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/bitops.h> 25#include <linux/preempt.h> 26#include <linux/types.h> 27#include <linux/bug.h> 28#include <linux/kernel.h> 29#include <linux/rcupdate.h> 30 31/* 32 * An indirect pointer (root->rnode pointing to a radix_tree_node, rather 33 * than a data item) is signalled by the low bit set in the root->rnode 34 * pointer. 35 * 36 * In this case root->height is > 0, but the indirect pointer tests are 37 * needed for RCU lookups (because root->height is unreliable). The only 38 * time callers need worry about this is when doing a lookup_slot under 39 * RCU. 40 * 41 * Indirect pointer in fact is also used to tag the last pointer of a node 42 * when it is shrunk, before we rcu free the node. See shrink code for 43 * details. 44 */ 45#define RADIX_TREE_INDIRECT_PTR 1 46/* 47 * A common use of the radix tree is to store pointers to struct pages; 48 * but shmem/tmpfs needs also to store swap entries in the same tree: 49 * those are marked as exceptional entries to distinguish them. 50 * EXCEPTIONAL_ENTRY tests the bit, EXCEPTIONAL_SHIFT shifts content past it. 51 */ 52#define RADIX_TREE_EXCEPTIONAL_ENTRY 2 53#define RADIX_TREE_EXCEPTIONAL_SHIFT 2 54 55#define RADIX_DAX_MASK 0xf 56#define RADIX_DAX_SHIFT 4 57#define RADIX_DAX_PTE (0x4 | RADIX_TREE_EXCEPTIONAL_ENTRY) 58#define RADIX_DAX_PMD (0x8 | RADIX_TREE_EXCEPTIONAL_ENTRY) 59#define RADIX_DAX_TYPE(entry) ((unsigned long)entry & RADIX_DAX_MASK) 60#define RADIX_DAX_SECTOR(entry) (((unsigned long)entry >> RADIX_DAX_SHIFT)) 61#define RADIX_DAX_ENTRY(sector, pmd) ((void *)((unsigned long)sector << \ 62 RADIX_DAX_SHIFT | (pmd ? RADIX_DAX_PMD : RADIX_DAX_PTE))) 63 64static inline int radix_tree_is_indirect_ptr(void *ptr) 65{ 66 return (int)((unsigned long)ptr & RADIX_TREE_INDIRECT_PTR); 67} 68 69/*** radix-tree API starts here ***/ 70 71#define RADIX_TREE_MAX_TAGS 3 72 73#ifdef __KERNEL__ 74#define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6) 75#else 76#define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */ 77#endif 78 79#define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT) 80#define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1) 81 82#define RADIX_TREE_TAG_LONGS \ 83 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG) 84 85#define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long)) 86#define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \ 87 RADIX_TREE_MAP_SHIFT)) 88 89/* Height component in node->path */ 90#define RADIX_TREE_HEIGHT_SHIFT (RADIX_TREE_MAX_PATH + 1) 91#define RADIX_TREE_HEIGHT_MASK ((1UL << RADIX_TREE_HEIGHT_SHIFT) - 1) 92 93/* Internally used bits of node->count */ 94#define RADIX_TREE_COUNT_SHIFT (RADIX_TREE_MAP_SHIFT + 1) 95#define RADIX_TREE_COUNT_MASK ((1UL << RADIX_TREE_COUNT_SHIFT) - 1) 96 97struct radix_tree_node { 98 unsigned int path; /* Offset in parent & height from the bottom */ 99 unsigned int count; 100 union { 101 struct { 102 /* Used when ascending tree */ 103 struct radix_tree_node *parent; 104 /* For tree user */ 105 void *private_data; 106 }; 107 /* Used when freeing node */ 108 struct rcu_head rcu_head; 109 }; 110 /* For tree user */ 111 struct list_head private_list; 112 void __rcu *slots[RADIX_TREE_MAP_SIZE]; 113 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS]; 114}; 115 116/* root tags are stored in gfp_mask, shifted by __GFP_BITS_SHIFT */ 117struct radix_tree_root { 118 unsigned int height; 119 gfp_t gfp_mask; 120 struct radix_tree_node __rcu *rnode; 121}; 122 123#define RADIX_TREE_INIT(mask) { \ 124 .height = 0, \ 125 .gfp_mask = (mask), \ 126 .rnode = NULL, \ 127} 128 129#define RADIX_TREE(name, mask) \ 130 struct radix_tree_root name = RADIX_TREE_INIT(mask) 131 132#define INIT_RADIX_TREE(root, mask) \ 133do { \ 134 (root)->height = 0; \ 135 (root)->gfp_mask = (mask); \ 136 (root)->rnode = NULL; \ 137} while (0) 138 139/** 140 * Radix-tree synchronization 141 * 142 * The radix-tree API requires that users provide all synchronisation (with 143 * specific exceptions, noted below). 