Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | #ifndef _LINUX_PAGEMAP_H |
| 2 | #define _LINUX_PAGEMAP_H |
| 3 | |
| 4 | /* |
| 5 | * Copyright 1995 Linus Torvalds |
| 6 | */ |
| 7 | #include <linux/mm.h> |
| 8 | #include <linux/fs.h> |
| 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> |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 14 | #include <linux/bitops.h> |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 15 | #include <linux/hardirq.h> /* for in_interrupt() */ |
Naoya Horiguchi | 8edf344 | 2010-05-28 09:29:15 +0900 | [diff] [blame] | 16 | #include <linux/hugetlb_inline.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | |
| 18 | /* |
| 19 | * Bits in mapping->flags. The lower __GFP_BITS_SHIFT bits are the page |
| 20 | * allocation mode flags. |
| 21 | */ |
Lee Schermerhorn | 9a896c9 | 2009-04-02 16:56:45 -0700 | [diff] [blame] | 22 | enum mapping_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() */ |
Lee Schermerhorn | 9a896c9 | 2009-04-02 16:56:45 -0700 | [diff] [blame] | 26 | AS_UNEVICTABLE = __GFP_BITS_SHIFT + 3, /* e.g., ramdisk, SHM_LOCK */ |
Rafael Aquini | 18468d9 | 2012-12-11 16:02:38 -0800 | [diff] [blame] | 27 | AS_BALLOON_MAP = __GFP_BITS_SHIFT + 4, /* balloon page special map */ |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 28 | AS_EXITING = __GFP_BITS_SHIFT + 5, /* final truncate in progress */ |
Lee Schermerhorn | 9a896c9 | 2009-04-02 16:56:45 -0700 | [diff] [blame] | 29 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 31 | static inline void mapping_set_error(struct address_space *mapping, int error) |
| 32 | { |
Andrew Morton | 2185e69 | 2008-07-23 21:27:19 -0700 | [diff] [blame] | 33 | if (unlikely(error)) { |
Guillaume Chazarain | 3e9f45b | 2007-05-08 00:23:25 -0700 | [diff] [blame] | 34 | if (error == -ENOSPC) |
| 35 | set_bit(AS_ENOSPC, &mapping->flags); |
| 36 | else |
| 37 | set_bit(AS_EIO, &mapping->flags); |
| 38 | } |
| 39 | } |
| 40 | |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 41 | static inline void mapping_set_unevictable(struct address_space *mapping) |
| 42 | { |
| 43 | set_bit(AS_UNEVICTABLE, &mapping->flags); |
| 44 | } |
| 45 | |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 46 | static inline void mapping_clear_unevictable(struct address_space *mapping) |
| 47 | { |
| 48 | clear_bit(AS_UNEVICTABLE, &mapping->flags); |
| 49 | } |
| 50 | |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 51 | static inline int mapping_unevictable(struct address_space *mapping) |
| 52 | { |
Steven Rostedt | 088e546 | 2011-01-13 15:46:16 -0800 | [diff] [blame] | 53 | if (mapping) |
Lee Schermerhorn | 89e004ea | 2008-10-18 20:26:43 -0700 | [diff] [blame] | 54 | return test_bit(AS_UNEVICTABLE, &mapping->flags); |
| 55 | return !!mapping; |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 56 | } |
Lee Schermerhorn | ba9ddf4 | 2008-10-18 20:26:42 -0700 | [diff] [blame] | 57 | |
Rafael Aquini | 18468d9 | 2012-12-11 16:02:38 -0800 | [diff] [blame] | 58 | static inline void mapping_set_balloon(struct address_space *mapping) |
| 59 | { |
| 60 | set_bit(AS_BALLOON_MAP, &mapping->flags); |
| 61 | } |
| 62 | |
| 63 | static inline void mapping_clear_balloon(struct address_space *mapping) |
| 64 | { |
| 65 | clear_bit(AS_BALLOON_MAP, &mapping->flags); |
| 66 | } |
| 67 | |
| 68 | static inline int mapping_balloon(struct address_space *mapping) |
| 69 | { |
| 70 | return mapping && test_bit(AS_BALLOON_MAP, &mapping->flags); |
| 71 | } |
| 72 | |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 73 | static inline void mapping_set_exiting(struct address_space *mapping) |
| 74 | { |
| 75 | set_bit(AS_EXITING, &mapping->flags); |
| 76 | } |
| 77 | |
| 78 | static inline int mapping_exiting(struct address_space *mapping) |
| 79 | { |
| 80 | return test_bit(AS_EXITING, &mapping->flags); |
| 81 | } |
