Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Elastic Binary Trees - macros to manipulate Indirect String data nodes. |
Willy Tarreau | f3bfede | 2011-07-25 11:38:17 +0200 | [diff] [blame] | 3 | * Version 6.0.6 |
Willy Tarreau | e1ee956 | 2011-01-04 14:33:13 +0100 | [diff] [blame] | 4 | * (C) 2002-2011 - Willy Tarreau <w@1wt.eu> |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 5 | * |
Willy Tarreau | f3bfede | 2011-07-25 11:38:17 +0200 | [diff] [blame] | 6 | * This library is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation, version 2.1 |
| 9 | * exclusively. |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 10 | * |
Willy Tarreau | f3bfede | 2011-07-25 11:38:17 +0200 | [diff] [blame] | 11 | * This library is distributed in the hope that it will be useful, |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
Willy Tarreau | f3bfede | 2011-07-25 11:38:17 +0200 | [diff] [blame] | 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 15 | * |
Willy Tarreau | f3bfede | 2011-07-25 11:38:17 +0200 | [diff] [blame] | 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with this library; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 19 | */ |
| 20 | |
| 21 | /* These functions and macros rely on Multi-Byte nodes */ |
| 22 | |
Willy Tarreau | 20a81c2 | 2014-03-15 07:43:05 +0100 | [diff] [blame] | 23 | #ifndef _EBISTREE_H |
| 24 | #define _EBISTREE_H |
| 25 | |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 26 | #include <string.h> |
| 27 | #include "ebtree.h" |
| 28 | #include "ebpttree.h" |
Willy Tarreau | e1ee956 | 2011-01-04 14:33:13 +0100 | [diff] [blame] | 29 | #include "ebimtree.h" |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 30 | |
| 31 | /* These functions and macros rely on Pointer nodes and use the <key> entry as |
| 32 | * a pointer to an indirect key. Most operations are performed using ebpt_*. |
| 33 | */ |
| 34 | |
| 35 | /* The following functions are not inlined by default. They are declared |
| 36 | * in ebistree.c, which simply relies on their inline version. |
| 37 | */ |
| 38 | REGPRM2 struct ebpt_node *ebis_lookup(struct eb_root *root, const char *x); |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 39 | REGPRM2 struct ebpt_node *ebis_insert(struct eb_root *root, struct ebpt_node *new); |
| 40 | |
Willy Tarreau | e1ee956 | 2011-01-04 14:33:13 +0100 | [diff] [blame] | 41 | /* Find the first occurence of a length <len> string <x> in the tree <root>. |
| 42 | * It's the caller's reponsibility to use this function only on trees which |
| 43 | * only contain zero-terminated strings, and that no null character is present |
| 44 | * in string <x> in the first <len> chars. If none can be found, return NULL. |
| 45 | */ |
| 46 | static forceinline struct ebpt_node * |
| 47 | ebis_lookup_len(struct eb_root *root, const char *x, unsigned int len) |
| 48 | { |
| 49 | struct ebpt_node *node; |
| 50 | |
| 51 | node = ebim_lookup(root, x, len); |
| 52 | if (!node || ((const char *)node->key)[len] != 0) |
| 53 | return NULL; |
| 54 | return node; |
| 55 | } |
| 56 | |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 57 | /* Find the first occurence of a zero-terminated string <x> in the tree <root>. |
| 58 | * It's the caller's reponsibility to use this function only on trees which |
| 59 | * only contain zero-terminated strings. If none can be found, return NULL. |
| 60 | */ |
| 61 | static forceinline struct ebpt_node *__ebis_lookup(struct eb_root *root, const void *x) |
| 62 | { |
| 63 | struct ebpt_node *node; |
| 64 | eb_troot_t *troot; |
Willy Tarreau | 3a93244 | 2010-05-09 19:29:23 +0200 | [diff] [blame] | 65 | int bit; |
| 66 | int node_bit; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 67 | |
| 68 | troot = root->b[EB_LEFT]; |
| 69 | if (unlikely(troot == NULL)) |
| 70 | return NULL; |
| 71 | |
| 72 | bit = 0; |
| 73 | while (1) { |
| 74 | if ((eb_gettag(troot) == EB_LEAF)) { |
| 75 | node = container_of(eb_untag(troot, EB_LEAF), |
| 76 | struct ebpt_node, node.branches); |
| 77 | if (strcmp(node->key, x) == 0) |
| 78 | return node; |
| 79 | else |
| 80 | return NULL; |
| 81 | } |
| 82 | node = container_of(eb_untag(troot, EB_NODE), |
