include/linux/radix-tree.h

   1 /*
   2  * Copyright (C) 2001 Momchil Velikov
   3  * Portions Copyright (C) 2001 Christoph Hellwig
   4  * Copyright (C) 2006 Nick Piggin
   5  * Copyright (C) 2012 Konstantin Khlebnikov
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License as
   9  * published by the Free Software Foundation; either version 2, or (at
  10  * your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21 #ifndef _LINUX_RADIX_TREE_H
  22 #define _LINUX_RADIX_TREE_H
  23
  24 #include <linux/preempt.h>
  25 #include <linux/types.h>
  26 #include <linux/bug.h>
  27 #include <linux/kernel.h>
  28 #include <linux/rcupdate.h>
  29
  30 /*
  31  * An indirect pointer (root->rnode pointing to a radix_tree_node, rather
  32  * than a data item) is signalled by the low bit set in the root->rnode
  33  * pointer.
  34  *
  35  * In this case root->height is > 0, but the indirect pointer tests are
  36  * needed for RCU lookups (because root->height is unreliable). The only
  37  * time callers need worry about this is when doing a lookup_slot under
  38  * RCU.
  39  *
  40  * Indirect pointer in fact is also used to tag the last pointer of a node
  41  * when it is shrunk, before we rcu free the node. See shrink code for
  42  * details.
  43  */
  44 #define RADIX_TREE_INDIRECT_PTR         1
  45 /*
  46  * A common use of the radix tree is to store pointers to struct pages;
  47  * but shmem/tmpfs needs also to store swap entries in the same tree:
  48  * those are marked as exceptional entries to distinguish them.
  49  * EXCEPTIONAL_ENTRY tests the bit, EXCEPTIONAL_SHIFT shifts content past it.
  50  */
  51 #define RADIX_TREE_EXCEPTIONAL_ENTRY    2
  52 #define RADIX_TREE_EXCEPTIONAL_SHIFT    2
  53
  54 static inline int radix_tree_is_indirect_ptr(void *ptr)
  55 {
  56         return (int)((unsigned long)ptr & RADIX_TREE_INDIRECT_PTR);
  57 }
  58
  59 /*** radix-tree API starts here ***/
  60
  61 #define RADIX_TREE_MAX_TAGS 3
  62
  63 /* root tags are stored in gfp_mask, shifted by __GFP_BITS_SHIFT */
  64 struct radix_tree_root {
  65         unsigned int            height;
  66         gfp_t                   gfp_mask;
  67         struct radix_tree_node  __rcu *rnode;
  68 };
  69
  70 #define RADIX_TREE_INIT(mask)   {                                       \
  71         .height = 0,                                                    \
  72         .gfp_mask = (mask),                                             \
  73         .rnode = NULL,                                                  \
  74 }
  75
  76 #define RADIX_TREE(name, mask) \
  77         struct radix_tree_root name = RADIX_TREE_INIT(mask)
  78
  79 #define INIT_RADIX_TREE(root, mask)                                     \
  80 do {                                                                    \
  81         (root)->height = 0;                                             \
  82         (root)->gfp_mask = (mask);                                      \
  83         (root)->rnode = NULL;                                           \
  84 } while (0)
  85
  86 /**
  87  * Radix-tree synchronization
  88  *
  89  * The radix-tree API requires that users provide all synchronisation (with
  90  * specific exceptions, noted below).
  91  *
  92  * Synchronization of access to the data items being stored in the tree, and
  93  * management of their lifetimes must be completely managed by API users.
  94  *
  95  * For API usage, in general,
  96  * - any function _modifying_ the tree or tags (inserting or deleting
  97  *   items, setting or clearing tags) must exclude other modifications, and
  98  *   exclude any functions reading the tree.
  99  * - any function _reading_ the tree or tags (looking up items or tags,
 100  *   gang lookups) must exclude modifications to the tree, but may occur
 101  *   concurrently with other readers.
