]>
git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - include/linux/pagemap.h
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 */
29 static inline void mapping_set_error(struct address_space
*mapping
, int error
)
31 if (unlikely(error
)) {
33 set_bit(AS_ENOSPC
, &mapping
->flags
);
35 set_bit(AS_EIO
, &mapping
->flags
);
39 static inline void mapping_set_unevictable(struct address_space
*mapping
)
41 set_bit(AS_UNEVICTABLE
, &mapping
->flags
);
44 static inline void mapping_clear_unevictable(struct address_space
*mapping
)
46 clear_bit(AS_UNEVICTABLE
, &mapping
->flags
);
49 static inline int mapping_unevictable(struct address_space
*mapping
)
52 return test_bit(AS_UNEVICTABLE
, &mapping
->flags
);
56 static inline gfp_t
mapping_gfp_mask(struct address_space
* mapping
)
58 return (__force gfp_t
)mapping
->flags
& __GFP_BITS_MASK
;
62 * This is non-atomic. Only to be used before the mapping is activated.
63 * Probably needs a barrier...
65 static inline void mapping_set_gfp_mask(struct address_space
*m
, gfp_t mask
)
67 m
->flags
= (m
->flags
& ~(__force
unsigned long)__GFP_BITS_MASK
) |
68 (__force
unsigned long)mask
;
72 * The page cache can done in larger chunks than
73 * one page, because it allows for more efficient
74 * throughput (it can then be mapped into user
75 * space in smaller chunks for same flexibility).
77 * Or rather, it _will_ be done in larger chunks.
79 #define PAGE_CACHE_SHIFT PAGE_SHIFT
80 #define PAGE_CACHE_SIZE PAGE_SIZE
81 #define PAGE_CACHE_MASK PAGE_MASK
82 #define PAGE_CACHE_ALIGN(addr) (((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
84 #define page_cache_get(page) get_page(page)
85 #define page_cache_release(page) put_page(page)
86 void release_pages(struct page
**pages
, int nr
, int cold
);
89 * speculatively take a reference to a page.
90 * If the page is free (_count == 0), then _count is untouched, and 0
91 * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
93 * This function must be called inside the same rcu_read_lock() section as has
94 * been used to lookup the page in the pagecache radix-tree (or page table):
95 * this allows allocators to use a synchronize_rcu() to stabilize _count.
97 * Unless an RCU grace period has passed, the count of all pages coming out
98 * of the allocator must be considered unstable. page_count may return higher
99 * than expected, and put_page must be able to do the right thing when the
100 * page has been finished with, no matter what it is subsequently allocated
101 * for (because put_page is what is used here to drop an invalid speculative
104 * This is the interesting part of the lockless pagecache (and lockless
105 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
106 * has the following pattern:
107 * 1. find page in radix tree
108 * 2. conditionally increment refcount
109 * 3. check the page is still in pagecache (if no, goto 1)
111 * Remove-side that cares about stability of _count (eg. reclaim) has the
112 * following (with tree_lock held for write):
113 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
114 * B. remove page from pagecache
117 * There are 2 critical interleavings that matter:
118 * - 2 runs before A: in this case, A sees elevated refcount and bails out
119 * - A runs before 2: in this case, 2 sees zero refcount and retries;
120 * subsequently, B will complete and 1 will find no page, causing the
121 * lookup to return NULL.
123 * It is possible that between 1 and 2, the page is removed then the exact same
124 * page is inserted into the same position in pagecache. That's OK: the
125 * old find_get_page using tree_lock could equally have run before or after
126 * such a re-insertion, depending on order that locks are granted.
128 * Lookups racing against pagecache insertion isn't a big problem: either 1
129 * will find the page or it will not. Likewise, the old find_get_page could run
130 * either before the insertion or afterwards, depending on timing.
132 static inline int page_cache_get_speculative(struct page
*page
)
134 VM_BUG_ON(in_interrupt());
136 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
137 # ifdef CONFIG_PREEMPT
138 VM_BUG_ON(!in_atomic());
141 * Preempt must be disabled here - we rely on rcu_read_lock doing
144 * Pagecache won't be truncated from interrupt context, so if we have
145 * found a page in the radix tree here, we have pinned its refcount by
146 * disabling preempt, and hence no need for the "speculative get" that
149 VM_BUG_ON(page_count(page
) == 0);
150 atomic_inc(&page
->_count
);
153 if (unlikely(!get_page_unless_zero(page
))) {
155 * Either the page has been freed, or will be freed.
