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.
22 AS_EIO
= 0, /* IO error on async write */
23 AS_ENOSPC
= 1, /* ENOSPC on async write */
24 AS_MM_ALL_LOCKS
= 2, /* under mm_take_all_locks() */
25 AS_UNEVICTABLE
= 3, /* e.g., ramdisk, SHM_LOCK */
26 AS_EXITING
= 4, /* final truncate in progress */
27 /* writeback related tags are not used */
28 AS_NO_WRITEBACK_TAGS
= 5,
31 static inline void mapping_set_error(struct address_space
*mapping
, int error
)
33 if (unlikely(error
)) {
35 set_bit(AS_ENOSPC
, &mapping
->flags
);
37 set_bit(AS_EIO
, &mapping
->flags
);
41 static inline void mapping_set_unevictable(struct address_space
*mapping
)
43 set_bit(AS_UNEVICTABLE
, &mapping
->flags
);
46 static inline void mapping_clear_unevictable(struct address_space
*mapping
)
48 clear_bit(AS_UNEVICTABLE
, &mapping
->flags
);
51 static inline int mapping_unevictable(struct address_space
*mapping
)
54 return test_bit(AS_UNEVICTABLE
, &mapping
->flags
);
58 static inline void mapping_set_exiting(struct address_space
*mapping
)
60 set_bit(AS_EXITING
, &mapping
->flags
);
63 static inline int mapping_exiting(struct address_space
*mapping
)
65 return test_bit(AS_EXITING
, &mapping
->flags
);
68 static inline void mapping_set_no_writeback_tags(struct address_space
*mapping
)
70 set_bit(AS_NO_WRITEBACK_TAGS
, &mapping
->flags
);
73 static inline int mapping_use_writeback_tags(struct address_space
*mapping
)
75 return !test_bit(AS_NO_WRITEBACK_TAGS
, &mapping
->flags
);
78 static inline gfp_t
mapping_gfp_mask(struct address_space
* mapping
)
80 return mapping
->gfp_mask
;
83 /* Restricts the given gfp_mask to what the mapping allows. */
84 static inline gfp_t
mapping_gfp_constraint(struct address_space
*mapping
,
87 return mapping_gfp_mask(mapping
) & gfp_mask
;
91 * This is non-atomic. Only to be used before the mapping is activated.
92 * Probably needs a barrier...
94 static inline void mapping_set_gfp_mask(struct address_space
*m
, gfp_t mask
)
99 void release_pages(struct page
**pages
, int nr
, bool cold
);
102 * speculatively take a reference to a page.
103 * If the page is free (_refcount == 0), then _refcount is untouched, and 0
104 * is returned. Otherwise, _refcount is incremented by 1 and 1 is returned.
106 * This function must be called inside the same rcu_read_lock() section as has
107 * been used to lookup the page in the pagecache radix-tree (or page table):
108 * this allows allocators to use a synchronize_rcu() to stabilize _refcount.
110 * Unless an RCU grace period has passed, the count of all pages coming out
111 * of the allocator must be considered unstable. page_count may return higher
112 * than expected, and put_page must be able to do the right thing when the
113 * page has been finished with, no matter what it is subsequently allocated
114 * for (because put_page is what is used here to drop an invalid speculative
117 * This is the interesting part of the lockless pagecache (and lockless
118 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
119 * has the following pattern:
120 * 1. find page in radix tree
121 * 2. conditionally increment refcount
122 * 3. check the page is still in pagecache (if no, goto 1)
124 * Remove-side that cares about stability of _refcount (eg. reclaim) has the
125 * following (with tree_lock held for write):
126 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
127 * B. remove page from pagecache
130 * There are 2 critical interleavings that matter:
131 * - 2 runs before A: in this case, A sees elevated refcount and bails out
132 * - A runs before 2: in this case, 2 sees zero refcount and retries;
133 * subsequently, B will complete and 1 will find no page, causing the
134 * lookup to return NULL.
136 * It is possible that between 1 and 2, the page is removed then the exact same
137 * page is inserted into the same position in pagecache. That's OK: the
138 * old find_get_page using tree_lock could equally have run before or after
139 * such a re-insertion, depending on order that locks are granted.
141 * Lookups racing against pagecache insertion isn't a big problem: either 1
142 * will find the page or it will not. Likewise, the old find_get_page could run
143 * either before the insertion or afterwards, depending on timing.