| 82 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 83 | static inline gfp_t mapping_gfp_mask(struct address_space * mapping) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | { |
Al Viro | 260b236 | 2005-10-21 03:22:44 -0400 | [diff] [blame] | 85 | return (__force gfp_t)mapping->flags & __GFP_BITS_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | } |
| 87 | |
| 88 | /* |
| 89 | * This is non-atomic. Only to be used before the mapping is activated. |
| 90 | * Probably needs a barrier... |
| 91 | */ |
Al Viro | 260b236 | 2005-10-21 03:22:44 -0400 | [diff] [blame] | 92 | static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | { |
Al Viro | 260b236 | 2005-10-21 03:22:44 -0400 | [diff] [blame] | 94 | m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) | |
| 95 | (__force unsigned long)mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | } |
| 97 | |
| 98 | /* |
| 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). |
| 103 | * |
| 104 | * Or rather, it _will_ be done in larger chunks. |
| 105 | */ |
| 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) |
| 110 | |
| 111 | #define page_cache_get(page) get_page(page) |
| 112 | #define page_cache_release(page) put_page(page) |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 113 | void release_pages(struct page **pages, int nr, bool cold); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 115 | /* |
| 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. |
| 119 | * |
| 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. |
| 123 | * |
| 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 |
| 129 | * reference). |
| 130 | * |
| 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) |
| 137 | * |
| 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 |
| 142 | * C. free the page |
| 143 | * |
| 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. |
| 149 | * |
| 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. |
| 154 | * |
| 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. |
| 158 | */ |
| 159 | static inline int page_cache_get_speculative(struct page *page) |
| 160 | { |
| 161 | VM_BUG_ON(in_interrupt()); |
| 162 | |
Paul E. McKenney | 8375ad9 | 2013-04-29 15:06:13 -0700 | [diff] [blame] | 163 | #ifdef CONFIG_TINY_RCU |
Frederic Weisbecker | bdd4e85 | 2011-06-08 01:13:27 +0200 | [diff] [blame] | 164 | # ifdef CONFIG_PREEMPT_COUNT |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 165 | VM_BUG_ON(!in_atomic()); |
| 166 | # endif |
| 167 | /* |
| 168 | * Preempt must be disabled here - we rely on rcu_read_lock doing |
| 169 | * this for us. |
| 170 | * |
| 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 |
| 174 | * SMP requires. |
| 175 | */ |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 176 | VM_BUG_ON_PAGE(page_count(page) == 0, page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 177 | atomic_inc(&page->_count); |
| 178 | |
| 179 | #else |
| 180 | if (unlikely(!get_page_unless_zero(page))) { |
| 181 | /* |
| 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). |
| 185 | */ |
| 186 | return 0; |
| 187 | } |
| 188 | #endif |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 189 | VM_BUG_ON_PAGE(PageTail(page), page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 190 | |
| 191 | return 1; |
| 192 | } |
| 193 | |
Nick Piggin | ce0ad7f | 2008-07-30 15:23:13 +1000 | [diff] [blame] | 194 | /* |
| 195 | * Same as above, but add instead of inc (could just be merged) |
| 196 | */ |
| 197 | static inline int page_cache_add_speculative(struct page *page, int count) |
| 198 | { |
| 199 | VM_BUG_ON(in_interrupt()); |
| 200 | |
Paul E. McKenney | b560d8a | 2009-08-21 22:08:51 -0700 | [diff] [blame] | 201 | #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU) |
Frederic Weisbecker | bdd4e85 | 2011-06-08 01:13:27 +0200 | [diff] [blame] | 202 | # ifdef CONFIG_PREEMPT_COUNT |
Nick Piggin | ce0ad7f | 2008-07-30 15:23:13 +1000 | [diff] [blame] | 203 | VM_BUG_ON(!in_atomic()); |
