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 */
30 static inline void mapping_set_error(struct address_space *mapping, int error)
32 if (unlikely(error)) {
34 set_bit(AS_ENOSPC, &mapping->flags);
36 set_bit(AS_EIO, &mapping->flags);
40 static inline void mapping_set_unevictable(struct address_space *mapping)
42 set_bit(AS_UNEVICTABLE, &mapping->flags);
45 static inline void mapping_clear_unevictable(struct address_space *mapping)
47 clear_bit(AS_UNEVICTABLE, &mapping->flags);
50 static inline int mapping_unevictable(struct address_space *mapping)
53 return test_bit(AS_UNEVICTABLE, &mapping->flags);
57 static inline void mapping_set_balloon(struct address_space *mapping)
59 set_bit(AS_BALLOON_MAP, &mapping->flags);
62 static inline void mapping_clear_balloon(struct address_space *mapping)
64 clear_bit(AS_BALLOON_MAP, &mapping->flags);
67 static inline int mapping_balloon(struct address_space *mapping)
69 return mapping && test_bit(AS_BALLOON_MAP, &mapping->flags);
72 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
74 return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
78 * This is non-atomic. Only to be used before the mapping is activated.
79 * Probably needs a barrier...
81 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
83 m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
84 (__force unsigned long)mask;
88 * The page cache can done in larger chunks than
89 * one page, because it allows for more efficient
90 * throughput (it can then be mapped into user
91 * space in smaller chunks for same flexibility).
93 * Or rather, it _will_ be done in larger chunks.
95 #define PAGE_CACHE_SHIFT PAGE_SHIFT
96 #define PAGE_CACHE_SIZE PAGE_SIZE
97 #define PAGE_CACHE_MASK PAGE_MASK
98 #define PAGE_CACHE_ALIGN(addr) (((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
100 #define page_cache_get(page) get_page(page)
101 #define page_cache_release(page) put_page(page)
102 void release_pages(struct page **pages, int nr, int cold);
105 * speculatively take a reference to a page.
106 * If the page is free (_count == 0), then _count is untouched, and 0
107 * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
109 * This function must be called inside the same rcu_read_lock() section as has
110 * been used to lookup the page in the pagecache radix-tree (or page table):
111 * this allows allocators to use a synchronize_rcu() to stabilize _count.
113 * Unless an RCU grace period has passed, the count of all pages coming out
114 * of the allocator must be considered unstable. page_count may return higher
115 * than expected, and put_page must be able to do the right thing when the
116 * page has been finished with, no matter what it is subsequently allocated
117 * for (because put_page is what is used here to drop an invalid speculative
120 * This is the interesting part of the lockless pagecache (and lockless
121 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
122 * has the following pattern:
123 * 1. find page in radix tree
124 * 2. conditionally increment refcount
125 * 3. check the page is still in pagecache (if no, goto 1)
127 * Remove-side that cares about stability of _count (eg. reclaim) has the
128 * following (with tree_lock held for write):
129 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
130 * B. remove page from pagecache
133 * There are 2 critical interleavings that matter:
134 * - 2 runs before A: in this case, A sees elevated refcount and bails out
135 * - A runs before 2: in this case, 2 sees zero refcount and retries;
136 * subsequently, B will complete and 1 will find no page, causing the
137 * lookup to return NULL.
139 * It is possible that between 1 and 2, the page is removed then the exact same
140 * page is inserted into the same position in pagecache. That's OK: the
141 * old find_get_page using tree_lock could equally have run before or after
142 * such a re-insertion, depending on order that locks are granted.
144 * Lookups racing against pagecache insertion isn't a big problem: either 1
145 * will find the page or it will not. Likewise, the old find_get_page could run
146 * either before the insertion or afterwards, depending on timing.
148 static inline int page_cache_get_speculative(struct page *page)
150 VM_BUG_ON(in_interrupt());
152 #ifdef CONFIG_TINY_RCU
153 # ifdef CONFIG_PREEMPT_COUNT
154 VM_BUG_ON(!in_atomic());
157 * Preempt must be disabled here - we rely on rcu_read_lock doing
160 * Pagecache won't be truncated from interrupt context, so if we have
161 * found a page in the radix tree here, we have pinned its refcount by
162 * disabling preempt, and hence no need for the "speculative get" that
165 VM_BUG_ON_PAGE(page_count(page) == 0, page);
166 atomic_inc(&page->_count);
169 if (unlikely(!get_page_unless_zero(page))) {
171 * Either the page has been freed, or will be freed.
172 * In either case, retry here and the caller should
173 * do the right thing (see comments above).
