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 <linux/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,
32 * mapping_set_error - record a writeback error in the address_space
33 * @mapping - the mapping in which an error should be set
34 * @error - the error to set in the mapping
36 * When writeback fails in some way, we must record that error so that
37 * userspace can be informed when fsync and the like are called. We endeavor
38 * to report errors on any file that was open at the time of the error. Some
39 * internal callers also need to know when writeback errors have occurred.
41 * When a writeback error occurs, most filesystems will want to call
42 * mapping_set_error to record the error in the mapping so that it can be
43 * reported when the application calls fsync(2).
45 static inline void mapping_set_error(struct address_space
*mapping
, int error
)
50 /* Record in wb_err for checkers using errseq_t based tracking */
51 filemap_set_wb_err(mapping
, error
);
53 /* Record it in flags for now, for legacy callers */
55 set_bit(AS_ENOSPC
, &mapping
->flags
);
57 set_bit(AS_EIO
, &mapping
->flags
);
60 static inline void mapping_set_unevictable(struct address_space
*mapping
)
62 set_bit(AS_UNEVICTABLE
, &mapping
->flags
);
65 static inline void mapping_clear_unevictable(struct address_space
*mapping
)
67 clear_bit(AS_UNEVICTABLE
, &mapping
->flags
);
70 static inline int mapping_unevictable(struct address_space
*mapping
)
73 return test_bit(AS_UNEVICTABLE
, &mapping
->flags
);
77 static inline void mapping_set_exiting(struct address_space
*mapping
)
79 set_bit(AS_EXITING
, &mapping
->flags
);
82 static inline int mapping_exiting(struct address_space
*mapping
)
84 return test_bit(AS_EXITING
, &mapping
->flags
);
87 static inline void mapping_set_no_writeback_tags(struct address_space
*mapping
)
89 set_bit(AS_NO_WRITEBACK_TAGS
, &mapping
->flags
);
92 static inline int mapping_use_writeback_tags(struct address_space
*mapping
)
94 return !test_bit(AS_NO_WRITEBACK_TAGS
, &mapping
->flags
);
97 static inline gfp_t
mapping_gfp_mask(struct address_space
* mapping
)
99 return mapping
->gfp_mask
;
102 /* Restricts the given gfp_mask to what the mapping allows. */
103 static inline gfp_t
mapping_gfp_constraint(struct address_space
*mapping
,
106 return mapping_gfp_mask(mapping
) & gfp_mask
;
110 * This is non-atomic. Only to be used before the mapping is activated.
111 * Probably needs a barrier...
113 static inline void mapping_set_gfp_mask(struct address_space
*m
, gfp_t mask
)
118 void release_pages(struct page
**pages
, int nr
, bool cold
);
121 * speculatively take a reference to a page.
122 * If the page is free (_refcount == 0), then _refcount is untouched, and 0
123 * is returned. Otherwise, _refcount is incremented by 1 and 1 is returned.
125 * This function must be called inside the same rcu_read_lock() section as has
126 * been used to lookup the page in the pagecache radix-tree (or page table):
127 * this allows allocators to use a synchronize_rcu() to stabilize _refcount.
129 * Unless an RCU grace period has passed, the count of all pages coming out
130 * of the allocator must be considered unstable. page_count may return higher
131 * than expected, and put_page must be able to do the right thing when the
132 * page has been finished with, no matter what it is subsequently allocated
133 * for (because put_page is what is used here to drop an invalid speculative
136 * This is the interesting part of the lockless pagecache (and lockless
137 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
138 * has the following pattern:
139 * 1. find page in radix tree
140 * 2. conditionally increment refcount
141 * 3. check the page is still in pagecache (if no, goto 1)
143 * Remove-side that cares about stability of _refcount (eg. reclaim) has the
144 * following (with tree_lock held for write):
145 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
146 * B. remove page from pagecache
149 * There are 2 critical interleavings that matter:
150 * - 2 runs before A: in this case, A sees elevated refcount and bails out
151 * - A runs before 2: in this case, 2 sees zero refcount and retries;
152 * subsequently, B will complete and 1 will find no page, causing the
153 * lookup to return NULL.
155 * It is possible that between 1 and 2, the page is removed then the exact same
156 * page is inserted into the same position in pagecache. That's OK: the
157 * old find_get_page using tree_lock could equally have run before or after
158 * such a re-insertion, depending on order that locks are granted.
160 * Lookups racing against pagecache insertion isn't a big problem: either 1
161 * will find the page or it will not. Likewise, the old find_get_page could run
162 * either before the insertion or afterwards, depending on timing.