144 * 145 * Synchronization of access to the data items being stored in the tree, and 146 * management of their lifetimes must be completely managed by API users. 147 * 148 * For API usage, in general, 149 * - any function _modifying_ the tree or tags (inserting or deleting 150 * items, setting or clearing tags) must exclude other modifications, and 151 * exclude any functions reading the tree. 152 * - any function _reading_ the tree or tags (looking up items or tags, 153 * gang lookups) must exclude modifications to the tree, but may occur 154 * concurrently with other readers. 155 * 156 * The notable exceptions to this rule are the following functions: 157 * __radix_tree_lookup 158 * radix_tree_lookup 159 * radix_tree_lookup_slot 160 * radix_tree_tag_get 161 * radix_tree_gang_lookup 162 * radix_tree_gang_lookup_slot 163 * radix_tree_gang_lookup_tag 164 * radix_tree_gang_lookup_tag_slot 165 * radix_tree_tagged 166 * 167 * The first 8 functions are able to be called locklessly, using RCU. The 168 * caller must ensure calls to these functions are made within rcu_read_lock() 169 * regions. Other readers (lock-free or otherwise) and modifications may be 170 * running concurrently. 171 * 172 * It is still required that the caller manage the synchronization and lifetimes 173 * of the items. So if RCU lock-free lookups are used, typically this would mean 174 * that the items have their own locks, or are amenable to lock-free access; and 175 * that the items are freed by RCU (or only freed after having been deleted from 176 * the radix tree *and* a synchronize_rcu() grace period). 177 * 178 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control 179 * access to data items when inserting into or looking up from the radix tree) 180 * 181 * Note that the value returned by radix_tree_tag_get() may not be relied upon 182 * if only the RCU read lock is held. Functions to set/clear tags and to 183 * delete nodes running concurrently with it may affect its result such that 184 * two consecutive reads in the same locked section may return different 185 * values. If reliability is required, modification functions must also be 186 * excluded from concurrency. 187 * 188 * radix_tree_tagged is able to be called without locking or RCU. 189 */ 190 191/** 192 * radix_tree_deref_slot - dereference a slot 193 * @pslot: pointer to slot, returned by radix_tree_lookup_slot 194 * Returns: item that was stored in that slot with any direct pointer flag 195 * removed. 196 * 197 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read 198 * locked across slot lookup and dereference. Not required if write lock is 199 * held (ie. items cannot be concurrently inserted). 200 * 201 * radix_tree_deref_retry must be used to confirm validity of the pointer if 202 * only the read lock is held. 203 */ 204static inline void *radix_tree_deref_slot(void **pslot) 205{ 206 return rcu_dereference(*pslot); 207} 208 209/** 210 * radix_tree_deref_slot_protected - dereference a slot without RCU lock but with tree lock held 211 * @pslot: pointer to slot, returned by radix_tree_lookup_slot 212 * Returns: item that was stored in that slot with any direct pointer flag 213 * removed. 214 * 215 * Similar to radix_tree_deref_slot but only used during migration when a pages 216 * mapping is being moved. The caller does not hold the RCU read lock but it 217 * must hold the tree lock to prevent parallel updates. 218 */ 219static inline void *radix_tree_deref_slot_protected(void **pslot, 220 spinlock_t *treelock) 221{ 222 return rcu_dereference_protected(*pslot, lockdep_is_held(treelock)); 223} 224 225/** 226 * radix_tree_deref_retry - check radix_tree_deref_slot 227 * @arg: pointer returned by radix_tree_deref_slot 228 * Returns: 0 if retry is not required, otherwise retry is required 229 * 230 * radix_tree_deref_retry must be used with radix_tree_deref_slot. 231 */ 232static inline int radix_tree_deref_retry(void *arg) 233{ 234 return unlikely((unsigned long)arg & RADIX_TREE_INDIRECT_PTR); 235} 236 237/** 238 * radix_tree_exceptional_entry - radix_tree_deref_slot gave exceptional entry? 