1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_PAGEMAP_H
3 #define _LINUX_PAGEMAP_H
6 * Copyright 1995 Linus Torvalds
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>
22 * Bits in mapping->flags.
25 AS_EIO = 0, /* IO error on async write */
26 AS_ENOSPC = 1, /* ENOSPC on async write */
27 AS_MM_ALL_LOCKS = 2, /* under mm_take_all_locks() */
28 AS_UNEVICTABLE = 3, /* e.g., ramdisk, SHM_LOCK */
29 AS_EXITING = 4, /* final truncate in progress */
30 /* writeback related tags are not used */
31 AS_NO_WRITEBACK_TAGS = 5,
35 * mapping_set_error - record a writeback error in the address_space
36 * @mapping: the mapping in which an error should be set
37 * @error: the error to set in the mapping
39 * When writeback fails in some way, we must record that error so that
40 * userspace can be informed when fsync and the like are called. We endeavor
41 * to report errors on any file that was open at the time of the error. Some
42 * internal callers also need to know when writeback errors have occurred.
44 * When a writeback error occurs, most filesystems will want to call
45 * mapping_set_error to record the error in the mapping so that it can be
46 * reported when the application calls fsync(2).
48 static inline void mapping_set_error(struct address_space *mapping, int error)
53 /* Record in wb_err for checkers using errseq_t based tracking */
54 __filemap_set_wb_err(mapping, error);
56 /* Record it in superblock */
57 errseq_set(&mapping->host->i_sb->s_wb_err, error);
59 /* Record it in flags for now, for legacy callers */
61 set_bit(AS_ENOSPC, &mapping->flags);
63 set_bit(AS_EIO, &mapping->flags);
66 static inline void mapping_set_unevictable(struct address_space *mapping)
68 set_bit(AS_UNEVICTABLE, &mapping->flags);
71 static inline void mapping_clear_unevictable(struct address_space *mapping)
73 clear_bit(AS_UNEVICTABLE, &mapping->flags);
76 static inline bool mapping_unevictable(struct address_space *mapping)
78 return mapping && test_bit(AS_UNEVICTABLE, &mapping->flags);
81 static inline void mapping_set_exiting(struct address_space *mapping)
83 set_bit(AS_EXITING, &mapping->flags);
86 static inline int mapping_exiting(struct address_space *mapping)
88 return test_bit(AS_EXITING, &mapping->flags);
91 static inline void mapping_set_no_writeback_tags(struct address_space *mapping)
93 set_bit(AS_NO_WRITEBACK_TAGS, &mapping->flags);
96 static inline int mapping_use_writeback_tags(struct address_space *mapping)
98 return !test_bit(AS_NO_WRITEBACK_TAGS, &mapping->flags);
101 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
103 return mapping->gfp_mask;
106 /* Restricts the given gfp_mask to what the mapping allows. */
107 static inline gfp_t mapping_gfp_constraint(struct address_space *mapping,
110 return mapping_gfp_mask(mapping) & gfp_mask;
114 * This is non-atomic. Only to be used before the mapping is activated.
115 * Probably needs a barrier...
117 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
122 void release_pages(struct page **pages, int nr);
125 * speculatively take a reference to a page.
126 * If the page is free (_refcount == 0), then _refcount is untouched, and 0
127 * is returned. Otherwise, _refcount is incremented by 1 and 1 is returned.
129 * This function must be called inside the same rcu_read_lock() section as has
130 * been used to lookup the page in the pagecache radix-tree (or page table):
131 * this allows allocators to use a synchronize_rcu() to stabilize _refcount.
133 * Unless an RCU grace period has passed, the count of all pages coming out
134 * of the allocator must be considered unstable. page_count may return higher
135 * than expected, and put_page must be able to do the right thing when the
136 * page has been finished with, no matter what it is subsequently allocated
137 * for (because put_page is what is used here to drop an invalid speculative
140 * This is the interesting part of the lockless pagecache (and lockless
141 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
142 * has the following pattern:
143 * 1. find page in radix tree
144 * 2. conditionally increment refcount
145 * 3. check the page is still in pagecache (if no, goto 1)
147 * Remove-side that cares about stability of _refcount (eg. reclaim) has the
148 * following (with the i_pages lock held):
149 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
150 * B. remove page from pagecache
153 * There are 2 critical interleavings that matter:
154 * - 2 runs before A: in this case, A sees elevated refcount and bails out
155 * - A runs before 2: in this case, 2 sees zero refcount and retries;
156 * subsequently, B will complete and 1 will find no page, causing the
157 * lookup to return NULL.
