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         /* Update the access time for the aging algorithm. */
  15         val->lru = server.lruclock;
  16
  17         if (server.vm_enabled) {
  18             if (val->storage == REDIS_VM_MEMORY ||
  19                 val->storage == REDIS_VM_SWAPPING)
  20             {
  21                 /* If we were swapping the object out, cancel the operation */
  22                 if (val->storage == REDIS_VM_SWAPPING)
  23                     vmCancelThreadedIOJob(val);
  24             } else {
  25                 int notify = (val->storage == REDIS_VM_LOADING);
  26
  27                 /* Our value was swapped on disk. Bring it at home. */
  28                 redisAssert(val->type == REDIS_VMPOINTER);
  29                 val = vmLoadObject(val);
  30                 dictGetEntryVal(de) = val;
  31
  32                 /* Clients blocked by the VM subsystem may be waiting for
  33                  * this key... */
  34                 if (notify) handleClientsBlockedOnSwappedKey(db,key);
  35             }
  36         }
  37         server.stat_keyspace_hits++;
  38         return val;
  39     } else {
  40         server.stat_keyspace_misses++;
  41         return NULL;
  42     }
  43 }
  44
  45 robj *lookupKeyRead(redisDb *db, robj *key) {
  46     expireIfNeeded(db,key);
  47     return lookupKey(db,key);
  48 }
  49
  50 robj *lookupKeyWrite(redisDb *db, robj *key) {
  51     expireIfNeeded(db,key);
  52     return lookupKey(db,key);
  53 }
  54
  55 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply) {
  56     robj *o = lookupKeyRead(c->db, key);
  57     if (!o) addReply(c,reply);
  58     return o;
  59 }
  60
  61 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply) {
  62     robj *o = lookupKeyWrite(c->db, key);
  63     if (!o) addReply(c,reply);
  64     return o;
  65 }
  66
  67 /* Add the key to the DB. If the key already exists REDIS_ERR is returned,
  68  * otherwise REDIS_OK is returned, and the caller should increment the
  69  * refcount of 'val'. */
  70 int dbAdd(redisDb *db, robj *key, robj *val) {
  71     /* Perform a lookup before adding the key, as we need to copy the
  72      * key value. */
  73     if (dictFind(db->dict, key->ptr) != NULL) {
  74         return REDIS_ERR;
  75     } else {
  76         sds copy = sdsdup(key->ptr);
  77         dictAdd(db->dict, copy, val);
  78         return REDIS_OK;
  79     }
  80 }
  81
  82 /* If the key does not exist, this is just like dbAdd(). Otherwise
  83  * the value associated to the key is replaced with the new one.
  84  *
  85  * On update (key already existed) 0 is returned. Otherwise 1. */
  86 int dbReplace(redisDb *db, robj *key, robj *val) {
  87     if (dictFind(db->dict,key->ptr) == NULL) {
  88         sds copy = sdsdup(key->ptr);
  89         dictAdd(db->dict, copy, val);
  90         return 1;
  91     } else {
  92         dictReplace(db->dict, key->ptr, val);
  93         return 0;
  94     }
  95 }
  96
  97 int dbExists(redisDb *db, robj *key) {
  98     return dictFind(db->dict,key->ptr) != NULL;
  99 }
 100
 101 /* Return a random key, in form of a Redis object.
 102  * If there are no keys, NULL is returned.
 103  *
 104  * The function makes sure to return keys not already expired. */
 105 robj *dbRandomKey(redisDb *db) {
 106     struct dictEntry *de;
 107
 108     while(1) {
 109         sds key;
 110         robj *keyobj;
 111
 112         de = dictGetRandomKey(db->dict);
 113         if (de == NULL) return NULL;
 114
 115         key = dictGetEntryKey(de);
 116         keyobj = createStringObject(key,sdslen(key));
 117         if (dictFind(db->expires,key)) {
 118             if (expireIfNeeded(db,keyobj)) {
 119                 decrRefCount(keyobj);
 120                 continue; /* search for another key. This expired. */
 121             }
 122         }
 123         return keyobj;
 124     }
 125 }
 126
 127 /* Delete a key, value, and associated expiration entry if any, from the DB */
 128 int dbDelete(redisDb *db, robj *key) {
 129     /* If VM is enabled make sure to awake waiting clients for this key:
 130      * deleting the key will kill the I/O thread bringing the key from swap
 131      * to memory, so the client will never be notified and unblocked if we
 132      * don't do it now. */
 133     if (server.vm_enabled) handleClientsBlockedOnSwappedKey(db,key);
 134     /* Deleting an entry from the expires dict will not free the sds of
 135      * the key, because it is shared with the main dictionary. */
