Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
torvalds
GitHub Repository: torvalds/linux
Path: blob/master/mm/filemap.c
29501 views
1
// SPDX-License-Identifier: GPL-2.0-only
2
/*
3
* linux/mm/filemap.c
4
*
5
* Copyright (C) 1994-1999 Linus Torvalds
6
*/
7
8
/*
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)
12
*/
13
#include <linux/export.h>
14
#include <linux/compiler.h>
15
#include <linux/dax.h>
16
#include <linux/fs.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>
22
#include <linux/mm.h>
23
#include <linux/swap.h>
24
#include <linux/swapops.h>
25
#include <linux/syscalls.h>
26
#include <linux/mman.h>
27
#include <linux/pagemap.h>
28
#include <linux/file.h>
29
#include <linux/uio.h>
30
#include <linux/error-injection.h>
31
#include <linux/hash.h>
32
#include <linux/writeback.h>
33
#include <linux/backing-dev.h>
34
#include <linux/pagevec.h>
35
#include <linux/security.h>
36
#include <linux/cpuset.h>
37
#include <linux/hugetlb.h>
38
#include <linux/memcontrol.h>
39
#include <linux/shmem_fs.h>
40
#include <linux/rmap.h>
41
#include <linux/delayacct.h>
42
#include <linux/psi.h>
43
#include <linux/ramfs.h>
44
#include <linux/page_idle.h>
45
#include <linux/migrate.h>
46
#include <linux/pipe_fs_i.h>
47
#include <linux/splice.h>
48
#include <linux/rcupdate_wait.h>
49
#include <linux/sched/mm.h>
50
#include <linux/sysctl.h>
51
#include <asm/pgalloc.h>
52
#include <asm/tlbflush.h>
53
#include "internal.h"
54
55
#define CREATE_TRACE_POINTS
56
#include <trace/events/filemap.h>
57
58
/*
59
* FIXME: remove all knowledge of the buffer layer from the core VM
60
*/
61
#include <linux/buffer_head.h> /* for try_to_free_buffers */
62
63
#include <asm/mman.h>
64
65
#include "swap.h"
66
67
/*
68
* Shared mappings implemented 30.11.1994. It's not fully working yet,
69
* though.
70
*
71
* Shared mappings now work. 15.8.1995 Bruno.
72
*
73
* finished 'unifying' the page and buffer cache and SMP-threaded the
74
* page-cache, 21.05.1999, Ingo Molnar <[email protected]>
75
*
76
* SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <[email protected]>
77
*/
78
79
/*
80
* Lock ordering:
81
*
82
* ->i_mmap_rwsem (truncate_pagecache)
83
* ->private_lock (__free_pte->block_dirty_folio)
84
* ->swap_lock (exclusive_swap_page, others)
85
* ->i_pages lock
86
*
87
* ->i_rwsem
88
* ->invalidate_lock (acquired by fs in truncate path)
89
* ->i_mmap_rwsem (truncate->unmap_mapping_range)
90
*
91
* ->mmap_lock
92
* ->i_mmap_rwsem
93
* ->page_table_lock or pte_lock (various, mainly in memory.c)
94
* ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
95
*
96
* ->mmap_lock
97
* ->invalidate_lock (filemap_fault)
98
* ->lock_page (filemap_fault, access_process_vm)
99
*
100
* ->i_rwsem (generic_perform_write)
101
* ->mmap_lock (fault_in_readable->do_page_fault)
102
*
103
* bdi->wb.list_lock
104
* sb_lock (fs/fs-writeback.c)
105
* ->i_pages lock (__sync_single_inode)
106
*
107
* ->i_mmap_rwsem
108
* ->anon_vma.lock (vma_merge)
109
*
110
* ->anon_vma.lock
111
* ->page_table_lock or pte_lock (anon_vma_prepare and various)
112
*
113
* ->page_table_lock or pte_lock
114
* ->swap_lock (try_to_unmap_one)
115
* ->private_lock (try_to_unmap_one)
116
* ->i_pages lock (try_to_unmap_one)
117
* ->lruvec->lru_lock (follow_page_mask->mark_page_accessed)
118
* ->lruvec->lru_lock (check_pte_range->folio_isolate_lru)
119
* ->private_lock (folio_remove_rmap_pte->set_page_dirty)
120
* ->i_pages lock (folio_remove_rmap_pte->set_page_dirty)
121
* bdi.wb->list_lock (folio_remove_rmap_pte->set_page_dirty)
122
* ->inode->i_lock (folio_remove_rmap_pte->set_page_dirty)
123
* bdi.wb->list_lock (zap_pte_range->set_page_dirty)
124
* ->inode->i_lock (zap_pte_range->set_page_dirty)
125
* ->private_lock (zap_pte_range->block_dirty_folio)
126
*/
127
128
static void page_cache_delete(struct address_space *mapping,
129
struct folio *folio, void *shadow)
130
{
131
XA_STATE(xas, &mapping->i_pages, folio->index);
132
long nr = 1;
133
134
mapping_set_update(&xas, mapping);
135
136
xas_set_order(&xas, folio->index, folio_order(folio));
137
nr = folio_nr_pages(folio);
138
139
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
140
141
xas_store(&xas, shadow);
142
xas_init_marks(&xas);
143
144
folio->mapping = NULL;
145
/* Leave folio->index set: truncation lookup relies upon it */
146
mapping->nrpages -= nr;
147
}
148
149
static void filemap_unaccount_folio(struct address_space *mapping,
150
struct folio *folio)
151
{
152
long nr;
153
154
VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
155
if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
156
pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
157
current->comm, folio_pfn(folio));
158
dump_page(&folio->page, "still mapped when deleted");
159
dump_stack();
160
add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
161
162
if (mapping_exiting(mapping) && !folio_test_large(folio)) {
163
int mapcount = folio_mapcount(folio);
164
165
if (folio_ref_count(folio) >= mapcount + 2) {
166
/*
167
* All vmas have already been torn down, so it's
168
* a good bet that actually the page is unmapped
169
* and we'd rather not leak it: if we're wrong,
170
* another bad page check should catch it later.
171
*/
172
atomic_set(&folio->_mapcount, -1);
173
folio_ref_sub(folio, mapcount);
174
}
175
}
176
}
177
178
/* hugetlb folios do not participate in page cache accounting. */
179
if (folio_test_hugetlb(folio))
180
return;
181
182
nr = folio_nr_pages(folio);
183
184
__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
185
if (folio_test_swapbacked(folio)) {
186
__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
187
if (folio_test_pmd_mappable(folio))
188
__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
189
} else if (folio_test_pmd_mappable(folio)) {
190
__lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
191
filemap_nr_thps_dec(mapping);
192
}
193
if (test_bit(AS_KERNEL_FILE, &folio->mapping->flags))
194
mod_node_page_state(folio_pgdat(folio),
195
NR_KERNEL_FILE_PAGES, -nr);
196
197
/*
198
* At this point folio must be either written or cleaned by
199
* truncate. Dirty folio here signals a bug and loss of
200
* unwritten data - on ordinary filesystems.
201
*
202
* But it's harmless on in-memory filesystems like tmpfs; and can
203
* occur when a driver which did get_user_pages() sets page dirty
204
* before putting it, while the inode is being finally evicted.
205
*
206
* Below fixes dirty accounting after removing the folio entirely
207
* but leaves the dirty flag set: it has no effect for truncated
208
* folio and anyway will be cleared before returning folio to
209
* buddy allocator.
210
*/
211
if (WARN_ON_ONCE(folio_test_dirty(folio) &&
212
mapping_can_writeback(mapping)))
213
folio_account_cleaned(folio, inode_to_wb(mapping->host));
214
}
215
216
/*
217
* Delete a page from the page cache and free it. Caller has to make
218
* sure the page is locked and that nobody else uses it - or that usage
219
* is safe. The caller must hold the i_pages lock.
220
*/
221
void __filemap_remove_folio(struct folio *folio, void *shadow)
222
{
223
struct address_space *mapping = folio->mapping;
224
225
trace_mm_filemap_delete_from_page_cache(folio);
226
filemap_unaccount_folio(mapping, folio);
227
page_cache_delete(mapping, folio, shadow);
228
}
229
230
void filemap_free_folio(struct address_space *mapping, struct folio *folio)
231
{
232
void (*free_folio)(struct folio *);
233
234
free_folio = mapping->a_ops->free_folio;
235
if (free_folio)
236
free_folio(folio);
237
238
folio_put_refs(folio, folio_nr_pages(folio));
239
}
240
241
/**
242
* filemap_remove_folio - Remove folio from page cache.
243
* @folio: The folio.
244
*
245
* This must be called only on folios that are locked and have been
246
* verified to be in the page cache. It will never put the folio into
247
* the free list because the caller has a reference on the page.
248
*/
249
void filemap_remove_folio(struct folio *folio)
250
{
251
struct address_space *mapping = folio->mapping;
252
253
BUG_ON(!folio_test_locked(folio));
254
spin_lock(&mapping->host->i_lock);
255
xa_lock_irq(&mapping->i_pages);
256
__filemap_remove_folio(folio, NULL);
257
xa_unlock_irq(&mapping->i_pages);
258
if (mapping_shrinkable(mapping))
259
inode_add_lru(mapping->host);
260
spin_unlock(&mapping->host->i_lock);
261
262
filemap_free_folio(mapping, folio);
263
}
264
265
/*
266
* page_cache_delete_batch - delete several folios from page cache
267
* @mapping: the mapping to which folios belong
268
* @fbatch: batch of folios to delete
269
*
270
* The function walks over mapping->i_pages and removes folios passed in
271
* @fbatch from the mapping. The function expects @fbatch to be sorted
272
* by page index and is optimised for it to be dense.
273
* It tolerates holes in @fbatch (mapping entries at those indices are not
274
* modified).
275
*
276
* The function expects the i_pages lock to be held.
277
*/
278
static void page_cache_delete_batch(struct address_space *mapping,
279
struct folio_batch *fbatch)
280
{
281
XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
282
long total_pages = 0;
283
int i = 0;
284
struct folio *folio;
285
286
mapping_set_update(&xas, mapping);
287
xas_for_each(&xas, folio, ULONG_MAX) {
288
if (i >= folio_batch_count(fbatch))
289
break;
290
291
/* A swap/dax/shadow entry got inserted? Skip it. */
292
if (xa_is_value(folio))
293
continue;
294
/*
295
* A page got inserted in our range? Skip it. We have our
296
* pages locked so they are protected from being removed.
297
* If we see a page whose index is higher than ours, it
298
* means our page has been removed, which shouldn't be
299
* possible because we're holding the PageLock.
300
*/
301
if (folio != fbatch->folios[i]) {
302
VM_BUG_ON_FOLIO(folio->index >
303
fbatch->folios[i]->index, folio);
304
continue;
305
}
306
307
WARN_ON_ONCE(!folio_test_locked(folio));
308
309
folio->mapping = NULL;
310
/* Leave folio->index set: truncation lookup relies on it */
311
312
i++;
313
xas_store(&xas, NULL);
314
total_pages += folio_nr_pages(folio);
315
}
316
mapping->nrpages -= total_pages;
317
}
318
319
void delete_from_page_cache_batch(struct address_space *mapping,
320
struct folio_batch *fbatch)
321
{
322
int i;
323
324
if (!folio_batch_count(fbatch))
325
return;
326
327
spin_lock(&mapping->host->i_lock);
328
xa_lock_irq(&mapping->i_pages);
329
for (i = 0; i < folio_batch_count(fbatch); i++) {
330
struct folio *folio = fbatch->folios[i];
331
332
trace_mm_filemap_delete_from_page_cache(folio);
333
filemap_unaccount_folio(mapping, folio);
334
}
335
page_cache_delete_batch(mapping, fbatch);
336
xa_unlock_irq(&mapping->i_pages);
337
if (mapping_shrinkable(mapping))
338
inode_add_lru(mapping->host);
339
spin_unlock(&mapping->host->i_lock);
340
341
for (i = 0; i < folio_batch_count(fbatch); i++)
342
filemap_free_folio(mapping, fbatch->folios[i]);
343
}
344
345
int filemap_check_errors(struct address_space *mapping)
346
{
347
int ret = 0;
348
/* Check for outstanding write errors */
349
if (test_bit(AS_ENOSPC, &mapping->flags) &&
350
test_and_clear_bit(AS_ENOSPC, &mapping->flags))
351
ret = -ENOSPC;
352
if (test_bit(AS_EIO, &mapping->flags) &&
353
test_and_clear_bit(AS_EIO, &mapping->flags))
354
ret = -EIO;
355
return ret;
356
}
357
EXPORT_SYMBOL(filemap_check_errors);
358
359
static int filemap_check_and_keep_errors(struct address_space *mapping)
360
{
361
/* Check for outstanding write errors */
362
if (test_bit(AS_EIO, &mapping->flags))
363
return -EIO;
364
if (test_bit(AS_ENOSPC, &mapping->flags))
365
return -ENOSPC;
366
return 0;
367
}
368
369
/**
370
* filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
371
* @mapping: address space structure to write
372
* @wbc: the writeback_control controlling the writeout
373
*
374
* Call writepages on the mapping using the provided wbc to control the
375
* writeout.
376
*
377
* Return: %0 on success, negative error code otherwise.
378
*/
379
int filemap_fdatawrite_wbc(struct address_space *mapping,
380
struct writeback_control *wbc)
381
{
382
int ret;
383
384
if (!mapping_can_writeback(mapping) ||
385
!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
386
return 0;
387
388
wbc_attach_fdatawrite_inode(wbc, mapping->host);
389
ret = do_writepages(mapping, wbc);
390
wbc_detach_inode(wbc);
391
return ret;
392
}
393
EXPORT_SYMBOL(filemap_fdatawrite_wbc);
394
395
/**
396
* __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
397
* @mapping: address space structure to write
398
* @start: offset in bytes where the range starts
399
* @end: offset in bytes where the range ends (inclusive)
400
* @sync_mode: enable synchronous operation
401
*
402
* Start writeback against all of a mapping's dirty pages that lie
403
* within the byte offsets <start, end> inclusive.
404
*
405
* If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
406
* opposed to a regular memory cleansing writeback. The difference between
407
* these two operations is that if a dirty page/buffer is encountered, it must
408
* be waited upon, and not just skipped over.
409
*
410
* Return: %0 on success, negative error code otherwise.
411
*/
412
int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
413
loff_t end, int sync_mode)
414
{
415
struct writeback_control wbc = {
416
.sync_mode = sync_mode,
417
.nr_to_write = LONG_MAX,
418
.range_start = start,
419
.range_end = end,
420
};
421
422
return filemap_fdatawrite_wbc(mapping, &wbc);
423
}
424
425
static inline int __filemap_fdatawrite(struct address_space *mapping,
426
int sync_mode)
427
{
428
return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
429
}
430
431
int filemap_fdatawrite(struct address_space *mapping)
432
{
433
return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
434
}
435
EXPORT_SYMBOL(filemap_fdatawrite);
436
437
int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
438
loff_t end)
439
{
440
return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
441
}
442
EXPORT_SYMBOL(filemap_fdatawrite_range);
443
444
/**
445
* filemap_fdatawrite_range_kick - start writeback on a range
446
* @mapping: target address_space
447
* @start: index to start writeback on
448
* @end: last (inclusive) index for writeback
449
*
450
* This is a non-integrity writeback helper, to start writing back folios
451
* for the indicated range.
452
*
453
* Return: %0 on success, negative error code otherwise.
454
*/
455
int filemap_fdatawrite_range_kick(struct address_space *mapping, loff_t start,
456
loff_t end)
457
{
458
return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_NONE);
459
}
460
EXPORT_SYMBOL_GPL(filemap_fdatawrite_range_kick);
461
462
/**
463
* filemap_flush - mostly a non-blocking flush
464
* @mapping: target address_space
465
*
466
* This is a mostly non-blocking flush. Not suitable for data-integrity
467
* purposes - I/O may not be started against all dirty pages.
468
*
469
* Return: %0 on success, negative error code otherwise.
470
*/
471
int filemap_flush(struct address_space *mapping)
472
{
473
return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
474
}
475
EXPORT_SYMBOL(filemap_flush);
476
477
/**
478
* filemap_range_has_page - check if a page exists in range.
479
* @mapping: address space within which to check
480
* @start_byte: offset in bytes where the range starts
481
* @end_byte: offset in bytes where the range ends (inclusive)
482
*
483
* Find at least one page in the range supplied, usually used to check if
484
* direct writing in this range will trigger a writeback.
485
*
486
* Return: %true if at least one page exists in the specified range,
487
* %false otherwise.
488
*/
489
bool filemap_range_has_page(struct address_space *mapping,
490
loff_t start_byte, loff_t end_byte)
491
{
492
struct folio *folio;
493
XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
494
pgoff_t max = end_byte >> PAGE_SHIFT;
495
496
if (end_byte < start_byte)
497
return false;
498
499
rcu_read_lock();
500
for (;;) {
501
folio = xas_find(&xas, max);
502
if (xas_retry(&xas, folio))
503
continue;
504
/* Shadow entries don't count */
505
if (xa_is_value(folio))
506
continue;
507
/*
508
* We don't need to try to pin this page; we're about to
509
* release the RCU lock anyway. It is enough to know that
510
* there was a page here recently.
511
*/
512
break;
513
}
514
rcu_read_unlock();
515
516
return folio != NULL;
517
}
518
EXPORT_SYMBOL(filemap_range_has_page);
519
520
static void __filemap_fdatawait_range(struct address_space *mapping,
521
loff_t start_byte, loff_t end_byte)
522
{
523
pgoff_t index = start_byte >> PAGE_SHIFT;
524
pgoff_t end = end_byte >> PAGE_SHIFT;
525
struct folio_batch fbatch;
526
unsigned nr_folios;
527
528
folio_batch_init(&fbatch);
529
530
while (index <= end) {
531
unsigned i;
532
533
nr_folios = filemap_get_folios_tag(mapping, &index, end,
534
PAGECACHE_TAG_WRITEBACK, &fbatch);
535
536
if (!nr_folios)
537
break;
538
539
for (i = 0; i < nr_folios; i++) {
540
struct folio *folio = fbatch.folios[i];
541
542
folio_wait_writeback(folio);
543
}
544
folio_batch_release(&fbatch);
545
cond_resched();
546
}
547
}
548
549
/**
550
* filemap_fdatawait_range - wait for writeback to complete
551
* @mapping: address space structure to wait for
552
* @start_byte: offset in bytes where the range starts
553
* @end_byte: offset in bytes where the range ends (inclusive)
554
*
555
* Walk the list of under-writeback pages of the given address space
556
* in the given range and wait for all of them. Check error status of
557
* the address space and return it.
558
*
559
* Since the error status of the address space is cleared by this function,
560
* callers are responsible for checking the return value and handling and/or
561
* reporting the error.
562
*
563
* Return: error status of the address space.
564
*/
565
int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
566
loff_t end_byte)
567
{
568
__filemap_fdatawait_range(mapping, start_byte, end_byte);
569
return filemap_check_errors(mapping);
570
}
571
EXPORT_SYMBOL(filemap_fdatawait_range);
572
573
/**
574
* filemap_fdatawait_range_keep_errors - wait for writeback to complete
575
* @mapping: address space structure to wait for
576
* @start_byte: offset in bytes where the range starts
577
* @end_byte: offset in bytes where the range ends (inclusive)
578
*
579
* Walk the list of under-writeback pages of the given address space in the
580
* given range and wait for all of them. Unlike filemap_fdatawait_range(),
581
* this function does not clear error status of the address space.