 102  *
 103  * The notable exceptions to this rule are the following functions:
 104  * radix_tree_lookup
 105  * radix_tree_lookup_slot
 106  * radix_tree_tag_get
 107  * radix_tree_gang_lookup
 108  * radix_tree_gang_lookup_slot
 109  * radix_tree_gang_lookup_tag
 110  * radix_tree_gang_lookup_tag_slot
 111  * radix_tree_tagged
 112  *
 113  * The first 7 functions are able to be called locklessly, using RCU. The
 114  * caller must ensure calls to these functions are made within rcu_read_lock()
 115  * regions. Other readers (lock-free or otherwise) and modifications may be
 116  * running concurrently.
 117  *
 118  * It is still required that the caller manage the synchronization and lifetimes
 119  * of the items. So if RCU lock-free lookups are used, typically this would mean
 120  * that the items have their own locks, or are amenable to lock-free access; and
 121  * that the items are freed by RCU (or only freed after having been deleted from
 122  * the radix tree *and* a synchronize_rcu() grace period).
 123  *
 124  * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
 125  * access to data items when inserting into or looking up from the radix tree)
 126  *
 127  * Note that the value returned by radix_tree_tag_get() may not be relied upon
 128  * if only the RCU read lock is held.  Functions to set/clear tags and to
 129  * delete nodes running concurrently with it may affect its result such that
 130  * two consecutive reads in the same locked section may return different
 131  * values.  If reliability is required, modification functions must also be
 132  * excluded from concurrency.
 133  *
 134  * radix_tree_tagged is able to be called without locking or RCU.
 135  */
 136
 137 /**
 138  * radix_tree_deref_slot        - dereference a slot
 139  * @pslot:      pointer to slot, returned by radix_tree_lookup_slot
 140  * Returns:     item that was stored in that slot with any direct pointer flag
 141  *              removed.
 142  *
 143  * For use with radix_tree_lookup_slot().  Caller must hold tree at least read
 144  * locked across slot lookup and dereference. Not required if write lock is
 145  * held (ie. items cannot be concurrently inserted).
 146  *
 147  * radix_tree_deref_retry must be used to confirm validity of the pointer if
 148  * only the read lock is held.
 149  */
 150 static inline void *radix_tree_deref_slot(void **pslot)
 151 {
 152         return rcu_dereference(*pslot);
 153 }
 154
 155 /**
 156  * radix_tree_deref_slot_protected      - dereference a slot without RCU lock but with tree lock held
 157  * @pslot:      pointer to slot, returned by radix_tree_lookup_slot
 158  * Returns:     item that was stored in that slot with any direct pointer flag
 159  *              removed.
 160  *
 161  * Similar to radix_tree_deref_slot but only used during migration when a pages
 162  * mapping is being moved. The caller does not hold the RCU read lock but it
 163  * must hold the tree lock to prevent parallel updates.
 164  */
 165 static inline void *radix_tree_deref_slot_protected(void **pslot,
 166                                                         spinlock_t *treelock)
 167 {
 168         return rcu_dereference_protected(*pslot, lockdep_is_held(treelock));
 169 }
 170
 171 /**
 172  * radix_tree_deref_retry       - check radix_tree_deref_slot
 173  * @arg:        pointer returned by radix_tree_deref_slot
 174  * Returns:     0 if retry is not required, otherwise retry is required
 175  *
 176  * radix_tree_deref_retry must be used with radix_tree_deref_slot.
 177  */
 178 static inline int radix_tree_deref_retry(void *arg)
 179 {
 180         return unlikely((unsigned long)arg & RADIX_TREE_INDIRECT_PTR);
 181 }
 182
 183 /**
 184  * radix_tree_exceptional_entry - radix_tree_deref_slot gave exceptional entry?
 185  * @arg:        value returned by radix_tree_deref_slot
 186  * Returns:     0 if well-aligned pointer, non-0 if exceptional entry.
 187  */
 188 static inline int radix_tree_exceptional_entry(void *arg)
 189 {
 190         /* Not unlikely because radix_tree_exception often tested first */
 191         return (unsigned long)arg & RADIX_TREE_EXCEPTIONAL_ENTRY;
 192 }
 193
 194 /**
 195  * radix_tree_exception - radix_tree_deref_slot returned either exception?
