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/mman.h>
26 #include <linux/pagemap.h>
27 #include <linux/file.h>
28 #include <linux/uio.h>
29 #include <linux/error-injection.h>
30 #include <linux/hash.h>
31 #include <linux/writeback.h>
32 #include <linux/backing-dev.h>
33 #include <linux/pagevec.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/shmem_fs.h>
39 #include <linux/rmap.h>
40 #include <linux/delayacct.h>
41 #include <linux/psi.h>
42 #include <linux/ramfs.h>
43 #include <linux/page_idle.h>
44 #include <linux/migrate.h>
45 #include <linux/pipe_fs_i.h>
46 #include <linux/splice.h>
47 #include <asm/pgalloc.h>
48 #include <asm/tlbflush.h>
51 #define CREATE_TRACE_POINTS
52 #include <trace/events/filemap.h>
55 * FIXME: remove all knowledge of the buffer layer from the core VM
57 #include <linux/buffer_head.h> /* for try_to_free_buffers */
62 * Shared mappings implemented 30.11.1994. It's not fully working yet,
65 * Shared mappings now work. 15.8.1995 Bruno.
67 * finished 'unifying' the page and buffer cache and SMP-threaded the
68 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
70 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
76 * ->i_mmap_rwsem (truncate_pagecache)
77 * ->private_lock (__free_pte->block_dirty_folio)
78 * ->swap_lock (exclusive_swap_page, others)
82 * ->invalidate_lock (acquired by fs in truncate path)
83 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
87 * ->page_table_lock or pte_lock (various, mainly in memory.c)
88 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
91 * ->invalidate_lock (filemap_fault)
92 * ->lock_page (filemap_fault, access_process_vm)
94 * ->i_rwsem (generic_perform_write)
95 * ->mmap_lock (fault_in_readable->do_page_fault)
98 * sb_lock (fs/fs-writeback.c)
99 * ->i_pages lock (__sync_single_inode)
102 * ->anon_vma.lock (vma_merge)
105 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
107 * ->page_table_lock or pte_lock
108 * ->swap_lock (try_to_unmap_one)
109 * ->private_lock (try_to_unmap_one)
110 * ->i_pages lock (try_to_unmap_one)
111 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
112 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
113 * ->private_lock (page_remove_rmap->set_page_dirty)
114 * ->i_pages lock (page_remove_rmap->set_page_dirty)
115 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
116 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
117 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
118 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
119 * ->inode->i_lock (zap_pte_range->set_page_dirty)
120 * ->private_lock (zap_pte_range->block_dirty_folio)
123 * ->tasklist_lock (memory_failure, collect_procs_ao)
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 * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1363 * for pte entries, pass NULL for pmd entries.
1364 * @ptl: already locked ptl. This function will drop the lock.
1366 * Wait for a migration entry referencing the given page to be removed. This is
1367 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1368 * this can be called without taking a reference on the page. Instead this
1369 * should be called while holding the ptl for the migration entry referencing
1372 * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1374 * This follows the same logic as folio_wait_bit_common() so see the comments
1377 void migration_entry_wait_on_locked(swp_entry_t entry
, pte_t
*ptep
,
1380 struct wait_page_queue wait_page
;
1381 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1382 bool thrashing
= false;
1383 unsigned long pflags
;
1385 wait_queue_head_t
*q
;
1386 struct folio
*folio
= page_folio(pfn_swap_entry_to_page(entry
));
1388 q
= folio_waitqueue(folio
);
1389 if (!folio_test_uptodate(folio
) && folio_test_workingset(folio
)) {
1390 delayacct_thrashing_start(&in_thrashing
);
1391 psi_memstall_enter(&pflags
);
1396 wait
->func
= wake_page_function
;
1397 wait_page
.folio
= folio
;
1398 wait_page
.bit_nr
= PG_locked
;
1401 spin_lock_irq(&q
->lock
);
1402 folio_set_waiters(folio
);
1403 if (!folio_trylock_flag(folio
, PG_locked
, wait
))
1404 __add_wait_queue_entry_tail(q
, wait
);
1405 spin_unlock_irq(&q
->lock
);
1408 * If a migration entry exists for the page the migration path must hold
1409 * a valid reference to the page, and it must take the ptl to remove the
1410 * migration entry. So the page is valid until the ptl is dropped.
1413 pte_unmap_unlock(ptep
, ptl
);
1420 set_current_state(TASK_UNINTERRUPTIBLE
);
1422 /* Loop until we've been woken or interrupted */
1423 flags
= smp_load_acquire(&wait
->flags
);
1424 if (!(flags
& WQ_FLAG_WOKEN
)) {
1425 if (signal_pending_state(TASK_UNINTERRUPTIBLE
, current
))
1434 finish_wait(q
, wait
);
1437 delayacct_thrashing_end(&in_thrashing
);
1438 psi_memstall_leave(&pflags
);
1443 void folio_wait_bit(struct folio
*folio
, int bit_nr
)
1445 folio_wait_bit_common(folio
, bit_nr
, TASK_UNINTERRUPTIBLE
, SHARED
);
1447 EXPORT_SYMBOL(folio_wait_bit
);
1449 int folio_wait_bit_killable(struct folio
*folio
, int bit_nr
)
1451 return folio_wait_bit_common(folio
, bit_nr
, TASK_KILLABLE
, SHARED
);
1453 EXPORT_SYMBOL(folio_wait_bit_killable
);
1456 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1457 * @folio: The folio to wait for.
1458 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1460 * The caller should hold a reference on @folio. They expect the page to
1461 * become unlocked relatively soon, but do not wish to hold up migration
1462 * (for example) by holding the reference while waiting for the folio to
1463 * come unlocked. After this function returns, the caller should not
1464 * dereference @folio.
1466 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1468 static int folio_put_wait_locked(struct folio
*folio
, int state
)
1470 return folio_wait_bit_common(folio
, PG_locked
, state
, DROP
);
1474 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1475 * @folio: Folio defining the wait queue of interest
1476 * @waiter: Waiter to add to the queue
1478 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1480 void folio_add_wait_queue(struct folio
*folio
, wait_queue_entry_t
*waiter
)
1482 wait_queue_head_t
*q
= folio_waitqueue(folio
);
1483 unsigned long flags
;
1485 spin_lock_irqsave(&q
->lock
, flags
);
1486 __add_wait_queue_entry_tail(q
, waiter
);
1487 folio_set_waiters(folio
);
1488 spin_unlock_irqrestore(&q
->lock
, flags
);
1490 EXPORT_SYMBOL_GPL(folio_add_wait_queue
);
1492 #ifndef clear_bit_unlock_is_negative_byte
1495 * PG_waiters is the high bit in the same byte as PG_lock.
1497 * On x86 (and on many other architectures), we can clear PG_lock and
1498 * test the sign bit at the same time. But if the architecture does
1499 * not support that special operation, we just do this all by hand
1502 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1503 * being cleared, but a memory barrier should be unnecessary since it is
1504 * in the same byte as PG_locked.
1506 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1508 clear_bit_unlock(nr
, mem
);
1509 /* smp_mb__after_atomic(); */
1510 return test_bit(PG_waiters
, mem
);
1516 * folio_unlock - Unlock a locked folio.
1517 * @folio: The folio.
1519 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1521 * Context: May be called from interrupt or process context. May not be
1522 * called from NMI context.
1524 void folio_unlock(struct folio
*folio
)
1526 /* Bit 7 allows x86 to check the byte's sign bit */
1527 BUILD_BUG_ON(PG_waiters
!= 7);
1528 BUILD_BUG_ON(PG_locked
> 7);
1529 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
1530 if (clear_bit_unlock_is_negative_byte(PG_locked
, folio_flags(folio
, 0)))
1531 folio_wake_bit(folio
, PG_locked
);
1533 EXPORT_SYMBOL(folio_unlock
);
1536 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1537 * @folio: The folio.
1539 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1540 * it. The folio reference held for PG_private_2 being set is released.
1542 * This is, for example, used when a netfs folio is being written to a local
1543 * disk cache, thereby allowing writes to the cache for the same folio to be
1546 void folio_end_private_2(struct folio
*folio
)
1548 VM_BUG_ON_FOLIO(!folio_test_private_2(folio
), folio
);
1549 clear_bit_unlock(PG_private_2
, folio_flags(folio
, 0));
1550 folio_wake_bit(folio
, PG_private_2
);
1553 EXPORT_SYMBOL(folio_end_private_2
);
1556 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1557 * @folio: The folio to wait on.
1559 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1561 void folio_wait_private_2(struct folio
*folio
)
1563 while (folio_test_private_2(folio
))
1564 folio_wait_bit(folio
, PG_private_2
);
1566 EXPORT_SYMBOL(folio_wait_private_2
);
1569 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1570 * @folio: The folio to wait on.
1572 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1573 * fatal signal is received by the calling task.
1576 * - 0 if successful.
1577 * - -EINTR if a fatal signal was encountered.
1579 int folio_wait_private_2_killable(struct folio
*folio
)
1583 while (folio_test_private_2(folio
)) {
1584 ret
= folio_wait_bit_killable(folio
, PG_private_2
);
1591 EXPORT_SYMBOL(folio_wait_private_2_killable
);
1594 * folio_end_writeback - End writeback against a folio.
1595 * @folio: The folio.
1597 void folio_end_writeback(struct folio
*folio
)
1600 * folio_test_clear_reclaim() could be used here but it is an
1601 * atomic operation and overkill in this particular case. Failing
1602 * to shuffle a folio marked for immediate reclaim is too mild
1603 * a gain to justify taking an atomic operation penalty at the
1604 * end of every folio writeback.
1606 if (folio_test_reclaim(folio
)) {
1607 folio_clear_reclaim(folio
);
1608 folio_rotate_reclaimable(folio
);
1612 * Writeback does not hold a folio reference of its own, relying
1613 * on truncation to wait for the clearing of PG_writeback.
1614 * But here we must make sure that the folio is not freed and
1615 * reused before the folio_wake().
