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 #include <linux/hugetlb_inline.h>
  17
  18 /*
  19  * Bits in mapping->flags.  The lower __GFP_BITS_SHIFT bits are the page
  20  * allocation mode flags.
  21  */
  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() */
  26         AS_UNEVICTABLE  = __GFP_BITS_SHIFT + 3, /* e.g., ramdisk, SHM_LOCK */
  27 };
  28
  29 static inline void mapping_set_error(struct address_space *mapping, int error)
  30 {
  31         if (unlikely(error)) {
  32                 if (error == -ENOSPC)
  33                         set_bit(AS_ENOSPC, &mapping->flags);
  34                 else
  35                         set_bit(AS_EIO, &mapping->flags);
  36         }
  37 }
  38
  39 static inline void mapping_set_unevictable(struct address_space *mapping)
  40 {
  41         set_bit(AS_UNEVICTABLE, &mapping->flags);
  42 }
  43
  44 static inline void mapping_clear_unevictable(struct address_space *mapping)
  45 {
  46         clear_bit(AS_UNEVICTABLE, &mapping->flags);
  47 }
  48
  49 static inline int mapping_unevictable(struct address_space *mapping)
  50 {
  51         if (mapping)
  52                 return test_bit(AS_UNEVICTABLE, &mapping->flags);
  53         return !!mapping;
  54 }
  55
  56 static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
  57 {
  58         return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
  59 }
  60
  61 /*
  62  * This is non-atomic.  Only to be used before the mapping is activated.
  63  * Probably needs a barrier...
  64  */
  65 static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
  66 {
  67         m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
  68                                 (__force unsigned long)mask;
  69 }
  70
  71 /*
  72  * The page cache can done in larger chunks than
  73  * one page, because it allows for more efficient
  74  * throughput (it can then be mapped into user
  75  * space in smaller chunks for same flexibility).
  76  *
  77  * Or rather, it _will_ be done in larger chunks.
  78  */
  79 #define PAGE_CACHE_SHIFT        PAGE_SHIFT
  80 #define PAGE_CACHE_SIZE         PAGE_SIZE
  81 #define PAGE_CACHE_MASK         PAGE_MASK
  82 #define PAGE_CACHE_ALIGN(addr)  (((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
  83
  84 #define page_cache_get(page)            get_page(page)
  85 #define page_cache_release(page)        put_page(page)
  86 void release_pages(struct page **pages, int nr, int cold);
  87
  88 /*
  89  * speculatively take a reference to a page.
  90  * If the page is free (_count == 0), then _count is untouched, and 0
  91  * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
  92  *
  93  * This function must be called inside the same rcu_read_lock() section as has
  94  * been used to lookup the page in the pagecache radix-tree (or page table):
  95  * this allows allocators to use a synchronize_rcu() to stabilize _count.
  96  *
  97  * Unless an RCU grace period has passed, the count of all pages coming out
  98  * of the allocator must be considered unstable. page_count may return higher
  99  * than expected, and put_page must be able to do the right thing when the
 100  * page has been finished with, no matter what it is subsequently allocated
 101  * for (because put_page is what is used here to drop an invalid speculative
 102  * reference).
 103  *
 104  * This is the interesting part of the lockless pagecache (and lockless
 105  * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
 106  * has the following pattern:
 107  * 1. find page in radix tree
 108  * 2. conditionally increment refcount
 109  * 3. check the page is still in pagecache (if no, goto 1)
 110  *
 111  * Remove-side that cares about stability of _count (eg. reclaim) has the
 112  * following (with tree_lock held for write):
 113  * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
 114  * B. remove page from pagecache
 115  * C. free the page
 116  *
 117  * There are 2 critical interleavings that matter:
 118  * - 2 runs before A: in this case, A sees elevated refcount and bails out
 119  * - A runs before 2: in this case, 2 sees zero refcount and retries;
 120  *   subsequently, B will complete and 1 will find no page, causing the
 121  *   lookup to return NULL.