| 204 | # endif |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 205 | VM_BUG_ON_PAGE(page_count(page) == 0, page); |
Nick Piggin | ce0ad7f | 2008-07-30 15:23:13 +1000 | [diff] [blame] | 206 | atomic_add(count, &page->_count); |
| 207 | |
| 208 | #else |
| 209 | if (unlikely(!atomic_add_unless(&page->_count, count, 0))) |
| 210 | return 0; |
| 211 | #endif |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 212 | VM_BUG_ON_PAGE(PageCompound(page) && page != compound_head(page), page); |
Nick Piggin | ce0ad7f | 2008-07-30 15:23:13 +1000 | [diff] [blame] | 213 | |
| 214 | return 1; |
| 215 | } |
| 216 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 217 | static inline int page_freeze_refs(struct page *page, int count) |
| 218 | { |
| 219 | return likely(atomic_cmpxchg(&page->_count, count, 0) == count); |
| 220 | } |
| 221 | |
| 222 | static inline void page_unfreeze_refs(struct page *page, int count) |
| 223 | { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 224 | VM_BUG_ON_PAGE(page_count(page) != 0, page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 225 | VM_BUG_ON(count == 0); |
| 226 | |
| 227 | atomic_set(&page->_count, count); |
| 228 | } |
| 229 | |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 230 | #ifdef CONFIG_NUMA |
Nick Piggin | 2ae8814 | 2006-10-28 10:38:23 -0700 | [diff] [blame] | 231 | extern struct page *__page_cache_alloc(gfp_t gfp); |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 232 | #else |
Nick Piggin | 2ae8814 | 2006-10-28 10:38:23 -0700 | [diff] [blame] | 233 | static inline struct page *__page_cache_alloc(gfp_t gfp) |
| 234 | { |
| 235 | return alloc_pages(gfp, 0); |
| 236 | } |
| 237 | #endif |
| 238 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | static inline struct page *page_cache_alloc(struct address_space *x) |
| 240 | { |
Nick Piggin | 2ae8814 | 2006-10-28 10:38:23 -0700 | [diff] [blame] | 241 | return __page_cache_alloc(mapping_gfp_mask(x)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 242 | } |
| 243 | |
| 244 | static inline struct page *page_cache_alloc_cold(struct address_space *x) |
| 245 | { |
Nick Piggin | 2ae8814 | 2006-10-28 10:38:23 -0700 | [diff] [blame] | 246 | return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | } |
| 248 | |
Wu Fengguang | 7b1de58 | 2011-05-24 17:12:25 -0700 | [diff] [blame] | 249 | static inline struct page *page_cache_alloc_readahead(struct address_space *x) |
| 250 | { |
| 251 | return __page_cache_alloc(mapping_gfp_mask(x) | |
| 252 | __GFP_COLD | __GFP_NORETRY | __GFP_NOWARN); |
| 253 | } |
| 254 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | typedef int filler_t(void *, struct page *); |
| 256 | |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 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); |
| 261 | |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 262 | #define FGP_ACCESSED 0x00000001 |
| 263 | #define FGP_LOCK 0x00000002 |
| 264 | #define FGP_CREAT 0x00000004 |
| 265 | #define FGP_WRITE 0x00000008 |
| 266 | #define FGP_NOFS 0x00000010 |
| 267 | #define FGP_NOWAIT 0x00000020 |
| 268 | |
| 269 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
| 270 | int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask); |
| 271 | |
| 272 | /** |
| 273 | * find_get_page - find and get a page reference |
| 274 | * @mapping: the address_space to search |
| 275 | * @offset: the page index |
| 276 | * |
| 277 | * Looks up the page cache slot at @mapping & @offset. If there is a |
| 278 | * page cache page, it is returned with an increased refcount. |
| 279 | * |
| 280 | * Otherwise, %NULL is returned. |
| 281 | */ |
| 282 | static inline struct page *find_get_page(struct address_space *mapping, |
| 283 | pgoff_t offset) |
| 284 | { |
| 285 | return pagecache_get_page(mapping, offset, 0, 0, 0); |
| 286 | } |
| 287 | |