582
*
583
* Use this function if callers don't handle errors themselves. Expected
584
* call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
585
* fsfreeze(8)
586
*/
587
int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
588
loff_t start_byte, loff_t end_byte)
589
{
590
__filemap_fdatawait_range(mapping, start_byte, end_byte);
591
return filemap_check_and_keep_errors(mapping);
592
}
593
EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
594
595
/**
596
* file_fdatawait_range - wait for writeback to complete
597
* @file: file pointing to address space structure to wait for
598
* @start_byte: offset in bytes where the range starts
599
* @end_byte: offset in bytes where the range ends (inclusive)
600
*
601
* Walk the list of under-writeback pages of the address space that file
602
* refers to, in the given range and wait for all of them. Check error
603
* status of the address space vs. the file->f_wb_err cursor and return it.
604
*
605
* Since the error status of the file is advanced by this function,
606
* callers are responsible for checking the return value and handling and/or
607
* reporting the error.
608
*
609
* Return: error status of the address space vs. the file->f_wb_err cursor.
610
*/
611
int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
612
{
613
struct address_space *mapping = file->f_mapping;
614
615
__filemap_fdatawait_range(mapping, start_byte, end_byte);
616
return file_check_and_advance_wb_err(file);
617
}
618
EXPORT_SYMBOL(file_fdatawait_range);
619
620
/**
621
* filemap_fdatawait_keep_errors - wait for writeback without clearing errors
622
* @mapping: address space structure to wait for
623
*
624
* Walk the list of under-writeback pages of the given address space
625
* and wait for all of them. Unlike filemap_fdatawait(), this function
626
* does not clear error status of the address space.
627
*
628
* Use this function if callers don't handle errors themselves. Expected
629
* call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
630
* fsfreeze(8)
631
*
632
* Return: error status of the address space.
633
*/
634
int filemap_fdatawait_keep_errors(struct address_space *mapping)
635
{
636
__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
637
return filemap_check_and_keep_errors(mapping);
638
}
639
EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
640
641
/* Returns true if writeback might be needed or already in progress. */
642
static bool mapping_needs_writeback(struct address_space *mapping)
643
{
644
return mapping->nrpages;
645
}
646
647
bool filemap_range_has_writeback(struct address_space *mapping,
648
loff_t start_byte, loff_t end_byte)
649
{
650
XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
651
pgoff_t max = end_byte >> PAGE_SHIFT;
652
struct folio *folio;
653
654
if (end_byte < start_byte)
655
return false;
656
657
rcu_read_lock();
658
xas_for_each(&xas, folio, max) {
659
if (xas_retry(&xas, folio))
660
continue;
661
if (xa_is_value(folio))
662
continue;
663
if (folio_test_dirty(folio) || folio_test_locked(folio) ||
664
folio_test_writeback(folio))
665
break;
666
}
667
rcu_read_unlock();
668
return folio != NULL;
669
}
670
EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
671
672
/**
673
* filemap_write_and_wait_range - write out & wait on a file range
674
* @mapping: the address_space for the pages
675
* @lstart: offset in bytes where the range starts
676
* @lend: offset in bytes where the range ends (inclusive)
677
*
678
* Write out and wait upon file offsets lstart->lend, inclusive.
679
*
680
* Note that @lend is inclusive (describes the last byte to be written) so
681
* that this function can be used to write to the very end-of-file (end = -1).
682
*
683
* Return: error status of the address space.
684
*/
685
int filemap_write_and_wait_range(struct address_space *mapping,
686
loff_t lstart, loff_t lend)
687
{
688
int err = 0, err2;
689
690
if (lend < lstart)
691
return 0;
692
693
if (mapping_needs_writeback(mapping)) {
694
err = __filemap_fdatawrite_range(mapping, lstart, lend,
695
WB_SYNC_ALL);
696
/*
697
* Even if the above returned error, the pages may be
698
* written partially (e.g. -ENOSPC), so we wait for it.
699
* But the -EIO is special case, it may indicate the worst
700
* thing (e.g. bug) happened, so we avoid waiting for it.
701
*/
702
if (err != -EIO)
703
__filemap_fdatawait_range(mapping, lstart, lend);
704
}
705
err2 = filemap_check_errors(mapping);
706
if (!err)
707
err = err2;
708
return err;
709
}
710
EXPORT_SYMBOL(filemap_write_and_wait_range);
711
712
void __filemap_set_wb_err(struct address_space *mapping, int err)
713
{
714
errseq_t eseq = errseq_set(&mapping->wb_err, err);
715
716
trace_filemap_set_wb_err(mapping, eseq);
717
}
718
EXPORT_SYMBOL(__filemap_set_wb_err);
719
720
/**
721
* file_check_and_advance_wb_err - report wb error (if any) that was previously
722
* and advance wb_err to current one
723
* @file: struct file on which the error is being reported
724
*
725
* When userland calls fsync (or something like nfsd does the equivalent), we
726
* want to report any writeback errors that occurred since the last fsync (or
727
* since the file was opened if there haven't been any).
728
*
729
* Grab the wb_err from the mapping. If it matches what we have in the file,
730
* then just quickly return 0. The file is all caught up.
731
*
732
* If it doesn't match, then take the mapping value, set the "seen" flag in
733
* it and try to swap it into place. If it works, or another task beat us
734
* to it with the new value, then update the f_wb_err and return the error
735
* portion. The error at this point must be reported via proper channels
736
* (a'la fsync, or NFS COMMIT operation, etc.).
737
*
738
* While we handle mapping->wb_err with atomic operations, the f_wb_err
739
* value is protected by the f_lock since we must ensure that it reflects
740
* the latest value swapped in for this file descriptor.
741
*
742
* Return: %0 on success, negative error code otherwise.
743
*/
744
int file_check_and_advance_wb_err(struct file *file)
745
{
746
int err = 0;
747
errseq_t old = READ_ONCE(file->f_wb_err);
748
struct address_space *mapping = file->f_mapping;
749
750
/* Locklessly handle the common case where nothing has changed */
751
if (errseq_check(&mapping->wb_err, old)) {
752
/* Something changed, must use slow path */
753
spin_lock(&file->f_lock);
754
old = file->f_wb_err;
755
err = errseq_check_and_advance(&mapping->wb_err,
756
&file->f_wb_err);
757
trace_file_check_and_advance_wb_err(file, old);
758
spin_unlock(&file->f_lock);
759
}
760
761
/*
762
* We're mostly using this function as a drop in replacement for
763
* filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
764
* that the legacy code would have had on these flags.
765
*/
766
clear_bit(AS_EIO, &mapping->flags);
767
clear_bit(AS_ENOSPC, &mapping->flags);
768
return err;
769
}
770
EXPORT_SYMBOL(file_check_and_advance_wb_err);
771
772
/**
773
* file_write_and_wait_range - write out & wait on a file range
774
* @file: file pointing to address_space with pages
775
* @lstart: offset in bytes where the range starts
776
* @lend: offset in bytes where the range ends (inclusive)
777
*
778
* Write out and wait upon file offsets lstart->lend, inclusive.
779
*
780
* Note that @lend is inclusive (describes the last byte to be written) so
781
* that this function can be used to write to the very end-of-file (end = -1).
782
*
783
* After writing out and waiting on the data, we check and advance the
784
* f_wb_err cursor to the latest value, and return any errors detected there.
785
*
786
* Return: %0 on success, negative error code otherwise.
787
*/
788
int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
789
{
790
int err = 0, err2;
791
struct address_space *mapping = file->f_mapping;
792
793
if (lend < lstart)
794
return 0;
795
796
if (mapping_needs_writeback(mapping)) {
797
err = __filemap_fdatawrite_range(mapping, lstart, lend,
798
WB_SYNC_ALL);
799
/* See comment of filemap_write_and_wait() */
800
if (err != -EIO)
801
__filemap_fdatawait_range(mapping, lstart, lend);
802
}
803
err2 = file_check_and_advance_wb_err(file);
804
if (!err)
805
err = err2;
806
return err;
807
}
808
EXPORT_SYMBOL(file_write_and_wait_range);
809
810
/**
811
* replace_page_cache_folio - replace a pagecache folio with a new one
812
* @old: folio to be replaced
813
* @new: folio to replace with
814
*
815
* This function replaces a folio in the pagecache with a new one. On
816
* success it acquires the pagecache reference for the new folio and
817
* drops it for the old folio. Both the old and new folios must be
818
* locked. This function does not add the new folio to the LRU, the
819
* caller must do that.
820
*
821
* The remove + add is atomic. This function cannot fail.
822
*/
823
void replace_page_cache_folio(struct folio *old, struct folio *new)
824
{
825
struct address_space *mapping = old->mapping;
826
void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
827
pgoff_t offset = old->index;
828
XA_STATE(xas, &mapping->i_pages, offset);
829
830
VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
831
VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
832
VM_BUG_ON_FOLIO(new->mapping, new);
833
834
folio_get(new);
835
new->mapping = mapping;
836
new->index = offset;
837
838
mem_cgroup_replace_folio(old, new);
839
840
xas_lock_irq(&xas);
841
xas_store(&xas, new);
842
843
old->mapping = NULL;
844
/* hugetlb pages do not participate in page cache accounting. */
845
if (!folio_test_hugetlb(old))
846
__lruvec_stat_sub_folio(old, NR_FILE_PAGES);
847
if (!folio_test_hugetlb(new))
848
__lruvec_stat_add_folio(new, NR_FILE_PAGES);
849
if (folio_test_swapbacked(old))
850
__lruvec_stat_sub_folio(old, NR_SHMEM);
851
if (folio_test_swapbacked(new))
852
__lruvec_stat_add_folio(new, NR_SHMEM);
853
xas_unlock_irq(&xas);
854
if (free_folio)
855
free_folio(old);
856
folio_put(old);
857
}
858
EXPORT_SYMBOL_GPL(replace_page_cache_folio);
859
860
noinline int __filemap_add_folio(struct address_space *mapping,
861
struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
862
{
863
XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
864
bool huge;
865
long nr;
866
unsigned int forder = folio_order(folio);
867
868
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
869
VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
870
VM_BUG_ON_FOLIO(folio_order(folio) < mapping_min_folio_order(mapping),
871
folio);
872
mapping_set_update(&xas, mapping);
873
874
VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
875
huge = folio_test_hugetlb(folio);
876
nr = folio_nr_pages(folio);
877
878
gfp &= GFP_RECLAIM_MASK;
879
folio_ref_add(folio, nr);
880
folio->mapping = mapping;
881
folio->index = xas.xa_index;
882
883
for (;;) {
884
int order = -1;
885
void *entry, *old = NULL;
886
887
xas_lock_irq(&xas);
888
xas_for_each_conflict(&xas, entry) {
889
old = entry;
890
if (!xa_is_value(entry)) {
891
xas_set_err(&xas, -EEXIST);
892
goto unlock;
893
}
894
/*
895
* If a larger entry exists,
896
* it will be the first and only entry iterated.
897
*/
898
if (order == -1)
899
order = xas_get_order(&xas);
900
}
901
902
if (old) {
903
if (order > 0 && order > forder) {
904
unsigned int split_order = max(forder,
905
xas_try_split_min_order(order));
906
907
/* How to handle large swap entries? */
908
BUG_ON(shmem_mapping(mapping));
909
910
while (order > forder) {
911
xas_set_order(&xas, index, split_order);
912
xas_try_split(&xas, old, order);
913
if (xas_error(&xas))
914
goto unlock;
915
order = split_order;
916
split_order =
917
max(xas_try_split_min_order(
918
split_order),
919
forder);
920
}
921
xas_reset(&xas);
922
}
923
if (shadowp)
924
*shadowp = old;
925
}
926
927
xas_store(&xas, folio);
928
if (xas_error(&xas))
929
goto unlock;
930
931
mapping->nrpages += nr;
932
933
/* hugetlb pages do not participate in page cache accounting */
934
if (!huge) {
935
__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
936
if (folio_test_pmd_mappable(folio))
937
__lruvec_stat_mod_folio(folio,
938
NR_FILE_THPS, nr);
939
}
940
941
unlock:
942
xas_unlock_irq(&xas);
943
944
if (!xas_nomem(&xas, gfp))
945
break;
946
}
947
948
if (xas_error(&xas))
949
goto error;
950
951
trace_mm_filemap_add_to_page_cache(folio);
952
return 0;
953
error:
954
folio->mapping = NULL;
955
/* Leave folio->index set: truncation relies upon it */
956
folio_put_refs(folio, nr);
957
return xas_error(&xas);
958
}
959
ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
960
961
int filemap_add_folio(struct address_space *mapping, struct folio *folio,
962
pgoff_t index, gfp_t gfp)
963
{
964
void *shadow = NULL;
965
int ret;
966
struct mem_cgroup *tmp;
967
bool kernel_file = test_bit(AS_KERNEL_FILE, &mapping->flags);
968
969
if (kernel_file)
970
tmp = set_active_memcg(root_mem_cgroup);
971
ret = mem_cgroup_charge(folio, NULL, gfp);
972
if (kernel_file)
973
set_active_memcg(tmp);
974
if (ret)
975
return ret;
976
977
__folio_set_locked(folio);
978
ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
979
if (unlikely(ret)) {
980
mem_cgroup_uncharge(folio);
981
__folio_clear_locked(folio);
982
} else {
983
/*
984
* The folio might have been evicted from cache only
985
* recently, in which case it should be activated like
986
* any other repeatedly accessed folio.
987
* The exception is folios getting rewritten; evicting other
988
* data from the working set, only to cache data that will
989
* get overwritten with something else, is a waste of memory.
990
*/
991
WARN_ON_ONCE(folio_test_active(folio));
992
if (!(gfp & __GFP_WRITE) && shadow)
993
workingset_refault(folio, shadow);
994
folio_add_lru(folio);
995
if (kernel_file)
996
mod_node_page_state(folio_pgdat(folio),
997
NR_KERNEL_FILE_PAGES,
998
folio_nr_pages(folio));
999
}
1000
return ret;
1001
}
1002
EXPORT_SYMBOL_GPL(filemap_add_folio);
1003
1004
#ifdef CONFIG_NUMA
1005
struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order)
1006
{
1007
int n;
1008
struct folio *folio;
1009
1010
if (cpuset_do_page_mem_spread()) {
1011
unsigned int cpuset_mems_cookie;
1012
do {
1013
cpuset_mems_cookie = read_mems_allowed_begin();
1014
n = cpuset_mem_spread_node();
1015
folio = __folio_alloc_node_noprof(gfp, order, n);
1016
} while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
1017
1018
return folio;
1019
}
1020
return folio_alloc_noprof(gfp, order);
1021
}
1022
EXPORT_SYMBOL(filemap_alloc_folio_noprof);
1023
#endif
1024
1025
/*
1026
* filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1027
*
1028
* Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1029
*
1030
* @mapping1: the first mapping to lock
1031
* @mapping2: the second mapping to lock
1032
*/
1033
void filemap_invalidate_lock_two(struct address_space *mapping1,
1034
struct address_space *mapping2)
1035
{
1036
if (mapping1 > mapping2)
1037
swap(mapping1, mapping2);
1038
if (mapping1)
1039
down_write(&mapping1->invalidate_lock);
1040
if (mapping2 && mapping1 != mapping2)
1041
down_write_nested(&mapping2->invalidate_lock, 1);
1042
}
1043
EXPORT_SYMBOL(filemap_invalidate_lock_two);
1044
1045
/*
1046
* filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1047
*
1048
* Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1049
*
1050
* @mapping1: the first mapping to unlock
1051
* @mapping2: the second mapping to unlock
1052
*/
1053
void filemap_invalidate_unlock_two(struct address_space *mapping1,
1054
struct address_space *mapping2)
1055
{
1056
if (mapping1)
1057
up_write(&mapping1->invalidate_lock);
1058
if (mapping2 && mapping1 != mapping2)
1059
up_write(&mapping2->invalidate_lock);
1060
}
1061
EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1062
1063
/*
1064
* In order to wait for pages to become available there must be
1065
* waitqueues associated with pages. By using a hash table of
1066
* waitqueues where the bucket discipline is to maintain all
1067
* waiters on the same queue and wake all when any of the pages
1068
* become available, and for the woken contexts to check to be
1069
* sure the appropriate page became available, this saves space
1070
* at a cost of "thundering herd" phenomena during rare hash
1071
* collisions.
1072
*/
1073
#define PAGE_WAIT_TABLE_BITS 8
1074
#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1075
static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1076
1077
static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1078
{
1079
return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1080
}
1081
1082
/* How many times do we accept lock stealing from under a waiter? */
1083
static int sysctl_page_lock_unfairness = 5;
1084
static const struct ctl_table filemap_sysctl_table[] = {
1085
{
1086
.procname = "page_lock_unfairness",
1087
.data = &sysctl_page_lock_unfairness,
1088
.maxlen = sizeof(sysctl_page_lock_unfairness),
1089
.mode = 0644,
1090
.proc_handler = proc_dointvec_minmax,
1091
.extra1 = SYSCTL_ZERO,
1092
}
1093
};
1094
1095
void __init pagecache_init(void)
1096
{
1097
int i;
1098
1099
for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1100
init_waitqueue_head(&folio_wait_table[i]);
1101
1102
page_writeback_init();
1103
register_sysctl_init("vm", filemap_sysctl_table);
1104
}
1105
1106
/*
1107
* The page wait code treats the "wait->flags" somewhat unusually, because
1108
* we have multiple different kinds of waits, not just the usual "exclusive"
1109
* one.
1110
*
1111
* We have:
1112
*
1113
* (a) no special bits set:
1114
*
1115
* We're just waiting for the bit to be released, and when a waker
1116
* calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1117
* and remove it from the wait queue.
1118
*
1119
* Simple and straightforward.
1120
*
1121
* (b) WQ_FLAG_EXCLUSIVE:
1122
*
1123
* The waiter is waiting to get the lock, and only one waiter should
1124
* be woken up to avoid any thundering herd behavior. We'll set the
1125
* WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1126
*
1127
* This is the traditional exclusive wait.
1128
*
1129
* (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1130
*
1131
* The waiter is waiting to get the bit, and additionally wants the
1132
* lock to be transferred to it for fair lock behavior. If the lock
1133
* cannot be taken, we stop walking the wait queue without waking
1134
* the waiter.
1135
*
1136
* This is the "fair lock handoff" case, and in addition to setting
1137
* WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1138
* that it now has the lock.
1139
*/
1140
static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1141
{
1142
unsigned int flags;
1143
struct wait_page_key *key = arg;
1144
struct wait_page_queue *wait_page
1145
= container_of(wait, struct wait_page_queue, wait);
1146
1147
if (!wake_page_match(wait_page, key))
1148
return 0;
1149
1150
/*
1151
* If it's a lock handoff wait, we get the bit for it, and
1152
* stop walking (and do not wake it up) if we can't.
1153
*/
1154
flags = wait->flags;
1155
if (flags & WQ_FLAG_EXCLUSIVE) {
1156
if (test_bit(key->bit_nr, &key->folio->flags.f))
1157
return -1;
1158
if (flags & WQ_FLAG_CUSTOM) {
1159
if (test_and_set_bit(key->bit_nr, &key->folio->flags.f))
1160
return -1;
1161
flags |= WQ_FLAG_DONE;
1162
}
1163
}
1164
1165
/*
1166
* We are holding the wait-queue lock, but the waiter that
1167
* is waiting for this will be checking the flags without
1168
* any locking.
1169
*
1170
* So update the flags atomically, and wake up the waiter
1171
* afterwards to avoid any races. This store-release pairs
1172
* with the load-acquire in folio_wait_bit_common().
1173
*/
1174
smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1175
wake_up_state(wait->private, mode);
1176
1177
/*
1178
* Ok, we have successfully done what we're waiting for,
1179
* and we can unconditionally remove the wait entry.
1180
*
1181
* Note that this pairs with the "finish_wait()" in the
1182
* waiter, and has to be the absolute last thing we do.
1183
* After this list_del_init(&wait->entry) the wait entry
1184
* might be de-allocated and the process might even have
1185
* exited.
