lib/dbwrap/dbwrap_rbt.c

   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/dbwrap.h"
  22 #include "dbwrap/dbwrap_private.h"
  23 #include "dbwrap/dbwrap_rbt.h"
  24 #include "../lib/util/rbtree.h"
  25 #include "../lib/util/dlinklist.h"
  26
  27 #define DBWRAP_RBT_ALIGN(_size_) (((_size_)+15)&~15)
  28
  29 struct db_rbt_ctx {
  30         struct rb_root tree;
  31         struct db_rbt_node *nodes;
  32         size_t traverse_read;
  33         struct db_rbt_node **traverse_nextp;
  34 };
  35
  36 struct db_rbt_rec {
  37         struct db_rbt_node *node;
  38 };
  39
  40 /* The structure that ends up in the tree */
  41
  42 struct db_rbt_node {
  43         struct rb_node rb_node;
  44         size_t keysize, valuesize;
  45         struct db_rbt_node *prev, *next;
  46 };
  47
  48 /*
  49  * Hide the ugly pointer calculations in a function
  50  */
  51
  52 static struct db_rbt_node *db_rbt2node(struct rb_node *node)
  53 {
  54         return (struct db_rbt_node *)
  55                 ((char *)node - offsetof(struct db_rbt_node, rb_node));
  56 }
  57
  58 /*
  59  * Compare two keys
  60  */
  61
  62 static int db_rbt_compare(TDB_DATA a, TDB_DATA b)
  63 {
  64         int res;
  65
  66         res = memcmp(a.dptr, b.dptr, MIN(a.dsize, b.dsize));
  67
  68         if ((res < 0) || ((res == 0) && (a.dsize < b.dsize))) {
  69                 return -1;
  70         }
  71         if ((res > 0) || ((res == 0) && (a.dsize > b.dsize))) {
  72                 return 1;
  73         }
  74         return 0;
  75 }
  76
  77 /*
  78  * dissect a db_rbt_node into its implicit key and value parts
  79  */
  80
  81 static void db_rbt_parse_node(struct db_rbt_node *node,
  82                               TDB_DATA *key, TDB_DATA *value)
  83 {
  84         size_t key_offset, value_offset;
  85
  86         key_offset = DBWRAP_RBT_ALIGN(sizeof(struct db_rbt_node));
  87         key->dptr = ((uint8_t *)node) + key_offset;
  88         key->dsize = node->keysize;
  89
  90         value_offset = DBWRAP_RBT_ALIGN(node->keysize);
  91         value->dptr = key->dptr + value_offset;
  92         value->dsize = node->valuesize;
  93 }
  94
  95 static ssize_t db_rbt_reclen(size_t keylen, size_t valuelen)
  96 {
  97         size_t len, tmp;
  98
  99         len = DBWRAP_RBT_ALIGN(sizeof(struct db_rbt_node));
 100
 101         tmp = DBWRAP_RBT_ALIGN(keylen);
 102         if (tmp < keylen) {
 103                 goto overflow;
 104         }
 105
 106         len += tmp;
 107         if (len < tmp) {
 108                 goto overflow;
 109         }
 110
 111         len += valuelen;
 112         if (len < valuelen) {
 113                 goto overflow;
 114         }
 115
 116         return len;
 117 overflow:
 118         return -1;
 119 }
 120
 121 static NTSTATUS db_rbt_storev(struct db_record *rec,
 122                               const TDB_DATA *dbufs, int num_dbufs, int flag)
 123 {
 124         struct db_rbt_ctx *db_ctx = talloc_get_type_abort(
 125                 rec->db->private_data, struct db_rbt_ctx);
 126         struct db_rbt_rec *rec_priv = (struct db_rbt_rec *)rec->private_data;
 127         struct db_rbt_node *node;
 128
 129         struct rb_node ** p;
 130         struct rb_node *parent = NULL;
 131         struct db_rbt_node *parent_node = NULL;
 132
 133         ssize_t reclen;
 134         TDB_DATA data, this_key, this_val;
 135         void *to_free = NULL;
 136
 137         if (db_ctx->traverse_read > 0) {
 138                 return NT_STATUS_MEDIA_WRITE_PROTECTED;
 139         }
 140
 141         if (num_dbufs == 1) {
 142                 data = dbufs[0];
 143         } else {
 144                 data = dbwrap_merge_dbufs(rec, dbufs, num_dbufs);
 145                 if (data.dptr == NULL) {
 146                         return NT_STATUS_NO_MEMORY;
 147                 }
 148                 to_free = data.dptr;
 149         }
 150
 151         if (rec_priv->node != NULL) {
 152
 153                 /*
 154                  * The record was around previously
 155                  */
 156
 157                 db_rbt_parse_node(rec_priv->node, &this_key, &this_val);
 158
 159                 SMB_ASSERT(this_key.dsize == rec->key.dsize);
 160                 SMB_ASSERT(memcmp(this_key.dptr, rec->key.dptr,
 161                                   this_key.dsize) == 0);
 162
 163                 if (this_val.dsize >= data.dsize) {
 164                         /*
 165                          * The new value fits into the old space