1618 if (!__folio_end_writeback(folio
))
1621 smp_mb__after_atomic();
1622 folio_wake(folio
, PG_writeback
);
1623 acct_reclaim_writeback(folio
);
1626 EXPORT_SYMBOL(folio_end_writeback
);
1629 * After completing I/O on a page, call this routine to update the page
1630 * flags appropriately
1632 void page_endio(struct page
*page
, bool is_write
, int err
)
1634 struct folio
*folio
= page_folio(page
);
1638 folio_mark_uptodate(folio
);
1640 folio_clear_uptodate(folio
);
1641 folio_set_error(folio
);
1643 folio_unlock(folio
);
1646 struct address_space
*mapping
;
1648 folio_set_error(folio
);
1649 mapping
= folio_mapping(folio
);
1651 mapping_set_error(mapping
, err
);
1653 folio_end_writeback(folio
);
1656 EXPORT_SYMBOL_GPL(page_endio
);
1659 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1660 * @folio: The folio to lock
1662 void __folio_lock(struct folio
*folio
)
1664 folio_wait_bit_common(folio
, PG_locked
, TASK_UNINTERRUPTIBLE
,
1667 EXPORT_SYMBOL(__folio_lock
);
1669 int __folio_lock_killable(struct folio
*folio
)
1671 return folio_wait_bit_common(folio
, PG_locked
, TASK_KILLABLE
,
1674 EXPORT_SYMBOL_GPL(__folio_lock_killable
);
1676 static int __folio_lock_async(struct folio
*folio
, struct wait_page_queue
*wait
)
1678 struct wait_queue_head
*q
= folio_waitqueue(folio
);
1681 wait
->folio
= folio
;
1682 wait
->bit_nr
= PG_locked
;
1684 spin_lock_irq(&q
->lock
);
1685 __add_wait_queue_entry_tail(q
, &wait
->wait
);
1686 folio_set_waiters(folio
);
1687 ret
= !folio_trylock(folio
);
1689 * If we were successful now, we know we're still on the
1690 * waitqueue as we're still under the lock. This means it's
1691 * safe to remove and return success, we know the callback
1692 * isn't going to trigger.
1695 __remove_wait_queue(q
, &wait
->wait
);
1698 spin_unlock_irq(&q
->lock
);
1704 * true - folio is locked; mmap_lock is still held.
1705 * false - folio is not locked.
1706 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1707 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1708 * which case mmap_lock is still held.
1710 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1711 * with the folio locked and the mmap_lock unperturbed.
1713 bool __folio_lock_or_retry(struct folio
*folio
, struct mm_struct
*mm
,
1716 if (fault_flag_allow_retry_first(flags
)) {
1718 * CAUTION! In this case, mmap_lock is not released
1719 * even though return 0.
1721 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1724 mmap_read_unlock(mm
);
1725 if (flags
& FAULT_FLAG_KILLABLE
)
1726 folio_wait_locked_killable(folio
);
1728 folio_wait_locked(folio
);
1731 if (flags
& FAULT_FLAG_KILLABLE
) {
1734 ret
= __folio_lock_killable(folio
);
1736 mmap_read_unlock(mm
);
1740 __folio_lock(folio
);
1747 * page_cache_next_miss() - Find the next gap in the page cache.
1748 * @mapping: Mapping.
1750 * @max_scan: Maximum range to search.
1752 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1753 * gap with the lowest index.
1755 * This function may be called under the rcu_read_lock. However, this will
1756 * not atomically search a snapshot of the cache at a single point in time.
1757 * For example, if a gap is created at index 5, then subsequently a gap is
1758 * created at index 10, page_cache_next_miss covering both indices may
1759 * return 10 if called under the rcu_read_lock.
1761 * Return: The index of the gap if found, otherwise an index outside the
1762 * range specified (in which case 'return - index >= max_scan' will be true).
1763 * In the rare case of index wrap-around, 0 will be returned.
1765 pgoff_t
page_cache_next_miss(struct address_space
*mapping
,
1766 pgoff_t index
, unsigned long max_scan
)
1768 XA_STATE(xas
, &mapping
->i_pages
, index
);
1770 while (max_scan
--) {
1771 void *entry
= xas_next(&xas
);
1772 if (!entry
|| xa_is_value(entry
))
1774 if (xas
.xa_index
== 0)
1778 return xas
.xa_index
;
1780 EXPORT_SYMBOL(page_cache_next_miss
);
1783 * page_cache_prev_miss() - Find the previous gap in the page cache.
1784 * @mapping: Mapping.
1786 * @max_scan: Maximum range to search.
1788 * Search the range [max(index - max_scan + 1, 0), index] for the
1789 * gap with the highest index.
1791 * This function may be called under the rcu_read_lock. However, this will
1792 * not atomically search a snapshot of the cache at a single point in time.
1793 * For example, if a gap is created at index 10, then subsequently a gap is
1794 * created at index 5, page_cache_prev_miss() covering both indices may
1795 * return 5 if called under the rcu_read_lock.
1797 * Return: The index of the gap if found, otherwise an index outside the
1798 * range specified (in which case 'index - return >= max_scan' will be true).
1799 * In the rare case of wrap-around, ULONG_MAX will be returned.
1801 pgoff_t
page_cache_prev_miss(struct address_space
*mapping
,
1802 pgoff_t index
, unsigned long max_scan
)
1804 XA_STATE(xas
, &mapping
->i_pages
, index
);
1806 while (max_scan
--) {
1807 void *entry
= xas_prev(&xas
);
1808 if (!entry
|| xa_is_value(entry
))
1810 if (xas
.xa_index
== ULONG_MAX
)
1814 return xas
.xa_index
;
1816 EXPORT_SYMBOL(page_cache_prev_miss
);
1819 * Lockless page cache protocol:
1820 * On the lookup side:
1821 * 1. Load the folio from i_pages
1822 * 2. Increment the refcount if it's not zero
1823 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1825 * On the removal side:
1826 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1827 * B. Remove the page from i_pages
1828 * C. Return the page to the page allocator
1830 * This means that any page may have its reference count temporarily
1831 * increased by a speculative page cache (or fast GUP) lookup as it can
1832 * be allocated by another user before the RCU grace period expires.
1833 * Because the refcount temporarily acquired here may end up being the
1834 * last refcount on the page, any page allocation must be freeable by
1839 * filemap_get_entry - Get a page cache entry.
1840 * @mapping: the address_space to search
1841 * @index: The page cache index.
1843 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1844 * it is returned with an increased refcount. If it is a shadow entry
1845 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1846 * it is returned without further action.
1848 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1850 void *filemap_get_entry(struct address_space
*mapping
, pgoff_t index
)
1852 XA_STATE(xas
, &mapping
->i_pages
, index
);
1853 struct folio
*folio
;
1858 folio
= xas_load(&xas
);
1859 if (xas_retry(&xas
, folio
))
1862 * A shadow entry of a recently evicted page, or a swap entry from
1863 * shmem/tmpfs. Return it without attempting to raise page count.
1865 if (!folio
|| xa_is_value(folio
))
1868 if (!folio_try_get_rcu(folio
))
1871 if (unlikely(folio
!= xas_reload(&xas
))) {
1882 * __filemap_get_folio - Find and get a reference to a folio.
1883 * @mapping: The address_space to search.
1884 * @index: The page index.
1885 * @fgp_flags: %FGP flags modify how the folio is returned.
1886 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1888 * Looks up the page cache entry at @mapping & @index.
1890 * @fgp_flags can be zero or more of these flags:
1892 * * %FGP_ACCESSED - The folio will be marked accessed.
1893 * * %FGP_LOCK - The folio is returned locked.
1894 * * %FGP_CREAT - If no page is present then a new page is allocated using
1895 * @gfp and added to the page cache and the VM's LRU list.
1896 * The page is returned locked and with an increased refcount.
1897 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1898 * page is already in cache. If the page was allocated, unlock it before
1899 * returning so the caller can do the same dance.
1900 * * %FGP_WRITE - The page will be written to by the caller.
1901 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1902 * * %FGP_NOWAIT - Don't get blocked by page lock.
1903 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1905 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1906 * if the %GFP flags specified for %FGP_CREAT are atomic.
1908 * If there is a page cache page, it is returned with an increased refcount.
1910 * Return: The found folio or an ERR_PTR() otherwise.
1912 struct folio
*__filemap_get_folio(struct address_space
*mapping
, pgoff_t index
,
1913 int fgp_flags
, gfp_t gfp
)
1915 struct folio
*folio
;
1918 folio
= filemap_get_entry(mapping
, index
);
1919 if (xa_is_value(folio
))
1924 if (fgp_flags
& FGP_LOCK
) {
1925 if (fgp_flags
& FGP_NOWAIT
) {
1926 if (!folio_trylock(folio
)) {
1928 return ERR_PTR(-EAGAIN
);
1934 /* Has the page been truncated? */
1935 if (unlikely(folio
->mapping
!= mapping
)) {
1936 folio_unlock(folio
);
1940 VM_BUG_ON_FOLIO(!folio_contains(folio
, index
), folio
);
1943 if (fgp_flags
& FGP_ACCESSED
)
1944 folio_mark_accessed(folio
);
1945 else if (fgp_flags
& FGP_WRITE
) {
1946 /* Clear idle flag for buffer write */
1947 if (folio_test_idle(folio
))
1948 folio_clear_idle(folio
);
1951 if (fgp_flags
& FGP_STABLE
)
1952 folio_wait_stable(folio
);
1954 if (!folio
&& (fgp_flags
& FGP_CREAT
)) {
1956 if ((fgp_flags
& FGP_WRITE
) && mapping_can_writeback(mapping
))
1958 if (fgp_flags
& FGP_NOFS
)
1960 if (fgp_flags
& FGP_NOWAIT
) {
1962 gfp
|= GFP_NOWAIT
| __GFP_NOWARN
;
1965 folio
= filemap_alloc_folio(gfp
, 0);
1967 return ERR_PTR(-ENOMEM
);
1969 if (WARN_ON_ONCE(!(fgp_flags
& (FGP_LOCK
| FGP_FOR_MMAP
))))
1970 fgp_flags
|= FGP_LOCK
;
1972 /* Init accessed so avoid atomic mark_page_accessed later */
1973 if (fgp_flags
& FGP_ACCESSED
)
1974 __folio_set_referenced(folio
);
1976 err
= filemap_add_folio(mapping
, folio
, index
, gfp
);
1977 if (unlikely(err
)) {
1985 * filemap_add_folio locks the page, and for mmap
1986 * we expect an unlocked page.
