| 1 | /*
|
|---|
| 2 | Unix SMB/CIFS implementation.
|
|---|
| 3 | Database interface wrapper around red-black trees
|
|---|
| 4 | Copyright (C) Volker Lendecke 2007, 2008
|
|---|
| 5 |
|
|---|
| 6 | This program is free software; you can redistribute it and/or modify
|
|---|
| 7 | it under the terms of the GNU General Public License as published by
|
|---|
| 8 | the Free Software Foundation; either version 3 of the License, or
|
|---|
| 9 | (at your option) any later version.
|
|---|
| 10 |
|
|---|
| 11 | This program is distributed in the hope that it will be useful,
|
|---|
| 12 | but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|---|
| 13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|---|
| 14 | GNU General Public License for more details.
|
|---|
| 15 |
|
|---|
| 16 | You should have received a copy of the GNU General Public License
|
|---|
| 17 | along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|---|
| 18 | */
|
|---|
| 19 |
|
|---|
| 20 | #include "includes.h"
|
|---|
| 21 | #include "dbwrap.h"
|
|---|
| 22 | #include "../lib/util/rbtree.h"
|
|---|
| 23 |
|
|---|
| 24 | #define DBWRAP_RBT_ALIGN(_size_) (((_size_)+15)&~15)
|
|---|
| 25 |
|
|---|
| 26 | struct db_rbt_ctx {
|
|---|
| 27 | struct rb_root tree;
|
|---|
| 28 | };
|
|---|
| 29 |
|
|---|
| 30 | struct db_rbt_rec {
|
|---|
| 31 | struct db_rbt_ctx *db_ctx;
|
|---|
| 32 | struct db_rbt_node *node;
|
|---|
| 33 | };
|
|---|
| 34 |
|
|---|
| 35 | /* The structure that ends up in the tree */
|
|---|
| 36 |
|
|---|
| 37 | struct db_rbt_node {
|
|---|
| 38 | struct rb_node rb_node;
|
|---|
| 39 | size_t keysize, valuesize;
|
|---|
| 40 |
|
|---|
| 41 | /*
|
|---|
| 42 | * key and value are appended implicitly, "data" is only here as a
|
|---|
| 43 | * target for offsetof()
|
|---|
| 44 | */
|
|---|
| 45 |
|
|---|
| 46 | char data[1];
|
|---|
| 47 | };
|
|---|
| 48 |
|
|---|
| 49 | /*
|
|---|
| 50 | * Hide the ugly pointer calculations in a function
|
|---|
| 51 | */
|
|---|
| 52 |
|
|---|
| 53 | static struct db_rbt_node *db_rbt2node(struct rb_node *node)
|
|---|
| 54 | {
|
|---|
| 55 | return (struct db_rbt_node *)
|
|---|
| 56 | ((char *)node - offsetof(struct db_rbt_node, rb_node));
|
|---|
| 57 | }
|
|---|
| 58 |
|
|---|
| 59 | /*
|
|---|
| 60 | * Compare two keys
|
|---|
| 61 | */
|
|---|
| 62 |
|
|---|
| 63 | static int db_rbt_compare(TDB_DATA a, TDB_DATA b)
|
|---|
| 64 | {
|
|---|
| 65 | int res;
|
|---|
| 66 |
|
|---|
| 67 | res = memcmp(a.dptr, b.dptr, MIN(a.dsize, b.dsize));
|
|---|
| 68 |
|
|---|
| 69 | if ((res < 0) || ((res == 0) && (a.dsize < b.dsize))) {
|
|---|
| 70 | return -1;
|
|---|
| 71 | }
|
|---|
| 72 | if ((res > 0) || ((res == 0) && (a.dsize > b.dsize))) {
|
|---|
| 73 | return 1;
|
|---|
| 74 | }
|
|---|
| 75 | return 0;
|
|---|
| 76 | }
|
|---|
| 77 |
|
|---|
| 78 | /*
|
|---|
| 79 | * dissect a db_rbt_node into its implicit key and value parts
|
|---|
| 80 | */
|
|---|
| 81 |
|
|---|
