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_BALLOON_MAP = __GFP_BITS_SHIFT + 4, /* balloon page special map */
28 AS_EXITING = __GFP_BITS_SHIFT + 5, /* final truncate in progress */
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_balloon(struct address_space *mapping)
60 set_bit(AS_BALLOON_MAP, &mapping->flags);
63 static inline void mapping_clear_balloon(struct address_space *mapping)
65 clear_bit(AS_BALLOON_MAP, &mapping->flags);
68 static inline int mapping_balloon(struct address_space *mapping)
70 return mapping && test_bit(AS_BALLOON_MAP, &mapping->flags);
73 static inline void mapping_set_exiting(struct address_space *mapping)
75 set_bit(AS_EXITING, &mapping->flags);
78 static inline int mapping_exiting(struct address_space *mapping)
80 return test_bit(AS_EXITING, &mapping->flags);
83 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
85 return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
89 * This is non-atomic. Only to be used before the mapping is activated.
90 * Probably needs a barrier...
92 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
94 m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
95 (__force unsigned long)mask;
99 * The page cache can done in larger chunks than
100 * one page, because it allows for more efficient
101 * throughput (it can then be mapped into user
102 * space in smaller chunks for same flexibility).
104 * Or rather, it _will_ be done in larger chunks.
106 #define PAGE_CACHE_SHIFT PAGE_SHIFT
107 #define PAGE_CACHE_SIZE PAGE_SIZE
108 #define PAGE_CACHE_MASK PAGE_MASK
109 #define PAGE_CACHE_ALIGN(addr) (((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
111 #define page_cache_get(page) get_page(page)
112 #define page_cache_release(page) put_page(page)
113 void release_pages(struct page **pages, int nr, int cold);
116 * speculatively take a reference to a page.
117 * If the page is free (_count == 0), then _count is untouched, and 0
118 * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
120 * This function must be called inside the same rcu_read_lock() section as has
121 * been used to lookup the page in the pagecache radix-tree (or page table):
122 * this allows allocators to use a synchronize_rcu() to stabilize _count.
124 * Unless an RCU grace period has passed, the count of all pages coming out
125 * of the allocator must be considered unstable. page_count may return higher
126 * than expected, and put_page must be able to do the right thing when the
127 * page has been finished with, no matter what it is subsequently allocated
128 * for (because put_page is what is used here to drop an invalid speculative
131 * This is the interesting part of the lockless pagecache (and lockless
132 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
133 * has the following pattern:
134 * 1. find page in radix tree
135 * 2. conditionally increment refcount
136 * 3. check the page is still in pagecache (if no, goto 1)
138 * Remove-side that cares about stability of _count (eg. reclaim) has the
139 * following (with tree_lock held for write):
140 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
141 * B. remove page from pagecache
144 * There are 2 critical interleavings that matter:
145 * - 2 runs before A: in this case, A sees elevated refcount and bails out
146 * - A runs before 2: in this case, 2 sees zero refcount and retries;
147 * subsequently, B will complete and 1 will find no page, causing the
148 * lookup to return NULL.
150 * It is possible that between 1 and 2, the page is removed then the exact same
151 * page is inserted into the same position in pagecache. That's OK: the
152 * old find_get_page using tree_lock could equally have run before or after
153 * such a re-insertion, depending on order that locks are granted.
155 * Lookups racing against pagecache insertion isn't a big problem: either 1
156 * will find the page or it will not. Likewise, the old find_get_page could run
157 * either before the insertion or afterwards, depending on timing.
159 static inline int page_cache_get_speculative(struct page *page)
161 VM_BUG_ON(in_interrupt());
163 #ifdef CONFIG_TINY_RCU
164 # ifdef CONFIG_PREEMPT_COUNT
165 VM_BUG_ON(!in_atomic());
168 * Preempt must be disabled here - we rely on rcu_read_lock doing
171 * Pagecache won't be truncated from interrupt context, so if we have
172 * found a page in the radix tree here, we have pinned its refcount by
173 * disabling preempt, and hence no need for the "speculative get" that
176 VM_BUG_ON_PAGE(page_count(page) == 0, page);
177 atomic_inc(&page->_count);
180 if (unlikely(!get_page_unless_zero(page))) {
182 * Either the page has been freed, or will be freed.
183 * In either case, retry here and the caller should
184 * do the right thing (see comments above).