 166                          */
 167                         memcpy(this_val.dptr, data.dptr, data.dsize);
 168                         rec_priv->node->valuesize = data.dsize;
 169                         TALLOC_FREE(to_free);
 170                         return NT_STATUS_OK;
 171                 }
 172         }
 173
 174         reclen = db_rbt_reclen(rec->key.dsize, data.dsize);
 175         if (reclen == -1) {
 176                 TALLOC_FREE(to_free);
 177                 return NT_STATUS_INSUFFICIENT_RESOURCES;
 178         }
 179
 180         node = talloc_zero_size(db_ctx, reclen);
 181         if (node == NULL) {
 182                 TALLOC_FREE(to_free);
 183                 return NT_STATUS_NO_MEMORY;
 184         }
 185
 186         if (rec_priv->node != NULL) {
 187                 if (db_ctx->traverse_nextp != NULL) {
 188                         if (*db_ctx->traverse_nextp == rec_priv->node) {
 189                                 *db_ctx->traverse_nextp = node;
 190                         }
 191                 }
 192
 193                 /*
 194                  * We need to delete the key from the tree and start fresh,
 195                  * there's not enough space in the existing record
 196                  */
 197
 198                 rb_erase(&rec_priv->node->rb_node, &db_ctx->tree);
 199                 DLIST_REMOVE(db_ctx->nodes, rec_priv->node);
 200
 201                 /*
 202                  * Keep the existing node around for a while: If the record
 203                  * existed before, we reference the key data in there.
 204                  */
 205         }
 206
 207         node->keysize = rec->key.dsize;
 208         node->valuesize = data.dsize;
 209
 210         db_rbt_parse_node(node, &this_key, &this_val);
 211
 212         memcpy(this_key.dptr, rec->key.dptr, node->keysize);
 213         TALLOC_FREE(rec_priv->node);
 214         rec_priv->node = node;
 215
 216         memcpy(this_val.dptr, data.dptr, node->valuesize);
 217
 218         parent = NULL;
 219         p = &db_ctx->tree.rb_node;
 220
 221         while (*p) {
 222                 struct db_rbt_node *r;
 223                 TDB_DATA search_key, search_val;
 224                 int res;
 225
 226                 r = db_rbt2node(*p);
 227
 228                 parent = (*p);
 229                 parent_node = r;
 230
 231                 db_rbt_parse_node(r, &search_key, &search_val);
 232
 233                 res = db_rbt_compare(this_key, search_key);
 234
 235                 if (res == -1) {
 236                         p = &(*p)->rb_left;
 237                 }
 238                 else if (res == 1) {
 239                         p = &(*p)->rb_right;
 240                 }
 241                 else {
 242                         smb_panic("someone messed with the tree");
 243                 }
 244         }
 245
 246         rb_link_node(&node->rb_node, parent, p);
 247         DLIST_ADD_AFTER(db_ctx->nodes, node, parent_node);
 248         rb_insert_color(&node->rb_node, &db_ctx->tree);
 249
 250         TALLOC_FREE(to_free);
 251
 252         return NT_STATUS_OK;
 253 }
 254
 255 static NTSTATUS db_rbt_delete(struct db_record *rec)
 256 {
 257         struct db_rbt_ctx *db_ctx = talloc_get_type_abort(
 258                 rec->db->private_data, struct db_rbt_ctx);
 259         struct db_rbt_rec *rec_priv = (struct db_rbt_rec *)rec->private_data;
 260
 261         if (db_ctx->traverse_read > 0) {
 262                 return NT_STATUS_MEDIA_WRITE_PROTECTED;
 263         }
 264
 265         if (rec_priv->node == NULL) {
 266                 return NT_STATUS_OK;
 267         }
 268
 269         if (db_ctx->traverse_nextp != NULL) {
 270                 if (*db_ctx->traverse_nextp == rec_priv->node) {
 271                         *db_ctx->traverse_nextp = rec_priv->node->next;
 272                 }
 273         }
 274
 275         rb_erase(&rec_priv->node->rb_node, &db_ctx->tree);
 276         DLIST_REMOVE(db_ctx->nodes, rec_priv->node);
 277         TALLOC_FREE(rec_priv->node);
 278
 279         return NT_STATUS_OK;
 280 }
 281
 282 struct db_rbt_search_result {
 283         TDB_DATA key;
 284         TDB_DATA val;
 285         struct db_rbt_node* node;
 286 };
 287
 288 static bool db_rbt_search_internal(struct db_context *db, TDB_DATA key,
 289                                    struct db_rbt_search_result *result)
 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 = { 0 };