1988 if (folio
&& (fgp_flags
& FGP_FOR_MMAP
))
1989 folio_unlock(folio
);
1993 return ERR_PTR(-ENOENT
);
1996 EXPORT_SYMBOL(__filemap_get_folio
);
1998 static inline struct folio
*find_get_entry(struct xa_state
*xas
, pgoff_t max
,
2001 struct folio
*folio
;
2004 if (mark
== XA_PRESENT
)
2005 folio
= xas_find(xas
, max
);
2007 folio
= xas_find_marked(xas
, max
, mark
);
2009 if (xas_retry(xas
, folio
))
2012 * A shadow entry of a recently evicted page, a swap
2013 * entry from shmem/tmpfs or a DAX entry. Return it
2014 * without attempting to raise page count.
2016 if (!folio
|| xa_is_value(folio
))
2019 if (!folio_try_get_rcu(folio
))
2022 if (unlikely(folio
!= xas_reload(xas
))) {
2034 * find_get_entries - gang pagecache lookup
2035 * @mapping: The address_space to search
2036 * @start: The starting page cache index
2037 * @end: The final page index (inclusive).
2038 * @fbatch: Where the resulting entries are placed.
2039 * @indices: The cache indices corresponding to the entries in @entries
2041 * find_get_entries() will search for and return a batch of entries in
2042 * the mapping. The entries are placed in @fbatch. find_get_entries()
2043 * takes a reference on any actual folios it returns.
2045 * The entries have ascending indexes. The indices may not be consecutive
2046 * due to not-present entries or large folios.
2048 * Any shadow entries of evicted folios, or swap entries from
2049 * shmem/tmpfs, are included in the returned array.
2051 * Return: The number of entries which were found.
2053 unsigned find_get_entries(struct address_space
*mapping
, pgoff_t
*start
,
2054 pgoff_t end
, struct folio_batch
*fbatch
, pgoff_t
*indices
)
2056 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2057 struct folio
*folio
;
2060 while ((folio
= find_get_entry(&xas
, end
, XA_PRESENT
)) != NULL
) {
2061 indices
[fbatch
->nr
] = xas
.xa_index
;
2062 if (!folio_batch_add(fbatch
, folio
))
2067 if (folio_batch_count(fbatch
)) {
2068 unsigned long nr
= 1;
2069 int idx
= folio_batch_count(fbatch
) - 1;
2071 folio
= fbatch
->folios
[idx
];
2072 if (!xa_is_value(folio
) && !folio_test_hugetlb(folio
))
2073 nr
= folio_nr_pages(folio
);
2074 *start
= indices
[idx
] + nr
;
2076 return folio_batch_count(fbatch
);
2080 * find_lock_entries - Find a batch of pagecache entries.
2081 * @mapping: The address_space to search.
2082 * @start: The starting page cache index.
2083 * @end: The final page index (inclusive).
2084 * @fbatch: Where the resulting entries are placed.
2085 * @indices: The cache indices of the entries in @fbatch.
2087 * find_lock_entries() will return a batch of entries from @mapping.
2088 * Swap, shadow and DAX entries are included. Folios are returned
2089 * locked and with an incremented refcount. Folios which are locked
2090 * by somebody else or under writeback are skipped. Folios which are
2091 * partially outside the range are not returned.
2093 * The entries have ascending indexes. The indices may not be consecutive
2094 * due to not-present entries, large folios, folios which could not be
2095 * locked or folios under writeback.
2097 * Return: The number of entries which were found.
2099 unsigned find_lock_entries(struct address_space
*mapping
, pgoff_t
*start
,
2100 pgoff_t end
, struct folio_batch
*fbatch
, pgoff_t
*indices
)
2102 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2103 struct folio
*folio
;
2106 while ((folio
= find_get_entry(&xas
, end
, XA_PRESENT
))) {
2107 if (!xa_is_value(folio
)) {
2108 if (folio
->index
< *start
)
2110 if (folio
->index
+ folio_nr_pages(folio
) - 1 > end
)
2112 if (!folio_trylock(folio
))
2114 if (folio
->mapping
!= mapping
||
2115 folio_test_writeback(folio
))
2117 VM_BUG_ON_FOLIO(!folio_contains(folio
, xas
.xa_index
),
2120 indices
[fbatch
->nr
] = xas
.xa_index
;
2121 if (!folio_batch_add(fbatch
, folio
))
2125 folio_unlock(folio
);
2131 if (folio_batch_count(fbatch
)) {
2132 unsigned long nr
= 1;
2133 int idx
= folio_batch_count(fbatch
) - 1;
2135 folio
= fbatch
->folios
[idx
];
2136 if (!xa_is_value(folio
) && !folio_test_hugetlb(folio
))
2137 nr
= folio_nr_pages(folio
);
2138 *start
= indices
[idx
] + nr
;
2140 return folio_batch_count(fbatch
);
2144 * filemap_get_folios - Get a batch of folios
2145 * @mapping: The address_space to search
2146 * @start: The starting page index
2147 * @end: The final page index (inclusive)
2148 * @fbatch: The batch to fill.
2150 * Search for and return a batch of folios in the mapping starting at
2151 * index @start and up to index @end (inclusive). The folios are returned
2152 * in @fbatch with an elevated reference count.
2154 * The first folio may start before @start; if it does, it will contain
2155 * @start. The final folio may extend beyond @end; if it does, it will
2156 * contain @end. The folios have ascending indices. There may be gaps
2157 * between the folios if there are indices which have no folio in the
2158 * page cache. If folios are added to or removed from the page cache
2159 * while this is running, they may or may not be found by this call.
2161 * Return: The number of folios which were found.
2162 * We also update @start to index the next folio for the traversal.
2164 unsigned filemap_get_folios(struct address_space
*mapping
, pgoff_t
*start
,
2165 pgoff_t end
, struct folio_batch
*fbatch
)
2167 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2168 struct folio
*folio
;
2171 while ((folio
= find_get_entry(&xas
, end
, XA_PRESENT
)) != NULL
) {
2172 /* Skip over shadow, swap and DAX entries */
2173 if (xa_is_value(folio
))
2175 if (!folio_batch_add(fbatch
, folio
)) {
2176 unsigned long nr
= folio_nr_pages(folio
);
2178 if (folio_test_hugetlb(folio
))
2180 *start
= folio
->index
+ nr
;
2186 * We come here when there is no page beyond @end. We take care to not
2187 * overflow the index @start as it confuses some of the callers. This
2188 * breaks the iteration when there is a page at index -1 but that is
2189 * already broken anyway.
2191 if (end
== (pgoff_t
)-1)
2192 *start
= (pgoff_t
)-1;
2198 return folio_batch_count(fbatch
);
2200 EXPORT_SYMBOL(filemap_get_folios
);
2203 bool folio_more_pages(struct folio
*folio
, pgoff_t index
, pgoff_t max
)
2205 if (!folio_test_large(folio
) || folio_test_hugetlb(folio
))
2209 return index
< folio
->index
+ folio_nr_pages(folio
) - 1;
2213 * filemap_get_folios_contig - Get a batch of contiguous folios
2214 * @mapping: The address_space to search
2215 * @start: The starting page index
2216 * @end: The final page index (inclusive)
2217 * @fbatch: The batch to fill
2219 * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2220 * except the returned folios are guaranteed to be contiguous. This may
2221 * not return all contiguous folios if the batch gets filled up.
2223 * Return: The number of folios found.
2224 * Also update @start to be positioned for traversal of the next folio.
2227 unsigned filemap_get_folios_contig(struct address_space
*mapping
,
2228 pgoff_t
*start
, pgoff_t end
, struct folio_batch
*fbatch
)
2230 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2232 struct folio
*folio
;
2236 for (folio
= xas_load(&xas
); folio
&& xas
.xa_index
<= end
;
2237 folio
= xas_next(&xas
)) {
2238 if (xas_retry(&xas
, folio
))
2241 * If the entry has been swapped out, we can stop looking.
2242 * No current caller is looking for DAX entries.
2244 if (xa_is_value(folio
))
2247 if (!folio_try_get_rcu(folio
))
2250 if (unlikely(folio
!= xas_reload(&xas
)))
2253 if (!folio_batch_add(fbatch
, folio
)) {
2254 nr
= folio_nr_pages(folio
);
2256 if (folio_test_hugetlb(folio
))
2258 *start
= folio
->index
+ nr
;
2270 nr
= folio_batch_count(fbatch
);
2273 folio
= fbatch
->folios
[nr
- 1];
2274 if (folio_test_hugetlb(folio
))
2275 *start
= folio
->index
+ 1;
2277 *start
= folio
->index
+ folio_nr_pages(folio
);
2281 return folio_batch_count(fbatch
);
2283 EXPORT_SYMBOL(filemap_get_folios_contig
);
2286 * filemap_get_folios_tag - Get a batch of folios matching @tag
2287 * @mapping: The address_space to search
2288 * @start: The starting page index
2289 * @end: The final page index (inclusive)
2290 * @tag: The tag index
2291 * @fbatch: The batch to fill
2293 * Same as filemap_get_folios(), but only returning folios tagged with @tag.
2295 * Return: The number of folios found.
2296 * Also update @start to index the next folio for traversal.
2298 unsigned filemap_get_folios_tag(struct address_space
*mapping
, pgoff_t
*start
,
2299 pgoff_t end
, xa_mark_t tag
, struct folio_batch
*fbatch
)
2301 XA_STATE(xas
, &mapping
->i_pages
, *start
);
2302 struct folio
*folio
;
2305 while ((folio
= find_get_entry(&xas
, end
, tag
)) != NULL
) {
2307 * Shadow entries should never be tagged, but this iteration
2308 * is lockless so there is a window for page reclaim to evict
2309 * a page we saw tagged. Skip over it.
2311 if (xa_is_value(folio
))
2313 if (!folio_batch_add(fbatch
, folio
)) {
2314 unsigned long nr
= folio_nr_pages(folio
);
2316 if (folio_test_hugetlb(folio
))
2318 *start
= folio
->index
+ nr
;
2323 * We come here when there is no page beyond @end. We take care to not
2324 * overflow the index @start as it confuses some of the callers. This
2325 * breaks the iteration when there is a page at index -1 but that is
2326 * already broke anyway.