189 VM_BUG_ON_PAGE(PageTail(page), page);
195 * Same as above, but add instead of inc (could just be merged)
197 static inline int page_cache_add_speculative(struct page *page, int count)
199 VM_BUG_ON(in_interrupt());
201 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
202 # ifdef CONFIG_PREEMPT_COUNT
203 VM_BUG_ON(!in_atomic());
205 VM_BUG_ON_PAGE(page_count(page) == 0, page);
206 atomic_add(count, &page->_count);
209 if (unlikely(!atomic_add_unless(&page->_count, count, 0)))
212 VM_BUG_ON_PAGE(PageCompound(page) && page != compound_head(page), page);
217 static inline int page_freeze_refs(struct page *page, int count)
219 return likely(atomic_cmpxchg(&page->_count, count, 0) == count);
222 static inline void page_unfreeze_refs(struct page *page, int count)
224 VM_BUG_ON_PAGE(page_count(page) != 0, page);
225 VM_BUG_ON(count == 0);
227 atomic_set(&page->_count, count);
231 extern struct page *__page_cache_alloc(gfp_t gfp);
233 static inline struct page *__page_cache_alloc(gfp_t gfp)
235 return alloc_pages(gfp, 0);
239 static inline struct page *page_cache_alloc(struct address_space *x)
241 return __page_cache_alloc(mapping_gfp_mask(x));
244 static inline struct page *page_cache_alloc_cold(struct address_space *x)
246 return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
249 static inline struct page *page_cache_alloc_readahead(struct address_space *x)
251 return __page_cache_alloc(mapping_gfp_mask(x) |
252 __GFP_COLD | __GFP_NORETRY | __GFP_NOWARN);
255 typedef int filler_t(void *, struct page *);
257 pgoff_t page_cache_next_hole(struct address_space *mapping,
258 pgoff_t index, unsigned long max_scan);
259 pgoff_t page_cache_prev_hole(struct address_space *mapping,
260 pgoff_t index, unsigned long max_scan);
262 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
263 struct page *find_get_page(struct address_space *mapping, pgoff_t offset);
264 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
265 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset);
266 struct page *find_or_create_page(struct address_space *mapping, pgoff_t index,
268 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
269 unsigned int nr_entries, struct page **entries,
271 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
272 unsigned int nr_pages, struct page **pages);
273 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
274 unsigned int nr_pages, struct page **pages);
275 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
276 int tag, unsigned int nr_pages, struct page **pages);
278 struct page *grab_cache_page_write_begin(struct address_space *mapping,
279 pgoff_t index, unsigned flags);
282 * Returns locked page at given index in given cache, creating it if needed.
284 static inline struct page *grab_cache_page(struct address_space *mapping,
287 return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
290 extern struct page * grab_cache_page_nowait(struct address_space *mapping,
292 extern struct page * read_cache_page(struct address_space *mapping,
293 pgoff_t index, filler_t *filler, void *data);
294 extern struct page * read_cache_page_gfp(struct address_space *mapping,
295 pgoff_t index, gfp_t gfp_mask);
296 extern int read_cache_pages(struct address_space *mapping,
297 struct list_head *pages, filler_t *filler, void *data);
299 static inline struct page *read_mapping_page(struct address_space *mapping,
300 pgoff_t index, void *data)
302 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
303 return read_cache_page(mapping, index, filler, data);
307 * Return byte-offset into filesystem object for page.
309 static inline loff_t page_offset(struct page *page)
311 return ((loff_t)page->index) << PAGE_CACHE_SHIFT;
314 static inline loff_t page_file_offset(struct page *page)
316 return ((loff_t)page_file_index(page)) << PAGE_CACHE_SHIFT;
319 extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
320 unsigned long address);
322 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
323 unsigned long address)
326 if (unlikely(is_vm_hugetlb_page(vma)))
327 return linear_hugepage_index(vma, address);
328 pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
329 pgoff += vma->vm_pgoff;
330 return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT);
333 extern void __lock_page(struct page *page);
334 extern int __lock_page_killable(struct page *page);
335 extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
337 extern void unlock_page(struct page *page);
339 static inline void __set_page_locked(struct page *page)
341 __set_bit(PG_locked, &page->flags);
344 static inline void __clear_page_locked(struct page *page)
346 __clear_bit(PG_locked, &page->flags);
349 static inline int trylock_page(struct page *page)
351 return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
355 * lock_page may only be called if we have the page's inode pinned.
