src/db.c

   1 #include "redis.h"
   2
   3 #include <signal.h>
   4
   5 /*-----------------------------------------------------------------------------
   6  * C-level DB API
   7  *----------------------------------------------------------------------------*/
   8
   9 robj *lookupKey(redisDb *db, robj *key) {
  10     dictEntry *de = dictFind(db->dict,key->ptr);
  11     if (de) {
  12         robj *val = dictGetEntryVal(de);
  13
  14         if (server.vm_enabled) {
  15             if (val->storage == REDIS_VM_MEMORY ||
  16                 val->storage == REDIS_VM_SWAPPING)
  17             {
  18                 /* If we were swapping the object out, cancel the operation */
  19                 if (val->storage == REDIS_VM_SWAPPING)
  20                     vmCancelThreadedIOJob(val);
  21                 /* Update the access time for the aging algorithm. */
  22                 val->lru = server.lruclock;
  23             } else {
  24                 int notify = (val->storage == REDIS_VM_LOADING);
  25
  26                 /* Our value was swapped on disk. Bring it at home. */
  27                 redisAssert(val->type == REDIS_VMPOINTER);
  28                 val = vmLoadObject(val);
  29                 dictGetEntryVal(de) = val;
  30
  31                 /* Clients blocked by the VM subsystem may be waiting for
  32                  * this key... */
  33                 if (notify) handleClientsBlockedOnSwappedKey(db,key);
  34             }
  35         }
  36         return val;
  37     } else {
  38         return NULL;
  39     }
  40 }
  41
  42 robj *lookupKeyRead(redisDb *db, robj *key) {
  43     expireIfNeeded(db,key);
  44     return lookupKey(db,key);
  45 }
  46
  47 robj *lookupKeyWrite(redisDb *db, robj *key) {
  48     expireIfNeeded(db,key);
  49     return lookupKey(db,key);
  50 }
  51
  52 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply) {
  53     robj *o = lookupKeyRead(c->db, key);
  54     if (!o) addReply(c,reply);
  55     return o;
  56 }
  57
  58 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply) {
  59     robj *o = lookupKeyWrite(c->db, key);
  60     if (!o) addReply(c,reply);
  61     return o;
  62 }
  63
  64 /* Add the key to the DB. If the key already exists REDIS_ERR is returned,
  65  * otherwise REDIS_OK is returned, and the caller should increment the
  66  * refcount of 'val'. */
  67 int dbAdd(redisDb *db, robj *key, robj *val) {
  68     /* Perform a lookup before adding the key, as we need to copy the
  69      * key value. */
  70     if (dictFind(db->dict, key->ptr) != NULL) {
  71         return REDIS_ERR;
  72     } else {
  73         sds copy = sdsdup(key->ptr);
  74         dictAdd(db->dict, copy, val);
  75         return REDIS_OK;
  76     }
  77 }
  78
  79 /* If the key does not exist, this is just like dbAdd(). Otherwise
  80  * the value associated to the key is replaced with the new one.
  81  *
  82  * On update (key already existed) 0 is returned. Otherwise 1. */
  83 int dbReplace(redisDb *db, robj *key, robj *val) {
  84     if (dictFind(db->dict,key->ptr) == NULL) {
  85         sds copy = sdsdup(key->ptr);
  86         dictAdd(db->dict, copy, val);
  87         return 1;
  88     } else {
  89         dictReplace(db->dict, key->ptr, val);
  90         return 0;
  91     }
  92 }
  93
  94 int dbExists(redisDb *db, robj *key) {
  95     return dictFind(db->dict,key->ptr) != NULL;
  96 }
  97
  98 /* Return a random key, in form of a Redis object.
  99  * If there are no keys, NULL is returned.