159 * It is possible that between 1 and 2, the page is removed then the exact same
160 * page is inserted into the same position in pagecache. That's OK: the
161 * old find_get_page using a lock could equally have run before or after
162 * such a re-insertion, depending on order that locks are granted.
164 * Lookups racing against pagecache insertion isn't a big problem: either 1
165 * will find the page or it will not. Likewise, the old find_get_page could run
166 * either before the insertion or afterwards, depending on timing.
168 static inline int __page_cache_add_speculative(struct page *page, int count)
170 #ifdef CONFIG_TINY_RCU
171 # ifdef CONFIG_PREEMPT_COUNT
172 VM_BUG_ON(!in_atomic() && !irqs_disabled());
175 * Preempt must be disabled here - we rely on rcu_read_lock doing
178 * Pagecache won't be truncated from interrupt context, so if we have
179 * found a page in the radix tree here, we have pinned its refcount by
180 * disabling preempt, and hence no need for the "speculative get" that
183 VM_BUG_ON_PAGE(page_count(page) == 0, page);
184 page_ref_add(page, count);
187 if (unlikely(!page_ref_add_unless(page, count, 0))) {
189 * Either the page has been freed, or will be freed.
190 * In either case, retry here and the caller should
191 * do the right thing (see comments above).
196 VM_BUG_ON_PAGE(PageTail(page), page);
201 static inline int page_cache_get_speculative(struct page *page)
203 return __page_cache_add_speculative(page, 1);
206 static inline int page_cache_add_speculative(struct page *page, int count)
208 return __page_cache_add_speculative(page, count);
212 * attach_page_private - Attach private data to a page.
213 * @page: Page to attach data to.
214 * @data: Data to attach to page.
216 * Attaching private data to a page increments the page's reference count.
217 * The data must be detached before the page will be freed.
219 static inline void attach_page_private(struct page *page, void *data)
222 set_page_private(page, (unsigned long)data);
223 SetPagePrivate(page);
227 * detach_page_private - Detach private data from a page.
228 * @page: Page to detach data from.
230 * Removes the data that was previously attached to the page and decrements
231 * the refcount on the page.
233 * Return: Data that was attached to the page.
235 static inline void *detach_page_private(struct page *page)
237 void *data = (void *)page_private(page);
239 if (!PagePrivate(page))
241 ClearPagePrivate(page);
242 set_page_private(page, 0);
249 extern struct page *__page_cache_alloc(gfp_t gfp);
251 static inline struct page *__page_cache_alloc(gfp_t gfp)
253 return alloc_pages(gfp, 0);
257 static inline struct page *page_cache_alloc(struct address_space *x)
259 return __page_cache_alloc(mapping_gfp_mask(x));
262 static inline gfp_t readahead_gfp_mask(struct address_space *x)
264 return mapping_gfp_mask(x) | __GFP_NORETRY | __GFP_NOWARN;
267 typedef int filler_t(void *, struct page *);
269 pgoff_t page_cache_next_miss(struct address_space *mapping,
270 pgoff_t index, unsigned long max_scan);
271 pgoff_t page_cache_prev_miss(struct address_space *mapping,
272 pgoff_t index, unsigned long max_scan);
274 #define FGP_ACCESSED 0x00000001
275 #define FGP_LOCK 0x00000002
276 #define FGP_CREAT 0x00000004
277 #define FGP_WRITE 0x00000008
278 #define FGP_NOFS 0x00000010
279 #define FGP_NOWAIT 0x00000020
280 #define FGP_FOR_MMAP 0x00000040
282 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
283 int fgp_flags, gfp_t cache_gfp_mask);
286 * find_get_page - find and get a page reference
287 * @mapping: the address_space to search
288 * @offset: the page index
290 * Looks up the page cache slot at @mapping & @offset. If there is a
291 * page cache page, it is returned with an increased refcount.
293 * Otherwise, %NULL is returned.
295 static inline struct page *find_get_page(struct address_space *mapping,
298 return pagecache_get_page(mapping, offset, 0, 0);
301 static inline struct page *find_get_page_flags(struct address_space *mapping,
302 pgoff_t offset, int fgp_flags)
304 return pagecache_get_page(mapping, offset, fgp_flags, 0);
308 * find_lock_page - locate, pin and lock a pagecache page
309 * @mapping: the address_space to search
310 * @offset: the page index
312 * Looks up the page cache slot at @mapping & @offset. If there is a
313 * page cache page, it is returned locked and with an increased
316 * Otherwise, %NULL is returned.
318 * find_lock_page() may sleep.