| 288 | static inline struct page *find_get_page_flags(struct address_space *mapping, |
| 289 | pgoff_t offset, int fgp_flags) |
| 290 | { |
| 291 | return pagecache_get_page(mapping, offset, fgp_flags, 0, 0); |
| 292 | } |
| 293 | |
| 294 | /** |
| 295 | * find_lock_page - locate, pin and lock a pagecache page |
| 296 | * pagecache_get_page - find and get a page reference |
| 297 | * @mapping: the address_space to search |
| 298 | * @offset: the page index |
| 299 | * |
| 300 | * Looks up the page cache slot at @mapping & @offset. If there is a |
| 301 | * page cache page, it is returned locked and with an increased |
| 302 | * refcount. |
| 303 | * |
| 304 | * Otherwise, %NULL is returned. |
| 305 | * |
| 306 | * find_lock_page() may sleep. |
| 307 | */ |
| 308 | static inline struct page *find_lock_page(struct address_space *mapping, |
| 309 | pgoff_t offset) |
| 310 | { |
| 311 | return pagecache_get_page(mapping, offset, FGP_LOCK, 0, 0); |
| 312 | } |
| 313 | |
| 314 | /** |
| 315 | * find_or_create_page - locate or add a pagecache page |
| 316 | * @mapping: the page's address_space |
| 317 | * @index: the page's index into the mapping |
| 318 | * @gfp_mask: page allocation mode |
| 319 | * |
| 320 | * Looks up the page cache slot at @mapping & @offset. If there is a |
| 321 | * page cache page, it is returned locked and with an increased |
| 322 | * refcount. |
| 323 | * |
| 324 | * If the page is not present, a new page is allocated using @gfp_mask |
| 325 | * and added to the page cache and the VM's LRU list. The page is |
| 326 | * returned locked and with an increased refcount. |
| 327 | * |
| 328 | * On memory exhaustion, %NULL is returned. |
| 329 | * |
| 330 | * find_or_create_page() may sleep, even if @gfp_flags specifies an |
| 331 | * atomic allocation! |
| 332 | */ |
| 333 | static inline struct page *find_or_create_page(struct address_space *mapping, |
| 334 | pgoff_t offset, gfp_t gfp_mask) |
| 335 | { |
| 336 | return pagecache_get_page(mapping, offset, |
| 337 | FGP_LOCK|FGP_ACCESSED|FGP_CREAT, |
| 338 | gfp_mask, gfp_mask & GFP_RECLAIM_MASK); |
| 339 | } |
| 340 | |
| 341 | /** |
| 342 | * grab_cache_page_nowait - returns locked page at given index in given cache |
| 343 | * @mapping: target address_space |
| 344 | * @index: the page index |
| 345 | * |
| 346 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
| 347 | * This is intended for speculative data generators, where the data can |
| 348 | * be regenerated if the page couldn't be grabbed. This routine should |
| 349 | * be safe to call while holding the lock for another page. |
| 350 | * |
| 351 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs |
| 352 | * and deadlock against the caller's locked page. |
| 353 | */ |
| 354 | static inline struct page *grab_cache_page_nowait(struct address_space *mapping, |
| 355 | pgoff_t index) |
| 356 | { |
| 357 | return pagecache_get_page(mapping, index, |
| 358 | FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT, |
| 359 | mapping_gfp_mask(mapping), |
| 360 | GFP_NOFS); |
| 361 | } |
| 362 | |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 363 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset); |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 364 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset); |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 365 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
| 366 | unsigned int nr_entries, struct page **entries, |
| 367 | pgoff_t *indices); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, |
| 369 | unsigned int nr_pages, struct page **pages); |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 370 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start, |
| 371 | unsigned int nr_pages, struct page **pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, |
| 373 | int tag, unsigned int nr_pages, struct page **pages); |
| 374 | |