1186
*/
1187
list_del_init_careful(&wait->entry);
1188
return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1189
}
1190
1191
static void folio_wake_bit(struct folio *folio, int bit_nr)
1192
{
1193
wait_queue_head_t *q = folio_waitqueue(folio);
1194
struct wait_page_key key;
1195
unsigned long flags;
1196
1197
key.folio = folio;
1198
key.bit_nr = bit_nr;
1199
key.page_match = 0;
1200
1201
spin_lock_irqsave(&q->lock, flags);
1202
__wake_up_locked_key(q, TASK_NORMAL, &key);
1203
1204
/*
1205
* It's possible to miss clearing waiters here, when we woke our page
1206
* waiters, but the hashed waitqueue has waiters for other pages on it.
1207
* That's okay, it's a rare case. The next waker will clear it.
1208
*
1209
* Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1210
* other), the flag may be cleared in the course of freeing the page;
1211
* but that is not required for correctness.
1212
*/
1213
if (!waitqueue_active(q) || !key.page_match)
1214
folio_clear_waiters(folio);
1215
1216
spin_unlock_irqrestore(&q->lock, flags);
1217
}
1218
1219
/*
1220
* A choice of three behaviors for folio_wait_bit_common():
1221
*/
1222
enum behavior {
1223
EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1224
* __folio_lock() waiting on then setting PG_locked.
1225
*/
1226
SHARED, /* Hold ref to page and check the bit when woken, like
1227
* folio_wait_writeback() waiting on PG_writeback.
1228
*/
1229
DROP, /* Drop ref to page before wait, no check when woken,
1230
* like folio_put_wait_locked() on PG_locked.
1231
*/
1232
};
1233
1234
/*
1235
* Attempt to check (or get) the folio flag, and mark us done
1236
* if successful.
1237
*/
1238
static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1239
struct wait_queue_entry *wait)
1240
{
1241
if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1242
if (test_and_set_bit(bit_nr, &folio->flags.f))
1243
return false;
1244
} else if (test_bit(bit_nr, &folio->flags.f))
1245
return false;
1246
1247
wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1248
return true;
1249
}
1250
1251
static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1252
int state, enum behavior behavior)
1253
{
1254
wait_queue_head_t *q = folio_waitqueue(folio);
1255
int unfairness = sysctl_page_lock_unfairness;
1256
struct wait_page_queue wait_page;
1257
wait_queue_entry_t *wait = &wait_page.wait;
1258
bool thrashing = false;
1259
unsigned long pflags;
1260
bool in_thrashing;
1261
1262
if (bit_nr == PG_locked &&
1263
!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1264
delayacct_thrashing_start(&in_thrashing);
1265
psi_memstall_enter(&pflags);
1266
thrashing = true;
1267
}
1268
1269
init_wait(wait);
1270
wait->func = wake_page_function;
1271
wait_page.folio = folio;
1272
wait_page.bit_nr = bit_nr;
1273
1274
repeat:
1275
wait->flags = 0;
1276
if (behavior == EXCLUSIVE) {
1277
wait->flags = WQ_FLAG_EXCLUSIVE;
1278
if (--unfairness < 0)
1279
wait->flags |= WQ_FLAG_CUSTOM;
1280
}
1281
1282
/*
1283
* Do one last check whether we can get the
1284
* page bit synchronously.
1285
*
1286
* Do the folio_set_waiters() marking before that
1287
* to let any waker we _just_ missed know they
1288
* need to wake us up (otherwise they'll never
1289
* even go to the slow case that looks at the
1290
* page queue), and add ourselves to the wait
1291
* queue if we need to sleep.
1292
*
1293
* This part needs to be done under the queue
1294
* lock to avoid races.
1295
*/
1296
spin_lock_irq(&q->lock);
1297
folio_set_waiters(folio);
1298
if (!folio_trylock_flag(folio, bit_nr, wait))
1299
__add_wait_queue_entry_tail(q, wait);
1300
spin_unlock_irq(&q->lock);
1301
1302
/*
1303
* From now on, all the logic will be based on
1304
* the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1305
* see whether the page bit testing has already
1306
* been done by the wake function.
1307
*
1308
* We can drop our reference to the folio.
1309
*/
1310
if (behavior == DROP)
1311
folio_put(folio);
1312
1313
/*
1314
* Note that until the "finish_wait()", or until
1315
* we see the WQ_FLAG_WOKEN flag, we need to
1316
* be very careful with the 'wait->flags', because
1317
* we may race with a waker that sets them.
1318
*/
1319
for (;;) {
1320
unsigned int flags;
1321
1322
set_current_state(state);
1323
1324
/* Loop until we've been woken or interrupted */
1325
flags = smp_load_acquire(&wait->flags);
1326
if (!(flags & WQ_FLAG_WOKEN)) {
1327
if (signal_pending_state(state, current))
1328
break;
1329
1330
io_schedule();
1331
continue;
1332
}
1333
1334
/* If we were non-exclusive, we're done */
1335
if (behavior != EXCLUSIVE)
1336
break;
1337
1338
/* If the waker got the lock for us, we're done */
1339
if (flags & WQ_FLAG_DONE)
1340
break;
1341
1342
/*
1343
* Otherwise, if we're getting the lock, we need to
1344
* try to get it ourselves.
1345
*
1346
* And if that fails, we'll have to retry this all.
1347
*/
1348
if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1349
goto repeat;
1350
1351
wait->flags |= WQ_FLAG_DONE;
1352
break;
1353
}
1354
1355
/*
1356
* If a signal happened, this 'finish_wait()' may remove the last
1357
* waiter from the wait-queues, but the folio waiters bit will remain
1358
* set. That's ok. The next wakeup will take care of it, and trying
1359
* to do it here would be difficult and prone to races.
1360
*/
1361
finish_wait(q, wait);
1362
1363
if (thrashing) {
1364
delayacct_thrashing_end(&in_thrashing);
1365
psi_memstall_leave(&pflags);
1366
}
1367
1368
/*
1369
* NOTE! The wait->flags weren't stable until we've done the
1370
* 'finish_wait()', and we could have exited the loop above due
1371
* to a signal, and had a wakeup event happen after the signal
1372
* test but before the 'finish_wait()'.
1373
*
1374
* So only after the finish_wait() can we reliably determine
1375
* if we got woken up or not, so we can now figure out the final
1376
* return value based on that state without races.
1377
*
1378
* Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1379
* waiter, but an exclusive one requires WQ_FLAG_DONE.
1380
*/
1381
if (behavior == EXCLUSIVE)
1382
return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1383
1384
return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1385
}
1386
1387
#ifdef CONFIG_MIGRATION
1388
/**
1389
* migration_entry_wait_on_locked - Wait for a migration entry to be removed
1390
* @entry: migration swap entry.
1391
* @ptl: already locked ptl. This function will drop the lock.
1392
*
1393
* Wait for a migration entry referencing the given page to be removed. This is
1394
* equivalent to folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE) except
1395
* this can be called without taking a reference on the page. Instead this
1396
* should be called while holding the ptl for the migration entry referencing
1397
* the page.
1398
*
1399
* Returns after unlocking the ptl.
1400
*
1401
* This follows the same logic as folio_wait_bit_common() so see the comments
1402
* there.
1403
*/
1404
void migration_entry_wait_on_locked(swp_entry_t entry, spinlock_t *ptl)
1405
__releases(ptl)
1406
{
1407
struct wait_page_queue wait_page;
1408
wait_queue_entry_t *wait = &wait_page.wait;
1409
bool thrashing = false;
1410
unsigned long pflags;
1411
bool in_thrashing;
1412
wait_queue_head_t *q;
1413
struct folio *folio = pfn_swap_entry_folio(entry);
1414
1415
q = folio_waitqueue(folio);
1416
if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1417
delayacct_thrashing_start(&in_thrashing);
1418
psi_memstall_enter(&pflags);
1419
thrashing = true;
1420
}
1421
1422
init_wait(wait);
1423
wait->func = wake_page_function;
1424
wait_page.folio = folio;
1425
wait_page.bit_nr = PG_locked;
1426
wait->flags = 0;
1427
1428
spin_lock_irq(&q->lock);
1429
folio_set_waiters(folio);
1430
if (!folio_trylock_flag(folio, PG_locked, wait))
1431
__add_wait_queue_entry_tail(q, wait);
1432
spin_unlock_irq(&q->lock);
1433
1434
/*
1435
* If a migration entry exists for the page the migration path must hold
1436
* a valid reference to the page, and it must take the ptl to remove the
1437
* migration entry. So the page is valid until the ptl is dropped.
1438
*/
1439
spin_unlock(ptl);
1440
1441
for (;;) {
1442
unsigned int flags;
1443
1444
set_current_state(TASK_UNINTERRUPTIBLE);
1445
1446
/* Loop until we've been woken or interrupted */
1447
flags = smp_load_acquire(&wait->flags);
1448
if (!(flags & WQ_FLAG_WOKEN)) {
1449
if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1450
break;
1451
1452
io_schedule();
1453
continue;
1454
}
1455
break;
1456
}
1457
1458
finish_wait(q, wait);
1459
1460
if (thrashing) {
1461
delayacct_thrashing_end(&in_thrashing);
1462
psi_memstall_leave(&pflags);
1463
}
1464
}
1465
#endif
1466
1467
void folio_wait_bit(struct folio *folio, int bit_nr)
1468
{
1469
folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1470
}
1471
EXPORT_SYMBOL(folio_wait_bit);
1472
1473
int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1474
{
1475
return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1476
}
1477
EXPORT_SYMBOL(folio_wait_bit_killable);
1478
1479
/**
1480
* folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1481
* @folio: The folio to wait for.
1482
* @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1483
*
1484
* The caller should hold a reference on @folio. They expect the page to
1485
* become unlocked relatively soon, but do not wish to hold up migration
1486
* (for example) by holding the reference while waiting for the folio to
1487
* come unlocked. After this function returns, the caller should not
1488
* dereference @folio.
1489
*
1490
* Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1491
*/
1492
static int folio_put_wait_locked(struct folio *folio, int state)
1493
{
1494
return folio_wait_bit_common(folio, PG_locked, state, DROP);
1495
}
1496
1497
/**
1498
* folio_unlock - Unlock a locked folio.
1499
* @folio: The folio.
1500
*
1501
* Unlocks the folio and wakes up any thread sleeping on the page lock.
1502
*
1503
* Context: May be called from interrupt or process context. May not be
1504
* called from NMI context.
1505
*/
1506
void folio_unlock(struct folio *folio)
1507
{
1508
/* Bit 7 allows x86 to check the byte's sign bit */
1509
BUILD_BUG_ON(PG_waiters != 7);
1510
BUILD_BUG_ON(PG_locked > 7);
1511
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1512
if (folio_xor_flags_has_waiters(folio, 1 << PG_locked))
1513
folio_wake_bit(folio, PG_locked);
1514
}
1515
EXPORT_SYMBOL(folio_unlock);
1516
1517
/**
1518
* folio_end_read - End read on a folio.
1519
* @folio: The folio.
1520
* @success: True if all reads completed successfully.
1521
*
1522
* When all reads against a folio have completed, filesystems should
1523
* call this function to let the pagecache know that no more reads
1524
* are outstanding. This will unlock the folio and wake up any thread
1525
* sleeping on the lock. The folio will also be marked uptodate if all
1526
* reads succeeded.
1527
*
1528
* Context: May be called from interrupt or process context. May not be
1529
* called from NMI context.
1530
*/
1531
void folio_end_read(struct folio *folio, bool success)
1532
{
1533
unsigned long mask = 1 << PG_locked;
1534
1535
/* Must be in bottom byte for x86 to work */
1536
BUILD_BUG_ON(PG_uptodate > 7);
1537
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1538
VM_BUG_ON_FOLIO(success && folio_test_uptodate(folio), folio);
1539
1540
if (likely(success))
1541
mask |= 1 << PG_uptodate;
1542
if (folio_xor_flags_has_waiters(folio, mask))
1543
folio_wake_bit(folio, PG_locked);
1544
}
1545
EXPORT_SYMBOL(folio_end_read);
1546
1547
/**
1548
* folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1549
* @folio: The folio.
1550
*
1551
* Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1552
* it. The folio reference held for PG_private_2 being set is released.
1553
*
1554
* This is, for example, used when a netfs folio is being written to a local
1555
* disk cache, thereby allowing writes to the cache for the same folio to be
1556
* serialised.
1557
*/
1558
void folio_end_private_2(struct folio *folio)
1559
{
1560
VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1561
clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1562
folio_wake_bit(folio, PG_private_2);
1563
folio_put(folio);
1564
}
1565
EXPORT_SYMBOL(folio_end_private_2);
1566
1567
/**
1568
* folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1569
* @folio: The folio to wait on.
1570
*
1571
* Wait for PG_private_2 to be cleared on a folio.
1572
*/
1573
void folio_wait_private_2(struct folio *folio)
1574
{
1575
while (folio_test_private_2(folio))
1576
folio_wait_bit(folio, PG_private_2);
1577
}
1578
EXPORT_SYMBOL(folio_wait_private_2);
1579
1580
/**
1581
* folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1582
* @folio: The folio to wait on.
1583
*
1584
* Wait for PG_private_2 to be cleared on a folio or until a fatal signal is
1585
* received by the calling task.
1586
*
1587
* Return:
1588
* - 0 if successful.
1589
* - -EINTR if a fatal signal was encountered.
1590
*/
1591
int folio_wait_private_2_killable(struct folio *folio)
1592
{
1593
int ret = 0;
1594
1595
while (folio_test_private_2(folio)) {
1596
ret = folio_wait_bit_killable(folio, PG_private_2);
1597
if (ret < 0)
1598
break;
1599
}
1600
1601
return ret;
1602
}
1603
EXPORT_SYMBOL(folio_wait_private_2_killable);
1604
1605
static void filemap_end_dropbehind(struct folio *folio)
1606
{
1607
struct address_space *mapping = folio->mapping;
1608
1609
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1610
1611
if (folio_test_writeback(folio) || folio_test_dirty(folio))
1612
return;
1613
if (!folio_test_clear_dropbehind(folio))
1614
return;
1615
if (mapping)
1616
folio_unmap_invalidate(mapping, folio, 0);
1617
}
1618
1619
/*
1620
* If folio was marked as dropbehind, then pages should be dropped when writeback
1621
* completes. Do that now. If we fail, it's likely because of a big folio -
1622
* just reset dropbehind for that case and latter completions should invalidate.
1623
*/
1624
void folio_end_dropbehind(struct folio *folio)
1625
{
1626
if (!folio_test_dropbehind(folio))
1627
return;
1628
1629
/*
1630
* Hitting !in_task() should not happen off RWF_DONTCACHE writeback,
1631
* but can happen if normal writeback just happens to find dirty folios
1632
* that were created as part of uncached writeback, and that writeback
1633
* would otherwise not need non-IRQ handling. Just skip the
1634
* invalidation in that case.
1635
*/
1636
if (in_task() && folio_trylock(folio)) {
1637
filemap_end_dropbehind(folio);
1638
folio_unlock(folio);
1639
}
1640
}
1641
EXPORT_SYMBOL_GPL(folio_end_dropbehind);
1642
1643
/**
1644
* folio_end_writeback_no_dropbehind - End writeback against a folio.
1645
* @folio: The folio.
1646
*
1647
* The folio must actually be under writeback.
1648
* This call is intended for filesystems that need to defer dropbehind.
1649
*
1650
* Context: May be called from process or interrupt context.
1651
*/
1652
void folio_end_writeback_no_dropbehind(struct folio *folio)
1653
{
1654
VM_BUG_ON_FOLIO(!folio_test_writeback(folio), folio);
1655
1656
/*
1657
* folio_test_clear_reclaim() could be used here but it is an
1658
* atomic operation and overkill in this particular case. Failing
1659
* to shuffle a folio marked for immediate reclaim is too mild
1660
* a gain to justify taking an atomic operation penalty at the
1661
* end of every folio writeback.
1662
*/
1663
if (folio_test_reclaim(folio)) {
1664
folio_clear_reclaim(folio);
1665
folio_rotate_reclaimable(folio);
1666
}
1667
1668
if (__folio_end_writeback(folio))
1669
folio_wake_bit(folio, PG_writeback);
1670
1671
acct_reclaim_writeback(folio);
1672
}
1673
EXPORT_SYMBOL_GPL(folio_end_writeback_no_dropbehind);
1674
1675
/**
1676
* folio_end_writeback - End writeback against a folio.
1677
* @folio: The folio.
1678
*
1679
* The folio must actually be under writeback.
1680
*
1681
* Context: May be called from process or interrupt context.
1682
*/
1683
void folio_end_writeback(struct folio *folio)
1684
{
1685
VM_BUG_ON_FOLIO(!folio_test_writeback(folio), folio);
1686
1687
/*
1688
* Writeback does not hold a folio reference of its own, relying
1689
* on truncation to wait for the clearing of PG_writeback.
1690
* But here we must make sure that the folio is not freed and
1691
* reused before the folio_wake_bit().
1692
*/
1693
folio_get(folio);
1694
folio_end_writeback_no_dropbehind(folio);
1695
folio_end_dropbehind(folio);
1696
folio_put(folio);
1697
}
1698
EXPORT_SYMBOL(folio_end_writeback);
1699
1700
/**
1701
* __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1702
* @folio: The folio to lock
1703
*/
1704
void __folio_lock(struct folio *folio)
1705
{
1706
folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1707
EXCLUSIVE);
1708
}
1709
EXPORT_SYMBOL(__folio_lock);
1710
1711
int __folio_lock_killable(struct folio *folio)
1712
{
1713
return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1714
EXCLUSIVE);
1715
}
1716
EXPORT_SYMBOL_GPL(__folio_lock_killable);
1717
1718
static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1719
{
1720
struct wait_queue_head *q = folio_waitqueue(folio);
1721
int ret;
1722
1723
wait->folio = folio;
1724
wait->bit_nr = PG_locked;
1725
1726
spin_lock_irq(&q->lock);
1727
__add_wait_queue_entry_tail(q, &wait->wait);
1728
folio_set_waiters(folio);
1729
ret = !folio_trylock(folio);
1730
/*
1731
* If we were successful now, we know we're still on the
1732
* waitqueue as we're still under the lock. This means it's
1733
* safe to remove and return success, we know the callback
1734
* isn't going to trigger.
1735
*/
1736
if (!ret)
1737
__remove_wait_queue(q, &wait->wait);
1738
else
1739
ret = -EIOCBQUEUED;
1740
spin_unlock_irq(&q->lock);
1741
return ret;
1742
}
1743
1744
/*
1745
* Return values:
1746
* 0 - folio is locked.
1747
* non-zero - folio is not locked.
1748
* mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
1749
* vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
1750
* FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
1751
*
1752
* If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
1753
* with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
1754
*/
1755
vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
1756
{
1757
unsigned int flags = vmf->flags;
1758
1759
if (fault_flag_allow_retry_first(flags)) {
1760
/*
1761
* CAUTION! In this case, mmap_lock/per-VMA lock is not
1762
* released even though returning VM_FAULT_RETRY.
1763
*/
1764
if (flags & FAULT_FLAG_RETRY_NOWAIT)
1765
return VM_FAULT_RETRY;
1766
1767
release_fault_lock(vmf);
1768
if (flags & FAULT_FLAG_KILLABLE)
1769
folio_wait_locked_killable(folio);
1770
else
1771
folio_wait_locked(folio);
1772
return VM_FAULT_RETRY;
1773
}
1774
if (flags & FAULT_FLAG_KILLABLE) {
1775
bool ret;
1776
1777
ret = __folio_lock_killable(folio);
1778
if (ret) {
1779
release_fault_lock(vmf);
1780
return VM_FAULT_RETRY;
1781
}
1782
} else {
1783
__folio_lock(folio);
1784
}
1785
1786
return 0;
1787
}
1788
1789
/**
1790
* page_cache_next_miss() - Find the next gap in the page cache.
