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 flags for now, for legacy callers */
58 set_bit(AS_ENOSPC, &mapping->flags);
60 set_bit(AS_EIO, &mapping->flags);
63 static inline void mapping_set_unevictable(struct address_space *mapping)
65 set_bit(AS_UNEVICTABLE, &mapping->flags);
68 static inline void mapping_clear_unevictable(struct address_space *mapping)
70 clear_bit(AS_UNEVICTABLE, &mapping->flags);
73 static inline int mapping_unevictable(struct address_space *mapping)
76 return test_bit(AS_UNEVICTABLE, &mapping->flags);
80 static inline void mapping_set_exiting(struct address_space *mapping)
82 set_bit(AS_EXITING, &mapping->flags);
85 static inline int mapping_exiting(struct address_space *mapping)
87 return test_bit(AS_EXITING, &mapping->flags);
90 static inline void mapping_set_no_writeback_tags(struct address_space *mapping)
92 set_bit(AS_NO_WRITEBACK_TAGS, &mapping->flags);
95 static inline int mapping_use_writeback_tags(struct address_space *mapping)
97 return !test_bit(AS_NO_WRITEBACK_TAGS, &mapping->flags);
100 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
102 return mapping->gfp_mask;
105 /* Restricts the given gfp_mask to what the mapping allows. */
106 static inline gfp_t mapping_gfp_constraint(struct address_space *mapping,
109 return mapping_gfp_mask(mapping) & gfp_mask;
113 * This is non-atomic. Only to be used before the mapping is activated.
114 * Probably needs a barrier...
116 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
121 void release_pages(struct page **pages, int nr);
124 * speculatively take a reference to a page.
125 * If the page is free (_refcount == 0), then _refcount is untouched, and 0
126 * is returned. Otherwise, _refcount is incremented by 1 and 1 is returned.
128 * This function must be called inside the same rcu_read_lock() section as has
129 * been used to lookup the page in the pagecache radix-tree (or page table):
130 * this allows allocators to use a synchronize_rcu() to stabilize _refcount.
132 * Unless an RCU grace period has passed, the count of all pages coming out
133 * of the allocator must be considered unstable. page_count may return higher
134 * than expected, and put_page must be able to do the right thing when the
135 * page has been finished with, no matter what it is subsequently allocated
136 * for (because put_page is what is used here to drop an invalid speculative
139 * This is the interesting part of the lockless pagecache (and lockless
140 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
141 * has the following pattern:
142 * 1. find page in radix tree
143 * 2. conditionally increment refcount
144 * 3. check the page is still in pagecache (if no, goto 1)
146 * Remove-side that cares about stability of _refcount (eg. reclaim) has the
147 * following (with tree_lock held for write):
148 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
149 * B. remove page from pagecache
152 * There are 2 critical interleavings that matter:
153 * - 2 runs before A: in this case, A sees elevated refcount and bails out
154 * - A runs before 2: in this case, 2 sees zero refcount and retries;
155 * subsequently, B will complete and 1 will find no page, causing the
156 * lookup to return NULL.
158 * It is possible that between 1 and 2, the page is removed then the exact same
159 * page is inserted into the same position in pagecache. That's OK: the
160 * old find_get_page using tree_lock could equally have run before or after
161 * such a re-insertion, depending on order that locks are granted.
163 * Lookups racing against pagecache insertion isn't a big problem: either 1
164 * will find the page or it will not. Likewise, the old find_get_page could run
165 * either before the insertion or afterwards, depending on timing.
167 static inline int page_cache_get_speculative(struct page *page)
169 #ifdef CONFIG_TINY_RCU
170 # ifdef CONFIG_PREEMPT_COUNT
171 VM_BUG_ON(!in_atomic() && !irqs_disabled());
174 * Preempt must be disabled here - we rely on rcu_read_lock doing
177 * Pagecache won't be truncated from interrupt context, so if we have
178 * found a page in the radix tree here, we have pinned its refcount by
179 * disabling preempt, and hence no need for the "speculative get" that
182 VM_BUG_ON_PAGE(page_count(page) == 0, page);
186 if (unlikely(!get_page_unless_zero(page))) {
188 * Either the page has been freed, or will be freed.
