1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_PAGEMAP_H
3#define _LINUX_PAGEMAP_H
4
5/*
6 * Copyright 1995 Linus Torvalds
7 */
8#include <linux/mm.h>
9#include <linux/fs.h>
10#include <linux/list.h>
11#include <linux/highmem.h>
12#include <linux/compiler.h>
13#include <linux/uaccess.h>
14#include <linux/gfp.h>
15#include <linux/bitops.h>
16#include <linux/hardirq.h> /* for in_interrupt() */
17#include <linux/hugetlb_inline.h>
18
19struct folio_batch;
20
21unsigned long invalidate_mapping_pages(struct address_space *mapping,
22 pgoff_t start, pgoff_t end);
23
24static inline void invalidate_remote_inode(struct inode *inode)
25{
26 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
27 S_ISLNK(inode->i_mode))
28 invalidate_mapping_pages(mapping: inode->i_mapping, start: 0, end: -1);
29}
30int invalidate_inode_pages2(struct address_space *mapping);
31int invalidate_inode_pages2_range(struct address_space *mapping,
32 pgoff_t start, pgoff_t end);
33int kiocb_invalidate_pages(struct kiocb *iocb, size_t count);
34void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count);
35int filemap_invalidate_pages(struct address_space *mapping,
36 loff_t pos, loff_t end, bool nowait);
37
38int write_inode_now(struct inode *, int sync);
39int filemap_fdatawrite(struct address_space *);
40int filemap_flush(struct address_space *);
41int filemap_fdatawait_keep_errors(struct address_space *mapping);
42int filemap_fdatawait_range(struct address_space *, loff_t lstart, loff_t lend);
43int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
44 loff_t start_byte, loff_t end_byte);
45int filemap_invalidate_inode(struct inode *inode, bool flush,
46 loff_t start, loff_t end);
47
48static inline int filemap_fdatawait(struct address_space *mapping)
49{
50 return filemap_fdatawait_range(mapping, lstart: 0, LLONG_MAX);
51}
52
53bool filemap_range_has_page(struct address_space *, loff_t lstart, loff_t lend);
54int filemap_write_and_wait_range(struct address_space *mapping,
55 loff_t lstart, loff_t lend);
56int __filemap_fdatawrite_range(struct address_space *mapping,
57 loff_t start, loff_t end, int sync_mode);
58int filemap_fdatawrite_range(struct address_space *mapping,
59 loff_t start, loff_t end);
60int filemap_check_errors(struct address_space *mapping);
61void __filemap_set_wb_err(struct address_space *mapping, int err);
62int filemap_fdatawrite_wbc(struct address_space *mapping,
63 struct writeback_control *wbc);
64int kiocb_write_and_wait(struct kiocb *iocb, size_t count);
65
66static inline int filemap_write_and_wait(struct address_space *mapping)
67{
68 return filemap_write_and_wait_range(mapping, lstart: 0, LLONG_MAX);
69}
70
71/**
72 * filemap_set_wb_err - set a writeback error on an address_space
73 * @mapping: mapping in which to set writeback error
74 * @err: error to be set in mapping
75 *
76 * When writeback fails in some way, we must record that error so that
77 * userspace can be informed when fsync and the like are called. We endeavor
78 * to report errors on any file that was open at the time of the error. Some
79 * internal callers also need to know when writeback errors have occurred.
80 *
81 * When a writeback error occurs, most filesystems will want to call
82 * filemap_set_wb_err to record the error in the mapping so that it will be
83 * automatically reported whenever fsync is called on the file.
84 */
85static inline void filemap_set_wb_err(struct address_space *mapping, int err)
86{
87 /* Fastpath for common case of no error */
88 if (unlikely(err))
89 __filemap_set_wb_err(mapping, err);
90}
91
92/**
93 * filemap_check_wb_err - has an error occurred since the mark was sampled?
94 * @mapping: mapping to check for writeback errors
95 * @since: previously-sampled errseq_t
96 *
97 * Grab the errseq_t value from the mapping, and see if it has changed "since"
98 * the given value was sampled.
99 *
100 * If it has then report the latest error set, otherwise return 0.
101 */
102static inline int filemap_check_wb_err(struct address_space *mapping,
103 errseq_t since)
104{
105 return errseq_check(eseq: &mapping->wb_err, since);
106}
107
108/**
109 * filemap_sample_wb_err - sample the current errseq_t to test for later errors
110 * @mapping: mapping to be sampled
111 *
112 * Writeback errors are always reported relative to a particular sample point
113 * in the past. This function provides those sample points.
114 */
115static inline errseq_t filemap_sample_wb_err(struct address_space *mapping)
116{
117 return errseq_sample(eseq: &mapping->wb_err);
118}
119
120/**
121 * file_sample_sb_err - sample the current errseq_t to test for later errors
122 * @file: file pointer to be sampled
123 *
124 * Grab the most current superblock-level errseq_t value for the given
125 * struct file.
126 */
127static inline errseq_t file_sample_sb_err(struct file *file)
128{
129 return errseq_sample(eseq: &file->f_path.dentry->d_sb->s_wb_err);
130}
131
132/*
133 * Flush file data before changing attributes. Caller must hold any locks
134 * required to prevent further writes to this file until we're done setting
135 * flags.
136 */
137static inline int inode_drain_writes(struct inode *inode)
138{
139 inode_dio_wait(inode);
140 return filemap_write_and_wait(mapping: inode->i_mapping);
141}
142
143static inline bool mapping_empty(struct address_space *mapping)
144{
145 return xa_empty(xa: &mapping->i_pages);
146}
147
148/*
149 * mapping_shrinkable - test if page cache state allows inode reclaim
150 * @mapping: the page cache mapping
151 *
152 * This checks the mapping's cache state for the pupose of inode
153 * reclaim and LRU management.
154 *
155 * The caller is expected to hold the i_lock, but is not required to
156 * hold the i_pages lock, which usually protects cache state. That's
157 * because the i_lock and the list_lru lock that protect the inode and
158 * its LRU state don't nest inside the irq-safe i_pages lock.
159 *
160 * Cache deletions are performed under the i_lock, which ensures that
161 * when an inode goes empty, it will reliably get queued on the LRU.
162 *
163 * Cache additions do not acquire the i_lock and may race with this
164 * check, in which case we'll report the inode as shrinkable when it
165 * has cache pages. This is okay: the shrinker also checks the
166 * refcount and the referenced bit, which will be elevated or set in
167 * the process of adding new cache pages to an inode.
168 */
169static inline bool mapping_shrinkable(struct address_space *mapping)
170{
171 void *head;
172
173 /*
174 * On highmem systems, there could be lowmem pressure from the
175 * inodes before there is highmem pressure from the page
176 * cache. Make inodes shrinkable regardless of cache state.
177 */
178 if (IS_ENABLED(CONFIG_HIGHMEM))
179 return true;
180
181 /* Cache completely empty? Shrink away. */
182 head = rcu_access_pointer(mapping->i_pages.xa_head);
183 if (!head)
184 return true;
185
186 /*
187 * The xarray stores single offset-0 entries directly in the
188 * head pointer, which allows non-resident page cache entries
189 * to escape the shadow shrinker's list of xarray nodes. The
190 * inode shrinker needs to pick them up under memory pressure.
