| 1 | /* |
| 2 | * Copyright (C) 2001 Momchil Velikov |
| 3 | * Portions Copyright (C) 2001 Christoph Hellwig |
| 4 | * Copyright (C) 2005 SGI, Christoph Lameter |
| 5 | * Copyright (C) 2006 Nick Piggin |
| 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 | |
| 22 | #include <linux/errno.h> |
| 23 | #include <linux/init.h> |
| 24 | #include <linux/kernel.h> |
| 25 | #include <linux/module.h> |
| 26 | #include <linux/radix-tree.h> |
| 27 | #include <linux/percpu.h> |
| 28 | #include <linux/slab.h> |
| 29 | #include <linux/notifier.h> |
| 30 | #include <linux/cpu.h> |
| 31 | #include <linux/string.h> |
| 32 | #include <linux/bitops.h> |
| 33 | #include <linux/rcupdate.h> |
| 34 | |
| 35 | |
| 36 | #ifdef __KERNEL__ |
| 37 | #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6) |
| 38 | #else |
| 39 | #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */ |
| 40 | #endif |
| 41 | |
| 42 | #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT) |
| 43 | #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1) |
| 44 | |
| 45 | #define RADIX_TREE_TAG_LONGS \ |
| 46 | ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG) |
| 47 | |
| 48 | struct radix_tree_node { |
| 49 | unsigned int height; /* Height from the bottom */ |
| 50 | unsigned int count; |
| 51 | struct rcu_head rcu_head; |
| 52 | void __rcu *slots[RADIX_TREE_MAP_SIZE]; |
| 53 | unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS]; |
| 54 | }; |
| 55 | |
| 56 | struct radix_tree_path { |
| 57 | struct radix_tree_node *node; |
| 58 | int offset; |
| 59 | }; |
| 60 | |
| 61 | #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long)) |
| 62 | #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \ |
| 63 | RADIX_TREE_MAP_SHIFT)) |
| 64 | |
| 65 | /* |
| 66 | * The height_to_maxindex array needs to be one deeper than the maximum |
| 67 | * path as height 0 holds only 1 entry. |
| 68 | */ |
| 69 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; |
| 70 | |
| 71 | /* |
| 72 | * Radix tree node cache. |
| 73 | */ |
| 74 | static struct kmem_cache *radix_tree_node_cachep; |
| 75 | |
| 76 | /* |
| 77 | * Per-cpu pool of preloaded nodes |
| 78 | */ |
| 79 | struct radix_tree_preload { |
| 80 | int nr; |
| 81 | struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH]; |
| 82 | }; |
| 83 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
| 84 | |
| 85 | static inline void *ptr_to_indirect(void *ptr) |
| 86 | { |
| 87 | return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR); |
| 88 | } |
| 89 | |
| 90 | static inline void *indirect_to_ptr(void *ptr) |
| 91 | { |
| 92 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR); |
| 93 | } |
| 94 | |
| 95 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
| 96 | { |
| 97 | return root->gfp_mask & __GFP_BITS_MASK; |
| 98 | } |
| 99 | |
| 100 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
| 101 | int offset) |
| 102 | { |
| 103 | __set_bit(offset, node->tags[tag]); |
| 104 | } |
| 105 | |
| 106 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, |
| 107 | int offset) |
| 108 | { |
| 109 | __clear_bit(offset, node->tags[tag]); |
| 110 | } |
| 111 | |
| 112 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, |
| 113 | int offset) |
| 114 | { |
| 115 | return test_bit(offset, node->tags[tag]); |
| 116 | } |
| 117 | |
| 118 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) |
| 119 | { |
| 120 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); |
| 121 | } |
| 122 | |
| 123 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) |
| 124 | { |
| 125 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); |
| 126 | } |
| 127 | |
| 128 | static inline void root_tag_clear_all(struct radix_tree_root *root) |
| 129 | { |
| 130 | root->gfp_mask &= __GFP_BITS_MASK; |
| 131 | } |
| 132 | |
| 133 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) |
| 134 | { |
| 135 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); |
| 136 | } |
| 137 | |
| 138 | /* |
| 139 | * Returns 1 if any slot in the node has this tag set. |
| 140 | * Otherwise returns 0. |
| 141 | */ |
| 142 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) |
| 143 | { |
| 144 | int idx; |
| 145 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
| 146 | if (node->tags[tag][idx]) |
| 147 | return 1; |
| 148 | } |
| 149 | return 0; |
| 150 | } |
| 151 | /* |
| 152 | * This assumes that the caller has performed appropriate preallocation, and |
| 153 | * that the caller has pinned this thread of control to the current CPU. |
| 154 | */ |
| 155 | static struct radix_tree_node * |
| 156 | radix_tree_node_alloc(struct radix_tree_root *root) |
| 157 | { |
| 158 | struct radix_tree_node *ret = NULL; |
| 159 | gfp_t gfp_mask = root_gfp_mask(root); |
| 160 | |
| 161 | if (!(gfp_mask & __GFP_WAIT)) { |
| 162 | struct radix_tree_preload *rtp; |
| 163 | |
| 164 | /* |
| 165 | * Provided the caller has preloaded here, we will always |
| 166 | * succeed in getting a node here (and never reach |
| 167 | * kmem_cache_alloc) |
| 168 | */ |
| 169 | rtp = &__get_cpu_var(radix_tree_preloads); |
| 170 | if (rtp->nr) { |
| 171 | ret = rtp->nodes[rtp->nr - 1]; |
| 172 | rtp->nodes[rtp->nr - 1] = NULL; |
| 173 | rtp->nr--; |
| 174 | } |
| 175 | } |
| 176 | if (ret == NULL) |
| 177 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
| 178 | |
| 179 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
| 180 | return ret; |
| 181 | } |
| 182 | |
| 183 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
| 184 | { |
| 185 | struct radix_tree_node *node = |
| 186 | container_of(head, struct radix_tree_node, rcu_head); |
| 187 | int i; |
| 188 | |
| 189 | /* |
| 190 | * must only free zeroed nodes into the slab. radix_tree_shrink |
| 191 | * can leave us with a non-NULL entry in the first slot, so clear |
| 192 | * that here to make sure. |
| 193 | */ |
| 194 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
| 195 | tag_clear(node, i, 0); |
| 196 | |
| 197 | node->slots[0] = NULL; |
| 198 | node->count = 0; |
| 199 | |
| 200 | kmem_cache_free(radix_tree_node_cachep, node); |
| 201 | } |
| 202 | |
| 203 | static inline void |
| 204 | radix_tree_node_free(struct radix_tree_node *node) |
| 205 | { |
| 206 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
| 207 | } |
| 208 | |
| 209 | /* |
| 210 | * Load up this CPU's radix_tree_node buffer with sufficient objects to |
| 211 | * ensure that the addition of a single element in the tree cannot fail. On |
| 212 | * success, return zero, with preemption disabled. On error, return -ENOMEM |