320 static inline struct page *find_lock_page(struct address_space *mapping,
323 return pagecache_get_page(mapping, offset, FGP_LOCK, 0);
327 * find_or_create_page - locate or add a pagecache page
328 * @mapping: the page's address_space
329 * @index: the page's index into the mapping
330 * @gfp_mask: page allocation mode
332 * Looks up the page cache slot at @mapping & @offset. If there is a
333 * page cache page, it is returned locked and with an increased
336 * If the page is not present, a new page is allocated using @gfp_mask
337 * and added to the page cache and the VM's LRU list. The page is
338 * returned locked and with an increased refcount.
340 * On memory exhaustion, %NULL is returned.
342 * find_or_create_page() may sleep, even if @gfp_flags specifies an
345 static inline struct page *find_or_create_page(struct address_space *mapping,
346 pgoff_t index, gfp_t gfp_mask)
348 return pagecache_get_page(mapping, index,
349 FGP_LOCK|FGP_ACCESSED|FGP_CREAT,
354 * grab_cache_page_nowait - returns locked page at given index in given cache
355 * @mapping: target address_space
356 * @index: the page index
358 * Same as grab_cache_page(), but do not wait if the page is unavailable.
359 * This is intended for speculative data generators, where the data can
360 * be regenerated if the page couldn't be grabbed. This routine should
361 * be safe to call while holding the lock for another page.
363 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
364 * and deadlock against the caller's locked page.
366 static inline struct page *grab_cache_page_nowait(struct address_space *mapping,
369 return pagecache_get_page(mapping, index,
370 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
371 mapping_gfp_mask(mapping));
375 * Given the page we found in the page cache, return the page corresponding
376 * to this index in the file
378 static inline struct page *find_subpage(struct page *head, pgoff_t index)
380 /* HugeTLBfs wants the head page regardless */
384 return head + (index & (hpage_nr_pages(head) - 1));
387 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
388 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
389 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
390 unsigned int nr_entries, struct page **entries,
392 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
393 pgoff_t end, unsigned int nr_pages,
394 struct page **pages);
395 static inline unsigned find_get_pages(struct address_space *mapping,
396 pgoff_t *start, unsigned int nr_pages,
399 return find_get_pages_range(mapping, start, (pgoff_t)-1, nr_pages,
402 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
403 unsigned int nr_pages, struct page **pages);
404 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
405 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
406 struct page **pages);
407 static inline unsigned find_get_pages_tag(struct address_space *mapping,
408 pgoff_t *index, xa_mark_t tag, unsigned int nr_pages,
411 return find_get_pages_range_tag(mapping, index, (pgoff_t)-1, tag,
415 struct page *grab_cache_page_write_begin(struct address_space *mapping,
416 pgoff_t index, unsigned flags);
419 * Returns locked page at given index in given cache, creating it if needed.
421 static inline struct page *grab_cache_page(struct address_space *mapping,
424 return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
427 extern struct page * read_cache_page(struct address_space *mapping,
428 pgoff_t index, filler_t *filler, void *data);
429 extern struct page * read_cache_page_gfp(struct address_space *mapping,
430 pgoff_t index, gfp_t gfp_mask);
431 extern int read_cache_pages(struct address_space *mapping,
432 struct list_head *pages, filler_t *filler, void *data);
434 static inline struct page *read_mapping_page(struct address_space *mapping,
435 pgoff_t index, void *data)
437 return read_cache_page(mapping, index, NULL, data);
441 * Get index of the page with in radix-tree
442 * (TODO: remove once hugetlb pages will have ->index in PAGE_SIZE)
444 static inline pgoff_t page_to_index(struct page *page)
448 if (likely(!PageTransTail(page)))
452 * We don't initialize ->index for tail pages: calculate based on
455 pgoff = compound_head(page)->index;
456 pgoff += page - compound_head(page);
461 * Get the offset in PAGE_SIZE.
462 * (TODO: hugepage should have ->index in PAGE_SIZE)
464 static inline pgoff_t page_to_pgoff(struct page *page)
466 if (unlikely(PageHeadHuge(page)))
467 return page->index << compound_order(page);
469 return page_to_index(page);
473 * Return byte-offset into filesystem object for page.
475 static inline loff_t page_offset(struct page *page)
477 return ((loff_t)page->index) << PAGE_SHIFT;
480 static inline loff_t page_file_offset(struct page *page)
482 return ((loff_t)page_index(page)) << PAGE_SHIFT;
485 extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
486 unsigned long address);
488 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
489 unsigned long address)
492 if (unlikely(is_vm_hugetlb_page(vma)))
493 return linear_hugepage_index(vma, address);
494 pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
495 pgoff += vma->vm_pgoff;
499 extern void __lock_page(struct page *page);
500 extern int __lock_page_killable(struct page *page);
501 extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
503 extern void unlock_page(struct page *page);
506 * Return true if the page was successfully locked
508 static inline int trylock_page(struct page *page)
510 page = compound_head(page);
511 return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
515 * lock_page may only be called if we have the page's inode pinned.