2328 if (end
== (pgoff_t
)-1)
2329 *start
= (pgoff_t
)-1;
2335 return folio_batch_count(fbatch
);
2337 EXPORT_SYMBOL(filemap_get_folios_tag
);
2340 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2341 * a _large_ part of the i/o request. Imagine the worst scenario:
2343 * ---R__________________________________________B__________
2344 * ^ reading here ^ bad block(assume 4k)
2346 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2347 * => failing the whole request => read(R) => read(R+1) =>
2348 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2349 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2350 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2352 * It is going insane. Fix it by quickly scaling down the readahead size.
2354 static void shrink_readahead_size_eio(struct file_ra_state
*ra
)
2360 * filemap_get_read_batch - Get a batch of folios for read
2362 * Get a batch of folios which represent a contiguous range of bytes in
2363 * the file. No exceptional entries will be returned. If @index is in
2364 * the middle of a folio, the entire folio will be returned. The last
2365 * folio in the batch may have the readahead flag set or the uptodate flag
2366 * clear so that the caller can take the appropriate action.
2368 static void filemap_get_read_batch(struct address_space
*mapping
,
2369 pgoff_t index
, pgoff_t max
, struct folio_batch
*fbatch
)
2371 XA_STATE(xas
, &mapping
->i_pages
, index
);
2372 struct folio
*folio
;
2375 for (folio
= xas_load(&xas
); folio
; folio
= xas_next(&xas
)) {
2376 if (xas_retry(&xas
, folio
))
2378 if (xas
.xa_index
> max
|| xa_is_value(folio
))
2380 if (xa_is_sibling(folio
))
2382 if (!folio_try_get_rcu(folio
))
2385 if (unlikely(folio
!= xas_reload(&xas
)))
2388 if (!folio_batch_add(fbatch
, folio
))
2390 if (!folio_test_uptodate(folio
))
2392 if (folio_test_readahead(folio
))
2394 xas_advance(&xas
, folio
->index
+ folio_nr_pages(folio
) - 1);
2404 static int filemap_read_folio(struct file
*file
, filler_t filler
,
2405 struct folio
*folio
)
2407 bool workingset
= folio_test_workingset(folio
);
2408 unsigned long pflags
;
2412 * A previous I/O error may have been due to temporary failures,
2413 * eg. multipath errors. PG_error will be set again if read_folio
2416 folio_clear_error(folio
);
2418 /* Start the actual read. The read will unlock the page. */
2419 if (unlikely(workingset
))
2420 psi_memstall_enter(&pflags
);
2421 error
= filler(file
, folio
);
2422 if (unlikely(workingset
))
2423 psi_memstall_leave(&pflags
);
2427 error
= folio_wait_locked_killable(folio
);
2430 if (folio_test_uptodate(folio
))
2433 shrink_readahead_size_eio(&file
->f_ra
);
2437 static bool filemap_range_uptodate(struct address_space
*mapping
,
2438 loff_t pos
, size_t count
, struct folio
*folio
,
2441 if (folio_test_uptodate(folio
))
2443 /* pipes can't handle partially uptodate pages */
2446 if (!mapping
->a_ops
->is_partially_uptodate
)
2448 if (mapping
->host
->i_blkbits
>= folio_shift(folio
))
2451 if (folio_pos(folio
) > pos
) {
2452 count
-= folio_pos(folio
) - pos
;
2455 pos
-= folio_pos(folio
);
2458 return mapping
->a_ops
->is_partially_uptodate(folio
, pos
, count
);
2461 static int filemap_update_page(struct kiocb
*iocb
,
2462 struct address_space
*mapping
, size_t count
,
2463 struct folio
*folio
, bool need_uptodate
)
2467 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2468 if (!filemap_invalidate_trylock_shared(mapping
))
2471 filemap_invalidate_lock_shared(mapping
);
2474 if (!folio_trylock(folio
)) {
2476 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_NOIO
))
2477 goto unlock_mapping
;
2478 if (!(iocb
->ki_flags
& IOCB_WAITQ
)) {
2479 filemap_invalidate_unlock_shared(mapping
);
2481 * This is where we usually end up waiting for a
2482 * previously submitted readahead to finish.
2484 folio_put_wait_locked(folio
, TASK_KILLABLE
);
2485 return AOP_TRUNCATED_PAGE
;
2487 error
= __folio_lock_async(folio
, iocb
->ki_waitq
);
2489 goto unlock_mapping
;
2492 error
= AOP_TRUNCATED_PAGE
;
2493 if (!folio
->mapping
)
2497 if (filemap_range_uptodate(mapping
, iocb
->ki_pos
, count
, folio
,
2502 if (iocb
->ki_flags
& (IOCB_NOIO
| IOCB_NOWAIT
| IOCB_WAITQ
))
2505 error
= filemap_read_folio(iocb
->ki_filp
, mapping
->a_ops
->read_folio
,
2507 goto unlock_mapping
;
2509 folio_unlock(folio
);
2511 filemap_invalidate_unlock_shared(mapping
);
2512 if (error
== AOP_TRUNCATED_PAGE
)
2517 static int filemap_create_folio(struct file
*file
,
2518 struct address_space
*mapping
, pgoff_t index
,
2519 struct folio_batch
*fbatch
)
2521 struct folio
*folio
;
2524 folio
= filemap_alloc_folio(mapping_gfp_mask(mapping
), 0);
2529 * Protect against truncate / hole punch. Grabbing invalidate_lock
2530 * here assures we cannot instantiate and bring uptodate new
2531 * pagecache folios after evicting page cache during truncate
2532 * and before actually freeing blocks. Note that we could
2533 * release invalidate_lock after inserting the folio into
2534 * the page cache as the locked folio would then be enough to
2535 * synchronize with hole punching. But there are code paths
2536 * such as filemap_update_page() filling in partially uptodate
2537 * pages or ->readahead() that need to hold invalidate_lock
2538 * while mapping blocks for IO so let's hold the lock here as
2539 * well to keep locking rules simple.
2541 filemap_invalidate_lock_shared(mapping
);
2542 error
= filemap_add_folio(mapping
, folio
, index
,
2543 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2544 if (error
== -EEXIST
)
2545 error
= AOP_TRUNCATED_PAGE
;
2549 error
= filemap_read_folio(file
, mapping
->a_ops
->read_folio
, folio
);
2553 filemap_invalidate_unlock_shared(mapping
);
2554 folio_batch_add(fbatch
, folio
);
2557 filemap_invalidate_unlock_shared(mapping
);
2562 static int filemap_readahead(struct kiocb
*iocb
, struct file
*file
,
2563 struct address_space
*mapping
, struct folio
*folio
,
2566 DEFINE_READAHEAD(ractl
, file
, &file
->f_ra
, mapping
, folio
->index
);
2568 if (iocb
->ki_flags
& IOCB_NOIO
)
2570 page_cache_async_ra(&ractl
, folio
, last_index
- folio
->index
);
2574 static int filemap_get_pages(struct kiocb
*iocb
, size_t count
,
2575 struct folio_batch
*fbatch
, bool need_uptodate
)
2577 struct file
*filp
= iocb
->ki_filp
;
2578 struct address_space
*mapping
= filp
->f_mapping
;
2579 struct file_ra_state
*ra
= &filp
->f_ra
;
2580 pgoff_t index
= iocb
->ki_pos
>> PAGE_SHIFT
;
2582 struct folio
*folio
;
2585 /* "last_index" is the index of the page beyond the end of the read */
2586 last_index
= DIV_ROUND_UP(iocb
->ki_pos
+ count
, PAGE_SIZE
);
2588 if (fatal_signal_pending(current
))
2591 filemap_get_read_batch(mapping
, index
, last_index
- 1, fbatch
);
2592 if (!folio_batch_count(fbatch
)) {
2593 if (iocb
->ki_flags
& IOCB_NOIO
)
2595 page_cache_sync_readahead(mapping
, ra
, filp
, index
,
2596 last_index
- index
);
2597 filemap_get_read_batch(mapping
, index
, last_index
- 1, fbatch
);
2599 if (!folio_batch_count(fbatch
)) {
2600 if (iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_WAITQ
))
2602 err
= filemap_create_folio(filp
, mapping
,
2603 iocb
->ki_pos
>> PAGE_SHIFT
, fbatch
);
2604 if (err
== AOP_TRUNCATED_PAGE
)
2609 folio
= fbatch
->folios
[folio_batch_count(fbatch
) - 1];
2610 if (folio_test_readahead(folio
)) {
2611 err
= filemap_readahead(iocb
, filp
, mapping
, folio
, last_index
);
2615 if (!folio_test_uptodate(folio
)) {
2616 if ((iocb
->ki_flags
& IOCB_WAITQ
) &&
2617 folio_batch_count(fbatch
) > 1)
2618 iocb
->ki_flags
|= IOCB_NOWAIT
;
2619 err
= filemap_update_page(iocb
, mapping
, count
, folio
,
2629 if (likely(--fbatch
->nr
))
2631 if (err
== AOP_TRUNCATED_PAGE
)
2636 static inline bool pos_same_folio(loff_t pos1
, loff_t pos2
, struct folio
*folio
)
2638 unsigned int shift
= folio_shift(folio
);
2640 return (pos1
>> shift
== pos2
>> shift
);
2644 * filemap_read - Read data from the page cache.
2645 * @iocb: The iocb to read.
2646 * @iter: Destination for the data.
2647 * @already_read: Number of bytes already read by the caller.
2649 * Copies data from the page cache. If the data is not currently present,
2650 * uses the readahead and read_folio address_space operations to fetch it.
2652 * Return: Total number of bytes copied, including those already read by
2653 * the caller. If an error happens before any bytes are copied, returns
2654 * a negative error number.
2656 ssize_t
filemap_read(struct kiocb
*iocb
, struct iov_iter
*iter
,
2657 ssize_t already_read
)
2659 struct file
*filp
= iocb
->ki_filp
;
2660 struct file_ra_state
*ra
= &filp
->f_ra
;
2661 struct address_space
*mapping
= filp
->f_mapping
;
2662 struct inode
*inode
= mapping
->host
;
2663 struct folio_batch fbatch
;
2665 bool writably_mapped
;
2666 loff_t isize
, end_offset
;
2668 if (unlikely(iocb
->ki_pos
>= inode
->i_sb
->s_maxbytes
))
2670 if (unlikely(!iov_iter_count(iter
)))
2673 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2674 folio_batch_init(&fbatch
);
2680 * If we've already successfully copied some data, then we
2681 * can no longer safely return -EIOCBQUEUED. Hence mark
2682 * an async read NOWAIT at that point.
