio
))
1828 * A shadow entry of a recently evicted page, or a swap entry from
1829 * shmem/tmpfs. Return it without attempting to raise page count.
1831 if (!folio
|| xa_is_value(folio
))
1834 if (!folio_try_get_rcu(folio
))
1837 if (unlikely(folio
!= xas_reload(&xas
))) {
1848 * __filemap_get_folio - Find and get a reference to a folio.
1849 * @mapping: The address_space to search.
1850 * @index: The page index.
1851 * @fgp_flags: %FGP flags modify how the folio is returned.
1852 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1854 * Looks up the page cache entry at @mapping & @index.
1856 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1857 * if the %GFP flags specified for %FGP_CREAT are atomic.
1859 * If this function returns a folio, it is returned with an increased refcount.
1861 * Return: The found folio or an ERR_PTR() otherwise.
1863 struct folio
*__filemap_get_folio(struct address_space
*mapping
, pgoff_t index
,
1864 fgf_t fgp_flags
, gfp_t gfp
)
1866 struct folio
*folio
;
1869 folio
= filemap_get_entry(mapping
, index
);
1870 if (xa_is_value(folio
))
1875 if (fgp_flags
& FGP_LOCK
) {
1876 if (fgp_flags
& FGP_NOWAIT
) {
1877 if (!folio_trylock(folio
)) {
1879 return ERR_PTR(-EAGAIN
);
1885 /* Has the page been truncated? */
1886 if (unlikely(folio
->mapping
!= mapping
)) {
1887 folio_unlock(folio
);
1891 VM_BUG_ON_FOLIO(!folio_contains(folio
, index
), folio
);
1894 if (fgp_flags
& FGP_ACCESSED
)
1895 folio_mark_accessed(folio
);
1896 else if (fgp_flags
& FGP_WRITE
) {
1897 /* Clear idle flag for buffer write */
1898 if (folio_test_idle(folio
))
1899 folio_clear_idle(folio
);
1902 if (fgp_flags
& FGP_STABLE
)
1903 folio_wait_stable(folio
);
1905 if (!folio
&& (fgp_flags
& FGP_CREAT
)) {
1906 unsigned order
= FGF_GET_ORDER(fgp_flags
);
1909 if ((fgp_flags
& FGP_WRITE
) && mapping_can_writeback(mapping
))
1911 if (fgp_flags
& FGP_NOFS
)
1913 if (fgp_flags
& FGP_NOWAIT
) {
1915 gfp
|= GFP_NOWAIT
| __GFP_NOWARN
;
1917 if (WARN_ON_ONCE(!(fgp_flags
& (FGP_LOCK
| FGP_FOR_MMAP
))))
1918 fgp_flags
|= FGP_LOCK
;
1920 if (!mapping_large_folio_support(mapping
))
1922 if (order
> MAX_PAGECACHE_ORDER
)
1923 order
= MAX_PAGECACHE_ORDER
;
1924 /* If we're not aligned, allocate a smaller folio */
1925 if (index
& ((1UL << order
) - 1))
1926 order
= __ffs(index
);
1929 gfp_t alloc_gfp
= gfp
;
1935 alloc_gfp
|= __GFP_NORETRY
| __GFP_NOWARN
;
1936 folio
= filemap_alloc_folio(alloc_gfp
, order
);
1940 /* Init accessed so avoid atomic mark_page_accessed later */
1941 if (fgp_flags
& FGP_ACCESSED
)
1942 __folio_set_referenced(folio
);
1944 err
= filemap_add_folio(mapping
, folio
, index
, gfp
);
1949 } while (order
-- > 0);
1954 return ERR_PTR(err
);
1956 * filemap_add_folio locks the page, and for mmap
1957 * we expect an unlocked page.
1959 if (folio
&& (fgp_flags
& FGP_FOR_MMAP
))
1960 folio_unlock(folio
);
1964 return ERR_PTR(-ENOENT
);
1967 EXPORT_SYMBOL(__filemap_get_folio
);
1969 static inline struct folio
*find_get_entry(struct xa_state
*xas
, pgoff_t max
,
1972 struct folio
*folio
;
1975 if (mark
== XA_PRESENT
)
1976 folio
= xas_find(xas
, max
);
1978 folio
= xas_find_marked(xas
, max
, mark
);
1980 if (xas_retry(xas
, folio
))
1983 * A shadow entry of a recently evicted page, a swap
1984 * entry from shmem/tmpfs or a DAX entry. Return it
1985 * without attempting to raise page count.
1987 if (!folio
|| xa_is_value(folio
))
1990 if (!folio_try_get_rcu(folio
))
1993 if (unlikely(folio
!= xas_reload(xas
))) {
2005 * find_get_entries - gang pagecache lookup
2006 * @mapping: The address_space to search
2007 * @start: The starting page cache index
2008 * @end: The final page index (inclusive).
2009 * @fbatch: Where the resulting entries are placed.
2010 * @indices: The cache indices corresponding to the entries in @entries
2012 * find_get_entries() will search for and return a batch of entries in
2013 * the mapping. The entries are placed in @fbatch. find_get_entries()
2014 * takes a reference on any actual folios it returns.
2016 * The entries have ascending indexes. The indices may not be consecutive
2017 * due to not-present entries or large folios.
2019 * Any shadow entries of evicted folios, or swap entries from
2020 * shmem/tmpfs, are included in the returned array.
2022 * Return: The number of entries which were found.
2024 unsigned find_get_entries(struct address_space
*mapping
, pgoff_t
*start
,
2025 pgoff_t end
, struct folio_batch
*fbatch
, pgoff_t
*indices
)
2027 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2028 struct folio
*folio
;
2031 while ((folio
= find_get_entry(&xas
, end
, XA_PRESENT
)) != NULL
) {
2032 indices
[fbatch
->nr
] = xas
.xa_index
;
2033 if (!folio_batch_add(fbatch
, folio
))
2038 if (folio_batch_count(fbatch
)) {
2039 unsigned long nr
= 1;
2040 int idx
= folio_batch_count(fbatch
) - 1;
2042 folio
= fbatch
->folios
[idx
];
2043 if (!xa_is_value(folio
) && !folio_test_hugetlb(folio
))
2044 nr
= folio_nr_pages(folio
);
2045 *start
= indices
[idx
] + nr
;
2047 return folio_batch_count(fbatch
);
2051 * find_lock_entries - Find a batch of pagecache entries.
2052 * @mapping: The address_space to search.
2053 * @start: The starting page cache index.
2054 * @end: The final page index (inclusive).
2055 * @fbatch: Where the resulting entries are placed.
2056 * @indices: The cache indices of the entries in @fbatch.
2058 * find_lock_entries() will return a batch of entries from @mapping.
2059 * Swap, shadow and DAX entries are included. Folios are returned
2060 * locked and with an incremented refcount. Folios which are locked
2061 * by somebody else or under writeback are skipped. Folios which are
2062 * partially outside the range are not returned.
2064 * The entries have ascending indexes. The indices may not be consecutive
2065 * due to not-present entries, large folios, folios which could not be
2066 * locked or folios under writeback.
2068 * Return: The number of entries which were found.
2070 unsigned find_lock_entries(struct address_space
*mapping
, pgoff_t
*start
,
2071 pgoff_t end
, struct folio_batch
*fbatch
, pgoff_t
*indices
)
2073 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2074 struct folio
*folio
;
2077 while ((folio
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
2078 if (!xa_is_value(folio
)) {
2079 if (folio
->index
< *start
)
2081 if (folio_next_index(folio
) - 1 > end
)
2083 if (!folio_trylock(folio
))
2085 if (folio
->mapping
!= mapping
||
2086 folio_test_writeback(folio
))
2088 VM_BUG_ON_FOLIO(!folio_contains(folio
, xas
.xa_index
),
2091 indices
[fbatch
->nr
] = xas
.xa_index
;
2092 if (!folio_batch_add(fbatch
, folio
))
2096 folio_unlock(folio
);
2102 if (folio_batch_count(fbatch
)) {
2103 unsigned long nr
= 1;
2104 int idx
= folio_batch_count(fbatch
) - 1;
2106 folio
= fbatch
->folios
[idx
];
2107 if (!xa_is_value(folio
) && !folio_test_hugetlb(folio
))
2108 nr
= folio_nr_pages(folio
);
2109 *start
= indices
[idx
] + nr
;
2111 return folio_batch_count(fbatch
);
2115 * filemap_get_folios - Get a batch of folios
2116 * @mapping: The address_space to search
2117 * @start: The starting page index
2118 * @end: The final page index (inclusive)
2119 * @fbatch: The batch to fill.
2121 * Search for and return a batch of folios in the mapping starting at
2122 * index @start and up to index @end (inclusive). The folios are returned
2123 * in @fbatch with an elevated reference count.
2125 * The first folio may start before @start; if it does, it will contain
2126 * @start. The final folio may extend beyond @end; if it does, it will
2127 * contain @end. The folios have ascending indices. There may be gaps
2128 * between the folios if there are indices which have no folio in the
2129 * page cache. If folios are added to or removed from the page cache
2130 * while this is running, they may or may not be found by this call.
2132 * Return: The number of folios which were found.