517 static inline void lock_page(struct page *page)
520 if (!trylock_page(page))
525 * lock_page_killable is like lock_page but can be interrupted by fatal
526 * signals. It returns 0 if it locked the page and -EINTR if it was
527 * killed while waiting.
529 static inline int lock_page_killable(struct page *page)
532 if (!trylock_page(page))
533 return __lock_page_killable(page);
538 * lock_page_or_retry - Lock the page, unless this would block and the
539 * caller indicated that it can handle a retry.
541 * Return value and mmap_sem implications depend on flags; see
542 * __lock_page_or_retry().
544 static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
548 return trylock_page(page) || __lock_page_or_retry(page, mm, flags);
552 * This is exported only for wait_on_page_locked/wait_on_page_writeback, etc.,
553 * and should not be used directly.
555 extern void wait_on_page_bit(struct page *page, int bit_nr);
556 extern int wait_on_page_bit_killable(struct page *page, int bit_nr);
559 * Wait for a page to be unlocked.
561 * This must be called with the caller "holding" the page,
562 * ie with increased "page->count" so that the page won't
563 * go away during the wait..
565 static inline void wait_on_page_locked(struct page *page)
567 if (PageLocked(page))
568 wait_on_page_bit(compound_head(page), PG_locked);
571 static inline int wait_on_page_locked_killable(struct page *page)
573 if (!PageLocked(page))
575 return wait_on_page_bit_killable(compound_head(page), PG_locked);
578 extern void put_and_wait_on_page_locked(struct page *page);
580 void wait_on_page_writeback(struct page *page);
581 extern void end_page_writeback(struct page *page);
582 void wait_for_stable_page(struct page *page);
584 void page_endio(struct page *page, bool is_write, int err);
587 * Add an arbitrary waiter to a page's wait queue
589 extern void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter);
592 * Fault everything in given userspace address range in.
594 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
596 char __user *end = uaddr + size - 1;
598 if (unlikely(size == 0))
601 if (unlikely(uaddr > end))
604 * Writing zeroes into userspace here is OK, because we know that if
605 * the zero gets there, we'll be overwriting it.
608 if (unlikely(__put_user(0, uaddr) != 0))
611 } while (uaddr <= end);
613 /* Check whether the range spilled into the next page. */
614 if (((unsigned long)uaddr & PAGE_MASK) ==
615 ((unsigned long)end & PAGE_MASK))
616 return __put_user(0, end);
621 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
624 const char __user *end = uaddr + size - 1;
626 if (unlikely(size == 0))
629 if (unlikely(uaddr > end))
633 if (unlikely(__get_user(c, uaddr) != 0))
636 } while (uaddr <= end);
638 /* Check whether the range spilled into the next page. */
639 if (((unsigned long)uaddr & PAGE_MASK) ==
640 ((unsigned long)end & PAGE_MASK)) {
641 return __get_user(c, end);
648 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
649 pgoff_t index, gfp_t gfp_mask);
650 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
651 pgoff_t index, gfp_t gfp_mask);
652 extern void delete_from_page_cache(struct page *page);
653 extern void __delete_from_page_cache(struct page *page, void *shadow);
654 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
655 void delete_from_page_cache_batch(struct address_space *mapping,
656 struct pagevec *pvec);
658 #define VM_READAHEAD_PAGES (SZ_128K / PAGE_SIZE)
660 void page_cache_sync_readahead(struct address_space *, struct file_ra_state *,
661 struct file *, pgoff_t index, unsigned long req_count);
662 void page_cache_async_readahead(struct address_space *, struct file_ra_state *,
663 struct file *, struct page *, pgoff_t index,
664 unsigned long req_count);
665 void page_cache_readahead_unbounded(struct address_space *, struct file *,
666 pgoff_t index, unsigned long nr_to_read,
667 unsigned long lookahead_count);
670 * Like add_to_page_cache_locked, but used to add newly allocated pages:
671 * the page is new, so we can just run __SetPageLocked() against it.
673 static inline int add_to_page_cache(struct page *page,
674 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
678 __SetPageLocked(page);
679 error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
681 __ClearPageLocked(page);
686 * struct readahead_control - Describes a readahead request.