 122  *
 123  * It is possible that between 1 and 2, the page is removed then the exact same
 124  * page is inserted into the same position in pagecache. That's OK: the
 125  * old find_get_page using tree_lock could equally have run before or after
 126  * such a re-insertion, depending on order that locks are granted.
 127  *
 128  * Lookups racing against pagecache insertion isn't a big problem: either 1
 129  * will find the page or it will not. Likewise, the old find_get_page could run
 130  * either before the insertion or afterwards, depending on timing.
 131  */
 132 static inline int page_cache_get_speculative(struct page *page)
 133 {
 134         VM_BUG_ON(in_interrupt());
 135
 136 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
 137 # ifdef CONFIG_PREEMPT
 138         VM_BUG_ON(!in_atomic());
 139 # endif
 140         /*
 141          * Preempt must be disabled here - we rely on rcu_read_lock doing
 142          * this for us.
 143          *
 144          * Pagecache won't be truncated from interrupt context, so if we have
 145          * found a page in the radix tree here, we have pinned its refcount by
 146          * disabling preempt, and hence no need for the "speculative get" that
 147          * SMP requires.
 148          */
 149         VM_BUG_ON(page_count(page) == 0);
 150         atomic_inc(&page->_count);
 151
 152 #else
 153         if (unlikely(!get_page_unless_zero(page))) {
 154                 /*
 155                  * Either the page has been freed, or will be freed.
 156                  * In either case, retry here and the caller should
 157                  * do the right thing (see comments above).
 158                  */
 159                 return 0;
 160         }
 161 #endif
 162         VM_BUG_ON(PageTail(page));
 163
 164         return 1;
 165 }
 166
 167 /*
 168  * Same as above, but add instead of inc (could just be merged)
 169  */
 170 static inline int page_cache_add_speculative(struct page *page, int count)
 171 {
 172         VM_BUG_ON(in_interrupt());
 173
 174 #if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
 175 # ifdef CONFIG_PREEMPT
 176         VM_BUG_ON(!in_atomic());
 177 # endif
 178         VM_BUG_ON(page_count(page) == 0);
 179         atomic_add(count, &page->_count);
 180
 181 #else
 182         if (unlikely(!atomic_add_unless(&page->_count, count, 0)))
 183                 return 0;
 184 #endif
 185         VM_BUG_ON(PageCompound(page) && page != compound_head(page));
 186
 187         return 1;
 188 }
 189
 190 static inline int page_freeze_refs(struct page *page, int count)
 191 {
 192         return likely(atomic_cmpxchg(&page->_count, count, 0) == count);
 193 }
 194
 195 static inline void page_unfreeze_refs(struct page *page, int count)
 196 {
 197         VM_BUG_ON(page_count(page) != 0);
 198         VM_BUG_ON(count == 0);
 199
 200         atomic_set(&page->_count, count);
 201 }
 202
 203 #ifdef CONFIG_NUMA
 204 extern struct page *__page_cache_alloc(gfp_t gfp);
 205 #else
 206 static inline struct page *__page_cache_alloc(gfp_t gfp)
 207 {
 208         return alloc_pages(gfp, 0);
 209 }
 210 #endif
 211
 212 static inline struct page *page_cache_alloc(struct address_space *x)
 213 {
 214         return __page_cache_alloc(mapping_gfp_mask(x));
 215 }
 216
 217 static inline struct page *page_cache_alloc_cold(struct address_space *x)
 218 {
 219         return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
 220 }
 221
 222 typedef int filler_t(void *, struct page *);
 223
 224 extern struct page * find_get_page(struct address_space *mapping,
 225                                 pgoff_t index);
 226 extern struct page * find_lock_page(struct address_space *mapping,
 227                                 pgoff_t index);
 228 extern struct page * find_or_create_page(struct address_space *mapping,
 229                                 pgoff_t index, gfp_t gfp_mask);
 230 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
 231                         unsigned int nr_pages, struct page **pages);
 232 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
 233                                unsigned int nr_pages, struct page **pages);
 234 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
 235                         int tag, unsigned int nr_pages, struct page **pages);
 236
 237 struct page *grab_cache_page_write_begin(struct address_space *mapping,
 238                         pgoff_t index, unsigned flags);
 239
 240 /*
 241  * Returns locked page at given index in given cache, creating it if needed.