1791
* @mapping: Mapping.
1792
* @index: Index.
1793
* @max_scan: Maximum range to search.
1794
*
1795
* Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1796
* gap with the lowest index.
1797
*
1798
* This function may be called under the rcu_read_lock. However, this will
1799
* not atomically search a snapshot of the cache at a single point in time.
1800
* For example, if a gap is created at index 5, then subsequently a gap is
1801
* created at index 10, page_cache_next_miss covering both indices may
1802
* return 10 if called under the rcu_read_lock.
1803
*
1804
* Return: The index of the gap if found, otherwise an index outside the
1805
* range specified (in which case 'return - index >= max_scan' will be true).
1806
* In the rare case of index wrap-around, 0 will be returned.
1807
*/
1808
pgoff_t page_cache_next_miss(struct address_space *mapping,
1809
pgoff_t index, unsigned long max_scan)
1810
{
1811
XA_STATE(xas, &mapping->i_pages, index);
1812
unsigned long nr = max_scan;
1813
1814
while (nr--) {
1815
void *entry = xas_next(&xas);
1816
if (!entry || xa_is_value(entry))
1817
return xas.xa_index;
1818
if (xas.xa_index == 0)
1819
return 0;
1820
}
1821
1822
return index + max_scan;
1823
}
1824
EXPORT_SYMBOL(page_cache_next_miss);
1825
1826
/**
1827
* page_cache_prev_miss() - Find the previous gap in the page cache.
1828
* @mapping: Mapping.
1829
* @index: Index.
1830
* @max_scan: Maximum range to search.
1831
*
1832
* Search the range [max(index - max_scan + 1, 0), index] for the
1833
* gap with the highest index.
1834
*
1835
* This function may be called under the rcu_read_lock. However, this will
1836
* not atomically search a snapshot of the cache at a single point in time.
1837
* For example, if a gap is created at index 10, then subsequently a gap is
1838
* created at index 5, page_cache_prev_miss() covering both indices may
1839
* return 5 if called under the rcu_read_lock.
1840
*
1841
* Return: The index of the gap if found, otherwise an index outside the
1842
* range specified (in which case 'index - return >= max_scan' will be true).
1843
* In the rare case of wrap-around, ULONG_MAX will be returned.
1844
*/
1845
pgoff_t page_cache_prev_miss(struct address_space *mapping,
1846
pgoff_t index, unsigned long max_scan)
1847
{
1848
XA_STATE(xas, &mapping->i_pages, index);
1849
1850
while (max_scan--) {
1851
void *entry = xas_prev(&xas);
1852
if (!entry || xa_is_value(entry))
1853
break;
1854
if (xas.xa_index == ULONG_MAX)
1855
break;
1856
}
1857
1858
return xas.xa_index;
1859
}
1860
EXPORT_SYMBOL(page_cache_prev_miss);
1861
1862
/*
1863
* Lockless page cache protocol:
1864
* On the lookup side:
1865
* 1. Load the folio from i_pages
1866
* 2. Increment the refcount if it's not zero
1867
* 3. If the folio is not found by xas_reload(), put the refcount and retry
1868
*
1869
* On the removal side:
1870
* A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1871
* B. Remove the page from i_pages
1872
* C. Return the page to the page allocator
1873
*
1874
* This means that any page may have its reference count temporarily
1875
* increased by a speculative page cache (or GUP-fast) lookup as it can
1876
* be allocated by another user before the RCU grace period expires.
1877
* Because the refcount temporarily acquired here may end up being the
1878
* last refcount on the page, any page allocation must be freeable by
1879
* folio_put().
1880
*/
1881
1882
/*
1883
* filemap_get_entry - Get a page cache entry.
1884
* @mapping: the address_space to search
1885
* @index: The page cache index.
1886
*
1887
* Looks up the page cache entry at @mapping & @index. If it is a folio,
1888
* it is returned with an increased refcount. If it is a shadow entry
1889
* of a previously evicted folio, or a swap entry from shmem/tmpfs,
1890
* it is returned without further action.
1891
*
1892
* Return: The folio, swap or shadow entry, %NULL if nothing is found.
1893
*/
1894
void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1895
{
1896
XA_STATE(xas, &mapping->i_pages, index);
1897
struct folio *folio;
1898
1899
rcu_read_lock();
1900
repeat:
1901
xas_reset(&xas);
1902
folio = xas_load(&xas);
1903
if (xas_retry(&xas, folio))
1904
goto repeat;
1905
/*
1906
* A shadow entry of a recently evicted page, or a swap entry from
1907
* shmem/tmpfs. Return it without attempting to raise page count.
1908
*/
1909
if (!folio || xa_is_value(folio))
1910
goto out;
1911
1912
if (!folio_try_get(folio))
1913
goto repeat;
1914
1915
if (unlikely(folio != xas_reload(&xas))) {
1916
folio_put(folio);
1917
goto repeat;
1918
}
1919
out:
1920
rcu_read_unlock();
1921
1922
return folio;
1923
}
1924
1925
/**
1926
* __filemap_get_folio - Find and get a reference to a folio.
1927
* @mapping: The address_space to search.
1928
* @index: The page index.
1929
* @fgp_flags: %FGP flags modify how the folio is returned.
1930
* @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1931
*
1932
* Looks up the page cache entry at @mapping & @index.
1933
*
1934
* If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1935
* if the %GFP flags specified for %FGP_CREAT are atomic.
1936
*
1937
* If this function returns a folio, it is returned with an increased refcount.
1938
*
1939
* Return: The found folio or an ERR_PTR() otherwise.
1940
*/
1941
struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1942
fgf_t fgp_flags, gfp_t gfp)
1943
{
1944
struct folio *folio;
1945
1946
repeat:
1947
folio = filemap_get_entry(mapping, index);
1948
if (xa_is_value(folio))
1949
folio = NULL;
1950
if (!folio)
1951
goto no_page;
1952
1953
if (fgp_flags & FGP_LOCK) {
1954
if (fgp_flags & FGP_NOWAIT) {
1955
if (!folio_trylock(folio)) {
1956
folio_put(folio);
1957
return ERR_PTR(-EAGAIN);
1958
}
1959
} else {
1960
folio_lock(folio);
1961
}
1962
1963
/* Has the page been truncated? */
1964
if (unlikely(folio->mapping != mapping)) {
1965
folio_unlock(folio);
1966
folio_put(folio);
1967
goto repeat;
1968
}
1969
VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1970
}
1971
1972
if (fgp_flags & FGP_ACCESSED)
1973
folio_mark_accessed(folio);
1974
else if (fgp_flags & FGP_WRITE) {
1975
/* Clear idle flag for buffer write */
1976
if (folio_test_idle(folio))
1977
folio_clear_idle(folio);
1978
}
1979
1980
if (fgp_flags & FGP_STABLE)
1981
folio_wait_stable(folio);
1982
no_page:
1983
if (!folio && (fgp_flags & FGP_CREAT)) {
1984
unsigned int min_order = mapping_min_folio_order(mapping);
1985
unsigned int order = max(min_order, FGF_GET_ORDER(fgp_flags));
1986
int err;
1987
index = mapping_align_index(mapping, index);
1988
1989
if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1990
gfp |= __GFP_WRITE;
1991
if (fgp_flags & FGP_NOFS)
1992
gfp &= ~__GFP_FS;
1993
if (fgp_flags & FGP_NOWAIT) {
1994
gfp &= ~GFP_KERNEL;
1995
gfp |= GFP_NOWAIT;
1996
}
1997
if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1998
fgp_flags |= FGP_LOCK;
1999
2000
if (order > mapping_max_folio_order(mapping))
2001
order = mapping_max_folio_order(mapping);
2002
/* If we're not aligned, allocate a smaller folio */
2003
if (index & ((1UL << order) - 1))
2004
order = __ffs(index);
2005
2006
do {
2007
gfp_t alloc_gfp = gfp;
2008
2009
err = -ENOMEM;
2010
if (order > min_order)
2011
alloc_gfp |= __GFP_NORETRY | __GFP_NOWARN;
2012
folio = filemap_alloc_folio(alloc_gfp, order);
2013
if (!folio)
2014
continue;
2015
2016
/* Init accessed so avoid atomic mark_page_accessed later */
2017
if (fgp_flags & FGP_ACCESSED)
2018
__folio_set_referenced(folio);
2019
if (fgp_flags & FGP_DONTCACHE)
2020
__folio_set_dropbehind(folio);
2021
2022
err = filemap_add_folio(mapping, folio, index, gfp);
2023
if (!err)
2024
break;
2025
folio_put(folio);
2026
folio = NULL;
2027
} while (order-- > min_order);
2028
2029
if (err == -EEXIST)
2030
goto repeat;
2031
if (err) {
2032
/*
2033
* When NOWAIT I/O fails to allocate folios this could
2034
* be due to a nonblocking memory allocation and not
2035
* because the system actually is out of memory.
2036
* Return -EAGAIN so that there caller retries in a
2037
* blocking fashion instead of propagating -ENOMEM
2038
* to the application.
2039
*/
2040
if ((fgp_flags & FGP_NOWAIT) && err == -ENOMEM)
2041
err = -EAGAIN;
2042
return ERR_PTR(err);
2043
}
2044
/*
2045
* filemap_add_folio locks the page, and for mmap
2046
* we expect an unlocked page.
2047
*/
2048
if (folio && (fgp_flags & FGP_FOR_MMAP))
2049
folio_unlock(folio);
2050
}
2051
2052
if (!folio)
2053
return ERR_PTR(-ENOENT);
2054
/* not an uncached lookup, clear uncached if set */
2055
if (folio_test_dropbehind(folio) && !(fgp_flags & FGP_DONTCACHE))
2056
folio_clear_dropbehind(folio);
2057
return folio;
2058
}
2059
EXPORT_SYMBOL(__filemap_get_folio);
2060
2061
static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
2062
xa_mark_t mark)
2063
{
2064
struct folio *folio;
2065
2066
retry:
2067
if (mark == XA_PRESENT)
2068
folio = xas_find(xas, max);
2069
else
2070
folio = xas_find_marked(xas, max, mark);
2071
2072
if (xas_retry(xas, folio))
2073
goto retry;
2074
/*
2075
* A shadow entry of a recently evicted page, a swap
2076
* entry from shmem/tmpfs or a DAX entry. Return it
2077
* without attempting to raise page count.
2078
*/
2079
if (!folio || xa_is_value(folio))
2080
return folio;
2081
2082
if (!folio_try_get(folio))
2083
goto reset;
2084
2085
if (unlikely(folio != xas_reload(xas))) {
2086
folio_put(folio);
2087
goto reset;
2088
}
2089
2090
return folio;
2091
reset:
2092
xas_reset(xas);
2093
goto retry;
2094
}
2095
2096
/**
2097
* find_get_entries - gang pagecache lookup
2098
* @mapping: The address_space to search
2099
* @start: The starting page cache index
2100
* @end: The final page index (inclusive).
2101
* @fbatch: Where the resulting entries are placed.
2102
* @indices: The cache indices corresponding to the entries in @entries
2103
*
2104
* find_get_entries() will search for and return a batch of entries in
2105
* the mapping. The entries are placed in @fbatch. find_get_entries()
2106
* takes a reference on any actual folios it returns.
2107
*
2108
* The entries have ascending indexes. The indices may not be consecutive
2109
* due to not-present entries or large folios.
2110
*
2111
* Any shadow entries of evicted folios, or swap entries from
2112
* shmem/tmpfs, are included in the returned array.
2113
*
2114
* Return: The number of entries which were found.
2115
*/
2116
unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2117
pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2118
{
2119
XA_STATE(xas, &mapping->i_pages, *start);
2120
struct folio *folio;
2121
2122
rcu_read_lock();
2123
while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2124
indices[fbatch->nr] = xas.xa_index;
2125
if (!folio_batch_add(fbatch, folio))
2126
break;
2127
}
2128
2129
if (folio_batch_count(fbatch)) {
2130
unsigned long nr;
2131
int idx = folio_batch_count(fbatch) - 1;
2132
2133
folio = fbatch->folios[idx];
2134
if (!xa_is_value(folio))
2135
nr = folio_nr_pages(folio);
2136
else
2137
nr = 1 << xa_get_order(&mapping->i_pages, indices[idx]);
2138
*start = round_down(indices[idx] + nr, nr);
2139
}
2140
rcu_read_unlock();
2141
2142
return folio_batch_count(fbatch);
2143
}
2144
2145
/**
2146
* find_lock_entries - Find a batch of pagecache entries.
2147
* @mapping: The address_space to search.
2148
* @start: The starting page cache index.
2149
* @end: The final page index (inclusive).
2150
* @fbatch: Where the resulting entries are placed.
2151
* @indices: The cache indices of the entries in @fbatch.
2152
*
2153
* find_lock_entries() will return a batch of entries from @mapping.
2154
* Swap, shadow and DAX entries are included. Folios are returned
2155
* locked and with an incremented refcount. Folios which are locked
2156
* by somebody else or under writeback are skipped. Folios which are
2157
* partially outside the range are not returned.
2158
*
2159
* The entries have ascending indexes. The indices may not be consecutive
2160
* due to not-present entries, large folios, folios which could not be
2161
* locked or folios under writeback.
2162
*
2163
* Return: The number of entries which were found.
2164
*/
2165
unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2166
pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2167
{
2168
XA_STATE(xas, &mapping->i_pages, *start);
2169
struct folio *folio;
2170
2171
rcu_read_lock();
2172
while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2173
unsigned long base;
2174
unsigned long nr;
2175
2176
if (!xa_is_value(folio)) {
2177
nr = folio_nr_pages(folio);
2178
base = folio->index;
2179
/* Omit large folio which begins before the start */
2180
if (base < *start)
2181
goto put;
2182
/* Omit large folio which extends beyond the end */
2183
if (base + nr - 1 > end)
2184
goto put;
2185
if (!folio_trylock(folio))
2186
goto put;
2187
if (folio->mapping != mapping ||
2188
folio_test_writeback(folio))
2189
goto unlock;
2190
VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2191
folio);
2192
} else {
2193
nr = 1 << xas_get_order(&xas);
2194
base = xas.xa_index & ~(nr - 1);
2195
/* Omit order>0 value which begins before the start */
2196
if (base < *start)
2197
continue;
2198
/* Omit order>0 value which extends beyond the end */
2199
if (base + nr - 1 > end)
2200
break;
2201
}
2202
2203
/* Update start now so that last update is correct on return */
2204
*start = base + nr;
2205
indices[fbatch->nr] = xas.xa_index;
2206
if (!folio_batch_add(fbatch, folio))
2207
break;
2208
continue;
2209
unlock:
2210
folio_unlock(folio);
2211
put:
2212
folio_put(folio);
2213
}
2214
rcu_read_unlock();
2215
2216
return folio_batch_count(fbatch);
2217
}
2218
2219
/**
2220
* filemap_get_folios - Get a batch of folios
2221
* @mapping: The address_space to search
2222
* @start: The starting page index
2223
* @end: The final page index (inclusive)
2224
* @fbatch: The batch to fill.
2225
*
2226
* Search for and return a batch of folios in the mapping starting at
2227
* index @start and up to index @end (inclusive). The folios are returned
2228
* in @fbatch with an elevated reference count.
2229
*
2230
* Return: The number of folios which were found.
2231
* We also update @start to index the next folio for the traversal.
2232
*/
2233
unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2234
pgoff_t end, struct folio_batch *fbatch)
2235
{
2236
return filemap_get_folios_tag(mapping, start, end, XA_PRESENT, fbatch);
2237
}
2238
EXPORT_SYMBOL(filemap_get_folios);
2239
2240
/**
2241
* filemap_get_folios_contig - Get a batch of contiguous folios
2242
* @mapping: The address_space to search
2243
* @start: The starting page index
2244
* @end: The final page index (inclusive)
2245
* @fbatch: The batch to fill
2246
*
2247
* filemap_get_folios_contig() works exactly like filemap_get_folios(),
2248
* except the returned folios are guaranteed to be contiguous. This may
2249
* not return all contiguous folios if the batch gets filled up.
2250
*
2251
* Return: The number of folios found.
2252
* Also update @start to be positioned for traversal of the next folio.
2253
*/
2254
2255
unsigned filemap_get_folios_contig(struct address_space *mapping,
2256
pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2257
{
2258
XA_STATE(xas, &mapping->i_pages, *start);
2259
unsigned long nr;
2260
struct folio *folio;
2261
2262
rcu_read_lock();
2263
2264
for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2265
folio = xas_next(&xas)) {
2266
if (xas_retry(&xas, folio))
2267
continue;
2268
/*
2269
* If the entry has been swapped out, we can stop looking.
2270
* No current caller is looking for DAX entries.
2271
*/
2272
if (xa_is_value(folio))
2273
goto update_start;
2274
2275
/* If we landed in the middle of a THP, continue at its end. */
2276
if (xa_is_sibling(folio))
2277
goto update_start;
2278
2279
if (!folio_try_get(folio))
2280
goto retry;
2281
2282
if (unlikely(folio != xas_reload(&xas)))
2283
goto put_folio;
2284
2285
if (!folio_batch_add(fbatch, folio)) {
2286
nr = folio_nr_pages(folio);
2287
*start = folio->index + nr;
2288
goto out;
2289
}
2290
xas_advance(&xas, folio_next_index(folio) - 1);
2291
continue;
2292
put_folio:
2293
folio_put(folio);
2294
2295
retry:
2296
xas_reset(&xas);
2297
}
2298
2299
update_start:
2300
nr = folio_batch_count(fbatch);
2301
2302
if (nr) {
2303
folio = fbatch->folios[nr - 1];
2304
*start = folio_next_index(folio);
2305
}
2306
out:
2307
rcu_read_unlock();
2308
return folio_batch_count(fbatch);
2309
}
2310
EXPORT_SYMBOL(filemap_get_folios_contig);
2311
2312
/**
2313
* filemap_get_folios_tag - Get a batch of folios matching @tag
2314
* @mapping: The address_space to search
2315
* @start: The starting page index
2316
* @end: The final page index (inclusive)
2317
* @tag: The tag index
2318
* @fbatch: The batch to fill
2319
*
2320
* The first folio may start before @start; if it does, it will contain
2321
* @start. The final folio may extend beyond @end; if it does, it will
2322
* contain @end. The folios have ascending indices. There may be gaps
2323
* between the folios if there are indices which have no folio in the
2324
* page cache. If folios are added to or removed from the page cache
2325
* while this is running, they may or may not be found by this call.
2326
* Only returns folios that are tagged with @tag.
2327
*
2328
* Return: The number of folios found.
2329
* Also update @start to index the next folio for traversal.
2330
*/
2331
unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2332
pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
2333
{
2334
XA_STATE(xas, &mapping->i_pages, *start);
2335
struct folio *folio;
2336
2337
rcu_read_lock();
2338
while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
2339
/*
2340
* Shadow entries should never be tagged, but this iteration
2341
* is lockless so there is a window for page reclaim to evict
2342
* a page we saw tagged. Skip over it.
2343
*/
2344
if (xa_is_value(folio))
2345
continue;
2346
if (!folio_batch_add(fbatch, folio)) {
2347
unsigned long nr = folio_nr_pages(folio);
2348
*start = folio->index + nr;
2349
goto out;
2350
}
2351
}
2352
/*
2353
* We come here when there is no page beyond @end. We take care to not
2354
* overflow the index @start as it confuses some of the callers. This
2355
* breaks the iteration when there is a page at index -1 but that is
2356
* already broke anyway.