189 * In either case, retry here and the caller should
190 * do the right thing (see comments above).
195 VM_BUG_ON_PAGE(PageTail(page), page);
201 * Same as above, but add instead of inc (could just be merged)
203 static inline int page_cache_add_speculative(struct page *page, int count)
205 VM_BUG_ON(in_interrupt());
207 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
208 # ifdef CONFIG_PREEMPT_COUNT
209 VM_BUG_ON(!in_atomic() && !irqs_disabled());
211 VM_BUG_ON_PAGE(page_count(page) == 0, page);
212 page_ref_add(page, count);
215 if (unlikely(!page_ref_add_unless(page, count, 0)))
218 VM_BUG_ON_PAGE(PageCompound(page) && page != compound_head(page), page);
224 extern struct page *__page_cache_alloc(gfp_t gfp);
226 static inline struct page *__page_cache_alloc(gfp_t gfp)
228 return alloc_pages(gfp, 0);
232 static inline struct page *page_cache_alloc(struct address_space *x)
234 return __page_cache_alloc(mapping_gfp_mask(x));
237 static inline gfp_t readahead_gfp_mask(struct address_space *x)
239 return mapping_gfp_mask(x) | __GFP_NORETRY | __GFP_NOWARN;
242 typedef int filler_t(void *, struct page *);
244 pgoff_t page_cache_next_hole(struct address_space *mapping,
245 pgoff_t index, unsigned long max_scan);
246 pgoff_t page_cache_prev_hole(struct address_space *mapping,
247 pgoff_t index, unsigned long max_scan);
249 #define FGP_ACCESSED 0x00000001
250 #define FGP_LOCK 0x00000002
251 #define FGP_CREAT 0x00000004
252 #define FGP_WRITE 0x00000008
253 #define FGP_NOFS 0x00000010
254 #define FGP_NOWAIT 0x00000020
256 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
257 int fgp_flags, gfp_t cache_gfp_mask);
260 * find_get_page - find and get a page reference
261 * @mapping: the address_space to search
262 * @offset: the page index
264 * Looks up the page cache slot at @mapping & @offset. If there is a
265 * page cache page, it is returned with an increased refcount.
267 * Otherwise, %NULL is returned.
269 static inline struct page *find_get_page(struct address_space *mapping,
272 return pagecache_get_page(mapping, offset, 0, 0);
275 static inline struct page *find_get_page_flags(struct address_space *mapping,
276 pgoff_t offset, int fgp_flags)
278 return pagecache_get_page(mapping, offset, fgp_flags, 0);
282 * find_lock_page - locate, pin and lock a pagecache page
283 * @mapping: the address_space to search
284 * @offset: the page index
286 * Looks up the page cache slot at @mapping & @offset. If there is a
287 * page cache page, it is returned locked and with an increased
290 * Otherwise, %NULL is returned.
292 * find_lock_page() may sleep.
294 static inline struct page *find_lock_page(struct address_space *mapping,
297 return pagecache_get_page(mapping, offset, FGP_LOCK, 0);
301 * find_or_create_page - locate or add a pagecache page
302 * @mapping: the page's address_space
303 * @index: the page's index into the mapping
304 * @gfp_mask: page allocation mode
306 * Looks up the page cache slot at @mapping & @offset. If there is a
307 * page cache page, it is returned locked and with an increased
310 * If the page is not present, a new page is allocated using @gfp_mask
311 * and added to the page cache and the VM's LRU list. The page is
312 * returned locked and with an increased refcount.
314 * On memory exhaustion, %NULL is returned.