 242  */
 243 static inline struct page *grab_cache_page(struct address_space *mapping,
 244                                                                 pgoff_t index)
 245 {
 246         return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
 247 }
 248
 249 extern struct page * grab_cache_page_nowait(struct address_space *mapping,
 250                                 pgoff_t index);
 251 extern struct page * read_cache_page_async(struct address_space *mapping,
 252                                 pgoff_t index, filler_t *filler,
 253                                 void *data);
 254 extern struct page * read_cache_page(struct address_space *mapping,
 255                                 pgoff_t index, filler_t *filler,
 256                                 void *data);
 257 extern struct page * read_cache_page_gfp(struct address_space *mapping,
 258                                 pgoff_t index, gfp_t gfp_mask);
 259 extern int read_cache_pages(struct address_space *mapping,
 260                 struct list_head *pages, filler_t *filler, void *data);
 261
 262 static inline struct page *read_mapping_page_async(
 263                                                 struct address_space *mapping,
 264                                                      pgoff_t index, void *data)
 265 {
 266         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
 267         return read_cache_page_async(mapping, index, filler, data);
 268 }
 269
 270 static inline struct page *read_mapping_page(struct address_space *mapping,
 271                                              pgoff_t index, void *data)
 272 {
 273         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
 274         return read_cache_page(mapping, index, filler, data);
 275 }
 276
 277 /*
 278  * Return byte-offset into filesystem object for page.
 279  */
 280 static inline loff_t page_offset(struct page *page)
 281 {
 282         return ((loff_t)page->index) << PAGE_CACHE_SHIFT;
 283 }
 284
 285 extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
 286                                      unsigned long address);
 287
 288 static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
 289                                         unsigned long address)
 290 {
 291         pgoff_t pgoff;
 292         if (unlikely(is_vm_hugetlb_page(vma)))
 293                 return linear_hugepage_index(vma, address);
 294         pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
 295         pgoff += vma->vm_pgoff;
 296         return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT);
 297 }
 298
 299 extern void __lock_page(struct page *page);
 300 extern int __lock_page_killable(struct page *page);
 301 extern void __lock_page_nosync(struct page *page);
 302 extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
 303                                 unsigned int flags);
 304 extern void unlock_page(struct page *page);
 305
 306 static inline void __set_page_locked(struct page *page)
 307 {
 308         __set_bit(PG_locked, &page->flags);
 309 }
 310
 311 static inline void __clear_page_locked(struct page *page)
 312 {
 313         __clear_bit(PG_locked, &page->flags);
 314 }
 315
 316 static inline int trylock_page(struct page *page)
 317 {
 318         return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
 319 }
 320
 321 /*
 322  * lock_page may only be called if we have the page's inode pinned.
 323  */
 324 static inline void lock_page(struct page *page)
 325 {
 326         might_sleep();
 327         if (!trylock_page(page))
 328                 __lock_page(page);
 329 }
 330
 331 /*
 332  * lock_page_killable is like lock_page but can be interrupted by fatal
 333  * signals.  It returns 0 if it locked the page and -EINTR if it was
 334  * killed while waiting.
 335  */
 336 static inline int lock_page_killable(struct page *page)
 337 {
 338         might_sleep();
 339         if (!trylock_page(page))
 340                 return __lock_page_killable(page);
 341         return 0;
 342 }
 343
 344 /*
 345  * lock_page_nosync should only be used if we can't pin the page's inode.
 346  * Doesn't play quite so well with block device plugging.