191 */
192 if (!xa_is_node(entry: head) && xa_is_value(entry: head))
193 return true;
194
195 return false;
196}
197
198/*
199 * Bits in mapping->flags.
200 */
201enum mapping_flags {
202 AS_EIO = 0, /* IO error on async write */
203 AS_ENOSPC = 1, /* ENOSPC on async write */
204 AS_MM_ALL_LOCKS = 2, /* under mm_take_all_locks() */
205 AS_UNEVICTABLE = 3, /* e.g., ramdisk, SHM_LOCK */
206 AS_EXITING = 4, /* final truncate in progress */
207 /* writeback related tags are not used */
208 AS_NO_WRITEBACK_TAGS = 5,
209 AS_RELEASE_ALWAYS = 6, /* Call ->release_folio(), even if no private data */
210 AS_STABLE_WRITES = 7, /* must wait for writeback before modifying
211 folio contents */
212 AS_INACCESSIBLE = 8, /* Do not attempt direct R/W access to the mapping */
213 AS_WRITEBACK_MAY_DEADLOCK_ON_RECLAIM = 9,
214 /* Bits 16-25 are used for FOLIO_ORDER */
215 AS_FOLIO_ORDER_BITS = 5,
216 AS_FOLIO_ORDER_MIN = 16,
217 AS_FOLIO_ORDER_MAX = AS_FOLIO_ORDER_MIN + AS_FOLIO_ORDER_BITS,
218};
219
220#define AS_FOLIO_ORDER_BITS_MASK ((1u << AS_FOLIO_ORDER_BITS) - 1)
221#define AS_FOLIO_ORDER_MIN_MASK (AS_FOLIO_ORDER_BITS_MASK << AS_FOLIO_ORDER_MIN)
222#define AS_FOLIO_ORDER_MAX_MASK (AS_FOLIO_ORDER_BITS_MASK << AS_FOLIO_ORDER_MAX)
223#define AS_FOLIO_ORDER_MASK (AS_FOLIO_ORDER_MIN_MASK | AS_FOLIO_ORDER_MAX_MASK)
224
225/**
226 * mapping_set_error - record a writeback error in the address_space
227 * @mapping: the mapping in which an error should be set
228 * @error: the error to set in the mapping
229 *
230 * When writeback fails in some way, we must record that error so that
231 * userspace can be informed when fsync and the like are called. We endeavor
232 * to report errors on any file that was open at the time of the error. Some
233 * internal callers also need to know when writeback errors have occurred.
234 *
235 * When a writeback error occurs, most filesystems will want to call
236 * mapping_set_error to record the error in the mapping so that it can be
237 * reported when the application calls fsync(2).
238 */
239static inline void mapping_set_error(struct address_space *mapping, int error)
240{
241 if (likely(!error))
242 return;
243
244 /* Record in wb_err for checkers using errseq_t based tracking */
245 __filemap_set_wb_err(mapping, err: error);
246
247 /* Record it in superblock */
248 if (mapping->host)
249 errseq_set(eseq: &mapping->host->i_sb->s_wb_err, err: error);
250
251 /* Record it in flags for now, for legacy callers */
252 if (error == -ENOSPC)
253 set_bit(nr: AS_ENOSPC, addr: &mapping->flags);
254 else
255 set_bit(nr: AS_EIO, addr: &mapping->flags);
256}
257
258static inline void mapping_set_unevictable(struct address_space *mapping)
259{
260 set_bit(nr: AS_UNEVICTABLE, addr: &mapping->flags);
261}
262
263static inline void mapping_clear_unevictable(struct address_space *mapping)
264{
265 clear_bit(nr: AS_UNEVICTABLE, addr: &mapping->flags);
266}
267
268static inline bool mapping_unevictable(struct address_space *mapping)
269{
270 return mapping && test_bit(AS_UNEVICTABLE, &mapping->flags);
271}
272
273static inline void mapping_set_exiting(struct address_space *mapping)
274{
275 set_bit(nr: AS_EXITING, addr: &mapping->flags);
276}
277
278static inline int mapping_exiting(struct address_space *mapping)
279{
280 return test_bit(AS_EXITING, &mapping->flags);
281}
282
283static inline void mapping_set_no_writeback_tags(struct address_space *mapping)
284{
285 set_bit(nr: AS_NO_WRITEBACK_TAGS, addr: &mapping->flags);
286}
287
288static inline int mapping_use_writeback_tags(struct address_space *mapping)
289{
290 return !test_bit(AS_NO_WRITEBACK_TAGS, &mapping->flags);
291}
292
293static inline bool mapping_release_always(const struct address_space *mapping)
294{
295 return test_bit(AS_RELEASE_ALWAYS, &mapping->flags);
296}
297
298static inline void mapping_set_release_always(struct address_space *mapping)
299{
300 set_bit(nr: AS_RELEASE_ALWAYS, addr: &mapping->flags);
301}
302
303static inline void mapping_clear_release_always(struct address_space *mapping)
304{
305 clear_bit(nr: AS_RELEASE_ALWAYS, addr: &mapping->flags);
306}
307
308static inline bool mapping_stable_writes(const struct address_space *mapping)
309{
310 return test_bit(AS_STABLE_WRITES, &mapping->flags);
311}
312
313static inline void mapping_set_stable_writes(struct address_space *mapping)
314{
315 set_bit(nr: AS_STABLE_WRITES, addr: &mapping->flags);
316}
317
318static inline void mapping_clear_stable_writes(struct address_space *mapping)
319{
320 clear_bit(nr: AS_STABLE_WRITES, addr: &mapping->flags);
321}
322
323static inline void mapping_set_inaccessible(struct address_space *mapping)
324{
325 /*
326 * It's expected inaccessible mappings are also unevictable. Compaction
327 * migrate scanner (isolate_migratepages_block()) relies on this to
328 * reduce page locking.
329 */
330 set_bit(nr: AS_UNEVICTABLE, addr: &mapping->flags);
331 set_bit(nr: AS_INACCESSIBLE, addr: &mapping->flags);
332}
333
334static inline bool mapping_inaccessible(struct address_space *mapping)
335{
336 return test_bit(AS_INACCESSIBLE, &mapping->flags);
337}
338
339static inline void mapping_set_writeback_may_deadlock_on_reclaim(struct address_space *mapping)
340{
341 set_bit(nr: AS_WRITEBACK_MAY_DEADLOCK_ON_RECLAIM, addr: &mapping->flags);
342}
343
344static inline bool mapping_writeback_may_deadlock_on_reclaim(struct address_space *mapping)
345{
346 return test_bit(AS_WRITEBACK_MAY_DEADLOCK_ON_RECLAIM, &mapping->flags);
347}
348
349static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
350{
351 return mapping->gfp_mask;
352}
353
354/* Restricts the given gfp_mask to what the mapping allows. */
355static inline gfp_t mapping_gfp_constraint(struct address_space *mapping,
356 gfp_t gfp_mask)
357{
358 return mapping_gfp_mask(mapping) & gfp_mask;
359}
360
361/*
362 * This is non-atomic. Only to be used before the mapping is activated.
