1 #ifndef _LINUX_PAGEMAP_H
2 #define _LINUX_PAGEMAP_H
5 * Copyright 1995 Linus Torvalds
9 #include <linux/list.h>
10 #include <linux/highmem.h>
11 #include <linux/compiler.h>
12 #include <asm/uaccess.h>
13 #include <linux/gfp.h>
14 #include <linux/bitops.h>
15 #include <linux/hardirq.h> /* for in_interrupt() */
16 #include <linux/hugetlb_inline.h>
19 * Bits in mapping->flags. The lower __GFP_BITS_SHIFT bits are the page
20 * allocation mode flags.
23 AS_EIO = __GFP_BITS_SHIFT + 0, /* IO error on async write */
24 AS_ENOSPC = __GFP_BITS_SHIFT + 1, /* ENOSPC on async write */
25 AS_MM_ALL_LOCKS = __GFP_BITS_SHIFT + 2, /* under mm_take_all_locks() */
26 AS_UNEVICTABLE = __GFP_BITS_SHIFT + 3, /* e.g., ramdisk, SHM_LOCK */
27 AS_EXITING = __GFP_BITS_SHIFT + 4, /* final truncate in progress */
28 /* writeback related tags are not used */
29 AS_NO_WRITEBACK_TAGS = __GFP_BITS_SHIFT + 5,
32 static inline void mapping_set_error(struct address_space *mapping, int error)
34 if (unlikely(error)) {
36 set_bit(AS_ENOSPC, &mapping->flags);
38 set_bit(AS_EIO, &mapping->flags);
42 static inline void mapping_set_unevictable(struct address_space *mapping)
44 set_bit(AS_UNEVICTABLE, &mapping->flags);
47 static inline void mapping_clear_unevictable(struct address_space *mapping)
49 clear_bit(AS_UNEVICTABLE, &mapping->flags);
52 static inline int mapping_unevictable(struct address_space *mapping)
55 return test_bit(AS_UNEVICTABLE, &mapping->flags);
59 static inline void mapping_set_exiting(struct address_space *mapping)
61 set_bit(AS_EXITING, &mapping->flags);
64 static inline int mapping_exiting(struct address_space *mapping)
66 return test_bit(AS_EXITING, &mapping->flags);
69 static inline void mapping_set_no_writeback_tags(struct address_space *mapping)
71 set_bit(AS_NO_WRITEBACK_TAGS, &mapping->flags);
74 static inline int mapping_use_writeback_tags(struct address_space *mapping)
76 return !test_bit(AS_NO_WRITEBACK_TAGS, &mapping->flags);
79 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
81 return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
84 /* Restricts the given gfp_mask to what the mapping allows. */
85 static inline gfp_t mapping_gfp_constraint(struct address_space *mapping,
88 return mapping_gfp_mask(mapping) & gfp_mask;
92 * This is non-atomic. Only to be used before the mapping is activated.
93 * Probably needs a barrier...
95 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
97 m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
98 (__force unsigned long)mask;
101 void release_pages(struct page **pages, int nr, bool cold);
104 * speculatively take a reference to a page.
105 * If the page is free (_refcount == 0), then _refcount is untouched, and 0
106 * is returned. Otherwise, _refcount is incremented by 1 and 1 is returned.
108 * This function must be called inside the same rcu_read_lock() section as has
109 * been used to lookup the page in the pagecache radix-tree (or page table):
110 * this allows allocators to use a synchronize_rcu() to stabilize _refcount.
112 * Unless an RCU grace period has passed, the count of all pages coming out
113 * of the allocator must be considered unstable. page_count may return higher
114 * than expected, and put_page must be able to do the right thing when the
115 * page has been finished with, no matter what it is subsequently allocated
116 * for (because put_page is what is used here to drop an invalid speculative
119 * This is the interesting part of the lockless pagecache (and lockless
120 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
121 * has the following pattern:
122 * 1. find page in radix tree
123 * 2. conditionally increment refcount
124 * 3. check the page is still in pagecache (if no, goto 1)
126 * Remove-side that cares about stability of _refcount (eg. reclaim) has the
127 * following (with tree_lock held for write):
128 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
129 * B. remove page from pagecache
132 * There are 2 critical interleavings that matter:
133 * - 2 runs before A: in this case, A sees elevated refcount and bails out
134 * - A runs before 2: in this case, 2 sees zero refcount and retries;
135 * subsequently, B will complete and 1 will find no page, causing the
136 * lookup to return NULL.