2684 if ((iocb
->ki_flags
& IOCB_WAITQ
) && already_read
)
2685 iocb
->ki_flags
|= IOCB_NOWAIT
;
2687 if (unlikely(iocb
->ki_pos
>= i_size_read(inode
)))
2690 error
= filemap_get_pages(iocb
, iter
->count
, &fbatch
,
2691 iov_iter_is_pipe(iter
));
2696 * i_size must be checked after we know the pages are Uptodate.
2698 * Checking i_size after the check allows us to calculate
2699 * the correct value for "nr", which means the zero-filled
2700 * part of the page is not copied back to userspace (unless
2701 * another truncate extends the file - this is desired though).
2703 isize
= i_size_read(inode
);
2704 if (unlikely(iocb
->ki_pos
>= isize
))
2706 end_offset
= min_t(loff_t
, isize
, iocb
->ki_pos
+ iter
->count
);
2709 * Once we start copying data, we don't want to be touching any
2710 * cachelines that might be contended:
2712 writably_mapped
= mapping_writably_mapped(mapping
);
2715 * When a read accesses the same folio several times, only
2716 * mark it as accessed the first time.
2718 if (!pos_same_folio(iocb
->ki_pos
, ra
->prev_pos
- 1,
2720 folio_mark_accessed(fbatch
.folios
[0]);
2722 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
2723 struct folio
*folio
= fbatch
.folios
[i
];
2724 size_t fsize
= folio_size(folio
);
2725 size_t offset
= iocb
->ki_pos
& (fsize
- 1);
2726 size_t bytes
= min_t(loff_t
, end_offset
- iocb
->ki_pos
,
2730 if (end_offset
< folio_pos(folio
))
2733 folio_mark_accessed(folio
);
2735 * If users can be writing to this folio using arbitrary
2736 * virtual addresses, take care of potential aliasing
2737 * before reading the folio on the kernel side.
2739 if (writably_mapped
)
2740 flush_dcache_folio(folio
);
2742 copied
= copy_folio_to_iter(folio
, offset
, bytes
, iter
);
2744 already_read
+= copied
;
2745 iocb
->ki_pos
+= copied
;
2746 ra
->prev_pos
= iocb
->ki_pos
;
2748 if (copied
< bytes
) {
2754 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++)
2755 folio_put(fbatch
.folios
[i
]);
2756 folio_batch_init(&fbatch
);
2757 } while (iov_iter_count(iter
) && iocb
->ki_pos
< isize
&& !error
);
2759 file_accessed(filp
);
2761 return already_read
? already_read
: error
;
2763 EXPORT_SYMBOL_GPL(filemap_read
);
2766 * generic_file_read_iter - generic filesystem read routine
2767 * @iocb: kernel I/O control block
2768 * @iter: destination for the data read
2770 * This is the "read_iter()" routine for all filesystems
2771 * that can use the page cache directly.
2773 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2774 * be returned when no data can be read without waiting for I/O requests
2775 * to complete; it doesn't prevent readahead.
2777 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2778 * requests shall be made for the read or for readahead. When no data
2779 * can be read, -EAGAIN shall be returned. When readahead would be
2780 * triggered, a partial, possibly empty read shall be returned.
2783 * * number of bytes copied, even for partial reads
2784 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2787 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2789 size_t count
= iov_iter_count(iter
);
2793 return 0; /* skip atime */
2795 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2796 struct file
*file
= iocb
->ki_filp
;
2797 struct address_space
*mapping
= file
->f_mapping
;
2798 struct inode
*inode
= mapping
->host
;
2800 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2801 if (filemap_range_needs_writeback(mapping
, iocb
->ki_pos
,
2802 iocb
->ki_pos
+ count
- 1))
2805 retval
= filemap_write_and_wait_range(mapping
,
2807 iocb
->ki_pos
+ count
- 1);
2812 file_accessed(file
);
2814 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2816 iocb
->ki_pos
+= retval
;
2819 if (retval
!= -EIOCBQUEUED
)
2820 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2823 * Btrfs can have a short DIO read if we encounter
2824 * compressed extents, so if there was an error, or if
2825 * we've already read everything we wanted to, or if
2826 * there was a short read because we hit EOF, go ahead
2827 * and return. Otherwise fallthrough to buffered io for
2828 * the rest of the read. Buffered reads will not work for
2829 * DAX files, so don't bother trying.
2831 if (retval
< 0 || !count
|| IS_DAX(inode
))
2833 if (iocb
->ki_pos
>= i_size_read(inode
))
2837 return filemap_read(iocb
, iter
, retval
);
2839 EXPORT_SYMBOL(generic_file_read_iter
);
2842 * Splice subpages from a folio into a pipe.
2844 size_t splice_folio_into_pipe(struct pipe_inode_info
*pipe
,
2845 struct folio
*folio
, loff_t fpos
, size_t size
)
2848 size_t spliced
= 0, offset
= offset_in_folio(folio
, fpos
);
2850 page
= folio_page(folio
, offset
/ PAGE_SIZE
);
2851 size
= min(size
, folio_size(folio
) - offset
);
2852 offset
%= PAGE_SIZE
;
2854 while (spliced
< size
&&
2855 !pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
)) {
2856 struct pipe_buffer
*buf
= pipe_head_buf(pipe
);
2857 size_t part
= min_t(size_t, PAGE_SIZE
- offset
, size
- spliced
);
2859 *buf
= (struct pipe_buffer
) {
2860 .ops
= &page_cache_pipe_buf_ops
,
2876 * Splice folios from the pagecache of a buffered (ie. non-O_DIRECT) file into
2879 ssize_t
filemap_splice_read(struct file
*in
, loff_t
*ppos
,
2880 struct pipe_inode_info
*pipe
,
2881 size_t len
, unsigned int flags
)
2883 struct folio_batch fbatch
;
2885 size_t total_spliced
= 0, used
, npages
;
2886 loff_t isize
, end_offset
;
2887 bool writably_mapped
;
2890 init_sync_kiocb(&iocb
, in
);
2891 iocb
.ki_pos
= *ppos
;
2893 /* Work out how much data we can actually add into the pipe */
2894 used
= pipe_occupancy(pipe
->head
, pipe
->tail
);
2895 npages
= max_t(ssize_t
, pipe
->max_usage
- used
, 0);
2896 len
= min_t(size_t, len
, npages
* PAGE_SIZE
);
2898 folio_batch_init(&fbatch
);
2903 if (*ppos
>= i_size_read(file_inode(in
)))
2906 iocb
.ki_pos
= *ppos
;
2907 error
= filemap_get_pages(&iocb
, len
, &fbatch
, true);
2912 * i_size must be checked after we know the pages are Uptodate.
2914 * Checking i_size after the check allows us to calculate
2915 * the correct value for "nr", which means the zero-filled
2916 * part of the page is not copied back to userspace (unless
2917 * another truncate extends the file - this is desired though).
2919 isize
= i_size_read(file_inode(in
));
2920 if (unlikely(*ppos
>= isize
))
2922 end_offset
= min_t(loff_t
, isize
, *ppos
+ len
);
2925 * Once we start copying data, we don't want to be touching any
2926 * cachelines that might be contended:
2928 writably_mapped
= mapping_writably_mapped(in
->f_mapping
);
2930 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
2931 struct folio
*folio
= fbatch
.folios
[i
];
2934 if (folio_pos(folio
) >= end_offset
)
2936 folio_mark_accessed(folio
);
2939 * If users can be writing to this folio using arbitrary
2940 * virtual addresses, take care of potential aliasing
2941 * before reading the folio on the kernel side.
2943 if (writably_mapped
)
2944 flush_dcache_folio(folio
);
2946 n
= min_t(loff_t
, len
, isize
- *ppos
);
2947 n
= splice_folio_into_pipe(pipe
, folio
, *ppos
, n
);
2953 in
->f_ra
.prev_pos
= *ppos
;
2954 if (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
2958 folio_batch_release(&fbatch
);
2962 folio_batch_release(&fbatch
);
2965 return total_spliced
? total_spliced
: error
;
2967 EXPORT_SYMBOL(filemap_splice_read
);
2969 static inline loff_t
folio_seek_hole_data(struct xa_state
*xas
,
2970 struct address_space
*mapping
, struct folio
*folio
,
2971 loff_t start
, loff_t end
, bool seek_data
)
2973 const struct address_space_operations
*ops
= mapping
->a_ops
;
2974 size_t offset
, bsz
= i_blocksize(mapping
->host
);
2976 if (xa_is_value(folio
) || folio_test_uptodate(folio
))
2977 return seek_data
? start
: end
;
2978 if (!ops
->is_partially_uptodate
)
2979 return seek_data
? end
: start
;
2984 if (unlikely(folio
->mapping
!= mapping
))
2987 offset
= offset_in_folio(folio
, start
) & ~(bsz
- 1);
2990 if (ops
->is_partially_uptodate(folio
, offset
, bsz
) ==
2993 start
= (start
+ bsz
) & ~(bsz
- 1);
2995 } while (offset
< folio_size(folio
));
2997 folio_unlock(folio
);
3002 static inline size_t seek_folio_size(struct xa_state
*xas
, struct folio
*folio
)
3004 if (xa_is_value(folio
))
3005 return PAGE_SIZE
<< xa_get_order(xas
->xa
, xas
->xa_index
);
3006 return folio_size(folio
);
3010 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3011 * @mapping: Address space to search.
3012 * @start: First byte to consider.
3013 * @end: Limit of search (exclusive).
3014 * @whence: Either SEEK_HOLE or SEEK_DATA.
3016 * If the page cache knows which blocks contain holes and which blocks
3017 * contain data, your filesystem can use this function to implement
3018 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3019 * entirely memory-based such as tmpfs, and filesystems which support
3020 * unwritten extents.
3022 * Return: The requested offset on success, or -ENXIO if @whence specifies
3023 * SEEK_DATA and there is no data after @start. There is an implicit hole
3024 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3025 * and @end contain data.
