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/kernel.h> 13#include <linux/list.h> 14#include <linux/preempt.h> 15#include <linux/rcupdate.h> 16#include <linux/spinlock.h> 17#include <linux/types.h> 18#include <linux/xarray.h> 19 20/* Keep unconverted code working */ 21#define radix_tree_root xarray 22#define radix_tree_node xa_node 23 24/* 25 * The bottom two bits of the slot determine how the remaining bits in the 26 * slot are interpreted: 27 * 28 * 00 - data pointer 29 * 10 - internal entry 30 * x1 - value entry 31 * 32 * The internal entry may be a pointer to the next level in the tree, a 33 * sibling entry, or an indicator that the entry in this slot has been moved 34 * to another location in the tree and the lookup should be restarted. While 35 * NULL fits the 'data pointer' pattern, it means that there is no entry in 36 * the tree for this index (no matter what level of the tree it is found at). 37 * This means that storing a NULL entry in the tree is the same as deleting 38 * the entry from the tree. 39 */ 40#define RADIX_TREE_ENTRY_MASK 3UL 41#define RADIX_TREE_INTERNAL_NODE 2UL 42 43static inline bool radix_tree_is_internal_node(void *ptr) 44{ 45 return ((unsigned long)ptr & RADIX_TREE_ENTRY_MASK) == 46 RADIX_TREE_INTERNAL_NODE; 47} 48 49/*** radix-tree API starts here ***/ 50 51#define RADIX_TREE_MAP_SHIFT XA_CHUNK_SHIFT 52#define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT) 53#define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1) 54 55#define RADIX_TREE_MAX_TAGS XA_MAX_MARKS 56#define RADIX_TREE_TAG_LONGS XA_MARK_LONGS 57 58#define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long)) 59#define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \ 60 RADIX_TREE_MAP_SHIFT)) 61 62/* The IDR tag is stored in the low bits of xa_flags */ 63#define ROOT_IS_IDR ((__force gfp_t)4) 64/* The top bits of xa_flags are used to store the root tags */ 65#define ROOT_TAG_SHIFT (__GFP_BITS_SHIFT) 66 67#define RADIX_TREE_INIT(name, mask) XARRAY_INIT(name, mask) 68 69#define RADIX_TREE(name, mask) \ 70 struct radix_tree_root name = RADIX_TREE_INIT(name, mask) 71 72#define INIT_RADIX_TREE(root, mask) xa_init_flags(root, mask) 73 74static inline bool radix_tree_empty(const struct radix_tree_root *root) 75{ 76 return root->xa_head == NULL; 77} 78 79/** 80 * struct radix_tree_iter - radix tree iterator state 81 * 82 * @index: index of current slot 83 * @next_index: one beyond the last index for this chunk 84 * @tags: bit-mask for tag-iterating 85 * @node: node that contains current slot 86 * 87 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a 88 * subinterval of slots contained within one radix tree leaf node. It is 89 * described by a pointer to its first slot and a struct radix_tree_iter 90 * which holds the chunk's position in the tree and its size. For tagged 91 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen 92 * radix tree tag. 93 */ 94struct radix_tree_iter { 95 unsigned long index; 96 unsigned long next_index; 97 unsigned long tags; 98 struct radix_tree_node *node; 99}; 100 101/** 102 * Radix-tree synchronization 103 * 104 * The radix-tree API requires that users provide all synchronisation (with 105 * specific exceptions, noted below). 106 * 107 * Synchronization of access to the data items being stored in the tree, and 108 * management of their lifetimes must be completely managed by API users. 109 * 110 * For API usage, in general, 111 * - any function _modifying_ the tree or tags (inserting or deleting 112 * items, setting or clearing tags) must exclude other modifications, and 113 * exclude any functions reading the tree. 114 * - any function _reading_ the tree or tags (looking up items or tags, 115 * gang lookups) must exclude modifications to the tree, but may occur 116 * concurrently with other readers. 117 * 118 * The notable exceptions to this rule are the following functions: 119 * __radix_tree_lookup 120 * radix_tree_lookup 121 * radix_tree_lookup_slot 122 * radix_tree_tag_get 123 * radix_tree_gang_lookup 124 * radix_tree_gang_lookup_tag 125 * radix_tree_gang_lookup_tag_slot 126 * radix_tree_tagged 127 * 128 * The first 7 functions are able to be called locklessly, using RCU. The 129 * caller must ensure calls to these functions are made within rcu_read_lock() 130 * regions. Other readers (lock-free or otherwise) and modifications may be 131 * running concurrently. 