138 * It is possible that between 1 and 2, the page is removed then the exact same
139 * page is inserted into the same position in pagecache. That's OK: the
140 * old find_get_page using tree_lock could equally have run before or after
141 * such a re-insertion, depending on order that locks are granted.
143 * Lookups racing against pagecache insertion isn't a big problem: either 1
144 * will find the page or it will not. Likewise, the old find_get_page could run
145 * either before the insertion or afterwards, depending on timing.
147 static inline int page_cache_get_speculative(struct page *page)
149 VM_BUG_ON(in_interrupt());
151 #ifdef CONFIG_TINY_RCU
152 # ifdef CONFIG_PREEMPT_COUNT
153 VM_BUG_ON(!in_atomic());
156 * Preempt must be disabled here - we rely on rcu_read_lock doing
159 * Pagecache won't be truncated from interrupt context, so if we have
160 * found a page in the radix tree here, we have pinned its refcount by
161 * disabling preempt, and hence no need for the "speculative get" that
164 VM_BUG_ON_PAGE(page_count(page) == 0, page);
168 if (unlikely(!get_page_unless_zero(page))) {
170 * Either the page has been freed, or will be freed.
171 * In either case, retry here and the caller should
172 * do the right thing (see comments above).
177 VM_BUG_ON_PAGE(PageTail(page), page);
183 * Same as above, but add instead of inc (could just be merged)
185 static inline int page_cache_add_speculative(struct page *page, int count)
187 VM_BUG_ON(in_interrupt());
189 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
190 # ifdef CONFIG_PREEMPT_COUNT
191 VM_BUG_ON(!in_atomic());
193 VM_BUG_ON_PAGE(page_count(page) == 0, page);
194 page_ref_add(page, count);
197 if (unlikely(!page_ref_add_unless(page, count, 0)))
200 VM_BUG_ON_PAGE(PageCompound(page) && page != compound_head(page), page);
206 extern struct page *__page_cache_alloc(gfp_t gfp);
208 static inline struct page *__page_cache_alloc(gfp_t gfp)
210 return alloc_pages(gfp, 0);
214 static inline struct page *page_cache_alloc(struct address_space *x)
216 return __page_cache_alloc(mapping_gfp_mask(x));
219 static inline struct page *page_cache_alloc_cold(struct address_space *x)
221 return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
224 static inline gfp_t readahead_gfp_mask(struct address_space *x)
226 return mapping_gfp_mask(x) |
227 __GFP_COLD | __GFP_NORETRY | __GFP_NOWARN;
230 typedef int filler_t(void *, struct page *);
232 pgoff_t page_cache_next_hole(struct address_space *mapping,
233 pgoff_t index, unsigned long max_scan);
234 pgoff_t page_cache_prev_hole(struct address_space *mapping,
235 pgoff_t index, unsigned long max_scan);
237 #define FGP_ACCESSED 0x00000001
238 #define FGP_LOCK 0x00000002
239 #define FGP_CREAT 0x00000004
240 #define FGP_WRITE 0x00000008
241 #define FGP_NOFS 0x00000010
242 #define FGP_NOWAIT 0x00000020
244 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
245 int fgp_flags, gfp_t cache_gfp_mask);
248 * find_get_page - find and get a page reference
249 * @mapping: the address_space to search
250 * @offset: the page index
252 * Looks up the page cache slot at @mapping & @offset. If there is a
253 * page cache page, it is returned with an increased refcount.
255 * Otherwise, %NULL is returned.
257 static inline struct page *find_get_page(struct address_space *mapping,
260 return pagecache_get_page(mapping, offset, 0, 0);
263 static inline struct page *find_get_page_flags(struct address_space *mapping,
264 pgoff_t offset, int fgp_flags)
266 return pagecache_get_page(mapping, offset, fgp_flags, 0);
270 * find_lock_page - locate, pin and lock a pagecache page
271 * pagecache_get_page - find and get a page reference
272 * @mapping: the address_space to search
273 * @offset: the page index
275 * Looks up the page cache slot at @mapping & @offset. If there is a
276 * page cache page, it is returned locked and with an increased
279 * Otherwise, %NULL is returned.