2133 * We also update @start to index the next folio for the traversal.
2135 unsigned filemap_get_folios(struct address_space
*mapping
, pgoff_t
*start
,
2136 pgoff_t end
, struct folio_batch
*fbatch
)
2138 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2139 struct folio
*folio
;
2142 while ((folio
= find_get_entry(&xas
, end
, XA_PRESENT
)) != NULL
) {
2143 /* Skip over shadow, swap and DAX entries */
2144 if (xa_is_value(folio
))
2146 if (!folio_batch_add(fbatch
, folio
)) {
2147 unsigned long nr
= folio_nr_pages(folio
);
2149 if (folio_test_hugetlb(folio
))
2151 *start
= folio
->index
+ nr
;
2157 * We come here when there is no page beyond @end. We take care to not
2158 * overflow the index @start as it confuses some of the callers. This
2159 * breaks the iteration when there is a page at index -1 but that is
2160 * already broken anyway.
2162 if (end
== (pgoff_t
)-1)
2163 *start
= (pgoff_t
)-1;
2169 return folio_batch_count(fbatch
);
2171 EXPORT_SYMBOL(filemap_get_folios
);
2174 * filemap_get_folios_contig - Get a batch of contiguous folios
2175 * @mapping: The address_space to search
2176 * @start: The starting page index
2177 * @end: The final page index (inclusive)
2178 * @fbatch: The batch to fill
2180 * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2181 * except the returned folios are guaranteed to be contiguous. This may
2182 * not return all contiguous folios if the batch gets filled up.
2184 * Return: The number of folios found.
2185 * Also update @start to be positioned for traversal of the next folio.
2188 unsigned filemap_get_folios_contig(struct address_space
*mapping
,
2189 pgoff_t
*start
, pgoff_t end
, struct folio_batch
*fbatch
)
2191 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2193 struct folio
*folio
;
2197 for (folio
= xas_load(&xas
); folio
&& xas
.xa_index
<= end
;
2198 folio
= xas_next(&xas
)) {
2199 if (xas_retry(&xas
, folio
))
2202 * If the entry has been swapped out, we can stop looking.
2203 * No current caller is looking for DAX entries.
2205 if (xa_is_value(folio
))
2208 if (!folio_try_get_rcu(folio
))
2211 if (unlikely(folio
!= xas_reload(&xas
)))
2214 if (!folio_batch_add(fbatch
, folio
)) {
2215 nr
= folio_nr_pages(folio
);
2217 if (folio_test_hugetlb(folio
))
2219 *start
= folio
->index
+ nr
;
2231 nr
= folio_batch_count(fbatch
);
2234 folio
= fbatch
->folios
[nr
- 1];
2235 if (folio_test_hugetlb(folio
))
2236 *start
= folio
->index
+ 1;
2238 *start
= folio_next_index(folio
);
2242 return folio_batch_count(fbatch
);
2244 EXPORT_SYMBOL(filemap_get_folios_contig
);
2247 * filemap_get_folios_tag - Get a batch of folios matching @tag
2248 * @mapping: The address_space to search
2249 * @start: The starting page index
2250 * @end: The final page index (inclusive)
2251 * @tag: The tag index
2252 * @fbatch: The batch to fill
2254 * Same as filemap_get_folios(), but only returning folios tagged with @tag.
2256 * Return: The number of folios found.
2257 * Also update @start to index the next folio for traversal.
2259 unsigned filemap_get_folios_tag(struct address_space
*mapping
, pgoff_t
*start
,
2260 pgoff_t end
, xa_mark_t tag
, struct folio_batch
*fbatch
)
2262 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2263 struct folio
*folio
;
2266 while ((folio
= find_get_entry(&xas
, end
, tag
)) != NULL
) {
2268 * Shadow entries should never be tagged, but this iteration
2269 * is lockless so there is a window for page reclaim to evict
2270 * a page we saw tagged. Skip over it.
2272 if (xa_is_value(folio
))
2274 if (!folio_batch_add(fbatch
, folio
)) {
2275 unsigned long nr
= folio_nr_pages(folio
);
2277 if (folio_test_hugetlb(folio
))
2279 *start
= folio
->index
+ nr
;
2284 * We come here when there is no page beyond @end. We take care to not
2285 * overflow the index @start as it confuses some of the callers. This
2286 * breaks the iteration when there is a page at index -1 but that is
2287 * already broke anyway.
2289 if (end
== (pgoff_t
)-1)
2290 *start
= (pgoff_t
)-1;
2296 return folio_batch_count(fbatch
);
2298 EXPORT_SYMBOL(filemap_get_folios_tag
);
2301 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2302 * a _large_ part of the i/o request. Imagine the worst scenario:
2304 * ---R__________________________________________B__________
2305 * ^ reading here ^ bad block(assume 4k)
2307 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2308 * => failing the whole request => read(R) => read(R+1) =>
2309 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2310 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2311 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2313 * It is going insane. Fix it by quickly scaling down the readahead size.
2315 static void shrink_readahead_size_eio(struct file_ra_state
*ra
)
2321 * filemap_get_read_batch - Get a batch of folios for read
2323 * Get a batch of folios which represent a contiguous range of bytes in
2324 * the file. No exceptional entries will be returned. If @index is in
2325 * the middle of a folio, the entire folio will be returned. The last
2326 * folio in the batch may have the readahead flag set or the uptodate flag
2327 * clear so that the caller can take the appropriate action.
2329 static void filemap_get_read_batch(struct address_space
*mapping
,
2330 pgoff_t index
, pgoff_t max
, struct folio_batch
*fbatch
)
2332 XA_STATE(xas
, &mapping
->i_pages
, index
);
2333 struct folio
*folio
;
2336 for (folio
= xas_load(&xas
); folio
; folio
= xas_next(&xas
)) {
2337 if (xas_retry(&xas
, folio
))
2339 if (xas
.xa_index
> max
|| xa_is_value(folio
))
2341 if (xa_is_sibling(folio
))
2343 if (!folio_try_get_rcu(folio
))
2346 if (unlikely(folio
!= xas_reload(&xas
)))
2349 if (!folio_batch_add(fbatch
, folio
))
2351 if (!folio_test_uptodate(folio
))
2353 if (folio_test_readahead(folio
))
2355 xas_advance(&xas
, folio_next_index(folio
) - 1);
2365 static int filemap_read_folio(struct file
*file
, filler_t filler
,
2366 struct folio
*folio
)
2368 bool workingset
= folio_test_workingset(folio
);
2369 unsigned long pflags
;
2373 * A previous I/O error may have been due to temporary failures,
2374 * eg. multipath errors. PG_error will be set again if read_folio
2377 folio_clear_error(folio
);
2379 /* Start the actual read. The read will unlock the page. */
2380 if (unlikely(workingset
))
2381 psi_memstall_enter(&pflags
);
2382 error
= filler(file
, folio
);
2383 if (unlikely(workingset
))
2384 psi_memstall_leave(&pflags
);
2388 error
= folio_wait_locked_killable(folio
);
2391 if (folio_test_uptodate(folio
))
2394 shrink_readahead_size_eio(&file
->f_ra
);
2398 static bool filemap_range_uptodate(struct address_space
*mapping
,
2399 loff_t pos
, size_t count
, struct folio
*folio
,
2402 if (folio_test_uptodate(folio
))
2404 /* pipes can't handle partially uptodate pages */
2407 if (!mapping
->a_ops
->is_partially_uptodate
)
2409 if (mapping
->host
->i_blkbits
>= folio_shift(folio
))
2412 if (folio_pos(folio
) > pos
) {
2413 count
-= folio_pos(folio
) - pos
;
2416 pos
-= folio_pos(folio
);
2419 return mapping
->a_ops
->is_partially_uptodate(folio
, pos
, count
);
2422 static int filemap_update_page(struct kiocb
*iocb
,
2423 struct address_space
*mapping
, size_t count
,
2424 struct folio
*folio
, bool need_uptodate
)
2428 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2429 if (!filemap_invalidate_trylock_shared(mapping
))
2432 filemap_invalidate_lock_shared(mapping
);
2435 if (!folio_trylock(folio
)) {
2437 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_NOIO
))
2438 goto unlock_mapping
;
2439 if (!(iocb
->ki_flags
& IOCB_WAITQ
)) {
2440 filemap_invalidate_unlock_shared(mapping
);
2442 * This is where we usually end up waiting for a
2443 * previously submitted readahead to finish.