| 82 | static void db_rbt_parse_node(struct db_rbt_node *node,
|
|---|
| 83 | TDB_DATA *key, TDB_DATA *value)
|
|---|
| 84 | {
|
|---|
| 85 | key->dptr = ((uint8 *)node) + offsetof(struct db_rbt_node, data);
|
|---|
| 86 | key->dsize = node->keysize;
|
|---|
| 87 | value->dptr = key->dptr + node->keysize;
|
|---|
| 88 | value->dsize = node->valuesize;
|
|---|
| 89 | }
|
|---|
| 90 |
|
|---|
| 91 | static NTSTATUS db_rbt_store(struct db_record *rec, TDB_DATA data, int flag)
|
|---|
| 92 | {
|
|---|
| 93 | struct db_rbt_rec *rec_priv = (struct db_rbt_rec *)rec->private_data;
|
|---|
| 94 | struct db_rbt_node *node;
|
|---|
| 95 |
|
|---|
| 96 | struct rb_node ** p;
|
|---|
| 97 | struct rb_node * parent;
|
|---|
| 98 |
|
|---|
| 99 | TDB_DATA this_key, this_val;
|
|---|
| 100 |
|
|---|
| 101 | if (rec_priv->node != NULL) {
|
|---|
| 102 |
|
|---|
| 103 | /*
|
|---|
| 104 | * The record was around previously
|
|---|
| 105 | */
|
|---|
| 106 |
|
|---|
| 107 | db_rbt_parse_node(rec_priv->node, &this_key, &this_val);
|
|---|
| 108 |
|
|---|
| 109 | SMB_ASSERT(this_key.dsize == rec->key.dsize);
|
|---|
| 110 | SMB_ASSERT(memcmp(this_key.dptr, rec->key.dptr,
|
|---|
| 111 | this_key.dsize) == 0);
|
|---|
| 112 |
|
|---|
| 113 | if (this_val.dsize >= data.dsize) {
|
|---|
| 114 | /*
|
|---|
| 115 | * The new value fits into the old space
|
|---|
| 116 | */
|
|---|
| 117 | memcpy(this_val.dptr, data.dptr, data.dsize);
|
|---|
| 118 | rec_priv->node->valuesize = data.dsize;
|
|---|
| 119 | return NT_STATUS_OK;
|
|---|
| 120 | }
|
|---|
| 121 |
|
|---|
| 122 | /*
|
|---|
| 123 | * We need to delete the key from the tree and start fresh,
|
|---|
| 124 | * there's not enough space in the existing record
|
|---|
| 125 | */
|
|---|
| 126 |
|
|---|
| 127 | rb_erase(&rec_priv->node->rb_node, &rec_priv->db_ctx->tree);
|
|---|
| 128 |
|
|---|
| 129 | /*
|
|---|
| 130 | * Keep the existing node around for a while: If the record
|
|---|
| 131 | * existed before, we reference the key data in there.
|
|---|
| 132 | */
|
|---|
| 133 | }
|
|---|
| 134 |
|
|---|
| 135 | node = (struct db_rbt_node *)talloc_size(rec_priv->db_ctx,
|
|---|
| 136 | offsetof(struct db_rbt_node, data) + rec->key.dsize
|
|---|
| 137 | + data.dsize);
|
|---|
| 138 |
|
|---|
| 139 | if (node == NULL) {
|
|---|
| 140 | TALLOC_FREE(rec_priv->node);
|
|---|
| 141 | return NT_STATUS_NO_MEMORY;
|
|---|
| 142 | }
|
|---|
| 143 |
|
|---|
| 144 | ZERO_STRUCT(node->rb_node);
|
|---|
| 145 |
|
|---|
| 146 | node->keysize = rec->key.dsize;
|
|---|
| 147 | node->valuesize = data.dsize;
|
|---|
| 148 |
|
|---|
| 149 | db_rbt_parse_node(node, &this_key, &this_val);
|
|---|
| 150 |
|
|---|
| 151 | memcpy(this_key.dptr, rec->key.dptr, node->keysize);
|
|---|
| 152 | TALLOC_FREE(rec_priv->node);
|
|---|
| 153 |
|
|---|
| 154 | memcpy(this_val.dptr, data.dptr, node->valuesize);
|
|---|
| 155 |
|
|---|
| 156 | parent = NULL;
|
|---|
| 157 | p = &rec_priv->db_ctx->tree.rb_node;