2357
*/
2358
if (end == (pgoff_t)-1)
2359
*start = (pgoff_t)-1;
2360
else
2361
*start = end + 1;
2362
out:
2363
rcu_read_unlock();
2364
2365
return folio_batch_count(fbatch);
2366
}
2367
EXPORT_SYMBOL(filemap_get_folios_tag);
2368
2369
/*
2370
* CD/DVDs are error prone. When a medium error occurs, the driver may fail
2371
* a _large_ part of the i/o request. Imagine the worst scenario:
2372
*
2373
* ---R__________________________________________B__________
2374
* ^ reading here ^ bad block(assume 4k)
2375
*
2376
* read(R) => miss => readahead(R...B) => media error => frustrating retries
2377
* => failing the whole request => read(R) => read(R+1) =>
2378
* readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2379
* readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2380
* readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2381
*
2382
* It is going insane. Fix it by quickly scaling down the readahead size.
2383
*/
2384
static void shrink_readahead_size_eio(struct file_ra_state *ra)
2385
{
2386
ra->ra_pages /= 4;
2387
}
2388
2389
/*
2390
* filemap_get_read_batch - Get a batch of folios for read
2391
*
2392
* Get a batch of folios which represent a contiguous range of bytes in
2393
* the file. No exceptional entries will be returned. If @index is in
2394
* the middle of a folio, the entire folio will be returned. The last
2395
* folio in the batch may have the readahead flag set or the uptodate flag
2396
* clear so that the caller can take the appropriate action.
2397
*/
2398
static void filemap_get_read_batch(struct address_space *mapping,
2399
pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2400
{
2401
XA_STATE(xas, &mapping->i_pages, index);
2402
struct folio *folio;
2403
2404
rcu_read_lock();
2405
for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2406
if (xas_retry(&xas, folio))
2407
continue;
2408
if (xas.xa_index > max || xa_is_value(folio))
2409
break;
2410
if (xa_is_sibling(folio))
2411
break;
2412
if (!folio_try_get(folio))
2413
goto retry;
2414
2415
if (unlikely(folio != xas_reload(&xas)))
2416
goto put_folio;
2417
2418
if (!folio_batch_add(fbatch, folio))
2419
break;
2420
if (!folio_test_uptodate(folio))
2421
break;
2422
if (folio_test_readahead(folio))
2423
break;
2424
xas_advance(&xas, folio_next_index(folio) - 1);
2425
continue;
2426
put_folio:
2427
folio_put(folio);
2428
retry:
2429
xas_reset(&xas);
2430
}
2431
rcu_read_unlock();
2432
}
2433
2434
static int filemap_read_folio(struct file *file, filler_t filler,
2435
struct folio *folio)
2436
{
2437
bool workingset = folio_test_workingset(folio);
2438
unsigned long pflags;
2439
int error;
2440
2441
/* Start the actual read. The read will unlock the page. */
2442
if (unlikely(workingset))
2443
psi_memstall_enter(&pflags);
2444
error = filler(file, folio);
2445
if (unlikely(workingset))
2446
psi_memstall_leave(&pflags);
2447
if (error)
2448
return error;
2449
2450
error = folio_wait_locked_killable(folio);
2451
if (error)
2452
return error;
2453
if (folio_test_uptodate(folio))
2454
return 0;
2455
if (file)
2456
shrink_readahead_size_eio(&file->f_ra);
2457
return -EIO;
2458
}
2459
2460
static bool filemap_range_uptodate(struct address_space *mapping,
2461
loff_t pos, size_t count, struct folio *folio,
2462
bool need_uptodate)
2463
{
2464
if (folio_test_uptodate(folio))
2465
return true;
2466
/* pipes can't handle partially uptodate pages */
2467
if (need_uptodate)
2468
return false;
2469
if (!mapping->a_ops->is_partially_uptodate)
2470
return false;
2471
if (mapping->host->i_blkbits >= folio_shift(folio))
2472
return false;
2473
2474
if (folio_pos(folio) > pos) {
2475
count -= folio_pos(folio) - pos;
2476
pos = 0;
2477
} else {
2478
pos -= folio_pos(folio);
2479
}
2480
2481
if (pos == 0 && count >= folio_size(folio))
2482
return false;
2483
2484
return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2485
}
2486
2487
static int filemap_update_page(struct kiocb *iocb,
2488
struct address_space *mapping, size_t count,
2489
struct folio *folio, bool need_uptodate)
2490
{
2491
int error;
2492
2493
if (iocb->ki_flags & IOCB_NOWAIT) {
2494
if (!filemap_invalidate_trylock_shared(mapping))
2495
return -EAGAIN;
2496
} else {
2497
filemap_invalidate_lock_shared(mapping);
2498
}
2499
2500
if (!folio_trylock(folio)) {
2501
error = -EAGAIN;
2502
if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2503
goto unlock_mapping;
2504
if (!(iocb->ki_flags & IOCB_WAITQ)) {
2505
filemap_invalidate_unlock_shared(mapping);
2506
/*
2507
* This is where we usually end up waiting for a
2508
* previously submitted readahead to finish.
2509
*/
2510
folio_put_wait_locked(folio, TASK_KILLABLE);
2511
return AOP_TRUNCATED_PAGE;
2512
}
2513
error = __folio_lock_async(folio, iocb->ki_waitq);
2514
if (error)
2515
goto unlock_mapping;
2516
}
2517
2518
error = AOP_TRUNCATED_PAGE;
2519
if (!folio->mapping)
2520
goto unlock;
2521
2522
error = 0;
2523
if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2524
need_uptodate))
2525
goto unlock;
2526
2527
error = -EAGAIN;
2528
if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2529
goto unlock;
2530
2531
error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2532
folio);
2533
goto unlock_mapping;
2534
unlock:
2535
folio_unlock(folio);
2536
unlock_mapping:
2537
filemap_invalidate_unlock_shared(mapping);
2538
if (error == AOP_TRUNCATED_PAGE)
2539
folio_put(folio);
2540
return error;
2541
}
2542
2543
static int filemap_create_folio(struct kiocb *iocb, struct folio_batch *fbatch)
2544
{
2545
struct address_space *mapping = iocb->ki_filp->f_mapping;
2546
struct folio *folio;
2547
int error;
2548
unsigned int min_order = mapping_min_folio_order(mapping);
2549
pgoff_t index;
2550
2551
if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2552
return -EAGAIN;
2553
2554
folio = filemap_alloc_folio(mapping_gfp_mask(mapping), min_order);
2555
if (!folio)
2556
return -ENOMEM;
2557
if (iocb->ki_flags & IOCB_DONTCACHE)
2558
__folio_set_dropbehind(folio);
2559
2560
/*
2561
* Protect against truncate / hole punch. Grabbing invalidate_lock
2562
* here assures we cannot instantiate and bring uptodate new
2563
* pagecache folios after evicting page cache during truncate
2564
* and before actually freeing blocks. Note that we could
2565
* release invalidate_lock after inserting the folio into
2566
* the page cache as the locked folio would then be enough to
2567
* synchronize with hole punching. But there are code paths
2568
* such as filemap_update_page() filling in partially uptodate
2569
* pages or ->readahead() that need to hold invalidate_lock
2570
* while mapping blocks for IO so let's hold the lock here as
2571
* well to keep locking rules simple.
2572
*/
2573
filemap_invalidate_lock_shared(mapping);
2574
index = (iocb->ki_pos >> (PAGE_SHIFT + min_order)) << min_order;
2575
error = filemap_add_folio(mapping, folio, index,
2576
mapping_gfp_constraint(mapping, GFP_KERNEL));
2577
if (error == -EEXIST)
2578
error = AOP_TRUNCATED_PAGE;
2579
if (error)
2580
goto error;
2581
2582
error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2583
folio);
2584
if (error)
2585
goto error;
2586
2587
filemap_invalidate_unlock_shared(mapping);
2588
folio_batch_add(fbatch, folio);
2589
return 0;
2590
error:
2591
filemap_invalidate_unlock_shared(mapping);
2592
folio_put(folio);
2593
return error;
2594
}
2595
2596
static int filemap_readahead(struct kiocb *iocb, struct file *file,
2597
struct address_space *mapping, struct folio *folio,
2598
pgoff_t last_index)
2599
{
2600
DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2601
2602
if (iocb->ki_flags & IOCB_NOIO)
2603
return -EAGAIN;
2604
if (iocb->ki_flags & IOCB_DONTCACHE)
2605
ractl.dropbehind = 1;
2606
page_cache_async_ra(&ractl, folio, last_index - folio->index);
2607
return 0;
2608
}
2609
2610
static int filemap_get_pages(struct kiocb *iocb, size_t count,
2611
struct folio_batch *fbatch, bool need_uptodate)
2612
{
2613
struct file *filp = iocb->ki_filp;
2614
struct address_space *mapping = filp->f_mapping;
2615
pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2616
pgoff_t last_index;
2617
struct folio *folio;
2618
unsigned int flags;
2619
int err = 0;
2620
2621
/* "last_index" is the index of the folio beyond the end of the read */
2622
last_index = round_up(iocb->ki_pos + count,
2623
mapping_min_folio_nrbytes(mapping)) >> PAGE_SHIFT;
2624
retry:
2625
if (fatal_signal_pending(current))
2626
return -EINTR;
2627
2628
filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2629
if (!folio_batch_count(fbatch)) {
2630
DEFINE_READAHEAD(ractl, filp, &filp->f_ra, mapping, index);
2631
2632
if (iocb->ki_flags & IOCB_NOIO)
2633
return -EAGAIN;
2634
if (iocb->ki_flags & IOCB_NOWAIT)
2635
flags = memalloc_noio_save();
2636
if (iocb->ki_flags & IOCB_DONTCACHE)
2637
ractl.dropbehind = 1;
2638
page_cache_sync_ra(&ractl, last_index - index);
2639
if (iocb->ki_flags & IOCB_NOWAIT)
2640
memalloc_noio_restore(flags);
2641
filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2642
}
2643
if (!folio_batch_count(fbatch)) {
2644
err = filemap_create_folio(iocb, fbatch);
2645
if (err == AOP_TRUNCATED_PAGE)
2646
goto retry;
2647
return err;
2648
}
2649
2650
folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2651
if (folio_test_readahead(folio)) {
2652
err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2653
if (err)
2654
goto err;
2655
}
2656
if (!folio_test_uptodate(folio)) {
2657
if (folio_batch_count(fbatch) > 1) {
2658
err = -EAGAIN;
2659
goto err;
2660
}
2661
err = filemap_update_page(iocb, mapping, count, folio,
2662
need_uptodate);
2663
if (err)
2664
goto err;
2665
}
2666
2667
trace_mm_filemap_get_pages(mapping, index, last_index - 1);
2668
return 0;
2669
err:
2670
if (err < 0)
2671
folio_put(folio);
2672
if (likely(--fbatch->nr))
2673
return 0;
2674
if (err == AOP_TRUNCATED_PAGE)
2675
goto retry;
2676
return err;
2677
}
2678
2679
static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2680
{
2681
unsigned int shift = folio_shift(folio);
2682
2683
return (pos1 >> shift == pos2 >> shift);
2684
}
2685
2686
static void filemap_end_dropbehind_read(struct folio *folio)
2687
{
2688
if (!folio_test_dropbehind(folio))
2689
return;
2690
if (folio_test_writeback(folio) || folio_test_dirty(folio))
2691
return;
2692
if (folio_trylock(folio)) {
2693
filemap_end_dropbehind(folio);
2694
folio_unlock(folio);
2695
}
2696
}
2697
2698
/**
2699
* filemap_read - Read data from the page cache.
2700
* @iocb: The iocb to read.
2701
* @iter: Destination for the data.
2702
* @already_read: Number of bytes already read by the caller.
2703
*
2704
* Copies data from the page cache. If the data is not currently present,
2705
* uses the readahead and read_folio address_space operations to fetch it.
2706
*
2707
* Return: Total number of bytes copied, including those already read by
2708
* the caller. If an error happens before any bytes are copied, returns
2709
* a negative error number.
2710
*/
2711
ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2712
ssize_t already_read)
2713
{
2714
struct file *filp = iocb->ki_filp;
2715
struct file_ra_state *ra = &filp->f_ra;
2716
struct address_space *mapping = filp->f_mapping;
2717
struct inode *inode = mapping->host;
2718
struct folio_batch fbatch;
2719
int i, error = 0;
2720
bool writably_mapped;
2721
loff_t isize, end_offset;
2722
loff_t last_pos = ra->prev_pos;
2723
2724
if (unlikely(iocb->ki_pos < 0))
2725
return -EINVAL;
2726
if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2727
return 0;
2728
if (unlikely(!iov_iter_count(iter)))
2729
return 0;
2730
2731
iov_iter_truncate(iter, inode->i_sb->s_maxbytes - iocb->ki_pos);
2732
folio_batch_init(&fbatch);
2733
2734
do {
2735
cond_resched();
2736
2737
/*
2738
* If we've already successfully copied some data, then we
2739
* can no longer safely return -EIOCBQUEUED. Hence mark
2740
* an async read NOWAIT at that point.
2741
*/
2742
if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2743
iocb->ki_flags |= IOCB_NOWAIT;
2744
2745
if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2746
break;
2747
2748
error = filemap_get_pages(iocb, iter->count, &fbatch, false);
2749
if (error < 0)
2750
break;
2751
2752
/*
2753
* i_size must be checked after we know the pages are Uptodate.
2754
*
2755
* Checking i_size after the check allows us to calculate
2756
* the correct value for "nr", which means the zero-filled
2757
* part of the page is not copied back to userspace (unless
2758
* another truncate extends the file - this is desired though).
2759
*/
2760
isize = i_size_read(inode);
2761
if (unlikely(iocb->ki_pos >= isize))
2762
goto put_folios;
2763
end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2764
2765
/*
2766
* Once we start copying data, we don't want to be touching any
2767
* cachelines that might be contended:
2768
*/
2769
writably_mapped = mapping_writably_mapped(mapping);
2770
2771
/*
2772
* When a read accesses the same folio several times, only
2773
* mark it as accessed the first time.
2774
*/
2775
if (!pos_same_folio(iocb->ki_pos, last_pos - 1,
2776
fbatch.folios[0]))
2777
folio_mark_accessed(fbatch.folios[0]);
2778
2779
for (i = 0; i < folio_batch_count(&fbatch); i++) {
2780
struct folio *folio = fbatch.folios[i];
2781
size_t fsize = folio_size(folio);
2782
size_t offset = iocb->ki_pos & (fsize - 1);
2783
size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2784
fsize - offset);
2785
size_t copied;
2786
2787
if (end_offset < folio_pos(folio))
2788
break;
2789
if (i > 0)
2790
folio_mark_accessed(folio);
2791
/*
2792
* If users can be writing to this folio using arbitrary
2793
* virtual addresses, take care of potential aliasing
2794
* before reading the folio on the kernel side.
2795
*/
2796
if (writably_mapped)
2797
flush_dcache_folio(folio);
2798
2799
copied = copy_folio_to_iter(folio, offset, bytes, iter);
2800
2801
already_read += copied;
2802
iocb->ki_pos += copied;
2803
last_pos = iocb->ki_pos;
2804
2805
if (copied < bytes) {
2806
error = -EFAULT;
2807
break;
2808
}
2809
}
2810
put_folios:
2811
for (i = 0; i < folio_batch_count(&fbatch); i++) {
2812
struct folio *folio = fbatch.folios[i];
2813
2814
filemap_end_dropbehind_read(folio);
2815
folio_put(folio);
2816
}
2817
folio_batch_init(&fbatch);
2818
} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2819
2820
file_accessed(filp);
2821
ra->prev_pos = last_pos;
2822
return already_read ? already_read : error;
2823
}
2824
EXPORT_SYMBOL_GPL(filemap_read);
2825
2826
int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
2827
{
2828
struct address_space *mapping = iocb->ki_filp->f_mapping;
2829
loff_t pos = iocb->ki_pos;
2830
loff_t end = pos + count - 1;
2831
2832
if (iocb->ki_flags & IOCB_NOWAIT) {
2833
if (filemap_range_needs_writeback(mapping, pos, end))
2834
return -EAGAIN;
2835
return 0;
2836
}
2837
2838
return filemap_write_and_wait_range(mapping, pos, end);
2839
}
2840
EXPORT_SYMBOL_GPL(kiocb_write_and_wait);
2841
2842
int filemap_invalidate_pages(struct address_space *mapping,
2843
loff_t pos, loff_t end, bool nowait)
2844
{
2845
int ret;
2846
2847
if (nowait) {
2848
/* we could block if there are any pages in the range */
2849
if (filemap_range_has_page(mapping, pos, end))
2850
return -EAGAIN;
2851
} else {
2852
ret = filemap_write_and_wait_range(mapping, pos, end);
2853
if (ret)
2854
return ret;
2855
}
2856
2857
/*
2858
* After a write we want buffered reads to be sure to go to disk to get
2859
* the new data. We invalidate clean cached page from the region we're
2860
* about to write. We do this *before* the write so that we can return
2861
* without clobbering -EIOCBQUEUED from ->direct_IO().
2862
*/
2863
return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
2864
end >> PAGE_SHIFT);
2865
}
2866
2867
int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
2868
{
2869
struct address_space *mapping = iocb->ki_filp->f_mapping;
2870
2871
return filemap_invalidate_pages(mapping, iocb->ki_pos,
2872
iocb->ki_pos + count - 1,
2873
iocb->ki_flags & IOCB_NOWAIT);
2874
}
2875
EXPORT_SYMBOL_GPL(kiocb_invalidate_pages);
2876
2877
/**
2878
* generic_file_read_iter - generic filesystem read routine
2879
* @iocb: kernel I/O control block
2880
* @iter: destination for the data read
2881
*
2882
* This is the "read_iter()" routine for all filesystems
2883
* that can use the page cache directly.
2884
*
2885
* The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2886
* be returned when no data can be read without waiting for I/O requests
2887
* to complete; it doesn't prevent readahead.
2888
*
2889
* The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2890
* requests shall be made for the read or for readahead. When no data
2891
* can be read, -EAGAIN shall be returned. When readahead would be
2892
* triggered, a partial, possibly empty read shall be returned.
2893
*
2894
* Return:
2895
* * number of bytes copied, even for partial reads
2896
* * negative error code (or 0 if IOCB_NOIO) if nothing was read
2897
*/
2898
ssize_t
2899
generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2900
{
2901
size_t count = iov_iter_count(iter);
2902
ssize_t retval = 0;
2903
2904
if (!count)
2905
return 0; /* skip atime */
2906
2907
if (iocb->ki_flags & IOCB_DIRECT) {
2908
struct file *file = iocb->ki_filp;
2909
struct address_space *mapping = file->f_mapping;
2910
struct inode *inode = mapping->host;
2911
2912
retval = kiocb_write_and_wait(iocb, count);
2913
if (retval < 0)
2914
return retval;
2915
file_accessed(file);
2916
2917
retval = mapping->a_ops->direct_IO(iocb, iter);
2918
if (retval >= 0) {
2919
iocb->ki_pos += retval;
2920
count -= retval;
2921
}
2922
if (retval != -EIOCBQUEUED)
2923
iov_iter_revert(iter, count - iov_iter_count(iter));
2924
2925
/*
2926
* Btrfs can have a short DIO read if we encounter
2927
* compressed extents, so if there was an error, or if
2928
* we've already read everything we wanted to, or if
2929
* there was a short read because we hit EOF, go ahead
2930
* and return. Otherwise fallthrough to buffered io for
2931
* the rest of the read. Buffered reads will not work for
2932
* DAX files, so don't bother trying.
2933
*/
2934
if (retval < 0 || !count || IS_DAX(inode))
2935
return retval;
2936
if (iocb->ki_pos >= i_size_read(inode))
2937
return retval;
2938
}
2939
2940
return filemap_read(iocb, iter, retval);
2941
}
2942
EXPORT_SYMBOL(generic_file_read_iter);
2943
2944
/*
2945
* Splice subpages from a folio into a pipe.