239 * @arg: value returned by radix_tree_deref_slot 240 * Returns: 0 if well-aligned pointer, non-0 if exceptional entry. 241 */ 242static inline int radix_tree_exceptional_entry(void *arg) 243{ 244 /* Not unlikely because radix_tree_exception often tested first */ 245 return (unsigned long)arg & RADIX_TREE_EXCEPTIONAL_ENTRY; 246} 247 248/** 249 * radix_tree_exception - radix_tree_deref_slot returned either exception? 250 * @arg: value returned by radix_tree_deref_slot 251 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception. 252 */ 253static inline int radix_tree_exception(void *arg) 254{ 255 return unlikely((unsigned long)arg & 256 (RADIX_TREE_INDIRECT_PTR | RADIX_TREE_EXCEPTIONAL_ENTRY)); 257} 258 259/** 260 * radix_tree_replace_slot - replace item in a slot 261 * @pslot: pointer to slot, returned by radix_tree_lookup_slot 262 * @item: new item to store in the slot. 263 * 264 * For use with radix_tree_lookup_slot(). Caller must hold tree write locked 265 * across slot lookup and replacement. 266 */ 267static inline void radix_tree_replace_slot(void **pslot, void *item) 268{ 269 BUG_ON(radix_tree_is_indirect_ptr(item)); 270 rcu_assign_pointer(*pslot, item); 271} 272 273int __radix_tree_create(struct radix_tree_root *root, unsigned long index, 274 unsigned order, struct radix_tree_node **nodep, 275 void ***slotp); 276int __radix_tree_insert(struct radix_tree_root *, unsigned long index, 277 unsigned order, void *); 278static inline int radix_tree_insert(struct radix_tree_root *root, 279 unsigned long index, void *entry) 280{ 281 return __radix_tree_insert(root, index, 0, entry); 282} 283void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, 284 struct radix_tree_node **nodep, void ***slotp); 285void *radix_tree_lookup(struct radix_tree_root *, unsigned long); 286void **radix_tree_lookup_slot(struct radix_tree_root *, unsigned long); 287bool __radix_tree_delete_node(struct radix_tree_root *root, 288 struct radix_tree_node *node); 289void *radix_tree_delete_item(struct radix_tree_root *, unsigned long, void *); 290void *radix_tree_delete(struct radix_tree_root *, unsigned long); 291unsigned int 292radix_tree_gang_lookup(struct radix_tree_root *root, void **results, 293 unsigned long first_index, unsigned int max_items); 294unsigned int radix_tree_gang_lookup_slot(struct radix_tree_root *root, 295 void ***results, unsigned long *indices, 296 unsigned long first_index, unsigned int max_items); 297int radix_tree_preload(gfp_t gfp_mask); 298int radix_tree_maybe_preload(gfp_t gfp_mask); 299void radix_tree_init(void); 300void *radix_tree_tag_set(struct radix_tree_root *root, 301 unsigned long index, unsigned int tag); 302void *radix_tree_tag_clear(struct radix_tree_root *root, 303 unsigned long index, unsigned int tag); 304int radix_tree_tag_get(struct radix_tree_root *root, 305 unsigned long index, unsigned int tag); 306unsigned int 307radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, 308 unsigned long first_index, unsigned int max_items, 309 unsigned int tag); 310unsigned int 311radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, 312 unsigned long first_index, unsigned int max_items, 313 unsigned int tag); 314unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, 315 unsigned long *first_indexp, unsigned long last_index, 316 unsigned long nr_to_tag, 317 unsigned int fromtag, unsigned int totag); 318int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag); 319unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item); 320 321static inline void radix_tree_preload_end(void) 322{ 323 preempt_enable(); 324} 325 326/** 327 * struct radix_tree_iter - radix tree iterator state 328 * 329 * @index: index of current slot 330 * @next_index: next-to-last index for this chunk 331 * @tags: bit-mask for tag-iterating 332 * 333 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a 334 * subinterval of slots contained within one radix tree leaf node. It is 335 * described by a pointer to its first slot and a struct radix_tree_iter 336 * which holds the chunk's position in the tree and its size. For tagged 337 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen 338 * radix tree tag. 339 */ 340struct radix_tree_iter { 341 unsigned long index; 342 unsigned long next_index; 343 unsigned long tags; 344}; 345 346#define RADIX_TREE_ITER_TAG_MASK 0x00FF /* tag index in lower byte */ 347#define RADIX_TREE_ITER_TAGGED 0x0100 /* lookup tagged slots */ 348#define RADIX_TREE_ITER_CONTIG 0x0200 /* stop at first hole */ 349 350/** 351 * radix_tree_iter_init - initialize radix tree iterator 352 * 353 * @iter: pointer to iterator state 354 * @start: iteration starting index 355 * Returns: NULL 356 */ 357static __always_inline void ** 358radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start) 359{ 360 /* 361 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it 362 * in the case of a successful tagged chunk lookup. If the lookup was 363 * unsuccessful or non-tagged then nobody cares about ->tags. 364 * 365 * Set index to zero to bypass next_index overflow protection. 366 * See the comment in radix_tree_next_chunk() for details. 367 */ 368 iter->index = 0; 369 iter->next_index = start; 370 return NULL; 371} 372 373/** 374 * radix_tree_next_chunk - find next chunk of slots for iteration 375 * 376 * @root: radix tree root 377 * @iter: iterator state 378 * @flags: RADIX_TREE_ITER_* flags and tag index 379 * Returns: pointer to chunk first slot, or NULL if there no more left 380 * 381 * This function looks up the next chunk in the radix tree starting from 382 * @iter->next_index. It returns a pointer to the chunk's first slot. 383 * Also it fills @iter with data about chunk: position in the tree (index), 384 * its end (next_index), and constructs a bit mask for tagged iterating (tags). 385 */ 386void **radix_tree_next_chunk(struct radix_tree_root *root, 387 struct radix_tree_iter *iter, unsigned flags); 388 389/** 390 * radix_tree_iter_retry - retry this chunk of the iteration 391 * @iter: iterator state 392 * 393 * If we iterate over a tree protected only by the RCU lock, a race 394 * against deletion or creation may result in seeing a slot for which 395 * radix_tree_deref_retry() returns true. If so, call this function 396 * and continue the iteration. 397 */ 398static inline __must_check 399void **radix_tree_iter_retry(struct radix_tree_iter *iter) 400{ 401 iter->next_index = iter->index; 402 return NULL; 403} 404 405/** 406 * radix_tree_iter_next - resume iterating when the chunk may be invalid 407 * @iter: iterator state 408 * 409 * If the iterator needs to release then reacquire a lock, the chunk may 410 * have been invalidated by an insertion or deletion. Call this function 411 * to continue the iteration from the next index. 412 */ 413static inline __must_check 414void **radix_tree_iter_next(struct radix_tree_iter *iter) 415{ 416 iter->next_index = iter->index + 1; 417 iter->tags = 0; 418 return NULL; 419} 420 421/** 422 * radix_tree_chunk_size - get current chunk size 423 * 424 * @iter: pointer to radix tree iterator 425 * Returns: current chunk size 426 */ 427static __always_inline long 428radix_tree_chunk_size(struct radix_tree_iter *iter) 429{ 430 return iter->next_index - iter->index; 431} 432 433/** 434 * radix_tree_next_slot - find next slot in chunk 435 * 436 * @slot: pointer to current slot 437 * @iter: pointer to interator state 438 * @flags: RADIX_TREE_ITER_*, should be constant 439 * Returns: pointer to next slot, or NULL if there no more left 440 * 441 * This function updates @iter->index in the case of a successful lookup. 