include/linux/pagemap.h

   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>
  14 #include <linux/bitops.h>
  15 #include <linux/hardirq.h> /* for in_interrupt() */
  16
  17 /*
  18  * Bits in mapping->flags.  The lower __GFP_BITS_SHIFT bits are the page
  19  * allocation mode flags.
  20  */
  21 enum mapping_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 #ifdef CONFIG_UNEVICTABLE_LRU
  26         AS_UNEVICTABLE  = __GFP_BITS_SHIFT + 3, /* e.g., ramdisk, SHM_LOCK */
  27 #endif
  28 };
  29
  30 static inline void mapping_set_error(struct address_space *mapping, int error)
  31 {
  32         if (unlikely(error)) {
  33                 if (error == -ENOSPC)
  34                         set_bit(AS_ENOSPC, &mapping->flags);
  35                 else
  36                         set_bit(AS_EIO, &mapping->flags);
  37         }
  38 }
  39
  40 #ifdef CONFIG_UNEVICTABLE_LRU
  41
  42 static inline void mapping_set_unevictable(struct address_space *mapping)
  43 {
  44         set_bit(AS_UNEVICTABLE, &mapping->flags);
  45 }
  46
  47 static inline void mapping_clear_unevictable(struct address_space *mapping)
  48 {
  49         clear_bit(AS_UNEVICTABLE, &mapping->flags);
  50 }
  51
  52 static inline int mapping_unevictable(struct address_space *mapping)
  53 {
  54         if (likely(mapping))
  55                 return test_bit(AS_UNEVICTABLE, &mapping->flags);
  56         return !!mapping;
  57 }
  58 #else
  59 static inline void mapping_set_unevictable(struct address_space *mapping) { }
  60 static inline void mapping_clear_unevictable(struct address_space *mapping) { }
  61 static inline int mapping_unevictable(struct address_space *mapping)
  62 {
  63         return 0;
  64 }
  65 #endif
  66
  67 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
  68 {
  69         return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
  70 }
  71
  72 /*
  73  * This is non-atomic.  Only to be used before the mapping is activated.
  74  * Probably needs a barrier...
  75  */
  76 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
  77 {
  78         m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
  79                                 (__force unsigned long)mask;
  80 }
  81
  82 /*
  83  * The page cache can done in larger chunks than
  84  * one page, because it allows for more efficient
  85  * throughput (it can then be mapped into user
  86  * space in smaller chunks for same flexibility).
  87  *
  88  * Or rather, it _will_ be done in larger chunks.
  89  */
  90 #define PAGE_CACHE_SHIFT        PAGE_SHIFT
  91 #define PAGE_CACHE_SIZE         PAGE_SIZE
  92 #define PAGE_CACHE_MASK         PAGE_MASK
  93 #define PAGE_CACHE_ALIGN(addr)  (((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
  94
  95 #define page_cache_get(page)            get_page(page)
  96 #define page_cache_release(page)        put_page(page)
  97 void release_pages(struct page **pages, int nr, int cold);
  98
  99 /*
 100  * speculatively take a reference to a page.
 101  * If the page is free (_count == 0), then _count is untouched, and 0
 102  * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
 103  *
 104  * This function must be called inside the same rcu_read_lock() section as has
 105  * been used to lookup the page in the pagecache radix-tree (or page table):
 106  * this allows allocators to use a synchronize_rcu() to stabilize _count.
 107  *
 108  * Unless an RCU grace period has passed, the count of all pages coming out
 109  * of the allocator must be considered unstable. page_count may return higher
 110  * than expected, and put_page must be able to do the right thing when the
 111  * page has been finished with, no matter what it is subsequently allocated
 112  * for (because put_page is what is used here to drop an invalid speculative
 113  * reference).