 136     if (dictSize(db->expires) > 0) dictDelete(db->expires,key->ptr);
 137     return dictDelete(db->dict,key->ptr) == DICT_OK;
 138 }
 139
 140 /* Empty the whole database */
 141 long long emptyDb() {
 142     int j;
 143     long long removed = 0;
 144
 145     for (j = 0; j < server.dbnum; j++) {
 146         removed += dictSize(server.db[j].dict);
 147         dictEmpty(server.db[j].dict);
 148         dictEmpty(server.db[j].expires);
 149     }
 150     return removed;
 151 }
 152
 153 int selectDb(redisClient *c, int id) {
 154     if (id < 0 || id >= server.dbnum)
 155         return REDIS_ERR;
 156     c->db = &server.db[id];
 157     return REDIS_OK;
 158 }
 159
 160 /*-----------------------------------------------------------------------------
 161  * Type agnostic commands operating on the key space
 162  *----------------------------------------------------------------------------*/
 163
 164 void flushdbCommand(redisClient *c) {
 165     server.dirty += dictSize(c->db->dict);
 166     touchWatchedKeysOnFlush(c->db->id);
 167     dictEmpty(c->db->dict);
 168     dictEmpty(c->db->expires);
 169     addReply(c,shared.ok);
 170 }
 171
 172 void flushallCommand(redisClient *c) {
 173     touchWatchedKeysOnFlush(-1);
 174     server.dirty += emptyDb();
 175     addReply(c,shared.ok);
 176     if (server.bgsavechildpid != -1) {
 177         kill(server.bgsavechildpid,SIGKILL);
 178         rdbRemoveTempFile(server.bgsavechildpid);
 179     }
 180     rdbSave(server.dbfilename);
 181     server.dirty++;
 182 }
 183
 184 void delCommand(redisClient *c) {
 185     int deleted = 0, j;
 186
 187     for (j = 1; j < c->argc; j++) {
 188         if (dbDelete(c->db,c->argv[j])) {
 189             touchWatchedKey(c->db,c->argv[j]);
 190             server.dirty++;
 191             deleted++;
 192         }
 193     }
 194     addReplyLongLong(c,deleted);
 195 }
 196
 197 void existsCommand(redisClient *c) {
 198     expireIfNeeded(c->db,c->argv[1]);
 199     if (dbExists(c->db,c->argv[1])) {
 200         addReply(c, shared.cone);
 201     } else {
 202         addReply(c, shared.czero);
 203     }
 204 }
 205
 206 void selectCommand(redisClient *c) {
 207     int id = atoi(c->argv[1]->ptr);
 208
 209     if (selectDb(c,id) == REDIS_ERR) {
 210         addReplyError(c,"invalid DB index");
 211     } else {
 212         addReply(c,shared.ok);
 213     }
 214 }
 215
 216 void randomkeyCommand(redisClient *c) {
 217     robj *key;
 218
 219     if ((key = dbRandomKey(c->db)) == NULL) {
 220         addReply(c,shared.nullbulk);
 221         return;
 222     }
 223
 224     addReplyBulk(c,key);
 225     decrRefCount(key);
 226 }
 227
 228 void keysCommand(redisClient *c) {
 229     dictIterator *di;
 230     dictEntry *de;
 231     sds pattern = c->argv[1]->ptr;
 232     int plen = sdslen(pattern), allkeys;
 233     unsigned long numkeys = 0;
 234     void *replylen = addDeferredMultiBulkLength(c);
 235
 236     di = dictGetIterator(c->db->dict);
 237     allkeys = (pattern[0] == '*' && pattern[1] == '\0');
 238     while((de = dictNext(di)) != NULL) {
 239         sds key = dictGetEntryKey(de);
 240         robj *keyobj;
 241
 242         if (allkeys || stringmatchlen(pattern,plen,key,sdslen(key),0)) {
 243             keyobj = createStringObject(key,sdslen(key));
 244             if (expireIfNeeded(c->db,keyobj) == 0) {
 245                 addReplyBulk(c,keyobj);
 246                 numkeys++;
 247             }
 248             decrRefCount(keyobj);
 249         }
 250     }
 251     dictReleaseIterator(di);
 252     setDeferredMultiBulkLength(c,replylen,numkeys);
 253 }
 254
 255 void dbsizeCommand(redisClient *c) {
 256     addReplyLongLong(c,dictSize(c->db->dict));
 257 }
 258
 259 void lastsaveCommand(redisClient *c) {
 260     addReplyLongLong(c,server.lastsave);
 261 }
 262
 263 void typeCommand(redisClient *c) {
 264     robj *o;
 265     char *type;
 266
 267     o = lookupKeyRead(c->db,c->argv[1]);
 268     if (o == NULL) {
 269         type = "none";
 270     } else {
 271         switch(o->type) {
 272         case REDIS_STRING: type = "string"; break;
 273         case REDIS_LIST: type = "list"; break;
 274         case REDIS_SET: type = "set"; break;
 275         case REDIS_ZSET: type = "zset"; break;
 276         case REDIS_HASH: type = "hash"; break;
 277         default: type = "unknown"; break;
 278         }
 279     }
 280     addReplyStatus(c,type);
 281 }
 282
 283 void saveCommand(redisClient *c) {
 284     if (server.bgsavechildpid != -1) {
 285         addReplyError(c,"Background save already in progress");
 286         return;