 196  * @arg:        value returned by radix_tree_deref_slot
 197  * Returns:     0 if well-aligned pointer, non-0 if either kind of exception.
 198  */
 199 static inline int radix_tree_exception(void *arg)
 200 {
 201         return unlikely((unsigned long)arg &
 202                 (RADIX_TREE_INDIRECT_PTR | RADIX_TREE_EXCEPTIONAL_ENTRY));
 203 }
 204
 205 /**
 206  * radix_tree_replace_slot      - replace item in a slot
 207  * @pslot:      pointer to slot, returned by radix_tree_lookup_slot
 208  * @item:       new item to store in the slot.
 209  *
 210  * For use with radix_tree_lookup_slot().  Caller must hold tree write locked
 211  * across slot lookup and replacement.
 212  */
 213 static inline void radix_tree_replace_slot(void **pslot, void *item)
 214 {
 215         BUG_ON(radix_tree_is_indirect_ptr(item));
 216         rcu_assign_pointer(*pslot, item);
 217 }
 218
 219 int radix_tree_insert(struct radix_tree_root *, unsigned long, void *);
 220 void *radix_tree_lookup(struct radix_tree_root *, unsigned long);
 221 void **radix_tree_lookup_slot(struct radix_tree_root *, unsigned long);
 222 void *radix_tree_delete(struct radix_tree_root *, unsigned long);
 223 unsigned int
 224 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
 225                         unsigned long first_index, unsigned int max_items);
 226 unsigned int radix_tree_gang_lookup_slot(struct radix_tree_root *root,
 227                         void ***results, unsigned long *indices,
 228                         unsigned long first_index, unsigned int max_items);
 229 unsigned long radix_tree_next_hole(struct radix_tree_root *root,
 230                                 unsigned long index, unsigned long max_scan);
 231 unsigned long radix_tree_prev_hole(struct radix_tree_root *root,
 232                                 unsigned long index, unsigned long max_scan);
 233 int radix_tree_preload(gfp_t gfp_mask);
 234 void radix_tree_init(void);
 235 void *radix_tree_tag_set(struct radix_tree_root *root,
 236                         unsigned long index, unsigned int tag);
 237 void *radix_tree_tag_clear(struct radix_tree_root *root,
 238                         unsigned long index, unsigned int tag);
 239 int radix_tree_tag_get(struct radix_tree_root *root,
 240                         unsigned long index, unsigned int tag);
 241 unsigned int
 242 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
 243                 unsigned long first_index, unsigned int max_items,
 244                 unsigned int tag);
 245 unsigned int
 246 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
 247                 unsigned long first_index, unsigned int max_items,
 248                 unsigned int tag);
 249 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
 250                 unsigned long *first_indexp, unsigned long last_index,
 251                 unsigned long nr_to_tag,
 252                 unsigned int fromtag, unsigned int totag);
 253 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag);
 254 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item);
 255
 256 static inline void radix_tree_preload_end(void)
 257 {
 258         preempt_enable();
 259 }
 260
 261 /**
 262  * struct radix_tree_iter - radix tree iterator state
 263  *
 264  * @index:      index of current slot
 265  * @next_index: next-to-last index for this chunk
 266  * @tags:       bit-mask for tag-iterating
 267  *
 268  * This radix tree iterator works in terms of "chunks" of slots.  A chunk is a
 269  * subinterval of slots contained within one radix tree leaf node.  It is
 270  * described by a pointer to its first slot and a struct radix_tree_iter
 271  * which holds the chunk's position in the tree and its size.  For tagged
 272  * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
 273  * radix tree tag.