145 static inline int page_cache_get_speculative(struct page
*page
)
147 VM_BUG_ON(in_interrupt());
149 #ifdef CONFIG_TINY_RCU
150 # ifdef CONFIG_PREEMPT_COUNT
151 VM_BUG_ON(!in_atomic());
154 * Preempt must be disabled here - we rely on rcu_read_lock doing
157 * Pagecache won't be truncated from interrupt context, so if we have
158 * found a page in the radix tree here, we have pinned its refcount by
159 * disabling preempt, and hence no need for the "speculative get" that
162 VM_BUG_ON_PAGE(page_count(page
) == 0, page
);
166 if (unlikely(!get_page_unless_zero(page
))) {
168 * Either the page has been freed, or will be freed.
169 * In either case, retry here and the caller should
170 * do the right thing (see comments above).
175 VM_BUG_ON_PAGE(PageTail(page
), page
);
181 * Same as above, but add instead of inc (could just be merged)
183 static inline int page_cache_add_speculative(struct page
*page
, int count
)
185 VM_BUG_ON(in_interrupt());
187 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
188 # ifdef CONFIG_PREEMPT_COUNT
189 VM_BUG_ON(!in_atomic());
191 VM_BUG_ON_PAGE(page_count(page
) == 0, page
);
192 page_ref_add(page
, count
);
195 if (unlikely(!page_ref_add_unless(page
, count
, 0)))
198 VM_BUG_ON_PAGE(PageCompound(page
) && page
!= compound_head(page
), page
);
204 extern struct page
*__page_cache_alloc(gfp_t gfp
);
206 static inline struct page
*__page_cache_alloc(gfp_t gfp
)
208 return alloc_pages(gfp
, 0);
212 static inline struct page
*page_cache_alloc(struct address_space
*x
)
214 return __page_cache_alloc(mapping_gfp_mask(x
));
217 static inline struct page
*page_cache_alloc_cold(struct address_space
*x
)
219 return __page_cache_alloc(mapping_gfp_mask(x
)|__GFP_COLD
);
222 static inline gfp_t
readahead_gfp_mask(struct address_space
*x
)
224 return mapping_gfp_mask(x
) |
225 __GFP_COLD
| __GFP_NORETRY
| __GFP_NOWARN
;
228 typedef int filler_t(void *, struct page
*);
230 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
231 pgoff_t index
, unsigned long max_scan
);
232 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
233 pgoff_t index
, unsigned long max_scan
);
235 #define FGP_ACCESSED 0x00000001
236 #define FGP_LOCK 0x00000002
237 #define FGP_CREAT 0x00000004
238 #define FGP_WRITE 0x00000008
239 #define FGP_NOFS 0x00000010
240 #define FGP_NOWAIT 0x00000020
242 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
243 int fgp_flags
, gfp_t cache_gfp_mask
);
246 * find_get_page - find and get a page reference
247 * @mapping: the address_space to search
248 * @offset: the page index
250 * Looks up the page cache slot at @mapping & @offset. If there is a
251 * page cache page, it is returned with an increased refcount.
253 * Otherwise, %NULL is returned.
255 static inline struct page
*find_get_page(struct address_space
*mapping
,
258 return pagecache_get_page(mapping
, offset
, 0, 0);
261 static inline struct page
*find_get_page_flags(struct address_space
*mapping
,
262 pgoff_t offset
, int fgp_flags
)
264 return pagecache_get_page(mapping
, offset
, fgp_flags
, 0);
268 * find_lock_page - locate, pin and lock a pagecache page
269 * pagecache_get_page - find and get a page reference
270 * @mapping: the address_space to search
271 * @offset: the page index
273 * Looks up the page cache slot at @mapping & @offset. If there is a
274 * page cache page, it is returned locked and with an increased
277 * Otherwise, %NULL is returned.
279 * find_lock_page() may sleep.
281 static inline struct page
*find_lock_page(struct address_space
*mapping
,
284 return pagecache_get_page(mapping
, offset
, FGP_LOCK
, 0);
288 * find_or_create_page - locate or add a pagecache page
289 * @mapping: the page's address_space
290 * @index: the page's index into the mapping
291 * @gfp_mask: page allocation mode
293 * Looks up the page cache slot at @mapping & @offset. If there is a
294 * page cache page, it is returned locked and with an increased
297 * If the page is not present, a new page is allocated using @gfp_mask
298 * and added to the page cache and the VM's LRU list. The page is
299 * returned locked and with an increased refcount.
301 * On memory exhaustion, %NULL is returned.
