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 #define AS_EIO          (__GFP_BITS_SHIFT + 0)  /* IO error on async write */
  22 #define AS_ENOSPC       (__GFP_BITS_SHIFT + 1)  /* ENOSPC on async write */
  23 #define AS_MM_ALL_LOCKS (__GFP_BITS_SHIFT + 2)  /* under mm_take_all_locks() */
  24
  25 static inline void mapping_set_error(struct address_space *mapping, int error)
  26 {
  27         if (unlikely(error)) {
  28                 if (error == -ENOSPC)
  29                         set_bit(AS_ENOSPC, &mapping->flags);
  30                 else
  31                         set_bit(AS_EIO, &mapping->flags);
  32         }
  33 }
  34
  35 #ifdef CONFIG_UNEVICTABLE_LRU
  36 #define AS_UNEVICTABLE  (__GFP_BITS_SHIFT + 2)  /* e.g., ramdisk, SHM_LOCK */
  37
  38 static inline void mapping_set_unevictable(struct address_space *mapping)
  39 {
  40         set_bit(AS_UNEVICTABLE, &mapping->flags);
  41 }
  42
  43 static inline void mapping_clear_unevictable(struct address_space *mapping)
  44 {
  45         clear_bit(AS_UNEVICTABLE, &mapping->flags);
  46 }
  47
  48 static inline int mapping_unevictable(struct address_space *mapping)
  49 {
  50         if (likely(mapping))
  51                 return test_bit(AS_UNEVICTABLE, &mapping->flags);
  52         return !!mapping;
  53 }
  54 #else
  55 static inline void mapping_set_unevictable(struct address_space *mapping) { }
  56 static inline void mapping_clear_unevictable(struct address_space *mapping) { }
  57 static inline int mapping_unevictable(struct address_space *mapping)
  58 {
  59         return 0;
  60 }
  61 #endif
  62
  63 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
  64 {
  65         return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
  66 }
  67
  68 /*
  69  * This is non-atomic.  Only to be used before the mapping is activated.
  70  * Probably needs a barrier...
  71  */
  72 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
  73 {
  74         m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
  75                                 (__force unsigned long)mask;
  76 }
  77
  78 /*
  79  * The page cache can done in larger chunks than
  80  * one page, because it allows for more efficient
  81  * throughput (it can then be mapped into user
  82  * space in smaller chunks for same flexibility).
  83  *
  84  * Or rather, it _will_ be done in larger chunks.
  85  */
  86 #define PAGE_CACHE_SHIFT        PAGE_SHIFT
  87 #define PAGE_CACHE_SIZE         PAGE_SIZE
  88 #define PAGE_CACHE_MASK         PAGE_MASK
  89 #define PAGE_CACHE_ALIGN(addr)  (((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
  90
  91 #define page_cache_get(page)            get_page(page)
  92 #define page_cache_release(page)        put_page(page)
  93 void release_pages(struct page **pages, int nr, int cold);
  94
  95 /*
  96  * speculatively take a reference to a page.
  97  * If the page is free (_count == 0), then _count is untouched, and 0
  98  * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
  99  *
 100  * This function must be called inside the same rcu_read_lock() section as has
 101  * been used to lookup the page in the pagecache radix-tree (or page table):
 102  * this allows allocators to use a synchronize_rcu() to stabilize _count.
 103  *
 104  * Unless an RCU grace period has passed, the count of all pages coming out
 105  * of the allocator must be considered unstable. page_count may return higher
 106  * than expected, and put_page must be able to do the right thing when the
 107  * page has been finished with, no matter what it is subsequently allocated
 108  * for (because put_page is what is used here to drop an invalid speculative
 109  * reference).