164 static inline int page_cache_get_speculative(struct page
*page
)
166 VM_BUG_ON(in_interrupt());
168 #ifdef CONFIG_TINY_RCU
169 # ifdef CONFIG_PREEMPT_COUNT
170 VM_BUG_ON(!in_atomic() && !irqs_disabled());
173 * Preempt must be disabled here - we rely on rcu_read_lock doing
176 * Pagecache won't be truncated from interrupt context, so if we have
177 * found a page in the radix tree here, we have pinned its refcount by
178 * disabling preempt, and hence no need for the "speculative get" that
181 VM_BUG_ON_PAGE(page_count(page
) == 0, page
);
185 if (unlikely(!get_page_unless_zero(page
))) {
187 * Either the page has been freed, or will be freed.
188 * In either case, retry here and the caller should
189 * do the right thing (see comments above).
194 VM_BUG_ON_PAGE(PageTail(page
), page
);
200 * Same as above, but add instead of inc (could just be merged)
202 static inline int page_cache_add_speculative(struct page
*page
, int count
)
204 VM_BUG_ON(in_interrupt());
206 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
207 # ifdef CONFIG_PREEMPT_COUNT
208 VM_BUG_ON(!in_atomic() && !irqs_disabled());
210 VM_BUG_ON_PAGE(page_count(page
) == 0, page
);
211 page_ref_add(page
, count
);
214 if (unlikely(!page_ref_add_unless(page
, count
, 0)))
217 VM_BUG_ON_PAGE(PageCompound(page
) && page
!= compound_head(page
), page
);
223 extern struct page
*__page_cache_alloc(gfp_t gfp
);
225 static inline struct page
*__page_cache_alloc(gfp_t gfp
)
227 return alloc_pages(gfp
, 0);
231 static inline struct page
*page_cache_alloc(struct address_space
*x
)
233 return __page_cache_alloc(mapping_gfp_mask(x
));
236 static inline struct page
*page_cache_alloc_cold(struct address_space
*x
)
238 return __page_cache_alloc(mapping_gfp_mask(x
)|__GFP_COLD
);
241 static inline gfp_t
readahead_gfp_mask(struct address_space
*x
)
243 return mapping_gfp_mask(x
) |
244 __GFP_COLD
| __GFP_NORETRY
| __GFP_NOWARN
;
247 typedef int filler_t(void *, struct page
*);
249 pgoff_t
page_cache_next_hole(struct address_space
*mapping
,
250 pgoff_t index
, unsigned long max_scan
);
251 pgoff_t
page_cache_prev_hole(struct address_space
*mapping
,
252 pgoff_t index
, unsigned long max_scan
);
254 #define FGP_ACCESSED 0x00000001
255 #define FGP_LOCK 0x00000002
256 #define FGP_CREAT 0x00000004
257 #define FGP_WRITE 0x00000008
258 #define FGP_NOFS 0x00000010
259 #define FGP_NOWAIT 0x00000020
261 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t offset
,
262 int fgp_flags
, gfp_t cache_gfp_mask
);
265 * find_get_page - find and get a page reference
266 * @mapping: the address_space to search
267 * @offset: the page index
269 * Looks up the page cache slot at @mapping & @offset. If there is a
270 * page cache page, it is returned with an increased refcount.
272 * Otherwise, %NULL is returned.
274 static inline struct page
*find_get_page(struct address_space
*mapping
,
277 return pagecache_get_page(mapping
, offset
, 0, 0);
280 static inline struct page
*find_get_page_flags(struct address_space
*mapping
,
281 pgoff_t offset
, int fgp_flags
)
283 return pagecache_get_page(mapping
, offset
, fgp_flags
, 0);
287 * find_lock_page - locate, pin and lock a pagecache page
288 * @mapping: the address_space to search
289 * @offset: the page index
291 * Looks up the page cache slot at @mapping & @offset. If there is a
292 * page cache page, it is returned locked and with an increased
295 * Otherwise, %NULL is returned.
297 * find_lock_page() may sleep.
299 static inline struct page
*find_lock_page(struct address_space
*mapping
,
302 return pagecache_get_page(mapping
, offset
, FGP_LOCK
, 0);
306 * find_or_create_page - locate or add a pagecache page
307 * @mapping: the page's address_space
308 * @index: the page's index into the mapping
309 * @gfp_mask: page allocation mode
311 * Looks up the page cache slot at @mapping & @offset. If there is a
312 * page cache page, it is returned locked and with an increased
315 * If the page is not present, a new page is allocated using @gfp_mask
316 * and added to the page cache and the VM's LRU list. The page is
317 * returned locked and with an increased refcount.