2946
*/
2947
size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2948
struct folio *folio, loff_t fpos, size_t size)
2949
{
2950
struct page *page;
2951
size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2952
2953
page = folio_page(folio, offset / PAGE_SIZE);
2954
size = min(size, folio_size(folio) - offset);
2955
offset %= PAGE_SIZE;
2956
2957
while (spliced < size && !pipe_is_full(pipe)) {
2958
struct pipe_buffer *buf = pipe_head_buf(pipe);
2959
size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2960
2961
*buf = (struct pipe_buffer) {
2962
.ops = &page_cache_pipe_buf_ops,
2963
.page = page,
2964
.offset = offset,
2965
.len = part,
2966
};
2967
folio_get(folio);
2968
pipe->head++;
2969
page++;
2970
spliced += part;
2971
offset = 0;
2972
}
2973
2974
return spliced;
2975
}
2976
2977
/**
2978
* filemap_splice_read - Splice data from a file's pagecache into a pipe
2979
* @in: The file to read from
2980
* @ppos: Pointer to the file position to read from
2981
* @pipe: The pipe to splice into
2982
* @len: The amount to splice
2983
* @flags: The SPLICE_F_* flags
2984
*
2985
* This function gets folios from a file's pagecache and splices them into the
2986
* pipe. Readahead will be called as necessary to fill more folios. This may
2987
* be used for blockdevs also.
2988
*
2989
* Return: On success, the number of bytes read will be returned and *@ppos
2990
* will be updated if appropriate; 0 will be returned if there is no more data
2991
* to be read; -EAGAIN will be returned if the pipe had no space, and some
2992
* other negative error code will be returned on error. A short read may occur
2993
* if the pipe has insufficient space, we reach the end of the data or we hit a
2994
* hole.
2995
*/
2996
ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2997
struct pipe_inode_info *pipe,
2998
size_t len, unsigned int flags)
2999
{
3000
struct folio_batch fbatch;
3001
struct kiocb iocb;
3002
size_t total_spliced = 0, used, npages;
3003
loff_t isize, end_offset;
3004
bool writably_mapped;
3005
int i, error = 0;
3006
3007
if (unlikely(*ppos >= in->f_mapping->host->i_sb->s_maxbytes))
3008
return 0;
3009
3010
init_sync_kiocb(&iocb, in);
3011
iocb.ki_pos = *ppos;
3012
3013
/* Work out how much data we can actually add into the pipe */
3014
used = pipe_buf_usage(pipe);
3015
npages = max_t(ssize_t, pipe->max_usage - used, 0);
3016
len = min_t(size_t, len, npages * PAGE_SIZE);
3017
3018
folio_batch_init(&fbatch);
3019
3020
do {
3021
cond_resched();
3022
3023
if (*ppos >= i_size_read(in->f_mapping->host))
3024
break;
3025
3026
iocb.ki_pos = *ppos;
3027
error = filemap_get_pages(&iocb, len, &fbatch, true);
3028
if (error < 0)
3029
break;
3030
3031
/*
3032
* i_size must be checked after we know the pages are Uptodate.
3033
*
3034
* Checking i_size after the check allows us to calculate
3035
* the correct value for "nr", which means the zero-filled
3036
* part of the page is not copied back to userspace (unless
3037
* another truncate extends the file - this is desired though).
3038
*/
3039
isize = i_size_read(in->f_mapping->host);
3040
if (unlikely(*ppos >= isize))
3041
break;
3042
end_offset = min_t(loff_t, isize, *ppos + len);
3043
3044
/*
3045
* Once we start copying data, we don't want to be touching any
3046
* cachelines that might be contended:
3047
*/
3048
writably_mapped = mapping_writably_mapped(in->f_mapping);
3049
3050
for (i = 0; i < folio_batch_count(&fbatch); i++) {
3051
struct folio *folio = fbatch.folios[i];
3052
size_t n;
3053
3054
if (folio_pos(folio) >= end_offset)
3055
goto out;
3056
folio_mark_accessed(folio);
3057
3058
/*
3059
* If users can be writing to this folio using arbitrary
3060
* virtual addresses, take care of potential aliasing
3061
* before reading the folio on the kernel side.
3062
*/
3063
if (writably_mapped)
3064
flush_dcache_folio(folio);
3065
3066
n = min_t(loff_t, len, isize - *ppos);
3067
n = splice_folio_into_pipe(pipe, folio, *ppos, n);
3068
if (!n)
3069
goto out;
3070
len -= n;
3071
total_spliced += n;
3072
*ppos += n;
3073
in->f_ra.prev_pos = *ppos;
3074
if (pipe_is_full(pipe))
3075
goto out;
3076
}
3077
3078
folio_batch_release(&fbatch);
3079
} while (len);
3080
3081
out:
3082
folio_batch_release(&fbatch);
3083
file_accessed(in);
3084
3085
return total_spliced ? total_spliced : error;
3086
}
3087
EXPORT_SYMBOL(filemap_splice_read);
3088
3089
static inline loff_t folio_seek_hole_data(struct xa_state *xas,
3090
struct address_space *mapping, struct folio *folio,
3091
loff_t start, loff_t end, bool seek_data)
3092
{
3093
const struct address_space_operations *ops = mapping->a_ops;
3094
size_t offset, bsz = i_blocksize(mapping->host);
3095
3096
if (xa_is_value(folio) || folio_test_uptodate(folio))
3097
return seek_data ? start : end;
3098
if (!ops->is_partially_uptodate)
3099
return seek_data ? end : start;
3100
3101
xas_pause(xas);
3102
rcu_read_unlock();
3103
folio_lock(folio);
3104
if (unlikely(folio->mapping != mapping))
3105
goto unlock;
3106
3107
offset = offset_in_folio(folio, start) & ~(bsz - 1);
3108
3109
do {
3110
if (ops->is_partially_uptodate(folio, offset, bsz) ==
3111
seek_data)
3112
break;
3113
start = (start + bsz) & ~((u64)bsz - 1);
3114
offset += bsz;
3115
} while (offset < folio_size(folio));
3116
unlock:
3117
folio_unlock(folio);
3118
rcu_read_lock();
3119
return start;
3120
}
3121
3122
static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
3123
{
3124
if (xa_is_value(folio))
3125
return PAGE_SIZE << xas_get_order(xas);
3126
return folio_size(folio);
3127
}
3128
3129
/**
3130
* mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3131
* @mapping: Address space to search.
3132
* @start: First byte to consider.
3133
* @end: Limit of search (exclusive).
3134
* @whence: Either SEEK_HOLE or SEEK_DATA.
3135
*
3136
* If the page cache knows which blocks contain holes and which blocks
3137
* contain data, your filesystem can use this function to implement
3138
* SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3139
* entirely memory-based such as tmpfs, and filesystems which support
3140
* unwritten extents.
3141
*
3142
* Return: The requested offset on success, or -ENXIO if @whence specifies
3143
* SEEK_DATA and there is no data after @start. There is an implicit hole
3144
* after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3145
* and @end contain data.
3146
*/
3147
loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3148
loff_t end, int whence)
3149
{
3150
XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3151
pgoff_t max = (end - 1) >> PAGE_SHIFT;
3152
bool seek_data = (whence == SEEK_DATA);
3153
struct folio *folio;
3154
3155
if (end <= start)
3156
return -ENXIO;
3157
3158
rcu_read_lock();
3159
while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3160
loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3161
size_t seek_size;
3162
3163
if (start < pos) {
3164
if (!seek_data)
3165
goto unlock;
3166
start = pos;
3167
}
3168
3169
seek_size = seek_folio_size(&xas, folio);
3170
pos = round_up((u64)pos + 1, seek_size);
3171
start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3172
seek_data);
3173
if (start < pos)
3174
goto unlock;
3175
if (start >= end)
3176
break;
3177
if (seek_size > PAGE_SIZE)
3178
xas_set(&xas, pos >> PAGE_SHIFT);
3179
if (!xa_is_value(folio))
3180
folio_put(folio);
3181
}
3182
if (seek_data)
3183
start = -ENXIO;
3184
unlock:
3185
rcu_read_unlock();
3186
if (folio && !xa_is_value(folio))
3187
folio_put(folio);
3188
if (start > end)
3189
return end;
3190
return start;
3191
}
3192
3193
#ifdef CONFIG_MMU
3194
#define MMAP_LOTSAMISS (100)
3195
/*
3196
* lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3197
* @vmf - the vm_fault for this fault.
3198
* @folio - the folio to lock.
3199
* @fpin - the pointer to the file we may pin (or is already pinned).
3200
*
3201
* This works similar to lock_folio_or_retry in that it can drop the
3202
* mmap_lock. It differs in that it actually returns the folio locked
3203
* if it returns 1 and 0 if it couldn't lock the folio. If we did have
3204
* to drop the mmap_lock then fpin will point to the pinned file and
3205
* needs to be fput()'ed at a later point.
3206
*/
3207
static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3208
struct file **fpin)
3209
{
3210
if (folio_trylock(folio))
3211
return 1;
3212
3213
/*
3214
* NOTE! This will make us return with VM_FAULT_RETRY, but with
3215
* the fault lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3216
* is supposed to work. We have way too many special cases..
3217
*/
3218
if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3219
return 0;
3220
3221
*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3222
if (vmf->flags & FAULT_FLAG_KILLABLE) {
3223
if (__folio_lock_killable(folio)) {
3224
/*
3225
* We didn't have the right flags to drop the
3226
* fault lock, but all fault_handlers only check
3227
* for fatal signals if we return VM_FAULT_RETRY,
3228
* so we need to drop the fault lock here and
3229
* return 0 if we don't have a fpin.
3230
*/
3231
if (*fpin == NULL)
3232
release_fault_lock(vmf);
3233
return 0;
3234
}
3235
} else
3236
__folio_lock(folio);
3237
3238
return 1;
3239
}
3240
3241
/*
3242
* Synchronous readahead happens when we don't even find a page in the page
3243
* cache at all. We don't want to perform IO under the mmap sem, so if we have
3244
* to drop the mmap sem we return the file that was pinned in order for us to do
3245
* that. If we didn't pin a file then we return NULL. The file that is
3246
* returned needs to be fput()'ed when we're done with it.
3247
*/
3248
static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3249
{
3250
struct file *file = vmf->vma->vm_file;
3251
struct file_ra_state *ra = &file->f_ra;
3252
struct address_space *mapping = file->f_mapping;
3253
DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3254
struct file *fpin = NULL;
3255
vm_flags_t vm_flags = vmf->vma->vm_flags;
3256
unsigned short mmap_miss;
3257
3258
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3259
/* Use the readahead code, even if readahead is disabled */
3260
if ((vm_flags & VM_HUGEPAGE) && HPAGE_PMD_ORDER <= MAX_PAGECACHE_ORDER) {
3261
fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3262
ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3263
ra->size = HPAGE_PMD_NR;
3264
/*
3265
* Fetch two PMD folios, so we get the chance to actually
3266
* readahead, unless we've been told not to.
3267
*/
3268
if (!(vm_flags & VM_RAND_READ))
3269
ra->size *= 2;
3270
ra->async_size = HPAGE_PMD_NR;
3271
ra->order = HPAGE_PMD_ORDER;
3272
page_cache_ra_order(&ractl, ra);
3273
return fpin;
3274
}
3275
#endif
3276
3277
/*
3278
* If we don't want any read-ahead, don't bother. VM_EXEC case below is
3279
* already intended for random access.
3280
*/
3281
if ((vm_flags & (VM_RAND_READ | VM_EXEC)) == VM_RAND_READ)
3282
return fpin;
3283
if (!ra->ra_pages)
3284
return fpin;
3285
3286
if (vm_flags & VM_SEQ_READ) {
3287
fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3288
page_cache_sync_ra(&ractl, ra->ra_pages);
3289
return fpin;
3290
}
3291
3292
/* Avoid banging the cache line if not needed */
3293
mmap_miss = READ_ONCE(ra->mmap_miss);
3294
if (mmap_miss < MMAP_LOTSAMISS * 10)
3295
WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3296
3297
/*
3298
* Do we miss much more than hit in this file? If so,
3299
* stop bothering with read-ahead. It will only hurt.
3300
*/
3301
if (mmap_miss > MMAP_LOTSAMISS)
3302
return fpin;
3303
3304
if (vm_flags & VM_EXEC) {
3305
/*
3306
* Allow arch to request a preferred minimum folio order for
3307
* executable memory. This can often be beneficial to
3308
* performance if (e.g.) arm64 can contpte-map the folio.
3309
* Executable memory rarely benefits from readahead, due to its
3310
* random access nature, so set async_size to 0.
3311
*
3312
* Limit to the boundaries of the VMA to avoid reading in any
3313
* pad that might exist between sections, which would be a waste
3314
* of memory.
3315
*/
3316
struct vm_area_struct *vma = vmf->vma;
3317
unsigned long start = vma->vm_pgoff;
3318
unsigned long end = start + vma_pages(vma);
3319
unsigned long ra_end;
3320
3321
ra->order = exec_folio_order();
3322
ra->start = round_down(vmf->pgoff, 1UL << ra->order);
3323
ra->start = max(ra->start, start);
3324
ra_end = round_up(ra->start + ra->ra_pages, 1UL << ra->order);
3325
ra_end = min(ra_end, end);
3326
ra->size = ra_end - ra->start;
3327
ra->async_size = 0;
3328
} else {
3329
/*
3330
* mmap read-around
3331
*/
3332
ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3333
ra->size = ra->ra_pages;
3334
ra->async_size = ra->ra_pages / 4;
3335
ra->order = 0;
3336
}
3337
3338
fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3339
ractl._index = ra->start;
3340
page_cache_ra_order(&ractl, ra);
3341
return fpin;
3342
}
3343
3344
/*
3345
* Asynchronous readahead happens when we find the page and PG_readahead,
3346
* so we want to possibly extend the readahead further. We return the file that
3347
* was pinned if we have to drop the mmap_lock in order to do IO.
3348
*/
3349
static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3350
struct folio *folio)
3351
{
3352
struct file *file = vmf->vma->vm_file;
3353
struct file_ra_state *ra = &file->f_ra;
3354
DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3355
struct file *fpin = NULL;
3356
unsigned short mmap_miss;
3357
3358
/* If we don't want any read-ahead, don't bother */
3359
if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3360
return fpin;
3361
3362
/*
3363
* If the folio is locked, we're likely racing against another fault.
3364
* Don't touch the mmap_miss counter to avoid decreasing it multiple
3365
* times for a single folio and break the balance with mmap_miss
3366
* increase in do_sync_mmap_readahead().
3367
*/
3368
if (likely(!folio_test_locked(folio))) {
3369
mmap_miss = READ_ONCE(ra->mmap_miss);
3370
if (mmap_miss)
3371
WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3372
}
3373
3374
if (folio_test_readahead(folio)) {
3375
fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3376
page_cache_async_ra(&ractl, folio, ra->ra_pages);
3377
}
3378
return fpin;
3379
}
3380
3381
static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault *vmf)
3382
{
3383
struct vm_area_struct *vma = vmf->vma;
3384
vm_fault_t ret = 0;
3385
pte_t *ptep;
3386
3387
/*
3388
* We might have COW'ed a pagecache folio and might now have an mlocked
3389
* anon folio mapped. The original pagecache folio is not mlocked and
3390
* might have been evicted. During a read+clear/modify/write update of
3391
* the PTE, such as done in do_numa_page()/change_pte_range(), we
3392
* temporarily clear the PTE under PT lock and might detect it here as
3393
* "none" when not holding the PT lock.
3394
*
3395
* Not rechecking the PTE under PT lock could result in an unexpected
3396
* major fault in an mlock'ed region. Recheck only for this special
3397
* scenario while holding the PT lock, to not degrade non-mlocked
3398
* scenarios. Recheck the PTE without PT lock firstly, thereby reducing
3399
* the number of times we hold PT lock.
3400
*/
3401
if (!(vma->vm_flags & VM_LOCKED))
3402
return 0;
3403
3404
if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
3405
return 0;
3406
3407
ptep = pte_offset_map_ro_nolock(vma->vm_mm, vmf->pmd, vmf->address,
3408
&vmf->ptl);
3409
if (unlikely(!ptep))
3410
return VM_FAULT_NOPAGE;
3411
3412
if (unlikely(!pte_none(ptep_get_lockless(ptep)))) {
3413
ret = VM_FAULT_NOPAGE;
3414
} else {
3415
spin_lock(vmf->ptl);
3416
if (unlikely(!pte_none(ptep_get(ptep))))
3417
ret = VM_FAULT_NOPAGE;
3418
spin_unlock(vmf->ptl);
3419
}
3420
pte_unmap(ptep);
3421
return ret;
3422
}
3423
3424
/**
3425
* filemap_fault - read in file data for page fault handling
3426
* @vmf: struct vm_fault containing details of the fault
3427
*
3428
* filemap_fault() is invoked via the vma operations vector for a
3429
* mapped memory region to read in file data during a page fault.
3430
*
3431
* The goto's are kind of ugly, but this streamlines the normal case of having
3432
* it in the page cache, and handles the special cases reasonably without
3433
* having a lot of duplicated code.
3434
*
3435
* vma->vm_mm->mmap_lock must be held on entry.
3436
*
3437
* If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3438
* may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3439
*
3440
* If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3441
* has not been released.
3442
*
3443
* We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3444
*
3445
* Return: bitwise-OR of %VM_FAULT_ codes.
3446
*/
3447
vm_fault_t filemap_fault(struct vm_fault *vmf)
3448
{
3449
int error;
3450
struct file *file = vmf->vma->vm_file;
3451
struct file *fpin = NULL;
3452
struct address_space *mapping = file->f_mapping;
3453
struct inode *inode = mapping->host;
3454
pgoff_t max_idx, index = vmf->pgoff;
3455
struct folio *folio;
3456
vm_fault_t ret = 0;
3457
bool mapping_locked = false;
3458
3459
max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3460
if (unlikely(index >= max_idx))
3461
return VM_FAULT_SIGBUS;
3462
3463
trace_mm_filemap_fault(mapping, index);
3464
3465
/*
3466
* Do we have something in the page cache already?
3467
*/
3468
folio = filemap_get_folio(mapping, index);
3469
if (likely(!IS_ERR(folio))) {
3470
/*
3471
* We found the page, so try async readahead before waiting for
3472
* the lock.
3473
*/
3474
if (!(vmf->flags & FAULT_FLAG_TRIED))
3475
fpin = do_async_mmap_readahead(vmf, folio);
3476
if (unlikely(!folio_test_uptodate(folio))) {
3477
filemap_invalidate_lock_shared(mapping);
3478
mapping_locked = true;
3479
}
3480
} else {
3481
ret = filemap_fault_recheck_pte_none(vmf);
3482
if (unlikely(ret))
3483
return ret;
3484
3485
/* No page in the page cache at all */
3486
count_vm_event(PGMAJFAULT);
3487
count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3488
ret = VM_FAULT_MAJOR;
3489
fpin = do_sync_mmap_readahead(vmf);
3490
retry_find:
3491
/*
3492
* See comment in filemap_create_folio() why we need
3493
* invalidate_lock
3494
*/
3495
if (!mapping_locked) {
3496
filemap_invalidate_lock_shared(mapping);
3497
mapping_locked = true;
3498
}
3499
folio = __filemap_get_folio(mapping, index,
3500
FGP_CREAT|FGP_FOR_MMAP,
3501
vmf->gfp_mask);
3502
if (IS_ERR(folio)) {
3503
if (fpin)
3504
goto out_retry;
3505
filemap_invalidate_unlock_shared(mapping);
3506
return VM_FAULT_OOM;
3507
}
3508
}
3509
3510
if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3511
goto out_retry;
3512
3513
/* Did it get truncated? */
3514
if (unlikely(folio->mapping != mapping)) {
3515
folio_unlock(folio);
3516
folio_put(folio);
3517
goto retry_find;
3518
}
3519
VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3520
3521
/*
3522
* We have a locked folio in the page cache, now we need to check
3523
* that it's up-to-date. If not, it is going to be due to an error,
3524
* or because readahead was otherwise unable to retrieve it.