363 * Probably needs a barrier...
364 */
365static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
366{
367 m->gfp_mask = mask;
368}
369
370/*
371 * There are some parts of the kernel which assume that PMD entries
372 * are exactly HPAGE_PMD_ORDER. Those should be fixed, but until then,
373 * limit the maximum allocation order to PMD size. I'm not aware of any
374 * assumptions about maximum order if THP are disabled, but 8 seems like
375 * a good order (that's 1MB if you're using 4kB pages)
376 */
377#ifdef CONFIG_TRANSPARENT_HUGEPAGE
378#define PREFERRED_MAX_PAGECACHE_ORDER HPAGE_PMD_ORDER
379#else
380#define PREFERRED_MAX_PAGECACHE_ORDER 8
381#endif
382
383/*
384 * xas_split_alloc() does not support arbitrary orders. This implies no
385 * 512MB THP on ARM64 with 64KB base page size.
386 */
387#define MAX_XAS_ORDER (XA_CHUNK_SHIFT * 2 - 1)
388#define MAX_PAGECACHE_ORDER min(MAX_XAS_ORDER, PREFERRED_MAX_PAGECACHE_ORDER)
389
390/*
391 * mapping_max_folio_size_supported() - Check the max folio size supported
392 *
393 * The filesystem should call this function at mount time if there is a
394 * requirement on the folio mapping size in the page cache.
395 */
396static inline size_t mapping_max_folio_size_supported(void)
397{
398 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
399 return 1U << (PAGE_SHIFT + MAX_PAGECACHE_ORDER);
400 return PAGE_SIZE;
401}
402
403/*
404 * mapping_set_folio_order_range() - Set the orders supported by a file.
405 * @mapping: The address space of the file.
406 * @min: Minimum folio order (between 0-MAX_PAGECACHE_ORDER inclusive).
407 * @max: Maximum folio order (between @min-MAX_PAGECACHE_ORDER inclusive).
408 *
409 * The filesystem should call this function in its inode constructor to
410 * indicate which base size (min) and maximum size (max) of folio the VFS
411 * can use to cache the contents of the file. This should only be used
412 * if the filesystem needs special handling of folio sizes (ie there is
413 * something the core cannot know).
414 * Do not tune it based on, eg, i_size.
415 *
416 * Context: This should not be called while the inode is active as it
417 * is non-atomic.
418 */
419static inline void mapping_set_folio_order_range(struct address_space *mapping,
420 unsigned int min,
421 unsigned int max)
422{
423 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
424 return;
425
426 if (min > MAX_PAGECACHE_ORDER)
427 min = MAX_PAGECACHE_ORDER;
428
429 if (max > MAX_PAGECACHE_ORDER)
430 max = MAX_PAGECACHE_ORDER;
431
432 if (max < min)
433 max = min;
434
435 mapping->flags = (mapping->flags & ~AS_FOLIO_ORDER_MASK) |
436 (min << AS_FOLIO_ORDER_MIN) | (max << AS_FOLIO_ORDER_MAX);
437}
438
439static inline void mapping_set_folio_min_order(struct address_space *mapping,
440 unsigned int min)
441{
442 mapping_set_folio_order_range(mapping, min, MAX_PAGECACHE_ORDER);
443}
444
445/**
446 * mapping_set_large_folios() - Indicate the file supports large folios.
447 * @mapping: The address space of the file.
448 *
449 * The filesystem should call this function in its inode constructor to
450 * indicate that the VFS can use large folios to cache the contents of
451 * the file.
452 *
453 * Context: This should not be called while the inode is active as it
454 * is non-atomic.
455 */
456static inline void mapping_set_large_folios(struct address_space *mapping)
457{
458 mapping_set_folio_order_range(mapping, min: 0, MAX_PAGECACHE_ORDER);
459}
460
461static inline unsigned int
462mapping_max_folio_order(const struct address_space *mapping)
463{
464 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
465 return 0;
466 return (mapping->flags & AS_FOLIO_ORDER_MAX_MASK) >> AS_FOLIO_ORDER_MAX;
467}
468
469static inline unsigned int
470mapping_min_folio_order(const struct address_space *mapping)
471{
472 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
473 return 0;
474 return (mapping->flags & AS_FOLIO_ORDER_MIN_MASK) >> AS_FOLIO_ORDER_MIN;
475}
476
477static inline unsigned long
478mapping_min_folio_nrpages(struct address_space *mapping)
479{
480 return 1UL << mapping_min_folio_order(mapping);
481}
482
483/**
484 * mapping_align_index() - Align index for this mapping.
485 * @mapping: The address_space.
486 * @index: The page index.
487 *
488 * The index of a folio must be naturally aligned. If you are adding a
489 * new folio to the page cache and need to know what index to give it,
490 * call this function.
491 */
492static inline pgoff_t mapping_align_index(struct address_space *mapping,
493 pgoff_t index)
494{
495 return round_down(index, mapping_min_folio_nrpages(mapping));
496}
497
498/*
499 * Large folio support currently depends on THP. These dependencies are
500 * being worked on but are not yet fixed.
501 */
502static inline bool mapping_large_folio_support(struct address_space *mapping)
503{
504 /* AS_FOLIO_ORDER is only reasonable for pagecache folios */
505 VM_WARN_ONCE((unsigned long)mapping & PAGE_MAPPING_ANON,
506 "Anonymous mapping always supports large folio");
507
508 return mapping_max_folio_order(mapping) > 0;
509}
510
511/* Return the maximum folio size for this pagecache mapping, in bytes. */
512static inline size_t mapping_max_folio_size(const struct address_space *mapping)
513{
514 return PAGE_SIZE << mapping_max_folio_order(mapping);
515}
516
517static inline int filemap_nr_thps(struct address_space *mapping)
518{
519#ifdef CONFIG_READ_ONLY_THP_FOR_FS
520 return atomic_read(v: &mapping->nr_thps);
521#else
522 return 0;
523#endif
524}
525
526static inline void filemap_nr_thps_inc(struct address_space *mapping)
527{
528#ifdef CONFIG_READ_ONLY_THP_FOR_FS
529 if (!mapping_large_folio_support(mapping))
530 atomic_inc(v: &mapping->nr_thps);
531#else
532 WARN_ON_ONCE(mapping_large_folio_support(mapping) == 0);
533#endif
534}
535
536static inline void filemap_nr_thps_dec(struct address_space *mapping)
537{
538#ifdef CONFIG_READ_ONLY_THP_FOR_FS
539 if (!mapping_large_folio_support(mapping))
540 atomic_dec(v: &mapping->nr_thps);
541#else
542 WARN_ON_ONCE(mapping_large_folio_support(mapping) == 0);
543#endif
544}
545
546struct address_space *folio_mapping(struct folio *);
547
548/**
549 * folio_flush_mapping - Find the file mapping this folio belongs to.
550 * @folio: The folio.
551 *
552 * For folios which are in the page cache, return the mapping that this
553 * page belongs to. Anonymous folios return NULL, even if they're in
554 * the swap cache. Other kinds of folio also return NULL.
