1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Copyright (C) 2001 Momchil Velikov 4 * Portions Copyright (C) 2001 Christoph Hellwig 5 * Copyright (C) 2006 Nick Piggin 6 * Copyright (C) 2012 Konstantin Khlebnikov 7 */ 8 #ifndef _LINUX_RADIX_TREE_H 9 #define _LINUX_RADIX_TREE_H 10 11 #include <linux/bitops.h> 12 #include <linux/gfp_types.h> 13 #include <linux/list.h> 14 #include <linux/lockdep.h> 15 #include <linux/math.h> 16 #include <linux/percpu.h> 17 #include <linux/preempt.h> 18 #include <linux/rcupdate.h> 19 #include <linux/spinlock.h> 20 #include <linux/types.h> 21 #include <linux/xarray.h> 22 #include <linux/local_lock.h> 23 24 /* Keep unconverted code working */ 25 #define radix_tree_root xarray 26 #define radix_tree_node xa_node 27 28 struct radix_tree_preload { 29 local_lock_t lock; 30 unsigned nr; 31 /* nodes->parent points to next preallocated node */ 32 struct radix_tree_node *nodes; 33 }; 34 DECLARE_PER_CPU(struct radix_tree_preload, radix_tree_preloads); 35 36 /* 37 * The bottom two bits of the slot determine how the remaining bits in the 38 * slot are interpreted: 39 * 40 * 00 - data pointer 41 * 10 - internal entry 42 * x1 - value entry 43 * 44 * The internal entry may be a pointer to the next level in the tree, a 45 * sibling entry, or an indicator that the entry in this slot has been moved 46 * to another location in the tree and the lookup should be restarted. While 47 * NULL fits the 'data pointer' pattern, it means that there is no entry in 48 * the tree for this index (no matter what level of the tree it is found at). 49 * This means that storing a NULL entry in the tree is the same as deleting 50 * the entry from the tree. 51 */ 52 #define RADIX_TREE_ENTRY_MASK 3UL 53 #define RADIX_TREE_INTERNAL_NODE 2UL 54 55 static inline bool radix_tree_is_internal_node(void *ptr) 56 { 57 return ((unsigned long)ptr & RADIX_TREE_ENTRY_MASK) == 58 RADIX_TREE_INTERNAL_NODE; 59 } 60 61 /*** radix-tree API starts here ***/ 62 63 #define RADIX_TREE_MAP_SHIFT XA_CHUNK_SHIFT 64 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT) 65 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1) 66 67 #define RADIX_TREE_MAX_TAGS XA_MAX_MARKS 68 #define RADIX_TREE_TAG_LONGS XA_MARK_LONGS 69 70 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long)) 71 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \ 72 RADIX_TREE_MAP_SHIFT)) 73 74 /* The IDR tag is stored in the low bits of xa_flags */ 75 #define ROOT_IS_IDR ((__force gfp_t)4) 76 /* The top bits of xa_flags are used to store the root tags */ 77 #define ROOT_TAG_SHIFT (__GFP_BITS_SHIFT) 78 79 #define RADIX_TREE_INIT(name, mask) XARRAY_INIT(name, mask) 80 81 #define RADIX_TREE(name, mask) \ 82 struct radix_tree_root name = RADIX_TREE_INIT(name, mask) 83 84 #define INIT_RADIX_TREE(root, mask) xa_init_flags(root, mask) 85 86 static inline bool radix_tree_empty(const struct radix_tree_root *root) 87 { 88 return root->xa_head == NULL; 89 } 90 91 /** 92 * struct radix_tree_iter - radix tree iterator state 93 * 94 * @index: index of current slot 95 * @next_index: one beyond the last index for this chunk 96 * @tags: bit-mask for tag-iterating 97 * @node: node that contains current slot 98 * 99 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a 100 * subinterval of slots contained within one radix tree leaf node. It is 101 * described by a pointer to its first slot and a struct radix_tree_iter 102 * which holds the chunk's position in the tree and its size. For tagged 103 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen 104 * radix tree tag. 105 */ 106 struct radix_tree_iter { 107 unsigned long index; 108 unsigned long next_index; 109 unsigned long tags; 110 struct radix_tree_node *node; 111 }; 112 113 /** 114 * DOC: Radix-tree synchronization 115 * 116 * The radix-tree API requires that users provide all synchronisation (with 117 * specific exceptions, noted below). 118 * 119 * Synchronization of access to the data items being stored in the tree, and 120 * management of their lifetimes must be completely managed by API users. 121 * 122 * For API usage, in general, 123 * - any function _modifying_ the tree or tags (inserting or deleting 124 * items, setting or clearing tags) must exclude other modifications, and 125 * exclude any functions reading the tree. 126 * - any function _reading_ the tree or tags (looking up items or tags, 127 * gang lookups) must exclude modifications to the tree, but may occur 128 * concurrently with other readers. 