688 * A readahead request is for consecutive pages. Filesystems which
689 * implement the ->readahead method should call readahead_page() or
690 * readahead_page_batch() in a loop and attempt to start I/O against
691 * each page in the request.
693 * Most of the fields in this struct are private and should be accessed
694 * by the functions below.
696 * @file: The file, used primarily by network filesystems for authentication.
697 * May be NULL if invoked internally by the filesystem.
698 * @mapping: Readahead this filesystem object.
700 struct readahead_control {
702 struct address_space *mapping;
703 /* private: use the readahead_* accessors instead */
705 unsigned int _nr_pages;
706 unsigned int _batch_count;
710 * readahead_page - Get the next page to read.
711 * @rac: The current readahead request.
713 * Context: The page is locked and has an elevated refcount. The caller
714 * should decreases the refcount once the page has been submitted for I/O
715 * and unlock the page once all I/O to that page has completed.
716 * Return: A pointer to the next page, or %NULL if we are done.
718 static inline struct page *readahead_page(struct readahead_control *rac)
722 BUG_ON(rac->_batch_count > rac->_nr_pages);
723 rac->_nr_pages -= rac->_batch_count;
724 rac->_index += rac->_batch_count;
726 if (!rac->_nr_pages) {
727 rac->_batch_count = 0;
731 page = xa_load(&rac->mapping->i_pages, rac->_index);
732 VM_BUG_ON_PAGE(!PageLocked(page), page);
733 rac->_batch_count = hpage_nr_pages(page);
738 static inline unsigned int __readahead_batch(struct readahead_control *rac,
739 struct page **array, unsigned int array_sz)
742 XA_STATE(xas, &rac->mapping->i_pages, 0);
745 BUG_ON(rac->_batch_count > rac->_nr_pages);
746 rac->_nr_pages -= rac->_batch_count;
747 rac->_index += rac->_batch_count;
748 rac->_batch_count = 0;
750 xas_set(&xas, rac->_index);
752 xas_for_each(&xas, page, rac->_index + rac->_nr_pages - 1) {
753 VM_BUG_ON_PAGE(!PageLocked(page), page);
754 VM_BUG_ON_PAGE(PageTail(page), page);
756 rac->_batch_count += hpage_nr_pages(page);
759 * The page cache isn't using multi-index entries yet,
760 * so the xas cursor needs to be manually moved to the
761 * next index. This can be removed once the page cache
765 xas_set(&xas, rac->_index + rac->_batch_count);
776 * readahead_page_batch - Get a batch of pages to read.
777 * @rac: The current readahead request.
778 * @array: An array of pointers to struct page.
780 * Context: The pages are locked and have an elevated refcount. The caller
781 * should decreases the refcount once the page has been submitted for I/O
782 * and unlock the page once all I/O to that page has completed.
783 * Return: The number of pages placed in the array. 0 indicates the request
786 #define readahead_page_batch(rac, array) \
787 __readahead_batch(rac, array, ARRAY_SIZE(array))
790 * readahead_pos - The byte offset into the file of this readahead request.
791 * @rac: The readahead request.
793 static inline loff_t readahead_pos(struct readahead_control *rac)
795 return (loff_t)rac->_index * PAGE_SIZE;
799 * readahead_length - The number of bytes in this readahead request.
800 * @rac: The readahead request.
802 static inline loff_t readahead_length(struct readahead_control *rac)
804 return (loff_t)rac->_nr_pages * PAGE_SIZE;
808 * readahead_index - The index of the first page in this readahead request.
809 * @rac: The readahead request.
811 static inline pgoff_t readahead_index(struct readahead_control *rac)
817 * readahead_count - The number of pages in this readahead request.
818 * @rac: The readahead request.
820 static inline unsigned int readahead_count(struct readahead_control *rac)
822 return rac->_nr_pages;
825 static inline unsigned long dir_pages(struct inode *inode)
827 return (unsigned long)(inode->i_size + PAGE_SIZE - 1) >>
832 * page_mkwrite_check_truncate - check if page was truncated
833 * @page: the page to check
834 * @inode: the inode to check the page against
836 * Returns the number of bytes in the page up to EOF,
837 * or -EFAULT if the page was truncated.
839 static inline int page_mkwrite_check_truncate(struct page *page,
842 loff_t size = i_size_read(inode);
843 pgoff_t index = size >> PAGE_SHIFT;
844 int offset = offset_in_page(size);
846 if (page->mapping != inode->i_mapping)
849 /* page is wholly inside EOF */
850 if (page->index < index)
852 /* page is wholly past EOF */
853 if (page->index > index || !offset)
855 /* page is partially inside EOF */
859 #endif /* _LINUX_PAGEMAP_H */