 298         TDB_DATA search_val = { 0 };
 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         if (result != NULL) {
 323                 if (found) {
 324                         result->key = search_key;
 325                         result->val = search_val;
 326                         result->node = r;
 327                 } else {
 328                         ZERO_STRUCT(*result);
 329                 }
 330         }
 331         return found;
 332 }
 333
 334 static struct db_record *db_rbt_fetch_locked(struct db_context *db_ctx,
 335                                              TALLOC_CTX *mem_ctx,
 336                                              TDB_DATA key)
 337 {
 338         struct db_rbt_rec *rec_priv;
 339         struct db_record *result;
 340         size_t size;
 341         bool found;
 342         struct db_rbt_search_result res;
 343
 344         found = db_rbt_search_internal(db_ctx, key, &res);
 345
 346         /*
 347          * In this low-level routine, play tricks to reduce the number of
 348          * tallocs to one. Not recommened for general use, but here it pays
 349          * off.
 350          */
 351
 352         size = DBWRAP_RBT_ALIGN(sizeof(struct db_record))
 353                 + sizeof(struct db_rbt_rec);
 354
 355         if (!found) {
 356                 /*
 357                  * We need to keep the key around for later store
 358                  */
 359                 size += key.dsize;
 360         }
 361
 362         result = (struct db_record *)talloc_size(mem_ctx, size);
 363         if (result == NULL) {
 364                 return NULL;
 365         }
 366
 367         rec_priv = (struct db_rbt_rec *)
 368                 ((char *)result + DBWRAP_RBT_ALIGN(sizeof(struct db_record)));
 369
 370         result->storev = db_rbt_storev;
 371         result->delete_rec = db_rbt_delete;
 372         result->private_data = rec_priv;
 373
 374         rec_priv->node = res.node;
 375         result->value  = res.val;
 376
 377         if (found) {
 378                 result->key = res.key;
 379         }
 380         else {
 381                 result->key.dptr = (uint8_t *)
 382                         ((char *)rec_priv + sizeof(*rec_priv));
 383                 result->key.dsize = key.dsize;
 384                 memcpy(result->key.dptr, key.dptr, key.dsize);
 385         }
 386
 387         return result;
 388 }
 389
 390 static int db_rbt_exists(struct db_context *db, TDB_DATA key)
 391 {
 392         return db_rbt_search_internal(db, key, NULL);
 393 }
 394
 395 static int db_rbt_wipe(struct db_context *db)
 396 {
 397         struct db_rbt_ctx *old_ctx = talloc_get_type_abort(
 398                 db->private_data, struct db_rbt_ctx);
 399         struct db_rbt_ctx *new_ctx = talloc_zero(db, struct db_rbt_ctx);
 400         if (new_ctx == NULL) {
 401                 return -1;
 402         }
 403         db->private_data = new_ctx;
 404         talloc_free(old_ctx);
 405         return 0;
 406 }
 407
 408 static NTSTATUS db_rbt_parse_record(struct db_context *db, TDB_DATA key,
 409                                     void (*parser)(TDB_DATA key, TDB_DATA data,
 410                                                    void *private_data),
 411                                     void *private_data)
 412 {
 413         struct db_rbt_search_result res;
 414         bool found = db_rbt_search_internal(db, key, &res);
 415
 416         if (!found) {
 417                 return NT_STATUS_NOT_FOUND;
 418         }
 419         parser(res.key, res.val, private_data);
 420         return NT_STATUS_OK;
 421 }
 422
 423 static int db_rbt_traverse_internal(struct db_context *db,
 424                                     int (*f)(struct db_record *db,
 425                                              void *private_data),
 426                                     void *private_data, uint32_t* count,
 427                                     bool rw)
 428 {
 429         struct db_rbt_ctx *ctx = talloc_get_type_abort(
 430                 db->private_data, struct db_rbt_ctx);
 431         struct db_rbt_node *cur = NULL;
 432         struct db_rbt_node *next = NULL;
 433         int ret;
 434
 435         for (cur = ctx->nodes; cur != NULL; cur = next) {
 436                 struct db_record rec;
 437                 struct db_rbt_rec rec_priv;
 438
 439                 rec_priv.node = cur;
 440                 next = rec_priv.node->next;
 441
 442                 ZERO_STRUCT(rec);
 443                 rec.db = db;