129 * 130 * The notable exceptions to this rule are the following functions: 131 * __radix_tree_lookup 132 * radix_tree_lookup 133 * radix_tree_lookup_slot 134 * radix_tree_tag_get 135 * radix_tree_gang_lookup 136 * radix_tree_gang_lookup_tag 137 * radix_tree_gang_lookup_tag_slot 138 * radix_tree_tagged 139 * 140 * The first 7 functions are able to be called locklessly, using RCU. The 141 * caller must ensure calls to these functions are made within rcu_read_lock() 142 * regions. Other readers (lock-free or otherwise) and modifications may be 143 * running concurrently. 144 * 145 * It is still required that the caller manage the synchronization and lifetimes 146 * of the items. So if RCU lock-free lookups are used, typically this would mean 147 * that the items have their own locks, or are amenable to lock-free access; and 148 * that the items are freed by RCU (or only freed after having been deleted from 149 * the radix tree *and* a synchronize_rcu() grace period). 150 * 151 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control 152 * access to data items when inserting into or looking up from the radix tree) 153 * 154 * Note that the value returned by radix_tree_tag_get() may not be relied upon 155 * if only the RCU read lock is held. Functions to set/clear tags and to 156 * delete nodes running concurrently with it may affect its result such that 157 * two consecutive reads in the same locked section may return different 158 * values. If reliability is required, modification functions must also be 159 * excluded from concurrency. 160 * 161 * radix_tree_tagged is able to be called without locking or RCU. 162 */ 163 164 /** 165 * radix_tree_deref_slot - dereference a slot 166 * @slot: slot pointer, returned by radix_tree_lookup_slot 167 * 168 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read 169 * locked across slot lookup and dereference. Not required if write lock is 170 * held (ie. items cannot be concurrently inserted). 171 * 172 * radix_tree_deref_retry must be used to confirm validity of the pointer if 173 * only the read lock is held. 174 * 175 * Return: entry stored in that slot. 176 */ 177 static inline void *radix_tree_deref_slot(void __rcu **slot) 178 { 179 return rcu_dereference(*slot); 180 } 181 182 /** 183 * radix_tree_deref_slot_protected - dereference a slot with tree lock held 184 * @slot: slot pointer, returned by radix_tree_lookup_slot 185 * @treelock: caller must hold this spinlock 186 * 187 * Similar to radix_tree_deref_slot. The caller does not hold the RCU read 188 * lock but it must hold the tree lock to prevent parallel updates. 189 * 190 * Return: entry stored in that slot. 191 */ 192 static inline void *radix_tree_deref_slot_protected(void __rcu **slot, 193 spinlock_t *treelock) 194 { 195 return rcu_dereference_protected(*slot, lockdep_is_held(treelock)); 196 } 197 198 /** 199 * radix_tree_deref_retry - check radix_tree_deref_slot 200 * @arg: pointer returned by radix_tree_deref_slot 201 * Returns: 0 if retry is not required, otherwise retry is required 202 * 203 * radix_tree_deref_retry must be used with radix_tree_deref_slot. 204 */ 205 static inline int radix_tree_deref_retry(void *arg) 206 { 207 return unlikely(radix_tree_is_internal_node(arg)); 208 } 209 210 /** 211 * radix_tree_exception - radix_tree_deref_slot returned either exception? 212 * @arg: value returned by radix_tree_deref_slot 213 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception. 