 110  *
 111  * This is the interesting part of the lockless pagecache (and lockless
 112  * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
 113  * has the following pattern:
 114  * 1. find page in radix tree
 115  * 2. conditionally increment refcount
 116  * 3. check the page is still in pagecache (if no, goto 1)
 117  *
 118  * Remove-side that cares about stability of _count (eg. reclaim) has the
 119  * following (with tree_lock held for write):
 120  * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
 121  * B. remove page from pagecache
 122  * C. free the page
 123  *
 124  * There are 2 critical interleavings that matter:
 125  * - 2 runs before A: in this case, A sees elevated refcount and bails out
 126  * - A runs before 2: in this case, 2 sees zero refcount and retries;
 127  *   subsequently, B will complete and 1 will find no page, causing the
 128  *   lookup to return NULL.
 129  *
 130  * It is possible that between 1 and 2, the page is removed then the exact same
 131  * page is inserted into the same position in pagecache. That's OK: the
 132  * old find_get_page using tree_lock could equally have run before or after
 133  * such a re-insertion, depending on order that locks are granted.
 134  *
 135  * Lookups racing against pagecache insertion isn't a big problem: either 1
 136  * will find the page or it will not. Likewise, the old find_get_page could run
 137  * either before the insertion or afterwards, depending on timing.
 138  */
 139 static inline int page_cache_get_speculative(struct page *page)
 140 {
 141         VM_BUG_ON(in_interrupt());
 142
 143 #if !defined(CONFIG_SMP) && defined(CONFIG_CLASSIC_RCU)
 144 # ifdef CONFIG_PREEMPT
 145         VM_BUG_ON(!in_atomic());
 146 # endif
 147         /*
 148          * Preempt must be disabled here - we rely on rcu_read_lock doing
 149          * this for us.
 150          *
 151          * Pagecache won't be truncated from interrupt context, so if we have
 152          * found a page in the radix tree here, we have pinned its refcount by
 153          * disabling preempt, and hence no need for the "speculative get" that
 154          * SMP requires.
 155          */
 156         VM_BUG_ON(page_count(page) == 0);
 157         atomic_inc(&page->_count);
 158
 159 #else
 160         if (unlikely(!get_page_unless_zero(page))) {
 161                 /*
 162                  * Either the page has been freed, or will be freed.
 163                  * In either case, retry here and the caller should
 164                  * do the right thing (see comments above).
 165                  */
 166                 return 0;
 167         }
 168 #endif
 169         VM_BUG_ON(PageTail(page));
 170
 171         return 1;
 172 }
 173
 174 /*
 175  * Same as above, but add instead of inc (could just be merged)
 176  */
 177 static inline int page_cache_add_speculative(struct page *page, int count)
 178 {
 179         VM_BUG_ON(in_interrupt());
 180
 181 #if !defined(CONFIG_SMP) && defined(CONFIG_CLASSIC_RCU)
 182 # ifdef CONFIG_PREEMPT
 183         VM_BUG_ON(!in_atomic());
 184 # endif
 185         VM_BUG_ON(page_count(page) == 0);
 186         atomic_add(count, &page->_count);
 187
 188 #else
 189         if (unlikely(!atomic_add_unless(&page->_count, count, 0)))
 190                 return 0;
 191 #endif
 192         VM_BUG_ON(PageCompound(page) && page != compound_head(page));
 193
 194         return 1;
 195 }
 196
 197 static inline int page_freeze_refs(struct page *page, int count)
 198 {
 199         return likely(atomic_cmpxchg(&page->_count, count, 0) == count);
 200 }
 201
 202 static inline void page_unfreeze_refs(struct page *page, int count)
 203 {
 204         VM_BUG_ON(page_count(page) != 0);
 205         VM_BUG_ON(count == 0);
 206
 207         atomic_set(&page->_count, count);
 208 }
 209
 210 #ifdef CONFIG_NUMA
 211 extern struct page *__page_cache_alloc(gfp_t gfp);
 212 #else
 213 static inline struct page *__page_cache_alloc(gfp_t gfp)
 214 {
 215         return alloc_pages(gfp, 0);
 216 }
 217 #endif
 218
 219 static inline struct page *page_cache_alloc(struct address_space *x)
 220 {
 221         return __page_cache_alloc(mapping_gfp_mask(x));
 222 }
 223
 224 static inline struct page *page_cache_alloc_cold(struct address_space *x)
 225 {
 226         return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