Nick Piggin | 54566b2 | 2009-01-04 12:00:53 -0800 | [diff] [blame] | 375 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
| 376 | pgoff_t index, unsigned flags); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 377 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | /* |
| 379 | * Returns locked page at given index in given cache, creating it if needed. |
| 380 | */ |
Fengguang Wu | 57f6b96 | 2007-10-16 01:24:37 -0700 | [diff] [blame] | 381 | static inline struct page *grab_cache_page(struct address_space *mapping, |
| 382 | pgoff_t index) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | { |
| 384 | return find_or_create_page(mapping, index, mapping_gfp_mask(mapping)); |
| 385 | } |
| 386 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 387 | extern struct page * read_cache_page(struct address_space *mapping, |
Hugh Dickins | 5e5358e | 2011-07-25 17:12:23 -0700 | [diff] [blame] | 388 | pgoff_t index, filler_t *filler, void *data); |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 389 | extern struct page * read_cache_page_gfp(struct address_space *mapping, |
| 390 | pgoff_t index, gfp_t gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | extern int read_cache_pages(struct address_space *mapping, |
| 392 | struct list_head *pages, filler_t *filler, void *data); |
| 393 | |
Pekka Enberg | 090d2b1 | 2006-06-23 02:05:08 -0700 | [diff] [blame] | 394 | static inline struct page *read_mapping_page(struct address_space *mapping, |
Hugh Dickins | 5e5358e | 2011-07-25 17:12:23 -0700 | [diff] [blame] | 395 | pgoff_t index, void *data) |
Pekka Enberg | 090d2b1 | 2006-06-23 02:05:08 -0700 | [diff] [blame] | 396 | { |
| 397 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; |
| 398 | return read_cache_page(mapping, index, filler, data); |
| 399 | } |
| 400 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 401 | /* |
Naoya Horiguchi | a0f7a75 | 2014-07-23 14:00:01 -0700 | [diff] [blame] | 402 | * Get the offset in PAGE_SIZE. |
| 403 | * (TODO: hugepage should have ->index in PAGE_SIZE) |
| 404 | */ |
| 405 | static inline pgoff_t page_to_pgoff(struct page *page) |
| 406 | { |
| 407 | if (unlikely(PageHeadHuge(page))) |
| 408 | return page->index << compound_order(page); |
| 409 | else |
| 410 | return page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 411 | } |
| 412 | |
| 413 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | * Return byte-offset into filesystem object for page. |
| 415 | */ |
| 416 | static inline loff_t page_offset(struct page *page) |
| 417 | { |
| 418 | return ((loff_t)page->index) << PAGE_CACHE_SHIFT; |
| 419 | } |
| 420 | |
Mel Gorman | f981c59 | 2012-07-31 16:44:47 -0700 | [diff] [blame] | 421 | static inline loff_t page_file_offset(struct page *page) |
| 422 | { |
| 423 | return ((loff_t)page_file_index(page)) << PAGE_CACHE_SHIFT; |
| 424 | } |
| 425 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 426 | extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma, |
| 427 | unsigned long address); |
| 428 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 429 | static inline pgoff_t linear_page_index(struct vm_area_struct *vma, |
| 430 | unsigned long address) |
| 431 | { |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 432 | pgoff_t pgoff; |
| 433 | if (unlikely(is_vm_hugetlb_page(vma))) |
| 434 | return linear_hugepage_index(vma, address); |
| 435 | pgoff = (address - vma->vm_start) >> PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | pgoff += vma->vm_pgoff; |
| 437 | return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 438 | } |
| 439 | |
Harvey Harrison | b3c9752 | 2008-02-13 15:03:15 -0800 | [diff] [blame] | 440 | extern void __lock_page(struct page *page); |
| 441 | extern int __lock_page_killable(struct page *page); |
Michel Lespinasse | d065bd8 | 2010-10-26 14:21:57 -0700 | [diff] [blame] | 442 | extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