 100  *
 101  * The function makes sure to return keys not already expired. */
 102 robj *dbRandomKey(redisDb *db) {
 103     struct dictEntry *de;
 104
 105     while(1) {
 106         sds key;
 107         robj *keyobj;
 108
 109         de = dictGetRandomKey(db->dict);
 110         if (de == NULL) return NULL;
 111
 112         key = dictGetEntryKey(de);
 113         keyobj = createStringObject(key,sdslen(key));
 114         if (dictFind(db->expires,key)) {
 115             if (expireIfNeeded(db,keyobj)) {
 116                 decrRefCount(keyobj);
 117                 continue; /* search for another key. This expired. */
 118             }
 119         }
 120         return keyobj;
 121     }
 122 }
 123
 124 /* Delete a key, value, and associated expiration entry if any, from the DB */
 125 int dbDelete(redisDb *db, robj *key) {
 126     /* Deleting an entry from the expires dict will not free the sds of
 127      * the key, because it is shared with the main dictionary. */
 128     if (dictSize(db->expires) > 0) dictDelete(db->expires,key->ptr);
 129     return dictDelete(db->dict,key->ptr) == DICT_OK;
 130 }
 131
 132 /* Empty the whole database */
 133 long long emptyDb() {
 134     int j;
 135     long long removed = 0;
 136
 137     for (j = 0; j < server.dbnum; j++) {
 138         removed += dictSize(server.db[j].dict);
 139         dictEmpty(server.db[j].dict);
 140         dictEmpty(server.db[j].expires);
 141     }
 142     return removed;
 143 }
 144
 145 int selectDb(redisClient *c, int id) {
 146     if (id < 0 || id >= server.dbnum)
 147         return REDIS_ERR;
 148     c->db = &server.db[id];
 149     return REDIS_OK;
 150 }
 151
 152 /*-----------------------------------------------------------------------------
 153  * Type agnostic commands operating on the key space
 154  *----------------------------------------------------------------------------*/
 155
 156 void flushdbCommand(redisClient *c) {
 157     server.dirty += dictSize(c->db->dict);
 158     touchWatchedKeysOnFlush(c->db->id);
 159     dictEmpty(c->db->dict);
 160     dictEmpty(c->db->expires);
 161     addReply(c,shared.ok);
 162 }
 163
 164 void flushallCommand(redisClient *c) {
 165     touchWatchedKeysOnFlush(-1);
 166     server.dirty += emptyDb();
 167     addReply(c,shared.ok);
 168     if (server.bgsavechildpid != -1) {
 169         kill(server.bgsavechildpid,SIGKILL);
 170         rdbRemoveTempFile(server.bgsavechildpid);
 171     }
 172     rdbSave(server.dbfilename);
 173     server.dirty++;
 174 }
 175
 176 void delCommand(redisClient *c) {
 177     int deleted = 0, j;
 178
 179     for (j = 1; j < c->argc; j++) {
 180         if (dbDelete(c->db,c->argv[j])) {
 181             touchWatchedKey(c->db,c->argv[j]);
 182             server.dirty++;
 183             deleted++;
 184         }
 185     }
 186     addReplyLongLong(c,deleted);
 187 }
 188
 189 void existsCommand(redisClient *c) {
 190     expireIfNeeded(c->db,c->argv[1]);
 191     if (dbExists(c->db,c->argv[1])) {
 192         addReply(c, shared.cone);
 193     } else {
 194         addReply(c, shared.czero);
 195     }
 196 }
 197
 198 void selectCommand(redisClient *c) {
 199     int id = atoi(c->argv[1]->ptr);
 200
 201     if (selectDb(c,id) == REDIS_ERR) {
 202         addReplySds(c,sdsnew("-ERR invalid DB index\r\n"));
 203     } else {
 204         addReply(c,shared.ok);
 205     }
 206 }
 207
 208 void randomkeyCommand(redisClient *c) {
 209     robj *key;
 210
 211     if ((key = dbRandomKey(c->db)) == NULL) {