 287     }
 288     if (rdbSave(server.dbfilename) == REDIS_OK) {
 289         addReply(c,shared.ok);
 290     } else {
 291         addReply(c,shared.err);
 292     }
 293 }
 294
 295 void bgsaveCommand(redisClient *c) {
 296     if (server.bgsavechildpid != -1) {
 297         addReplyError(c,"Background save already in progress");
 298         return;
 299     }
 300     if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
 301         addReplyStatus(c,"Background saving started");
 302     } else {
 303         addReply(c,shared.err);
 304     }
 305 }
 306
 307 void shutdownCommand(redisClient *c) {
 308     if (prepareForShutdown() == REDIS_OK)
 309         exit(0);
 310     addReplyError(c,"Errors trying to SHUTDOWN. Check logs.");
 311 }
 312
 313 void renameGenericCommand(redisClient *c, int nx) {
 314     robj *o;
 315
 316     /* To use the same key as src and dst is probably an error */
 317     if (sdscmp(c->argv[1]->ptr,c->argv[2]->ptr) == 0) {
 318         addReply(c,shared.sameobjecterr);
 319         return;
 320     }
 321
 322     if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nokeyerr)) == NULL)
 323         return;
 324
 325     incrRefCount(o);
 326     if (dbAdd(c->db,c->argv[2],o) == REDIS_ERR) {
 327         if (nx) {
 328             decrRefCount(o);
 329             addReply(c,shared.czero);
 330             return;
 331         }
 332         dbReplace(c->db,c->argv[2],o);
 333     }
 334     dbDelete(c->db,c->argv[1]);
 335     touchWatchedKey(c->db,c->argv[1]);
 336     touchWatchedKey(c->db,c->argv[2]);
 337     server.dirty++;
 338     addReply(c,nx ? shared.cone : shared.ok);
 339 }
 340
 341 void renameCommand(redisClient *c) {
 342     renameGenericCommand(c,0);
 343 }
 344
 345 void renamenxCommand(redisClient *c) {
 346     renameGenericCommand(c,1);
 347 }
 348
 349 void moveCommand(redisClient *c) {
 350     robj *o;
 351     redisDb *src, *dst;
 352     int srcid;
 353
 354     /* Obtain source and target DB pointers */
 355     src = c->db;
 356     srcid = c->db->id;
 357     if (selectDb(c,atoi(c->argv[2]->ptr)) == REDIS_ERR) {
 358         addReply(c,shared.outofrangeerr);
 359         return;
 360     }
 361     dst = c->db;
 362     selectDb(c,srcid); /* Back to the source DB */
 363
 364     /* If the user is moving using as target the same
 365      * DB as the source DB it is probably an error. */
 366     if (src == dst) {
 367         addReply(c,shared.sameobjecterr);
 368         return;
 369     }
 370
 371     /* Check if the element exists and get a reference */
 372     o = lookupKeyWrite(c->db,c->argv[1]);
 373     if (!o) {
 374         addReply(c,shared.czero);
 375         return;
 376     }
 377
 378     /* Try to add the element to the target DB */
 379     if (dbAdd(dst,c->argv[1],o) == REDIS_ERR) {
 380         addReply(c,shared.czero);
 381         return;
 382     }
 383     incrRefCount(o);
 384
 385     /* OK! key moved, free the entry in the source DB */
 386     dbDelete(src,c->argv[1]);
 387     server.dirty++;
 388     addReply(c,shared.cone);
 389 }
 390
 391 /*-----------------------------------------------------------------------------
 392  * Expires API
 393  *----------------------------------------------------------------------------*/
 394
 395 int removeExpire(redisDb *db, robj *key) {
 396     /* An expire may only be removed if there is a corresponding entry in the
 397      * main dict. Otherwise, the key will never be freed. */
 398     redisAssert(dictFind(db->dict,key->ptr) != NULL);
 399     return dictDelete(db->expires,key->ptr) == DICT_OK;
 400 }
 401
 402 void setExpire(redisDb *db, robj *key, time_t when) {
 403     dictEntry *de;
 404
 405     /* Reuse the sds from the main dict in the expire dict */
 406     de = dictFind(db->dict,key->ptr);
 407     redisAssert(de != NULL);
 408     dictReplace(db->expires,dictGetEntryKey(de),(void*)when);
 409 }
 410
 411 /* Return the expire time of the specified key, or -1 if no expire
 412  * is associated with this key (i.e. the key is non volatile) */
 413 time_t getExpire(redisDb *db, robj *key) {
 414     dictEntry *de;
 415
 416     /* No expire? return ASAP */
 417     if (dictSize(db->expires) == 0 ||
 418        (de = dictFind(db->expires,key->ptr)) == NULL) return -1;
 419
 420     /* The entry was found in the expire dict, this means it should also
 421      * be present in the main dict (safety check). */
 422     redisAssert(dictFind(db->dict,key->ptr) != NULL);
 423     return (time_t) dictGetEntryVal(de);
 424 }
 425
 426 /* Propagate expires into slaves and the AOF file.