555 *
556 * This is ONLY used by architecture cache flushing code. If you aren't
557 * writing cache flushing code, you want either folio_mapping() or
558 * folio_file_mapping().
559 */
560static inline struct address_space *folio_flush_mapping(struct folio *folio)
561{
562 if (unlikely(folio_test_swapcache(folio)))
563 return NULL;
564
565 return folio_mapping(folio);
566}
567
568/**
569 * folio_inode - Get the host inode for this folio.
570 * @folio: The folio.
571 *
572 * For folios which are in the page cache, return the inode that this folio
573 * belongs to.
574 *
575 * Do not call this for folios which aren't in the page cache.
576 */
577static inline struct inode *folio_inode(struct folio *folio)
578{
579 return folio->mapping->host;
580}
581
582/**
583 * folio_attach_private - Attach private data to a folio.
584 * @folio: Folio to attach data to.
585 * @data: Data to attach to folio.
586 *
587 * Attaching private data to a folio increments the page's reference count.
588 * The data must be detached before the folio will be freed.
589 */
590static inline void folio_attach_private(struct folio *folio, void *data)
591{
592 folio_get(folio);
593 folio->private = data;
594 folio_set_private(folio);
595}
596
597/**
598 * folio_change_private - Change private data on a folio.
599 * @folio: Folio to change the data on.
600 * @data: Data to set on the folio.
601 *
602 * Change the private data attached to a folio and return the old
603 * data. The page must previously have had data attached and the data
604 * must be detached before the folio will be freed.
605 *
606 * Return: Data that was previously attached to the folio.
607 */
608static inline void *folio_change_private(struct folio *folio, void *data)
609{
610 void *old = folio_get_private(folio);
611
612 folio->private = data;
613 return old;
614}
615
616/**
617 * folio_detach_private - Detach private data from a folio.
618 * @folio: Folio to detach data from.
619 *
620 * Removes the data that was previously attached to the folio and decrements
621 * the refcount on the page.
622 *
623 * Return: Data that was attached to the folio.
624 */
625static inline void *folio_detach_private(struct folio *folio)
626{
627 void *data = folio_get_private(folio);
628
629 if (!folio_test_private(folio))
630 return NULL;
631 folio_clear_private(folio);
632 folio->private = NULL;
633 folio_put(folio);
634
635 return data;
636}
637
638static inline void attach_page_private(struct page *page, void *data)
639{
640 folio_attach_private(page_folio(page), data);
641}
642
643static inline void *detach_page_private(struct page *page)
644{
645 return folio_detach_private(page_folio(page));
646}
647
648#ifdef CONFIG_NUMA
649struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order);
650#else
651static inline struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order)
652{
653 return folio_alloc_noprof(gfp, order);
654}
655#endif
656
657#define filemap_alloc_folio(...) \
658 alloc_hooks(filemap_alloc_folio_noprof(__VA_ARGS__))
659
660static inline struct page *__page_cache_alloc(gfp_t gfp)
661{
662 return &filemap_alloc_folio(gfp, 0)->page;
663}
664
665static inline gfp_t readahead_gfp_mask(struct address_space *x)
666{
667 return mapping_gfp_mask(mapping: x) | __GFP_NORETRY | __GFP_NOWARN;
668}
669
670typedef int filler_t(struct file *, struct folio *);
671
672pgoff_t page_cache_next_miss(struct address_space *mapping,
673 pgoff_t index, unsigned long max_scan);
674pgoff_t page_cache_prev_miss(struct address_space *mapping,
675 pgoff_t index, unsigned long max_scan);
676
677/**
678 * typedef fgf_t - Flags for getting folios from the page cache.
679 *
680 * Most users of the page cache will not need to use these flags;
681 * there are convenience functions such as filemap_get_folio() and
682 * filemap_lock_folio(). For users which need more control over exactly
683 * what is done with the folios, these flags to __filemap_get_folio()
684 * are available.
685 *
686 * * %FGP_ACCESSED - The folio will be marked accessed.
687 * * %FGP_LOCK - The folio is returned locked.
688 * * %FGP_CREAT - If no folio is present then a new folio is allocated,
689 * added to the page cache and the VM's LRU list. The folio is
690 * returned locked.
691 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
692 * folio is already in cache. If the folio was allocated, unlock it
693 * before returning so the caller can do the same dance.
694 * * %FGP_WRITE - The folio will be written to by the caller.
695 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
696 * * %FGP_NOWAIT - Don't block on the folio lock.
697 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
698 * * %FGP_DONTCACHE - Uncached buffered IO
699 * * %FGP_WRITEBEGIN - The flags to use in a filesystem write_begin()
700 * implementation.
701 */
702typedef unsigned int __bitwise fgf_t;
703
704#define FGP_ACCESSED ((__force fgf_t)0x00000001)
705#define FGP_LOCK ((__force fgf_t)0x00000002)
706#define FGP_CREAT ((__force fgf_t)0x00000004)
707#define FGP_WRITE ((__force fgf_t)0x00000008)
708#define FGP_NOFS ((__force fgf_t)0x00000010)
709#define FGP_NOWAIT ((__force fgf_t)0x00000020)
710#define FGP_FOR_MMAP ((__force fgf_t)0x00000040)
711#define FGP_STABLE ((__force fgf_t)0x00000080)
712#define FGP_DONTCACHE ((__force fgf_t)0x00000100)
713#define FGF_GET_ORDER(fgf) (((__force unsigned)fgf) >> 26) /* top 6 bits */
714
715#define FGP_WRITEBEGIN (FGP_LOCK | FGP_WRITE | FGP_CREAT | FGP_STABLE)
716
717static inline unsigned int filemap_get_order(size_t size)
718{
719 unsigned int shift = ilog2(size);
720
721 if (shift <= PAGE_SHIFT)
722 return 0;
723
724 return shift - PAGE_SHIFT;
725}
726
727/**
728 * fgf_set_order - Encode a length in the fgf_t flags.
729 * @size: The suggested size of the folio to create.
730 *
731 * The caller of __filemap_get_folio() can use this to suggest a preferred
732 * size for the folio that is created. If there is already a folio at
733 * the index, it will be returned, no matter what its size. If a folio
734 * is freshly created, it may be of a different size than requested
735 * due to alignment constraints, memory pressure, or the presence of
736 * other folios at nearby indices.
737 */
738static inline fgf_t fgf_set_order(size_t size)
739{
740 unsigned int order = filemap_get_order(size);
741
742 if (!order)
743 return 0;
744 return (__force fgf_t)(order << 26);
745}
746
747void *filemap_get_entry(struct address_space *mapping, pgoff_t index);
748struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
749 fgf_t fgp_flags, gfp_t gfp);
750struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
751 fgf_t fgp_flags, gfp_t gfp);
752
753/**
754 * filemap_get_folio - Find and get a folio.
755 * @mapping: The address_space to search.
756 * @index: The page index.
757 *
758 * Looks up the page cache entry at @mapping & @index. If a folio is
759 * present, it is returned with an increased refcount.