| 213 | * with preemption not disabled. |
| 214 | * |
| 215 | * To make use of this facility, the radix tree must be initialised without |
| 216 | * __GFP_WAIT being passed to INIT_RADIX_TREE(). |
| 217 | */ |
| 218 | int radix_tree_preload(gfp_t gfp_mask) |
| 219 | { |
| 220 | struct radix_tree_preload *rtp; |
| 221 | struct radix_tree_node *node; |
| 222 | int ret = -ENOMEM; |
| 223 | |
| 224 | preempt_disable(); |
| 225 | rtp = &__get_cpu_var(radix_tree_preloads); |
| 226 | while (rtp->nr < ARRAY_SIZE(rtp->nodes)) { |
| 227 | preempt_enable(); |
| 228 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
| 229 | if (node == NULL) |
| 230 | goto out; |
| 231 | preempt_disable(); |
| 232 | rtp = &__get_cpu_var(radix_tree_preloads); |
| 233 | if (rtp->nr < ARRAY_SIZE(rtp->nodes)) |
| 234 | rtp->nodes[rtp->nr++] = node; |
| 235 | else |
| 236 | kmem_cache_free(radix_tree_node_cachep, node); |
| 237 | } |
| 238 | ret = 0; |
| 239 | out: |
| 240 | return ret; |
| 241 | } |
| 242 | EXPORT_SYMBOL(radix_tree_preload); |
| 243 | |
| 244 | /* |
| 245 | * Return the maximum key which can be store into a |
| 246 | * radix tree with height HEIGHT. |
| 247 | */ |
| 248 | static inline unsigned long radix_tree_maxindex(unsigned int height) |
| 249 | { |
| 250 | return height_to_maxindex[height]; |
| 251 | } |
| 252 | |
| 253 | /* |
| 254 | * Extend a radix tree so it can store key @index. |
| 255 | */ |
| 256 | static int radix_tree_extend(struct radix_tree_root *root, unsigned long index) |
| 257 | { |
| 258 | struct radix_tree_node *node; |
| 259 | unsigned int height; |
| 260 | int tag; |
| 261 | |
| 262 | /* Figure out what the height should be. */ |
| 263 | height = root->height + 1; |
| 264 | while (index > radix_tree_maxindex(height)) |
| 265 | height++; |
| 266 | |
| 267 | if (root->rnode == NULL) { |
| 268 | root->height = height; |
| 269 | goto out; |
| 270 | } |
| 271 | |
| 272 | do { |
| 273 | unsigned int newheight; |
| 274 | if (!(node = radix_tree_node_alloc(root))) |
| 275 | return -ENOMEM; |
| 276 | |
| 277 | /* Increase the height. */ |
| 278 | node->slots[0] = indirect_to_ptr(root->rnode); |
| 279 | |
| 280 | /* Propagate the aggregated tag info into the new root */ |
| 281 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
| 282 | if (root_tag_get(root, tag)) |
| 283 | tag_set(node, tag, 0); |
| 284 | } |
| 285 | |
| 286 | newheight = root->height+1; |
| 287 | node->height = newheight; |
| 288 | node->count = 1; |
| 289 | node = ptr_to_indirect(node); |
| 290 | rcu_assign_pointer(root->rnode, node); |
| 291 | root->height = newheight; |
| 292 | } while (height > root->height); |
| 293 | out: |
| 294 | return 0; |
| 295 | } |
| 296 | |
| 297 | /** |
| 298 | * radix_tree_insert - insert into a radix tree |
| 299 | * @root: radix tree root |
| 300 | * @index: index key |
| 301 | * @item: item to insert |
| 302 | * |
| 303 | * Insert an item into the radix tree at position @index. |
| 304 | */ |
| 305 | int radix_tree_insert(struct radix_tree_root *root, |
| 306 | unsigned long index, void *item) |
| 307 | { |
| 308 | struct radix_tree_node *node = NULL, *slot; |
| 309 | unsigned int height, shift; |
| 310 | int offset; |
| 311 | int error; |
| 312 | |
| 313 | BUG_ON(radix_tree_is_indirect_ptr(item)); |
| 314 | |
| 315 | /* Make sure the tree is high enough. */ |
| 316 | if (index > radix_tree_maxindex(root->height)) { |
| 317 | error = radix_tree_extend(root, index); |
| 318 | if (error) |
| 319 | return error; |
| 320 | } |
| 321 | |
| 322 | slot = indirect_to_ptr(root->rnode); |
| 323 | |
| 324 | height = root->height; |
| 325 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
| 326 | |
| 327 | offset = 0; /* uninitialised var warning */ |
| 328 | while (height > 0) { |
| 329 | if (slot == NULL) { |
| 330 | /* Have to add a child node. */ |
| 331 | if (!(slot = radix_tree_node_alloc(root))) |
| 332 | return -ENOMEM; |
| 333 | slot->height = height; |
| 334 | if (node) { |
| 335 | rcu_assign_pointer(node->slots[offset], slot); |
| 336 | node->count++; |
| 337 | } else |
| 338 | rcu_assign_pointer(root->rnode, ptr_to_indirect(slot)); |
| 339 | } |
| 340 | |
| 341 | /* Go a level down */ |
| 342 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 343 | node = slot; |
| 344 | slot = node->slots[offset]; |
| 345 | shift -= RADIX_TREE_MAP_SHIFT; |
| 346 | height--; |
| 347 | } |
| 348 | |
| 349 | if (slot != NULL) |
| 350 | return -EEXIST; |
| 351 | |
| 352 | if (node) { |
| 353 | node->count++; |
| 354 | rcu_assign_pointer(node->slots[offset], item); |
| 355 | BUG_ON(tag_get(node, 0, offset)); |
| 356 | BUG_ON(tag_get(node, 1, offset)); |
| 357 | } else { |
| 358 | rcu_assign_pointer(root->rnode, item); |
| 359 | BUG_ON(root_tag_get(root, 0)); |
| 360 | BUG_ON(root_tag_get(root, 1)); |
| 361 | } |
| 362 | |
| 363 | return 0; |
| 364 | } |
| 365 | EXPORT_SYMBOL(radix_tree_insert); |
| 366 | |
| 367 | /* |
| 368 | * is_slot == 1 : search for the slot. |
| 369 | * is_slot == 0 : search for the node. |
| 370 | */ |
| 371 | static void *radix_tree_lookup_element(struct radix_tree_root *root, |
| 372 | unsigned long index, int is_slot) |
| 373 | { |
| 374 | unsigned int height, shift; |
| 375 | struct radix_tree_node *node, **slot; |
| 376 | |
| 377 | node = rcu_dereference_raw(root->rnode); |
| 378 | if (node == NULL) |
| 379 | return NULL; |
| 380 | |
| 381 | if (!radix_tree_is_indirect_ptr(node)) { |
| 382 | if (index > 0) |
| 383 | return NULL; |
| 384 | return is_slot ? (void *)&root->rnode : node; |
| 385 | } |
| 386 | node = indirect_to_ptr(node); |
| 387 | |
| 388 | height = node->height; |
| 389 | if (index > radix_tree_maxindex(height)) |
| 390 | return NULL; |
| 391 | |
| 392 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
| 393 | |
| 394 | do { |
| 395 | slot = (struct radix_tree_node **) |
| 396 | (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK)); |
| 397 | node = rcu_dereference_raw(*slot); |
| 398 | if (node == NULL) |
| 399 | return NULL; |
| 400 | |
| 401 | shift -= RADIX_TREE_MAP_SHIFT; |
| 402 | height--; |
| 403 | } while (height > 0); |
| 404 | |
| 405 | return is_slot ? (void *)slot : indirect_to_ptr(node); |
| 406 | } |
| 407 | |
| 408 | /** |
| 409 | * radix_tree_lookup_slot - lookup a slot in a radix tree |
| 410 | * @root: radix tree root |
| 411 | * @index: index key |
| 412 | * |
| 413 | * Returns: the slot corresponding to the position @index in the |
| 414 | * radix tree @root. This is useful for update-if-exists operations. |