178 VM_BUG_ON_PAGE(PageTail(page), page);
184 * Same as above, but add instead of inc (could just be merged)
186 static inline int page_cache_add_speculative(struct page *page, int count)
188 VM_BUG_ON(in_interrupt());
190 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
191 # ifdef CONFIG_PREEMPT_COUNT
192 VM_BUG_ON(!in_atomic());
194 VM_BUG_ON_PAGE(page_count(page) == 0, page);
195 atomic_add(count, &page->_count);
198 if (unlikely(!atomic_add_unless(&page->_count, count, 0)))
201 VM_BUG_ON_PAGE(PageCompound(page) && page != compound_head(page), page);
206 static inline int page_freeze_refs(struct page *page, int count)
208 return likely(atomic_cmpxchg(&page->_count, count, 0) == count);
211 static inline void page_unfreeze_refs(struct page *page, int count)
213 VM_BUG_ON_PAGE(page_count(page) != 0, page);
214 VM_BUG_ON(count == 0);
216 atomic_set(&page->_count, count);
220 extern struct page *__page_cache_alloc(gfp_t gfp);
222 static inline struct page *__page_cache_alloc(gfp_t gfp)
224 return alloc_pages(gfp, 0);
228 static inline struct page *page_cache_alloc(struct address_space *x)
230 return __page_cache_alloc(mapping_gfp_mask(x));
233 static inline struct page *page_cache_alloc_cold(struct address_space *x)
235 return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
238 static inline struct page *page_cache_alloc_readahead(struct address_space *x)
240 return __page_cache_alloc(mapping_gfp_mask(x) |
241 __GFP_COLD | __GFP_NORETRY | __GFP_NOWARN);
244 typedef int filler_t(void *, struct page *);
246 pgoff_t page_cache_next_hole(struct address_space *mapping,
247 pgoff_t index, unsigned long max_scan);
248 pgoff_t page_cache_prev_hole(struct address_space *mapping,
249 pgoff_t index, unsigned long max_scan);
251 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
252 struct page *find_get_page(struct address_space *mapping, pgoff_t offset);
253 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
254 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset);
255 struct page *find_or_create_page(struct address_space *mapping, pgoff_t index,
257 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
258 unsigned int nr_entries, struct page **entries,
260 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
261 unsigned int nr_pages, struct page **pages);
262 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
263 unsigned int nr_pages, struct page **pages);
264 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
265 int tag, unsigned int nr_pages, struct page **pages);
267 struct page *grab_cache_page_write_begin(struct address_space *mapping,
268 pgoff_t index, unsigned flags);
271 * Returns locked page at given index in given cache, creating it if needed.
273 static inline struct page *grab_cache_page(struct address_space *mapping,
276 return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
279 extern struct page * grab_cache_page_nowait(struct address_space *mapping,
281 extern struct page * read_cache_page_async(struct address_space *mapping,
282 pgoff_t index, filler_t *filler, void *data);
283 extern struct page * read_cache_page(struct address_space *mapping,
284 pgoff_t index, filler_t *filler, void *data);
285 extern struct page * read_cache_page_gfp(struct address_space *mapping,
286 pgoff_t index, gfp_t gfp_mask);
287 extern int read_cache_pages(struct address_space *mapping,
288 struct list_head *pages, filler_t *filler, void *data);
290 static inline struct page *read_mapping_page_async(
291 struct address_space *mapping,
292 pgoff_t index, void *data)
294 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
295 return read_cache_page_async(mapping, index, filler, data);
298 static inline struct page *read_mapping_page(struct address_space *mapping,
299 pgoff_t index, void *data)
301 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
302 return read_cache_page(mapping, index, filler, data);
306 * Return byte-offset into filesystem object for page.
308 static inline loff_t page_offset(struct page *page)
310 return ((loff_t)page->index) << PAGE_CACHE_SHIFT;
313 static inline loff_t page_file_offset(struct page *page)
315 return ((loff_t)page_file_index(page)) << PAGE_CACHE_SHIFT;
318 extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
319 unsigned long address);
321 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
322 unsigned long address)
325 if (unlikely(is_vm_hugetlb_page(vma)))
326 return linear_hugepage_index(vma, address);
327 pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
328 pgoff += vma->vm_pgoff;
329 return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT);
332 extern void __lock_page(struct page *page);
333 extern int __lock_page_killable(struct page *page);
334 extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
336 extern void unlock_page(struct page *page);
338 static inline void __set_page_locked(struct page *page)
340 __set_bit(PG_locked, &page->flags);
343 static inline void __clear_page_locked(struct page *page)
345 __clear_bit(PG_locked, &page->flags);
348 static inline int trylock_page(struct page *page)
350 return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
354 * lock_page may only be called if we have the page's inode pinned.