|
|---|
| 158 |
|
|---|
| 159 | while (*p) {
|
|---|
| 160 | struct db_rbt_node *r;
|
|---|
| 161 | TDB_DATA search_key, search_val;
|
|---|
| 162 | int res;
|
|---|
| 163 |
|
|---|
| 164 | parent = (*p);
|
|---|
| 165 |
|
|---|
| 166 | r = db_rbt2node(*p);
|
|---|
| 167 |
|
|---|
| 168 | db_rbt_parse_node(r, &search_key, &search_val);
|
|---|
| 169 |
|
|---|
| 170 | res = db_rbt_compare(this_key, search_key);
|
|---|
| 171 |
|
|---|
| 172 | if (res == -1) {
|
|---|
| 173 | p = &(*p)->rb_left;
|
|---|
| 174 | }
|
|---|
| 175 | else if (res == 1) {
|
|---|
| 176 | p = &(*p)->rb_right;
|
|---|
| 177 | }
|
|---|
| 178 | else {
|
|---|
| 179 | smb_panic("someone messed with the tree");
|
|---|
| 180 | }
|
|---|
| 181 | }
|
|---|
| 182 |
|
|---|
| 183 | rb_link_node(&node->rb_node, parent, p);
|
|---|
| 184 | rb_insert_color(&node->rb_node, &rec_priv->db_ctx->tree);
|
|---|
| 185 |
|
|---|
| 186 | return NT_STATUS_OK;
|
|---|
| 187 | }
|
|---|
| 188 |
|
|---|
| 189 | static NTSTATUS db_rbt_delete(struct db_record *rec)
|
|---|
| 190 | {
|
|---|
| 191 | struct db_rbt_rec *rec_priv = (struct db_rbt_rec *)rec->private_data;
|
|---|
| 192 |
|
|---|
| 193 | if (rec_priv->node == NULL) {
|
|---|
| 194 | return NT_STATUS_OK;
|
|---|
| 195 | }
|
|---|
| 196 |
|
|---|
| 197 | rb_erase(&rec_priv->node->rb_node, &rec_priv->db_ctx->tree);
|
|---|
| 198 | TALLOC_FREE(rec_priv->node);
|
|---|
| 199 |
|
|---|
| 200 | return NT_STATUS_OK;
|
|---|
| 201 | }
|
|---|
| 202 |
|
|---|
| 203 | static struct db_record *db_rbt_fetch_locked(struct db_context *db_ctx,
|
|---|
| 204 | TALLOC_CTX *mem_ctx,
|
|---|
| 205 | TDB_DATA key)
|
|---|
| 206 | {
|
|---|
| 207 | struct db_rbt_ctx *ctx = talloc_get_type_abort(
|
|---|
| 208 | db_ctx->private_data, struct db_rbt_ctx);
|
|---|
| 209 |
|
|---|
| 210 | struct db_rbt_rec *rec_priv;
|
|---|
| 211 | struct db_record *result;
|
|---|
| 212 | struct rb_node *n;
|
|---|
| 213 | size_t size;
|
|---|
| 214 | bool found = false;
|
|---|
| 215 | struct db_rbt_node *r = NULL;
|
|---|
| 216 | TDB_DATA search_key = tdb_null, search_val = tdb_null;
|
|---|
| 217 |
|
|---|
| 218 | n = ctx->tree.rb_node;
|
|---|
| 219 |
|
|---|
| 220 | while (n != NULL) {
|
|---|
| 221 | int res;
|
|---|
| 222 |
|
|---|
| 223 | r = db_rbt2node(n);
|
|---|
| 224 |
|
|---|
| 225 | db_rbt_parse_node(r, &search_key, &search_val);
|
|---|
| 226 |
|
|---|
| 227 | res = db_rbt_compare(key, search_key);
|
|---|
| 228 |
|
|---|
| 229 | if (res == -1) {
|
|---|
| 230 | n = n->rb_left;
|
|---|
| 231 | }
|
|---|
| 232 | else if (res == 1) {
|
|---|
| 233 | n = n->rb_right;
|
|---|
| 234 | }
|
|---|
| 235 | else {
|
|---|
| 236 | found = true;
|
|---|
| 237 | break;
|
|---|
| 238 | }
|
|---|
| 239 | }
|
|---|
| 240 |
|
|---|
| 241 | /*
|
|---|
| 242 | * In this low-level routine, play tricks to reduce the number of
|
|---|
| 243 | * tallocs to one. Not recommened for general use, but here it pays
|
|---|
| 244 | * off.