| 415 | * |
| 416 | * This function can be called under rcu_read_lock iff the slot is not |
| 417 | * modified by radix_tree_replace_slot, otherwise it must be called |
| 418 | * exclusive from other writers. Any dereference of the slot must be done |
| 419 | * using radix_tree_deref_slot. |
| 420 | */ |
| 421 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) |
| 422 | { |
| 423 | return (void **)radix_tree_lookup_element(root, index, 1); |
| 424 | } |
| 425 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
| 426 | |
| 427 | /** |
| 428 | * radix_tree_lookup - perform lookup operation on a radix tree |
| 429 | * @root: radix tree root |
| 430 | * @index: index key |
| 431 | * |
| 432 | * Lookup the item at the position @index in the radix tree @root. |
| 433 | * |
| 434 | * This function can be called under rcu_read_lock, however the caller |
| 435 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free |
| 436 | * them safely). No RCU barriers are required to access or modify the |
| 437 | * returned item, however. |
| 438 | */ |
| 439 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) |
| 440 | { |
| 441 | return radix_tree_lookup_element(root, index, 0); |
| 442 | } |
| 443 | EXPORT_SYMBOL(radix_tree_lookup); |
| 444 | |
| 445 | /** |
| 446 | * radix_tree_tag_set - set a tag on a radix tree node |
| 447 | * @root: radix tree root |
| 448 | * @index: index key |
| 449 | * @tag: tag index |
| 450 | * |
| 451 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
| 452 | * corresponding to @index in the radix tree. From |
| 453 | * the root all the way down to the leaf node. |
| 454 | * |
| 455 | * Returns the address of the tagged item. Setting a tag on a not-present |
| 456 | * item is a bug. |
| 457 | */ |
| 458 | void *radix_tree_tag_set(struct radix_tree_root *root, |
| 459 | unsigned long index, unsigned int tag) |
| 460 | { |
| 461 | unsigned int height, shift; |
| 462 | struct radix_tree_node *slot; |
| 463 | |
| 464 | height = root->height; |
| 465 | BUG_ON(index > radix_tree_maxindex(height)); |
| 466 | |
| 467 | slot = indirect_to_ptr(root->rnode); |
| 468 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
| 469 | |
| 470 | while (height > 0) { |
| 471 | int offset; |
| 472 | |
| 473 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 474 | if (!tag_get(slot, tag, offset)) |
| 475 | tag_set(slot, tag, offset); |
| 476 | slot = slot->slots[offset]; |
| 477 | BUG_ON(slot == NULL); |
| 478 | shift -= RADIX_TREE_MAP_SHIFT; |
| 479 | height--; |
| 480 | } |
| 481 | |
| 482 | /* set the root's tag bit */ |
| 483 | if (slot && !root_tag_get(root, tag)) |
| 484 | root_tag_set(root, tag); |
| 485 | |
| 486 | return slot; |
| 487 | } |
| 488 | EXPORT_SYMBOL(radix_tree_tag_set); |
| 489 | |
| 490 | /** |
| 491 | * radix_tree_tag_clear - clear a tag on a radix tree node |
| 492 | * @root: radix tree root |
| 493 | * @index: index key |
| 494 | * @tag: tag index |
| 495 | * |
| 496 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
| 497 | * corresponding to @index in the radix tree. If |
| 498 | * this causes the leaf node to have no tags set then clear the tag in the |
| 499 | * next-to-leaf node, etc. |
| 500 | * |
| 501 | * Returns the address of the tagged item on success, else NULL. ie: |
| 502 | * has the same return value and semantics as radix_tree_lookup(). |
| 503 | */ |
| 504 | void *radix_tree_tag_clear(struct radix_tree_root *root, |
| 505 | unsigned long index, unsigned int tag) |
| 506 | { |
| 507 | /* |
| 508 | * The radix tree path needs to be one longer than the maximum path |
| 509 | * since the "list" is null terminated. |
| 510 | */ |
| 511 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; |
| 512 | struct radix_tree_node *slot = NULL; |
| 513 | unsigned int height, shift; |
| 514 | |
| 515 | height = root->height; |
| 516 | if (index > radix_tree_maxindex(height)) |
| 517 | goto out; |
| 518 | |
| 519 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
| 520 | pathp->node = NULL; |
| 521 | slot = indirect_to_ptr(root->rnode); |
| 522 | |
| 523 | while (height > 0) { |
| 524 | int offset; |
| 525 | |
| 526 | if (slot == NULL) |
| 527 | goto out; |
| 528 | |
| 529 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 530 | pathp[1].offset = offset; |
| 531 | pathp[1].node = slot; |
| 532 | slot = slot->slots[offset]; |
| 533 | pathp++; |
| 534 | shift -= RADIX_TREE_MAP_SHIFT; |
| 535 | height--; |
| 536 | } |
| 537 | |
| 538 | if (slot == NULL) |
| 539 | goto out; |
| 540 | |
| 541 | while (pathp->node) { |
| 542 | if (!tag_get(pathp->node, tag, pathp->offset)) |
| 543 | goto out; |
| 544 | tag_clear(pathp->node, tag, pathp->offset); |
| 545 | if (any_tag_set(pathp->node, tag)) |
| 546 | goto out; |
| 547 | pathp--; |
| 548 | } |
| 549 | |
| 550 | /* clear the root's tag bit */ |
| 551 | if (root_tag_get(root, tag)) |
| 552 | root_tag_clear(root, tag); |
| 553 | |
| 554 | out: |
| 555 | return slot; |
| 556 | } |
| 557 | EXPORT_SYMBOL(radix_tree_tag_clear); |
| 558 | |
| 559 | /** |
| 560 | * radix_tree_tag_get - get a tag on a radix tree node |
| 561 | * @root: radix tree root |
| 562 | * @index: index key |
| 563 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
| 564 | * |
| 565 | * Return values: |
| 566 | * |
| 567 | * 0: tag not present or not set |
| 568 | * 1: tag set |
| 569 | * |
| 570 | * Note that the return value of this function may not be relied on, even if |
| 571 | * the RCU lock is held, unless tag modification and node deletion are excluded |
| 572 | * from concurrency. |
| 573 | */ |
| 574 | int radix_tree_tag_get(struct radix_tree_root *root, |
| 575 | unsigned long index, unsigned int tag) |
| 576 | { |
| 577 | unsigned int height, shift; |
| 578 | struct radix_tree_node *node; |
| 579 | int saw_unset_tag = 0; |
| 580 | |
| 581 | /* check the root's tag bit */ |
| 582 | if (!root_tag_get(root, tag)) |
| 583 | return 0; |
| 584 | |
| 585 | node = rcu_dereference_raw(root->rnode); |
| 586 | if (node == NULL) |
| 587 | return 0; |
| 588 | |
| 589 | if (!radix_tree_is_indirect_ptr(node)) |
| 590 | return (index == 0); |
| 591 | node = indirect_to_ptr(node); |
| 592 | |
| 593 | height = node->height; |
| 594 | if (index > radix_tree_maxindex(height)) |
| 595 | return 0; |
| 596 | |
| 597 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
| 598 | |
| 599 | for ( ; ; ) { |
| 600 | int offset; |
| 601 | |
| 602 | if (node == NULL) |
| 603 | return 0; |
| 604 | |
| 605 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 606 | |