2445 folio_put_wait_locked(folio
, TASK_KILLABLE
);
2446 return AOP_TRUNCATED_PAGE
;
2448 error
= __folio_lock_async(folio
, iocb
->ki_waitq
);
2450 goto unlock_mapping
;
2453 error
= AOP_TRUNCATED_PAGE
;
2454 if (!folio
->mapping
)
2458 if (filemap_range_uptodate(mapping
, iocb
->ki_pos
, count
, folio
,
2463 if (iocb
->ki_flags
& (IOCB_NOIO
| IOCB_NOWAIT
| IOCB_WAITQ
))
2466 error
= filemap_read_folio(iocb
->ki_filp
, mapping
->a_ops
->read_folio
,
2468 goto unlock_mapping
;
2470 folio_unlock(folio
);
2472 filemap_invalidate_unlock_shared(mapping
);
2473 if (error
== AOP_TRUNCATED_PAGE
)
2478 static int filemap_create_folio(struct file
*file
,
2479 struct address_space
*mapping
, pgoff_t index
,
2480 struct folio_batch
*fbatch
)
2482 struct folio
*folio
;
2485 folio
= filemap_alloc_folio(mapping_gfp_mask(mapping
), 0);
2490 * Protect against truncate / hole punch. Grabbing invalidate_lock
2491 * here assures we cannot instantiate and bring uptodate new
2492 * pagecache folios after evicting page cache during truncate
2493 * and before actually freeing blocks. Note that we could
2494 * release invalidate_lock after inserting the folio into
2495 * the page cache as the locked folio would then be enough to
2496 * synchronize with hole punching. But there are code paths
2497 * such as filemap_update_page() filling in partially uptodate
2498 * pages or ->readahead() that need to hold invalidate_lock
2499 * while mapping blocks for IO so let's hold the lock here as
2500 * well to keep locking rules simple.
2502 filemap_invalidate_lock_shared(mapping
);
2503 error
= filemap_add_folio(mapping
, folio
, index
,
2504 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2505 if (error
== -EEXIST
)
2506 error
= AOP_TRUNCATED_PAGE
;
2510 error
= filemap_read_folio(file
, mapping
->a_ops
->read_folio
, folio
);
2514 filemap_invalidate_unlock_shared(mapping
);
2515 folio_batch_add(fbatch
, folio
);
2518 filemap_invalidate_unlock_shared(mapping
);
2523 static int filemap_readahead(struct kiocb
*iocb
, struct file
*file
,
2524 struct address_space
*mapping
, struct folio
*folio
,
2527 DEFINE_READAHEAD(ractl
, file
, &file
->f_ra
, mapping
, folio
->index
);
2529 if (iocb
->ki_flags
& IOCB_NOIO
)
2531 page_cache_async_ra(&ractl
, folio
, last_index
- folio
->index
);
2535 static int filemap_get_pages(struct kiocb
*iocb
, size_t count
,
2536 struct folio_batch
*fbatch
, bool need_uptodate
)
2538 struct file
*filp
= iocb
->ki_filp
;
2539 struct address_space
*mapping
= filp
->f_mapping
;
2540 struct file_ra_state
*ra
= &filp
->f_ra
;
2541 pgoff_t index
= iocb
->ki_pos
>> PAGE_SHIFT
;
2543 struct folio
*folio
;
2546 /* "last_index" is the index of the page beyond the end of the read */
2547 last_index
= DIV_ROUND_UP(iocb
->ki_pos
+ count
, PAGE_SIZE
);
2549 if (fatal_signal_pending(current
))
2552 filemap_get_read_batch(mapping
, index
, last_index
- 1, fbatch
);
2553 if (!folio_batch_count(fbatch
)) {
2554 if (iocb
->ki_flags
& IOCB_NOIO
)
2556 page_cache_sync_readahead(mapping
, ra
, filp
, index
,
2557 last_index
- index
);
2558 filemap_get_read_batch(mapping
, index
, last_index
- 1, fbatch
);
2560 if (!folio_batch_count(fbatch
)) {
2561 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_WAITQ
))
2563 err
= filemap_create_folio(filp
, mapping
,
2564 iocb
->ki_pos
>> PAGE_SHIFT
, fbatch
);
2565 if (err
== AOP_TRUNCATED_PAGE
)
2570 folio
= fbatch
->folios
[folio_batch_count(fbatch
) - 1];
2571 if (folio_test_readahead(folio
)) {
2572 err
= filemap_readahead(iocb
, filp
, mapping
, folio
, last_index
);
2576 if (!folio_test_uptodate(folio
)) {
2577 if ((iocb
->ki_flags
& IOCB_WAITQ
) &&
2578 folio_batch_count(fbatch
) > 1)
2579 iocb
->ki_flags
|= IOCB_NOWAIT
;
2580 err
= filemap_update_page(iocb
, mapping
, count
, folio
,
2590 if (likely(--fbatch
->nr
))
2592 if (err
== AOP_TRUNCATED_PAGE
)
2597 static inline bool pos_same_folio(loff_t pos1
, loff_t pos2
, struct folio
*folio
)
2599 unsigned int shift
= folio_shift(folio
);
2601 return (pos1
>> shift
== pos2
>> shift
);
2605 * filemap_read - Read data from the page cache.
2606 * @iocb: The iocb to read.
2607 * @iter: Destination for the data.
2608 * @already_read: Number of bytes already read by the caller.
2610 * Copies data from the page cache. If the data is not currently present,
2611 * uses the readahead and read_folio address_space operations to fetch it.
2613 * Return: Total number of bytes copied, including those already read by
2614 * the caller. If an error happens before any bytes are copied, returns
2615 * a negative error number.
2617 ssize_t
filemap_read(struct kiocb
*iocb
, struct iov_iter
*iter
,
2618 ssize_t already_read
)
2620 struct file
*filp
= iocb
->ki_filp
;
2621 struct file_ra_state
*ra
= &filp
->f_ra
;
2622 struct address_space
*mapping
= filp
->f_mapping
;
2623 struct inode
*inode
= mapping
->host
;
2624 struct folio_batch fbatch
;
2626 bool writably_mapped
;
2627 loff_t isize
, end_offset
;
2628 loff_t last_pos
= ra
->prev_pos
;
2630 if (unlikely(iocb
->ki_pos
>= inode
->i_sb
->s_maxbytes
))
2632 if (unlikely(!iov_iter_count(iter
)))
2635 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2636 folio_batch_init(&fbatch
);
2642 * If we've already successfully copied some data, then we
2643 * can no longer safely return -EIOCBQUEUED. Hence mark
2644 * an async read NOWAIT at that point.
2646 if ((iocb
->ki_flags
& IOCB_WAITQ
) && already_read
)
2647 iocb
->ki_flags
|= IOCB_NOWAIT
;
2649 if (unlikely(iocb
->ki_pos
>= i_size_read(inode
)))
2652 error
= filemap_get_pages(iocb
, iter
->count
, &fbatch
, false);
2657 * i_size must be checked after we know the pages are Uptodate.
2659 * Checking i_size after the check allows us to calculate
2660 * the correct value for "nr", which means the zero-filled
2661 * part of the page is not copied back to userspace (unless
2662 * another truncate extends the file - this is desired though).
2664 isize
= i_size_read(inode
);
2665 if (unlikely(iocb
->ki_pos
>= isize
))
2667 end_offset
= min_t(loff_t
, isize
, iocb
->ki_pos
+ iter
->count
);
2670 * Once we start copying data, we don't want to be touching any
2671 * cachelines that might be contended:
2673 writably_mapped
= mapping_writably_mapped(mapping
);
2676 * When a read accesses the same folio several times, only
2677 * mark it as accessed the first time.
2679 if (!pos_same_folio(iocb
->ki_pos
, last_pos
- 1,
2681 folio_mark_accessed(fbatch
.folios
[0]);
2683 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
2684 struct folio
*folio
= fbatch
.folios
[i
];
2685 size_t fsize
= folio_size(folio
);
2686 size_t offset
= iocb
->ki_pos
& (fsize
- 1);
2687 size_t bytes
= min_t(loff_t
, end_offset
- iocb
->ki_pos
,
2691 if (end_offset
< folio_pos(folio
))
2694 folio_mark_accessed(folio
);
2696 * If users can be writing to this folio using arbitrary
2697 * virtual addresses, take care of potential aliasing
2698 * before reading the folio on the kernel side.
2700 if (writably_mapped
)
2701 flush_dcache_folio(folio
);
2703 copied
= copy_folio_to_iter(folio
, offset
, bytes
, iter
);
2705 already_read
+= copied
;
2706 iocb
->ki_pos
+= copied
;
2707 last_pos
= iocb
->ki_pos
;
2709 if (copied
< bytes
) {
2715 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++)
2716 folio_put(fbatch
.folios
[i
]);
2717 folio_batch_init(&fbatch
);
2718 } while (iov_iter_count(iter
) && iocb
->ki_pos
< isize
&& !error
);
2720 file_accessed(filp
);
2721 ra
->prev_pos
= last_pos
;
2722 return already_read
? already_read
: error
;
2724 EXPORT_SYMBOL_GPL(filemap_read
);
2726 int kiocb_write_and_wait(struct kiocb
*iocb
, size_t count
)
2728 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
2729 loff_t pos
= iocb
->ki_pos
;
2730 loff_t end
= pos
+ count
- 1;
2732 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2733 if (filemap_range_needs_writeback(mapping
, pos
, end
))
2738 return filemap_write_and_wait_range(mapping
, pos
, end
);
2741 int kiocb_invalidate_pages(struct kiocb
*iocb
, size_t count
)
2743 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
2744 loff_t pos
= iocb
->ki_pos
;
2745 loff_t end
= pos
+ count
- 1;
2748 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2749 /* we could block if there are any pages in the range */
2750 if (filemap_range_has_page(mapping
, pos
, end
))
2753 ret
= filemap_write_and_wait_range(mapping
, pos
, end
);
2759 * After a write we want buffered reads to be sure to go to disk to get
2760 * the new data. We invalidate clean cached page from the region we're
2761 * about to write. We do this *before* the write so that we can return
2762 * without clobbering -EIOCBQUEUED from ->direct_IO().