442 * For tagged lookup it also eats @iter->tags. 443 */ 444static __always_inline void ** 445radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, unsigned flags) 446{ 447 if (flags & RADIX_TREE_ITER_TAGGED) { 448 iter->tags >>= 1; 449 if (likely(iter->tags & 1ul)) { 450 iter->index++; 451 return slot + 1; 452 } 453 if (!(flags & RADIX_TREE_ITER_CONTIG) && likely(iter->tags)) { 454 unsigned offset = __ffs(iter->tags); 455 456 iter->tags >>= offset; 457 iter->index += offset + 1; 458 return slot + offset + 1; 459 } 460 } else { 461 long size = radix_tree_chunk_size(iter); 462 463 while (--size > 0) { 464 slot++; 465 iter->index++; 466 if (likely(*slot)) 467 return slot; 468 if (flags & RADIX_TREE_ITER_CONTIG) { 469 /* forbid switching to the next chunk */ 470 iter->next_index = 0; 471 break; 472 } 473 } 474 } 475 return NULL; 476} 477 478/** 479 * radix_tree_for_each_chunk - iterate over chunks 480 * 481 * @slot: the void** variable for pointer to chunk first slot 482 * @root: the struct radix_tree_root pointer 483 * @iter: the struct radix_tree_iter pointer 484 * @start: iteration starting index 485 * @flags: RADIX_TREE_ITER_* and tag index 486 * 487 * Locks can be released and reacquired between iterations. 488 */ 489#define radix_tree_for_each_chunk(slot, root, iter, start, flags) \ 490 for (slot = radix_tree_iter_init(iter, start) ; \ 491 (slot = radix_tree_next_chunk(root, iter, flags)) ;) 492 493/** 494 * radix_tree_for_each_chunk_slot - iterate over slots in one chunk 495 * 496 * @slot: the void** variable, at the beginning points to chunk first slot 497 * @iter: the struct radix_tree_iter pointer 498 * @flags: RADIX_TREE_ITER_*, should be constant 499 * 500 * This macro is designed to be nested inside radix_tree_for_each_chunk(). 501 * @slot points to the radix tree slot, @iter->index contains its index. 502 */ 503#define radix_tree_for_each_chunk_slot(slot, iter, flags) \ 504 for (; slot ; slot = radix_tree_next_slot(slot, iter, flags)) 505 506/** 507 * radix_tree_for_each_slot - iterate over non-empty slots 508 * 509 * @slot: the void** variable for pointer to slot 510 * @root: the struct radix_tree_root pointer 511 * @iter: the struct radix_tree_iter pointer 512 * @start: iteration starting index 513 * 514 * @slot points to radix tree slot, @iter->index contains its index. 515 */ 516#define radix_tree_for_each_slot(slot, root, iter, start) \ 517 for (slot = radix_tree_iter_init(iter, start) ; \ 518 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \ 519 slot = radix_tree_next_slot(slot, iter, 0)) 520 521/** 522 * radix_tree_for_each_contig - iterate over contiguous slots 523 * 524 * @slot: the void** variable for pointer to slot 525 * @root: the struct radix_tree_root pointer 526 * @iter: the struct radix_tree_iter pointer 527 * @start: iteration starting index 528 * 529 * @slot points to radix tree slot, @iter->index contains its index. 530 */ 531#define radix_tree_for_each_contig(slot, root, iter, start) \ 532 for (slot = radix_tree_iter_init(iter, start) ; \ 533 slot || (slot = radix_tree_next_chunk(root, iter, \ 534 RADIX_TREE_ITER_CONTIG)) ; \ 535 slot = radix_tree_next_slot(slot, iter, \ 536 RADIX_TREE_ITER_CONTIG)) 537 538/** 539 * radix_tree_for_each_tagged - iterate over tagged slots 540 * 541 * @slot: the void** variable for pointer to slot 542 * @root: the struct radix_tree_root pointer 543 * @iter: the struct radix_tree_iter pointer 544 * @start: iteration starting index 545 * @tag: tag index 546 * 547 * @slot points to radix tree slot, @iter->index contains its index. 548 */ 549#define radix_tree_for_each_tagged(slot, root, iter, start, tag) \ 550 for (slot = radix_tree_iter_init(iter, start) ; \ 551 slot || (slot = radix_tree_next_chunk(root, iter, \ 552 RADIX_TREE_ITER_TAGGED | tag)) ; \ 553 slot = radix_tree_next_slot(slot, iter, \ 554 RADIX_TREE_ITER_TAGGED)) 555 556#endif /* _LINUX_RADIX_TREE_H */ 557