 114  *
 115  * This is the interesting part of the lockless pagecache (and lockless
 116  * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
 117  * has the following pattern:
 118  * 1. find page in radix tree
 119  * 2. conditionally increment refcount
 120  * 3. check the page is still in pagecache (if no, goto 1)
 121  *
 122  * Remove-side that cares about stability of _count (eg. reclaim) has the
 123  * following (with tree_lock held for write):
 124  * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
 125  * B. remove page from pagecache
 126  * C. free the page
 127  *
 128  * There are 2 critical interleavings that matter:
 129  * - 2 runs before A: in this case, A sees elevated refcount and bails out
 130  * - A runs before 2: in this case, 2 sees zero refcount and retries;
 131  *   subsequently, B will complete and 1 will find no page, causing the
 132  *   lookup to return NULL.
 133  *
 134  * It is possible that between 1 and 2, the page is removed then the exact same
 135  * page is inserted into the same position in pagecache. That's OK: the
 136  * old find_get_page using tree_lock could equally have run before or after
 137  * such a re-insertion, depending on order that locks are granted.
 138  *
 139  * Lookups racing against pagecache insertion isn't a big problem: either 1
 140  * will find the page or it will not. Likewise, the old find_get_page could run
 141  * either before the insertion or afterwards, depending on timing.
 142  */
 143 static inline int page_cache_get_speculative(struct page *page)
 144 {
 145         VM_BUG_ON(in_interrupt());
 146
 147 #if !defined(CONFIG_SMP) && defined(CONFIG_CLASSIC_RCU)
 148 # ifdef CONFIG_PREEMPT
 149         VM_BUG_ON(!in_atomic());
 150 # endif
 151         /*
 152          * Preempt must be disabled here - we rely on rcu_read_lock doing
 153          * this for us.
 154          *
 155          * Pagecache won't be truncated from interrupt context, so if we have
 156          * found a page in the radix tree here, we have pinned its refcount by
 157          * disabling preempt, and hence no need for the "speculative get" that
 158          * SMP requires.
 159          */
 160         VM_BUG_ON(page_count(page) == 0);
 161         atomic_inc(&page->_count);
 162
 163 #else
 164         if (unlikely(!get_page_unless_zero(page))) {
 165                 /*
 166                  * Either the page has been freed, or will be freed.
 167                  * In either case, retry here and the caller should
 168                  * do the right thing (see comments above).
 169                  */
 170                 return 0;
 171         }
 172 #endif
 173         VM_BUG_ON(PageTail(page));
 174
 175         return 1;
 176 }
 177
 178 /*
 179  * Same as above, but add instead of inc (could just be merged)
 180  */
 181 static inline int page_cache_add_speculative(struct page *page, int count)
 182 {
 183         VM_BUG_ON(in_interrupt());
 184
 185 #if !defined(CONFIG_SMP) && defined(CONFIG_CLASSIC_RCU)
 186 # ifdef CONFIG_PREEMPT
 187         VM_BUG_ON(!in_atomic());
 188 # endif
 189         VM_BUG_ON(page_count(page) == 0);
 190         atomic_add(count, &page->_count);
 191
 192 #else
 193         if (unlikely(!atomic_add_unless(&page->_count, count, 0)))
 194                 return 0;
 195 #endif
 196         VM_BUG_ON(PageCompound(page) && page != compound_head(page));
 197
 198         return 1;
 199 }
 200
 201 static inline int page_freeze_refs(struct page *page, int count)
 202 {
 203         return likely(atomic_cmpxchg(&page->_count, count, 0) == count);
 204 }
 205
 206 static inline void page_unfreeze_refs(struct page *page, int count)
 207 {
 208         VM_BUG_ON(page_count(page) != 0);
 209         VM_BUG_ON(count == 0);
 210
 211         atomic_set(&page->_count, count);
 212 }
 213
 214 #ifdef CONFIG_NUMA
 215 extern struct page *__page_cache_alloc(gfp_t gfp);
 216 #else
 217 static inline struct page *__page_cache_alloc(gfp_t gfp)
 218 {
 219         return alloc_pages(gfp, 0);
 220 }
 221 #endif
 222
 223 static inline struct page *page_cache_alloc(struct address_space *x)
 224 {
 225         return __page_cache_alloc(mapping_gfp_mask(x));
 226 }
 227
 228 static inline struct page *page_cache_alloc_cold(struct address_space *x)
 229 {
 230         return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