 227 }
 228
 229 typedef int filler_t(void *, struct page *);
 230
 231 extern struct page * find_get_page(struct address_space *mapping,
 232                                 pgoff_t index);
 233 extern struct page * find_lock_page(struct address_space *mapping,
 234                                 pgoff_t index);
 235 extern struct page * find_or_create_page(struct address_space *mapping,
 236                                 pgoff_t index, gfp_t gfp_mask);
 237 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
 238                         unsigned int nr_pages, struct page **pages);
 239 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
 240                                unsigned int nr_pages, struct page **pages);
 241 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
 242                         int tag, unsigned int nr_pages, struct page **pages);
 243
 244 struct page *grab_cache_page_write_begin(struct address_space *mapping,
 245                         pgoff_t index, unsigned flags);
 246
 247 /*
 248  * Returns locked page at given index in given cache, creating it if needed.
 249  */
 250 static inline struct page *grab_cache_page(struct address_space *mapping,
 251                                                                 pgoff_t index)
 252 {
 253         return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
 254 }
 255
 256 extern struct page * grab_cache_page_nowait(struct address_space *mapping,
 257                                 pgoff_t index);
 258 extern struct page * read_cache_page_async(struct address_space *mapping,
 259                                 pgoff_t index, filler_t *filler,
 260                                 void *data);
 261 extern struct page * read_cache_page(struct address_space *mapping,
 262                                 pgoff_t index, filler_t *filler,
 263                                 void *data);
 264 extern int read_cache_pages(struct address_space *mapping,
 265                 struct list_head *pages, filler_t *filler, void *data);
 266
 267 static inline struct page *read_mapping_page_async(
 268                                                 struct address_space *mapping,
 269                                                      pgoff_t index, void *data)
 270 {
 271         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
 272         return read_cache_page_async(mapping, index, filler, data);
 273 }
 274
 275 static inline struct page *read_mapping_page(struct address_space *mapping,
 276                                              pgoff_t index, void *data)
 277 {
 278         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
 279         return read_cache_page(mapping, index, filler, data);
 280 }
 281
 282 /*
 283  * Return byte-offset into filesystem object for page.
 284  */
 285 static inline loff_t page_offset(struct page *page)
 286 {
 287         return ((loff_t)page->index) << PAGE_CACHE_SHIFT;
 288 }
 289
 290 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
 291                                         unsigned long address)
 292 {
 293         pgoff_t pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
 294         pgoff += vma->vm_pgoff;
 295         return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT);
 296 }
 297
 298 extern void __lock_page(struct page *page);
 299 extern int __lock_page_killable(struct page *page);
 300 extern void __lock_page_nosync(struct page *page);
 301 extern void unlock_page(struct page *page);
 302
 303 static inline void __set_page_locked(struct page *page)
 304 {
 305         __set_bit(PG_locked, &page->flags);
 306 }
 307
 308 static inline void __clear_page_locked(struct page *page)
 309 {
 310         __clear_bit(PG_locked, &page->flags);
 311 }
 312
 313 static inline int trylock_page(struct page *page)
 314 {
 315         return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
 316 }
 317
 318 /*
 319  * lock_page may only be called if we have the page's inode pinned.
 320  */
 321 static inline void lock_page(struct page *page)
 322 {
 323         might_sleep();
 324         if (!trylock_page(page))
 325                 __lock_page(page);
 326 }
 327
 328 /*
 329  * lock_page_killable is like lock_page but can be interrupted by fatal
 330  * signals.  It returns 0 if it locked the page and -EINTR if it was
 331  * killed while waiting.
