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