 231 }
 232
 233 typedef int filler_t(void *, struct page *);
 234
 235 extern struct page * find_get_page(struct address_space *mapping,
 236                                 pgoff_t index);
 237 extern struct page * find_lock_page(struct address_space *mapping,
 238                                 pgoff_t index);
 239 extern struct page * find_or_create_page(struct address_space *mapping,
 240                                 pgoff_t index, gfp_t gfp_mask);
 241 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
 242                         unsigned int nr_pages, struct page **pages);
 243 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
 244                                unsigned int nr_pages, struct page **pages);
 245 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
 246                         int tag, unsigned int nr_pages, struct page **pages);
 247
 248 struct page *grab_cache_page_write_begin(struct address_space *mapping,
 249                         pgoff_t index, unsigned flags);
 250
 251 /*
 252  * Returns locked page at given index in given cache, creating it if needed.
 253  */
 254 static inline struct page *grab_cache_page(struct address_space *mapping,
 255                                                                 pgoff_t index)
 256 {
 257         return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
 258 }
 259
 260 extern struct page * grab_cache_page_nowait(struct address_space *mapping,
 261                                 pgoff_t index);
 262 extern struct page * read_cache_page_async(struct address_space *mapping,
 263                                 pgoff_t index, filler_t *filler,
 264                                 void *data);
 265 extern struct page * read_cache_page(struct address_space *mapping,
 266                                 pgoff_t index, filler_t *filler,
 267                                 void *data);
 268 extern int read_cache_pages(struct address_space *mapping,
 269                 struct list_head *pages, filler_t *filler, void *data);
 270
 271 static inline struct page *read_mapping_page_async(
 272                                                 struct address_space *mapping,
 273                                                      pgoff_t index, void *data)
 274 {
 275         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
 276         return read_cache_page_async(mapping, index, filler, data);
 277 }
 278
 279 static inline struct page *read_mapping_page(struct address_space *mapping,
 280                                              pgoff_t index, void *data)
 281 {
 282         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
 283         return read_cache_page(mapping, index, filler, data);
 284 }
 285
 286 /*
 287  * Return byte-offset into filesystem object for page.
 288  */
 289 static inline loff_t page_offset(struct page *page)
 290 {
 291         return ((loff_t)page->index) << PAGE_CACHE_SHIFT;
 292 }
 293
 294 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
 295                                         unsigned long address)
 296 {
 297         pgoff_t pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
 298         pgoff += vma->vm_pgoff;
 299         return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT);
 300 }
 301
 302 extern void __lock_page(struct page *page);
 303 extern int __lock_page_killable(struct page *page);
 304 extern void __lock_page_nosync(struct page *page);
 305 extern void unlock_page(struct page *page);
 306
 307 static inline void __set_page_locked(struct page *page)
 308 {
 309         __set_bit(PG_locked, &page->flags);
 310 }
 311
 312 static inline void __clear_page_locked(struct page *page)
 313 {
 314         __clear_bit(PG_locked, &page->flags);
 315 }
 316
 317 static inline int trylock_page(struct page *page)
 318 {
 319         return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
 320 }
 321
 322 /*
 323  * lock_page may only be called if we have the page's inode pinned.
 324  */
 325 static inline void lock_page(struct page *page)
 326 {
 327         might_sleep();
 328         if (!trylock_page(page))
 329                 __lock_page(page);
 330 }
 331
 332 /*
 333  * lock_page_killable is like lock_page but can be interrupted by fatal
 334  * signals.  It returns 0 if it locked the page and -EINTR if it was
 335  * killed while waiting.