2764 return invalidate_inode_pages2_range(mapping
, pos
>> PAGE_SHIFT
,
2769 * generic_file_read_iter - generic filesystem read routine
2770 * @iocb: kernel I/O control block
2771 * @iter: destination for the data read
2773 * This is the "read_iter()" routine for all filesystems
2774 * that can use the page cache directly.
2776 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2777 * be returned when no data can be read without waiting for I/O requests
2778 * to complete; it doesn't prevent readahead.
2780 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2781 * requests shall be made for the read or for readahead. When no data
2782 * can be read, -EAGAIN shall be returned. When readahead would be
2783 * triggered, a partial, possibly empty read shall be returned.
2786 * * number of bytes copied, even for partial reads
2787 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2790 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2792 size_t count
= iov_iter_count(iter
);
2796 return 0; /* skip atime */
2798 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2799 struct file
*file
= iocb
->ki_filp
;
2800 struct address_space
*mapping
= file
->f_mapping
;
2801 struct inode
*inode
= mapping
->host
;
2803 retval
= kiocb_write_and_wait(iocb
, count
);
2806 file_accessed(file
);
2808 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2810 iocb
->ki_pos
+= retval
;
2813 if (retval
!= -EIOCBQUEUED
)
2814 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2817 * Btrfs can have a short DIO read if we encounter
2818 * compressed extents, so if there was an error, or if
2819 * we've already read everything we wanted to, or if
2820 * there was a short read because we hit EOF, go ahead
2821 * and return. Otherwise fallthrough to buffered io for
2822 * the rest of the read. Buffered reads will not work for
2823 * DAX files, so don't bother trying.
2825 if (retval
< 0 || !count
|| IS_DAX(inode
))
2827 if (iocb
->ki_pos
>= i_size_read(inode
))
2831 return filemap_read(iocb
, iter
, retval
);
2833 EXPORT_SYMBOL(generic_file_read_iter
);
2836 * Splice subpages from a folio into a pipe.
2838 size_t splice_folio_into_pipe(struct pipe_inode_info
*pipe
,
2839 struct folio
*folio
, loff_t fpos
, size_t size
)
2842 size_t spliced
= 0, offset
= offset_in_folio(folio
, fpos
);
2844 page
= folio_page(folio
, offset
/ PAGE_SIZE
);
2845 size
= min(size
, folio_size(folio
) - offset
);
2846 offset
%= PAGE_SIZE
;
2848 while (spliced
< size
&&
2849 !pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
)) {
2850 struct pipe_buffer
*buf
= pipe_head_buf(pipe
);
2851 size_t part
= min_t(size_t, PAGE_SIZE
- offset
, size
- spliced
);
2853 *buf
= (struct pipe_buffer
) {
2854 .ops
= &page_cache_pipe_buf_ops
,
2870 * filemap_splice_read - Splice data from a file's pagecache into a pipe
2871 * @in: The file to read from
2872 * @ppos: Pointer to the file position to read from
2873 * @pipe: The pipe to splice into
2874 * @len: The amount to splice
2875 * @flags: The SPLICE_F_* flags
2877 * This function gets folios from a file's pagecache and splices them into the
2878 * pipe. Readahead will be called as necessary to fill more folios. This may
2879 * be used for blockdevs also.
2881 * Return: On success, the number of bytes read will be returned and *@ppos
2882 * will be updated if appropriate; 0 will be returned if there is no more data
2883 * to be read; -EAGAIN will be returned if the pipe had no space, and some
2884 * other negative error code will be returned on error. A short read may occur
2885 * if the pipe has insufficient space, we reach the end of the data or we hit a
2888 ssize_t
filemap_splice_read(struct file
*in
, loff_t
*ppos
,
2889 struct pipe_inode_info
*pipe
,
2890 size_t len
, unsigned int flags
)
2892 struct folio_batch fbatch
;
2894 size_t total_spliced
= 0, used
, npages
;
2895 loff_t isize
, end_offset
;
2896 bool writably_mapped
;
2899 if (unlikely(*ppos
>= in
->f_mapping
->host
->i_sb
->s_maxbytes
))
2902 init_sync_kiocb(&iocb
, in
);
2903 iocb
.ki_pos
= *ppos
;
2905 /* Work out how much data we can actually add into the pipe */
2906 used
= pipe_occupancy(pipe
->head
, pipe
->tail
);
2907 npages
= max_t(ssize_t
, pipe
->max_usage
- used
, 0);
2908 len
= min_t(size_t, len
, npages
* PAGE_SIZE
);
2910 folio_batch_init(&fbatch
);
2915 if (*ppos
>= i_size_read(in
->f_mapping
->host
))
2918 iocb
.ki_pos
= *ppos
;
2919 error
= filemap_get_pages(&iocb
, len
, &fbatch
, true);
2924 * i_size must be checked after we know the pages are Uptodate.
2926 * Checking i_size after the check allows us to calculate
2927 * the correct value for "nr", which means the zero-filled
2928 * part of the page is not copied back to userspace (unless
2929 * another truncate extends the file - this is desired though).
2931 isize
= i_size_read(in
->f_mapping
->host
);
2932 if (unlikely(*ppos
>= isize
))
2934 end_offset
= min_t(loff_t
, isize
, *ppos
+ len
);
2937 * Once we start copying data, we don't want to be touching any
2938 * cachelines that might be contended:
2940 writably_mapped
= mapping_writably_mapped(in
->f_mapping
);
2942 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
2943 struct folio
*folio
= fbatch
.folios
[i
];
2946 if (folio_pos(folio
) >= end_offset
)
2948 folio_mark_accessed(folio
);
2951 * If users can be writing to this folio using arbitrary
2952 * virtual addresses, take care of potential aliasing
2953 * before reading the folio on the kernel side.
2955 if (writably_mapped
)
2956 flush_dcache_folio(folio
);
2958 n
= min_t(loff_t
, len
, isize
- *ppos
);
2959 n
= splice_folio_into_pipe(pipe
, folio
, *ppos
, n
);
2965 in
->f_ra
.prev_pos
= *ppos
;
2966 if (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
2970 folio_batch_release(&fbatch
);
2974 folio_batch_release(&fbatch
);
2977 return total_spliced
? total_spliced
: error
;
2979 EXPORT_SYMBOL(filemap_splice_read
);
2981 static inline loff_t
folio_seek_hole_data(struct xa_state
*xas
,
2982 struct address_space
*mapping
, struct folio
*folio
,
2983 loff_t start
, loff_t end
, bool seek_data
)
2985 const struct address_space_operations
*ops
= mapping
->a_ops
;
2986 size_t offset
, bsz
= i_blocksize(mapping
->host
);
2988 if (xa_is_value(folio
) || folio_test_uptodate(folio
))
2989 return seek_data
? start
: end
;
2990 if (!ops
->is_partially_uptodate
)
2991 return seek_data
? end
: start
;
2996 if (unlikely(folio
->mapping
!= mapping
))
2999 offset
= offset_in_folio(folio
, start
) & ~(bsz
- 1);
3002 if (ops
->is_partially_uptodate(folio
, offset
, bsz
) ==
3005 start
= (start
+ bsz
) & ~(bsz
- 1);
3007 } while (offset
< folio_size(folio
));
3009 folio_unlock(folio
);
3014 static inline size_t seek_folio_size(struct xa_state
*xas
, struct folio
*folio
)
3016 if (xa_is_value(folio
))
3017 return PAGE_SIZE
<< xa_get_order(xas
->xa
, xas
->xa_index
);
3018 return folio_size(folio
);
3022 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3023 * @mapping: Address space to search.
3024 * @start: First byte to consider.
3025 * @end: Limit of search (exclusive).
3026 * @whence: Either SEEK_HOLE or SEEK_DATA.
3028 * If the page cache knows which blocks contain holes and which blocks
3029 * contain data, your filesystem can use this function to implement
3030 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3031 * entirely memory-based such as tmpfs, and filesystems which support
3032 * unwritten extents.
3034 * Return: The requested offset on success, or -ENXIO if @whence specifies
3035 * SEEK_DATA and there is no data after @start. There is an implicit hole
3036 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3037 * and @end contain data.
3039 loff_t
mapping_seek_hole_data(struct address_space
*mapping
, loff_t start
,
3040 loff_t end
, int whence
)
3042 XA_STATE(xas
, &mapping
->i_pages
, start
>> PAGE_SHIFT
);
3043 pgoff_t max
= (end
- 1) >> PAGE_SHIFT
;
3044 bool seek_data
= (whence
== SEEK_DATA
);
3045 struct folio
*folio
;
3051 while ((folio
= find_get_entry(&xas
, max
, XA_PRESENT
))) {
3052 loff_t pos
= (u64
)xas
.xa_index
<< PAGE_SHIFT
;
3061 seek_size
= seek_folio_size(&xas
, folio
);
3062 pos
= round_up((u64
)pos
+ 1, seek_size
);
3063 start
= folio_seek_hole_data(&xas
, mapping
, folio
, start
, pos
,
3069 if (seek_size
> PAGE_SIZE
)
3070 xas_set(&xas
, pos
>> PAGE_SHIFT
);
3071 if (!xa_is_value(folio
))
3078 if (folio
&& !xa_is_value(folio
))
3086 #define MMAP_LOTSAMISS (100)
3088 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3089 * @vmf - the vm_fault for this fault.