| 83 | struct ebpt_node, node.branches); |
Willy Tarreau | 3a93244 | 2010-05-09 19:29:23 +0200 | [diff] [blame] | 84 | node_bit = node->node.bit; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 85 | |
Willy Tarreau | 3a93244 | 2010-05-09 19:29:23 +0200 | [diff] [blame] | 86 | if (node_bit < 0) { |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 87 | /* We have a dup tree now. Either it's for the same |
| 88 | * value, and we walk down left, or it's a different |
| 89 | * one and we don't have our key. |
| 90 | */ |
| 91 | if (strcmp(node->key, x) != 0) |
| 92 | return NULL; |
| 93 | |
| 94 | troot = node->node.branches.b[EB_LEFT]; |
| 95 | while (eb_gettag(troot) != EB_LEAF) |
| 96 | troot = (eb_untag(troot, EB_NODE))->b[EB_LEFT]; |
| 97 | node = container_of(eb_untag(troot, EB_LEAF), |
| 98 | struct ebpt_node, node.branches); |
| 99 | return node; |
| 100 | } |
| 101 | |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 102 | /* OK, normal data node, let's walk down but don't compare data |
| 103 | * if we already reached the end of the key. |
| 104 | */ |
| 105 | if (likely(bit >= 0)) { |
| 106 | bit = string_equal_bits(x, node->key, bit); |
| 107 | if (likely(bit < node_bit)) { |
| 108 | if (bit >= 0) |
| 109 | return NULL; /* no more common bits */ |
| 110 | |
| 111 | /* bit < 0 : we reached the end of the key. If we |
| 112 | * are in a tree with unique keys, we can return |
| 113 | * this node. Otherwise we have to walk it down |
| 114 | * and stop comparing bits. |
| 115 | */ |
| 116 | if (eb_gettag(root->b[EB_RGHT])) |
| 117 | return node; |
| 118 | } |
Willy Tarreau | 007257e | 2011-11-14 14:09:27 +0100 | [diff] [blame] | 119 | /* if the bit is larger than the node's, we must bound it |
| 120 | * because we might have compared too many bytes with an |
| 121 | * inappropriate leaf. For a test, build a tree from "0", |
| 122 | * "WW", "W", "S" inserted in this exact sequence and lookup |
| 123 | * "W" => "S" is returned without this assignment. |
| 124 | */ |
| 125 | else |
| 126 | bit = node_bit; |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 127 | } |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 128 | |
Willy Tarreau | 3a93244 | 2010-05-09 19:29:23 +0200 | [diff] [blame] | 129 | troot = node->node.branches.b[(((unsigned char*)x)[node_bit >> 3] >> |
| 130 | (~node_bit & 7)) & 1]; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 131 | } |
| 132 | } |
| 133 | |
| 134 | /* Insert ebpt_node <new> into subtree starting at node root <root>. Only |
| 135 | * new->key needs be set with the zero-terminated string key. The ebpt_node is |
| 136 | * returned. If root->b[EB_RGHT]==1, the tree may only contain unique keys. The |
| 137 | * caller is responsible for properly terminating the key with a zero. |
| 138 | */ |
| 139 | static forceinline struct ebpt_node * |
| 140 | __ebis_insert(struct eb_root *root, struct ebpt_node *new) |
| 141 | { |
| 142 | struct ebpt_node *old; |
| 143 | unsigned int side; |
| 144 | eb_troot_t *troot; |
Willy Tarreau | 6258f7b | 2011-09-19 20:48:00 +0200 | [diff] [blame] | 145 | eb_troot_t *root_right; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 146 | int diff; |
| 147 | int bit; |
Willy Tarreau | 3a93244 | 2010-05-09 19:29:23 +0200 | [diff] [blame] | 148 | int old_node_bit; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 149 | |
| 150 | side = EB_LEFT; |
| 151 | troot = root->b[EB_LEFT]; |
| 152 | root_right = root->b[EB_RGHT]; |
| 153 | if (unlikely(troot == NULL)) { |
| 154 | /* Tree is empty, insert the leaf part below the left branch */ |
| 155 | root->b[EB_LEFT] = eb_dotag(&new->node.branches, EB_LEAF); |
| 156 | new->node.leaf_p = eb_dotag(root, EB_LEFT); |
| 157 | new->node.node_p = NULL; /* node part unused */ |
| 158 | return new; |
| 159 | } |
| 160 | |
| 161 | /* The tree descent is fairly easy : |
| 162 | * - first, check if we have reached a leaf node |
| 163 | * - second, check if we have gone too far |
| 164 | * - third, reiterate |
| 165 | * Everywhere, we use <new> for the node node we are inserting, <root> |
| 166 | * for the node we attach it to, and <old> for the node we are |