760 *
761 * Return: A folio or ERR_PTR(-ENOENT) if there is no folio in the cache for
762 * this index. Will not return a shadow, swap or DAX entry.
763 */
764static inline struct folio *filemap_get_folio(struct address_space *mapping,
765 pgoff_t index)
766{
767 return __filemap_get_folio(mapping, index, fgp_flags: 0, gfp: 0);
768}
769
770/**
771 * filemap_lock_folio - Find and lock a folio.
772 * @mapping: The address_space to search.
773 * @index: The page index.
774 *
775 * Looks up the page cache entry at @mapping & @index. If a folio is
776 * present, it is returned locked with an increased refcount.
777 *
778 * Context: May sleep.
779 * Return: A folio or ERR_PTR(-ENOENT) if there is no folio in the cache for
780 * this index. Will not return a shadow, swap or DAX entry.
781 */
782static inline struct folio *filemap_lock_folio(struct address_space *mapping,
783 pgoff_t index)
784{
785 return __filemap_get_folio(mapping, index, FGP_LOCK, gfp: 0);
786}
787
788/**
789 * filemap_grab_folio - grab a folio from the page cache
790 * @mapping: The address space to search
791 * @index: The page index
792 *
793 * Looks up the page cache entry at @mapping & @index. If no folio is found,
794 * a new folio is created. The folio is locked, marked as accessed, and
795 * returned.
796 *
797 * Return: A found or created folio. ERR_PTR(-ENOMEM) if no folio is found
798 * and failed to create a folio.
799 */
800static inline struct folio *filemap_grab_folio(struct address_space *mapping,
801 pgoff_t index)
802{
803 return __filemap_get_folio(mapping, index,
804 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
805 gfp: mapping_gfp_mask(mapping));
806}
807
808/**
809 * find_get_page - find and get a page reference
810 * @mapping: the address_space to search
811 * @offset: the page index
812 *
813 * Looks up the page cache slot at @mapping & @offset. If there is a
814 * page cache page, it is returned with an increased refcount.
815 *
816 * Otherwise, %NULL is returned.
817 */
818static inline struct page *find_get_page(struct address_space *mapping,
819 pgoff_t offset)
820{
821 return pagecache_get_page(mapping, index: offset, fgp_flags: 0, gfp: 0);
822}
823
824static inline struct page *find_get_page_flags(struct address_space *mapping,
825 pgoff_t offset, fgf_t fgp_flags)
826{
827 return pagecache_get_page(mapping, index: offset, fgp_flags, gfp: 0);
828}
829
830/**
831 * find_lock_page - locate, pin and lock a pagecache page
832 * @mapping: the address_space to search
833 * @index: the page index
834 *
835 * Looks up the page cache entry at @mapping & @index. If there is a
836 * page cache page, it is returned locked and with an increased
837 * refcount.
838 *
839 * Context: May sleep.
840 * Return: A struct page or %NULL if there is no page in the cache for this
841 * index.
842 */
843static inline struct page *find_lock_page(struct address_space *mapping,
844 pgoff_t index)
845{
846 return pagecache_get_page(mapping, index, FGP_LOCK, gfp: 0);
847}
848
849/**
850 * find_or_create_page - locate or add a pagecache page
851 * @mapping: the page's address_space
852 * @index: the page's index into the mapping
853 * @gfp_mask: page allocation mode
854 *
855 * Looks up the page cache slot at @mapping & @offset. If there is a
856 * page cache page, it is returned locked and with an increased
857 * refcount.
858 *
859 * If the page is not present, a new page is allocated using @gfp_mask
860 * and added to the page cache and the VM's LRU list. The page is
861 * returned locked and with an increased refcount.
862 *
863 * On memory exhaustion, %NULL is returned.
864 *
865 * find_or_create_page() may sleep, even if @gfp_flags specifies an
866 * atomic allocation!
867 */
868static inline struct page *find_or_create_page(struct address_space *mapping,
869 pgoff_t index, gfp_t gfp_mask)
870{
871 return pagecache_get_page(mapping, index,
872 FGP_LOCK|FGP_ACCESSED|FGP_CREAT,
873 gfp: gfp_mask);
874}
875
876/**
877 * grab_cache_page_nowait - returns locked page at given index in given cache
878 * @mapping: target address_space
879 * @index: the page index
880 *
881 * Same as grab_cache_page(), but do not wait if the page is unavailable.
882 * This is intended for speculative data generators, where the data can
883 * be regenerated if the page couldn't be grabbed. This routine should
884 * be safe to call while holding the lock for another page.
885 *
886 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
887 * and deadlock against the caller's locked page.
888 */
889static inline struct page *grab_cache_page_nowait(struct address_space *mapping,
890 pgoff_t index)
891{
892 return pagecache_get_page(mapping, index,
893 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
894 gfp: mapping_gfp_mask(mapping));
895}
896
897/**
898 * folio_next_index - Get the index of the next folio.
899 * @folio: The current folio.
900 *
901 * Return: The index of the folio which follows this folio in the file.
902 */
903static inline pgoff_t folio_next_index(struct folio *folio)
904{
905 return folio->index + folio_nr_pages(folio);
906}
907
908/**
909 * folio_file_page - The page for a particular index.
910 * @folio: The folio which contains this index.
911 * @index: The index we want to look up.
912 *
913 * Sometimes after looking up a folio in the page cache, we need to
914 * obtain the specific page for an index (eg a page fault).
915 *
916 * Return: The page containing the file data for this index.
917 */
918static inline struct page *folio_file_page(struct folio *folio, pgoff_t index)
919{
920 return folio_page(folio, index & (folio_nr_pages(folio) - 1));
921}
922
923/**
924 * folio_contains - Does this folio contain this index?
925 * @folio: The folio.
926 * @index: The page index within the file.
927 *
928 * Context: The caller should have the folio locked and ensure
929 * e.g., shmem did not move this folio to the swap cache.
930 * Return: true or false.
931 */
932static inline bool folio_contains(struct folio *folio, pgoff_t index)
933{
934 VM_WARN_ON_ONCE_FOLIO(folio_test_swapcache(folio), folio);
935 return index - folio->index < folio_nr_pages(folio);
936}
937
938unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
939 pgoff_t end, struct folio_batch *fbatch);
940unsigned filemap_get_folios_contig(struct address_space *mapping,
941 pgoff_t *start, pgoff_t end, struct folio_batch *fbatch);
942unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
943 pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch);
944
945/*
946 * Returns locked page at given index in given cache, creating it if needed.
947 */
948static inline struct page *grab_cache_page(struct address_space *mapping,
949 pgoff_t index)
950{
951 return find_or_create_page(mapping, index, gfp_mask: mapping_gfp_mask(mapping));
952}
953
954struct folio *read_cache_folio(struct address_space *, pgoff_t index,
955 filler_t *filler, struct file *file);
956struct folio *mapping_read_folio_gfp(struct address_space *, pgoff_t index,
957 gfp_t flags);
958struct page *read_cache_page(struct address_space *, pgoff_t index,
959 filler_t *filler, struct file *file);
960extern struct page * read_cache_page_gfp(struct address_space *mapping,
961 pgoff_t index, gfp_t gfp_mask);
962
963static inline struct page *read_mapping_page(struct address_space *mapping,
964 pgoff_t index, struct file *file)
965{
966 return read_cache_page(mapping, index, NULL, file);
967}
968
969static inline struct folio *read_mapping_folio(struct address_space *mapping,
970 pgoff_t index, struct file *file)
971{
972 return read_cache_folio(mapping, index, NULL, file);
973}
974
975/**
976 * page_pgoff - Calculate the logical page offset of this page.