| 607 | /* |
| 608 | * This is just a debug check. Later, we can bale as soon as |
| 609 | * we see an unset tag. |
| 610 | */ |
| 611 | if (!tag_get(node, tag, offset)) |
| 612 | saw_unset_tag = 1; |
| 613 | if (height == 1) |
| 614 | return !!tag_get(node, tag, offset); |
| 615 | node = rcu_dereference_raw(node->slots[offset]); |
| 616 | shift -= RADIX_TREE_MAP_SHIFT; |
| 617 | height--; |
| 618 | } |
| 619 | } |
| 620 | EXPORT_SYMBOL(radix_tree_tag_get); |
| 621 | |
| 622 | /** |
| 623 | * radix_tree_range_tag_if_tagged - for each item in given range set given |
| 624 | * tag if item has another tag set |
| 625 | * @root: radix tree root |
| 626 | * @first_indexp: pointer to a starting index of a range to scan |
| 627 | * @last_index: last index of a range to scan |
| 628 | * @nr_to_tag: maximum number items to tag |
| 629 | * @iftag: tag index to test |
| 630 | * @settag: tag index to set if tested tag is set |
| 631 | * |
| 632 | * This function scans range of radix tree from first_index to last_index |
| 633 | * (inclusive). For each item in the range if iftag is set, the function sets |
| 634 | * also settag. The function stops either after tagging nr_to_tag items or |
| 635 | * after reaching last_index. |
| 636 | * |
| 637 | * The tags must be set from the leaf level only and propagated back up the |
| 638 | * path to the root. We must do this so that we resolve the full path before |
| 639 | * setting any tags on intermediate nodes. If we set tags as we descend, then |
| 640 | * we can get to the leaf node and find that the index that has the iftag |
| 641 | * set is outside the range we are scanning. This reults in dangling tags and |
| 642 | * can lead to problems with later tag operations (e.g. livelocks on lookups). |
| 643 | * |
| 644 | * The function returns number of leaves where the tag was set and sets |
| 645 | * *first_indexp to the first unscanned index. |
| 646 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
| 647 | * be prepared to handle that. |
| 648 | */ |
| 649 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, |
| 650 | unsigned long *first_indexp, unsigned long last_index, |
| 651 | unsigned long nr_to_tag, |
| 652 | unsigned int iftag, unsigned int settag) |
| 653 | { |
| 654 | unsigned int height = root->height; |
| 655 | struct radix_tree_path path[height]; |
| 656 | struct radix_tree_path *pathp = path; |
| 657 | struct radix_tree_node *slot; |
| 658 | unsigned int shift; |
| 659 | unsigned long tagged = 0; |
| 660 | unsigned long index = *first_indexp; |
| 661 | |
| 662 | last_index = min(last_index, radix_tree_maxindex(height)); |
| 663 | if (index > last_index) |
| 664 | return 0; |
| 665 | if (!nr_to_tag) |
| 666 | return 0; |
| 667 | if (!root_tag_get(root, iftag)) { |
| 668 | *first_indexp = last_index + 1; |
| 669 | return 0; |
| 670 | } |
| 671 | if (height == 0) { |
| 672 | *first_indexp = last_index + 1; |
| 673 | root_tag_set(root, settag); |
| 674 | return 1; |
| 675 | } |
| 676 | |
| 677 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
| 678 | slot = indirect_to_ptr(root->rnode); |
| 679 | |
| 680 | /* |
| 681 | * we fill the path from (root->height - 2) to 0, leaving the index at |
| 682 | * (root->height - 1) as a terminator. Zero the node in the terminator |
| 683 | * so that we can use this to end walk loops back up the path. |
| 684 | */ |
| 685 | path[height - 1].node = NULL; |
| 686 | |
| 687 | for (;;) { |
| 688 | int offset; |
| 689 | |
| 690 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 691 | if (!slot->slots[offset]) |
| 692 | goto next; |
| 693 | if (!tag_get(slot, iftag, offset)) |
| 694 | goto next; |
| 695 | if (height > 1) { |
| 696 | /* Go down one level */ |
| 697 | height--; |
| 698 | shift -= RADIX_TREE_MAP_SHIFT; |
| 699 | path[height - 1].node = slot; |
| 700 | path[height - 1].offset = offset; |
| 701 | slot = slot->slots[offset]; |
| 702 | continue; |
| 703 | } |
| 704 | |
| 705 | /* tag the leaf */ |
| 706 | tagged++; |
| 707 | tag_set(slot, settag, offset); |
| 708 | |
| 709 | /* walk back up the path tagging interior nodes */ |
| 710 | pathp = &path[0]; |
| 711 | while (pathp->node) { |
| 712 | /* stop if we find a node with the tag already set */ |
| 713 | if (tag_get(pathp->node, settag, pathp->offset)) |
| 714 | break; |
| 715 | tag_set(pathp->node, settag, pathp->offset); |
| 716 | pathp++; |
| 717 | } |
| 718 | |
| 719 | next: |
| 720 | /* Go to next item at level determined by 'shift' */ |
| 721 | index = ((index >> shift) + 1) << shift; |
| 722 | /* Overflow can happen when last_index is ~0UL... */ |
| 723 | if (index > last_index || !index) |
| 724 | break; |
| 725 | if (tagged >= nr_to_tag) |
| 726 | break; |
| 727 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { |
| 728 | /* |
| 729 | * We've fully scanned this node. Go up. Because |
| 730 | * last_index is guaranteed to be in the tree, what |
| 731 | * we do below cannot wander astray. |
| 732 | */ |
| 733 | slot = path[height - 1].node; |
| 734 | height++; |
| 735 | shift += RADIX_TREE_MAP_SHIFT; |
| 736 | } |
| 737 | } |
| 738 | /* |
| 739 | * We need not to tag the root tag if there is no tag which is set with |
| 740 | * settag within the range from *first_indexp to last_index. |
| 741 | */ |
| 742 | if (tagged > 0) |
| 743 | root_tag_set(root, settag); |
| 744 | *first_indexp = index; |
| 745 | |
| 746 | return tagged; |
| 747 | } |
| 748 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); |
| 749 | |
| 750 | |
| 751 | /** |
| 752 | * radix_tree_next_hole - find the next hole (not-present entry) |
| 753 | * @root: tree root |
| 754 | * @index: index key |
| 755 | * @max_scan: maximum range to search |
| 756 | * |
| 757 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest |
| 758 | * indexed hole. |
| 759 | * |
| 760 | * Returns: the index of the hole if found, otherwise returns an index |
| 761 | * outside of the set specified (in which case 'return - index >= max_scan' |
| 762 | * will be true). In rare cases of index wrap-around, 0 will be returned. |
| 763 | * |
| 764 | * radix_tree_next_hole may be called under rcu_read_lock. However, like |
| 765 | * radix_tree_gang_lookup, this will not atomically search a snapshot of |
| 766 | * the tree at a single point in time. For example, if a hole is created |
| 767 | * at index 5, then subsequently a hole is created at index 10, |
| 768 | * radix_tree_next_hole covering both indexes may return 10 if called |
| 769 | * under rcu_read_lock. |
| 770 | */ |