316 * find_or_create_page() may sleep, even if @gfp_flags specifies an
319 static inline struct page *find_or_create_page(struct address_space *mapping,
320 pgoff_t offset, gfp_t gfp_mask)
322 return pagecache_get_page(mapping, offset,
323 FGP_LOCK|FGP_ACCESSED|FGP_CREAT,
328 * grab_cache_page_nowait - returns locked page at given index in given cache
329 * @mapping: target address_space
330 * @index: the page index
332 * Same as grab_cache_page(), but do not wait if the page is unavailable.
333 * This is intended for speculative data generators, where the data can
334 * be regenerated if the page couldn't be grabbed. This routine should
335 * be safe to call while holding the lock for another page.
337 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
338 * and deadlock against the caller's locked page.
340 static inline struct page *grab_cache_page_nowait(struct address_space *mapping,
343 return pagecache_get_page(mapping, index,
344 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
345 mapping_gfp_mask(mapping));
348 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
349 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
350 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
351 unsigned int nr_entries, struct page **entries,
353 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
354 pgoff_t end, unsigned int nr_pages,
355 struct page **pages);
356 static inline unsigned find_get_pages(struct address_space *mapping,
357 pgoff_t *start, unsigned int nr_pages,
360 return find_get_pages_range(mapping, start, (pgoff_t)-1, nr_pages,
363 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
364 unsigned int nr_pages, struct page **pages);
365 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
366 pgoff_t end, int tag, unsigned int nr_pages,
367 struct page **pages);
368 static inline unsigned find_get_pages_tag(struct address_space *mapping,
369 pgoff_t *index, int tag, unsigned int nr_pages,
372 return find_get_pages_range_tag(mapping, index, (pgoff_t)-1, tag,
375 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
376 int tag, unsigned int nr_entries,
377 struct page **entries, pgoff_t *indices);
379 struct page *grab_cache_page_write_begin(struct address_space *mapping,
380 pgoff_t index, unsigned flags);
383 * Returns locked page at given index in given cache, creating it if needed.
385 static inline struct page *grab_cache_page(struct address_space *mapping,
388 return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
391 extern struct page * read_cache_page(struct address_space *mapping,
392 pgoff_t index, filler_t *filler, void *data);
393 extern struct page * read_cache_page_gfp(struct address_space *mapping,
394 pgoff_t index, gfp_t gfp_mask);
395 extern int read_cache_pages(struct address_space *mapping,
396 struct list_head *pages, filler_t *filler, void *data);
398 static inline struct page *read_mapping_page(struct address_space *mapping,
399 pgoff_t index, void *data)
401 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
402 return read_cache_page(mapping, index, filler, data);
406 * Get index of the page with in radix-tree
407 * (TODO: remove once hugetlb pages will have ->index in PAGE_SIZE)
409 static inline pgoff_t page_to_index(struct page *page)
413 if (likely(!PageTransTail(page)))
417 * We don't initialize ->index for tail pages: calculate based on
420 pgoff = compound_head(page)->index;
421 pgoff += page - compound_head(page);
426 * Get the offset in PAGE_SIZE.
427 * (TODO: hugepage should have ->index in PAGE_SIZE)
429 static inline pgoff_t page_to_pgoff(struct page *page)
431 if (unlikely(PageHeadHuge(page)))
432 return page->index << compound_order(page);
434 return page_to_index(page);
438 * Return byte-offset into filesystem object for page.
440 static inline loff_t page_offset(struct page *page)
442 return ((loff_t)page->index) << PAGE_SHIFT;
445 static inline loff_t page_file_offset(struct page *page)
447 return ((loff_t)page_index(page)) << PAGE_SHIFT;
450 extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
451 unsigned long address);
453 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
454 unsigned long address)
457 if (unlikely(is_vm_hugetlb_page(vma)))
458 return linear_hugepage_index(vma, address);
459 pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
460 pgoff += vma->vm_pgoff;
464 extern void __lock_page(struct page *page);
465 extern int __lock_page_killable(struct page *page);
466 extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
468 extern void unlock_page(struct page *page);
470 static inline int trylock_page(struct page *page)
472 page = compound_head(page);
473 return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
477 * lock_page may only be called if we have the page's inode pinned.