319 * On memory exhaustion, %NULL is returned.
321 * find_or_create_page() may sleep, even if @gfp_flags specifies an
324 static inline struct page
*find_or_create_page(struct address_space
*mapping
,
325 pgoff_t offset
, gfp_t gfp_mask
)
327 return pagecache_get_page(mapping
, offset
,
328 FGP_LOCK
|FGP_ACCESSED
|FGP_CREAT
,
333 * grab_cache_page_nowait - returns locked page at given index in given cache
334 * @mapping: target address_space
335 * @index: the page index
337 * Same as grab_cache_page(), but do not wait if the page is unavailable.
338 * This is intended for speculative data generators, where the data can
339 * be regenerated if the page couldn't be grabbed. This routine should
340 * be safe to call while holding the lock for another page.
342 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
343 * and deadlock against the caller's locked page.
345 static inline struct page
*grab_cache_page_nowait(struct address_space
*mapping
,
348 return pagecache_get_page(mapping
, index
,
349 FGP_LOCK
|FGP_CREAT
|FGP_NOFS
|FGP_NOWAIT
,
350 mapping_gfp_mask(mapping
));
353 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
);
354 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
);
355 unsigned find_get_entries(struct address_space
*mapping
, pgoff_t start
,
356 unsigned int nr_entries
, struct page
**entries
,
358 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
359 unsigned int nr_pages
, struct page
**pages
);
360 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t start
,
361 unsigned int nr_pages
, struct page
**pages
);
362 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
363 int tag
, unsigned int nr_pages
, struct page
**pages
);
364 unsigned find_get_entries_tag(struct address_space
*mapping
, pgoff_t start
,
365 int tag
, unsigned int nr_entries
,
366 struct page
**entries
, pgoff_t
*indices
);
368 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
369 pgoff_t index
, unsigned flags
);
372 * Returns locked page at given index in given cache, creating it if needed.
374 static inline struct page
*grab_cache_page(struct address_space
*mapping
,
377 return find_or_create_page(mapping
, index
, mapping_gfp_mask(mapping
));
380 extern struct page
* read_cache_page(struct address_space
*mapping
,
381 pgoff_t index
, filler_t
*filler
, void *data
);
382 extern struct page
* read_cache_page_gfp(struct address_space
*mapping
,
383 pgoff_t index
, gfp_t gfp_mask
);
384 extern int read_cache_pages(struct address_space
*mapping
,
385 struct list_head
*pages
, filler_t
*filler
, void *data
);
387 static inline struct page
*read_mapping_page(struct address_space
*mapping
,
388 pgoff_t index
, void *data
)
390 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
391 return read_cache_page(mapping
, index
, filler
, data
);
395 * Get index of the page with in radix-tree
396 * (TODO: remove once hugetlb pages will have ->index in PAGE_SIZE)
398 static inline pgoff_t
page_to_index(struct page
*page
)
402 if (likely(!PageTransTail(page
)))
406 * We don't initialize ->index for tail pages: calculate based on
409 pgoff
= compound_head(page
)->index
;
410 pgoff
+= page
- compound_head(page
);
415 * Get the offset in PAGE_SIZE.
416 * (TODO: hugepage should have ->index in PAGE_SIZE)
418 static inline pgoff_t
page_to_pgoff(struct page
*page
)
420 if (unlikely(PageHeadHuge(page
)))
421 return page
->index
<< compound_order(page
);
423 return page_to_index(page
);
427 * Return byte-offset into filesystem object for page.
429 static inline loff_t
page_offset(struct page
*page
)
431 return ((loff_t
)page
->index
) << PAGE_SHIFT
;
434 static inline loff_t
page_file_offset(struct page
*page
)
436 return ((loff_t
)page_index(page
)) << PAGE_SHIFT
;
439 extern pgoff_t
linear_hugepage_index(struct vm_area_struct
*vma
,
440 unsigned long address
);
442 static inline pgoff_t
linear_page_index(struct vm_area_struct
*vma
,
443 unsigned long address
)
446 if (unlikely(is_vm_hugetlb_page(vma
)))
447 return linear_hugepage_index(vma
, address
);
448 pgoff
= (address
- vma
->vm_start
) >> PAGE_SHIFT
;
449 pgoff
+= vma
->vm_pgoff
;
453 extern void __lock_page(struct page
*page
);
454 extern int __lock_page_killable(struct page
*page
);
455 extern int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
457 extern void unlock_page(struct page
*page
);
459 static inline int trylock_page(struct page
*page
)
461 page
= compound_head(page
);
462 return (likely(!test_and_set_bit_lock(PG_locked
, &page
->flags
)));
466 * lock_page may only be called if we have the page's inode pinned.