| 771 | unsigned long radix_tree_next_hole(struct radix_tree_root *root, |
| 772 | unsigned long index, unsigned long max_scan) |
| 773 | { |
| 774 | unsigned long i; |
| 775 | |
| 776 | for (i = 0; i < max_scan; i++) { |
| 777 | if (!radix_tree_lookup(root, index)) |
| 778 | break; |
| 779 | index++; |
| 780 | if (index == 0) |
| 781 | break; |
| 782 | } |
| 783 | |
| 784 | return index; |
| 785 | } |
| 786 | EXPORT_SYMBOL(radix_tree_next_hole); |
| 787 | |
| 788 | /** |
| 789 | * radix_tree_prev_hole - find the prev hole (not-present entry) |
| 790 | * @root: tree root |
| 791 | * @index: index key |
| 792 | * @max_scan: maximum range to search |
| 793 | * |
| 794 | * Search backwards in the range [max(index-max_scan+1, 0), index] |
| 795 | * for the first hole. |
| 796 | * |
| 797 | * Returns: the index of the hole if found, otherwise returns an index |
| 798 | * outside of the set specified (in which case 'index - return >= max_scan' |
| 799 | * will be true). In rare cases of wrap-around, ULONG_MAX will be returned. |
| 800 | * |
| 801 | * radix_tree_next_hole may be called under rcu_read_lock. However, like |
| 802 | * radix_tree_gang_lookup, this will not atomically search a snapshot of |
| 803 | * the tree at a single point in time. For example, if a hole is created |
| 804 | * at index 10, then subsequently a hole is created at index 5, |
| 805 | * radix_tree_prev_hole covering both indexes may return 5 if called under |
| 806 | * rcu_read_lock. |
| 807 | */ |
| 808 | unsigned long radix_tree_prev_hole(struct radix_tree_root *root, |
| 809 | unsigned long index, unsigned long max_scan) |
| 810 | { |
| 811 | unsigned long i; |
| 812 | |
| 813 | for (i = 0; i < max_scan; i++) { |
| 814 | if (!radix_tree_lookup(root, index)) |
| 815 | break; |
| 816 | index--; |
| 817 | if (index == ULONG_MAX) |
| 818 | break; |
| 819 | } |
| 820 | |
| 821 | return index; |
| 822 | } |
| 823 | EXPORT_SYMBOL(radix_tree_prev_hole); |
| 824 | |
| 825 | static unsigned int |
| 826 | __lookup(struct radix_tree_node *slot, void ***results, unsigned long *indices, |
| 827 | unsigned long index, unsigned int max_items, unsigned long *next_index) |
| 828 | { |
| 829 | unsigned int nr_found = 0; |
| 830 | unsigned int shift, height; |
| 831 | unsigned long i; |
| 832 | |
| 833 | height = slot->height; |
| 834 | if (height == 0) |
| 835 | goto out; |
| 836 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
| 837 | |
| 838 | for ( ; height > 1; height--) { |
| 839 | i = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 840 | for (;;) { |
| 841 | if (slot->slots[i] != NULL) |
| 842 | break; |
| 843 | index &= ~((1UL << shift) - 1); |
| 844 | index += 1UL << shift; |
| 845 | if (index == 0) |
| 846 | goto out; /* 32-bit wraparound */ |
| 847 | i++; |
| 848 | if (i == RADIX_TREE_MAP_SIZE) |
| 849 | goto out; |
| 850 | } |
| 851 | |
| 852 | shift -= RADIX_TREE_MAP_SHIFT; |
| 853 | slot = rcu_dereference_raw(slot->slots[i]); |
| 854 | if (slot == NULL) |
| 855 | goto out; |
| 856 | } |
| 857 | |
| 858 | /* Bottom level: grab some items */ |
| 859 | for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) { |
| 860 | if (slot->slots[i]) { |
| 861 | results[nr_found] = &(slot->slots[i]); |
| 862 | if (indices) |
| 863 | indices[nr_found] = index; |
| 864 | if (++nr_found == max_items) { |
| 865 | index++; |
| 866 | goto out; |
| 867 | } |
| 868 | } |
| 869 | index++; |
| 870 | } |
| 871 | out: |
| 872 | *next_index = index; |
| 873 | return nr_found; |
| 874 | } |
| 875 | |
| 876 | /** |
| 877 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree |
| 878 | * @root: radix tree root |
| 879 | * @results: where the results of the lookup are placed |
| 880 | * @first_index: start the lookup from this key |
| 881 | * @max_items: place up to this many items at *results |
| 882 | * |
| 883 | * Performs an index-ascending scan of the tree for present items. Places |
| 884 | * them at *@results and returns the number of items which were placed at |
| 885 | * *@results. |
| 886 | * |
| 887 | * The implementation is naive. |
| 888 | * |
| 889 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under |
| 890 | * rcu_read_lock. In this case, rather than the returned results being |
| 891 | * an atomic snapshot of the tree at a single point in time, the semantics |
| 892 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups |
| 893 | * have been issued in individual locks, and results stored in 'results'. |
| 894 | */ |
| 895 | unsigned int |
| 896 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, |
| 897 | unsigned long first_index, unsigned int max_items) |
| 898 | { |
| 899 | unsigned long max_index; |
| 900 | struct radix_tree_node *node; |
| 901 | unsigned long cur_index = first_index; |
| 902 | unsigned int ret; |
| 903 | |
| 904 | node = rcu_dereference_raw(root->rnode); |
| 905 | if (!node) |
| 906 | return 0; |
| 907 | |
| 908 | if (!radix_tree_is_indirect_ptr(node)) { |
| 909 | if (first_index > 0) |
| 910 | return 0; |
| 911 | results[0] = node; |
| 912 | return 1; |
| 913 | } |
| 914 | node = indirect_to_ptr(node); |
| 915 | |
| 916 | max_index = radix_tree_maxindex(node->height); |
| 917 | |
| 918 | ret = 0; |
| 919 | while (ret < max_items) { |
| 920 | unsigned int nr_found, slots_found, i; |
| 921 | unsigned long next_index; /* Index of next search */ |
| 922 | |
| 923 | if (cur_index > max_index) |
| 924 | break; |
| 925 | slots_found = __lookup(node, (void ***)results + ret, NULL, |
| 926 | cur_index, max_items - ret, &next_index); |
| 927 | nr_found = 0; |
| 928 | for (i = 0; i < slots_found; i++) { |
| 929 | struct radix_tree_node *slot; |
| 930 | slot = *(((void ***)results)[ret + i]); |
| 931 | if (!slot) |
| 932 | continue; |
| 933 | results[ret + nr_found] = |
| 934 | indirect_to_ptr(rcu_dereference_raw(slot)); |
| 935 | nr_found++; |
| 936 | } |
| 937 | ret += nr_found; |
| 938 | if (next_index == 0) |
| 939 | break; |
| 940 | cur_index = next_index; |
| 941 | } |
| 942 | |
| 943 | return ret; |
| 944 | } |
| 945 | EXPORT_SYMBOL(radix_tree_gang_lookup); |
| 946 | |
| 947 | /** |
| 948 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree |
| 949 | * @root: radix tree root |
| 950 | * @results: where the results of the lookup are placed |
| 951 | * @indices: where their indices should be placed (but usually NULL) |
| 952 | * @first_index: start the lookup from this key |
| 953 | * @max_items: place up to this many items at *results |
| 954 | * |
| 955 | * Performs an index-ascending scan of the tree for present items. Places |