3027 loff_t
mapping_seek_hole_data(struct address_space
*mapping
, loff_t start
,
3028 loff_t end
, int whence
)
3030 XA_STATE(xas
, &mapping
->i_pages
, start
>> PAGE_SHIFT
);
3031 pgoff_t max
= (end
- 1) >> PAGE_SHIFT
;
3032 bool seek_data
= (whence
== SEEK_DATA
);
3033 struct folio
*folio
;
3039 while ((folio
= find_get_entry(&xas
, max
, XA_PRESENT
))) {
3040 loff_t pos
= (u64
)xas
.xa_index
<< PAGE_SHIFT
;
3049 seek_size
= seek_folio_size(&xas
, folio
);
3050 pos
= round_up((u64
)pos
+ 1, seek_size
);
3051 start
= folio_seek_hole_data(&xas
, mapping
, folio
, start
, pos
,
3057 if (seek_size
> PAGE_SIZE
)
3058 xas_set(&xas
, pos
>> PAGE_SHIFT
);
3059 if (!xa_is_value(folio
))
3066 if (folio
&& !xa_is_value(folio
))
3074 #define MMAP_LOTSAMISS (100)
3076 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3077 * @vmf - the vm_fault for this fault.
3078 * @folio - the folio to lock.
3079 * @fpin - the pointer to the file we may pin (or is already pinned).
3081 * This works similar to lock_folio_or_retry in that it can drop the
3082 * mmap_lock. It differs in that it actually returns the folio locked
3083 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
3084 * to drop the mmap_lock then fpin will point to the pinned file and
3085 * needs to be fput()'ed at a later point.
3087 static int lock_folio_maybe_drop_mmap(struct vm_fault
*vmf
, struct folio
*folio
,
3090 if (folio_trylock(folio
))
3094 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3095 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3096 * is supposed to work. We have way too many special cases..
3098 if (vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)
3101 *fpin
= maybe_unlock_mmap_for_io(vmf
, *fpin
);
3102 if (vmf
->flags
& FAULT_FLAG_KILLABLE
) {
3103 if (__folio_lock_killable(folio
)) {
3105 * We didn't have the right flags to drop the mmap_lock,
3106 * but all fault_handlers only check for fatal signals
3107 * if we return VM_FAULT_RETRY, so we need to drop the
3108 * mmap_lock here and return 0 if we don't have a fpin.
3111 mmap_read_unlock(vmf
->vma
->vm_mm
);
3115 __folio_lock(folio
);
3121 * Synchronous readahead happens when we don't even find a page in the page
3122 * cache at all. We don't want to perform IO under the mmap sem, so if we have
3123 * to drop the mmap sem we return the file that was pinned in order for us to do
3124 * that. If we didn't pin a file then we return NULL. The file that is
3125 * returned needs to be fput()'ed when we're done with it.
3127 static struct file
*do_sync_mmap_readahead(struct vm_fault
*vmf
)
3129 struct file
*file
= vmf
->vma
->vm_file
;
3130 struct file_ra_state
*ra
= &file
->f_ra
;
3131 struct address_space
*mapping
= file
->f_mapping
;
3132 DEFINE_READAHEAD(ractl
, file
, ra
, mapping
, vmf
->pgoff
);
3133 struct file
*fpin
= NULL
;
3134 unsigned long vm_flags
= vmf
->vma
->vm_flags
;
3135 unsigned int mmap_miss
;
3137 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3138 /* Use the readahead code, even if readahead is disabled */
3139 if (vm_flags
& VM_HUGEPAGE
) {
3140 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3141 ractl
._index
&= ~((unsigned long)HPAGE_PMD_NR
- 1);
3142 ra
->size
= HPAGE_PMD_NR
;
3144 * Fetch two PMD folios, so we get the chance to actually
3145 * readahead, unless we've been told not to.
3147 if (!(vm_flags
& VM_RAND_READ
))
3149 ra
->async_size
= HPAGE_PMD_NR
;
3150 page_cache_ra_order(&ractl
, ra
, HPAGE_PMD_ORDER
);
3155 /* If we don't want any read-ahead, don't bother */
3156 if (vm_flags
& VM_RAND_READ
)
3161 if (vm_flags
& VM_SEQ_READ
) {
3162 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3163 page_cache_sync_ra(&ractl
, ra
->ra_pages
);
3167 /* Avoid banging the cache line if not needed */
3168 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
3169 if (mmap_miss
< MMAP_LOTSAMISS
* 10)
3170 WRITE_ONCE(ra
->mmap_miss
, ++mmap_miss
);
3173 * Do we miss much more than hit in this file? If so,
3174 * stop bothering with read-ahead. It will only hurt.
3176 if (mmap_miss
> MMAP_LOTSAMISS
)
3182 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3183 ra
->start
= max_t(long, 0, vmf
->pgoff
- ra
->ra_pages
/ 2);
3184 ra
->size
= ra
->ra_pages
;
3185 ra
->async_size
= ra
->ra_pages
/ 4;
3186 ractl
._index
= ra
->start
;
3187 page_cache_ra_order(&ractl
, ra
, 0);
3192 * Asynchronous readahead happens when we find the page and PG_readahead,
3193 * so we want to possibly extend the readahead further. We return the file that
3194 * was pinned if we have to drop the mmap_lock in order to do IO.
3196 static struct file
*do_async_mmap_readahead(struct vm_fault
*vmf
,
3197 struct folio
*folio
)
3199 struct file
*file
= vmf
->vma
->vm_file
;
3200 struct file_ra_state
*ra
= &file
->f_ra
;
3201 DEFINE_READAHEAD(ractl
, file
, ra
, file
->f_mapping
, vmf
->pgoff
);
3202 struct file
*fpin
= NULL
;
3203 unsigned int mmap_miss
;
3205 /* If we don't want any read-ahead, don't bother */
3206 if (vmf
->vma
->vm_flags
& VM_RAND_READ
|| !ra
->ra_pages
)
3209 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
3211 WRITE_ONCE(ra
->mmap_miss
, --mmap_miss
);
3213 if (folio_test_readahead(folio
)) {
3214 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3215 page_cache_async_ra(&ractl
, folio
, ra
->ra_pages
);
3221 * filemap_fault - read in file data for page fault handling
3222 * @vmf: struct vm_fault containing details of the fault
3224 * filemap_fault() is invoked via the vma operations vector for a
3225 * mapped memory region to read in file data during a page fault.
3227 * The goto's are kind of ugly, but this streamlines the normal case of having
3228 * it in the page cache, and handles the special cases reasonably without
3229 * having a lot of duplicated code.
3231 * vma->vm_mm->mmap_lock must be held on entry.
3233 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3234 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3236 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3237 * has not been released.
3239 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3241 * Return: bitwise-OR of %VM_FAULT_ codes.
3243 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
3246 struct file
*file
= vmf
->vma
->vm_file
;
3247 struct file
*fpin
= NULL
;
3248 struct address_space
*mapping
= file
->f_mapping
;
3249 struct inode
*inode
= mapping
->host
;
3250 pgoff_t max_idx
, index
= vmf
->pgoff
;
3251 struct folio
*folio
;
3253 bool mapping_locked
= false;
3255 max_idx
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
3256 if (unlikely(index
>= max_idx
))
3257 return VM_FAULT_SIGBUS
;
3260 * Do we have something in the page cache already?
3262 folio
= filemap_get_folio(mapping
, index
);
3263 if (likely(!IS_ERR(folio
))) {
3265 * We found the page, so try async readahead before waiting for
3268 if (!(vmf
->flags
& FAULT_FLAG_TRIED
))
3269 fpin
= do_async_mmap_readahead(vmf
, folio
);
3270 if (unlikely(!folio_test_uptodate(folio
))) {
3271 filemap_invalidate_lock_shared(mapping
);
3272 mapping_locked
= true;
3275 /* No page in the page cache at all */
3276 count_vm_event(PGMAJFAULT
);
3277 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
3278 ret
= VM_FAULT_MAJOR
;
3279 fpin
= do_sync_mmap_readahead(vmf
);
3282 * See comment in filemap_create_folio() why we need
3285 if (!mapping_locked
) {
3286 filemap_invalidate_lock_shared(mapping
);
3287 mapping_locked
= true;
3289 folio
= __filemap_get_folio(mapping
, index
,
3290 FGP_CREAT
|FGP_FOR_MMAP
,
3292 if (IS_ERR(folio
)) {
3295 filemap_invalidate_unlock_shared(mapping
);
3296 return VM_FAULT_OOM
;
3300 if (!lock_folio_maybe_drop_mmap(vmf
, folio
, &fpin
))
3303 /* Did it get truncated? */
3304 if (unlikely(folio
->mapping
!= mapping
)) {
3305 folio_unlock(folio
);
3309 VM_BUG_ON_FOLIO(!folio_contains(folio
, index
), folio
);
3312 * We have a locked page in the page cache, now we need to check
3313 * that it's up-to-date. If not, it is going to be due to an error.
3315 if (unlikely(!folio_test_uptodate(folio
))) {
3317 * The page was in cache and uptodate and now it is not.
3318 * Strange but possible since we didn't hold the page lock all
3319 * the time. Let's drop everything get the invalidate lock and
3322 if (!mapping_locked
) {
3323 folio_unlock(folio
);
3327 goto page_not_uptodate
;
3331 * We've made it this far and we had to drop our mmap_lock, now is the
3332 * time to return to the upper layer and have it re-find the vma and
3336 folio_unlock(folio
);
3340 filemap_invalidate_unlock_shared(mapping
);
3343 * Found the page and have a reference on it.
3344 * We must recheck i_size under page lock.
3346 max_idx
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
3347 if (unlikely(index
>= max_idx
)) {
3348 folio_unlock(folio
);
3350 return VM_FAULT_SIGBUS
;
3353 vmf
->page
= folio_file_page(folio
, index
);
3354 return ret
| VM_FAULT_LOCKED
;
3358 * Umm, take care of errors if the page isn't up-to-date.
3359 * Try to re-read it _once_. We do this synchronously,
3360 * because there really aren't any performance issues here
3361 * and we need to check for errors.