977 * @folio: The folio containing this page.
978 * @page: The page which we need the offset of.
979 *
980 * For file pages, this is the offset from the beginning of the file
981 * in units of PAGE_SIZE. For anonymous pages, this is the offset from
982 * the beginning of the anon_vma in units of PAGE_SIZE. This will
983 * return nonsense for KSM pages.
984 *
985 * Context: Caller must have a reference on the folio or otherwise
986 * prevent it from being split or freed.
987 *
988 * Return: The offset in units of PAGE_SIZE.
989 */
990static inline pgoff_t page_pgoff(const struct folio *folio,
991 const struct page *page)
992{
993 return folio->index + folio_page_idx(folio, page);
994}
995
996/**
997 * folio_pos - Returns the byte position of this folio in its file.
998 * @folio: The folio.
999 */
1000static inline loff_t folio_pos(const struct folio *folio)
1001{
1002 return ((loff_t)folio->index) * PAGE_SIZE;
1003}
1004
1005/*
1006 * Return byte-offset into filesystem object for page.
1007 */
1008static inline loff_t page_offset(struct page *page)
1009{
1010 struct folio *folio = page_folio(page);
1011
1012 return folio_pos(folio) + folio_page_idx(folio, page) * PAGE_SIZE;
1013}
1014
1015/*
1016 * Get the offset in PAGE_SIZE (even for hugetlb folios).
1017 */
1018static inline pgoff_t folio_pgoff(struct folio *folio)
1019{
1020 return folio->index;
1021}
1022
1023static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
1024 unsigned long address)
1025{
1026 pgoff_t pgoff;
1027 pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
1028 pgoff += vma->vm_pgoff;
1029 return pgoff;
1030}
1031
1032struct wait_page_key {
1033 struct folio *folio;
1034 int bit_nr;
1035 int page_match;
1036};
1037
1038struct wait_page_queue {
1039 struct folio *folio;
1040 int bit_nr;
1041 wait_queue_entry_t wait;
1042};
1043
1044static inline bool wake_page_match(struct wait_page_queue *wait_page,
1045 struct wait_page_key *key)
1046{
1047 if (wait_page->folio != key->folio)
1048 return false;
1049 key->page_match = 1;
1050
1051 if (wait_page->bit_nr != key->bit_nr)
1052 return false;
1053
1054 return true;
1055}
1056
1057void __folio_lock(struct folio *folio);
1058int __folio_lock_killable(struct folio *folio);
1059vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf);
1060void unlock_page(struct page *page);
1061void folio_unlock(struct folio *folio);
1062
1063/**
1064 * folio_trylock() - Attempt to lock a folio.
1065 * @folio: The folio to attempt to lock.
1066 *
1067 * Sometimes it is undesirable to wait for a folio to be unlocked (eg
1068 * when the locks are being taken in the wrong order, or if making
1069 * progress through a batch of folios is more important than processing
1070 * them in order). Usually folio_lock() is the correct function to call.
1071 *
1072 * Context: Any context.
1073 * Return: Whether the lock was successfully acquired.
1074 */
1075static inline bool folio_trylock(struct folio *folio)
1076{
1077 return likely(!test_and_set_bit_lock(PG_locked, folio_flags(folio, 0)));
1078}
1079
1080/*
1081 * Return true if the page was successfully locked
1082 */
1083static inline bool trylock_page(struct page *page)
1084{
1085 return folio_trylock(page_folio(page));
1086}
1087
1088/**
1089 * folio_lock() - Lock this folio.
1090 * @folio: The folio to lock.
1091 *
1092 * The folio lock protects against many things, probably more than it
1093 * should. It is primarily held while a folio is being brought uptodate,
1094 * either from its backing file or from swap. It is also held while a
1095 * folio is being truncated from its address_space, so holding the lock
1096 * is sufficient to keep folio->mapping stable.
1097 *
1098 * The folio lock is also held while write() is modifying the page to
1099 * provide POSIX atomicity guarantees (as long as the write does not
1100 * cross a page boundary). Other modifications to the data in the folio
1101 * do not hold the folio lock and can race with writes, eg DMA and stores
1102 * to mapped pages.
1103 *
1104 * Context: May sleep. If you need to acquire the locks of two or
1105 * more folios, they must be in order of ascending index, if they are
1106 * in the same address_space. If they are in different address_spaces,
1107 * acquire the lock of the folio which belongs to the address_space which
1108 * has the lowest address in memory first.
1109 */
1110static inline void folio_lock(struct folio *folio)
1111{
1112 might_sleep();
1113 if (!folio_trylock(folio))
1114 __folio_lock(folio);
1115}
1116
1117/**
1118 * lock_page() - Lock the folio containing this page.
1119 * @page: The page to lock.
1120 *
1121 * See folio_lock() for a description of what the lock protects.
1122 * This is a legacy function and new code should probably use folio_lock()
1123 * instead.
1124 *
1125 * Context: May sleep. Pages in the same folio share a lock, so do not
1126 * attempt to lock two pages which share a folio.
1127 */
1128static inline void lock_page(struct page *page)
1129{
1130 struct folio *folio;
1131 might_sleep();
1132
1133 folio = page_folio(page);
1134 if (!folio_trylock(folio))
1135 __folio_lock(folio);
1136}
1137
1138/**
1139 * folio_lock_killable() - Lock this folio, interruptible by a fatal signal.
1140 * @folio: The folio to lock.
1141 *
1142 * Attempts to lock the folio, like folio_lock(), except that the sleep
1143 * to acquire the lock is interruptible by a fatal signal.
1144 *
1145 * Context: May sleep; see folio_lock().
1146 * Return: 0 if the lock was acquired; -EINTR if a fatal signal was received.
1147 */
1148static inline int folio_lock_killable(struct folio *folio)
1149{
1150 might_sleep();
1151 if (!folio_trylock(folio))
1152 return __folio_lock_killable(folio);
1153 return 0;
1154}
1155
1156/*
1157 * folio_lock_or_retry - Lock the folio, unless this would block and the
1158 * caller indicated that it can handle a retry.
1159 *
1160 * Return value and mmap_lock implications depend on flags; see
1161 * __folio_lock_or_retry().
1162 */
1163static inline vm_fault_t folio_lock_or_retry(struct folio *folio,
1164 struct vm_fault *vmf)
1165{
1166 might_sleep();
1167 if (!folio_trylock(folio))
1168 return __folio_lock_or_retry(folio, vmf);
1169 return 0;
1170}
1171
1172/*
1173 * This is exported only for folio_wait_locked/folio_wait_writeback, etc.,
1174 * and should not be used directly.