 336  */
 337 static inline int lock_page_killable(struct page *page)
 338 {
 339         might_sleep();
 340         if (!trylock_page(page))
 341                 return __lock_page_killable(page);
 342         return 0;
 343 }
 344
 345 /*
 346  * lock_page_nosync should only be used if we can't pin the page's inode.
 347  * Doesn't play quite so well with block device plugging.
 348  */
 349 static inline void lock_page_nosync(struct page *page)
 350 {
 351         might_sleep();
 352         if (!trylock_page(page))
 353                 __lock_page_nosync(page);
 354 }
 355
 356 /*
 357  * This is exported only for wait_on_page_locked/wait_on_page_writeback.
 358  * Never use this directly!
 359  */
 360 extern void wait_on_page_bit(struct page *page, int bit_nr);
 361
 362 /*
 363  * Wait for a page to be unlocked.
 364  *
 365  * This must be called with the caller "holding" the page,
 366  * ie with increased "page->count" so that the page won't
 367  * go away during the wait..
 368  */
 369 static inline void wait_on_page_locked(struct page *page)
 370 {
 371         if (PageLocked(page))
 372                 wait_on_page_bit(page, PG_locked);
 373 }
 374
 375 /*
 376  * Wait for a page to complete writeback
 377  */
 378 static inline void wait_on_page_writeback(struct page *page)
 379 {
 380         if (PageWriteback(page))
 381                 wait_on_page_bit(page, PG_writeback);
 382 }
 383
 384 extern void end_page_writeback(struct page *page);
 385
 386 /*
 387  * Add an arbitrary waiter to a page's wait queue
 388  */
 389 extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter);
 390
 391 /*
 392  * Fault a userspace page into pagetables.  Return non-zero on a fault.
 393  *
 394  * This assumes that two userspace pages are always sufficient.  That's
 395  * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
 396  */
 397 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
 398 {
 399         int ret;
 400
 401         if (unlikely(size == 0))
 402                 return 0;
 403
 404         /*
 405          * Writing zeroes into userspace here is OK, because we know that if
 406          * the zero gets there, we'll be overwriting it.
 407          */
 408         ret = __put_user(0, uaddr);
 409         if (ret == 0) {
 410                 char __user *end = uaddr + size - 1;
 411
 412                 /*
 413                  * If the page was already mapped, this will get a cache miss
 414                  * for sure, so try to avoid doing it.
 415                  */
 416                 if (((unsigned long)uaddr & PAGE_MASK) !=
 417                                 ((unsigned long)end & PAGE_MASK))
 418                         ret = __put_user(0, end);
 419         }
 420         return ret;
 421 }
 422
 423 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
 424 {
 425         volatile char c;
 426         int ret;
 427
 428         if (unlikely(size == 0))
 429                 return 0;
 430
 431         ret = __get_user(c, uaddr);
 432         if (ret == 0) {
 433                 const char __user *end = uaddr + size - 1;
 434
 435                 if (((unsigned long)uaddr & PAGE_MASK) !=
 436                                 ((unsigned long)end & PAGE_MASK))
 437                         ret = __get_user(c, end);
 438         }
 439         return ret;
 440 }
 441
 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 remove_from_page_cache(struct page *page);
 447 extern void __remove_from_page_cache(struct page *page);
 448
 449 /*
 450  * Like add_to_page_cache_locked, but used to add newly allocated pages:
 451  * the page is new, so we can just run __set_page_locked() against it.
 452  */
 453 static inline int add_to_page_cache(struct page *page,
 454                 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
 455 {
 456         int error;
 457
 458         __set_page_locked(page);
 459         error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
 460         if (unlikely(error))
 461                 __clear_page_locked(page);
 462         return error;
 463 }
 464
 465 #endif /* _LINUX_PAGEMAP_H */