 274  */
 275 struct radix_tree_iter {
 276         unsigned long   index;
 277         unsigned long   next_index;
 278         unsigned long   tags;
 279 };
 280
 281 #define RADIX_TREE_ITER_TAG_MASK        0x00FF  /* tag index in lower byte */
 282 #define RADIX_TREE_ITER_TAGGED          0x0100  /* lookup tagged slots */
 283 #define RADIX_TREE_ITER_CONTIG          0x0200  /* stop at first hole */
 284
 285 /**
 286  * radix_tree_iter_init - initialize radix tree iterator
 287  *
 288  * @iter:       pointer to iterator state
 289  * @start:      iteration starting index
 290  * Returns:     NULL
 291  */
 292 static __always_inline void **
 293 radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start)
 294 {
 295         /*
 296          * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
 297          * in the case of a successful tagged chunk lookup.  If the lookup was
 298          * unsuccessful or non-tagged then nobody cares about ->tags.
 299          *
 300          * Set index to zero to bypass next_index overflow protection.
 301          * See the comment in radix_tree_next_chunk() for details.
 302          */
 303         iter->index = 0;
 304         iter->next_index = start;
 305         return NULL;
 306 }
 307
 308 /**
 309  * radix_tree_next_chunk - find next chunk of slots for iteration
 310  *
 311  * @root:       radix tree root
 312  * @iter:       iterator state
 313  * @flags:      RADIX_TREE_ITER_* flags and tag index
 314  * Returns:     pointer to chunk first slot, or NULL if there no more left
 315  *
 316  * This function looks up the next chunk in the radix tree starting from
 317  * @iter->next_index.  It returns a pointer to the chunk's first slot.
 318  * Also it fills @iter with data about chunk: position in the tree (index),
 319  * its end (next_index), and constructs a bit mask for tagged iterating (tags).
 320  */
 321 void **radix_tree_next_chunk(struct radix_tree_root *root,
 322                              struct radix_tree_iter *iter, unsigned flags);
 323
 324 /**
 325  * radix_tree_iter_retry - retry this chunk of the iteration
 326  * @iter:       iterator state
 327  *
 328  * If we iterate over a tree protected only by the RCU lock, a race
 329  * against deletion or creation may result in seeing a slot for which
 330  * radix_tree_deref_retry() returns true.  If so, call this function
 331  * and continue the iteration.
 332  */
 333 static inline __must_check
 334 void **radix_tree_iter_retry(struct radix_tree_iter *iter)
 335 {
 336         iter->next_index = iter->index;
 337         return NULL;
 338 }
 339
 340 /**
 341  * radix_tree_chunk_size - get current chunk size
 342  *
 343  * @iter:       pointer to radix tree iterator
 344  * Returns:     current chunk size
 345  */
 346 static __always_inline long
 347 radix_tree_chunk_size(struct radix_tree_iter *iter)
 348 {
 349         return iter->next_index - iter->index;
 350 }
 351
 352 /**
 353  * radix_tree_next_slot - find next slot in chunk
 354  *
 355  * @slot:       pointer to current slot
 356  * @iter:       pointer to interator state
 357  * @flags:      RADIX_TREE_ITER_*, should be constant
 358  * Returns:     pointer to next slot, or NULL if there no more left
 359  *
 360  * This function updates @iter->index in the case of a successful lookup.
 361  * For tagged lookup it also eats @iter->tags.
 362  */
 363 static __always_inline void **
 364 radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, unsigned flags)
 365 {
 366         if (flags & RADIX_TREE_ITER_TAGGED) {
 367                 iter->tags >>= 1;
 368                 if (likely(iter->tags & 1ul)) {
 369                         iter->index++;
 370                         return slot + 1;
 371                 }
 372                 if (!(flags & RADIX_TREE_ITER_CONTIG) && likely(iter->tags)) {
 373                         unsigned offset = __ffs(iter->tags);
 374
 375                         iter->tags >>= offset;
 376                         iter->index += offset + 1;
 377                         return slot + offset + 1;
 378                 }
 379         } else {
 380                 long size = radix_tree_chunk_size(iter);
 381
 382                 while (--size > 0) {
 383                         slot++;
 384                         iter->index++;
 385                         if (likely(*slot))
 386                                 return slot;
 387                         if (flags & RADIX_TREE_ITER_CONTIG) {
 388                                 /* forbid switching to the next chunk */
 389                                 iter->next_index = 0;
 390                                 break;
 391                         }
 392                 }
 393         }
 394         return NULL;
 395 }
 396
 397 /**
 398  * radix_tree_for_each_chunk - iterate over chunks
 399  *
 400  * @slot:       the void** variable for pointer to chunk first slot
 401  * @root:       the struct radix_tree_root pointer
 402  * @iter:       the struct radix_tree_iter pointer
 403  * @start:      iteration starting index
 404  * @flags:      RADIX_TREE_ITER_* and tag index
 405  *
 406  * Locks can be released and reacquired between iterations.