 212         addReply(c,shared.nullbulk);
 213         return;
 214     }
 215
 216     addReplyBulk(c,key);
 217     decrRefCount(key);
 218 }
 219
 220 void keysCommand(redisClient *c) {
 221     dictIterator *di;
 222     dictEntry *de;
 223     sds pattern = c->argv[1]->ptr;
 224     int plen = sdslen(pattern);
 225     unsigned long numkeys = 0;
 226     robj *lenobj = createObject(REDIS_STRING,NULL);
 227
 228     di = dictGetIterator(c->db->dict);
 229     addReply(c,lenobj);
 230     decrRefCount(lenobj);
 231     while((de = dictNext(di)) != NULL) {
 232         sds key = dictGetEntryKey(de);
 233         robj *keyobj;
 234
 235         if ((pattern[0] == '*' && pattern[1] == '\0') ||
 236             stringmatchlen(pattern,plen,key,sdslen(key),0)) {
 237             keyobj = createStringObject(key,sdslen(key));
 238             if (expireIfNeeded(c->db,keyobj) == 0) {
 239                 addReplyBulk(c,keyobj);
 240                 numkeys++;
 241             }
 242             decrRefCount(keyobj);
 243         }
 244     }
 245     dictReleaseIterator(di);
 246     lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",numkeys);
 247 }
 248
 249 void dbsizeCommand(redisClient *c) {
 250     addReplySds(c,
 251         sdscatprintf(sdsempty(),":%lu\r\n",dictSize(c->db->dict)));
 252 }
 253
 254 void lastsaveCommand(redisClient *c) {
 255     addReplySds(c,
 256         sdscatprintf(sdsempty(),":%lu\r\n",server.lastsave));
 257 }
 258
 259 void typeCommand(redisClient *c) {
 260     robj *o;
 261     char *type;
 262
 263     o = lookupKeyRead(c->db,c->argv[1]);
 264     if (o == NULL) {
 265         type = "+none";
 266     } else {
 267         switch(o->type) {
 268         case REDIS_STRING: type = "+string"; break;
 269         case REDIS_LIST: type = "+list"; break;
 270         case REDIS_SET: type = "+set"; break;
 271         case REDIS_ZSET: type = "+zset"; break;
 272         case REDIS_HASH: type = "+hash"; break;
 273         default: type = "+unknown"; break;
 274         }
 275     }
 276     addReplySds(c,sdsnew(type));
 277     addReply(c,shared.crlf);
 278 }
 279
 280 void saveCommand(redisClient *c) {
 281     if (server.bgsavechildpid != -1) {
 282         addReplySds(c,sdsnew("-ERR background save in progress\r\n"));
 283         return;
 284     }
 285     if (rdbSave(server.dbfilename) == REDIS_OK) {
 286         addReply(c,shared.ok);
 287     } else {
 288         addReply(c,shared.err);
 289     }
 290 }
 291
 292 void bgsaveCommand(redisClient *c) {
 293     if (server.bgsavechildpid != -1) {
 294         addReplySds(c,sdsnew("-ERR background save already in progress\r\n"));
 295         return;
 296     }
 297     if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
 298         char *status = "+Background saving started\r\n";
 299         addReplySds(c,sdsnew(status));
 300     } else {
 301         addReply(c,shared.err);
 302     }
 303 }
 304
 305 void shutdownCommand(redisClient *c) {
 306     if (prepareForShutdown() == REDIS_OK)
 307         exit(0);
 308     addReplySds(c, sdsnew("-ERR Errors trying to SHUTDOWN. Check logs.\r\n"));
 309 }
 310
 311 void renameGenericCommand(redisClient *c, int nx) {
 312     robj *o;
 313
 314     /* To use the same key as src and dst is probably an error */
 315     if (sdscmp(c->argv[1]->ptr,c->argv[2]->ptr) == 0) {
 316         addReply(c,shared.sameobjecterr);
 317         return;
 318     }
 319
 320     if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nokeyerr)) == NULL)
 321         return;
 322
 323     incrRefCount(o);