3090 * @folio - the folio to lock.
3091 * @fpin - the pointer to the file we may pin (or is already pinned).
3093 * This works similar to lock_folio_or_retry in that it can drop the
3094 * mmap_lock. It differs in that it actually returns the folio locked
3095 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
3096 * to drop the mmap_lock then fpin will point to the pinned file and
3097 * needs to be fput()'ed at a later point.
3099 static int lock_folio_maybe_drop_mmap(struct vm_fault
*vmf
, struct folio
*folio
,
3102 if (folio_trylock(folio
))
3106 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3107 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3108 * is supposed to work. We have way too many special cases..
3110 if (vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)
3113 *fpin
= maybe_unlock_mmap_for_io(vmf
, *fpin
);
3114 if (vmf
->flags
& FAULT_FLAG_KILLABLE
) {
3115 if (__folio_lock_killable(folio
)) {
3117 * We didn't have the right flags to drop the mmap_lock,
3118 * but all fault_handlers only check for fatal signals
3119 * if we return VM_FAULT_RETRY, so we need to drop the
3120 * mmap_lock here and return 0 if we don't have a fpin.
3123 mmap_read_unlock(vmf
->vma
->vm_mm
);
3127 __folio_lock(folio
);
3133 * Synchronous readahead happens when we don't even find a page in the page
3134 * cache at all. We don't want to perform IO under the mmap sem, so if we have
3135 * to drop the mmap sem we return the file that was pinned in order for us to do
3136 * that. If we didn't pin a file then we return NULL. The file that is
3137 * returned needs to be fput()'ed when we're done with it.
3139 static struct file
*do_sync_mmap_readahead(struct vm_fault
*vmf
)
3141 struct file
*file
= vmf
->vma
->vm_file
;
3142 struct file_ra_state
*ra
= &file
->f_ra
;
3143 struct address_space
*mapping
= file
->f_mapping
;
3144 DEFINE_READAHEAD(ractl
, file
, ra
, mapping
, vmf
->pgoff
);
3145 struct file
*fpin
= NULL
;
3146 unsigned long vm_flags
= vmf
->vma
->vm_flags
;
3147 unsigned int mmap_miss
;
3149 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3150 /* Use the readahead code, even if readahead is disabled */
3151 if (vm_flags
& VM_HUGEPAGE
) {
3152 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3153 ractl
._index
&= ~((unsigned long)HPAGE_PMD_NR
- 1);
3154 ra
->size
= HPAGE_PMD_NR
;
3156 * Fetch two PMD folios, so we get the chance to actually
3157 * readahead, unless we've been told not to.
3159 if (!(vm_flags
& VM_RAND_READ
))
3161 ra
->async_size
= HPAGE_PMD_NR
;
3162 page_cache_ra_order(&ractl
, ra
, HPAGE_PMD_ORDER
);
3167 /* If we don't want any read-ahead, don't bother */
3168 if (vm_flags
& VM_RAND_READ
)
3173 if (vm_flags
& VM_SEQ_READ
) {
3174 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3175 page_cache_sync_ra(&ractl
, ra
->ra_pages
);
3179 /* Avoid banging the cache line if not needed */
3180 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
3181 if (mmap_miss
< MMAP_LOTSAMISS
* 10)
3182 WRITE_ONCE(ra
->mmap_miss
, ++mmap_miss
);
3185 * Do we miss much more than hit in this file? If so,
3186 * stop bothering with read-ahead. It will only hurt.
3188 if (mmap_miss
> MMAP_LOTSAMISS
)
3194 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3195 ra
->start
= max_t(long, 0, vmf
->pgoff
- ra
->ra_pages
/ 2);
3196 ra
->size
= ra
->ra_pages
;
3197 ra
->async_size
= ra
->ra_pages
/ 4;
3198 ractl
._index
= ra
->start
;
3199 page_cache_ra_order(&ractl
, ra
, 0);
3204 * Asynchronous readahead happens when we find the page and PG_readahead,
3205 * so we want to possibly extend the readahead further. We return the file that
3206 * was pinned if we have to drop the mmap_lock in order to do IO.
3208 static struct file
*do_async_mmap_readahead(struct vm_fault
*vmf
,
3209 struct folio
*folio
)
3211 struct file
*file
= vmf
->vma
->vm_file
;
3212 struct file_ra_state
*ra
= &file
->f_ra
;
3213 DEFINE_READAHEAD(ractl
, file
, ra
, file
->f_mapping
, vmf
->pgoff
);
3214 struct file
*fpin
= NULL
;
3215 unsigned int mmap_miss
;
3217 /* If we don't want any read-ahead, don't bother */
3218 if (vmf
->vma
->vm_flags
& VM_RAND_READ
|| !ra
->ra_pages
)
3221 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
3223 WRITE_ONCE(ra
->mmap_miss
, --mmap_miss
);
3225 if (folio_test_readahead(folio
)) {
3226 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3227 page_cache_async_ra(&ractl
, folio
, ra
->ra_pages
);
3233 * filemap_fault - read in file data for page fault handling
3234 * @vmf: struct vm_fault containing details of the fault
3236 * filemap_fault() is invoked via the vma operations vector for a
3237 * mapped memory region to read in file data during a page fault.
3239 * The goto's are kind of ugly, but this streamlines the normal case of having
3240 * it in the page cache, and handles the special cases reasonably without
3241 * having a lot of duplicated code.
3243 * vma->vm_mm->mmap_lock must be held on entry.
3245 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3246 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3248 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3249 * has not been released.
3251 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3253 * Return: bitwise-OR of %VM_FAULT_ codes.
3255 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
3258 struct file
*file
= vmf
->vma
->vm_file
;
3259 struct file
*fpin
= NULL
;
3260 struct address_space
*mapping
= file
->f_mapping
;
3261 struct inode
*inode
= mapping
->host
;
3262 pgoff_t max_idx
, index
= vmf
->pgoff
;
3263 struct folio
*folio
;
3265 bool mapping_locked
= false;
3267 max_idx
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
3268 if (unlikely(index
>= max_idx
))
3269 return VM_FAULT_SIGBUS
;
3272 * Do we have something in the page cache already?
3274 folio
= filemap_get_folio(mapping
, index
);
3275 if (likely(!IS_ERR(folio
))) {
3277 * We found the page, so try async readahead before waiting for
3280 if (!(vmf
->flags
& FAULT_FLAG_TRIED
))
3281 fpin
= do_async_mmap_readahead(vmf
, folio
);
3282 if (unlikely(!folio_test_uptodate(folio
))) {
3283 filemap_invalidate_lock_shared(mapping
);
3284 mapping_locked
= true;
3287 /* No page in the page cache at all */
3288 count_vm_event(PGMAJFAULT
);
3289 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
3290 ret
= VM_FAULT_MAJOR
;
3291 fpin
= do_sync_mmap_readahead(vmf
);
3294 * See comment in filemap_create_folio() why we need
3297 if (!mapping_locked
) {
3298 filemap_invalidate_lock_shared(mapping
);
3299 mapping_locked
= true;
3301 folio
= __filemap_get_folio(mapping
, index
,
3302 FGP_CREAT
|FGP_FOR_MMAP
,
3304 if (IS_ERR(folio
)) {
3307 filemap_invalidate_unlock_shared(mapping
);
3308 return VM_FAULT_OOM
;
3312 if (!lock_folio_maybe_drop_mmap(vmf
, folio
, &fpin
))
3315 /* Did it get truncated? */
3316 if (unlikely(folio
->mapping
!= mapping
)) {
3317 folio_unlock(folio
);
3321 VM_BUG_ON_FOLIO(!folio_contains(folio
, index
), folio
);
3324 * We have a locked page in the page cache, now we need to check
3325 * that it's up-to-date. If not, it is going to be due to an error.
3327 if (unlikely(!folio_test_uptodate(folio
))) {
3329 * The page was in cache and uptodate and now it is not.
3330 * Strange but possible since we didn't hold the page lock all
3331 * the time. Let's drop everything get the invalidate lock and
3334 if (!mapping_locked
) {
3335 folio_unlock(folio
);
3339 goto page_not_uptodate
;
3343 * We've made it this far and we had to drop our mmap_lock, now is the
3344 * time to return to the upper layer and have it re-find the vma and
3348 folio_unlock(folio
);
3352 filemap_invalidate_unlock_shared(mapping
);
3355 * Found the page and have a reference on it.
3356 * We must recheck i_size under page lock.
3358 max_idx
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
3359 if (unlikely(index
>= max_idx
)) {
3360 folio_unlock(folio
);
3362 return VM_FAULT_SIGBUS
;
3365 vmf
->page
= folio_file_page(folio
, index
);
3366 return ret
| VM_FAULT_LOCKED
;
3370 * Umm, take care of errors if the page isn't up-to-date.