 332  */
 333 static inline int lock_page_killable(struct page *page)
 334 {
 335         might_sleep();
 336         if (!trylock_page(page))
 337                 return __lock_page_killable(page);
 338         return 0;
 339 }
 340
 341 /*
 342  * lock_page_nosync should only be used if we can't pin the page's inode.
 343  * Doesn't play quite so well with block device plugging.
 344  */
 345 static inline void lock_page_nosync(struct page *page)
 346 {
 347         might_sleep();
 348         if (!trylock_page(page))
 349                 __lock_page_nosync(page);
 350 }
 351
 352 /*
 353  * This is exported only for wait_on_page_locked/wait_on_page_writeback.
 354  * Never use this directly!
 355  */
 356 extern void wait_on_page_bit(struct page *page, int bit_nr);
 357
 358 /*
 359  * Wait for a page to be unlocked.
 360  *
 361  * This must be called with the caller "holding" the page,
 362  * ie with increased "page->count" so that the page won't
 363  * go away during the wait..
 364  */
 365 static inline void wait_on_page_locked(struct page *page)
 366 {
 367         if (PageLocked(page))
 368                 wait_on_page_bit(page, PG_locked);
 369 }
 370
 371 /*
 372  * Wait for a page to complete writeback
 373  */
 374 static inline void wait_on_page_writeback(struct page *page)
 375 {
 376         if (PageWriteback(page))
 377                 wait_on_page_bit(page, PG_writeback);
 378 }
 379
 380 extern void end_page_writeback(struct page *page);
 381
 382 /*
 383  * Fault a userspace page into pagetables.  Return non-zero on a fault.
 384  *
 385  * This assumes that two userspace pages are always sufficient.  That's
 386  * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
 387  */
 388 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
 389 {
 390         int ret;
 391
 392         if (unlikely(size == 0))
 393                 return 0;
 394
 395         /*
 396          * Writing zeroes into userspace here is OK, because we know that if
 397          * the zero gets there, we'll be overwriting it.
 398          */
 399         ret = __put_user(0, uaddr);
 400         if (ret == 0) {
 401                 char __user *end = uaddr + size - 1;
 402
 403                 /*
 404                  * If the page was already mapped, this will get a cache miss
 405                  * for sure, so try to avoid doing it.
 406                  */
 407                 if (((unsigned long)uaddr & PAGE_MASK) !=
 408                                 ((unsigned long)end & PAGE_MASK))
 409                         ret = __put_user(0, end);
 410         }
 411         return ret;
 412 }
 413
 414 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
 415 {
 416         volatile char c;
 417         int ret;
 418
 419         if (unlikely(size == 0))
 420                 return 0;
 421
 422         ret = __get_user(c, uaddr);
 423         if (ret == 0) {
 424                 const char __user *end = uaddr + size - 1;
 425
 426                 if (((unsigned long)uaddr & PAGE_MASK) !=
 427                                 ((unsigned long)end & PAGE_MASK))
 428                         ret = __get_user(c, end);
 429         }
 430         return ret;
 431 }
 432
 433 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 434                                 pgoff_t index, gfp_t gfp_mask);
 435 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 436                                 pgoff_t index, gfp_t gfp_mask);
 437 extern void remove_from_page_cache(struct page *page);
 438 extern void __remove_from_page_cache(struct page *page);
 439
 440 /*
 441  * Like add_to_page_cache_locked, but used to add newly allocated pages:
 442  * the page is new, so we can just run __set_page_locked() against it.
 443  */
 444 static inline int add_to_page_cache(struct page *page,
 445                 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
 446 {
 447         int error;
 448
 449         __set_page_locked(page);
 450         error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
 451         if (unlikely(error))
 452                 __clear_page_locked(page);
 453         return error;
 454 }
 455
 456 #endif /* _LINUX_PAGEMAP_H */