3363 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
3364 error
= filemap_read_folio(file
, mapping
->a_ops
->read_folio
, folio
);
3369 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
3371 filemap_invalidate_unlock_shared(mapping
);
3373 return VM_FAULT_SIGBUS
;
3377 * We dropped the mmap_lock, we need to return to the fault handler to
3378 * re-find the vma and come back and find our hopefully still populated
3384 filemap_invalidate_unlock_shared(mapping
);
3387 return ret
| VM_FAULT_RETRY
;
3389 EXPORT_SYMBOL(filemap_fault
);
3391 static bool filemap_map_pmd(struct vm_fault
*vmf
, struct folio
*folio
,
3394 struct mm_struct
*mm
= vmf
->vma
->vm_mm
;
3396 /* Huge page is mapped? No need to proceed. */
3397 if (pmd_trans_huge(*vmf
->pmd
)) {
3398 folio_unlock(folio
);
3403 if (pmd_none(*vmf
->pmd
) && folio_test_pmd_mappable(folio
)) {
3404 struct page
*page
= folio_file_page(folio
, start
);
3405 vm_fault_t ret
= do_set_pmd(vmf
, page
);
3407 /* The page is mapped successfully, reference consumed. */
3408 folio_unlock(folio
);
3413 if (pmd_none(*vmf
->pmd
))
3414 pmd_install(mm
, vmf
->pmd
, &vmf
->prealloc_pte
);
3416 /* See comment in handle_pte_fault() */
3417 if (pmd_devmap_trans_unstable(vmf
->pmd
)) {
3418 folio_unlock(folio
);
3426 static struct folio
*next_uptodate_page(struct folio
*folio
,
3427 struct address_space
*mapping
,
3428 struct xa_state
*xas
, pgoff_t end_pgoff
)
3430 unsigned long max_idx
;
3435 if (xas_retry(xas
, folio
))
3437 if (xa_is_value(folio
))
3439 if (folio_test_locked(folio
))
3441 if (!folio_try_get_rcu(folio
))
3443 /* Has the page moved or been split? */
3444 if (unlikely(folio
!= xas_reload(xas
)))
3446 if (!folio_test_uptodate(folio
) || folio_test_readahead(folio
))
3448 if (!folio_trylock(folio
))
3450 if (folio
->mapping
!= mapping
)
3452 if (!folio_test_uptodate(folio
))
3454 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
3455 if (xas
->xa_index
>= max_idx
)
3459 folio_unlock(folio
);
3462 } while ((folio
= xas_next_entry(xas
, end_pgoff
)) != NULL
);
3467 static inline struct folio
*first_map_page(struct address_space
*mapping
,
3468 struct xa_state
*xas
,
3471 return next_uptodate_page(xas_find(xas
, end_pgoff
),
3472 mapping
, xas
, end_pgoff
);
3475 static inline struct folio
*next_map_page(struct address_space
*mapping
,
3476 struct xa_state
*xas
,
3479 return next_uptodate_page(xas_next_entry(xas
, end_pgoff
),
3480 mapping
, xas
, end_pgoff
);
3483 vm_fault_t
filemap_map_pages(struct vm_fault
*vmf
,
3484 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
3486 struct vm_area_struct
*vma
= vmf
->vma
;
3487 struct file
*file
= vma
->vm_file
;
3488 struct address_space
*mapping
= file
->f_mapping
;
3489 pgoff_t last_pgoff
= start_pgoff
;
3491 XA_STATE(xas
, &mapping
->i_pages
, start_pgoff
);
3492 struct folio
*folio
;
3494 unsigned int mmap_miss
= READ_ONCE(file
->f_ra
.mmap_miss
);
3498 folio
= first_map_page(mapping
, &xas
, end_pgoff
);
3502 if (filemap_map_pmd(vmf
, folio
, start_pgoff
)) {
3503 ret
= VM_FAULT_NOPAGE
;
3507 addr
= vma
->vm_start
+ ((start_pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
3508 vmf
->pte
= pte_offset_map_lock(vma
->vm_mm
, vmf
->pmd
, addr
, &vmf
->ptl
);
3511 page
= folio_file_page(folio
, xas
.xa_index
);
3512 if (PageHWPoison(page
))
3518 addr
+= (xas
.xa_index
- last_pgoff
) << PAGE_SHIFT
;
3519 vmf
->pte
+= xas
.xa_index
- last_pgoff
;
3520 last_pgoff
= xas
.xa_index
;
3523 * NOTE: If there're PTE markers, we'll leave them to be
3524 * handled in the specific fault path, and it'll prohibit the
3525 * fault-around logic.
3527 if (!pte_none(*vmf
->pte
))
3530 /* We're about to handle the fault */
3531 if (vmf
->address
== addr
)
3532 ret
= VM_FAULT_NOPAGE
;
3534 do_set_pte(vmf
, page
, addr
);
3535 /* no need to invalidate: a not-present page won't be cached */
3536 update_mmu_cache(vma
, addr
, vmf
->pte
);
3537 if (folio_more_pages(folio
, xas
.xa_index
, end_pgoff
)) {
3539 folio_ref_inc(folio
);
3542 folio_unlock(folio
);
3545 if (folio_more_pages(folio
, xas
.xa_index
, end_pgoff
)) {
3549 folio_unlock(folio
);
3551 } while ((folio
= next_map_page(mapping
, &xas
, end_pgoff
)) != NULL
);
3552 pte_unmap_unlock(vmf
->pte
, vmf
->ptl
);
3555 WRITE_ONCE(file
->f_ra
.mmap_miss
, mmap_miss
);
3558 EXPORT_SYMBOL(filemap_map_pages
);
3560 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3562 struct address_space
*mapping
= vmf
->vma
->vm_file
->f_mapping
;
3563 struct folio
*folio
= page_folio(vmf
->page
);
3564 vm_fault_t ret
= VM_FAULT_LOCKED
;
3566 sb_start_pagefault(mapping
->host
->i_sb
);
3567 file_update_time(vmf
->vma
->vm_file
);
3569 if (folio
->mapping
!= mapping
) {
3570 folio_unlock(folio
);
3571 ret
= VM_FAULT_NOPAGE
;
3575 * We mark the folio dirty already here so that when freeze is in
3576 * progress, we are guaranteed that writeback during freezing will
3577 * see the dirty folio and writeprotect it again.
3579 folio_mark_dirty(folio
);
3580 folio_wait_stable(folio
);
3582 sb_end_pagefault(mapping
->host
->i_sb
);
3586 const struct vm_operations_struct generic_file_vm_ops
= {
3587 .fault
= filemap_fault
,
3588 .map_pages
= filemap_map_pages
,
3589 .page_mkwrite
= filemap_page_mkwrite
,
3592 /* This is used for a general mmap of a disk file */
3594 int generic_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3596 struct address_space
*mapping
= file
->f_mapping
;
3598 if (!mapping
->a_ops
->read_folio
)
3600 file_accessed(file
);
3601 vma
->vm_ops
= &generic_file_vm_ops
;
3606 * This is for filesystems which do not implement ->writepage.
3608 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3610 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
3612 return generic_file_mmap(file
, vma
);
3615 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
3617 return VM_FAULT_SIGBUS
;
3619 int generic_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3623 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3627 #endif /* CONFIG_MMU */
3629 EXPORT_SYMBOL(filemap_page_mkwrite
);
3630 EXPORT_SYMBOL(generic_file_mmap
);
3631 EXPORT_SYMBOL(generic_file_readonly_mmap
);
3633 static struct folio
*do_read_cache_folio(struct address_space
*mapping
,
3634 pgoff_t index
, filler_t filler
, struct file
*file
, gfp_t gfp
)
3636 struct folio
*folio
;
3640 filler
= mapping
->a_ops
->read_folio
;
3642 folio
= filemap_get_folio(mapping
, index
);
3643 if (IS_ERR(folio
)) {
3644 folio
= filemap_alloc_folio(gfp
, 0);
3646 return ERR_PTR(-ENOMEM
);
3647 err
= filemap_add_folio(mapping
, folio
, index
, gfp
);
3648 if (unlikely(err
)) {
3652 /* Presumably ENOMEM for xarray node */
3653 return ERR_PTR(err
);
3658 if (folio_test_uptodate(folio
))
3661 if (!folio_trylock(folio
)) {
3662 folio_put_wait_locked(folio
, TASK_UNINTERRUPTIBLE
);
3666 /* Folio was truncated from mapping */
3667 if (!folio
->mapping
) {
3668 folio_unlock(folio
);
3673 /* Someone else locked and filled the page in a very small window */
3674 if (folio_test_uptodate(folio
)) {
3675 folio_unlock(folio
);
3680 err
= filemap_read_folio(file
, filler
, folio
);
3683 if (err
== AOP_TRUNCATED_PAGE
)
3685 return ERR_PTR(err
);
3689 folio_mark_accessed(folio
);
3694 * read_cache_folio - Read into page cache, fill it if needed.
3695 * @mapping: The address_space to read from.
3696 * @index: The index to read.
3697 * @filler: Function to perform the read, or NULL to use aops->read_folio().
3698 * @file: Passed to filler function, may be NULL if not required.
3700 * Read one page into the page cache. If it succeeds, the folio returned
3701 * will contain @index, but it may not be the first page of the folio.
3703 * If the filler function returns an error, it will be returned to the
3706 * Context: May sleep. Expects mapping->invalidate_lock to be held.
3707 * Return: An uptodate folio on success, ERR_PTR() on failure.
3709 struct folio
*read_cache_folio(struct address_space
*mapping
, pgoff_t index
,
3710 filler_t filler
, struct file
*file
)
3712 return do_read_cache_folio(mapping
, index
, filler
, file
,
3713 mapping_gfp_mask(mapping
));
3715 EXPORT_SYMBOL(read_cache_folio
);
3718 * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3719 * @mapping: The address_space for the folio.
3720 * @index: The index that the allocated folio will contain.
3721 * @gfp: The page allocator flags to use if allocating.
3723 * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3724 * any new memory allocations done using the specified allocation flags.
3726 * The most likely error from this function is EIO, but ENOMEM is
3727 * possible and so is EINTR. If ->read_folio returns another error,
3728 * that will be returned to the caller.
3730 * The function expects mapping->invalidate_lock to be already held.
3732 * Return: Uptodate folio on success, ERR_PTR() on failure.
3734 struct folio
*mapping_read_folio_gfp(struct address_space
*mapping
,
3735 pgoff_t index
, gfp_t gfp
)
3737 return do_read_cache_folio(mapping
, index
, NULL
, NULL
, gfp
);
3739 EXPORT_SYMBOL(mapping_read_folio_gfp
);
3741 static struct page
*do_read_cache_page(struct address_space
*mapping
,
3742 pgoff_t index
, filler_t
*filler
, struct file
*file
, gfp_t gfp
)
3744 struct folio
*folio
;
3746 folio
= do_read_cache_folio(mapping
, index
, filler
, file
, gfp
);
3748 return &folio
->page
;
3749 return folio_file_page(folio
, index
);
3752 struct page
*read_cache_page(struct address_space
*mapping
,
3753 pgoff_t index
, filler_t
*filler
, struct file
*file
)
3755 return do_read_cache_page(mapping
, index
, filler
, file
,
3756 mapping_gfp_mask(mapping
));
3758 EXPORT_SYMBOL(read_cache_page
);
3761 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3762 * @mapping: the page's address_space
3763 * @index: the page index
3764 * @gfp: the page allocator flags to use if allocating
3766 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3767 * any new page allocations done using the specified allocation flags.