 347  */
 348 static inline void lock_page_nosync(struct page *page)
 349 {
 350         might_sleep();
 351         if (!trylock_page(page))
 352                 __lock_page_nosync(page);
 353 }
 354
 355 /*
 356  * lock_page_or_retry - Lock the page, unless this would block and the
 357  * caller indicated that it can handle a retry.
 358  */
 359 static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
 360                                      unsigned int flags)
 361 {
 362         might_sleep();
 363         return trylock_page(page) || __lock_page_or_retry(page, mm, flags);
 364 }
 365
 366 /*
 367  * This is exported only for wait_on_page_locked/wait_on_page_writeback.
 368  * Never use this directly!
 369  */
 370 extern void wait_on_page_bit(struct page *page, int bit_nr);
 371
 372 /*
 373  * Wait for a page to be unlocked.
 374  *
 375  * This must be called with the caller "holding" the page,
 376  * ie with increased "page->count" so that the page won't
 377  * go away during the wait..
 378  */
 379 static inline void wait_on_page_locked(struct page *page)
 380 {
 381         if (PageLocked(page))
 382                 wait_on_page_bit(page, PG_locked);
 383 }
 384
 385 /*
 386  * Wait for a page to complete writeback
 387  */
 388 static inline void wait_on_page_writeback(struct page *page)
 389 {
 390         if (PageWriteback(page))
 391                 wait_on_page_bit(page, PG_writeback);
 392 }
 393
 394 extern void end_page_writeback(struct page *page);
 395
 396 /*
 397  * Add an arbitrary waiter to a page's wait queue
 398  */
 399 extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter);
 400
 401 /*
 402  * Fault a userspace page into pagetables.  Return non-zero on a fault.
 403  *
 404  * This assumes that two userspace pages are always sufficient.  That's
 405  * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
 406  */
 407 static inline int fault_in_pages_writeable(char __user *uaddr, int size)
 408 {
 409         int ret;
 410
 411         if (unlikely(size == 0))
 412                 return 0;
 413
 414         /*
 415          * Writing zeroes into userspace here is OK, because we know that if
 416          * the zero gets there, we'll be overwriting it.
 417          */
 418         ret = __put_user(0, uaddr);
 419         if (ret == 0) {
 420                 char __user *end = uaddr + size - 1;
 421
 422                 /*
 423                  * If the page was already mapped, this will get a cache miss
 424                  * for sure, so try to avoid doing it.
 425                  */
 426                 if (((unsigned long)uaddr & PAGE_MASK) !=
 427                                 ((unsigned long)end & PAGE_MASK))
 428                         ret = __put_user(0, end);
 429         }
 430         return ret;
 431 }
 432
 433 static inline int fault_in_pages_readable(const char __user *uaddr, int size)
 434 {
 435         volatile char c;
 436         int ret;
 437
 438         if (unlikely(size == 0))
 439                 return 0;
 440
 441         ret = __get_user(c, uaddr);
 442         if (ret == 0) {
 443                 const char __user *end = uaddr + size - 1;
 444
 445                 if (((unsigned long)uaddr & PAGE_MASK) !=
 446                                 ((unsigned long)end & PAGE_MASK)) {
 447                         ret = __get_user(c, end);
 448                         (void)c;
 449                 }
 450         }
 451         return ret;
 452 }
 453
 454 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 455                                 pgoff_t index, gfp_t gfp_mask);
 456 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 457                                 pgoff_t index, gfp_t gfp_mask);
 458 extern void remove_from_page_cache(struct page *page);
 459 extern void __remove_from_page_cache(struct page *page);
 460
 461 /*
 462  * Like add_to_page_cache_locked, but used to add newly allocated pages:
 463  * the page is new, so we can just run __set_page_locked() against it.
 464  */
 465 static inline int add_to_page_cache(struct page *page,
 466                 struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
 467 {
 468         int error;
 469
 470         __set_page_locked(page);
 471         error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
 472         if (unlikely(error))
 473                 __clear_page_locked(page);
 474         return error;
 475 }
 476
 477 #endif /* _LINUX_PAGEMAP_H */