281 * find_lock_page() may sleep.
283 static inline struct page *find_lock_page(struct address_space *mapping,
286 return pagecache_get_page(mapping, offset, FGP_LOCK, 0);
290 * find_or_create_page - locate or add a pagecache page
291 * @mapping: the page's address_space
292 * @index: the page's index into the mapping
293 * @gfp_mask: page allocation mode
295 * Looks up the page cache slot at @mapping & @offset. If there is a
296 * page cache page, it is returned locked and with an increased
299 * If the page is not present, a new page is allocated using @gfp_mask
300 * and added to the page cache and the VM's LRU list. The page is
301 * returned locked and with an increased refcount.
303 * On memory exhaustion, %NULL is returned.
305 * find_or_create_page() may sleep, even if @gfp_flags specifies an
308 static inline struct page *find_or_create_page(struct address_space *mapping,
309 pgoff_t offset, gfp_t gfp_mask)
311 return pagecache_get_page(mapping, offset,
312 FGP_LOCK|FGP_ACCESSED|FGP_CREAT,
317 * grab_cache_page_nowait - returns locked page at given index in given cache
318 * @mapping: target address_space
319 * @index: the page index
321 * Same as grab_cache_page(), but do not wait if the page is unavailable.
322 * This is intended for speculative data generators, where the data can
323 * be regenerated if the page couldn't be grabbed. This routine should
324 * be safe to call while holding the lock for another page.
326 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
327 * and deadlock against the caller's locked page.
329 static inline struct page *grab_cache_page_nowait(struct address_space *mapping,
332 return pagecache_get_page(mapping, index,
333 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
334 mapping_gfp_mask(mapping));
337 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
338 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
339 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
340 unsigned int nr_entries, struct page **entries,
342 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
343 unsigned int nr_pages, struct page **pages);
344 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
345 unsigned int nr_pages, struct page **pages);
346 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
347 int tag, unsigned int nr_pages, struct page **pages);
348 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
349 int tag, unsigned int nr_entries,
350 struct page **entries, pgoff_t *indices);
352 struct page *grab_cache_page_write_begin(struct address_space *mapping,
353 pgoff_t index, unsigned flags);
356 * Returns locked page at given index in given cache, creating it if needed.
358 static inline struct page *grab_cache_page(struct address_space *mapping,
361 return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
364 extern struct page * read_cache_page(struct address_space *mapping,
365 pgoff_t index, filler_t *filler, void *data);
366 extern struct page * read_cache_page_gfp(struct address_space *mapping,
367 pgoff_t index, gfp_t gfp_mask);
368 extern int read_cache_pages(struct address_space *mapping,
369 struct list_head *pages, filler_t *filler, void *data);
371 static inline struct page *read_mapping_page(struct address_space *mapping,
372 pgoff_t index, void *data)
374 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
375 return read_cache_page(mapping, index, filler, data);
379 * Get the offset in PAGE_SIZE.
380 * (TODO: hugepage should have ->index in PAGE_SIZE)
382 static inline pgoff_t page_to_pgoff(struct page *page)
386 if (unlikely(PageHeadHuge(page)))
387 return page->index << compound_order(page);
389 if (likely(!PageTransTail(page)))
393 * We don't initialize ->index for tail pages: calculate based on
396 pgoff = compound_head(page)->index;
397 pgoff += page - compound_head(page);
402 * Return byte-offset into filesystem object for page.
404 static inline loff_t page_offset(struct page *page)
406 return ((loff_t)page->index) << PAGE_SHIFT;
409 static inline loff_t page_file_offset(struct page *page)
411 return ((loff_t)page_index(page)) << PAGE_SHIFT;
414 extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
415 unsigned long address);
417 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
418 unsigned long address)
421 if (unlikely(is_vm_hugetlb_page(vma)))
422 return linear_hugepage_index(vma, address);
423 pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
424 pgoff += vma->vm_pgoff;
428 extern void __lock_page(struct page *page);
429 extern int __lock_page_killable(struct page *page);
430 extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
432 extern void unlock_page(struct page *page);
434 static inline int trylock_page(struct page *page)
436 page = compound_head(page);
437 return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
441 * lock_page may only be called if we have the page's inode pinned.