303 * find_or_create_page() may sleep, even if @gfp_flags specifies an
306 static inline struct page
*find_or_create_page(struct address_space
*mapping
,
307 pgoff_t offset
, gfp_t gfp_mask
)
309 return pagecache_get_page(mapping
, offset
,
310 FGP_LOCK
|FGP_ACCESSED
|FGP_CREAT
,
315 * grab_cache_page_nowait - returns locked page at given index in given cache
316 * @mapping: target address_space
317 * @index: the page index
319 * Same as grab_cache_page(), but do not wait if the page is unavailable.
320 * This is intended for speculative data generators, where the data can
321 * be regenerated if the page couldn't be grabbed. This routine should
322 * be safe to call while holding the lock for another page.
324 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
325 * and deadlock against the caller's locked page.
327 static inline struct page
*grab_cache_page_nowait(struct address_space
*mapping
,
330 return pagecache_get_page(mapping
, index
,
331 FGP_LOCK
|FGP_CREAT
|FGP_NOFS
|FGP_NOWAIT
,
332 mapping_gfp_mask(mapping
));
335 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
);
336 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
);
337 unsigned find_get_entries(struct address_space
*mapping
, pgoff_t start
,
338 unsigned int nr_entries
, struct page
**entries
,
340 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
341 unsigned int nr_pages
, struct page
**pages
);
342 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t start
,
343 unsigned int nr_pages
, struct page
**pages
);
344 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
345 int tag
, unsigned int nr_pages
, struct page
**pages
);
346 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
347 int tag
, unsigned int nr_entries
,
348 struct page
**entries
, pgoff_t
*indices
);
350 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
351 pgoff_t index
, unsigned flags
);
354 * Returns locked page at given index in given cache, creating it if needed.
356 static inline struct page
*grab_cache_page(struct address_space
*mapping
,
359 return find_or_create_page(mapping
, index
, mapping_gfp_mask(mapping
));
362 extern struct page
* read_cache_page(struct address_space
*mapping
,
363 pgoff_t index
, filler_t
*filler
, void *data
);
364 extern struct page
* read_cache_page_gfp(struct address_space
*mapping
,
365 pgoff_t index
, gfp_t gfp_mask
);
366 extern int read_cache_pages(struct address_space
*mapping
,
367 struct list_head
*pages
, filler_t
*filler
, void *data
);
369 static inline struct page
*read_mapping_page(struct address_space
*mapping
,
370 pgoff_t index
, void *data
)
372 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
373 return read_cache_page(mapping
, index
, filler
, data
);
377 * Get the offset in PAGE_SIZE.
378 * (TODO: hugepage should have ->index in PAGE_SIZE)
380 static inline pgoff_t
page_to_pgoff(struct page
*page
)
384 if (unlikely(PageHeadHuge(page
)))
385 return page
->index
<< compound_order(page
);
387 if (likely(!PageTransTail(page
)))
391 * We don't initialize ->index for tail pages: calculate based on
394 pgoff
= compound_head(page
)->index
;
395 pgoff
+= page
- compound_head(page
);
400 * Return byte-offset into filesystem object for page.
402 static inline loff_t
page_offset(struct page
*page
)
404 return ((loff_t
)page
->index
) << PAGE_SHIFT
;
407 static inline loff_t
page_file_offset(struct page
*page
)
409 return ((loff_t
)page_index(page
)) << PAGE_SHIFT
;
412 extern pgoff_t
linear_hugepage_index(struct vm_area_struct
*vma
,
413 unsigned long address
);
415 static inline pgoff_t
linear_page_index(struct vm_area_struct
*vma
,
416 unsigned long address
)
419 if (unlikely(is_vm_hugetlb_page(vma
)))
420 return linear_hugepage_index(vma
, address
);
421 pgoff
= (address
- vma
->vm_start
) >> PAGE_SHIFT
;
422 pgoff
+= vma
->vm_pgoff
;
426 extern void __lock_page(struct page
*page
);
427 extern int __lock_page_killable(struct page
*page
);
428 extern int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
430 extern void unlock_page(struct page
*page
);
432 static inline int trylock_page(struct page
*page
)
434 page
= compound_head(page
);
435 return (likely(!test_and_set_bit_lock(PG_locked
, &page
->flags
)));
439 * lock_page may only be called if we have the page's inode pinned.
441 static inline void lock_page(struct page
*page
)
444 if (!trylock_page(page
))
449 * lock_page_killable is like lock_page but can be interrupted by fatal
450 * signals. It returns 0 if it locked the page and -EINTR if it was
451 * killed while waiting.