| 167 | * displacing below <new>. <troot> will always point to the future node |
| 168 | * (tagged with its type). <side> carries the side the node <new> is |
| 169 | * attached to below its parent, which is also where previous node |
| 170 | * was attached. |
| 171 | */ |
| 172 | |
| 173 | bit = 0; |
| 174 | while (1) { |
| 175 | if (unlikely(eb_gettag(troot) == EB_LEAF)) { |
| 176 | eb_troot_t *new_left, *new_rght; |
| 177 | eb_troot_t *new_leaf, *old_leaf; |
| 178 | |
| 179 | old = container_of(eb_untag(troot, EB_LEAF), |
| 180 | struct ebpt_node, node.branches); |
| 181 | |
| 182 | new_left = eb_dotag(&new->node.branches, EB_LEFT); |
| 183 | new_rght = eb_dotag(&new->node.branches, EB_RGHT); |
| 184 | new_leaf = eb_dotag(&new->node.branches, EB_LEAF); |
| 185 | old_leaf = eb_dotag(&old->node.branches, EB_LEAF); |
| 186 | |
| 187 | new->node.node_p = old->node.leaf_p; |
| 188 | |
| 189 | /* Right here, we have 3 possibilities : |
| 190 | * - the tree does not contain the key, and we have |
| 191 | * new->key < old->key. We insert new above old, on |
| 192 | * the left ; |
| 193 | * |
| 194 | * - the tree does not contain the key, and we have |
| 195 | * new->key > old->key. We insert new above old, on |
| 196 | * the right ; |
| 197 | * |
| 198 | * - the tree does contain the key, which implies it |
| 199 | * is alone. We add the new key next to it as a |
| 200 | * first duplicate. |
| 201 | * |
| 202 | * The last two cases can easily be partially merged. |
| 203 | */ |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 204 | if (bit >= 0) |
| 205 | bit = string_equal_bits(new->key, old->key, bit); |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 206 | |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 207 | if (bit < 0) { |
| 208 | /* key was already there */ |
| 209 | |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 210 | /* we may refuse to duplicate this key if the tree is |
| 211 | * tagged as containing only unique keys. |
| 212 | */ |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 213 | if (eb_gettag(root_right)) |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 214 | return old; |
| 215 | |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 216 | /* new arbitrarily goes to the right and tops the dup tree */ |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 217 | old->node.leaf_p = new_left; |
| 218 | new->node.leaf_p = new_rght; |
| 219 | new->node.branches.b[EB_LEFT] = old_leaf; |
| 220 | new->node.branches.b[EB_RGHT] = new_leaf; |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 221 | new->node.bit = -1; |
| 222 | root->b[side] = eb_dotag(&new->node.branches, EB_NODE); |
| 223 | return new; |
| 224 | } |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 225 | |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 226 | diff = cmp_bits(new->key, old->key, bit); |
| 227 | if (diff < 0) { |
| 228 | /* new->key < old->key, new takes the left */ |
| 229 | new->node.leaf_p = new_left; |
| 230 | old->node.leaf_p = new_rght; |
| 231 | new->node.branches.b[EB_LEFT] = new_leaf; |
| 232 | new->node.branches.b[EB_RGHT] = old_leaf; |
| 233 | } else { |
| 234 | /* new->key > old->key, new takes the right */ |
| 235 | old->node.leaf_p = new_left; |
| 236 | new->node.leaf_p = new_rght; |
| 237 | new->node.branches.b[EB_LEFT] = old_leaf; |
| 238 | new->node.branches.b[EB_RGHT] = new_leaf; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 239 | } |
| 240 | break; |
| 241 | } |
| 242 | |
| 243 | /* OK we're walking down this link */ |
| 244 | old = container_of(eb_untag(troot, EB_NODE), |
| 245 | struct ebpt_node, node.branches); |
Willy Tarreau | 3a93244 | 2010-05-09 19:29:23 +0200 | [diff] [blame] | 246 | old_node_bit = old->node.bit; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 247 | |
| 248 | /* Stop going down when we don't have common bits anymore. We |
| 249 | * also stop in front of a duplicates tree because it means we |
| 250 | * have to insert above. Note: we can compare more bits than |
| 251 | * the current node's because as long as they are identical, we |
| 252 | * know we descend along the correct side. |