132 * 133 * It is still required that the caller manage the synchronization and lifetimes 134 * of the items. So if RCU lock-free lookups are used, typically this would mean 135 * that the items have their own locks, or are amenable to lock-free access; and 136 * that the items are freed by RCU (or only freed after having been deleted from 137 * the radix tree *and* a synchronize_rcu() grace period). 138 * 139 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control 140 * access to data items when inserting into or looking up from the radix tree) 141 * 142 * Note that the value returned by radix_tree_tag_get() may not be relied upon 143 * if only the RCU read lock is held. Functions to set/clear tags and to 144 * delete nodes running concurrently with it may affect its result such that 145 * two consecutive reads in the same locked section may return different 146 * values. If reliability is required, modification functions must also be 147 * excluded from concurrency. 148 * 149 * radix_tree_tagged is able to be called without locking or RCU. 150 */ 151 152/** 153 * radix_tree_deref_slot - dereference a slot 154 * @slot: slot pointer, returned by radix_tree_lookup_slot 155 * 156 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read 157 * locked across slot lookup and dereference. Not required if write lock is 158 * held (ie. items cannot be concurrently inserted). 159 * 160 * radix_tree_deref_retry must be used to confirm validity of the pointer if 161 * only the read lock is held. 162 * 163 * Return: entry stored in that slot. 164 */ 165static inline void *radix_tree_deref_slot(void __rcu **slot) 166{ 167 return rcu_dereference(*slot); 168} 169 170/** 171 * radix_tree_deref_slot_protected - dereference a slot with tree lock held 172 * @slot: slot pointer, returned by radix_tree_lookup_slot 173 * 174 * Similar to radix_tree_deref_slot. The caller does not hold the RCU read 175 * lock but it must hold the tree lock to prevent parallel updates. 176 * 177 * Return: entry stored in that slot. 178 */ 179static inline void *radix_tree_deref_slot_protected(void __rcu **slot, 180 spinlock_t *treelock) 181{ 182 return rcu_dereference_protected(*slot, lockdep_is_held(treelock)); 183} 184 185/** 186 * radix_tree_deref_retry - check radix_tree_deref_slot 187 * @arg: pointer returned by radix_tree_deref_slot 188 * Returns: 0 if retry is not required, otherwise retry is required 189 * 190 * radix_tree_deref_retry must be used with radix_tree_deref_slot. 191 */ 192static inline int radix_tree_deref_retry(void *arg) 193{ 194 return unlikely(radix_tree_is_internal_node(arg)); 195} 196 197/** 198 * radix_tree_exception - radix_tree_deref_slot returned either exception? 199 * @arg: value returned by radix_tree_deref_slot 200 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception. 201 */ 202static inline int radix_tree_exception(void *arg) 203{ 204 return unlikely((unsigned long)arg & RADIX_TREE_ENTRY_MASK); 205} 206 207int radix_tree_insert(struct radix_tree_root *, unsigned long index, 208 void *); 209void *__radix_tree_lookup(const struct radix_tree_root *, unsigned long index, 210 struct radix_tree_node **nodep, void __rcu ***slotp); 211void *radix_tree_lookup(const struct radix_tree_root *, unsigned long); 212void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *, 213 unsigned long index); 214void __radix_tree_replace(struct radix_tree_root *, struct radix_tree_node *, 215 void __rcu **slot, void *entry); 216void radix_tree_iter_replace(struct radix_tree_root *, 217 const struct radix_tree_iter *, void __rcu **slot, void *entry); 218void radix_tree_replace_slot(struct radix_tree_root *, 219 void __rcu **slot, void *entry); 220void radix_tree_iter_delete(struct radix_tree_root *, 221 struct radix_tree_iter *iter, void __rcu **slot); 222void *radix_tree_delete_item(struct radix_tree_root *, unsigned long, void *); 223void *radix_tree_delete(struct radix_tree_root *, unsigned long); 224unsigned int radix_tree_gang_lookup(const struct radix_tree_root *, 225 void **results, unsigned long first_index, 226 unsigned int max_items); 227int radix_tree_preload(gfp_t gfp_mask); 228int radix_tree_maybe_preload(gfp_t gfp_mask); 229void radix_tree_init(void); 230void *radix_tree_tag_set(struct radix_tree_root *, 231 unsigned long index, unsigned int tag); 232void *radix_tree_tag_clear(struct radix_tree_root *, 233 unsigned long index, unsigned int tag); 234int radix_tree_tag_get(const struct radix_tree_root *, 235 unsigned long index, unsigned int tag); 236void radix_tree_iter_tag_clear(struct radix_tree_root *, 237 const struct radix_tree_iter *iter, unsigned int tag); 238unsigned int radix_tree_gang_lookup_tag(const struct radix_tree_root *, 239 void **results, unsigned long first_index, 240 unsigned int max_items, unsigned int tag); 241unsigned int radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *, 242 void __rcu ***results, unsigned long first_index, 243 unsigned int max_items, unsigned int tag); 244int radix_tree_tagged(const struct radix_tree_root *, unsigned int tag); 245 246static inline void radix_tree_preload_end(void) 247{ 248 preempt_enable(); 249} 250 251void __rcu **idr_get_free(struct radix_tree_root *root, 252 struct radix_tree_iter *iter, gfp_t gfp, 253 unsigned long max); 254 255enum { 256 RADIX_TREE_ITER_TAG_MASK = 0x0f, /* tag index in lower nybble */ 257 RADIX_TREE_ITER_TAGGED = 0x10, /* lookup tagged slots */ 258 RADIX_TREE_ITER_CONTIG = 0x20, /* stop at first hole */ 259}; 260 261/** 262 * radix_tree_iter_init - initialize radix tree iterator 263 * 264 * @iter: pointer to iterator state 265 * @start: iteration starting index 266 * Returns: NULL 267 */ 268static __always_inline void __rcu ** 269radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start) 270{ 271 /* 272 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it 273 * in the case of a successful tagged chunk lookup. If the lookup was 274 * unsuccessful or non-tagged then nobody cares about ->tags. 275 * 276 * Set index to zero to bypass next_index overflow protection. 277 * See the comment in radix_tree_next_chunk() for details. 278 */ 279 iter->index = 0; 280 iter->next_index = start; 281 return NULL; 282} 283 284/** 285 * radix_tree_next_chunk - find next chunk of slots for iteration 286 * 287 * @root: radix tree root 288 * @iter: iterator state 289 * @flags: RADIX_TREE_ITER_* flags and tag index 290 * Returns: pointer to chunk first slot, or NULL if there no more left 291 * 292 * This function looks up the next chunk in the radix tree starting from 293 * @iter->next_index. It returns a pointer to the chunk's first slot. 294 * Also it fills @iter with data about chunk: position in the tree (index), 295 * its end (next_index), and constructs a bit mask for tagged iterating (tags). 296 */ 297void __rcu **radix_tree_next_chunk(const struct radix_tree_root *, 298 struct radix_tree_iter *iter, unsigned flags); 299 300/** 301 * radix_tree_iter_lookup - look up an index in the radix tree 302 * @root: radix tree root 303 * @iter: iterator state 304 * @index: key to look up 305 * 306 * If @index is present in the radix tree, this function returns the slot 307 * containing it and updates @iter to describe the entry. If @index is not 308 * present, it returns NULL. 309 */ 310static inline void __rcu ** 311radix_tree_iter_lookup(const struct radix_tree_root *root, 312 struct radix_tree_iter *iter, unsigned long index) 313{ 314 radix_tree_iter_init(iter, index); 315 return radix_tree_next_chunk(root, iter, RADIX_TREE_ITER_CONTIG); 316} 317 318/** 319 * radix_tree_iter_find - find a present entry 320 * @root: radix tree root 321 * @iter: iterator state 322 * @index: start location 323 * 324 * This function returns the slot containing the entry with the lowest index 325 * which is at least @index. If @index is larger than any present entry, this 326 * function returns NULL. The @iter is updated to describe the entry found. 327 */ 328static inline void __rcu ** 329radix_tree_iter_find(const struct radix_tree_root *root, 330 struct radix_tree_iter *iter, unsigned long index) 331{ 332 radix_tree_iter_init(iter, index); 333 return radix_tree_next_chunk(root, iter, 0); 334} 335 336/** 337 * radix_tree_iter_retry - retry this chunk of the iteration 338 * @iter: iterator state 339 * 340 * If we iterate over a tree protected only by the RCU lock, a race 341 * against deletion or creation may result in seeing a slot for which 342 * radix_tree_deref_retry() returns true. If so, call this function 343 * and continue the iteration. 