|
|---|
| 245 | */
|
|---|
| 246 |
|
|---|
| 247 | size = DBWRAP_RBT_ALIGN(sizeof(struct db_record))
|
|---|
| 248 | + sizeof(struct db_rbt_rec);
|
|---|
| 249 |
|
|---|
| 250 | if (!found) {
|
|---|
| 251 | /*
|
|---|
| 252 | * We need to keep the key around for later store
|
|---|
| 253 | */
|
|---|
| 254 | size += key.dsize;
|
|---|
| 255 | }
|
|---|
| 256 |
|
|---|
| 257 | result = (struct db_record *)talloc_size(mem_ctx, size);
|
|---|
| 258 | if (result == NULL) {
|
|---|
| 259 | return NULL;
|
|---|
| 260 | }
|
|---|
| 261 |
|
|---|
| 262 | rec_priv = (struct db_rbt_rec *)
|
|---|
| 263 | ((char *)result + DBWRAP_RBT_ALIGN(sizeof(struct db_record)));
|
|---|
| 264 | rec_priv->db_ctx = ctx;
|
|---|
| 265 |
|
|---|
| 266 | result->store = db_rbt_store;
|
|---|
| 267 | result->delete_rec = db_rbt_delete;
|
|---|
| 268 | result->private_data = rec_priv;
|
|---|
| 269 |
|
|---|
| 270 | if (found) {
|
|---|
| 271 | rec_priv->node = r;
|
|---|
| 272 | result->key = search_key;
|
|---|
| 273 | result->value = search_val;
|
|---|
| 274 | }
|
|---|
| 275 | else {
|
|---|
| 276 | rec_priv->node = NULL;
|
|---|
| 277 | result->key.dptr = (uint8 *)
|
|---|
| 278 | ((char *)rec_priv + sizeof(*rec_priv));
|
|---|
| 279 | result->key.dsize = key.dsize;
|
|---|
| 280 | memcpy(result->key.dptr, key.dptr, key.dsize);
|
|---|
| 281 |
|
|---|
| 282 | result->value = tdb_null;
|
|---|
| 283 | }
|
|---|
| 284 |
|
|---|
| 285 | return result;
|
|---|
| 286 | }
|
|---|
| 287 |
|
|---|
| 288 | static int db_rbt_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
|
|---|
| 289 | TDB_DATA key, TDB_DATA *data)
|
|---|
| 290 | {
|
|---|
| 291 | struct db_rbt_ctx *ctx = talloc_get_type_abort(
|
|---|
| 292 | db->private_data, struct db_rbt_ctx);
|
|---|
| 293 |
|
|---|
| 294 | struct rb_node *n;
|
|---|
| 295 | bool found = false;
|
|---|
| 296 | struct db_rbt_node *r = NULL;
|
|---|
| 297 | TDB_DATA search_key, search_val;
|
|---|
| 298 | uint8_t *result;
|
|---|
| 299 |
|
|---|
| 300 | n = ctx->tree.rb_node;
|
|---|
| 301 |
|
|---|
| 302 | while (n != NULL) {
|
|---|
| 303 | int res;
|
|---|
| 304 |
|
|---|
| 305 | r = db_rbt2node(n);
|
|---|
| 306 |
|
|---|
| 307 | db_rbt_parse_node(r, &search_key, &search_val);
|
|---|
| 308 |
|
|---|
| 309 | res = db_rbt_compare(key, search_key);
|
|---|
| 310 |
|
|---|
| 311 | if (res == -1) {
|
|---|
| 312 | n = n->rb_left;
|
|---|
| 313 | }
|
|---|
| 314 | else if (res == 1) {
|
|---|
| 315 | n = n->rb_right;
|
|---|
| 316 | }
|
|---|
| 317 | else {
|
|---|
| 318 | found = true;
|
|---|
| 319 | break;
|
|---|
| 320 | }
|
|---|
| 321 | }
|
|---|
| 322 |
|
|---|
| 323 | if (!found) {
|
|---|
| 324 | *data = tdb_null;
|
|---|
| 325 | return 0;
|
|---|
| 326 | }
|
|---|
| 327 |
|
|---|
| 328 | result = (uint8 *)talloc_memdup(mem_ctx, search_val.dptr,
|
|---|
| 329 | search_val.dsize);
|
|---|
| 330 | if (result == NULL) {
|
|---|
| 331 | return -1;
|
|---|
| 332 | }
|
|---|
| 333 |
|
|---|
| 334 | data->dptr = result;
|
|---|
| 335 | data->dsize = search_val.dsize;