| 956 | * their slots at *@results and returns the number of items which were |
| 957 | * placed at *@results. |
| 958 | * |
| 959 | * The implementation is naive. |
| 960 | * |
| 961 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must |
| 962 | * be dereferenced with radix_tree_deref_slot, and if using only RCU |
| 963 | * protection, radix_tree_deref_slot may fail requiring a retry. |
| 964 | */ |
| 965 | unsigned int |
| 966 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
| 967 | void ***results, unsigned long *indices, |
| 968 | unsigned long first_index, unsigned int max_items) |
| 969 | { |
| 970 | unsigned long max_index; |
| 971 | struct radix_tree_node *node; |
| 972 | unsigned long cur_index = first_index; |
| 973 | unsigned int ret; |
| 974 | |
| 975 | node = rcu_dereference_raw(root->rnode); |
| 976 | if (!node) |
| 977 | return 0; |
| 978 | |
| 979 | if (!radix_tree_is_indirect_ptr(node)) { |
| 980 | if (first_index > 0) |
| 981 | return 0; |
| 982 | results[0] = (void **)&root->rnode; |
| 983 | if (indices) |
| 984 | indices[0] = 0; |
| 985 | return 1; |
| 986 | } |
| 987 | node = indirect_to_ptr(node); |
| 988 | |
| 989 | max_index = radix_tree_maxindex(node->height); |
| 990 | |
| 991 | ret = 0; |
| 992 | while (ret < max_items) { |
| 993 | unsigned int slots_found; |
| 994 | unsigned long next_index; /* Index of next search */ |
| 995 | |
| 996 | if (cur_index > max_index) |
| 997 | break; |
| 998 | slots_found = __lookup(node, results + ret, |
| 999 | indices ? indices + ret : NULL, |
| 1000 | cur_index, max_items - ret, &next_index); |
| 1001 | ret += slots_found; |
| 1002 | if (next_index == 0) |
| 1003 | break; |
| 1004 | cur_index = next_index; |
| 1005 | } |
| 1006 | |
| 1007 | return ret; |
| 1008 | } |
| 1009 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); |
| 1010 | |
| 1011 | /* |
| 1012 | * FIXME: the two tag_get()s here should use find_next_bit() instead of |
| 1013 | * open-coding the search. |
| 1014 | */ |
| 1015 | static unsigned int |
| 1016 | __lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index, |
| 1017 | unsigned int max_items, unsigned long *next_index, unsigned int tag) |
| 1018 | { |
| 1019 | unsigned int nr_found = 0; |
| 1020 | unsigned int shift, height; |
| 1021 | |
| 1022 | height = slot->height; |
| 1023 | if (height == 0) |
| 1024 | goto out; |
| 1025 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
| 1026 | |
| 1027 | while (height > 0) { |
| 1028 | unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ; |
| 1029 | |
| 1030 | for (;;) { |
| 1031 | if (tag_get(slot, tag, i)) |
| 1032 | break; |
| 1033 | index &= ~((1UL << shift) - 1); |
| 1034 | index += 1UL << shift; |
| 1035 | if (index == 0) |
| 1036 | goto out; /* 32-bit wraparound */ |
| 1037 | i++; |
| 1038 | if (i == RADIX_TREE_MAP_SIZE) |
| 1039 | goto out; |
| 1040 | } |
| 1041 | height--; |
| 1042 | if (height == 0) { /* Bottom level: grab some items */ |
| 1043 | unsigned long j = index & RADIX_TREE_MAP_MASK; |
| 1044 | |
| 1045 | for ( ; j < RADIX_TREE_MAP_SIZE; j++) { |
| 1046 | index++; |
| 1047 | if (!tag_get(slot, tag, j)) |
| 1048 | continue; |
| 1049 | /* |
| 1050 | * Even though the tag was found set, we need to |
| 1051 | * recheck that we have a non-NULL node, because |
| 1052 | * if this lookup is lockless, it may have been |
| 1053 | * subsequently deleted. |
| 1054 | * |
| 1055 | * Similar care must be taken in any place that |
| 1056 | * lookup ->slots[x] without a lock (ie. can't |
| 1057 | * rely on its value remaining the same). |
| 1058 | */ |
| 1059 | if (slot->slots[j]) { |
| 1060 | results[nr_found++] = &(slot->slots[j]); |
| 1061 | if (nr_found == max_items) |
| 1062 | goto out; |
| 1063 | } |
| 1064 | } |
| 1065 | } |
| 1066 | shift -= RADIX_TREE_MAP_SHIFT; |
| 1067 | slot = rcu_dereference_raw(slot->slots[i]); |
| 1068 | if (slot == NULL) |
| 1069 | break; |
| 1070 | } |
| 1071 | out: |
| 1072 | *next_index = index; |
| 1073 | return nr_found; |
| 1074 | } |
| 1075 | |
| 1076 | /** |
| 1077 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree |
| 1078 | * based on a tag |
| 1079 | * @root: radix tree root |
| 1080 | * @results: where the results of the lookup are placed |
| 1081 | * @first_index: start the lookup from this key |
| 1082 | * @max_items: place up to this many items at *results |
| 1083 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
| 1084 | * |
| 1085 | * Performs an index-ascending scan of the tree for present items which |
| 1086 | * have the tag indexed by @tag set. Places the items at *@results and |
| 1087 | * returns the number of items which were placed at *@results. |
| 1088 | */ |
| 1089 | unsigned int |
| 1090 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, |
| 1091 | unsigned long first_index, unsigned int max_items, |
| 1092 | unsigned int tag) |
| 1093 | { |
| 1094 | struct radix_tree_node *node; |
| 1095 | unsigned long max_index; |
| 1096 | unsigned long cur_index = first_index; |
| 1097 | unsigned int ret; |
| 1098 | |
| 1099 | /* check the root's tag bit */ |
| 1100 | if (!root_tag_get(root, tag)) |
| 1101 | return 0; |
| 1102 | |
| 1103 | node = rcu_dereference_raw(root->rnode); |
| 1104 | if (!node) |
| 1105 | return 0; |
| 1106 | |
| 1107 | if (!radix_tree_is_indirect_ptr(node)) { |
| 1108 | if (first_index > 0) |
| 1109 | return 0; |
| 1110 | results[0] = node; |
| 1111 | return 1; |
| 1112 | } |
| 1113 | node = indirect_to_ptr(node); |
| 1114 | |
| 1115 | max_index = radix_tree_maxindex(node->height); |
| 1116 | |
| 1117 | ret = 0; |
| 1118 | while (ret < max_items) { |
| 1119 | unsigned int nr_found, slots_found, i; |
| 1120 | unsigned long next_index; /* Index of next search */ |
| 1121 | |
| 1122 | if (cur_index > max_index) |
| 1123 | break; |
| 1124 | slots_found = __lookup_tag(node, (void ***)results + ret, |
| 1125 | cur_index, max_items - ret, &next_index, tag); |
| 1126 | nr_found = 0; |
| 1127 | for (i = 0; i < slots_found; i++) { |
| 1128 | struct radix_tree_node *slot; |
| 1129 | slot = *(((void ***)results)[ret + i]); |
| 1130 | if (!slot) |
| 1131 | continue; |
| 1132 | results[ret + nr_found] = |
| 1133 | indirect_to_ptr(rcu_dereference_raw(slot)); |
| 1134 | nr_found++; |
| 1135 | } |
| 1136 | ret += nr_found; |
| 1137 | if (next_index == 0) |
| 1138 | break; |
| 1139 | cur_index = next_index; |
| 1140 | } |
| 1141 | |
| 1142 | return ret; |
| 1143 | } |
| 1144 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); |
| 1145 | |
| 1146 | /** |