214 */ 215 static inline int radix_tree_exception(void *arg) 216 { 217 return unlikely((unsigned long)arg & RADIX_TREE_ENTRY_MASK); 218 } 219 220 int radix_tree_insert(struct radix_tree_root *, unsigned long index, 221 void *); 222 void *__radix_tree_lookup(const struct radix_tree_root *, unsigned long index, 223 struct radix_tree_node **nodep, void __rcu ***slotp); 224 void *radix_tree_lookup(const struct radix_tree_root *, unsigned long); 225 void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *, 226 unsigned long index); 227 void __radix_tree_replace(struct radix_tree_root *, struct radix_tree_node *, 228 void __rcu **slot, void *entry); 229 void radix_tree_iter_replace(struct radix_tree_root *, 230 const struct radix_tree_iter *, void __rcu **slot, void *entry); 231 void radix_tree_replace_slot(struct radix_tree_root *, 232 void __rcu **slot, void *entry); 233 void radix_tree_iter_delete(struct radix_tree_root *, 234 struct radix_tree_iter *iter, void __rcu **slot); 235 void *radix_tree_delete_item(struct radix_tree_root *, unsigned long, void *); 236 void *radix_tree_delete(struct radix_tree_root *, unsigned long); 237 unsigned int radix_tree_gang_lookup(const struct radix_tree_root *, 238 void **results, unsigned long first_index, 239 unsigned int max_items); 240 int radix_tree_preload(gfp_t gfp_mask); 241 int radix_tree_maybe_preload(gfp_t gfp_mask); 242 void radix_tree_init(void); 243 void *radix_tree_tag_set(struct radix_tree_root *, 244 unsigned long index, unsigned int tag); 245 void *radix_tree_tag_clear(struct radix_tree_root *, 246 unsigned long index, unsigned int tag); 247 int radix_tree_tag_get(const struct radix_tree_root *, 248 unsigned long index, unsigned int tag); 249 void radix_tree_iter_tag_clear(struct radix_tree_root *, 250 const struct radix_tree_iter *iter, unsigned int tag); 251 unsigned int radix_tree_gang_lookup_tag(const struct radix_tree_root *, 252 void **results, unsigned long first_index, 253 unsigned int max_items, unsigned int tag); 254 unsigned int radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *, 255 void __rcu ***results, unsigned long first_index, 256 unsigned int max_items, unsigned int tag); 257 int radix_tree_tagged(const struct radix_tree_root *, unsigned int tag); 258 259 static inline void radix_tree_preload_end(void) 260 { 261 local_unlock(&radix_tree_preloads.lock); 262 } 263 264 void __rcu **idr_get_free(struct radix_tree_root *root, 265 struct radix_tree_iter *iter, gfp_t gfp, 266 unsigned long max); 267 268 enum { 269 RADIX_TREE_ITER_TAG_MASK = 0x0f, /* tag index in lower nybble */ 270 RADIX_TREE_ITER_TAGGED = 0x10, /* lookup tagged slots */ 271 RADIX_TREE_ITER_CONTIG = 0x20, /* stop at first hole */ 272 }; 273 274 /** 275 * radix_tree_iter_init - initialize radix tree iterator 276 * 277 * @iter: pointer to iterator state 278 * @start: iteration starting index 279 * Returns: NULL 280 */ 281 static __always_inline void __rcu ** 282 radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start) 283 { 284 /* 285 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it 286 * in the case of a successful tagged chunk lookup. If the lookup was 287 * unsuccessful or non-tagged then nobody cares about ->tags. 288 * 289 * Set index to zero to bypass next_index overflow protection. 290 * See the comment in radix_tree_next_chunk() for details. 291 */ 292 iter->index = 0; 293 iter->next_index = start; 294 return NULL; 295 } 296 297 /** 298 * radix_tree_next_chunk - find next chunk of slots for iteration 299 * 300 * @root: radix tree root 301 * @iter: iterator state 302 * @flags: RADIX_TREE_ITER_* flags and tag index 303 * Returns: pointer to chunk first slot, or NULL if there no more left 304 * 305 * This function looks up the next chunk in the radix tree starting from 306 * @iter->next_index. It returns a pointer to the chunk's first slot. 307 * Also it fills @iter with data about chunk: position in the tree (index), 308 * its end (next_index), and constructs a bit mask for tagged iterating (tags). 309 */ 310 void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root, 311 struct radix_tree_iter *iter, unsigned flags); 312 313 /** 314 * radix_tree_iter_lookup - look up an index in the radix tree 315 * @root: radix tree root 316 * @iter: iterator state 317 * @index: key to look up 318 * 319 * If @index is present in the radix tree, this function returns the slot 320 * containing it and updates @iter to describe the entry. If @index is not 321 * present, it returns NULL. 322 */ 323 static inline void __rcu ** 324 radix_tree_iter_lookup(const struct radix_tree_root *root, 325 struct radix_tree_iter *iter, unsigned long index) 326 { 327 radix_tree_iter_init(iter, index); 328 return radix_tree_next_chunk(root, iter, RADIX_TREE_ITER_CONTIG); 329 } 330 331 /** 332 * radix_tree_iter_retry - retry this chunk of the iteration 333 * @iter: iterator state 334 * 335 * If we iterate over a tree protected only by the RCU lock, a race 336 * against deletion or creation may result in seeing a slot for which 337 * radix_tree_deref_retry() returns true. If so, call this function 338 * and continue the iteration. 