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