344 */ 345static inline __must_check 346void __rcu **radix_tree_iter_retry(struct radix_tree_iter *iter) 347{ 348 iter->next_index = iter->index; 349 iter->tags = 0; 350 return NULL; 351} 352 353static inline unsigned long 354__radix_tree_iter_add(struct radix_tree_iter *iter, unsigned long slots) 355{ 356 return iter->index + slots; 357} 358 359/** 360 * radix_tree_iter_resume - resume iterating when the chunk may be invalid 361 * @slot: pointer to current slot 362 * @iter: iterator state 363 * Returns: New slot pointer 364 * 365 * If the iterator needs to release then reacquire a lock, the chunk may 366 * have been invalidated by an insertion or deletion. Call this function 367 * before releasing the lock to continue the iteration from the next index. 368 */ 369void __rcu **__must_check radix_tree_iter_resume(void __rcu **slot, 370 struct radix_tree_iter *iter); 371 372/** 373 * radix_tree_chunk_size - get current chunk size 374 * 375 * @iter: pointer to radix tree iterator 376 * Returns: current chunk size 377 */ 378static __always_inline long 379radix_tree_chunk_size(struct radix_tree_iter *iter) 380{ 381 return iter->next_index - iter->index; 382} 383 384/** 385 * radix_tree_next_slot - find next slot in chunk 386 * 387 * @slot: pointer to current slot 388 * @iter: pointer to interator state 389 * @flags: RADIX_TREE_ITER_*, should be constant 390 * Returns: pointer to next slot, or NULL if there no more left 391 * 392 * This function updates @iter->index in the case of a successful lookup. 393 * For tagged lookup it also eats @iter->tags. 394 * 395 * There are several cases where 'slot' can be passed in as NULL to this 396 * function. These cases result from the use of radix_tree_iter_resume() or 397 * radix_tree_iter_retry(). In these cases we don't end up dereferencing 398 * 'slot' because either: 399 * a) we are doing tagged iteration and iter->tags has been set to 0, or 400 * b) we are doing non-tagged iteration, and iter->index and iter->next_index 401 * have been set up so that radix_tree_chunk_size() returns 1 or 0. 402 */ 403static __always_inline void __rcu **radix_tree_next_slot(void __rcu **slot, 404 struct radix_tree_iter *iter, unsigned flags) 405{ 406 if (flags & RADIX_TREE_ITER_TAGGED) { 407 iter->tags >>= 1; 408 if (unlikely(!iter->tags)) 409 return NULL; 410 if (likely(iter->tags & 1ul)) { 411 iter->index = __radix_tree_iter_add(iter, 1); 412 slot++; 413 goto found; 414 } 415 if (!(flags & RADIX_TREE_ITER_CONTIG)) { 416 unsigned offset = __ffs(iter->tags); 417 418 iter->tags >>= offset++; 419 iter->index = __radix_tree_iter_add(iter, offset); 420 slot += offset; 421 goto found; 422 } 423 } else { 424 long count = radix_tree_chunk_size(iter); 425 426 while (--count > 0) { 427 slot++; 428 iter->index = __radix_tree_iter_add(iter, 1); 429 430 if (likely(*slot)) 431 goto found; 432 if (flags & RADIX_TREE_ITER_CONTIG) { 433 /* forbid switching to the next chunk */ 434 iter->next_index = 0; 435 break; 436 } 437 } 438 } 439 return NULL; 440 441 found: 442 return slot; 443} 444 445/** 446 * radix_tree_for_each_slot - iterate over non-empty slots 447 * 448 * @slot: the void** variable for pointer to slot 449 * @root: the struct radix_tree_root pointer 450 * @iter: the struct radix_tree_iter pointer 451 * @start: iteration starting index 452 * 453 * @slot points to radix tree slot, @iter->index contains its index. 454 */ 455#define radix_tree_for_each_slot(slot, root, iter, start) \ 456 for (slot = radix_tree_iter_init(iter, start) ; \ 457 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \ 458 slot = radix_tree_next_slot(slot, iter, 0)) 459 460/** 461 * radix_tree_for_each_tagged - iterate over tagged slots 462 * 463 * @slot: the void** variable for pointer to slot 464 * @root: the struct radix_tree_root pointer 465 * @iter: the struct radix_tree_iter pointer 466 * @start: iteration starting index 467 * @tag: tag index 468 * 469 * @slot points to radix tree slot, @iter->index contains its index. 470 */ 471#define radix_tree_for_each_tagged(slot, root, iter, start, tag) \ 472 for (slot = radix_tree_iter_init(iter, start) ; \ 473 slot || (slot = radix_tree_next_chunk(root, iter, \ 474 RADIX_TREE_ITER_TAGGED | tag)) ; \ 475 slot = radix_tree_next_slot(slot, iter, \ 476 RADIX_TREE_ITER_TAGGED | tag)) 477 478#endif /* _LINUX_RADIX_TREE_H */ 479