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() */
18 * Bits in mapping->flags. The lower __GFP_BITS_SHIFT bits are the page
19 * allocation mode flags.
22 AS_EIO = __GFP_BITS_SHIFT + 0, /* IO error on async write */
23 AS_ENOSPC = __GFP_BITS_SHIFT + 1, /* ENOSPC on async write */
24 AS_MM_ALL_LOCKS = __GFP_BITS_SHIFT + 2, /* under mm_take_all_locks() */
25 AS_UNEVICTABLE = __GFP_BITS_SHIFT + 3, /* e.g., ramdisk, SHM_LOCK */
28 static inline void mapping_set_error(struct address_space *mapping, int error)
30 if (unlikely(error)) {
32 set_bit(AS_ENOSPC, &mapping->flags);
34 set_bit(AS_EIO, &mapping->flags);
38 static inline void mapping_set_unevictable(struct address_space *mapping)
40 set_bit(AS_UNEVICTABLE, &mapping->flags);
43 static inline void mapping_clear_unevictable(struct address_space *mapping)
45 clear_bit(AS_UNEVICTABLE, &mapping->flags);
48 static inline int mapping_unevictable(struct address_space *mapping)
51 return test_bit(AS_UNEVICTABLE, &mapping->flags);
55 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
57 return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
61 * This is non-atomic. Only to be used before the mapping is activated.
62 * Probably needs a barrier...
64 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
66 m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
67 (__force unsigned long)mask;
71 * The page cache can done in larger chunks than
72 * one page, because it allows for more efficient
73 * throughput (it can then be mapped into user
74 * space in smaller chunks for same flexibility).
76 * Or rather, it _will_ be done in larger chunks.
78 #define PAGE_CACHE_SHIFT PAGE_SHIFT
79 #define PAGE_CACHE_SIZE PAGE_SIZE
80 #define PAGE_CACHE_MASK PAGE_MASK
81 #define PAGE_CACHE_ALIGN(addr) (((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
83 #define page_cache_get(page) get_page(page)
84 #define page_cache_release(page) put_page(page)
85 void release_pages(struct page **pages, int nr, int cold);
88 * speculatively take a reference to a page.
89 * If the page is free (_count == 0), then _count is untouched, and 0
90 * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
92 * This function must be called inside the same rcu_read_lock() section as has
93 * been used to lookup the page in the pagecache radix-tree (or page table):
94 * this allows allocators to use a synchronize_rcu() to stabilize _count.
96 * Unless an RCU grace period has passed, the count of all pages coming out
97 * of the allocator must be considered unstable. page_count may return higher
98 * than expected, and put_page must be able to do the right thing when the
99 * page has been finished with, no matter what it is subsequently allocated
100 * for (because put_page is what is used here to drop an invalid speculative
103 * This is the interesting part of the lockless pagecache (and lockless
104 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
105 * has the following pattern:
106 * 1. find page in radix tree
107 * 2. conditionally increment refcount
108 * 3. check the page is still in pagecache (if no, goto 1)
110 * Remove-side that cares about stability of _count (eg. reclaim) has the
111 * following (with tree_lock held for write):
112 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
113 * B. remove page from pagecache
116 * There are 2 critical interleavings that matter:
117 * - 2 runs before A: in this case, A sees elevated refcount and bails out
118 * - A runs before 2: in this case, 2 sees zero refcount and retries;
119 * subsequently, B will complete and 1 will find no page, causing the
120 * lookup to return NULL.
122 * It is possible that between 1 and 2, the page is removed then the exact same
123 * page is inserted into the same position in pagecache. That's OK: the
124 * old find_get_page using tree_lock could equally have run before or after
125 * such a re-insertion, depending on order that locks are granted.
127 * Lookups racing against pagecache insertion isn't a big problem: either 1
128 * will find the page or it will not. Likewise, the old find_get_page could run
129 * either before the insertion or afterwards, depending on timing.
131 static inline int page_cache_get_speculative(struct page *page)
133 VM_BUG_ON(in_interrupt());
135 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
136 # ifdef CONFIG_PREEMPT
137 VM_BUG_ON(!in_atomic());
140 * Preempt must be disabled here - we rely on rcu_read_lock doing
143 * Pagecache won't be truncated from interrupt context, so if we have
144 * found a page in the radix tree here, we have pinned its refcount by
145 * disabling preempt, and hence no need for the "speculative get" that
148 VM_BUG_ON(page_count(page) == 0);
149 atomic_inc(&page->_count);
152 if (unlikely(!get_page_unless_zero(page))) {
154 * Either the page has been freed, or will be freed.
155 * In either case, retry here and the caller should
156 * do the right thing (see comments above).
161 VM_BUG_ON(PageTail(page));
167 * Same as above, but add instead of inc (could just be merged)
169 static inline int page_cache_add_speculative(struct page *page, int count)
171 VM_BUG_ON(in_interrupt());
173 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
174 # ifdef CONFIG_PREEMPT
175 VM_BUG_ON(!in_atomic());
177 VM_BUG_ON(page_count(page) == 0);
178 atomic_add(count, &page->_count);
181 if (unlikely(!atomic_add_unless(&page->_count, count, 0)))
184 VM_BUG_ON(PageCompound(page) && page != compound_head(page));
189 static inline int page_freeze_refs(struct page *page, int count)
191 return likely(atomic_cmpxchg(&page->_count, count, 0) == count);
194 static inline void page_unfreeze_refs(struct page *page, int count)
196 VM_BUG_ON(page_count(page) != 0);
197 VM_BUG_ON(count == 0);
199 atomic_set(&page->_count, count);
203 extern struct page *__page_cache_alloc(gfp_t gfp);
205 static inline struct page *__page_cache_alloc(gfp_t gfp)
207 return alloc_pages(gfp, 0);
211 static inline struct page *page_cache_alloc(struct address_space *x)
213 return __page_cache_alloc(mapping_gfp_mask(x));
216 static inline struct page *page_cache_alloc_cold(struct address_space *x)
218 return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
221 typedef int filler_t(void *, struct page *);
223 extern struct page * find_get_page(struct address_space *mapping,
225 extern struct page * find_lock_page(struct address_space *mapping,
227 extern struct page * find_or_create_page(struct address_space *mapping,
228 pgoff_t index, gfp_t gfp_mask);
229 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
230 unsigned int nr_pages, struct page **pages);
231 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
232 unsigned int nr_pages, struct page **pages);
233 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
234 int tag, unsigned int nr_pages, struct page **pages);
236 struct page *grab_cache_page_write_begin(struct address_space *mapping,
237 pgoff_t index, unsigned flags);
240 * Returns locked page at given index in given cache, creating it if needed.