 427  * When a key expires in the master, a DEL operation for this key is sent
 428  * to all the slaves and the AOF file if enabled.
 429  *
 430  * This way the key expiry is centralized in one place, and since both
 431  * AOF and the master->slave link guarantee operation ordering, everything
 432  * will be consistent even if we allow write operations against expiring
 433  * keys. */
 434 void propagateExpire(redisDb *db, robj *key) {
 435     struct redisCommand *cmd;
 436     robj *argv[2];
 437
 438     cmd = lookupCommand("del");
 439     argv[0] = createStringObject("DEL",3);
 440     argv[1] = key;
 441     incrRefCount(key);
 442
 443     if (server.appendonly)
 444         feedAppendOnlyFile(cmd,db->id,argv,2);
 445     if (listLength(server.slaves))
 446         replicationFeedSlaves(server.slaves,db->id,argv,2);
 447
 448     decrRefCount(argv[0]);
 449     decrRefCount(argv[1]);
 450 }
 451
 452 int expireIfNeeded(redisDb *db, robj *key) {
 453     time_t when = getExpire(db,key);
 454
 455     /* If we are running in the context of a slave, return ASAP:
 456      * the slave key expiration is controlled by the master that will
 457      * send us synthesized DEL operations for expired keys.
 458      *
 459      * Still we try to return the right information to the caller,
 460      * that is, 0 if we think the key should be still valid, 1 if
 461      * we think the key is expired at this time. */
 462     if (server.masterhost != NULL) {
 463         return time(NULL) > when;
 464     }
 465
 466     if (when < 0) return 0;
 467
 468     /* Return when this key has not expired */
 469     if (time(NULL) <= when) return 0;
 470
 471     /* Delete the key */
 472     server.stat_expiredkeys++;
 473     propagateExpire(db,key);
 474     return dbDelete(db,key);
 475 }
 476
 477 /*-----------------------------------------------------------------------------
 478  * Expires Commands
 479  *----------------------------------------------------------------------------*/
 480
 481 void expireGenericCommand(redisClient *c, robj *key, robj *param, long offset) {
 482     dictEntry *de;
 483     long seconds;
 484
 485     if (getLongFromObjectOrReply(c, param, &seconds, NULL) != REDIS_OK) return;
 486
 487     seconds -= offset;
 488
 489     de = dictFind(c->db->dict,key->ptr);
 490     if (de == NULL) {
 491         addReply(c,shared.czero);
 492         return;
 493     }
 494     if (seconds <= 0) {
 495         if (dbDelete(c->db,key)) server.dirty++;
 496         addReply(c, shared.cone);
 497         touchWatchedKey(c->db,key);
 498         return;
 499     } else {
 500         time_t when = time(NULL)+seconds;
 501         setExpire(c->db,key,when);
 502         addReply(c,shared.cone);
 503         touchWatchedKey(c->db,key);
 504         server.dirty++;
 505         return;
 506     }
 507 }
 508
 509 void expireCommand(redisClient *c) {
 510     expireGenericCommand(c,c->argv[1],c->argv[2],0);
 511 }
 512
 513 void expireatCommand(redisClient *c) {
 514     expireGenericCommand(c,c->argv[1],c->argv[2],time(NULL));
 515 }
 516
 517 void ttlCommand(redisClient *c) {
 518     time_t expire, ttl = -1;
 519
 520     expire = getExpire(c->db,c->argv[1]);
 521     if (expire != -1) {
 522         ttl = (expire-time(NULL));
 523         if (ttl < 0) ttl = -1;
 524     }
 525     addReplyLongLong(c,(long long)ttl);
 526 }
 527
 528 void persistCommand(redisClient *c) {
 529     dictEntry *de;
 530
 531     de = dictFind(c->db->dict,c->argv[1]->ptr);
 532     if (de == NULL) {
 533         addReply(c,shared.czero);
 534     } else {
 535         if (removeExpire(c->db,c->argv[1])) {
 536             addReply(c,shared.cone);
 537             server.dirty++;
 538         } else {
 539             addReply(c,shared.czero);
 540         }
 541     }
 542 }