479 static inline void lock_page(struct page *page)
482 if (!trylock_page(page))
487 * lock_page_killable is like lock_page but can be interrupted by fatal
488 * signals. It returns 0 if it locked the page and -EINTR if it was
489 * killed while waiting.
491 static inline int lock_page_killable(struct page *page)
494 if (!trylock_page(page))
495 return __lock_page_killable(page);
500 * lock_page_or_retry - Lock the page, unless this would block and the
501 * caller indicated that it can handle a retry.
503 * Return value and mmap_sem implications depend on flags; see
504 * __lock_page_or_retry().
506 static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
510 return trylock_page(page) || __lock_page_or_retry(page, mm, flags);
514 * This is exported only for wait_on_page_locked/wait_on_page_writeback, etc.,
515 * and should not be used directly.
517 extern void wait_on_page_bit(struct page *page, int bit_nr);
518 extern int wait_on_page_bit_killable(struct page *page, int bit_nr);
521 * Wait for a page to be unlocked.
523 * This must be called with the caller "holding" the page,
524 * ie with increased "page->count" so that the page won't
525 * go away during the wait..
527 static inline void wait_on_page_locked(struct page *page)
529 if (PageLocked(page))
530 wait_on_page_bit(compound_head(page), PG_locked);
533 static inline int wait_on_page_locked_killable(struct page *page)
535 if (!PageLocked(page))
537 return wait_on_page_bit_killable(compound_head(page), PG_locked);
541 * Wait for a page to complete writeback
543 static inline void wait_on_page_writeback(struct page *page)
545 if (PageWriteback(page))
546 wait_on_page_bit(page, PG_writeback);
549 extern void end_page_writeback(struct page *page);
550 void wait_for_stable_page(struct page *page);
552 void page_endio(struct page *page, bool is_write, int err);
555 * Add an arbitrary waiter to a page's wait queue
557 extern void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter);
560 * Fault everything in given userspace address range in.
562 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
564 char __user *end = uaddr + size - 1;
566 if (unlikely(size == 0))
569 if (unlikely(uaddr > end))
572 * Writing zeroes into userspace here is OK, because we know that if
573 * the zero gets there, we'll be overwriting it.
576 if (unlikely(__put_user(0, uaddr) != 0))
579 } while (uaddr <= end);
581 /* Check whether the range spilled into the next page. */
582 if (((unsigned long)uaddr & PAGE_MASK) ==
583 ((unsigned long)end & PAGE_MASK))
584 return __put_user(0, end);
589 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
592 const char __user *end = uaddr + size - 1;
594 if (unlikely(size == 0))
597 if (unlikely(uaddr > end))
601 if (unlikely(__get_user(c, uaddr) != 0))
604 } while (uaddr <= end);
606 /* Check whether the range spilled into the next page. */
607 if (((unsigned long)uaddr & PAGE_MASK) ==
608 ((unsigned long)end & PAGE_MASK)) {
609 return __get_user(c, end);
616 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
617 pgoff_t index, gfp_t gfp_mask);
618 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
619 pgoff_t index, gfp_t gfp_mask);
620 extern void delete_from_page_cache(struct page *page);
621 extern void __delete_from_page_cache(struct page *page, void *shadow);
622 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
623 void delete_from_page_cache_batch(struct address_space *mapping,
624 struct pagevec *pvec);
627 * Like add_to_page_cache_locked, but used to add newly allocated pages:
628 * the page is new, so we can just run __SetPageLocked() against it.
630 static inline int add_to_page_cache(struct page *page,
631 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
635 __SetPageLocked(page);
636 error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
638 __ClearPageLocked(page);
642 static inline unsigned long dir_pages(struct inode *inode)
644 return (unsigned long)(inode->i_size + PAGE_SIZE - 1) >>
648 #endif /* _LINUX_PAGEMAP_H */