| 1147 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a |
| 1148 | * radix tree based on a tag |
| 1149 | * @root: radix tree root |
| 1150 | * @results: where the results of the lookup are placed |
| 1151 | * @first_index: start the lookup from this key |
| 1152 | * @max_items: place up to this many items at *results |
| 1153 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
| 1154 | * |
| 1155 | * Performs an index-ascending scan of the tree for present items which |
| 1156 | * have the tag indexed by @tag set. Places the slots at *@results and |
| 1157 | * returns the number of slots which were placed at *@results. |
| 1158 | */ |
| 1159 | unsigned int |
| 1160 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, |
| 1161 | unsigned long first_index, unsigned int max_items, |
| 1162 | unsigned int tag) |
| 1163 | { |
| 1164 | struct radix_tree_node *node; |
| 1165 | unsigned long max_index; |
| 1166 | unsigned long cur_index = first_index; |
| 1167 | unsigned int ret; |
| 1168 | |
| 1169 | /* check the root's tag bit */ |
| 1170 | if (!root_tag_get(root, tag)) |
| 1171 | return 0; |
| 1172 | |
| 1173 | node = rcu_dereference_raw(root->rnode); |
| 1174 | if (!node) |
| 1175 | return 0; |
| 1176 | |
| 1177 | if (!radix_tree_is_indirect_ptr(node)) { |
| 1178 | if (first_index > 0) |
| 1179 | return 0; |
| 1180 | results[0] = (void **)&root->rnode; |
| 1181 | return 1; |
| 1182 | } |
| 1183 | node = indirect_to_ptr(node); |
| 1184 | |
| 1185 | max_index = radix_tree_maxindex(node->height); |
| 1186 | |
| 1187 | ret = 0; |
| 1188 | while (ret < max_items) { |
| 1189 | unsigned int slots_found; |
| 1190 | unsigned long next_index; /* Index of next search */ |
| 1191 | |
| 1192 | if (cur_index > max_index) |
| 1193 | break; |
| 1194 | slots_found = __lookup_tag(node, results + ret, |
| 1195 | cur_index, max_items - ret, &next_index, tag); |
| 1196 | ret += slots_found; |
| 1197 | if (next_index == 0) |
| 1198 | break; |
| 1199 | cur_index = next_index; |
| 1200 | } |
| 1201 | |
| 1202 | return ret; |
| 1203 | } |
| 1204 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); |
| 1205 | |
| 1206 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
| 1207 | #include <linux/sched.h> /* for cond_resched() */ |
| 1208 | |
| 1209 | /* |
| 1210 | * This linear search is at present only useful to shmem_unuse_inode(). |
| 1211 | */ |
| 1212 | static unsigned long __locate(struct radix_tree_node *slot, void *item, |
| 1213 | unsigned long index, unsigned long *found_index) |
| 1214 | { |
| 1215 | unsigned int shift, height; |
| 1216 | unsigned long i; |
| 1217 | |
| 1218 | height = slot->height; |
| 1219 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
| 1220 | |
| 1221 | for ( ; height > 1; height--) { |
| 1222 | i = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 1223 | for (;;) { |
| 1224 | if (slot->slots[i] != NULL) |
| 1225 | break; |
| 1226 | index &= ~((1UL << shift) - 1); |
| 1227 | index += 1UL << shift; |
| 1228 | if (index == 0) |
| 1229 | goto out; /* 32-bit wraparound */ |
| 1230 | i++; |
| 1231 | if (i == RADIX_TREE_MAP_SIZE) |
| 1232 | goto out; |
| 1233 | } |
| 1234 | |
| 1235 | shift -= RADIX_TREE_MAP_SHIFT; |
| 1236 | slot = rcu_dereference_raw(slot->slots[i]); |
| 1237 | if (slot == NULL) |
| 1238 | goto out; |
| 1239 | } |
| 1240 | |
| 1241 | /* Bottom level: check items */ |
| 1242 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { |
| 1243 | if (slot->slots[i] == item) { |
| 1244 | *found_index = index + i; |
| 1245 | index = 0; |
| 1246 | goto out; |
| 1247 | } |
| 1248 | } |
| 1249 | index += RADIX_TREE_MAP_SIZE; |
| 1250 | out: |
| 1251 | return index; |
| 1252 | } |
| 1253 | |
| 1254 | /** |
| 1255 | * radix_tree_locate_item - search through radix tree for item |
| 1256 | * @root: radix tree root |
| 1257 | * @item: item to be found |
| 1258 | * |
| 1259 | * Returns index where item was found, or -1 if not found. |
| 1260 | * Caller must hold no lock (since this time-consuming function needs |
| 1261 | * to be preemptible), and must check afterwards if item is still there. |
| 1262 | */ |
| 1263 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) |
| 1264 | { |
| 1265 | struct radix_tree_node *node; |
| 1266 | unsigned long max_index; |
| 1267 | unsigned long cur_index = 0; |
| 1268 | unsigned long found_index = -1; |
| 1269 | |
| 1270 | do { |
| 1271 | rcu_read_lock(); |
| 1272 | node = rcu_dereference_raw(root->rnode); |
| 1273 | if (!radix_tree_is_indirect_ptr(node)) { |
| 1274 | rcu_read_unlock(); |
| 1275 | if (node == item) |
| 1276 | found_index = 0; |
| 1277 | break; |
| 1278 | } |
| 1279 | |
| 1280 | node = indirect_to_ptr(node); |
| 1281 | max_index = radix_tree_maxindex(node->height); |
| 1282 | if (cur_index > max_index) |
| 1283 | break; |
| 1284 | |
| 1285 | cur_index = __locate(node, item, cur_index, &found_index); |
| 1286 | rcu_read_unlock(); |
| 1287 | cond_resched(); |
| 1288 | } while (cur_index != 0 && cur_index <= max_index); |
| 1289 | |
| 1290 | return found_index; |
| 1291 | } |
| 1292 | #else |
| 1293 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) |
| 1294 | { |
| 1295 | return -1; |
| 1296 | } |
| 1297 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ |
| 1298 | |
| 1299 | /** |
| 1300 | * radix_tree_shrink - shrink height of a radix tree to minimal |
| 1301 | * @root radix tree root |
| 1302 | */ |
| 1303 | static inline void radix_tree_shrink(struct radix_tree_root *root) |
| 1304 | { |
| 1305 | /* try to shrink tree height */ |
| 1306 | while (root->height > 0) { |
| 1307 | struct radix_tree_node *to_free = root->rnode; |
| 1308 | void *newptr; |
| 1309 | |
| 1310 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
| 1311 | to_free = indirect_to_ptr(to_free); |
| 1312 | |
| 1313 | /* |
| 1314 | * The candidate node has more than one child, or its child |
| 1315 | * is not at the leftmost slot, we cannot shrink. |
| 1316 | */ |
| 1317 | if (to_free->count != 1) |
| 1318 | break; |
| 1319 | if (!to_free->slots[0]) |
| 1320 | break; |
| 1321 | |
| 1322 | /* |
| 1323 | * We don't need rcu_assign_pointer(), since we are simply |
| 1324 | * moving the node from one part of the tree to another: if it |
| 1325 | * was safe to dereference the old pointer to it |
| 1326 | * (to_free->slots[0]), it will be safe to dereference the new |
| 1327 | * one (root->rnode) as far as dependent read barriers go. |
| 1328 | */ |
| 1329 | newptr = to_free->slots[0]; |
| 1330 | if (root->height > 1) |
| 1331 | newptr = ptr_to_indirect(newptr); |