156 * In either case, retry here and the caller should
157 * do the right thing (see comments above).
162 VM_BUG_ON(PageTail(page
));
168 * Same as above, but add instead of inc (could just be merged)
170 static inline int page_cache_add_speculative(struct page
*page
, int count
)
172 VM_BUG_ON(in_interrupt());
174 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
175 # ifdef CONFIG_PREEMPT
176 VM_BUG_ON(!in_atomic());
178 VM_BUG_ON(page_count(page
) == 0);
179 atomic_add(count
, &page
->_count
);
182 if (unlikely(!atomic_add_unless(&page
->_count
, count
, 0)))
185 VM_BUG_ON(PageCompound(page
) && page
!= compound_head(page
));
190 static inline int page_freeze_refs(struct page
*page
, int count
)
192 return likely(atomic_cmpxchg(&page
->_count
, count
, 0) == count
);
195 static inline void page_unfreeze_refs(struct page
*page
, int count
)
197 VM_BUG_ON(page_count(page
) != 0);
198 VM_BUG_ON(count
== 0);
200 atomic_set(&page
->_count
, count
);
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 typedef int filler_t(void *, struct page
*);
224 extern struct page
* find_get_page(struct address_space
*mapping
,
226 extern struct page
* find_lock_page(struct address_space
*mapping
,
228 extern struct page
* find_or_create_page(struct address_space
*mapping
,
229 pgoff_t index
, gfp_t gfp_mask
);
230 unsigned find_get_pages(struct address_space
*mapping
, pgoff_t start
,
231 unsigned int nr_pages
, struct page
**pages
);
232 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t start
,
233 unsigned int nr_pages
, struct page
**pages
);
234 unsigned find_get_pages_tag(struct address_space
*mapping
, pgoff_t
*index
,
235 int tag
, unsigned int nr_pages
, struct page
**pages
);
237 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
238 pgoff_t index
, unsigned flags
);
241 * Returns locked page at given index in given cache, creating it if needed.
243 static inline struct page
*grab_cache_page(struct address_space
*mapping
,
246 return find_or_create_page(mapping
, index
, mapping_gfp_mask(mapping
));
249 extern struct page
* grab_cache_page_nowait(struct address_space
*mapping
,
251 extern struct page
* read_cache_page_async(struct address_space
*mapping
,
252 pgoff_t index
, filler_t
*filler
,
254 extern struct page
* read_cache_page(struct address_space
*mapping
,
255 pgoff_t index
, filler_t
*filler
,
257 extern struct page
* read_cache_page_gfp(struct address_space
*mapping
,
258 pgoff_t index
, gfp_t gfp_mask
);
259 extern int read_cache_pages(struct address_space
*mapping
,
260 struct list_head
*pages
, filler_t
*filler
, void *data
);
262 static inline struct page
*read_mapping_page_async(
263 struct address_space
*mapping
,
264 pgoff_t index
, void *data
)
266 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
267 return read_cache_page_async(mapping
, index
, filler
, data
);
270 static inline struct page
*read_mapping_page(struct address_space
*mapping
,
271 pgoff_t index
, void *data
)
273 filler_t
*filler
= (filler_t
*)mapping
->a_ops
->readpage
;
274 return read_cache_page(mapping
, index
, filler
, data
);
278 * Return byte-offset into filesystem object for page.