| 443 | unsigned int flags); |
Harvey Harrison | b3c9752 | 2008-02-13 15:03:15 -0800 | [diff] [blame] | 444 | extern void unlock_page(struct page *page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | |
Nick Piggin | f45840b | 2008-10-18 20:26:57 -0700 | [diff] [blame] | 446 | static inline void __set_page_locked(struct page *page) |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 447 | { |
Nick Piggin | f45840b | 2008-10-18 20:26:57 -0700 | [diff] [blame] | 448 | __set_bit(PG_locked, &page->flags); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 449 | } |
| 450 | |
Nick Piggin | f45840b | 2008-10-18 20:26:57 -0700 | [diff] [blame] | 451 | static inline void __clear_page_locked(struct page *page) |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 452 | { |
Nick Piggin | f45840b | 2008-10-18 20:26:57 -0700 | [diff] [blame] | 453 | __clear_bit(PG_locked, &page->flags); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 454 | } |
| 455 | |
| 456 | static inline int trylock_page(struct page *page) |
| 457 | { |
Nick Piggin | 8413ac9 | 2008-10-18 20:26:59 -0700 | [diff] [blame] | 458 | return (likely(!test_and_set_bit_lock(PG_locked, &page->flags))); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 459 | } |
| 460 | |
Nick Piggin | db37648 | 2006-09-25 23:31:24 -0700 | [diff] [blame] | 461 | /* |
| 462 | * lock_page may only be called if we have the page's inode pinned. |
| 463 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | static inline void lock_page(struct page *page) |
| 465 | { |
| 466 | might_sleep(); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 467 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 468 | __lock_page(page); |
| 469 | } |
Nick Piggin | db37648 | 2006-09-25 23:31:24 -0700 | [diff] [blame] | 470 | |
| 471 | /* |
Matthew Wilcox | 2687a35 | 2007-12-06 11:18:49 -0500 | [diff] [blame] | 472 | * lock_page_killable is like lock_page but can be interrupted by fatal |
| 473 | * signals. It returns 0 if it locked the page and -EINTR if it was |
| 474 | * killed while waiting. |
| 475 | */ |
| 476 | static inline int lock_page_killable(struct page *page) |
| 477 | { |
| 478 | might_sleep(); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 479 | if (!trylock_page(page)) |
Matthew Wilcox | 2687a35 | 2007-12-06 11:18:49 -0500 | [diff] [blame] | 480 | return __lock_page_killable(page); |
| 481 | return 0; |
| 482 | } |
| 483 | |
| 484 | /* |
Michel Lespinasse | d065bd8 | 2010-10-26 14:21:57 -0700 | [diff] [blame] | 485 | * lock_page_or_retry - Lock the page, unless this would block and the |
| 486 | * caller indicated that it can handle a retry. |
| 487 | */ |
| 488 | static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm, |
| 489 | unsigned int flags) |
| 490 | { |
| 491 | might_sleep(); |
| 492 | return trylock_page(page) || __lock_page_or_retry(page, mm, flags); |
| 493 | } |
| 494 | |
| 495 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 496 | * This is exported only for wait_on_page_locked/wait_on_page_writeback. |
| 497 | * Never use this directly! |
| 498 | */ |
Harvey Harrison | b3c9752 | 2008-02-13 15:03:15 -0800 | [diff] [blame] | 499 | extern void wait_on_page_bit(struct page *page, int bit_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | |
KOSAKI Motohiro | f62e00c | 2011-05-24 17:11:29 -0700 | [diff] [blame] | 501 | extern int wait_on_page_bit_killable(struct page *page, int bit_nr); |
| 502 | |
| 503 | static inline int wait_on_page_locked_killable(struct page *page) |
| 504 | { |
| 505 | if (PageLocked(page)) |
| 506 | return wait_on_page_bit_killable(page, PG_locked); |
| 507 | return 0; |
| 508 | } |
| 509 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | /* |
| 511 | * Wait for a page to be unlocked. |
| 512 | * |
| 513 | * This must be called with the caller "holding" the page, |
| 514 | * ie with increased "page->count" so that the page won't |