242 static inline struct page *grab_cache_page(struct address_space *mapping,
245 return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
248 extern struct page * grab_cache_page_nowait(struct address_space *mapping,
250 extern struct page * read_cache_page_async(struct address_space *mapping,
251 pgoff_t index, filler_t *filler,
253 extern struct page * read_cache_page(struct address_space *mapping,
254 pgoff_t index, filler_t *filler,
256 extern struct page * read_cache_page_gfp(struct address_space *mapping,
257 pgoff_t index, gfp_t gfp_mask);
258 extern int read_cache_pages(struct address_space *mapping,
259 struct list_head *pages, filler_t *filler, void *data);
261 static inline struct page *read_mapping_page_async(
262 struct address_space *mapping,
263 pgoff_t index, void *data)
265 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
266 return read_cache_page_async(mapping, index, filler, data);
269 static inline struct page *read_mapping_page(struct address_space *mapping,
270 pgoff_t index, void *data)
272 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
273 return read_cache_page(mapping, index, filler, data);
277 * Return byte-offset into filesystem object for page.
279 static inline loff_t page_offset(struct page *page)
281 return ((loff_t)page->index) << PAGE_CACHE_SHIFT;
284 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
285 unsigned long address)
287 pgoff_t pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
288 pgoff += vma->vm_pgoff;
289 return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT);
292 extern void __lock_page(struct page *page);
293 extern int __lock_page_killable(struct page *page);
294 extern void __lock_page_nosync(struct page *page);
295 extern void unlock_page(struct page *page);
297 static inline void __set_page_locked(struct page *page)
299 __set_bit(PG_locked, &page->flags);
302 static inline void __clear_page_locked(struct page *page)
304 __clear_bit(PG_locked, &page->flags);
307 static inline int trylock_page(struct page *page)
309 return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
313 * lock_page may only be called if we have the page's inode pinned.
315 static inline void lock_page(struct page *page)
318 if (!trylock_page(page))
323 * lock_page_killable is like lock_page but can be interrupted by fatal
324 * signals. It returns 0 if it locked the page and -EINTR if it was
325 * killed while waiting.
327 static inline int lock_page_killable(struct page *page)
330 if (!trylock_page(page))
331 return __lock_page_killable(page);
336 * lock_page_nosync should only be used if we can't pin the page's inode.
337 * Doesn't play quite so well with block device plugging.
339 static inline void lock_page_nosync(struct page *page)
342 if (!trylock_page(page))
343 __lock_page_nosync(page);
347 * This is exported only for wait_on_page_locked/wait_on_page_writeback.
348 * Never use this directly!
350 extern void wait_on_page_bit(struct page *page, int bit_nr);
353 * Wait for a page to be unlocked.
355 * This must be called with the caller "holding" the page,
356 * ie with increased "page->count" so that the page won't
357 * go away during the wait..
359 static inline void wait_on_page_locked(struct page *page)
361 if (PageLocked(page))
362 wait_on_page_bit(page, PG_locked);
366 * Wait for a page to complete writeback
368 static inline void wait_on_page_writeback(struct page *page)
370 if (PageWriteback(page))
371 wait_on_page_bit(page, PG_writeback);
374 extern void end_page_writeback(struct page *page);
377 * Add an arbitrary waiter to a page's wait queue
379 extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter);
382 * Fault a userspace page into pagetables. Return non-zero on a fault.
384 * This assumes that two userspace pages are always sufficient. That's
385 * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
387 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
391 if (unlikely(size == 0))
395 * Writing zeroes into userspace here is OK, because we know that if
396 * the zero gets there, we'll be overwriting it.
398 ret = __put_user(0, uaddr);
400 char __user *end = uaddr + size - 1;
403 * If the page was already mapped, this will get a cache miss
404 * for sure, so try to avoid doing it.
406 if (((unsigned long)uaddr & PAGE_MASK) !=
407 ((unsigned long)end & PAGE_MASK))
408 ret = __put_user(0, end);
413 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
418 if (unlikely(size == 0))
421 ret = __get_user(c, uaddr);
423 const char __user *end = uaddr + size - 1;
425 if (((unsigned long)uaddr & PAGE_MASK) !=
426 ((unsigned long)end & PAGE_MASK))
427 ret = __get_user(c, end);
432 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
433 pgoff_t index, gfp_t gfp_mask);
434 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
435 pgoff_t index, gfp_t gfp_mask);
436 extern void remove_from_page_cache(struct page *page);
437 extern void __remove_from_page_cache(struct page *page);
440 * Like add_to_page_cache_locked, but used to add newly allocated pages:
441 * the page is new, so we can just run __set_page_locked() against it.
443 static inline int add_to_page_cache(struct page *page,
444 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
448 __set_page_locked(page);
449 error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
451 __clear_page_locked(page);
455 #endif /* _LINUX_PAGEMAP_H */