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         return val;
  38     } else {
  39         return NULL;
  40     }
  41 }
  42
  43 robj *lookupKeyRead(redisDb *db, robj *key) {
  44     expireIfNeeded(db,key);
  45     return lookupKey(db,key);
  46 }
  47
  48 robj *lookupKeyWrite(redisDb *db, robj *key) {
  49     expireIfNeeded(db,key);
  50     return lookupKey(db,key);
  51 }
  52
  53 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply) {
  54     robj *o = lookupKeyRead(c->db, key);
  55     if (!o) addReply(c,reply);
  56     return o;
  57 }
  58
  59 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply) {
  60     robj *o = lookupKeyWrite(c->db, key);
  61     if (!o) addReply(c,reply);
  62     return o;
  63 }
  64
  65 /* Add the key to the DB. If the key already exists REDIS_ERR is returned,
  66  * otherwise REDIS_OK is returned, and the caller should increment the
  67  * refcount of 'val'. */
  68 int dbAdd(redisDb *db, robj *key, robj *val) {
  69     /* Perform a lookup before adding the key, as we need to copy the
  70      * key value. */
  71     if (dictFind(db->dict, key->ptr) != NULL) {
  72         return REDIS_ERR;
  73     } else {
  74         sds copy = sdsdup(key->ptr);
  75         dictAdd(db->dict, copy, val);
  76         return REDIS_OK;
  77     }
  78 }
  79
  80 /* If the key does not exist, this is just like dbAdd(). Otherwise
  81  * the value associated to the key is replaced with the new one.
  82  *
  83  * On update (key already existed) 0 is returned. Otherwise 1. */
  84 int dbReplace(redisDb *db, robj *key, robj *val) {
  85     if (dictFind(db->dict,key->ptr) == NULL) {
  86         sds copy = sdsdup(key->ptr);
  87         dictAdd(db->dict, copy, val);
  88         return 1;
  89     } else {
  90         dictReplace(db->dict, key->ptr, val);
  91         return 0;
  92     }
  93 }
  94
  95 int dbExists(redisDb *db, robj *key) {
  96     return dictFind(db->dict,key->ptr) != NULL;
  97 }
  98
  99 /* Return a random key, in form of a Redis object.
 100  * If there are no keys, NULL is returned.
 101  *
 102  * The function makes sure to return keys not already expired. */
 103 robj *dbRandomKey(redisDb *db) {
 104     struct dictEntry *de;
 105
 106     while(1) {
 107         sds key;
 108         robj *keyobj;
 109
 110         de = dictGetRandomKey(db->dict);
 111         if (de == NULL) return NULL;
 112
 113         key = dictGetEntryKey(de);
 114         keyobj = createStringObject(key,sdslen(key));
 115         if (dictFind(db->expires,key)) {
 116             if (expireIfNeeded(db,keyobj)) {
 117                 decrRefCount(keyobj);
 118                 continue; /* search for another key. This expired. */
 119             }
 120         }
 121         return keyobj;
 122     }
 123 }
 124
 125 /* Delete a key, value, and associated expiration entry if any, from the DB */
 126 int dbDelete(redisDb *db, robj *key) {
 127     /* If VM is enabled make sure to awake waiting clients for this key:
 128      * deleting the key will kill the I/O thread bringing the key from swap
 129      * to memory, so the client will never be notified and unblocked if we
 130      * don't do it now. */
 131     if (server.vm_enabled) handleClientsBlockedOnSwappedKey(db,key);
 132     /* Deleting an entry from the expires dict will not free the sds of
 133      * the key, because it is shared with the main dictionary. */
 134     if (dictSize(db->expires) > 0) dictDelete(db->expires,key->ptr);
 135     return dictDelete(db->dict,key->ptr) == DICT_OK;
 136 }
 137
 138 /* Empty the whole database */
 139 long long emptyDb() {
 140     int j;
 141     long long removed = 0;
 142
 143     for (j = 0; j < server.dbnum; j++) {
 144         removed += dictSize(server.db[j].dict);
 145         dictEmpty(server.db[j].dict);
 146         dictEmpty(server.db[j].expires);
 147     }
 148     return removed;
 149 }
 150
 151 int selectDb(redisClient *c, int id) {
 152     if (id < 0 || id >= server.dbnum)
 153         return REDIS_ERR;
 154     c->db = &server.db[id];
 155     return REDIS_OK;
 156 }
 157
 158 /*-----------------------------------------------------------------------------
 159  * Type agnostic commands operating on the key space
 160  *----------------------------------------------------------------------------*/
 161
 162 void flushdbCommand(redisClient *c) {
 163     server.dirty += dictSize(c->db->dict);
 164     touchWatchedKeysOnFlush(c->db->id);