443 static inline void lock_page(struct page *page)
446 if (!trylock_page(page))
451 * lock_page_killable is like lock_page but can be interrupted by fatal
452 * signals. It returns 0 if it locked the page and -EINTR if it was
453 * killed while waiting.
455 static inline int lock_page_killable(struct page *page)
458 if (!trylock_page(page))
459 return __lock_page_killable(page);
464 * lock_page_or_retry - Lock the page, unless this would block and the
465 * caller indicated that it can handle a retry.
467 * Return value and mmap_sem implications depend on flags; see
468 * __lock_page_or_retry().
470 static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
474 return trylock_page(page) || __lock_page_or_retry(page, mm, flags);
478 * This is exported only for wait_on_page_locked/wait_on_page_writeback,
479 * and for filesystems which need to wait on PG_private.
481 extern void wait_on_page_bit(struct page *page, int bit_nr);
483 extern int wait_on_page_bit_killable(struct page *page, int bit_nr);
484 extern int wait_on_page_bit_killable_timeout(struct page *page,
485 int bit_nr, unsigned long timeout);
487 static inline int wait_on_page_locked_killable(struct page *page)
489 if (!PageLocked(page))
491 return wait_on_page_bit_killable(compound_head(page), PG_locked);
494 extern wait_queue_head_t *page_waitqueue(struct page *page);
495 static inline void wake_up_page(struct page *page, int bit)
497 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
501 * Wait for a page to be unlocked.
503 * This must be called with the caller "holding" the page,
504 * ie with increased "page->count" so that the page won't
505 * go away during the wait..
507 static inline void wait_on_page_locked(struct page *page)
509 if (PageLocked(page))
510 wait_on_page_bit(compound_head(page), PG_locked);
514 * Wait for a page to complete writeback
516 static inline void wait_on_page_writeback(struct page *page)
518 if (PageWriteback(page))
519 wait_on_page_bit(page, PG_writeback);
522 extern void end_page_writeback(struct page *page);
523 void wait_for_stable_page(struct page *page);
525 void page_endio(struct page *page, bool is_write, int err);
528 * Add an arbitrary waiter to a page's wait queue
530 extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter);
533 * Fault everything in given userspace address range in.
535 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
537 char __user *end = uaddr + size - 1;
539 if (unlikely(size == 0))
542 if (unlikely(uaddr > end))
545 * Writing zeroes into userspace here is OK, because we know that if
546 * the zero gets there, we'll be overwriting it.
549 if (unlikely(__put_user(0, uaddr) != 0))
552 } while (uaddr <= end);
554 /* Check whether the range spilled into the next page. */
555 if (((unsigned long)uaddr & PAGE_MASK) ==
556 ((unsigned long)end & PAGE_MASK))
557 return __put_user(0, end);
562 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
565 const char __user *end = uaddr + size - 1;
567 if (unlikely(size == 0))
570 if (unlikely(uaddr > end))
574 if (unlikely(__get_user(c, uaddr) != 0))
577 } while (uaddr <= end);
579 /* Check whether the range spilled into the next page. */
580 if (((unsigned long)uaddr & PAGE_MASK) ==
581 ((unsigned long)end & PAGE_MASK)) {
582 return __get_user(c, end);
589 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
590 pgoff_t index, gfp_t gfp_mask);
591 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
592 pgoff_t index, gfp_t gfp_mask);
593 extern void delete_from_page_cache(struct page *page);
594 extern void __delete_from_page_cache(struct page *page, void *shadow);
595 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
598 * Like add_to_page_cache_locked, but used to add newly allocated pages:
599 * the page is new, so we can just run __SetPageLocked() against it.
601 static inline int add_to_page_cache(struct page *page,
602 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
606 __SetPageLocked(page);
607 error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
609 __ClearPageLocked(page);
613 static inline unsigned long dir_pages(struct inode *inode)
615 return (unsigned long)(inode->i_size + PAGE_SIZE - 1) >>
619 #endif /* _LINUX_PAGEMAP_H */