| 515 | * go away during the wait.. |
| 516 | */ |
| 517 | static inline void wait_on_page_locked(struct page *page) |
| 518 | { |
| 519 | if (PageLocked(page)) |
| 520 | wait_on_page_bit(page, PG_locked); |
| 521 | } |
| 522 | |
| 523 | /* |
| 524 | * Wait for a page to complete writeback |
| 525 | */ |
| 526 | static inline void wait_on_page_writeback(struct page *page) |
| 527 | { |
| 528 | if (PageWriteback(page)) |
| 529 | wait_on_page_bit(page, PG_writeback); |
| 530 | } |
| 531 | |
| 532 | extern void end_page_writeback(struct page *page); |
Darrick J. Wong | 1d1d1a7 | 2013-02-21 16:42:51 -0800 | [diff] [blame] | 533 | void wait_for_stable_page(struct page *page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 535 | void page_endio(struct page *page, int rw, int err); |
| 536 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 537 | /* |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 538 | * Add an arbitrary waiter to a page's wait queue |
| 539 | */ |
| 540 | extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter); |
| 541 | |
| 542 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | * Fault a userspace page into pagetables. Return non-zero on a fault. |
| 544 | * |
| 545 | * This assumes that two userspace pages are always sufficient. That's |
| 546 | * not true if PAGE_CACHE_SIZE > PAGE_SIZE. |
| 547 | */ |
| 548 | static inline int fault_in_pages_writeable(char __user *uaddr, int size) |
| 549 | { |
| 550 | int ret; |
| 551 | |
Nick Piggin | 0829142 | 2007-10-16 01:24:59 -0700 | [diff] [blame] | 552 | if (unlikely(size == 0)) |
| 553 | return 0; |
| 554 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | /* |
| 556 | * Writing zeroes into userspace here is OK, because we know that if |
| 557 | * the zero gets there, we'll be overwriting it. |
| 558 | */ |
| 559 | ret = __put_user(0, uaddr); |
| 560 | if (ret == 0) { |
| 561 | char __user *end = uaddr + size - 1; |
| 562 | |
| 563 | /* |
| 564 | * If the page was already mapped, this will get a cache miss |
| 565 | * for sure, so try to avoid doing it. |
| 566 | */ |
| 567 | if (((unsigned long)uaddr & PAGE_MASK) != |
| 568 | ((unsigned long)end & PAGE_MASK)) |
Daniel Vetter | f56f821 | 2012-03-25 19:47:41 +0200 | [diff] [blame] | 569 | ret = __put_user(0, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | } |
| 571 | return ret; |
| 572 | } |
| 573 | |
Nick Piggin | 0829142 | 2007-10-16 01:24:59 -0700 | [diff] [blame] | 574 | static inline int fault_in_pages_readable(const char __user *uaddr, int size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | { |
| 576 | volatile char c; |
| 577 | int ret; |
| 578 | |
Nick Piggin | 0829142 | 2007-10-16 01:24:59 -0700 | [diff] [blame] | 579 | if (unlikely(size == 0)) |
| 580 | return 0; |
| 581 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 582 | ret = __get_user(c, uaddr); |
| 583 | if (ret == 0) { |
| 584 | const char __user *end = uaddr + size - 1; |
| 585 | |
| 586 | if (((unsigned long)uaddr & PAGE_MASK) != |
Andi Kleen | 627295e | 2010-08-09 17:19:02 -0700 | [diff] [blame] | 587 | ((unsigned long)end & PAGE_MASK)) { |
Daniel Vetter | f56f821 | 2012-03-25 19:47:41 +0200 | [diff] [blame] | 588 | ret = __get_user(c, end); |
Andi Kleen | 627295e | 2010-08-09 17:19:02 -0700 | [diff] [blame] | 589 | (void)c; |
| 590 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | } |
Nick Piggin | 0829142 | 2007-10-16 01:24:59 -0700 | [diff] [blame] | 592 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | } |
| 594 | |
Daniel Vetter | f56f821 | 2012-03-25 19:47:41 +0200 | [diff] [blame] | 595 | /* |
| 596 | * Multipage variants of the above prefault helpers, useful if more than |
| 597 | * PAGE_SIZE of data needs to be prefaulted. These are separate from the above |
| 598 | * functions (which only handle up to PAGE_SIZE) to avoid clobbering the |
| 599 | * filemap.c hotpaths. |