| 1332 | root->rnode = newptr; |
| 1333 | root->height--; |
| 1334 | |
| 1335 | /* |
| 1336 | * We have a dilemma here. The node's slot[0] must not be |
| 1337 | * NULLed in case there are concurrent lookups expecting to |
| 1338 | * find the item. However if this was a bottom-level node, |
| 1339 | * then it may be subject to the slot pointer being visible |
| 1340 | * to callers dereferencing it. If item corresponding to |
| 1341 | * slot[0] is subsequently deleted, these callers would expect |
| 1342 | * their slot to become empty sooner or later. |
| 1343 | * |
| 1344 | * For example, lockless pagecache will look up a slot, deref |
| 1345 | * the page pointer, and if the page is 0 refcount it means it |
| 1346 | * was concurrently deleted from pagecache so try the deref |
| 1347 | * again. Fortunately there is already a requirement for logic |
| 1348 | * to retry the entire slot lookup -- the indirect pointer |
| 1349 | * problem (replacing direct root node with an indirect pointer |
| 1350 | * also results in a stale slot). So tag the slot as indirect |
| 1351 | * to force callers to retry. |
| 1352 | */ |
| 1353 | if (root->height == 0) |
| 1354 | *((unsigned long *)&to_free->slots[0]) |= |
| 1355 | RADIX_TREE_INDIRECT_PTR; |
| 1356 | |
| 1357 | radix_tree_node_free(to_free); |
| 1358 | } |
| 1359 | } |
| 1360 | |
| 1361 | /** |
| 1362 | * radix_tree_delete - delete an item from a radix tree |
| 1363 | * @root: radix tree root |
| 1364 | * @index: index key |
| 1365 | * |
| 1366 | * Remove the item at @index from the radix tree rooted at @root. |
| 1367 | * |
| 1368 | * Returns the address of the deleted item, or NULL if it was not present. |
| 1369 | */ |
| 1370 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) |
| 1371 | { |
| 1372 | /* |
| 1373 | * The radix tree path needs to be one longer than the maximum path |
| 1374 | * since the "list" is null terminated. |
| 1375 | */ |
| 1376 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; |
| 1377 | struct radix_tree_node *slot = NULL; |
| 1378 | struct radix_tree_node *to_free; |
| 1379 | unsigned int height, shift; |
| 1380 | int tag; |
| 1381 | int offset; |
| 1382 | |
| 1383 | height = root->height; |
| 1384 | if (index > radix_tree_maxindex(height)) |
| 1385 | goto out; |
| 1386 | |
| 1387 | slot = root->rnode; |
| 1388 | if (height == 0) { |
| 1389 | root_tag_clear_all(root); |
| 1390 | root->rnode = NULL; |
| 1391 | goto out; |
| 1392 | } |
| 1393 | slot = indirect_to_ptr(slot); |
| 1394 | |
| 1395 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
| 1396 | pathp->node = NULL; |
| 1397 | |
| 1398 | do { |
| 1399 | if (slot == NULL) |
| 1400 | goto out; |
| 1401 | |
| 1402 | pathp++; |
| 1403 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
| 1404 | pathp->offset = offset; |
| 1405 | pathp->node = slot; |
| 1406 | slot = slot->slots[offset]; |
| 1407 | shift -= RADIX_TREE_MAP_SHIFT; |
| 1408 | height--; |
| 1409 | } while (height > 0); |
| 1410 | |
| 1411 | if (slot == NULL) |
| 1412 | goto out; |
| 1413 | |
| 1414 | /* |
| 1415 | * Clear all tags associated with the just-deleted item |
| 1416 | */ |
| 1417 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
| 1418 | if (tag_get(pathp->node, tag, pathp->offset)) |
| 1419 | radix_tree_tag_clear(root, index, tag); |
| 1420 | } |
| 1421 | |
| 1422 | to_free = NULL; |
| 1423 | /* Now free the nodes we do not need anymore */ |
| 1424 | while (pathp->node) { |
| 1425 | pathp->node->slots[pathp->offset] = NULL; |
| 1426 | pathp->node->count--; |
| 1427 | /* |
| 1428 | * Queue the node for deferred freeing after the |
| 1429 | * last reference to it disappears (set NULL, above). |
| 1430 | */ |
| 1431 | if (to_free) |
| 1432 | radix_tree_node_free(to_free); |
| 1433 | |
| 1434 | if (pathp->node->count) { |
| 1435 | if (pathp->node == indirect_to_ptr(root->rnode)) |
| 1436 | radix_tree_shrink(root); |
| 1437 | goto out; |
| 1438 | } |
| 1439 | |
| 1440 | /* Node with zero slots in use so free it */ |
| 1441 | to_free = pathp->node; |
| 1442 | pathp--; |
| 1443 | |
| 1444 | } |
| 1445 | root_tag_clear_all(root); |
| 1446 | root->height = 0; |
| 1447 | root->rnode = NULL; |
| 1448 | if (to_free) |
| 1449 | radix_tree_node_free(to_free); |
| 1450 | |
| 1451 | out: |
| 1452 | return slot; |
| 1453 | } |
| 1454 | EXPORT_SYMBOL(radix_tree_delete); |
| 1455 | |
| 1456 | /** |
| 1457 | * radix_tree_tagged - test whether any items in the tree are tagged |
| 1458 | * @root: radix tree root |
| 1459 | * @tag: tag to test |
| 1460 | */ |
| 1461 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
| 1462 | { |
| 1463 | return root_tag_get(root, tag); |
| 1464 | } |
| 1465 | EXPORT_SYMBOL(radix_tree_tagged); |
| 1466 | |
| 1467 | static void |
| 1468 | radix_tree_node_ctor(void *node) |
| 1469 | { |
| 1470 | memset(node, 0, sizeof(struct radix_tree_node)); |
| 1471 | } |
| 1472 | |
| 1473 | static __init unsigned long __maxindex(unsigned int height) |
| 1474 | { |
| 1475 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
| 1476 | int shift = RADIX_TREE_INDEX_BITS - width; |
| 1477 | |
| 1478 | if (shift < 0) |
| 1479 | return ~0UL; |
| 1480 | if (shift >= BITS_PER_LONG) |
| 1481 | return 0UL; |
| 1482 | return ~0UL >> shift; |
| 1483 | } |
| 1484 | |
| 1485 | static __init void radix_tree_init_maxindex(void) |
| 1486 | { |
| 1487 | unsigned int i; |
| 1488 | |
| 1489 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) |
| 1490 | height_to_maxindex[i] = __maxindex(i); |
| 1491 | } |
| 1492 | |
| 1493 | static int radix_tree_callback(struct notifier_block *nfb, |
| 1494 | unsigned long action, |
| 1495 | void *hcpu) |
| 1496 | { |
| 1497 | int cpu = (long)hcpu; |
| 1498 | struct radix_tree_preload *rtp; |
| 1499 | |
| 1500 | /* Free per-cpu pool of perloaded nodes */ |
| 1501 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
| 1502 | rtp = &per_cpu(radix_tree_preloads, cpu); |
| 1503 | while (rtp->nr) { |
| 1504 | kmem_cache_free(radix_tree_node_cachep, |
| 1505 | rtp->nodes[rtp->nr-1]); |
| 1506 | rtp->nodes[rtp->nr-1] = NULL; |
| 1507 | rtp->nr--; |
| 1508 | } |
| 1509 | } |
| 1510 | return NOTIFY_OK; |
| 1511 | } |
| 1512 | |
| 1513 | void __init radix_tree_init(void) |
| 1514 | { |
| 1515 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
| 1516 | sizeof(struct radix_tree_node), 0, |
| 1517 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
| 1518 | radix_tree_node_ctor); |
| 1519 | radix_tree_init_maxindex(); |
| 1520 | hotcpu_notifier(radix_tree_callback, 0); |
| 1521 | } |