 324     if (dbAdd(c->db,c->argv[2],o) == REDIS_ERR) {
 325         if (nx) {
 326             decrRefCount(o);
 327             addReply(c,shared.czero);
 328             return;
 329         }
 330         dbReplace(c->db,c->argv[2],o);
 331     }
 332     dbDelete(c->db,c->argv[1]);
 333     touchWatchedKey(c->db,c->argv[1]);
 334     touchWatchedKey(c->db,c->argv[2]);
 335     server.dirty++;
 336     addReply(c,nx ? shared.cone : shared.ok);
 337 }
 338
 339 void renameCommand(redisClient *c) {
 340     renameGenericCommand(c,0);
 341 }
 342
 343 void renamenxCommand(redisClient *c) {
 344     renameGenericCommand(c,1);
 345 }
 346
 347 void moveCommand(redisClient *c) {
 348     robj *o;
 349     redisDb *src, *dst;
 350     int srcid;
 351
 352     /* Obtain source and target DB pointers */
 353     src = c->db;
 354     srcid = c->db->id;
 355     if (selectDb(c,atoi(c->argv[2]->ptr)) == REDIS_ERR) {
 356         addReply(c,shared.outofrangeerr);
 357         return;
 358     }
 359     dst = c->db;
 360     selectDb(c,srcid); /* Back to the source DB */
 361
 362     /* If the user is moving using as target the same
 363      * DB as the source DB it is probably an error. */
 364     if (src == dst) {
 365         addReply(c,shared.sameobjecterr);
 366         return;
 367     }
 368
 369     /* Check if the element exists and get a reference */
 370     o = lookupKeyWrite(c->db,c->argv[1]);
 371     if (!o) {
 372         addReply(c,shared.czero);
 373         return;
 374     }
 375
 376     /* Try to add the element to the target DB */
 377     if (dbAdd(dst,c->argv[1],o) == REDIS_ERR) {
 378         addReply(c,shared.czero);
 379         return;
 380     }
 381     incrRefCount(o);
 382
 383     /* OK! key moved, free the entry in the source DB */
 384     dbDelete(src,c->argv[1]);
 385     server.dirty++;
 386     addReply(c,shared.cone);
 387 }
 388
 389 /*-----------------------------------------------------------------------------
 390  * Expires API
 391  *----------------------------------------------------------------------------*/
 392
 393 int removeExpire(redisDb *db, robj *key) {
 394     /* An expire may only be removed if there is a corresponding entry in the
 395      * main dict. Otherwise, the key will never be freed. */
 396     redisAssert(dictFind(db->dict,key->ptr) != NULL);
 397     return dictDelete(db->expires,key->ptr) == DICT_OK;
 398 }
 399
 400 void setExpire(redisDb *db, robj *key, time_t when) {
 401     dictEntry *de;
 402
 403     /* Reuse the sds from the main dict in the expire dict */
 404     de = dictFind(db->dict,key->ptr);
 405     redisAssert(de != NULL);
 406     dictReplace(db->expires,dictGetEntryKey(de),(void*)when);
 407 }
 408
 409 /* Return the expire time of the specified key, or -1 if no expire
 410  * is associated with this key (i.e. the key is non volatile) */
 411 time_t getExpire(redisDb *db, robj *key) {
 412     dictEntry *de;
 413
 414     /* No expire? return ASAP */
 415     if (dictSize(db->expires) == 0 ||
 416        (de = dictFind(db->expires,key->ptr)) == NULL) return -1;
 417
 418     /* The entry was found in the expire dict, this means it should also
 419      * be present in the main dict (safety check). */
 420     redisAssert(dictFind(db->dict,key->ptr) != NULL);
 421     return (time_t) dictGetEntryVal(de);
 422 }
 423
 424 /* Propagate expires into slaves and the AOF file.
 425  * When a key expires in the master, a DEL operation for this key is sent
 426  * to all the slaves and the AOF file if enabled.