357 static inline void lock_page(struct page *page)
360 if (!trylock_page(page))
365 * lock_page_killable is like lock_page but can be interrupted by fatal
366 * signals. It returns 0 if it locked the page and -EINTR if it was
367 * killed while waiting.
369 static inline int lock_page_killable(struct page *page)
372 if (!trylock_page(page))
373 return __lock_page_killable(page);
378 * lock_page_or_retry - Lock the page, unless this would block and the
379 * caller indicated that it can handle a retry.
381 static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
385 return trylock_page(page) || __lock_page_or_retry(page, mm, flags);
389 * This is exported only for wait_on_page_locked/wait_on_page_writeback.
390 * Never use this directly!
392 extern void wait_on_page_bit(struct page *page, int bit_nr);
394 extern int wait_on_page_bit_killable(struct page *page, int bit_nr);
396 static inline int wait_on_page_locked_killable(struct page *page)
398 if (PageLocked(page))
399 return wait_on_page_bit_killable(page, PG_locked);
404 * Wait for a page to be unlocked.
406 * This must be called with the caller "holding" the page,
407 * ie with increased "page->count" so that the page won't
408 * go away during the wait..
410 static inline void wait_on_page_locked(struct page *page)
412 if (PageLocked(page))
413 wait_on_page_bit(page, PG_locked);
417 * Wait for a page to complete writeback
419 static inline void wait_on_page_writeback(struct page *page)
421 if (PageWriteback(page))
422 wait_on_page_bit(page, PG_writeback);
425 extern void end_page_writeback(struct page *page);
426 void wait_for_stable_page(struct page *page);
429 * Add an arbitrary waiter to a page's wait queue
431 extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter);
434 * Fault a userspace page into pagetables. Return non-zero on a fault.
436 * This assumes that two userspace pages are always sufficient. That's
437 * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
439 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
443 if (unlikely(size == 0))
447 * Writing zeroes into userspace here is OK, because we know that if
448 * the zero gets there, we'll be overwriting it.
450 ret = __put_user(0, uaddr);
452 char __user *end = uaddr + size - 1;
455 * If the page was already mapped, this will get a cache miss
456 * for sure, so try to avoid doing it.
458 if (((unsigned long)uaddr & PAGE_MASK) !=
459 ((unsigned long)end & PAGE_MASK))
460 ret = __put_user(0, end);
465 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
470 if (unlikely(size == 0))
473 ret = __get_user(c, uaddr);
475 const char __user *end = uaddr + size - 1;
477 if (((unsigned long)uaddr & PAGE_MASK) !=
478 ((unsigned long)end & PAGE_MASK)) {
479 ret = __get_user(c, end);
487 * Multipage variants of the above prefault helpers, useful if more than
488 * PAGE_SIZE of data needs to be prefaulted. These are separate from the above
489 * functions (which only handle up to PAGE_SIZE) to avoid clobbering the
490 * filemap.c hotpaths.
492 static inline int fault_in_multipages_writeable(char __user *uaddr, int size)
495 char __user *end = uaddr + size - 1;
497 if (unlikely(size == 0))
501 * Writing zeroes into userspace here is OK, because we know that if
502 * the zero gets there, we'll be overwriting it.
504 while (uaddr <= end) {
505 ret = __put_user(0, uaddr);
511 /* Check whether the range spilled into the next page. */
512 if (((unsigned long)uaddr & PAGE_MASK) ==
513 ((unsigned long)end & PAGE_MASK))
514 ret = __put_user(0, end);
519 static inline int fault_in_multipages_readable(const char __user *uaddr,
524 const char __user *end = uaddr + size - 1;
526 if (unlikely(size == 0))
529 while (uaddr <= end) {
530 ret = __get_user(c, uaddr);
536 /* Check whether the range spilled into the next page. */
537 if (((unsigned long)uaddr & PAGE_MASK) ==
538 ((unsigned long)end & PAGE_MASK)) {
539 ret = __get_user(c, end);
546 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
547 pgoff_t index, gfp_t gfp_mask);
548 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
549 pgoff_t index, gfp_t gfp_mask);
550 extern void delete_from_page_cache(struct page *page);
551 extern void __delete_from_page_cache(struct page *page, void *shadow);
552 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
555 * Like add_to_page_cache_locked, but used to add newly allocated pages:
556 * the page is new, so we can just run __set_page_locked() against it.
558 static inline int add_to_page_cache(struct page *page,
559 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
563 __set_page_locked(page);
564 error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
566 __clear_page_locked(page);
570 #endif /* _LINUX_PAGEMAP_H */