356 static inline void lock_page(struct page *page)
359 if (!trylock_page(page))
364 * lock_page_killable is like lock_page but can be interrupted by fatal
365 * signals. It returns 0 if it locked the page and -EINTR if it was
366 * killed while waiting.
368 static inline int lock_page_killable(struct page *page)
371 if (!trylock_page(page))
372 return __lock_page_killable(page);
377 * lock_page_or_retry - Lock the page, unless this would block and the
378 * caller indicated that it can handle a retry.
380 static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
384 return trylock_page(page) || __lock_page_or_retry(page, mm, flags);
388 * This is exported only for wait_on_page_locked/wait_on_page_writeback.
389 * Never use this directly!
391 extern void wait_on_page_bit(struct page *page, int bit_nr);
393 extern int wait_on_page_bit_killable(struct page *page, int bit_nr);
395 static inline int wait_on_page_locked_killable(struct page *page)
397 if (PageLocked(page))
398 return wait_on_page_bit_killable(page, PG_locked);
403 * Wait for a page to be unlocked.
405 * This must be called with the caller "holding" the page,
406 * ie with increased "page->count" so that the page won't
407 * go away during the wait..
409 static inline void wait_on_page_locked(struct page *page)
411 if (PageLocked(page))
412 wait_on_page_bit(page, PG_locked);
416 * Wait for a page to complete writeback
418 static inline void wait_on_page_writeback(struct page *page)
420 if (PageWriteback(page))
421 wait_on_page_bit(page, PG_writeback);
424 extern void end_page_writeback(struct page *page);
425 void wait_for_stable_page(struct page *page);
428 * Add an arbitrary waiter to a page's wait queue
430 extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter);
433 * Fault a userspace page into pagetables. Return non-zero on a fault.
435 * This assumes that two userspace pages are always sufficient. That's
436 * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
438 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
442 if (unlikely(size == 0))
446 * Writing zeroes into userspace here is OK, because we know that if
447 * the zero gets there, we'll be overwriting it.
449 ret = __put_user(0, uaddr);
451 char __user *end = uaddr + size - 1;
454 * If the page was already mapped, this will get a cache miss
455 * for sure, so try to avoid doing it.
457 if (((unsigned long)uaddr & PAGE_MASK) !=
458 ((unsigned long)end & PAGE_MASK))
459 ret = __put_user(0, end);
464 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
469 if (unlikely(size == 0))
472 ret = __get_user(c, uaddr);
474 const char __user *end = uaddr + size - 1;
476 if (((unsigned long)uaddr & PAGE_MASK) !=
477 ((unsigned long)end & PAGE_MASK)) {
478 ret = __get_user(c, end);
486 * Multipage variants of the above prefault helpers, useful if more than
487 * PAGE_SIZE of data needs to be prefaulted. These are separate from the above
488 * functions (which only handle up to PAGE_SIZE) to avoid clobbering the
489 * filemap.c hotpaths.
491 static inline int fault_in_multipages_writeable(char __user *uaddr, int size)
494 char __user *end = uaddr + size - 1;
496 if (unlikely(size == 0))
500 * Writing zeroes into userspace here is OK, because we know that if
501 * the zero gets there, we'll be overwriting it.
503 while (uaddr <= end) {
504 ret = __put_user(0, uaddr);
510 /* Check whether the range spilled into the next page. */
511 if (((unsigned long)uaddr & PAGE_MASK) ==
512 ((unsigned long)end & PAGE_MASK))
513 ret = __put_user(0, end);
518 static inline int fault_in_multipages_readable(const char __user *uaddr,
523 const char __user *end = uaddr + size - 1;
525 if (unlikely(size == 0))
528 while (uaddr <= end) {
529 ret = __get_user(c, uaddr);
535 /* Check whether the range spilled into the next page. */
536 if (((unsigned long)uaddr & PAGE_MASK) ==
537 ((unsigned long)end & PAGE_MASK)) {
538 ret = __get_user(c, end);
545 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
546 pgoff_t index, gfp_t gfp_mask);
547 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
548 pgoff_t index, gfp_t gfp_mask);
549 extern void delete_from_page_cache(struct page *page);
550 extern void __delete_from_page_cache(struct page *page);
551 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
554 * Like add_to_page_cache_locked, but used to add newly allocated pages:
555 * the page is new, so we can just run __set_page_locked() against it.
557 static inline int add_to_page_cache(struct page *page,
558 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
562 __set_page_locked(page);
563 error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
565 __clear_page_locked(page);
569 #endif /* _LINUX_PAGEMAP_H */