 407  */
 408 #define radix_tree_for_each_chunk(slot, root, iter, start, flags)       \
 409         for (slot = radix_tree_iter_init(iter, start) ;                 \
 410               (slot = radix_tree_next_chunk(root, iter, flags)) ;)
 411
 412 /**
 413  * radix_tree_for_each_chunk_slot - iterate over slots in one chunk
 414  *
 415  * @slot:       the void** variable, at the beginning points to chunk first slot
 416  * @iter:       the struct radix_tree_iter pointer
 417  * @flags:      RADIX_TREE_ITER_*, should be constant
 418  *
 419  * This macro is designed to be nested inside radix_tree_for_each_chunk().
 420  * @slot points to the radix tree slot, @iter->index contains its index.
 421  */
 422 #define radix_tree_for_each_chunk_slot(slot, iter, flags)               \
 423         for (; slot ; slot = radix_tree_next_slot(slot, iter, flags))
 424
 425 /**
 426  * radix_tree_for_each_slot - iterate over non-empty slots
 427  *
 428  * @slot:       the void** variable for pointer to slot
 429  * @root:       the struct radix_tree_root pointer
 430  * @iter:       the struct radix_tree_iter pointer
 431  * @start:      iteration starting index
 432  *
 433  * @slot points to radix tree slot, @iter->index contains its index.
 434  */
 435 #define radix_tree_for_each_slot(slot, root, iter, start)               \
 436         for (slot = radix_tree_iter_init(iter, start) ;                 \
 437              slot || (slot = radix_tree_next_chunk(root, iter, 0)) ;    \
 438              slot = radix_tree_next_slot(slot, iter, 0))
 439
 440 /**
 441  * radix_tree_for_each_contig - iterate over contiguous slots
 442  *
 443  * @slot:       the void** variable for pointer to slot
 444  * @root:       the struct radix_tree_root pointer
 445  * @iter:       the struct radix_tree_iter pointer
 446  * @start:      iteration starting index
 447  *
 448  * @slot points to radix tree slot, @iter->index contains its index.
 449  */
 450 #define radix_tree_for_each_contig(slot, root, iter, start)             \
 451         for (slot = radix_tree_iter_init(iter, start) ;                 \
 452              slot || (slot = radix_tree_next_chunk(root, iter,          \
 453                                 RADIX_TREE_ITER_CONTIG)) ;              \
 454              slot = radix_tree_next_slot(slot, iter,                    \
 455                                 RADIX_TREE_ITER_CONTIG))
 456
 457 /**
 458  * radix_tree_for_each_tagged - iterate over tagged slots
 459  *
 460  * @slot:       the void** variable for pointer to slot
 461  * @root:       the struct radix_tree_root pointer
 462  * @iter:       the struct radix_tree_iter pointer
 463  * @start:      iteration starting index
 464  * @tag:        tag index
 465  *
 466  * @slot points to radix tree slot, @iter->index contains its index.
 467  */
 468 #define radix_tree_for_each_tagged(slot, root, iter, start, tag)        \
 469         for (slot = radix_tree_iter_init(iter, start) ;                 \
 470              slot || (slot = radix_tree_next_chunk(root, iter,          \
 471                               RADIX_TREE_ITER_TAGGED | tag)) ;          \
 472              slot = radix_tree_next_slot(slot, iter,                    \
 473                                 RADIX_TREE_ITER_TAGGED))
 474
 475 #endif /* _LINUX_RADIX_TREE_H */