3525
*/
3526
if (unlikely(!folio_test_uptodate(folio))) {
3527
/*
3528
* If the invalidate lock is not held, the folio was in cache
3529
* and uptodate and now it is not. Strange but possible since we
3530
* didn't hold the page lock all the time. Let's drop
3531
* everything, get the invalidate lock and try again.
3532
*/
3533
if (!mapping_locked) {
3534
folio_unlock(folio);
3535
folio_put(folio);
3536
goto retry_find;
3537
}
3538
3539
/*
3540
* OK, the folio is really not uptodate. This can be because the
3541
* VMA has the VM_RAND_READ flag set, or because an error
3542
* arose. Let's read it in directly.
3543
*/
3544
goto page_not_uptodate;
3545
}
3546
3547
/*
3548
* We've made it this far and we had to drop our mmap_lock, now is the
3549
* time to return to the upper layer and have it re-find the vma and
3550
* redo the fault.
3551
*/
3552
if (fpin) {
3553
folio_unlock(folio);
3554
goto out_retry;
3555
}
3556
if (mapping_locked)
3557
filemap_invalidate_unlock_shared(mapping);
3558
3559
/*
3560
* Found the page and have a reference on it.
3561
* We must recheck i_size under page lock.
3562
*/
3563
max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3564
if (unlikely(index >= max_idx)) {
3565
folio_unlock(folio);
3566
folio_put(folio);
3567
return VM_FAULT_SIGBUS;
3568
}
3569
3570
vmf->page = folio_file_page(folio, index);
3571
return ret | VM_FAULT_LOCKED;
3572
3573
page_not_uptodate:
3574
/*
3575
* Umm, take care of errors if the page isn't up-to-date.
3576
* Try to re-read it _once_. We do this synchronously,
3577
* because there really aren't any performance issues here
3578
* and we need to check for errors.
3579
*/
3580
fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3581
error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3582
if (fpin)
3583
goto out_retry;
3584
folio_put(folio);
3585
3586
if (!error || error == AOP_TRUNCATED_PAGE)
3587
goto retry_find;
3588
filemap_invalidate_unlock_shared(mapping);
3589
3590
return VM_FAULT_SIGBUS;
3591
3592
out_retry:
3593
/*
3594
* We dropped the mmap_lock, we need to return to the fault handler to
3595
* re-find the vma and come back and find our hopefully still populated
3596
* page.
3597
*/
3598
if (!IS_ERR(folio))
3599
folio_put(folio);
3600
if (mapping_locked)
3601
filemap_invalidate_unlock_shared(mapping);
3602
if (fpin)
3603
fput(fpin);
3604
return ret | VM_FAULT_RETRY;
3605
}
3606
EXPORT_SYMBOL(filemap_fault);
3607
3608
static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3609
pgoff_t start)
3610
{
3611
struct mm_struct *mm = vmf->vma->vm_mm;
3612
3613
/* Huge page is mapped? No need to proceed. */
3614
if (pmd_trans_huge(*vmf->pmd)) {
3615
folio_unlock(folio);
3616
folio_put(folio);
3617
return true;
3618
}
3619
3620
if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3621
struct page *page = folio_file_page(folio, start);
3622
vm_fault_t ret = do_set_pmd(vmf, folio, page);
3623
if (!ret) {
3624
/* The page is mapped successfully, reference consumed. */
3625
folio_unlock(folio);
3626
return true;
3627
}
3628
}
3629
3630
if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
3631
pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3632
3633
return false;
3634
}
3635
3636
static struct folio *next_uptodate_folio(struct xa_state *xas,
3637
struct address_space *mapping, pgoff_t end_pgoff)
3638
{
3639
struct folio *folio = xas_next_entry(xas, end_pgoff);
3640
unsigned long max_idx;
3641
3642
do {
3643
if (!folio)
3644
return NULL;
3645
if (xas_retry(xas, folio))
3646
continue;
3647
if (xa_is_value(folio))
3648
continue;
3649
if (!folio_try_get(folio))
3650
continue;
3651
if (folio_test_locked(folio))
3652
goto skip;
3653
/* Has the page moved or been split? */
3654
if (unlikely(folio != xas_reload(xas)))
3655
goto skip;
3656
if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3657
goto skip;
3658
if (!folio_trylock(folio))
3659
goto skip;
3660
if (folio->mapping != mapping)
3661
goto unlock;
3662
if (!folio_test_uptodate(folio))
3663
goto unlock;
3664
max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3665
if (xas->xa_index >= max_idx)
3666
goto unlock;
3667
return folio;
3668
unlock:
3669
folio_unlock(folio);
3670
skip:
3671
folio_put(folio);
3672
} while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3673
3674
return NULL;
3675
}
3676
3677
/*
3678
* Map page range [start_page, start_page + nr_pages) of folio.
3679
* start_page is gotten from start by folio_page(folio, start)
3680
*/
3681
static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
3682
struct folio *folio, unsigned long start,
3683
unsigned long addr, unsigned int nr_pages,
3684
unsigned long *rss, unsigned short *mmap_miss)
3685
{
3686
unsigned int ref_from_caller = 1;
3687
vm_fault_t ret = 0;
3688
struct page *page = folio_page(folio, start);
3689
unsigned int count = 0;
3690
pte_t *old_ptep = vmf->pte;
3691
unsigned long addr0;
3692
3693
/*
3694
* Map the large folio fully where possible.
3695
*
3696
* The folio must not cross VMA or page table boundary.
3697
*/
3698
addr0 = addr - start * PAGE_SIZE;
3699
if (folio_within_vma(folio, vmf->vma) &&
3700
(addr0 & PMD_MASK) == ((addr0 + folio_size(folio) - 1) & PMD_MASK)) {
3701
vmf->pte -= start;
3702
page -= start;
3703
addr = addr0;
3704
nr_pages = folio_nr_pages(folio);
3705
}
3706
3707
do {
3708
if (PageHWPoison(page + count))
3709
goto skip;
3710
3711
/*
3712
* If there are too many folios that are recently evicted
3713
* in a file, they will probably continue to be evicted.
3714
* In such situation, read-ahead is only a waste of IO.
3715
* Don't decrease mmap_miss in this scenario to make sure
3716
* we can stop read-ahead.
3717
*/
3718
if (!folio_test_workingset(folio))
3719
(*mmap_miss)++;
3720
3721
/*
3722
* NOTE: If there're PTE markers, we'll leave them to be
3723
* handled in the specific fault path, and it'll prohibit the
3724
* fault-around logic.
3725
*/
3726
if (!pte_none(ptep_get(&vmf->pte[count])))
3727
goto skip;
3728
3729
count++;
3730
continue;
3731
skip:
3732
if (count) {
3733
set_pte_range(vmf, folio, page, count, addr);
3734
*rss += count;
3735
folio_ref_add(folio, count - ref_from_caller);
3736
ref_from_caller = 0;
3737
if (in_range(vmf->address, addr, count * PAGE_SIZE))
3738
ret = VM_FAULT_NOPAGE;
3739
}
3740
3741
count++;
3742
page += count;
3743
vmf->pte += count;
3744
addr += count * PAGE_SIZE;
3745
count = 0;
3746
} while (--nr_pages > 0);
3747
3748
if (count) {
3749
set_pte_range(vmf, folio, page, count, addr);
3750
*rss += count;
3751
folio_ref_add(folio, count - ref_from_caller);
3752
ref_from_caller = 0;
3753
if (in_range(vmf->address, addr, count * PAGE_SIZE))
3754
ret = VM_FAULT_NOPAGE;
3755
}
3756
3757
vmf->pte = old_ptep;
3758
if (ref_from_caller)
3759
/* Locked folios cannot get truncated. */
3760
folio_ref_dec(folio);
3761
3762
return ret;
3763
}
3764
3765
static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
3766
struct folio *folio, unsigned long addr,
3767
unsigned long *rss, unsigned short *mmap_miss)
3768
{
3769
vm_fault_t ret = 0;
3770
struct page *page = &folio->page;
3771
3772
if (PageHWPoison(page))
3773
goto out;
3774
3775
/* See comment of filemap_map_folio_range() */
3776
if (!folio_test_workingset(folio))
3777
(*mmap_miss)++;
3778
3779
/*
3780
* NOTE: If there're PTE markers, we'll leave them to be
3781
* handled in the specific fault path, and it'll prohibit
3782
* the fault-around logic.
3783
*/
3784
if (!pte_none(ptep_get(vmf->pte)))
3785
goto out;
3786
3787
if (vmf->address == addr)
3788
ret = VM_FAULT_NOPAGE;
3789
3790
set_pte_range(vmf, folio, page, 1, addr);
3791
(*rss)++;
3792
return ret;
3793
3794
out:
3795
/* Locked folios cannot get truncated. */
3796
folio_ref_dec(folio);
3797
return ret;
3798
}
3799
3800
vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3801
pgoff_t start_pgoff, pgoff_t end_pgoff)
3802
{
3803
struct vm_area_struct *vma = vmf->vma;
3804
struct file *file = vma->vm_file;
3805
struct address_space *mapping = file->f_mapping;
3806
pgoff_t file_end, last_pgoff = start_pgoff;
3807
unsigned long addr;
3808
XA_STATE(xas, &mapping->i_pages, start_pgoff);
3809
struct folio *folio;
3810
vm_fault_t ret = 0;
3811
unsigned long rss = 0;
3812
unsigned int nr_pages = 0, folio_type;
3813
unsigned short mmap_miss = 0, mmap_miss_saved;
3814
3815
rcu_read_lock();
3816
folio = next_uptodate_folio(&xas, mapping, end_pgoff);
3817
if (!folio)
3818
goto out;
3819
3820
if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3821
ret = VM_FAULT_NOPAGE;
3822
goto out;
3823
}
3824
3825
addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3826
vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3827
if (!vmf->pte) {
3828
folio_unlock(folio);
3829
folio_put(folio);
3830
goto out;
3831
}
3832
3833
file_end = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE) - 1;
3834
if (end_pgoff > file_end)
3835
end_pgoff = file_end;
3836
3837
folio_type = mm_counter_file(folio);
3838
do {
3839
unsigned long end;
3840
3841
addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3842
vmf->pte += xas.xa_index - last_pgoff;
3843
last_pgoff = xas.xa_index;
3844
end = folio_next_index(folio) - 1;
3845
nr_pages = min(end, end_pgoff) - xas.xa_index + 1;
3846
3847
if (!folio_test_large(folio))
3848
ret |= filemap_map_order0_folio(vmf,
3849
folio, addr, &rss, &mmap_miss);
3850
else
3851
ret |= filemap_map_folio_range(vmf, folio,
3852
xas.xa_index - folio->index, addr,
3853
nr_pages, &rss, &mmap_miss);
3854
3855
folio_unlock(folio);
3856
} while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
3857
add_mm_counter(vma->vm_mm, folio_type, rss);
3858
pte_unmap_unlock(vmf->pte, vmf->ptl);
3859
trace_mm_filemap_map_pages(mapping, start_pgoff, end_pgoff);
3860
out:
3861
rcu_read_unlock();
3862
3863
mmap_miss_saved = READ_ONCE(file->f_ra.mmap_miss);
3864
if (mmap_miss >= mmap_miss_saved)
3865
WRITE_ONCE(file->f_ra.mmap_miss, 0);
3866
else
3867
WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss_saved - mmap_miss);
3868
3869
return ret;
3870
}
3871
EXPORT_SYMBOL(filemap_map_pages);
3872
3873
vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3874
{
3875
struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3876
struct folio *folio = page_folio(vmf->page);
3877
vm_fault_t ret = VM_FAULT_LOCKED;
3878
3879
sb_start_pagefault(mapping->host->i_sb);
3880
file_update_time(vmf->vma->vm_file);
3881
folio_lock(folio);
3882
if (folio->mapping != mapping) {
3883
folio_unlock(folio);
3884
ret = VM_FAULT_NOPAGE;
3885
goto out;
3886
}
3887
/*
3888
* We mark the folio dirty already here so that when freeze is in
3889
* progress, we are guaranteed that writeback during freezing will
3890
* see the dirty folio and writeprotect it again.
3891
*/
3892
folio_mark_dirty(folio);
3893
folio_wait_stable(folio);
3894
out:
3895
sb_end_pagefault(mapping->host->i_sb);
3896
return ret;
3897
}
3898
3899
const struct vm_operations_struct generic_file_vm_ops = {
3900
.fault = filemap_fault,
3901
.map_pages = filemap_map_pages,
3902
.page_mkwrite = filemap_page_mkwrite,
3903
};
3904
3905
/* This is used for a general mmap of a disk file */
3906
3907
int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3908
{
3909
struct address_space *mapping = file->f_mapping;
3910
3911
if (!mapping->a_ops->read_folio)
3912
return -ENOEXEC;
3913
file_accessed(file);
3914
vma->vm_ops = &generic_file_vm_ops;
3915
return 0;
3916
}
3917
3918
int generic_file_mmap_prepare(struct vm_area_desc *desc)
3919
{
3920
struct file *file = desc->file;
3921
struct address_space *mapping = file->f_mapping;
3922
3923
if (!mapping->a_ops->read_folio)
3924
return -ENOEXEC;
3925
file_accessed(file);
3926
desc->vm_ops = &generic_file_vm_ops;
3927
return 0;
3928
}
3929
3930
/*
3931
* This is for filesystems which do not implement ->writepage.
3932
*/
3933
int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3934
{
3935
if (vma_is_shared_maywrite(vma))
3936
return -EINVAL;
3937
return generic_file_mmap(file, vma);
3938
}
3939
3940
int generic_file_readonly_mmap_prepare(struct vm_area_desc *desc)
3941
{
3942
if (is_shared_maywrite(desc->vm_flags))
3943
return -EINVAL;
3944
return generic_file_mmap_prepare(desc);
3945
}
3946
#else
3947
vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3948
{
3949
return VM_FAULT_SIGBUS;
3950
}
3951
int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3952
{
3953
return -ENOSYS;
3954
}
3955
int generic_file_mmap_prepare(struct vm_area_desc *desc)
3956
{
3957
return -ENOSYS;
3958
}
3959
int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3960
{
3961
return -ENOSYS;
3962
}
3963
int generic_file_readonly_mmap_prepare(struct vm_area_desc *desc)
3964
{
3965
return -ENOSYS;
3966
}
3967
#endif /* CONFIG_MMU */
3968
3969
EXPORT_SYMBOL(filemap_page_mkwrite);
3970
EXPORT_SYMBOL(generic_file_mmap);
3971
EXPORT_SYMBOL(generic_file_mmap_prepare);
3972
EXPORT_SYMBOL(generic_file_readonly_mmap);
3973
EXPORT_SYMBOL(generic_file_readonly_mmap_prepare);
3974
3975
static struct folio *do_read_cache_folio(struct address_space *mapping,
3976
pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3977
{
3978
struct folio *folio;
3979
int err;
3980
3981
if (!filler)
3982
filler = mapping->a_ops->read_folio;
3983
repeat:
3984
folio = filemap_get_folio(mapping, index);
3985
if (IS_ERR(folio)) {
3986
folio = filemap_alloc_folio(gfp,
3987
mapping_min_folio_order(mapping));
3988
if (!folio)
3989
return ERR_PTR(-ENOMEM);
3990
index = mapping_align_index(mapping, index);
3991
err = filemap_add_folio(mapping, folio, index, gfp);
3992
if (unlikely(err)) {
3993
folio_put(folio);
3994
if (err == -EEXIST)
3995
goto repeat;
3996
/* Presumably ENOMEM for xarray node */
3997
return ERR_PTR(err);
3998
}
3999
4000
goto filler;
4001
}
4002
if (folio_test_uptodate(folio))
4003
goto out;
4004
4005
if (!folio_trylock(folio)) {
4006
folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
4007
goto repeat;
4008
}
4009
4010
/* Folio was truncated from mapping */
4011
if (!folio->mapping) {
4012
folio_unlock(folio);
4013
folio_put(folio);
4014
goto repeat;
4015
}
4016
4017
/* Someone else locked and filled the page in a very small window */
4018
if (folio_test_uptodate(folio)) {
4019
folio_unlock(folio);
4020
goto out;
4021
}
4022
4023
filler:
4024
err = filemap_read_folio(file, filler, folio);
4025
if (err) {
4026
folio_put(folio);
4027
if (err == AOP_TRUNCATED_PAGE)
4028
goto repeat;
4029
return ERR_PTR(err);
4030
}
4031
4032
out:
4033
folio_mark_accessed(folio);
4034
return folio;
4035
}
4036
4037
/**
4038
* read_cache_folio - Read into page cache, fill it if needed.
4039
* @mapping: The address_space to read from.
4040
* @index: The index to read.
4041
* @filler: Function to perform the read, or NULL to use aops->read_folio().
4042
* @file: Passed to filler function, may be NULL if not required.
4043
*
4044
* Read one page into the page cache. If it succeeds, the folio returned
4045
* will contain @index, but it may not be the first page of the folio.
4046
*
4047
* If the filler function returns an error, it will be returned to the
4048
* caller.
4049
*
4050
* Context: May sleep. Expects mapping->invalidate_lock to be held.
4051
* Return: An uptodate folio on success, ERR_PTR() on failure.
4052
*/
4053
struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
4054
filler_t filler, struct file *file)
4055
{
4056
return do_read_cache_folio(mapping, index, filler, file,
4057
mapping_gfp_mask(mapping));
4058
}
4059
EXPORT_SYMBOL(read_cache_folio);
4060
4061
/**
4062
* mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
4063
* @mapping: The address_space for the folio.
4064
* @index: The index that the allocated folio will contain.
4065
* @gfp: The page allocator flags to use if allocating.
4066
*
4067
* This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
4068
* any new memory allocations done using the specified allocation flags.
4069
*
4070
* The most likely error from this function is EIO, but ENOMEM is
4071
* possible and so is EINTR. If ->read_folio returns another error,
4072
* that will be returned to the caller.
4073
*
4074
* The function expects mapping->invalidate_lock to be already held.
4075
*
4076
* Return: Uptodate folio on success, ERR_PTR() on failure.
4077
*/
4078
struct folio *mapping_read_folio_gfp(struct address_space *mapping,
4079
pgoff_t index, gfp_t gfp)
4080
{
4081
return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
4082
}
4083
EXPORT_SYMBOL(mapping_read_folio_gfp);
4084
4085
static struct page *do_read_cache_page(struct address_space *mapping,
4086
pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
4087
{
4088
struct folio *folio;
4089
4090
folio = do_read_cache_folio(mapping, index, filler, file, gfp);
4091
if (IS_ERR(folio))
4092
return &folio->page;
4093
return folio_file_page(folio, index);
4094
}
4095
4096
struct page *read_cache_page(struct address_space *mapping,
4097
pgoff_t index, filler_t *filler, struct file *file)
4098
{
4099
return do_read_cache_page(mapping, index, filler, file,
4100
mapping_gfp_mask(mapping));
4101
}
4102
EXPORT_SYMBOL(read_cache_page);
4103
4104
/**
4105
* read_cache_page_gfp - read into page cache, using specified page allocation flags.
4106
* @mapping: the page's address_space
4107
* @index: the page index
4108
* @gfp: the page allocator flags to use if allocating
4109
*
4110
* This is the same as "read_mapping_page(mapping, index, NULL)", but with
4111
* any new page allocations done using the specified allocation flags.
4112
*
4113
* If the page does not get brought uptodate, return -EIO.
4114
*
4115
* The function expects mapping->invalidate_lock to be already held.
4116
*
4117
* Return: up to date page on success, ERR_PTR() on failure.