1175 */
1176void folio_wait_bit(struct folio *folio, int bit_nr);
1177int folio_wait_bit_killable(struct folio *folio, int bit_nr);
1178
1179/*
1180 * Wait for a folio to be unlocked.
1181 *
1182 * This must be called with the caller "holding" the folio,
1183 * ie with increased folio reference count so that the folio won't
1184 * go away during the wait.
1185 */
1186static inline void folio_wait_locked(struct folio *folio)
1187{
1188 if (folio_test_locked(folio))
1189 folio_wait_bit(folio, bit_nr: PG_locked);
1190}
1191
1192static inline int folio_wait_locked_killable(struct folio *folio)
1193{
1194 if (!folio_test_locked(folio))
1195 return 0;
1196 return folio_wait_bit_killable(folio, bit_nr: PG_locked);
1197}
1198
1199void folio_end_read(struct folio *folio, bool success);
1200void wait_on_page_writeback(struct page *page);
1201void folio_wait_writeback(struct folio *folio);
1202int folio_wait_writeback_killable(struct folio *folio);
1203void end_page_writeback(struct page *page);
1204void folio_end_writeback(struct folio *folio);
1205void folio_wait_stable(struct folio *folio);
1206void __folio_mark_dirty(struct folio *folio, struct address_space *, int warn);
1207void folio_account_cleaned(struct folio *folio, struct bdi_writeback *wb);
1208void __folio_cancel_dirty(struct folio *folio);
1209static inline void folio_cancel_dirty(struct folio *folio)
1210{
1211 /* Avoid atomic ops, locking, etc. when not actually needed. */
1212 if (folio_test_dirty(folio))
1213 __folio_cancel_dirty(folio);
1214}
1215bool folio_clear_dirty_for_io(struct folio *folio);
1216bool clear_page_dirty_for_io(struct page *page);
1217void folio_invalidate(struct folio *folio, size_t offset, size_t length);
1218bool noop_dirty_folio(struct address_space *mapping, struct folio *folio);
1219
1220#ifdef CONFIG_MIGRATION
1221int filemap_migrate_folio(struct address_space *mapping, struct folio *dst,
1222 struct folio *src, enum migrate_mode mode);
1223#else
1224#define filemap_migrate_folio NULL
1225#endif
1226void folio_end_private_2(struct folio *folio);
1227void folio_wait_private_2(struct folio *folio);
1228int folio_wait_private_2_killable(struct folio *folio);
1229
1230/*
1231 * Fault in userspace address range.
1232 */
1233size_t fault_in_writeable(char __user *uaddr, size_t size);
1234size_t fault_in_subpage_writeable(char __user *uaddr, size_t size);
1235size_t fault_in_safe_writeable(const char __user *uaddr, size_t size);
1236size_t fault_in_readable(const char __user *uaddr, size_t size);
1237
1238int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
1239 pgoff_t index, gfp_t gfp);
1240int filemap_add_folio(struct address_space *mapping, struct folio *folio,
1241 pgoff_t index, gfp_t gfp);
1242void filemap_remove_folio(struct folio *folio);
1243void __filemap_remove_folio(struct folio *folio, void *shadow);
1244void replace_page_cache_folio(struct folio *old, struct folio *new);
1245void delete_from_page_cache_batch(struct address_space *mapping,
1246 struct folio_batch *fbatch);
1247bool filemap_release_folio(struct folio *folio, gfp_t gfp);
1248loff_t mapping_seek_hole_data(struct address_space *, loff_t start, loff_t end,
1249 int whence);
1250
1251/* Must be non-static for BPF error injection */
1252int __filemap_add_folio(struct address_space *mapping, struct folio *folio,
1253 pgoff_t index, gfp_t gfp, void **shadowp);
1254
1255bool filemap_range_has_writeback(struct address_space *mapping,
1256 loff_t start_byte, loff_t end_byte);
1257
1258/**
1259 * filemap_range_needs_writeback - check if range potentially needs writeback
1260 * @mapping: address space within which to check
1261 * @start_byte: offset in bytes where the range starts
1262 * @end_byte: offset in bytes where the range ends (inclusive)
1263 *
1264 * Find at least one page in the range supplied, usually used to check if
1265 * direct writing in this range will trigger a writeback. Used by O_DIRECT
1266 * read/write with IOCB_NOWAIT, to see if the caller needs to do
1267 * filemap_write_and_wait_range() before proceeding.
1268 *
1269 * Return: %true if the caller should do filemap_write_and_wait_range() before
1270 * doing O_DIRECT to a page in this range, %false otherwise.
1271 */
1272static inline bool filemap_range_needs_writeback(struct address_space *mapping,
1273 loff_t start_byte,
1274 loff_t end_byte)
1275{
1276 if (!mapping->nrpages)
1277 return false;
1278 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
1279 !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
1280 return false;
1281 return filemap_range_has_writeback(mapping, start_byte, end_byte);
1282}
1283
1284/**
1285 * struct readahead_control - Describes a readahead request.
1286 *
1287 * A readahead request is for consecutive pages. Filesystems which
1288 * implement the ->readahead method should call readahead_folio() or
1289 * __readahead_batch() in a loop and attempt to start reads into each
1290 * folio in the request.
1291 *
1292 * Most of the fields in this struct are private and should be accessed
1293 * by the functions below.
1294 *
1295 * @file: The file, used primarily by network filesystems for authentication.
1296 * May be NULL if invoked internally by the filesystem.
1297 * @mapping: Readahead this filesystem object.
1298 * @ra: File readahead state. May be NULL.
1299 */
1300struct readahead_control {
1301 struct file *file;
1302 struct address_space *mapping;
1303 struct file_ra_state *ra;
1304/* private: use the readahead_* accessors instead */
1305 pgoff_t _index;
1306 unsigned int _nr_pages;
1307 unsigned int _batch_count;
1308 bool dropbehind;
1309 bool _workingset;
1310 unsigned long _pflags;
1311};
1312
1313#define DEFINE_READAHEAD(ractl, f, r, m, i) \
1314 struct readahead_control ractl = { \
1315 .file = f, \
1316 .mapping = m, \
1317 .ra = r, \
1318 ._index = i, \
1319 }
1320
1321#define VM_READAHEAD_PAGES (SZ_128K / PAGE_SIZE)
1322
1323void page_cache_ra_unbounded(struct readahead_control *,
1324 unsigned long nr_to_read, unsigned long lookahead_count);
1325void page_cache_sync_ra(struct readahead_control *, unsigned long req_count);
1326void page_cache_async_ra(struct readahead_control *, struct folio *,
1327 unsigned long req_count);
1328void readahead_expand(struct readahead_control *ractl,
1329 loff_t new_start, size_t new_len);
1330
1331/**
1332 * page_cache_sync_readahead - generic file readahead
1333 * @mapping: address_space which holds the pagecache and I/O vectors
1334 * @ra: file_ra_state which holds the readahead state
1335 * @file: Used by the filesystem for authentication.
1336 * @index: Index of first page to be read.
1337 * @req_count: Total number of pages being read by the caller.