3371 * Try to re-read it _once_. We do this synchronously,
3372 * because there really aren't any performance issues here
3373 * and we need to check for errors.
3375 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3376 error
= filemap_read_folio(file
, mapping
->a_ops
->read_folio
, folio
);
3381 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
3383 filemap_invalidate_unlock_shared(mapping
);
3385 return VM_FAULT_SIGBUS
;
3389 * We dropped the mmap_lock, we need to return to the fault handler to
3390 * re-find the vma and come back and find our hopefully still populated
3396 filemap_invalidate_unlock_shared(mapping
);
3399 return ret
| VM_FAULT_RETRY
;
3401 EXPORT_SYMBOL(filemap_fault
);
3403 static bool filemap_map_pmd(struct vm_fault
*vmf
, struct folio
*folio
,
3406 struct mm_struct
*mm
= vmf
->vma
->vm_mm
;
3408 /* Huge page is mapped? No need to proceed. */
3409 if (pmd_trans_huge(*vmf
->pmd
)) {
3410 folio_unlock(folio
);
3415 if (pmd_none(*vmf
->pmd
) && folio_test_pmd_mappable(folio
)) {
3416 struct page
*page
= folio_file_page(folio
, start
);
3417 vm_fault_t ret
= do_set_pmd(vmf
, page
);
3419 /* The page is mapped successfully, reference consumed. */
3420 folio_unlock(folio
);
3425 if (pmd_none(*vmf
->pmd
))
3426 pmd_install(mm
, vmf
->pmd
, &vmf
->prealloc_pte
);
3431 static struct folio
*next_uptodate_folio(struct xa_state
*xas
,
3432 struct address_space
*mapping
, pgoff_t end_pgoff
)
3434 struct folio
*folio
= xas_next_entry(xas
, end_pgoff
);
3435 unsigned long max_idx
;
3440 if (xas_retry(xas
, folio
))
3442 if (xa_is_value(folio
))
3444 if (folio_test_locked(folio
))
3446 if (!folio_try_get_rcu(folio
))
3448 /* Has the page moved or been split? */
3449 if (unlikely(folio
!= xas_reload(xas
)))
3451 if (!folio_test_uptodate(folio
) || folio_test_readahead(folio
))
3453 if (!folio_trylock(folio
))
3455 if (folio
->mapping
!= mapping
)
3457 if (!folio_test_uptodate(folio
))
3459 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
3460 if (xas
->xa_index
>= max_idx
)
3464 folio_unlock(folio
);
3467 } while ((folio
= xas_next_entry(xas
, end_pgoff
)) != NULL
);
3473 * Map page range [start_page, start_page + nr_pages) of folio.
3474 * start_page is gotten from start by folio_page(folio, start)
3476 static vm_fault_t
filemap_map_folio_range(struct vm_fault
*vmf
,
3477 struct folio
*folio
, unsigned long start
,
3478 unsigned long addr
, unsigned int nr_pages
,
3479 unsigned int *mmap_miss
)
3482 struct page
*page
= folio_page(folio
, start
);
3483 unsigned int count
= 0;
3484 pte_t
*old_ptep
= vmf
->pte
;
3487 if (PageHWPoison(page
+ count
))
3493 * NOTE: If there're PTE markers, we'll leave them to be
3494 * handled in the specific fault path, and it'll prohibit the
3495 * fault-around logic.
3497 if (!pte_none(vmf
->pte
[count
]))
3504 set_pte_range(vmf
, folio
, page
, count
, addr
);
3505 folio_ref_add(folio
, count
);
3506 if (in_range(vmf
->address
, addr
, count
* PAGE_SIZE
))
3507 ret
= VM_FAULT_NOPAGE
;
3513 addr
+= count
* PAGE_SIZE
;
3515 } while (--nr_pages
> 0);
3518 set_pte_range(vmf
, folio
, page
, count
, addr
);
3519 folio_ref_add(folio
, count
);
3520 if (in_range(vmf
->address
, addr
, count
* PAGE_SIZE
))
3521 ret
= VM_FAULT_NOPAGE
;
3524 vmf
->pte
= old_ptep
;
3529 static vm_fault_t
filemap_map_order0_folio(struct vm_fault
*vmf
,
3530 struct folio
*folio
, unsigned long addr
,
3531 unsigned int *mmap_miss
)
3534 struct page
*page
= &folio
->page
;
3536 if (PageHWPoison(page
))
3542 * NOTE: If there're PTE markers, we'll leave them to be
3543 * handled in the specific fault path, and it'll prohibit
3544 * the fault-around logic.
3546 if (!pte_none(ptep_get(vmf
->pte
)))
3549 if (vmf
->address
== addr
)
3550 ret
= VM_FAULT_NOPAGE
;
3552 set_pte_range(vmf
, folio
, page
, 1, addr
);
3553 folio_ref_inc(folio
);
3558 vm_fault_t
filemap_map_pages(struct vm_fault
*vmf
,
3559 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
3561 struct vm_area_struct
*vma
= vmf
->vma
;
3562 struct file
*file
= vma
->vm_file
;
3563 struct address_space
*mapping
= file
->f_mapping
;
3564 pgoff_t last_pgoff
= start_pgoff
;
3566 XA_STATE(xas
, &mapping
->i_pages
, start_pgoff
);
3567 struct folio
*folio
;
3569 unsigned int nr_pages
= 0, mmap_miss
= 0, mmap_miss_saved
;
3572 folio
= next_uptodate_folio(&xas
, mapping
, end_pgoff
);
3576 if (filemap_map_pmd(vmf
, folio
, start_pgoff
)) {
3577 ret
= VM_FAULT_NOPAGE
;
3581 addr
= vma
->vm_start
+ ((start_pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
3582 vmf
->pte
= pte_offset_map_lock(vma
->vm_mm
, vmf
->pmd
, addr
, &vmf
->ptl
);
3584 folio_unlock(folio
);
3591 addr
+= (xas
.xa_index
- last_pgoff
) << PAGE_SHIFT
;
3592 vmf
->pte
+= xas
.xa_index
- last_pgoff
;
3593 last_pgoff
= xas
.xa_index
;
3594 end
= folio
->index
+ folio_nr_pages(folio
) - 1;
3595 nr_pages
= min(end
, end_pgoff
) - xas
.xa_index
+ 1;
3597 if (!folio_test_large(folio
))
3598 ret
|= filemap_map_order0_folio(vmf
,
3599 folio
, addr
, &mmap_miss
);
3601 ret
|= filemap_map_folio_range(vmf
, folio
,
3602 xas
.xa_index
- folio
->index
, addr
,
3603 nr_pages
, &mmap_miss
);
3605 folio_unlock(folio
);
3607 } while ((folio
= next_uptodate_folio(&xas
, mapping
, end_pgoff
)) != NULL
);
3608 pte_unmap_unlock(vmf
->pte
, vmf
->ptl
);
3612 mmap_miss_saved
= READ_ONCE(file
->f_ra
.mmap_miss
);
3613 if (mmap_miss
>= mmap_miss_saved
)
3614 WRITE_ONCE(file
->f_ra
.mmap_miss
, 0);
3616 WRITE_ONCE(file
->f_ra
.mmap_miss
, mmap_miss_saved
- mmap_miss
);
3620 EXPORT_SYMBOL(filemap_map_pages
);
3622 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3624 struct address_space
*mapping
= vmf
->vma
->vm_file
->f_mapping
;
3625 struct folio
*folio
= page_folio(vmf
->page
);
3626 vm_fault_t ret
= VM_FAULT_LOCKED
;
3628 sb_start_pagefault(mapping
->host
->i_sb
);
3629 file_update_time(vmf
->vma
->vm_file
);
3631 if (folio
->mapping
!= mapping
) {
3632 folio_unlock(folio
);
3633 ret
= VM_FAULT_NOPAGE
;
3637 * We mark the folio dirty already here so that when freeze is in
3638 * progress, we are guaranteed that writeback during freezing will
3639 * see the dirty folio and writeprotect it again.
3641 folio_mark_dirty(folio
);
3642 folio_wait_stable(folio
);
3644 sb_end_pagefault(mapping
->host
->i_sb
);
3648 const struct vm_operations_struct generic_file_vm_ops
= {
3649 .fault
= filemap_fault
,
3650 .map_pages
= filemap_map_pages
,
3651 .page_mkwrite
= filemap_page_mkwrite
,
3654 /* This is used for a general mmap of a disk file */
3656 int generic_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3658 struct address_space
*mapping
= file
->f_mapping
;
3660 if (!mapping
->a_ops
->read_folio
)
3662 file_accessed(file
);
3663 vma
->vm_ops
= &generic_file_vm_ops
;
3668 * This is for filesystems which do not implement ->writepage.