453 static inline int lock_page_killable(struct page
*page
)
456 if (!trylock_page(page
))
457 return __lock_page_killable(page
);
462 * lock_page_or_retry - Lock the page, unless this would block and the
463 * caller indicated that it can handle a retry.
465 * Return value and mmap_sem implications depend on flags; see
466 * __lock_page_or_retry().
468 static inline int lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
472 return trylock_page(page
) || __lock_page_or_retry(page
, mm
, flags
);
476 * This is exported only for wait_on_page_locked/wait_on_page_writeback,
477 * and for filesystems which need to wait on PG_private.
479 extern void wait_on_page_bit(struct page
*page
, int bit_nr
);
481 extern int wait_on_page_bit_killable(struct page
*page
, int bit_nr
);
482 extern int wait_on_page_bit_killable_timeout(struct page
*page
,
483 int bit_nr
, unsigned long timeout
);
485 static inline int wait_on_page_locked_killable(struct page
*page
)
487 if (!PageLocked(page
))
489 return wait_on_page_bit_killable(compound_head(page
), PG_locked
);
492 extern wait_queue_head_t
*page_waitqueue(struct page
*page
);
493 static inline void wake_up_page(struct page
*page
, int bit
)
495 __wake_up_bit(page_waitqueue(page
), &page
->flags
, bit
);
499 * Wait for a page to be unlocked.
501 * This must be called with the caller "holding" the page,
502 * ie with increased "page->count" so that the page won't
503 * go away during the wait..
505 static inline void wait_on_page_locked(struct page
*page
)
507 if (PageLocked(page
))
508 wait_on_page_bit(compound_head(page
), PG_locked
);
512 * Wait for a page to complete writeback
514 static inline void wait_on_page_writeback(struct page
*page
)
516 if (PageWriteback(page
))
517 wait_on_page_bit(page
, PG_writeback
);
520 extern void end_page_writeback(struct page
*page
);
521 void wait_for_stable_page(struct page
*page
);
523 void page_endio(struct page
*page
, bool is_write
, int err
);
526 * Add an arbitrary waiter to a page's wait queue
528 extern void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
);
531 * Fault everything in given userspace address range in.
533 static inline int fault_in_pages_writeable(char __user
*uaddr
, int size
)
535 char __user
*end
= uaddr
+ size
- 1;
537 if (unlikely(size
== 0))
540 if (unlikely(uaddr
> end
))
543 * Writing zeroes into userspace here is OK, because we know that if
544 * the zero gets there, we'll be overwriting it.
547 if (unlikely(__put_user(0, uaddr
) != 0))
550 } while (uaddr
<= end
);
552 /* Check whether the range spilled into the next page. */
553 if (((unsigned long)uaddr
& PAGE_MASK
) ==
554 ((unsigned long)end
& PAGE_MASK
))
555 return __put_user(0, end
);
560 static inline int fault_in_pages_readable(const char __user
*uaddr
, int size
)
563 const char __user
*end
= uaddr
+ size
- 1;
565 if (unlikely(size
== 0))
568 if (unlikely(uaddr
> end
))
572 if (unlikely(__get_user(c
, uaddr
) != 0))
575 } while (uaddr
<= end
);
577 /* Check whether the range spilled into the next page. */
578 if (((unsigned long)uaddr
& PAGE_MASK
) ==
579 ((unsigned long)end
& PAGE_MASK
)) {
580 return __get_user(c
, end
);
587 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
588 pgoff_t index
, gfp_t gfp_mask
);
589 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
590 pgoff_t index
, gfp_t gfp_mask
);
591 extern void delete_from_page_cache(struct page
*page
);
592 extern void __delete_from_page_cache(struct page
*page
, void *shadow
);
593 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
);
596 * Like add_to_page_cache_locked, but used to add newly allocated pages:
597 * the page is new, so we can just run __SetPageLocked() against it.
599 static inline int add_to_page_cache(struct page
*page
,
600 struct address_space
*mapping
, pgoff_t offset
, gfp_t gfp_mask
)
604 __SetPageLocked(page
);
605 error
= add_to_page_cache_locked(page
, mapping
, offset
, gfp_mask
);
607 __ClearPageLocked(page
);
611 static inline unsigned long dir_pages(struct inode
*inode
)
613 return (unsigned long)(inode
->i_size
+ PAGE_SIZE
- 1) >>
617 #endif /* _LINUX_PAGEMAP_H */