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