3670 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3672 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
3674 return generic_file_mmap(file
, vma
);
3677 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3679 return VM_FAULT_SIGBUS
;
3681 int generic_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3685 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3689 #endif /* CONFIG_MMU */
3691 EXPORT_SYMBOL(filemap_page_mkwrite
);
3692 EXPORT_SYMBOL(generic_file_mmap
);
3693 EXPORT_SYMBOL(generic_file_readonly_mmap
);
3695 static struct folio
*do_read_cache_folio(struct address_space
*mapping
,
3696 pgoff_t index
, filler_t filler
, struct file
*file
, gfp_t gfp
)
3698 struct folio
*folio
;
3702 filler
= mapping
->a_ops
->read_folio
;
3704 folio
= filemap_get_folio(mapping
, index
);
3705 if (IS_ERR(folio
)) {
3706 folio
= filemap_alloc_folio(gfp
, 0);
3708 return ERR_PTR(-ENOMEM
);
3709 err
= filemap_add_folio(mapping
, folio
, index
, gfp
);
3710 if (unlikely(err
)) {
3714 /* Presumably ENOMEM for xarray node */
3715 return ERR_PTR(err
);
3720 if (folio_test_uptodate(folio
))
3723 if (!folio_trylock(folio
)) {
3724 folio_put_wait_locked(folio
, TASK_UNINTERRUPTIBLE
);
3728 /* Folio was truncated from mapping */
3729 if (!folio
->mapping
) {
3730 folio_unlock(folio
);
3735 /* Someone else locked and filled the page in a very small window */
3736 if (folio_test_uptodate(folio
)) {
3737 folio_unlock(folio
);
3742 err
= filemap_read_folio(file
, filler
, folio
);
3745 if (err
== AOP_TRUNCATED_PAGE
)
3747 return ERR_PTR(err
);
3751 folio_mark_accessed(folio
);
3756 * read_cache_folio - Read into page cache, fill it if needed.
3757 * @mapping: The address_space to read from.
3758 * @index: The index to read.
3759 * @filler: Function to perform the read, or NULL to use aops->read_folio().
3760 * @file: Passed to filler function, may be NULL if not required.
3762 * Read one page into the page cache. If it succeeds, the folio returned
3763 * will contain @index, but it may not be the first page of the folio.
3765 * If the filler function returns an error, it will be returned to the
3768 * Context: May sleep. Expects mapping->invalidate_lock to be held.
3769 * Return: An uptodate folio on success, ERR_PTR() on failure.
3771 struct folio
*read_cache_folio(struct address_space
*mapping
, pgoff_t index
,
3772 filler_t filler
, struct file
*file
)
3774 return do_read_cache_folio(mapping
, index
, filler
, file
,
3775 mapping_gfp_mask(mapping
));
3777 EXPORT_SYMBOL(read_cache_folio
);
3780 * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3781 * @mapping: The address_space for the folio.
3782 * @index: The index that the allocated folio will contain.
3783 * @gfp: The page allocator flags to use if allocating.
3785 * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3786 * any new memory allocations done using the specified allocation flags.
3788 * The most likely error from this function is EIO, but ENOMEM is
3789 * possible and so is EINTR. If ->read_folio returns another error,
3790 * that will be returned to the caller.
3792 * The function expects mapping->invalidate_lock to be already held.
3794 * Return: Uptodate folio on success, ERR_PTR() on failure.
3796 struct folio
*mapping_read_folio_gfp(struct address_space
*mapping
,
3797 pgoff_t index
, gfp_t gfp
)
3799 return do_read_cache_folio(mapping
, index
, NULL
, NULL
, gfp
);
3801 EXPORT_SYMBOL(mapping_read_folio_gfp
);
3803 static struct page
*do_read_cache_page(struct address_space
*mapping
,
3804 pgoff_t index
, filler_t
*filler
, struct file
*file
, gfp_t gfp
)
3806 struct folio
*folio
;
3808 folio
= do_read_cache_folio(mapping
, index
, filler
, file
, gfp
);
3810 return &folio
->page
;
3811 return folio_file_page(folio
, index
);
3814 struct page
*read_cache_page(struct address_space
*mapping
,
3815 pgoff_t index
, filler_t
*filler
, struct file
*file
)
3817 return do_read_cache_page(mapping
, index
, filler
, file
,
3818 mapping_gfp_mask(mapping
));
3820 EXPORT_SYMBOL(read_cache_page
);
3823 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3824 * @mapping: the page's address_space
3825 * @index: the page index
3826 * @gfp: the page allocator flags to use if allocating
3828 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3829 * any new page allocations done using the specified allocation flags.
3831 * If the page does not get brought uptodate, return -EIO.
3833 * The function expects mapping->invalidate_lock to be already held.
3835 * Return: up to date page on success, ERR_PTR() on failure.
3837 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
3841 return do_read_cache_page(mapping
, index
, NULL
, NULL
, gfp
);
3843 EXPORT_SYMBOL(read_cache_page_gfp
);
3846 * Warn about a page cache invalidation failure during a direct I/O write.
3848 static void dio_warn_stale_pagecache(struct file
*filp
)
3850 static DEFINE_RATELIMIT_STATE(_rs
, 86400 * HZ
, DEFAULT_RATELIMIT_BURST
);
3854 errseq_set(&filp
->f_mapping
->wb_err
, -EIO
);
3855 if (__ratelimit(&_rs
)) {
3856 path
= file_path(filp
, pathname
, sizeof(pathname
));
3859 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3860 pr_crit("File: %s PID: %d Comm: %.20s\n", path
, current
->pid
,
3865 void kiocb_invalidate_post_direct_write(struct kiocb
*iocb
, size_t count
)
3867 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3869 if (mapping
->nrpages
&&
3870 invalidate_inode_pages2_range(mapping
,
3871 iocb
->ki_pos
>> PAGE_SHIFT
,
3872 (iocb
->ki_pos
+ count
- 1) >> PAGE_SHIFT
))
3873 dio_warn_stale_pagecache(iocb
->ki_filp
);
3877 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3879 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3880 size_t write_len
= iov_iter_count(from
);
3884 * If a page can not be invalidated, return 0 to fall back
3885 * to buffered write.
3887 written
= kiocb_invalidate_pages(iocb
, write_len
);
3889 if (written
== -EBUSY
)
3894 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3897 * Finally, try again to invalidate clean pages which might have been
3898 * cached by non-direct readahead, or faulted in by get_user_pages()
3899 * if the source of the write was an mmap'ed region of the file
3900 * we're writing. Either one is a pretty crazy thing to do,
3901 * so we don't support it 100%. If this invalidation
3902 * fails, tough, the write still worked...
3904 * Most of the time we do not need this since dio_complete() will do
3905 * the invalidation for us. However there are some file systems that
3906 * do not end up with dio_complete() being called, so let's not break
3907 * them by removing it completely.
3909 * Noticeable example is a blkdev_direct_IO().
3911 * Skip invalidation for async writes or if mapping has no pages.
3914 struct inode
*inode
= mapping
->host
;
3915 loff_t pos
= iocb
->ki_pos
;
3917 kiocb_invalidate_post_direct_write(iocb
, written
);
3919 write_len
-= written
;
3920 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3921 i_size_write(inode
, pos
);
3922 mark_inode_dirty(inode
);
3926 if (written
!= -EIOCBQUEUED
)
3927 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3930 EXPORT_SYMBOL(generic_file_direct_write
);
3932 ssize_t
generic_perform_write(struct kiocb
*iocb
, struct iov_iter
*i
)
3934 struct file
*file
= iocb
->ki_filp
;
3935 loff_t pos
= iocb
->ki_pos
;
3936 struct address_space
*mapping
= file
->f_mapping
;
3937 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3939 ssize_t written
= 0;
3943 unsigned long offset
; /* Offset into pagecache page */
3944 unsigned long bytes
; /* Bytes to write to page */
3945 size_t copied
; /* Bytes copied from user */
3946 void *fsdata
= NULL
;
3948 offset
= (pos
& (PAGE_SIZE
- 1));
3949 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3954 * Bring in the user page that we will copy from _first_.
3955 * Otherwise there's a nasty deadlock on copying from the
3956 * same page as we're writing to, without it being marked
3959 if (unlikely(fault_in_iov_iter_readable(i
, bytes
) == bytes
)) {
3964 if (fatal_signal_pending(current
)) {
3969 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
,
3971 if (unlikely(status
< 0))
3974 if (mapping_writably_mapped(mapping
))
3975 flush_dcache_page(page
);
3977 copied
= copy_page_from_iter_atomic(page
, offset
, bytes
, i
);
3978 flush_dcache_page(page
);
3980 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3982 if (unlikely(status
!= copied
)) {
3983 iov_iter_revert(i
, copied
- max(status
, 0L));
3984 if (unlikely(status
< 0))
3989 if (unlikely(status
== 0)) {
3991 * A short copy made ->write_end() reject the
3992 * thing entirely. Might be memory poisoning
3993 * halfway through, might be a race with munmap,
3994 * might be severe memory pressure.
4003 balance_dirty_pages_ratelimited(mapping
);
4004 } while (iov_iter_count(i
));
4008 iocb
->ki_pos
+= written
;
4011 EXPORT_SYMBOL(generic_perform_write
);
4014 * __generic_file_write_iter - write data to a file
4015 * @iocb: IO state structure (file, offset, etc.)