 165     dictEmpty(c->db->dict);
 166     dictEmpty(c->db->expires);
 167     addReply(c,shared.ok);
 168 }
 169
 170 void flushallCommand(redisClient *c) {
 171     touchWatchedKeysOnFlush(-1);
 172     server.dirty += emptyDb();
 173     addReply(c,shared.ok);
 174     if (server.bgsavechildpid != -1) {
 175         kill(server.bgsavechildpid,SIGKILL);
 176         rdbRemoveTempFile(server.bgsavechildpid);
 177     }
 178     rdbSave(server.dbfilename);
 179     server.dirty++;
 180 }
 181
 182 void delCommand(redisClient *c) {
 183     int deleted = 0, j;
 184
 185     for (j = 1; j < c->argc; j++) {
 186         if (dbDelete(c->db,c->argv[j])) {
 187             touchWatchedKey(c->db,c->argv[j]);
 188             server.dirty++;
 189             deleted++;
 190         }
 191     }
 192     addReplyLongLong(c,deleted);
 193 }
 194
 195 void existsCommand(redisClient *c) {
 196     expireIfNeeded(c->db,c->argv[1]);
 197     if (dbExists(c->db,c->argv[1])) {
 198         addReply(c, shared.cone);
 199     } else {
 200         addReply(c, shared.czero);
 201     }
 202 }
 203
 204 void selectCommand(redisClient *c) {
 205     int id = atoi(c->argv[1]->ptr);
 206
 207     if (selectDb(c,id) == REDIS_ERR) {
 208         addReplyError(c,"invalid DB index");
 209     } else {
 210         addReply(c,shared.ok);
 211     }
 212 }
 213
 214 void randomkeyCommand(redisClient *c) {
 215     robj *key;
 216
 217     if ((key = dbRandomKey(c->db)) == NULL) {
 218         addReply(c,shared.nullbulk);
 219         return;
 220     }
 221
 222     addReplyBulk(c,key);
 223     decrRefCount(key);
 224 }
 225
 226 void keysCommand(redisClient *c) {
 227     dictIterator *di;
 228     dictEntry *de;
 229     sds pattern = c->argv[1]->ptr;
 230     int plen = sdslen(pattern), allkeys;
 231     unsigned long numkeys = 0;
 232     void *replylen = addDeferredMultiBulkLength(c);
 233
 234     di = dictGetIterator(c->db->dict);
 235     allkeys = (pattern[0] == '*' && pattern[1] == '\0');
 236     while((de = dictNext(di)) != NULL) {
 237         sds key = dictGetEntryKey(de);
 238         robj *keyobj;
 239
 240         if (allkeys || stringmatchlen(pattern,plen,key,sdslen(key),0)) {
 241             keyobj = createStringObject(key,sdslen(key));
 242             if (expireIfNeeded(c->db,keyobj) == 0) {
 243                 addReplyBulk(c,keyobj);
 244                 numkeys++;
 245             }
 246             decrRefCount(keyobj);
 247         }
 248     }
 249     dictReleaseIterator(di);
 250     setDeferredMultiBulkLength(c,replylen,numkeys);
 251 }
 252
 253 void dbsizeCommand(redisClient *c) {
 254     addReplyLongLong(c,dictSize(c->db->dict));
 255 }
 256
 257 void lastsaveCommand(redisClient *c) {
 258     addReplyLongLong(c,server.lastsave);
 259 }
 260
 261 void typeCommand(redisClient *c) {
 262     robj *o;
 263     char *type;
 264
 265     o = lookupKeyRead(c->db,c->argv[1]);
 266     if (o == NULL) {
 267         type = "none";
 268     } else {
 269         switch(o->type) {
 270         case REDIS_STRING: type = "string"; break;
 271         case REDIS_LIST: type = "list"; break;
 272         case REDIS_SET: type = "set"; break;
 273         case REDIS_ZSET: type = "zset"; break;
 274         case REDIS_HASH: type = "hash"; break;
 275         default: type = "unknown"; break;
 276         }
 277     }
 278     addReplyStatus(c,type);
 279 }
 280
 281 void saveCommand(redisClient *c) {
 282     if (server.bgsavechildpid != -1) {
 283         addReplyError(c,"Background save already in progress");
 284         return;
 285     }
 286     if (rdbSave(server.dbfilename) == REDIS_OK) {
 287         addReply(c,shared.ok);
 288     } else {
 289         addReply(c,shared.err);
 290     }
 291 }
 292
 293 void bgsaveCommand(redisClient *c) {
 294     if (server.bgsavechildpid != -1) {
 295         addReplyError(c,"Background save already in progress");
 296         return;
 297     }
 298     if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
 299         addReplyStatus(c,"Background saving started");
 300     } else {
 301         addReply(c,shared.err);
 302     }
 303 }
 304
 305 void shutdownCommand(redisClient *c) {
 306     if (prepareForShutdown() == REDIS_OK)
 307         exit(0);
 308     addReplyError(c,"Errors trying to SHUTDOWN. Check logs.");
 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     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     long 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 }