1338 *
1339 * page_cache_sync_readahead() should be called when a cache miss happened:
1340 * it will submit the read. The readahead logic may decide to piggyback more
1341 * pages onto the read request if access patterns suggest it will improve
1342 * performance.
1343 */
1344static inline
1345void page_cache_sync_readahead(struct address_space *mapping,
1346 struct file_ra_state *ra, struct file *file, pgoff_t index,
1347 unsigned long req_count)
1348{
1349 DEFINE_READAHEAD(ractl, file, ra, mapping, index);
1350 page_cache_sync_ra(&ractl, req_count);
1351}
1352
1353/**
1354 * page_cache_async_readahead - file readahead for marked pages
1355 * @mapping: address_space which holds the pagecache and I/O vectors
1356 * @ra: file_ra_state which holds the readahead state
1357 * @file: Used by the filesystem for authentication.
1358 * @folio: The folio which triggered the readahead call.
1359 * @req_count: Total number of pages being read by the caller.
1360 *
1361 * page_cache_async_readahead() should be called when a page is used which
1362 * is marked as PageReadahead; this is a marker to suggest that the application
1363 * has used up enough of the readahead window that we should start pulling in
1364 * more pages.
1365 */
1366static inline
1367void page_cache_async_readahead(struct address_space *mapping,
1368 struct file_ra_state *ra, struct file *file,
1369 struct folio *folio, unsigned long req_count)
1370{
1371 DEFINE_READAHEAD(ractl, file, ra, mapping, folio->index);
1372 page_cache_async_ra(&ractl, folio, req_count);
1373}
1374
1375static inline struct folio *__readahead_folio(struct readahead_control *ractl)
1376{
1377 struct folio *folio;
1378
1379 BUG_ON(ractl->_batch_count > ractl->_nr_pages);
1380 ractl->_nr_pages -= ractl->_batch_count;
1381 ractl->_index += ractl->_batch_count;
1382
1383 if (!ractl->_nr_pages) {
1384 ractl->_batch_count = 0;
1385 return NULL;
1386 }
1387
1388 folio = xa_load(&ractl->mapping->i_pages, index: ractl->_index);
1389 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1390 ractl->_batch_count = folio_nr_pages(folio);
1391
1392 return folio;
1393}
1394
1395/**
1396 * readahead_folio - Get the next folio to read.
1397 * @ractl: The current readahead request.
1398 *
1399 * Context: The folio is locked. The caller should unlock the folio once
1400 * all I/O to that folio has completed.
1401 * Return: A pointer to the next folio, or %NULL if we are done.
1402 */
1403static inline struct folio *readahead_folio(struct readahead_control *ractl)
1404{
1405 struct folio *folio = __readahead_folio(ractl);
1406
1407 if (folio)
1408 folio_put(folio);
1409 return folio;
1410}
1411
1412static inline unsigned int __readahead_batch(struct readahead_control *rac,
1413 struct page **array, unsigned int array_sz)
1414{
1415 unsigned int i = 0;
1416 XA_STATE(xas, &rac->mapping->i_pages, 0);
1417 struct folio *folio;
1418
1419 BUG_ON(rac->_batch_count > rac->_nr_pages);
1420 rac->_nr_pages -= rac->_batch_count;
1421 rac->_index += rac->_batch_count;
1422 rac->_batch_count = 0;
1423
1424 xas_set(xas: &xas, index: rac->_index);
1425 rcu_read_lock();
1426 xas_for_each(&xas, folio, rac->_index + rac->_nr_pages - 1) {
1427 if (xas_retry(xas: &xas, entry: folio))
1428 continue;
1429 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1430 array[i++] = folio_page(folio, 0);
1431 rac->_batch_count += folio_nr_pages(folio);
1432 if (i == array_sz)
1433 break;
1434 }
1435 rcu_read_unlock();
1436
1437 return i;
1438}
1439
1440/**
1441 * readahead_pos - The byte offset into the file of this readahead request.
1442 * @rac: The readahead request.
1443 */
1444static inline loff_t readahead_pos(struct readahead_control *rac)
1445{
1446 return (loff_t)rac->_index * PAGE_SIZE;
1447}
1448
1449/**
1450 * readahead_length - The number of bytes in this readahead request.
1451 * @rac: The readahead request.
1452 */
1453static inline size_t readahead_length(struct readahead_control *rac)
1454{
1455 return rac->_nr_pages * PAGE_SIZE;
1456}
1457
1458/**
1459 * readahead_index - The index of the first page in this readahead request.
1460 * @rac: The readahead request.
1461 */
1462static inline pgoff_t readahead_index(struct readahead_control *rac)
1463{
1464 return rac->_index;
1465}
1466
1467/**
1468 * readahead_count - The number of pages in this readahead request.
1469 * @rac: The readahead request.
1470 */
1471static inline unsigned int readahead_count(struct readahead_control *rac)
1472{
1473 return rac->_nr_pages;
1474}
1475
1476/**
1477 * readahead_batch_length - The number of bytes in the current batch.
1478 * @rac: The readahead request.
1479 */
1480static inline size_t readahead_batch_length(struct readahead_control *rac)
1481{
1482 return rac->_batch_count * PAGE_SIZE;
1483}
1484
1485static inline unsigned long dir_pages(struct inode *inode)
1486{
1487 return (unsigned long)(inode->i_size + PAGE_SIZE - 1) >>
1488 PAGE_SHIFT;
1489}
1490
1491/**
1492 * folio_mkwrite_check_truncate - check if folio was truncated
1493 * @folio: the folio to check
1494 * @inode: the inode to check the folio against
1495 *
1496 * Return: the number of bytes in the folio up to EOF,
1497 * or -EFAULT if the folio was truncated.
1498 */
1499static inline ssize_t folio_mkwrite_check_truncate(struct folio *folio,
1500 struct inode *inode)
1501{
1502 loff_t size = i_size_read(inode);
1503 pgoff_t index = size >> PAGE_SHIFT;
1504 size_t offset = offset_in_folio(folio, size);
1505
1506 if (!folio->mapping)
1507 return -EFAULT;
1508
1509 /* folio is wholly inside EOF */
1510 if (folio_next_index(folio) - 1 < index)
1511 return folio_size(folio);
1512 /* folio is wholly past EOF */
1513 if (folio->index > index || !offset)
1514 return -EFAULT;
1515 /* folio is partially inside EOF */
1516 return offset;
1517}
1518
1519/**
1520 * i_blocks_per_folio - How many blocks fit in this folio.
1521 * @inode: The inode which contains the blocks.
1522 * @folio: The folio.
1523 *
1524 * If the block size is larger than the size of this folio, return zero.
1525 *
1526 * Context: The caller should hold a refcount on the folio to prevent it
1527 * from being split.
1528 * Return: The number of filesystem blocks covered by this folio.
1529 */
1530static inline
1531unsigned int i_blocks_per_folio(struct inode *inode, struct folio *folio)
1532{
1533 return folio_size(folio) >> inode->i_blkbits;
1534}
1535#endif /* _LINUX_PAGEMAP_H */
1536

source code of linux/include/linux/pagemap.h