4016 * @from: iov_iter with data to write
4018 * This function does all the work needed for actually writing data to a
4019 * file. It does all basic checks, removes SUID from the file, updates
4020 * modification times and calls proper subroutines depending on whether we
4021 * do direct IO or a standard buffered write.
4023 * It expects i_rwsem to be grabbed unless we work on a block device or similar
4024 * object which does not need locking at all.
4026 * This function does *not* take care of syncing data in case of O_SYNC write.
4027 * A caller has to handle it. This is mainly due to the fact that we want to
4028 * avoid syncing under i_rwsem.
4031 * * number of bytes written, even for truncated writes
4032 * * negative error code if no data has been written at all
4034 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
4036 struct file
*file
= iocb
->ki_filp
;
4037 struct address_space
*mapping
= file
->f_mapping
;
4038 struct inode
*inode
= mapping
->host
;
4041 ret
= file_remove_privs(file
);
4045 ret
= file_update_time(file
);
4049 if (iocb
->ki_flags
& IOCB_DIRECT
) {
4050 ret
= generic_file_direct_write(iocb
, from
);
4052 * If the write stopped short of completing, fall back to
4053 * buffered writes. Some filesystems do this for writes to
4054 * holes, for example. For DAX files, a buffered write will
4055 * not succeed (even if it did, DAX does not handle dirty
4056 * page-cache pages correctly).
4058 if (ret
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
4060 return direct_write_fallback(iocb
, from
, ret
,
4061 generic_perform_write(iocb
, from
));
4064 return generic_perform_write(iocb
, from
);
4066 EXPORT_SYMBOL(__generic_file_write_iter
);
4069 * generic_file_write_iter - write data to a file
4070 * @iocb: IO state structure
4071 * @from: iov_iter with data to write
4073 * This is a wrapper around __generic_file_write_iter() to be used by most
4074 * filesystems. It takes care of syncing the file in case of O_SYNC file
4075 * and acquires i_rwsem as needed.
4077 * * negative error code if no data has been written at all of
4078 * vfs_fsync_range() failed for a synchronous write
4079 * * number of bytes written, even for truncated writes
4081 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
4083 struct file
*file
= iocb
->ki_filp
;
4084 struct inode
*inode
= file
->f_mapping
->host
;
4088 ret
= generic_write_checks(iocb
, from
);
4090 ret
= __generic_file_write_iter(iocb
, from
);
4091 inode_unlock(inode
);
4094 ret
= generic_write_sync(iocb
, ret
);
4097 EXPORT_SYMBOL(generic_file_write_iter
);
4100 * filemap_release_folio() - Release fs-specific metadata on a folio.
4101 * @folio: The folio which the kernel is trying to free.
4102 * @gfp: Memory allocation flags (and I/O mode).
4104 * The address_space is trying to release any data attached to a folio
4105 * (presumably at folio->private).
4107 * This will also be called if the private_2 flag is set on a page,
4108 * indicating that the folio has other metadata associated with it.
4110 * The @gfp argument specifies whether I/O may be performed to release
4111 * this page (__GFP_IO), and whether the call may block
4112 * (__GFP_RECLAIM & __GFP_FS).
4114 * Return: %true if the release was successful, otherwise %false.
4116 bool filemap_release_folio(struct folio
*folio
, gfp_t gfp
)
4118 struct address_space
* const mapping
= folio
->mapping
;
4120 BUG_ON(!folio_test_locked(folio
));
4121 if (!folio_needs_release(folio
))
4123 if (folio_test_writeback(folio
))
4126 if (mapping
&& mapping
->a_ops
->release_folio
)
4127 return mapping
->a_ops
->release_folio(folio
, gfp
);
4128 return try_to_free_buffers(folio
);
4130 EXPORT_SYMBOL(filemap_release_folio
);
4132 #ifdef CONFIG_CACHESTAT_SYSCALL
4134 * filemap_cachestat() - compute the page cache statistics of a mapping
4135 * @mapping: The mapping to compute the statistics for.
4136 * @first_index: The starting page cache index.
4137 * @last_index: The final page index (inclusive).
4138 * @cs: the cachestat struct to write the result to.
4140 * This will query the page cache statistics of a mapping in the
4141 * page range of [first_index, last_index] (inclusive). The statistics
4142 * queried include: number of dirty pages, number of pages marked for
4143 * writeback, and the number of (recently) evicted pages.
4145 static void filemap_cachestat(struct address_space
*mapping
,
4146 pgoff_t first_index
, pgoff_t last_index
, struct cachestat
*cs
)
4148 XA_STATE(xas
, &mapping
->i_pages
, first_index
);
4149 struct folio
*folio
;
4152 xas_for_each(&xas
, folio
, last_index
) {
4153 unsigned long nr_pages
;
4154 pgoff_t folio_first_index
, folio_last_index
;
4156 if (xas_retry(&xas
, folio
))
4159 if (xa_is_value(folio
)) {
4160 /* page is evicted */
4161 void *shadow
= (void *)folio
;
4162 bool workingset
; /* not used */
4163 int order
= xa_get_order(xas
.xa
, xas
.xa_index
);
4165 nr_pages
= 1 << order
;
4166 folio_first_index
= round_down(xas
.xa_index
, 1 << order
);
4167 folio_last_index
= folio_first_index
+ nr_pages
- 1;
4169 /* Folios might straddle the range boundaries, only count covered pages */
4170 if (folio_first_index
< first_index
)
4171 nr_pages
-= first_index
- folio_first_index
;
4173 if (folio_last_index
> last_index
)
4174 nr_pages
-= folio_last_index
- last_index
;
4176 cs
->nr_evicted
+= nr_pages
;
4178 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4179 if (shmem_mapping(mapping
)) {
4180 /* shmem file - in swap cache */
4181 swp_entry_t swp
= radix_to_swp_entry(folio
);
4183 shadow
= get_shadow_from_swap_cache(swp
);
4186 if (workingset_test_recent(shadow
, true, &workingset
))
4187 cs
->nr_recently_evicted
+= nr_pages
;
4192 nr_pages
= folio_nr_pages(folio
);
4193 folio_first_index
= folio_pgoff(folio
);
4194 folio_last_index
= folio_first_index
+ nr_pages
- 1;
4196 /* Folios might straddle the range boundaries, only count covered pages */
4197 if (folio_first_index
< first_index
)
4198 nr_pages
-= first_index
- folio_first_index
;
4200 if (folio_last_index
> last_index
)
4201 nr_pages
-= folio_last_index
- last_index
;
4203 /* page is in cache */
4204 cs
->nr_cache
+= nr_pages
;
4206 if (folio_test_dirty(folio
))
4207 cs
->nr_dirty
+= nr_pages
;
4209 if (folio_test_writeback(folio
))
4210 cs
->nr_writeback
+= nr_pages
;
4213 if (need_resched()) {
4222 * The cachestat(2) system call.
4224 * cachestat() returns the page cache statistics of a file in the
4225 * bytes range specified by `off` and `len`: number of cached pages,
4226 * number of dirty pages, number of pages marked for writeback,
4227 * number of evicted pages, and number of recently evicted pages.
4229 * An evicted page is a page that is previously in the page cache
4230 * but has been evicted since. A page is recently evicted if its last
4231 * eviction was recent enough that its reentry to the cache would
4232 * indicate that it is actively being used by the system, and that
4233 * there is memory pressure on the system.
4235 * `off` and `len` must be non-negative integers. If `len` > 0,
4236 * the queried range is [`off`, `off` + `len`]. If `len` == 0,
4237 * we will query in the range from `off` to the end of the file.
4239 * The `flags` argument is unused for now, but is included for future
4240 * extensibility. User should pass 0 (i.e no flag specified).
4242 * Currently, hugetlbfs is not supported.
4244 * Because the status of a page can change after cachestat() checks it
4245 * but before it returns to the application, the returned values may
4246 * contain stale information.
4250 * -EFAULT - cstat or cstat_range points to an illegal address
4251 * -EINVAL - invalid flags
4252 * -EBADF - invalid file descriptor
4253 * -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4255 SYSCALL_DEFINE4(cachestat
, unsigned int, fd
,
4256 struct cachestat_range __user
*, cstat_range
,
4257 struct cachestat __user
*, cstat
, unsigned int, flags
)
4259 struct fd f
= fdget(fd
);
4260 struct address_space
*mapping
;
4261 struct cachestat_range csr
;
4262 struct cachestat cs
;
4263 pgoff_t first_index
, last_index
;
4268 if (copy_from_user(&csr
, cstat_range
,
4269 sizeof(struct cachestat_range
))) {
4274 /* hugetlbfs is not supported */
4275 if (is_file_hugepages(f
.file
)) {
4285 first_index
= csr
.off
>> PAGE_SHIFT
;
4287 csr
.len
== 0 ? ULONG_MAX
: (csr
.off
+ csr
.len
- 1) >> PAGE_SHIFT
;
4288 memset(&cs
, 0, sizeof(struct cachestat
));
4289 mapping
= f
.file
->f_mapping
;
4290 filemap_cachestat(mapping
, first_index
, last_index
, &cs
);
4293 if (copy_to_user(cstat
, &cs
, sizeof(struct cachestat
)))
4298 #endif /* CONFIG_CACHESTAT_SYSCALL */