339 */ 340 static inline __must_check 341 void __rcu **radix_tree_iter_retry(struct radix_tree_iter *iter) 342 { 343 iter->next_index = iter->index; 344 iter->tags = 0; 345 return NULL; 346 } 347 348 static inline unsigned long 349 __radix_tree_iter_add(struct radix_tree_iter *iter, unsigned long slots) 350 { 351 return iter->index + slots; 352 } 353 354 /** 355 * radix_tree_iter_resume - resume iterating when the chunk may be invalid 356 * @slot: pointer to current slot 357 * @iter: iterator state 358 * Returns: New slot pointer 359 * 360 * If the iterator needs to release then reacquire a lock, the chunk may 361 * have been invalidated by an insertion or deletion. Call this function 362 * before releasing the lock to continue the iteration from the next index. 363 */ 364 void __rcu **__must_check radix_tree_iter_resume(void __rcu **slot, 365 struct radix_tree_iter *iter); 366 367 /** 368 * radix_tree_chunk_size - get current chunk size 369 * 370 * @iter: pointer to radix tree iterator 371 * Returns: current chunk size 372 */ 373 static __always_inline long 374 radix_tree_chunk_size(struct radix_tree_iter *iter) 375 { 376 return iter->next_index - iter->index; 377 } 378 379 /** 380 * radix_tree_next_slot - find next slot in chunk 381 * 382 * @slot: pointer to current slot 383 * @iter: pointer to iterator state 384 * @flags: RADIX_TREE_ITER_*, should be constant 385 * Returns: pointer to next slot, or NULL if there no more left 386 * 387 * This function updates @iter->index in the case of a successful lookup. 388 * For tagged lookup it also eats @iter->tags. 389 * 390 * There are several cases where 'slot' can be passed in as NULL to this 391 * function. These cases result from the use of radix_tree_iter_resume() or 392 * radix_tree_iter_retry(). In these cases we don't end up dereferencing 393 * 'slot' because either: 394 * a) we are doing tagged iteration and iter->tags has been set to 0, or 395 * b) we are doing non-tagged iteration, and iter->index and iter->next_index 396 * have been set up so that radix_tree_chunk_size() returns 1 or 0. 397 */ 398 static __always_inline void __rcu **radix_tree_next_slot(void __rcu **slot, 399 struct radix_tree_iter *iter, unsigned flags) 400 { 401 if (flags & RADIX_TREE_ITER_TAGGED) { 402 iter->tags >>= 1; 403 if (unlikely(!iter->tags)) 404 return NULL; 405 if (likely(iter->tags & 1ul)) { 406 iter->index = __radix_tree_iter_add(iter, 1); 407 slot++; 408 goto found; 409 } 410 if (!(flags & RADIX_TREE_ITER_CONTIG)) { 411 unsigned offset = __ffs(iter->tags); 412 413 iter->tags >>= offset++; 414 iter->index = __radix_tree_iter_add(iter, offset); 415 slot += offset; 416 goto found; 417 } 418 } else { 419 long count = radix_tree_chunk_size(iter); 420 421 while (--count > 0) { 422 slot++; 423 iter->index = __radix_tree_iter_add(iter, 1); 424 425 if (likely(*slot)) 426 goto found; 427 if (flags & RADIX_TREE_ITER_CONTIG) { 428 /* forbid switching to the next chunk */ 429 iter->next_index = 0; 430 break; 431 } 432 } 433 } 434 return NULL; 435 436 found: 437 return slot; 438 } 439 440 /** 441 * radix_tree_for_each_slot - iterate over non-empty 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_slot(slot, root, iter, start) \ 451 for (slot = radix_tree_iter_init(iter, start) ; \ 452 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \ 453 slot = radix_tree_next_slot(slot, iter, 0)) 454 455 /** 456 * radix_tree_for_each_tagged - iterate over tagged slots 457 * 458 * @slot: the void** variable for pointer to slot 459 * @root: the struct radix_tree_root pointer 460 * @iter: the struct radix_tree_iter pointer 461 * @start: iteration starting index 462 * @tag: tag index 463 * 464 * @slot points to radix tree slot, @iter->index contains its index. 465 */ 466 #define radix_tree_for_each_tagged(slot, root, iter, start, tag) \ 467 for (slot = radix_tree_iter_init(iter, start) ; \ 468 slot || (slot = radix_tree_next_chunk(root, iter, \ 469 RADIX_TREE_ITER_TAGGED | tag)) ; \ 470 slot = radix_tree_next_slot(slot, iter, \ 471 RADIX_TREE_ITER_TAGGED | tag)) 472 473 #endif /* _LINUX_RADIX_TREE_H */ 474