| 253 | */ |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 254 | if (bit >= 0 && (bit < old_node_bit || old_node_bit < 0)) |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 255 | bit = string_equal_bits(new->key, old->key, bit); |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 256 | |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 257 | if (unlikely(bit < 0)) { |
| 258 | /* Perfect match, we must only stop on head of dup tree |
| 259 | * or walk down to a leaf. |
| 260 | */ |
| 261 | if (old_node_bit < 0) { |
| 262 | /* We know here that string_equal_bits matched all |
| 263 | * bits and that we're on top of a dup tree, then |
| 264 | * we can perform the dup insertion and return. |
| 265 | */ |
| 266 | struct eb_node *ret; |
| 267 | ret = eb_insert_dup(&old->node, &new->node); |
| 268 | return container_of(ret, struct ebpt_node, node); |
| 269 | } |
| 270 | /* OK so let's walk down */ |
| 271 | } |
| 272 | else if (bit < old_node_bit || old_node_bit < 0) { |
| 273 | /* The tree did not contain the key, or we stopped on top of a dup |
| 274 | * tree, possibly containing the key. In the former case, we insert |
| 275 | * <new> before the node <old>, and set ->bit to designate the lowest |
| 276 | * bit position in <new> which applies to ->branches.b[]. In the later |
| 277 | * case, we add the key to the existing dup tree. Note that we cannot |
| 278 | * enter here if we match an intermediate node's key that is not the |
| 279 | * head of a dup tree. |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 280 | */ |
| 281 | eb_troot_t *new_left, *new_rght; |
| 282 | eb_troot_t *new_leaf, *old_node; |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 283 | |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 284 | new_left = eb_dotag(&new->node.branches, EB_LEFT); |
| 285 | new_rght = eb_dotag(&new->node.branches, EB_RGHT); |
| 286 | new_leaf = eb_dotag(&new->node.branches, EB_LEAF); |
| 287 | old_node = eb_dotag(&old->node.branches, EB_NODE); |
| 288 | |
| 289 | new->node.node_p = old->node.node_p; |
| 290 | |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 291 | /* we can never match all bits here */ |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 292 | diff = cmp_bits(new->key, old->key, bit); |
| 293 | if (diff < 0) { |
| 294 | new->node.leaf_p = new_left; |
| 295 | old->node.node_p = new_rght; |
| 296 | new->node.branches.b[EB_LEFT] = new_leaf; |
| 297 | new->node.branches.b[EB_RGHT] = old_node; |
| 298 | } |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 299 | else { |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 300 | old->node.node_p = new_left; |
| 301 | new->node.leaf_p = new_rght; |
| 302 | new->node.branches.b[EB_LEFT] = old_node; |
| 303 | new->node.branches.b[EB_RGHT] = new_leaf; |
| 304 | } |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 305 | break; |
| 306 | } |
| 307 | |
| 308 | /* walk down */ |
| 309 | root = &old->node.branches; |
Willy Tarreau | 3a93244 | 2010-05-09 19:29:23 +0200 | [diff] [blame] | 310 | side = (((unsigned char *)new->key)[old_node_bit >> 3] >> (~old_node_bit & 7)) & 1; |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 311 | troot = root->b[side]; |
| 312 | } |
| 313 | |
| 314 | /* Ok, now we are inserting <new> between <root> and <old>. <old>'s |
| 315 | * parent is already set to <new>, and the <root>'s branch is still in |
| 316 | * <side>. Update the root's leaf till we have it. Note that we can also |
| 317 | * find the side by checking the side of new->node.node_p. |
| 318 | */ |
| 319 | |
| 320 | /* We need the common higher bits between new->key and old->key. |
| 321 | * This number of bits is already in <bit>. |
Willy Tarreau | b55fcf2 | 2010-10-28 22:48:29 +0200 | [diff] [blame] | 322 | * NOTE: we can't get here whit bit < 0 since we found a dup ! |
Willy Tarreau | c218602 | 2009-10-26 19:48:54 +0100 | [diff] [blame] | 323 | */ |
| 324 | new->node.bit = bit; |
| 325 | root->b[side] = eb_dotag(&new->node.branches, EB_NODE); |
| 326 | return new; |
| 327 | } |
| 328 | |
Willy Tarreau | 20a81c2 | 2014-03-15 07:43:05 +0100 | [diff] [blame] | 329 | #endif /* _EBISTREE_H */ |