3769 * If the page does not get brought uptodate, return -EIO.
3771 * The function expects mapping->invalidate_lock to be already held.
3773 * Return: up to date page on success, ERR_PTR() on failure.
3775 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
3779 return do_read_cache_page(mapping
, index
, NULL
, NULL
, gfp
);
3781 EXPORT_SYMBOL(read_cache_page_gfp
);
3784 * Warn about a page cache invalidation failure during a direct I/O write.
3786 void dio_warn_stale_pagecache(struct file
*filp
)
3788 static DEFINE_RATELIMIT_STATE(_rs
, 86400 * HZ
, DEFAULT_RATELIMIT_BURST
);
3792 errseq_set(&filp
->f_mapping
->wb_err
, -EIO
);
3793 if (__ratelimit(&_rs
)) {
3794 path
= file_path(filp
, pathname
, sizeof(pathname
));
3797 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3798 pr_crit("File: %s PID: %d Comm: %.20s\n", path
, current
->pid
,
3804 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3806 struct file
*file
= iocb
->ki_filp
;
3807 struct address_space
*mapping
= file
->f_mapping
;
3808 struct inode
*inode
= mapping
->host
;
3809 loff_t pos
= iocb
->ki_pos
;
3814 write_len
= iov_iter_count(from
);
3815 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
3817 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3818 /* If there are pages to writeback, return */
3819 if (filemap_range_has_page(file
->f_mapping
, pos
,
3820 pos
+ write_len
- 1))
3823 written
= filemap_write_and_wait_range(mapping
, pos
,
3824 pos
+ write_len
- 1);
3830 * After a write we want buffered reads to be sure to go to disk to get
3831 * the new data. We invalidate clean cached page from the region we're
3832 * about to write. We do this *before* the write so that we can return
3833 * without clobbering -EIOCBQUEUED from ->direct_IO().
3835 written
= invalidate_inode_pages2_range(mapping
,
3836 pos
>> PAGE_SHIFT
, end
);
3838 * If a page can not be invalidated, return 0 to fall back
3839 * to buffered write.
3842 if (written
== -EBUSY
)
3847 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3850 * Finally, try again to invalidate clean pages which might have been
3851 * cached by non-direct readahead, or faulted in by get_user_pages()
3852 * if the source of the write was an mmap'ed region of the file
3853 * we're writing. Either one is a pretty crazy thing to do,
3854 * so we don't support it 100%. If this invalidation
3855 * fails, tough, the write still worked...
3857 * Most of the time we do not need this since dio_complete() will do
3858 * the invalidation for us. However there are some file systems that
3859 * do not end up with dio_complete() being called, so let's not break
3860 * them by removing it completely.
3862 * Noticeable example is a blkdev_direct_IO().
3864 * Skip invalidation for async writes or if mapping has no pages.
3866 if (written
> 0 && mapping
->nrpages
&&
3867 invalidate_inode_pages2_range(mapping
, pos
>> PAGE_SHIFT
, end
))
3868 dio_warn_stale_pagecache(file
);
3872 write_len
-= written
;
3873 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3874 i_size_write(inode
, pos
);
3875 mark_inode_dirty(inode
);
3879 if (written
!= -EIOCBQUEUED
)
3880 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3884 EXPORT_SYMBOL(generic_file_direct_write
);
3886 ssize_t
generic_perform_write(struct kiocb
*iocb
, struct iov_iter
*i
)
3888 struct file
*file
= iocb
->ki_filp
;
3889 loff_t pos
= iocb
->ki_pos
;
3890 struct address_space
*mapping
= file
->f_mapping
;
3891 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3893 ssize_t written
= 0;
3897 unsigned long offset
; /* Offset into pagecache page */
3898 unsigned long bytes
; /* Bytes to write to page */
3899 size_t copied
; /* Bytes copied from user */
3900 void *fsdata
= NULL
;
3902 offset
= (pos
& (PAGE_SIZE
- 1));
3903 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3908 * Bring in the user page that we will copy from _first_.
3909 * Otherwise there's a nasty deadlock on copying from the
3910 * same page as we're writing to, without it being marked
3913 if (unlikely(fault_in_iov_iter_readable(i
, bytes
) == bytes
)) {
3918 if (fatal_signal_pending(current
)) {
3923 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
,
3925 if (unlikely(status
< 0))
3928 if (mapping_writably_mapped(mapping
))
3929 flush_dcache_page(page
);
3931 copied
= copy_page_from_iter_atomic(page
, offset
, bytes
, i
);
3932 flush_dcache_page(page
);
3934 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3936 if (unlikely(status
!= copied
)) {
3937 iov_iter_revert(i
, copied
- max(status
, 0L));
3938 if (unlikely(status
< 0))
3943 if (unlikely(status
== 0)) {
3945 * A short copy made ->write_end() reject the
3946 * thing entirely. Might be memory poisoning
3947 * halfway through, might be a race with munmap,
3948 * might be severe memory pressure.
3957 balance_dirty_pages_ratelimited(mapping
);
3958 } while (iov_iter_count(i
));
3960 return written
? written
: status
;
3962 EXPORT_SYMBOL(generic_perform_write
);
3965 * __generic_file_write_iter - write data to a file
3966 * @iocb: IO state structure (file, offset, etc.)
3967 * @from: iov_iter with data to write
3969 * This function does all the work needed for actually writing data to a
3970 * file. It does all basic checks, removes SUID from the file, updates
3971 * modification times and calls proper subroutines depending on whether we
3972 * do direct IO or a standard buffered write.
3974 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3975 * object which does not need locking at all.
3977 * This function does *not* take care of syncing data in case of O_SYNC write.
3978 * A caller has to handle it. This is mainly due to the fact that we want to
3979 * avoid syncing under i_rwsem.
3982 * * number of bytes written, even for truncated writes
3983 * * negative error code if no data has been written at all
3985 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3987 struct file
*file
= iocb
->ki_filp
;
3988 struct address_space
*mapping
= file
->f_mapping
;
3989 struct inode
*inode
= mapping
->host
;
3990 ssize_t written
= 0;
3994 /* We can write back this queue in page reclaim */
3995 current
->backing_dev_info
= inode_to_bdi(inode
);
3996 err
= file_remove_privs(file
);
4000 err
= file_update_time(file
);
4004 if (iocb
->ki_flags
& IOCB_DIRECT
) {
4005 loff_t pos
, endbyte
;
4007 written
= generic_file_direct_write(iocb
, from
);
4009 * If the write stopped short of completing, fall back to
4010 * buffered writes. Some filesystems do this for writes to
4011 * holes, for example. For DAX files, a buffered write will
4012 * not succeed (even if it did, DAX does not handle dirty
4013 * page-cache pages correctly).
4015 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
4019 status
= generic_perform_write(iocb
, from
);
4021 * If generic_perform_write() returned a synchronous error
4022 * then we want to return the number of bytes which were
4023 * direct-written, or the error code if that was zero. Note
4024 * that this differs from normal direct-io semantics, which
4025 * will return -EFOO even if some bytes were written.
4027 if (unlikely(status
< 0)) {
4032 * We need to ensure that the page cache pages are written to
4033 * disk and invalidated to preserve the expected O_DIRECT
4036 endbyte
= pos
+ status
- 1;
4037 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
4039 iocb
->ki_pos
= endbyte
+ 1;
4041 invalidate_mapping_pages(mapping
,
4043 endbyte
>> PAGE_SHIFT
);
4046 * We don't know how much we wrote, so just return
4047 * the number of bytes which were direct-written
4051 written
= generic_perform_write(iocb
, from
);
4052 if (likely(written
> 0))
4053 iocb
->ki_pos
+= written
;
4056 current
->backing_dev_info
= NULL
;
4057 return written
? written
: err
;
4059 EXPORT_SYMBOL(__generic_file_write_iter
);
4062 * generic_file_write_iter - write data to a file
4063 * @iocb: IO state structure
4064 * @from: iov_iter with data to write
4066 * This is a wrapper around __generic_file_write_iter() to be used by most
4067 * filesystems. It takes care of syncing the file in case of O_SYNC file
4068 * and acquires i_rwsem as needed.
4070 * * negative error code if no data has been written at all of
4071 * vfs_fsync_range() failed for a synchronous write
4072 * * number of bytes written, even for truncated writes
4074 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
4076 struct file
*file
= iocb
->ki_filp
;
4077 struct inode
*inode
= file
->f_mapping
->host
;
4081 ret
= generic_write_checks(iocb
, from
);
4083 ret
= __generic_file_write_iter(iocb
, from
);
4084 inode_unlock(inode
);
4087 ret
= generic_write_sync(iocb
, ret
);
4090 EXPORT_SYMBOL(generic_file_write_iter
);
4093 * filemap_release_folio() - Release fs-specific metadata on a folio.
4094 * @folio: The folio which the kernel is trying to free.
4095 * @gfp: Memory allocation flags (and I/O mode).
4097 * The address_space is trying to release any data attached to a folio
4098 * (presumably at folio->private).
4100 * This will also be called if the private_2 flag is set on a page,
4101 * indicating that the folio has other metadata associated with it.
4103 * The @gfp argument specifies whether I/O may be performed to release
4104 * this page (__GFP_IO), and whether the call may block
4105 * (__GFP_RECLAIM & __GFP_FS).
4107 * Return: %true if the release was successful, otherwise %false.
4109 bool filemap_release_folio(struct folio
*folio
, gfp_t gfp
)
4111 struct address_space
* const mapping
= folio
->mapping
;
4113 BUG_ON(!folio_test_locked(folio
));
4114 if (folio_test_writeback(folio
))
4117 if (mapping
&& mapping
->a_ops
->release_folio
)
4118 return mapping
->a_ops
->release_folio(folio
, gfp
);
4119 return try_to_free_buffers(folio
);
4121 EXPORT_SYMBOL(filemap_release_folio
);