 444                 rec.private_data = &rec_priv;
 445                 rec.storev = db_rbt_storev;
 446                 rec.delete_rec = db_rbt_delete;
 447                 db_rbt_parse_node(rec_priv.node, &rec.key, &rec.value);
 448
 449                 if (rw) {
 450                         ctx->traverse_nextp = &next;
 451                 }
 452                 ret = f(&rec, private_data);
 453                 (*count) ++;
 454                 if (rw) {
 455                         ctx->traverse_nextp = NULL;
 456                 }
 457                 if (ret != 0) {
 458                         return ret;
 459                 }
 460                 if (rec_priv.node != NULL) {
 461                         next = rec_priv.node->next;
 462                 }
 463         }
 464
 465         return 0;
 466 }
 467
 468 static int db_rbt_traverse_read(struct db_context *db,
 469                                 int (*f)(struct db_record *db,
 470                                          void *private_data),
 471                                 void *private_data)
 472 {
 473         struct db_rbt_ctx *ctx = talloc_get_type_abort(
 474                 db->private_data, struct db_rbt_ctx);
 475         uint32_t count = 0;
 476         int ret;
 477
 478         ctx->traverse_read++;
 479         ret = db_rbt_traverse_internal(db,
 480                                        f, private_data, &count,
 481                                        false /* rw */);
 482         ctx->traverse_read--;
 483         if (ret != 0) {
 484                 return -1;
 485         }
 486         if (count > INT_MAX) {
 487                 return -1;
 488         }
 489         return count;
 490 }
 491
 492 static int db_rbt_traverse(struct db_context *db,
 493                            int (*f)(struct db_record *db,
 494                                     void *private_data),
 495                            void *private_data)
 496 {
 497         struct db_rbt_ctx *ctx = talloc_get_type_abort(
 498                 db->private_data, struct db_rbt_ctx);
 499         uint32_t count = 0;
 500         int ret;
 501
 502         if (ctx->traverse_nextp != NULL) {
 503                 return -1;
 504         };
 505
 506         if (ctx->traverse_read > 0) {
 507                 return db_rbt_traverse_read(db, f, private_data);
 508         }
 509
 510         ret = db_rbt_traverse_internal(db,
 511                                        f, private_data, &count,
 512                                        true /* rw */);
 513         if (ret != 0) {
 514                 return -1;
 515         }
 516         if (count > INT_MAX) {
 517                 return -1;
 518         }
 519         return count;
 520 }
 521
 522 static int db_rbt_get_seqnum(struct db_context *db)
 523 {
 524         return 0;
 525 }
 526
 527 static int db_rbt_trans_dummy(struct db_context *db)
 528 {
 529         /*
 530          * Transactions are pretty pointless in-memory, just return success.
 531          */
 532         return 0;
 533 }
 534
 535 static size_t db_rbt_id(struct db_context *db, uint8_t *id, size_t idlen)
 536 {
 537         if (idlen >= sizeof(struct db_context *)) {
 538                 memcpy(id, &db, sizeof(struct db_context *));
 539         }
 540         return sizeof(struct db_context *);
 541 }
 542
 543 struct db_context *db_open_rbt(TALLOC_CTX *mem_ctx)
 544 {
 545         struct db_context *result;
 546
 547         result = talloc_zero(mem_ctx, struct db_context);
 548
 549         if (result == NULL) {
 550                 return NULL;
 551         }
 552
 553         result->private_data = talloc_zero(result, struct db_rbt_ctx);
 554
 555         if (result->private_data == NULL) {
 556                 TALLOC_FREE(result);
 557                 return NULL;
 558         }
 559
 560         result->fetch_locked = db_rbt_fetch_locked;
 561         result->traverse = db_rbt_traverse;
 562         result->traverse_read = db_rbt_traverse_read;
 563         result->get_seqnum = db_rbt_get_seqnum;
 564         result->transaction_start = db_rbt_trans_dummy;
 565         result->transaction_commit = db_rbt_trans_dummy;
 566         result->transaction_cancel = db_rbt_trans_dummy;
 567         result->exists = db_rbt_exists;
 568         result->wipe = db_rbt_wipe;
 569         result->parse_record = db_rbt_parse_record;
 570         result->id = db_rbt_id;
 571         result->name = "dbwrap rbt";
 572
 573         return result;
 574 }