280 static inline loff_t
page_offset(struct page
*page
)
282 return ((loff_t
)page
->index
) << PAGE_CACHE_SHIFT
;
285 extern pgoff_t
linear_hugepage_index(struct vm_area_struct
*vma
,
286 unsigned long address
);
288 static inline pgoff_t
linear_page_index(struct vm_area_struct
*vma
,
289 unsigned long address
)
292 if (unlikely(is_vm_hugetlb_page(vma
)))
293 return linear_hugepage_index(vma
, address
);
294 pgoff
= (address
- vma
->vm_start
) >> PAGE_SHIFT
;
295 pgoff
+= vma
->vm_pgoff
;
296 return pgoff
>> (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
299 extern void __lock_page(struct page
*page
);
300 extern int __lock_page_killable(struct page
*page
);
301 extern int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
303 extern void unlock_page(struct page
*page
);
305 static inline void __set_page_locked(struct page
*page
)
307 __set_bit(PG_locked
, &page
->flags
);
310 static inline void __clear_page_locked(struct page
*page
)
312 __clear_bit(PG_locked
, &page
->flags
);
315 static inline int trylock_page(struct page
*page
)
317 return (likely(!test_and_set_bit_lock(PG_locked
, &page
->flags
)));
321 * lock_page may only be called if we have the page's inode pinned.
323 static inline void lock_page(struct page
*page
)
326 if (!trylock_page(page
))
331 * lock_page_killable is like lock_page but can be interrupted by fatal
332 * signals. It returns 0 if it locked the page and -EINTR if it was
333 * killed while waiting.
335 static inline int lock_page_killable(struct page
*page
)
338 if (!trylock_page(page
))
339 return __lock_page_killable(page
);
344 * lock_page_or_retry - Lock the page, unless this would block and the
345 * caller indicated that it can handle a retry.
347 static inline int lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
351 return trylock_page(page
) || __lock_page_or_retry(page
, mm
, flags
);
355 * This is exported only for wait_on_page_locked/wait_on_page_writeback.
356 * Never use this directly!
358 extern void wait_on_page_bit(struct page
*page
, int bit_nr
);
361 * Wait for a page to be unlocked.
363 * This must be called with the caller "holding" the page,
364 * ie with increased "page->count" so that the page won't
365 * go away during the wait..
367 static inline void wait_on_page_locked(struct page
*page
)
369 if (PageLocked(page
))
370 wait_on_page_bit(page
, PG_locked
);
374 * Wait for a page to complete writeback
376 static inline void wait_on_page_writeback(struct page
*page
)
378 if (PageWriteback(page
))
379 wait_on_page_bit(page
, PG_writeback
);
382 extern void end_page_writeback(struct page
*page
);
385 * Add an arbitrary waiter to a page's wait queue
387 extern void add_page_wait_queue(struct page
*page
, wait_queue_t
*waiter
);
390 * Fault a userspace page into pagetables. Return non-zero on a fault.
392 * This assumes that two userspace pages are always sufficient. That's
393 * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
395 static inline int fault_in_pages_writeable(char __user
*uaddr
, int size
)
399 if (unlikely(size
== 0))
403 * Writing zeroes into userspace here is OK, because we know that if
404 * the zero gets there, we'll be overwriting it.
406 ret
= __put_user(0, uaddr
);
408 char __user
*end
= uaddr
+ size
- 1;
411 * If the page was already mapped, this will get a cache miss
412 * for sure, so try to avoid doing it.
414 if (((unsigned long)uaddr
& PAGE_MASK
) !=
415 ((unsigned long)end
& PAGE_MASK
))
416 ret
= __put_user(0, end
);
421 static inline int fault_in_pages_readable(const char __user
*uaddr
, int size
)
426 if (unlikely(size
== 0))
429 ret
= __get_user(c
, uaddr
);
431 const char __user
*end
= uaddr
+ size
- 1;
433 if (((unsigned long)uaddr
& PAGE_MASK
) !=
434 ((unsigned long)end
& PAGE_MASK
)) {
435 ret
= __get_user(c
, end
);
442 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
443 pgoff_t index
, gfp_t gfp_mask
);
444 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
445 pgoff_t index
, gfp_t gfp_mask
);
446 extern void delete_from_page_cache(struct page
*page
);
447 extern void __delete_from_page_cache(struct page
*page
);
448 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
);
451 * Like add_to_page_cache_locked, but used to add newly allocated pages:
452 * the page is new, so we can just run __set_page_locked() against it.
454 static inline int add_to_page_cache(struct page
*page
,
455 struct address_space
*mapping
, pgoff_t offset
, gfp_t gfp_mask
)
459 __set_page_locked(page
);
460 error
= add_to_page_cache_locked(page
, mapping
, offset
, gfp_mask
);
462 __clear_page_locked(page
);
466 #endif /* _LINUX_PAGEMAP_H */