| 600 | */ |
| 601 | static inline int fault_in_multipages_writeable(char __user *uaddr, int size) |
| 602 | { |
Paul Gortmaker | af2e840 | 2012-05-29 15:06:14 -0700 | [diff] [blame] | 603 | int ret = 0; |
Daniel Vetter | 9923777 | 2012-04-14 18:03:10 +0200 | [diff] [blame] | 604 | char __user *end = uaddr + size - 1; |
Daniel Vetter | f56f821 | 2012-03-25 19:47:41 +0200 | [diff] [blame] | 605 | |
| 606 | if (unlikely(size == 0)) |
Paul Gortmaker | af2e840 | 2012-05-29 15:06:14 -0700 | [diff] [blame] | 607 | return ret; |
Daniel Vetter | f56f821 | 2012-03-25 19:47:41 +0200 | [diff] [blame] | 608 | |
| 609 | /* |
| 610 | * Writing zeroes into userspace here is OK, because we know that if |
| 611 | * the zero gets there, we'll be overwriting it. |
| 612 | */ |
| 613 | while (uaddr <= end) { |
| 614 | ret = __put_user(0, uaddr); |
| 615 | if (ret != 0) |
| 616 | return ret; |
| 617 | uaddr += PAGE_SIZE; |
| 618 | } |
| 619 | |
| 620 | /* Check whether the range spilled into the next page. */ |
| 621 | if (((unsigned long)uaddr & PAGE_MASK) == |
| 622 | ((unsigned long)end & PAGE_MASK)) |
| 623 | ret = __put_user(0, end); |
| 624 | |
| 625 | return ret; |
| 626 | } |
| 627 | |
| 628 | static inline int fault_in_multipages_readable(const char __user *uaddr, |
| 629 | int size) |
| 630 | { |
| 631 | volatile char c; |
Paul Gortmaker | af2e840 | 2012-05-29 15:06:14 -0700 | [diff] [blame] | 632 | int ret = 0; |
Daniel Vetter | f56f821 | 2012-03-25 19:47:41 +0200 | [diff] [blame] | 633 | const char __user *end = uaddr + size - 1; |
| 634 | |
| 635 | if (unlikely(size == 0)) |
Paul Gortmaker | af2e840 | 2012-05-29 15:06:14 -0700 | [diff] [blame] | 636 | return ret; |
Daniel Vetter | f56f821 | 2012-03-25 19:47:41 +0200 | [diff] [blame] | 637 | |
| 638 | while (uaddr <= end) { |
| 639 | ret = __get_user(c, uaddr); |
| 640 | if (ret != 0) |
| 641 | return ret; |
| 642 | uaddr += PAGE_SIZE; |
| 643 | } |
| 644 | |
| 645 | /* Check whether the range spilled into the next page. */ |
| 646 | if (((unsigned long)uaddr & PAGE_MASK) == |
| 647 | ((unsigned long)end & PAGE_MASK)) { |
| 648 | ret = __get_user(c, end); |
| 649 | (void)c; |
| 650 | } |
| 651 | |
| 652 | return ret; |
| 653 | } |
| 654 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 655 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
| 656 | pgoff_t index, gfp_t gfp_mask); |
| 657 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, |
| 658 | pgoff_t index, gfp_t gfp_mask); |
Minchan Kim | 97cecb5 | 2011-03-22 16:30:53 -0700 | [diff] [blame] | 659 | extern void delete_from_page_cache(struct page *page); |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 660 | extern void __delete_from_page_cache(struct page *page, void *shadow); |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 661 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 662 | |
| 663 | /* |
| 664 | * Like add_to_page_cache_locked, but used to add newly allocated pages: |
Nick Piggin | f45840b | 2008-10-18 20:26:57 -0700 | [diff] [blame] | 665 | * the page is new, so we can just run __set_page_locked() against it. |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 666 | */ |
| 667 | static inline int add_to_page_cache(struct page *page, |
| 668 | struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask) |
| 669 | { |
| 670 | int error; |
| 671 | |
Nick Piggin | f45840b | 2008-10-18 20:26:57 -0700 | [diff] [blame] | 672 | __set_page_locked(page); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 673 | error = add_to_page_cache_locked(page, mapping, offset, gfp_mask); |
| 674 | if (unlikely(error)) |
Nick Piggin | f45840b | 2008-10-18 20:26:57 -0700 | [diff] [blame] | 675 | __clear_page_locked(page); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 676 | return error; |
| 677 | } |
| 678 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | #endif /* _LINUX_PAGEMAP_H */ |