4118
*/
4119
struct page *read_cache_page_gfp(struct address_space *mapping,
4120
pgoff_t index,
4121
gfp_t gfp)
4122
{
4123
return do_read_cache_page(mapping, index, NULL, NULL, gfp);
4124
}
4125
EXPORT_SYMBOL(read_cache_page_gfp);
4126
4127
/*
4128
* Warn about a page cache invalidation failure during a direct I/O write.
4129
*/
4130
static void dio_warn_stale_pagecache(struct file *filp)
4131
{
4132
static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
4133
char pathname[128];
4134
char *path;
4135
4136
errseq_set(&filp->f_mapping->wb_err, -EIO);
4137
if (__ratelimit(&_rs)) {
4138
path = file_path(filp, pathname, sizeof(pathname));
4139
if (IS_ERR(path))
4140
path = "(unknown)";
4141
pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
4142
pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
4143
current->comm);
4144
}
4145
}
4146
4147
void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
4148
{
4149
struct address_space *mapping = iocb->ki_filp->f_mapping;
4150
4151
if (mapping->nrpages &&
4152
invalidate_inode_pages2_range(mapping,
4153
iocb->ki_pos >> PAGE_SHIFT,
4154
(iocb->ki_pos + count - 1) >> PAGE_SHIFT))
4155
dio_warn_stale_pagecache(iocb->ki_filp);
4156
}
4157
4158
ssize_t
4159
generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
4160
{
4161
struct address_space *mapping = iocb->ki_filp->f_mapping;
4162
size_t write_len = iov_iter_count(from);
4163
ssize_t written;
4164
4165
/*
4166
* If a page can not be invalidated, return 0 to fall back
4167
* to buffered write.
4168
*/
4169
written = kiocb_invalidate_pages(iocb, write_len);
4170
if (written) {
4171
if (written == -EBUSY)
4172
return 0;
4173
return written;
4174
}
4175
4176
written = mapping->a_ops->direct_IO(iocb, from);
4177
4178
/*
4179
* Finally, try again to invalidate clean pages which might have been
4180
* cached by non-direct readahead, or faulted in by get_user_pages()
4181
* if the source of the write was an mmap'ed region of the file
4182
* we're writing. Either one is a pretty crazy thing to do,
4183
* so we don't support it 100%. If this invalidation
4184
* fails, tough, the write still worked...
4185
*
4186
* Most of the time we do not need this since dio_complete() will do
4187
* the invalidation for us. However there are some file systems that
4188
* do not end up with dio_complete() being called, so let's not break
4189
* them by removing it completely.
4190
*
4191
* Noticeable example is a blkdev_direct_IO().
4192
*
4193
* Skip invalidation for async writes or if mapping has no pages.
4194
*/
4195
if (written > 0) {
4196
struct inode *inode = mapping->host;
4197
loff_t pos = iocb->ki_pos;
4198
4199
kiocb_invalidate_post_direct_write(iocb, written);
4200
pos += written;
4201
write_len -= written;
4202
if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
4203
i_size_write(inode, pos);
4204
mark_inode_dirty(inode);
4205
}
4206
iocb->ki_pos = pos;
4207
}
4208
if (written != -EIOCBQUEUED)
4209
iov_iter_revert(from, write_len - iov_iter_count(from));
4210
return written;
4211
}
4212
EXPORT_SYMBOL(generic_file_direct_write);
4213
4214
ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
4215
{
4216
struct file *file = iocb->ki_filp;
4217
loff_t pos = iocb->ki_pos;
4218
struct address_space *mapping = file->f_mapping;
4219
const struct address_space_operations *a_ops = mapping->a_ops;
4220
size_t chunk = mapping_max_folio_size(mapping);
4221
long status = 0;
4222
ssize_t written = 0;
4223
4224
do {
4225
struct folio *folio;
4226
size_t offset; /* Offset into folio */
4227
size_t bytes; /* Bytes to write to folio */
4228
size_t copied; /* Bytes copied from user */
4229
void *fsdata = NULL;
4230
4231
bytes = iov_iter_count(i);
4232
retry:
4233
offset = pos & (chunk - 1);
4234
bytes = min(chunk - offset, bytes);
4235
balance_dirty_pages_ratelimited(mapping);
4236
4237
if (fatal_signal_pending(current)) {
4238
status = -EINTR;
4239
break;
4240
}
4241
4242
status = a_ops->write_begin(iocb, mapping, pos, bytes,
4243
&folio, &fsdata);
4244
if (unlikely(status < 0))
4245
break;
4246
4247
offset = offset_in_folio(folio, pos);
4248
if (bytes > folio_size(folio) - offset)
4249
bytes = folio_size(folio) - offset;
4250
4251
if (mapping_writably_mapped(mapping))
4252
flush_dcache_folio(folio);
4253
4254
/*
4255
* Faults here on mmap()s can recurse into arbitrary
4256
* filesystem code. Lots of locks are held that can
4257
* deadlock. Use an atomic copy to avoid deadlocking
4258
* in page fault handling.
4259
*/
4260
copied = copy_folio_from_iter_atomic(folio, offset, bytes, i);
4261
flush_dcache_folio(folio);
4262
4263
status = a_ops->write_end(iocb, mapping, pos, bytes, copied,
4264
folio, fsdata);
4265
if (unlikely(status != copied)) {
4266
iov_iter_revert(i, copied - max(status, 0L));
4267
if (unlikely(status < 0))
4268
break;
4269
}
4270
cond_resched();
4271
4272
if (unlikely(status == 0)) {
4273
/*
4274
* A short copy made ->write_end() reject the
4275
* thing entirely. Might be memory poisoning
4276
* halfway through, might be a race with munmap,
4277
* might be severe memory pressure.
4278
*/
4279
if (chunk > PAGE_SIZE)
4280
chunk /= 2;
4281
if (copied) {
4282
bytes = copied;
4283
goto retry;
4284
}
4285
4286
/*
4287
* 'folio' is now unlocked and faults on it can be
4288
* handled. Ensure forward progress by trying to
4289
* fault it in now.
4290
*/
4291
if (fault_in_iov_iter_readable(i, bytes) == bytes) {
4292
status = -EFAULT;
4293
break;
4294
}
4295
} else {
4296
pos += status;
4297
written += status;
4298
}
4299
} while (iov_iter_count(i));
4300
4301
if (!written)
4302
return status;
4303
iocb->ki_pos += written;
4304
return written;
4305
}
4306
EXPORT_SYMBOL(generic_perform_write);
4307
4308
/**
4309
* __generic_file_write_iter - write data to a file
4310
* @iocb: IO state structure (file, offset, etc.)
4311
* @from: iov_iter with data to write
4312
*
4313
* This function does all the work needed for actually writing data to a
4314
* file. It does all basic checks, removes SUID from the file, updates
4315
* modification times and calls proper subroutines depending on whether we
4316
* do direct IO or a standard buffered write.
4317
*
4318
* It expects i_rwsem to be grabbed unless we work on a block device or similar
4319
* object which does not need locking at all.
4320
*
4321
* This function does *not* take care of syncing data in case of O_SYNC write.
4322
* A caller has to handle it. This is mainly due to the fact that we want to
4323
* avoid syncing under i_rwsem.
4324
*
4325
* Return:
4326
* * number of bytes written, even for truncated writes
4327
* * negative error code if no data has been written at all
4328
*/
4329
ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4330
{
4331
struct file *file = iocb->ki_filp;
4332
struct address_space *mapping = file->f_mapping;
4333
struct inode *inode = mapping->host;
4334
ssize_t ret;
4335
4336
ret = file_remove_privs(file);
4337
if (ret)
4338
return ret;
4339
4340
ret = file_update_time(file);
4341
if (ret)
4342
return ret;
4343
4344
if (iocb->ki_flags & IOCB_DIRECT) {
4345
ret = generic_file_direct_write(iocb, from);
4346
/*
4347
* If the write stopped short of completing, fall back to
4348
* buffered writes. Some filesystems do this for writes to
4349
* holes, for example. For DAX files, a buffered write will
4350
* not succeed (even if it did, DAX does not handle dirty
4351
* page-cache pages correctly).
4352
*/
4353
if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
4354
return ret;
4355
return direct_write_fallback(iocb, from, ret,
4356
generic_perform_write(iocb, from));
4357
}
4358
4359
return generic_perform_write(iocb, from);
4360
}
4361
EXPORT_SYMBOL(__generic_file_write_iter);
4362
4363
/**
4364
* generic_file_write_iter - write data to a file
4365
* @iocb: IO state structure
4366
* @from: iov_iter with data to write
4367
*
4368
* This is a wrapper around __generic_file_write_iter() to be used by most
4369
* filesystems. It takes care of syncing the file in case of O_SYNC file
4370
* and acquires i_rwsem as needed.
4371
* Return:
4372
* * negative error code if no data has been written at all of
4373
* vfs_fsync_range() failed for a synchronous write
4374
* * number of bytes written, even for truncated writes
4375
*/
4376
ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4377
{
4378
struct file *file = iocb->ki_filp;
4379
struct inode *inode = file->f_mapping->host;
4380
ssize_t ret;
4381
4382
inode_lock(inode);
4383
ret = generic_write_checks(iocb, from);
4384
if (ret > 0)
4385
ret = __generic_file_write_iter(iocb, from);
4386
inode_unlock(inode);
4387
4388
if (ret > 0)
4389
ret = generic_write_sync(iocb, ret);
4390
return ret;
4391
}
4392
EXPORT_SYMBOL(generic_file_write_iter);
4393
4394
/**
4395
* filemap_release_folio() - Release fs-specific metadata on a folio.
4396
* @folio: The folio which the kernel is trying to free.
4397
* @gfp: Memory allocation flags (and I/O mode).
4398
*
4399
* The address_space is trying to release any data attached to a folio
4400
* (presumably at folio->private).
4401
*
4402
* This will also be called if the private_2 flag is set on a page,
4403
* indicating that the folio has other metadata associated with it.
4404
*
4405
* The @gfp argument specifies whether I/O may be performed to release
4406
* this page (__GFP_IO), and whether the call may block
4407
* (__GFP_RECLAIM & __GFP_FS).
4408
*
4409
* Return: %true if the release was successful, otherwise %false.
4410
*/
4411
bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4412
{
4413
struct address_space * const mapping = folio->mapping;
4414
4415
BUG_ON(!folio_test_locked(folio));
4416
if (!folio_needs_release(folio))
4417
return true;
4418
if (folio_test_writeback(folio))
4419
return false;
4420
4421
if (mapping && mapping->a_ops->release_folio)
4422
return mapping->a_ops->release_folio(folio, gfp);
4423
return try_to_free_buffers(folio);
4424
}
4425
EXPORT_SYMBOL(filemap_release_folio);
4426
4427
/**
4428
* filemap_invalidate_inode - Invalidate/forcibly write back a range of an inode's pagecache
4429
* @inode: The inode to flush
4430
* @flush: Set to write back rather than simply invalidate.
4431
* @start: First byte to in range.
4432
* @end: Last byte in range (inclusive), or LLONG_MAX for everything from start
4433
* onwards.
4434
*
4435
* Invalidate all the folios on an inode that contribute to the specified
4436
* range, possibly writing them back first. Whilst the operation is
4437
* undertaken, the invalidate lock is held to prevent new folios from being
4438
* installed.
4439
*/
4440
int filemap_invalidate_inode(struct inode *inode, bool flush,
4441
loff_t start, loff_t end)
4442
{
4443
struct address_space *mapping = inode->i_mapping;
4444
pgoff_t first = start >> PAGE_SHIFT;
4445
pgoff_t last = end >> PAGE_SHIFT;
4446
pgoff_t nr = end == LLONG_MAX ? ULONG_MAX : last - first + 1;
4447
4448
if (!mapping || !mapping->nrpages || end < start)
4449
goto out;
4450
4451
/* Prevent new folios from being added to the inode. */
4452
filemap_invalidate_lock(mapping);
4453
4454
if (!mapping->nrpages)
4455
goto unlock;
4456
4457
unmap_mapping_pages(mapping, first, nr, false);
4458
4459
/* Write back the data if we're asked to. */
4460
if (flush) {
4461
struct writeback_control wbc = {
4462
.sync_mode = WB_SYNC_ALL,
4463
.nr_to_write = LONG_MAX,
4464
.range_start = start,
4465
.range_end = end,
4466
};
4467
4468
filemap_fdatawrite_wbc(mapping, &wbc);
4469
}
4470
4471
/* Wait for writeback to complete on all folios and discard. */
4472
invalidate_inode_pages2_range(mapping, start / PAGE_SIZE, end / PAGE_SIZE);
4473
4474
unlock:
4475
filemap_invalidate_unlock(mapping);
4476
out:
4477
return filemap_check_errors(mapping);
4478
}
4479
EXPORT_SYMBOL_GPL(filemap_invalidate_inode);
4480
4481
#ifdef CONFIG_CACHESTAT_SYSCALL
4482
/**
4483
* filemap_cachestat() - compute the page cache statistics of a mapping
4484
* @mapping: The mapping to compute the statistics for.
4485
* @first_index: The starting page cache index.
4486
* @last_index: The final page index (inclusive).
4487
* @cs: the cachestat struct to write the result to.
4488
*
4489
* This will query the page cache statistics of a mapping in the
4490
* page range of [first_index, last_index] (inclusive). The statistics
4491
* queried include: number of dirty pages, number of pages marked for
4492
* writeback, and the number of (recently) evicted pages.
4493
*/
4494
static void filemap_cachestat(struct address_space *mapping,
4495
pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
4496
{
4497
XA_STATE(xas, &mapping->i_pages, first_index);
4498
struct folio *folio;
4499
4500
/* Flush stats (and potentially sleep) outside the RCU read section. */
4501
mem_cgroup_flush_stats_ratelimited(NULL);
4502
4503
rcu_read_lock();
4504
xas_for_each(&xas, folio, last_index) {
4505
int order;
4506
unsigned long nr_pages;
4507
pgoff_t folio_first_index, folio_last_index;
4508
4509
/*
4510
* Don't deref the folio. It is not pinned, and might
4511
* get freed (and reused) underneath us.
4512
*
4513
* We *could* pin it, but that would be expensive for
4514
* what should be a fast and lightweight syscall.
4515
*
4516
* Instead, derive all information of interest from
4517
* the rcu-protected xarray.
4518
*/
4519
4520
if (xas_retry(&xas, folio))
4521
continue;
4522
4523
order = xas_get_order(&xas);
4524
nr_pages = 1 << order;
4525
folio_first_index = round_down(xas.xa_index, 1 << order);
4526
folio_last_index = folio_first_index + nr_pages - 1;
4527
4528
/* Folios might straddle the range boundaries, only count covered pages */
4529
if (folio_first_index < first_index)
4530
nr_pages -= first_index - folio_first_index;
4531
4532
if (folio_last_index > last_index)
4533
nr_pages -= folio_last_index - last_index;
4534
4535
if (xa_is_value(folio)) {
4536
/* page is evicted */
4537
void *shadow = (void *)folio;
4538
bool workingset; /* not used */
4539
4540
cs->nr_evicted += nr_pages;
4541
4542
#ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4543
if (shmem_mapping(mapping)) {
4544
/* shmem file - in swap cache */
4545
swp_entry_t swp = radix_to_swp_entry(folio);
4546
4547
/* swapin error results in poisoned entry */
4548
if (non_swap_entry(swp))
4549
goto resched;
4550
4551
/*
4552
* Getting a swap entry from the shmem
4553
* inode means we beat
4554
* shmem_unuse(). rcu_read_lock()
4555
* ensures swapoff waits for us before
4556
* freeing the swapper space. However,
4557
* we can race with swapping and
4558
* invalidation, so there might not be
4559
* a shadow in the swapcache (yet).
4560
*/
4561
shadow = swap_cache_get_shadow(swp);
4562
if (!shadow)
4563
goto resched;
4564
}
4565
#endif
4566
if (workingset_test_recent(shadow, true, &workingset, false))
4567
cs->nr_recently_evicted += nr_pages;
4568
4569
goto resched;
4570
}
4571
4572
/* page is in cache */
4573
cs->nr_cache += nr_pages;
4574
4575
if (xas_get_mark(&xas, PAGECACHE_TAG_DIRTY))
4576
cs->nr_dirty += nr_pages;
4577
4578
if (xas_get_mark(&xas, PAGECACHE_TAG_WRITEBACK))
4579
cs->nr_writeback += nr_pages;
4580
4581
resched:
4582
if (need_resched()) {
4583
xas_pause(&xas);
4584
cond_resched_rcu();
4585
}
4586
}
4587
rcu_read_unlock();
4588
}
4589
4590
/*
4591
* See mincore: reveal pagecache information only for files
4592
* that the calling process has write access to, or could (if
4593
* tried) open for writing.
4594
*/
4595
static inline bool can_do_cachestat(struct file *f)
4596
{
4597
if (f->f_mode & FMODE_WRITE)
4598
return true;
4599
if (inode_owner_or_capable(file_mnt_idmap(f), file_inode(f)))
4600
return true;
4601
return file_permission(f, MAY_WRITE) == 0;
4602
}
4603
4604
/*
4605
* The cachestat(2) system call.
4606
*
4607
* cachestat() returns the page cache statistics of a file in the
4608
* bytes range specified by `off` and `len`: number of cached pages,
4609
* number of dirty pages, number of pages marked for writeback,
4610
* number of evicted pages, and number of recently evicted pages.
4611
*
4612
* An evicted page is a page that is previously in the page cache
4613
* but has been evicted since. A page is recently evicted if its last
4614
* eviction was recent enough that its reentry to the cache would
4615
* indicate that it is actively being used by the system, and that
4616
* there is memory pressure on the system.
4617
*
4618
* `off` and `len` must be non-negative integers. If `len` > 0,
4619
* the queried range is [`off`, `off` + `len`]. If `len` == 0,
4620
* we will query in the range from `off` to the end of the file.
4621
*
4622
* The `flags` argument is unused for now, but is included for future
4623
* extensibility. User should pass 0 (i.e no flag specified).
4624
*
4625
* Currently, hugetlbfs is not supported.
4626
*
4627
* Because the status of a page can change after cachestat() checks it
4628
* but before it returns to the application, the returned values may
4629
* contain stale information.
4630
*
4631
* return values:
4632
* zero - success
4633
* -EFAULT - cstat or cstat_range points to an illegal address
4634
* -EINVAL - invalid flags
4635
* -EBADF - invalid file descriptor
4636
* -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4637
*/
4638
SYSCALL_DEFINE4(cachestat, unsigned int, fd,
4639
struct cachestat_range __user *, cstat_range,
4640
struct cachestat __user *, cstat, unsigned int, flags)
4641
{
4642
CLASS(fd, f)(fd);
4643
struct address_space *mapping;
4644
struct cachestat_range csr;
4645
struct cachestat cs;
4646
pgoff_t first_index, last_index;
4647
4648
if (fd_empty(f))
4649
return -EBADF;
4650
4651
if (copy_from_user(&csr, cstat_range,
4652
sizeof(struct cachestat_range)))
4653
return -EFAULT;
4654
4655
/* hugetlbfs is not supported */
4656
if (is_file_hugepages(fd_file(f)))
4657
return -EOPNOTSUPP;
4658
4659
if (!can_do_cachestat(fd_file(f)))
4660
return -EPERM;
4661
4662
if (flags != 0)
4663
return -EINVAL;
4664
4665
first_index = csr.off >> PAGE_SHIFT;
4666
last_index =
4667
csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
4668
memset(&cs, 0, sizeof(struct cachestat));
4669
mapping = fd_file(f)->f_mapping;
4670
filemap_cachestat(mapping, first_index, last_index, &cs);
4671
4672
if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
4673
return -EFAULT;
4674
4675
return 0;
4676
}
4677
#endif /* CONFIG_CACHESTAT_SYSCALL */
4678
4679