|
|---|
| 336 | return 0;
|
|---|
| 337 | }
|
|---|
| 338 |
|
|---|
| 339 | static int db_rbt_traverse_internal(struct rb_node *n,
|
|---|
| 340 | int (*f)(struct db_record *db,
|
|---|
| 341 | void *private_data),
|
|---|
| 342 | void *private_data)
|
|---|
| 343 | {
|
|---|
| 344 | struct db_rbt_node *r;
|
|---|
| 345 | struct db_record rec;
|
|---|
| 346 | int ret;
|
|---|
| 347 |
|
|---|
| 348 | if (n == NULL) {
|
|---|
| 349 | return 0;
|
|---|
| 350 | }
|
|---|
| 351 |
|
|---|
| 352 | ret = db_rbt_traverse_internal(n->rb_left, f, private_data);
|
|---|
| 353 | if (ret != 0) {
|
|---|
| 354 | return ret;
|
|---|
| 355 | }
|
|---|
| 356 |
|
|---|
| 357 | r = db_rbt2node(n);
|
|---|
| 358 | ZERO_STRUCT(rec);
|
|---|
| 359 | db_rbt_parse_node(r, &rec.key, &rec.value);
|
|---|
| 360 |
|
|---|
| 361 | ret = f(&rec, private_data);
|
|---|
| 362 | if (ret != 0) {
|
|---|
| 363 | return ret;
|
|---|
| 364 | }
|
|---|
| 365 |
|
|---|
| 366 | return db_rbt_traverse_internal(n->rb_right, f, private_data);
|
|---|
| 367 | }
|
|---|
| 368 |
|
|---|
| 369 | static int db_rbt_traverse(struct db_context *db,
|
|---|
| 370 | int (*f)(struct db_record *db,
|
|---|
| 371 | void *private_data),
|
|---|
| 372 | void *private_data)
|
|---|
| 373 | {
|
|---|
| 374 | struct db_rbt_ctx *ctx = talloc_get_type_abort(
|
|---|
| 375 | db->private_data, struct db_rbt_ctx);
|
|---|
| 376 |
|
|---|
| 377 | return db_rbt_traverse_internal(ctx->tree.rb_node, f, private_data);
|
|---|
| 378 | }
|
|---|
| 379 |
|
|---|
| 380 | static int db_rbt_get_seqnum(struct db_context *db)
|
|---|
| 381 | {
|
|---|
| 382 | return 0;
|
|---|
| 383 | }
|
|---|
| 384 |
|
|---|
| 385 | static int db_rbt_trans_dummy(struct db_context *db)
|
|---|
| 386 | {
|
|---|
| 387 | /*
|
|---|
| 388 | * Transactions are pretty pointless in-memory, just return success.
|
|---|
| 389 | */
|
|---|
| 390 | return 0;
|
|---|
| 391 | }
|
|---|
| 392 |
|
|---|
| 393 | struct db_context *db_open_rbt(TALLOC_CTX *mem_ctx)
|
|---|
| 394 | {
|
|---|
| 395 | struct db_context *result;
|
|---|
| 396 |
|
|---|
| 397 | result = talloc(mem_ctx, struct db_context);
|
|---|
| 398 |
|
|---|
| 399 | if (result == NULL) {
|
|---|
| 400 | return NULL;
|
|---|
| 401 | }
|
|---|
| 402 |
|
|---|
| 403 | result->private_data = TALLOC_ZERO_P(result, struct db_rbt_ctx);
|
|---|
| 404 |
|
|---|
| 405 | if (result->private_data == NULL) {
|
|---|
| 406 | TALLOC_FREE(result);
|
|---|
| 407 | return NULL;
|
|---|
| 408 | }
|
|---|
| 409 |
|
|---|
| 410 | result->fetch_locked = db_rbt_fetch_locked;
|
|---|
| 411 | result->fetch = db_rbt_fetch;
|
|---|
| 412 | result->traverse = db_rbt_traverse;
|
|---|
| 413 | result->traverse_read = db_rbt_traverse;
|
|---|
| 414 | result->get_seqnum = db_rbt_get_seqnum;
|
|---|
| 415 | result->transaction_start = db_rbt_trans_dummy;
|
|---|
| 416 | result->transaction_commit = db_rbt_trans_dummy;
|
|---|
| 417 | result->transaction_cancel = db_rbt_trans_dummy;
|
|---|
| 418 |
|
|---|
| 419 | return result;
|
|---|
| 420 | }
|
|---|