 427  *
 428  * This way the key expiry is centralized in one place, and since both
 429  * AOF and the master->slave link guarantee operation ordering, everything
 430  * will be consistent even if we allow write operations against expiring
 431  * keys. */
 432 void propagateExpire(redisDb *db, robj *key) {
 433     struct redisCommand *cmd;
 434     robj *argv[2];
 435
 436     cmd = lookupCommand("del");
 437     argv[0] = createStringObject("DEL",3);
 438     argv[1] = key;
 439     incrRefCount(key);
 440
 441     if (server.appendonly)
 442         feedAppendOnlyFile(cmd,db->id,argv,2);
 443     if (listLength(server.slaves))
 444         replicationFeedSlaves(server.slaves,db->id,argv,2);
 445
 446     decrRefCount(argv[0]);
 447     decrRefCount(argv[1]);
 448 }
 449
 450 int expireIfNeeded(redisDb *db, robj *key) {
 451     time_t when = getExpire(db,key);
 452
 453     /* If we are running in the context of a slave, return ASAP:
 454      * the slave key expiration is controlled by the master that will
 455      * send us synthesized DEL operations for expired keys.
 456      *
 457      * Still we try to return the right information to the caller,
 458      * that is, 0 if we think the key should be still valid, 1 if
 459      * we think the key is expired at this time. */
 460     if (server.masterhost != NULL) {
 461         return time(NULL) > when;
 462     }
 463
 464     if (when < 0) return 0;
 465
 466     /* Return when this key has not expired */
 467     if (time(NULL) <= when) return 0;
 468
 469     /* Delete the key */
 470     server.stat_expiredkeys++;
 471     server.dirty++;
 472     propagateExpire(db,key);
 473     return dbDelete(db,key);
 474 }
 475
 476 /*-----------------------------------------------------------------------------
 477  * Expires Commands
 478  *----------------------------------------------------------------------------*/
 479
 480 void expireGenericCommand(redisClient *c, robj *key, robj *param, long offset) {
 481     dictEntry *de;
 482     time_t seconds;
 483
 484     if (getLongFromObjectOrReply(c, param, &seconds, NULL) != REDIS_OK) return;
 485
 486     seconds -= offset;
 487
 488     de = dictFind(c->db->dict,key->ptr);
 489     if (de == NULL) {
 490         addReply(c,shared.czero);
 491         return;
 492     }
 493     if (seconds <= 0) {
 494         if (dbDelete(c->db,key)) server.dirty++;
 495         addReply(c, shared.cone);
 496         touchWatchedKey(c->db,key);
 497         return;
 498     } else {
 499         time_t when = time(NULL)+seconds;
 500         setExpire(c->db,key,when);
 501         addReply(c,shared.cone);
 502         touchWatchedKey(c->db,key);
 503         server.dirty++;
 504         return;
 505     }
 506 }
 507
 508 void expireCommand(redisClient *c) {
 509     expireGenericCommand(c,c->argv[1],c->argv[2],0);
 510 }
 511
 512 void expireatCommand(redisClient *c) {
 513     expireGenericCommand(c,c->argv[1],c->argv[2],time(NULL));
 514 }
 515
 516 void ttlCommand(redisClient *c) {
 517     time_t expire, ttl = -1;
 518
 519     expire = getExpire(c->db,c->argv[1]);
 520     if (expire != -1) {
 521         ttl = (expire-time(NULL));
 522         if (ttl < 0) ttl = -1;
 523     }
 524     addReplyLongLong(c,(long long)ttl);
 525 }
 526
 527 void persistCommand(redisClient *c) {
 528     dictEntry *de;
 529
 530     de = dictFind(c->db->dict,c->argv[1]->ptr);
 531     if (de == NULL) {
 532         addReply(c,shared.czero);
 533     } else {
 534         if (removeExpire(c->db,c->argv[1])) {
 535             addReply(c,shared.cone);
 536             server.dirty++;
 537         } else {
 538             addReply(c,shared.czero);
 539         }
 540     }
 541 }