468 static inline void lock_page(struct page
*page
)
471 if (!trylock_page(page
))
476 * lock_page_killable is like lock_page but can be interrupted by fatal
477 * signals. It returns 0 if it locked the page and -EINTR if it was
478 * killed while waiting.
480 static inline int lock_page_killable(struct page
*page
)
483 if (!trylock_page(page
))
484 return __lock_page_killable(page
);
489 * lock_page_or_retry - Lock the page, unless this would block and the
490 * caller indicated that it can handle a retry.
492 * Return value and mmap_sem implications depend on flags; see
493 * __lock_page_or_retry().
495 static inline int lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
499 return trylock_page(page
) || __lock_page_or_retry(page
, mm
, flags
);
503 * This is exported only for wait_on_page_locked/wait_on_page_writeback, etc.,
504 * and should not be used directly.
506 extern void wait_on_page_bit(struct page
*page
, int bit_nr
);
507 extern int wait_on_page_bit_killable(struct page
*page
, int bit_nr
);
510 * Wait for a page to be unlocked.
512 * This must be called with the caller "holding" the page,
513 * ie with increased "page->count" so that the page won't
514 * go away during the wait..
516 static inline void wait_on_page_locked(struct page
*page
)
518 if (PageLocked(page
))
519 wait_on_page_bit(compound_head(page
), PG_locked
);
522 static inline int wait_on_page_locked_killable(struct page
*page
)
524 if (!PageLocked(page
))
526 return wait_on_page_bit_killable(compound_head(page
), PG_locked
);
530 * Wait for a page to complete writeback
532 static inline void wait_on_page_writeback(struct page
*page
)
534 if (PageWriteback(page
))
535 wait_on_page_bit(page
, PG_writeback
);
538 extern void end_page_writeback(struct page
*page
);
539 void wait_for_stable_page(struct page
*page
);
541 void page_endio(struct page
*page
, bool is_write
, int err
);
544 * Add an arbitrary waiter to a page's wait queue
546 extern void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
);
549 * Fault everything in given userspace address range in.
551 static inline int fault_in_pages_writeable(char __user
*uaddr
, int size
)
553 char __user
*end
= uaddr
+ size
- 1;
555 if (unlikely(size
== 0))
558 if (unlikely(uaddr
> end
))
561 * Writing zeroes into userspace here is OK, because we know that if
562 * the zero gets there, we'll be overwriting it.
565 if (unlikely(__put_user(0, uaddr
) != 0))
568 } while (uaddr
<= end
);
570 /* Check whether the range spilled into the next page. */
571 if (((unsigned long)uaddr
& PAGE_MASK
) ==
572 ((unsigned long)end
& PAGE_MASK
))
573 return __put_user(0, end
);
578 static inline int fault_in_pages_readable(const char __user
*uaddr
, int size
)
581 const char __user
*end
= uaddr
+ size
- 1;
583 if (unlikely(size
== 0))
586 if (unlikely(uaddr
> end
))
590 if (unlikely(__get_user(c
, uaddr
) != 0))
593 } while (uaddr
<= end
);
595 /* Check whether the range spilled into the next page. */
596 if (((unsigned long)uaddr
& PAGE_MASK
) ==
597 ((unsigned long)end
& PAGE_MASK
)) {
598 return __get_user(c
, end
);
605 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
606 pgoff_t index
, gfp_t gfp_mask
);
607 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
608 pgoff_t index
, gfp_t gfp_mask
);
609 extern void delete_from_page_cache(struct page
*page
);
610 extern void __delete_from_page_cache(struct page
*page
, void *shadow
);
611 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
);
614 * Like add_to_page_cache_locked, but used to add newly allocated pages:
615 * the page is new, so we can just run __SetPageLocked() against it.
617 static inline int add_to_page_cache(struct page
*page
,
618 struct address_space
*mapping
, pgoff_t offset
, gfp_t gfp_mask
)
622 __SetPageLocked(page
);
623 error
= add_to_page_cache_locked(page
, mapping
, offset
, gfp_mask
);
625 __ClearPageLocked(page
);
629 static inline unsigned long dir_pages(struct inode
*inode
)
631 return (unsigned long)(inode
->i_size
+ PAGE_SIZE
- 1) >>
635 #endif /* _LINUX_PAGEMAP_H */